projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: cleanup transaction on abort
[linux-2.6-block.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
163e783e 29#include <linux/crc32c.h>
5a0e3ad6 30#include <linux/slab.h>
784b4e29 31#include <linux/migrate.h>
7a36ddec 32#include <linux/ratelimit.h>
6463fe58 33#include <linux/uuid.h>
803b2f54 34#include <linux/semaphore.h>
7e75bf3f 35#include <asm/unaligned.h>
4b4e25f2 36#include "compat.h"
eb60ceac
CM		 37#include "ctree.h"
38#include "disk-io.h"
e089f05c 39#include "transaction.h"
0f7d52f4 40#include "btrfs_inode.h"
0b86a832 41#include "volumes.h"
db94535d 42#include "print-tree.h"
8b712842 43#include "async-thread.h"
925baedd 44#include "locking.h"
e02119d5 45#include "tree-log.h"
fa9c0d79 46#include "free-space-cache.h"
581bb050 47#include "inode-map.h"
21adbd5c 48#include "check-integrity.h"
606686ee 49#include "rcu-string.h"
8dabb742 50#include "dev-replace.h"
53b381b3 51#include "raid56.h"
eb60ceac 52
de0022b9
JB		 53#ifdef CONFIG_X86
54#include <asm/cpufeature.h>
55#endif
56
d1310b2e 57static struct extent_io_ops btree_extent_io_ops;
8b712842 58static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 59static void free_fs_root(struct btrfs_root *root);
fcd1f065 60static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 61 int read_only);
569e0f35
JB		 62static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
63 struct btrfs_root *root);
143bede5 64static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
acce952b 65static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
66 struct btrfs_root *root);
aec8030a 67static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
143bede5 68static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
acce952b 69static int btrfs_destroy_marked_extents(struct btrfs_root *root,
70 struct extent_io_tree *dirty_pages,
71 int mark);
72static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
73 struct extent_io_tree *pinned_extents);
48a3b636
ES		 74static int btrfs_cleanup_transaction(struct btrfs_root *root);
75static void btrfs_error_commit_super(struct btrfs_root *root);
ce9adaa5 76
d352ac68
CM		 77/*
78 * end_io_wq structs are used to do processing in task context when an IO is
79 * complete. This is used during reads to verify checksums, and it is used
80 * by writes to insert metadata for new file extents after IO is complete.
81 */
ce9adaa5
CM		 82struct end_io_wq {
83 struct bio *bio;
84 bio_end_io_t *end_io;
85 void *private;
86 struct btrfs_fs_info *info;
87 int error;
22c59948 88 int metadata;
ce9adaa5 89 struct list_head list;
8b712842 90 struct btrfs_work work;
ce9adaa5 91};
0da5468f 92
d352ac68
CM		 93/*
94 * async submit bios are used to offload expensive checksumming
95 * onto the worker threads. They checksum file and metadata bios
96 * just before they are sent down the IO stack.
97 */
44b8bd7e
CM		 98struct async_submit_bio {
99 struct inode *inode;
100 struct bio *bio;
101 struct list_head list;
4a69a410
CM		 102 extent_submit_bio_hook_t *submit_bio_start;
103 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 104 int rw;
105 int mirror_num;
c8b97818 106 unsigned long bio_flags;
eaf25d93
CM		 107 /*
108 * bio_offset is optional, can be used if the pages in the bio
109 * can't tell us where in the file the bio should go
110 */
111 u64 bio_offset;
8b712842 112 struct btrfs_work work;
79787eaa 113 int error;
44b8bd7e
CM		 114};
115
85d4e461
CM		 116/*
117 * Lockdep class keys for extent_buffer->lock's in this root. For a given
118 * eb, the lockdep key is determined by the btrfs_root it belongs to and
119 * the level the eb occupies in the tree.
120 *
121 * Different roots are used for different purposes and may nest inside each
122 * other and they require separate keysets. As lockdep keys should be
123 * static, assign keysets according to the purpose of the root as indicated
124 * by btrfs_root->objectid. This ensures that all special purpose roots
125 * have separate keysets.
4008c04a 126 *
85d4e461
CM		 127 * Lock-nesting across peer nodes is always done with the immediate parent
128 * node locked thus preventing deadlock. As lockdep doesn't know this, use
129 * subclass to avoid triggering lockdep warning in such cases.
4008c04a 130 *
85d4e461
CM		 131 * The key is set by the readpage_end_io_hook after the buffer has passed
132 * csum validation but before the pages are unlocked. It is also set by
133 * btrfs_init_new_buffer on freshly allocated blocks.
4008c04a 134 *
85d4e461
CM		 135 * We also add a check to make sure the highest level of the tree is the
136 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
137 * needs update as well.
4008c04a
CM		 138 */
139#ifdef CONFIG_DEBUG_LOCK_ALLOC
140# if BTRFS_MAX_LEVEL != 8
141# error
142# endif
85d4e461
CM		 143
144static struct btrfs_lockdep_keyset {
145 u64 id; /* root objectid */
146 const char *name_stem; /* lock name stem */
147 char names[BTRFS_MAX_LEVEL + 1][20];
148 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
149} btrfs_lockdep_keysets[] = {
150 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
151 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
152 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
153 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
154 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
155 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
60b62978 156 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
85d4e461
CM		 157 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
158 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
159 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
13fd8da9 160 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
85d4e461 161 { .id = 0, .name_stem = "tree" },
4008c04a 162};
85d4e461
CM		 163
164void __init btrfs_init_lockdep(void)
165{
166 int i, j;
167
168 /* initialize lockdep class names */
169 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
170 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
171
172 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
173 snprintf(ks->names[j], sizeof(ks->names[j]),
174 "btrfs-%s-%02d", ks->name_stem, j);
175 }
176}
177
178void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
179 int level)
180{
181 struct btrfs_lockdep_keyset *ks;
182
183 BUG_ON(level >= ARRAY_SIZE(ks->keys));
184
185 /* find the matching keyset, id 0 is the default entry */
186 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
187 if (ks->id == objectid)
188 break;
189
190 lockdep_set_class_and_name(&eb->lock,
191 &ks->keys[level], ks->names[level]);
192}
193
4008c04a
CM		 194#endif
195
d352ac68
CM		 196/*
197 * extents on the btree inode are pretty simple, there's one extent
198 * that covers the entire device
199 */
b2950863 200static struct extent_map *btree_get_extent(struct inode *inode,
306e16ce 201 struct page *page, size_t pg_offset, u64 start, u64 len,
b2950863 202 int create)
7eccb903 203{
5f39d397
CM		 204 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
205 struct extent_map *em;
206 int ret;
207
890871be 208 read_lock(&em_tree->lock);
d1310b2e 209 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 210 if (em) {
211 em->bdev =
212 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 213 read_unlock(&em_tree->lock);
5f39d397 214 goto out;
a061fc8d 215 }
890871be 216 read_unlock(&em_tree->lock);
7b13b7b1 217
172ddd60 218 em = alloc_extent_map();
5f39d397
CM		 219 if (!em) {
220 em = ERR_PTR(-ENOMEM);
221 goto out;
222 }
223 em->start = 0;
0afbaf8c 224 em->len = (u64)-1;
c8b97818 225 em->block_len = (u64)-1;
5f39d397 226 em->block_start = 0;
a061fc8d 227 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 228
890871be 229 write_lock(&em_tree->lock);
09a2a8f9 230 ret = add_extent_mapping(em_tree, em, 0);
5f39d397
CM		 231 if (ret == -EEXIST) {
232 free_extent_map(em);
7b13b7b1 233 em = lookup_extent_mapping(em_tree, start, len);
b4f359ab 234 if (!em)
0433f20d 235 em = ERR_PTR(-EIO);
5f39d397 236 } else if (ret) {
7b13b7b1 237 free_extent_map(em);
0433f20d 238 em = ERR_PTR(ret);
5f39d397 239 }
890871be 240 write_unlock(&em_tree->lock);
7b13b7b1 241
5f39d397
CM		 242out:
243 return em;
7eccb903
CM		 244}
245
b0496686 246u32 btrfs_csum_data(char *data, u32 seed, size_t len)
19c00ddc 247{
163e783e 248 return crc32c(seed, data, len);
19c00ddc
CM		 249}
250
251void btrfs_csum_final(u32 crc, char *result)
252{
7e75bf3f 253 put_unaligned_le32(~crc, result);
19c00ddc
CM		 254}
255
d352ac68
CM		 256/*
257 * compute the csum for a btree block, and either verify it or write it
258 * into the csum field of the block.
259 */
19c00ddc
CM		 260static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
261 int verify)
262{
6c41761f 263 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
607d432d 264 char *result = NULL;
19c00ddc
CM		 265 unsigned long len;
266 unsigned long cur_len;
267 unsigned long offset = BTRFS_CSUM_SIZE;
19c00ddc
CM		 268 char *kaddr;
269 unsigned long map_start;
270 unsigned long map_len;
271 int err;
272 u32 crc = ~(u32)0;
607d432d 273 unsigned long inline_result;
19c00ddc
CM		 274
275 len = buf->len - offset;
d397712b 276 while (len > 0) {
19c00ddc 277 err = map_private_extent_buffer(buf, offset, 32,
a6591715 278 &kaddr, &map_start, &map_len);
d397712b 279 if (err)
19c00ddc 280 return 1;
19c00ddc 281 cur_len = min(len, map_len - (offset - map_start));
b0496686 282 crc = btrfs_csum_data(kaddr + offset - map_start,
19c00ddc
CM		 283 crc, cur_len);
284 len -= cur_len;
285 offset += cur_len;
19c00ddc 286 }
607d432d
JB		 287 if (csum_size > sizeof(inline_result)) {
288 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
289 if (!result)
290 return 1;
291 } else {
292 result = (char *)&inline_result;
293 }
294
19c00ddc
CM		 295 btrfs_csum_final(crc, result);
296
297 if (verify) {
607d432d 298 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 299 u32 val;
300 u32 found = 0;
607d432d 301 memcpy(&found, result, csum_size);
e4204ded 302
607d432d 303 read_extent_buffer(buf, &val, 0, csum_size);
7a36ddec 304 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
193f284d
CM		 305 "failed on %llu wanted %X found %X "
306 "level %d\n",
c1c9ff7c
GU		 307 root->fs_info->sb->s_id, buf->start,
308 val, found, btrfs_header_level(buf));
607d432d
JB		 309 if (result != (char *)&inline_result)
310 kfree(result);
19c00ddc
CM		 311 return 1;
312 }
313 } else {
607d432d 314 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 315 }
607d432d
JB		 316 if (result != (char *)&inline_result)
317 kfree(result);
19c00ddc
CM		 318 return 0;
319}
320
d352ac68
CM		 321/*
322 * we can't consider a given block up to date unless the transid of the
323 * block matches the transid in the parent node's pointer. This is how we
324 * detect blocks that either didn't get written at all or got written
325 * in the wrong place.
326 */
1259ab75 327static int verify_parent_transid(struct extent_io_tree *io_tree,
b9fab919
CM		 328 struct extent_buffer *eb, u64 parent_transid,
329 int atomic)
1259ab75 330{
2ac55d41 331 struct extent_state *cached_state = NULL;
1259ab75
CM		 332 int ret;
333
334 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
335 return 0;
336
b9fab919
CM		 337 if (atomic)
338 return -EAGAIN;
339
2ac55d41 340 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
d0082371 341 0, &cached_state);
0b32f4bb 342 if (extent_buffer_uptodate(eb) &&
1259ab75
CM		 343 btrfs_header_generation(eb) == parent_transid) {
344 ret = 0;
345 goto out;
346 }
7a36ddec 347 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
193f284d 348 "found %llu\n",
c1c9ff7c 349 eb->start, parent_transid, btrfs_header_generation(eb));
1259ab75 350 ret = 1;
0b32f4bb 351 clear_extent_buffer_uptodate(eb);
33958dc6 352out:
2ac55d41
JB		 353 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
354 &cached_state, GFP_NOFS);
1259ab75 355 return ret;
1259ab75
CM		 356}
357
1104a885
DS		 358/*
359 * Return 0 if the superblock checksum type matches the checksum value of that
360 * algorithm. Pass the raw disk superblock data.
361 */
362static int btrfs_check_super_csum(char *raw_disk_sb)
363{
364 struct btrfs_super_block *disk_sb =
365 (struct btrfs_super_block *)raw_disk_sb;
366 u16 csum_type = btrfs_super_csum_type(disk_sb);
367 int ret = 0;
368
369 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
370 u32 crc = ~(u32)0;
371 const int csum_size = sizeof(crc);
372 char result[csum_size];
373
374 /*
375 * The super_block structure does not span the whole
376 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
377 * is filled with zeros and is included in the checkum.
378 */
379 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
380 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
381 btrfs_csum_final(crc, result);
382
383 if (memcmp(raw_disk_sb, result, csum_size))
384 ret = 1;
667e7d94
CM		 385
386 if (ret && btrfs_super_generation(disk_sb) < 10) {
387 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
388 ret = 0;
389 }
1104a885
DS		 390 }
391
392 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
393 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
394 csum_type);
395 ret = 1;
396 }
397
398 return ret;
399}
400
d352ac68
CM		 401/*
402 * helper to read a given tree block, doing retries as required when
403 * the checksums don't match and we have alternate mirrors to try.
404 */
f188591e
CM		 405static int btree_read_extent_buffer_pages(struct btrfs_root *root,
406 struct extent_buffer *eb,
ca7a79ad 407 u64 start, u64 parent_transid)
f188591e
CM		 408{
409 struct extent_io_tree *io_tree;
ea466794 410 int failed = 0;
f188591e
CM		 411 int ret;
412 int num_copies = 0;
413 int mirror_num = 0;
ea466794 414 int failed_mirror = 0;
f188591e 415
a826d6dc 416 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM		 417 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
418 while (1) {
bb82ab88
AJ		 419 ret = read_extent_buffer_pages(io_tree, eb, start,
420 WAIT_COMPLETE,
f188591e 421 btree_get_extent, mirror_num);
256dd1bb
SB		 422 if (!ret) {
423 if (!verify_parent_transid(io_tree, eb,
b9fab919 424 parent_transid, 0))
256dd1bb
SB		 425 break;
426 else
427 ret = -EIO;
428 }
d397712b 429
a826d6dc
JB		 430 /*
431 * This buffer's crc is fine, but its contents are corrupted, so
432 * there is no reason to read the other copies, they won't be
433 * any less wrong.
434 */
435 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
ea466794
JB		 436 break;
437
5d964051 438 num_copies = btrfs_num_copies(root->fs_info,
f188591e 439 eb->start, eb->len);
4235298e 440 if (num_copies == 1)
ea466794 441 break;
4235298e 442
5cf1ab56
JB		 443 if (!failed_mirror) {
444 failed = 1;
445 failed_mirror = eb->read_mirror;
446 }
447
f188591e 448 mirror_num++;
ea466794
JB		 449 if (mirror_num == failed_mirror)
450 mirror_num++;
451
4235298e 452 if (mirror_num > num_copies)
ea466794 453 break;
f188591e 454 }
ea466794 455
c0901581 456 if (failed && !ret && failed_mirror)
ea466794
JB		 457 repair_eb_io_failure(root, eb, failed_mirror);
458
459 return ret;
f188591e 460}
19c00ddc 461
d352ac68 462/*
d397712b
CM		 463 * checksum a dirty tree block before IO. This has extra checks to make sure
464 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 465 */
d397712b 466
b2950863 467static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 468{
d1310b2e 469 struct extent_io_tree *tree;
4eee4fa4 470 u64 start = page_offset(page);
19c00ddc 471 u64 found_start;
19c00ddc 472 struct extent_buffer *eb;
f188591e 473
d1310b2e 474 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc 475
4f2de97a
JB		 476 eb = (struct extent_buffer *)page->private;
477 if (page != eb->pages[0])
478 return 0;
19c00ddc
CM		 479 found_start = btrfs_header_bytenr(eb);
480 if (found_start != start) {
55c69072 481 WARN_ON(1);
4f2de97a 482 return 0;
55c69072 483 }
55c69072 484 if (!PageUptodate(page)) {
55c69072 485 WARN_ON(1);
4f2de97a 486 return 0;
19c00ddc 487 }
19c00ddc 488 csum_tree_block(root, eb, 0);
19c00ddc
CM		 489 return 0;
490}
491
2b82032c
YZ		 492static int check_tree_block_fsid(struct btrfs_root *root,
493 struct extent_buffer *eb)
494{
495 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
496 u8 fsid[BTRFS_UUID_SIZE];
497 int ret = 1;
498
fba6aa75 499 read_extent_buffer(eb, fsid, btrfs_header_fsid(eb), BTRFS_FSID_SIZE);
2b82032c
YZ		 500 while (fs_devices) {
501 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
502 ret = 0;
503 break;
504 }
505 fs_devices = fs_devices->seed;
506 }
507 return ret;
508}
509
a826d6dc
JB		 510#define CORRUPT(reason, eb, root, slot) \
511 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
512 "root=%llu, slot=%d\n", reason, \
c1c9ff7c 513 btrfs_header_bytenr(eb), root->objectid, slot)
a826d6dc
JB		 514
515static noinline int check_leaf(struct btrfs_root *root,
516 struct extent_buffer *leaf)
517{
518 struct btrfs_key key;
519 struct btrfs_key leaf_key;
520 u32 nritems = btrfs_header_nritems(leaf);
521 int slot;
522
523 if (nritems == 0)
524 return 0;
525
526 /* Check the 0 item */
527 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
528 BTRFS_LEAF_DATA_SIZE(root)) {
529 CORRUPT("invalid item offset size pair", leaf, root, 0);
530 return -EIO;
531 }
532
533 /*
534 * Check to make sure each items keys are in the correct order and their
535 * offsets make sense. We only have to loop through nritems-1 because
536 * we check the current slot against the next slot, which verifies the
537 * next slot's offset+size makes sense and that the current's slot
538 * offset is correct.
539 */
540 for (slot = 0; slot < nritems - 1; slot++) {
541 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
542 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
543
544 /* Make sure the keys are in the right order */
545 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
546 CORRUPT("bad key order", leaf, root, slot);
547 return -EIO;
548 }
549
550 /*
551 * Make sure the offset and ends are right, remember that the
552 * item data starts at the end of the leaf and grows towards the
553 * front.
554 */
555 if (btrfs_item_offset_nr(leaf, slot) !=
556 btrfs_item_end_nr(leaf, slot + 1)) {
557 CORRUPT("slot offset bad", leaf, root, slot);
558 return -EIO;
559 }
560
561 /*
562 * Check to make sure that we don't point outside of the leaf,
563 * just incase all the items are consistent to eachother, but
564 * all point outside of the leaf.
565 */
566 if (btrfs_item_end_nr(leaf, slot) >
567 BTRFS_LEAF_DATA_SIZE(root)) {
568 CORRUPT("slot end outside of leaf", leaf, root, slot);
569 return -EIO;
570 }
571 }
572
573 return 0;
574}
575
facc8a22
MX		 576static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
577 u64 phy_offset, struct page *page,
578 u64 start, u64 end, int mirror)
ce9adaa5
CM		 579{
580 struct extent_io_tree *tree;
581 u64 found_start;
582 int found_level;
ce9adaa5
CM		 583 struct extent_buffer *eb;
584 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 585 int ret = 0;
727011e0 586 int reads_done;
ce9adaa5 587
ce9adaa5
CM		 588 if (!page->private)
589 goto out;
d397712b 590
727011e0 591 tree = &BTRFS_I(page->mapping->host)->io_tree;
4f2de97a 592 eb = (struct extent_buffer *)page->private;
d397712b 593
0b32f4bb
JB		 594 /* the pending IO might have been the only thing that kept this buffer
595 * in memory. Make sure we have a ref for all this other checks
596 */
597 extent_buffer_get(eb);
598
599 reads_done = atomic_dec_and_test(&eb->io_pages);
727011e0
CM		 600 if (!reads_done)
601 goto err;
f188591e 602
5cf1ab56 603 eb->read_mirror = mirror;
ea466794
JB		 604 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
605 ret = -EIO;
606 goto err;
607 }
608
ce9adaa5 609 found_start = btrfs_header_bytenr(eb);
727011e0 610 if (found_start != eb->start) {
7a36ddec 611 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
193f284d 612 "%llu %llu\n",
c1c9ff7c 613 found_start, eb->start);
f188591e 614 ret = -EIO;
ce9adaa5
CM		 615 goto err;
616 }
2b82032c 617 if (check_tree_block_fsid(root, eb)) {
7a36ddec 618 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
c1c9ff7c 619 eb->start);
1259ab75
CM		 620 ret = -EIO;
621 goto err;
622 }
ce9adaa5 623 found_level = btrfs_header_level(eb);
1c24c3ce
JB		 624 if (found_level >= BTRFS_MAX_LEVEL) {
625 btrfs_info(root->fs_info, "bad tree block level %d\n",
626 (int)btrfs_header_level(eb));
627 ret = -EIO;
628 goto err;
629 }
ce9adaa5 630
85d4e461
CM		 631 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
632 eb, found_level);
4008c04a 633
ce9adaa5 634 ret = csum_tree_block(root, eb, 1);
a826d6dc 635 if (ret) {
f188591e 636 ret = -EIO;
a826d6dc
JB		 637 goto err;
638 }
639
640 /*
641 * If this is a leaf block and it is corrupt, set the corrupt bit so
642 * that we don't try and read the other copies of this block, just
643 * return -EIO.
644 */
645 if (found_level == 0 && check_leaf(root, eb)) {
646 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
647 ret = -EIO;
648 }
ce9adaa5 649
0b32f4bb
JB		 650 if (!ret)
651 set_extent_buffer_uptodate(eb);
ce9adaa5 652err:
79fb65a1
JB		 653 if (reads_done &&
654 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 655 btree_readahead_hook(root, eb, eb->start, ret);
4bb31e92 656
53b381b3
DW		 657 if (ret) {
658 /*
659 * our io error hook is going to dec the io pages
660 * again, we have to make sure it has something
661 * to decrement
662 */
663 atomic_inc(&eb->io_pages);
0b32f4bb 664 clear_extent_buffer_uptodate(eb);
53b381b3 665 }
0b32f4bb 666 free_extent_buffer(eb);
ce9adaa5 667out:
f188591e 668 return ret;
ce9adaa5
CM		 669}
670
ea466794 671static int btree_io_failed_hook(struct page *page, int failed_mirror)
4bb31e92 672{
4bb31e92
AJ		 673 struct extent_buffer *eb;
674 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
675
4f2de97a 676 eb = (struct extent_buffer *)page->private;
ea466794 677 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
5cf1ab56 678 eb->read_mirror = failed_mirror;
53b381b3 679 atomic_dec(&eb->io_pages);
ea466794 680 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 681 btree_readahead_hook(root, eb, eb->start, -EIO);
4bb31e92
AJ		 682 return -EIO; /* we fixed nothing */
683}
684
ce9adaa5 685static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 686{
687 struct end_io_wq *end_io_wq = bio->bi_private;
688 struct btrfs_fs_info *fs_info;
ce9adaa5 689
ce9adaa5 690 fs_info = end_io_wq->info;
ce9adaa5 691 end_io_wq->error = err;
8b712842
CM		 692 end_io_wq->work.func = end_workqueue_fn;
693 end_io_wq->work.flags = 0;
d20f7043 694
7b6d91da 695 if (bio->bi_rw & REQ_WRITE) {
53b381b3 696 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
cad321ad
CM		 697 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
698 &end_io_wq->work);
53b381b3 699 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
0cb59c99
JB		 700 btrfs_queue_worker(&fs_info->endio_freespace_worker,
701 &end_io_wq->work);
53b381b3
DW		 702 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
703 btrfs_queue_worker(&fs_info->endio_raid56_workers,
704 &end_io_wq->work);
cad321ad
CM		 705 else
706 btrfs_queue_worker(&fs_info->endio_write_workers,
707 &end_io_wq->work);
d20f7043 708 } else {
53b381b3
DW		 709 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
710 btrfs_queue_worker(&fs_info->endio_raid56_workers,
711 &end_io_wq->work);
712 else if (end_io_wq->metadata)
d20f7043
CM		 713 btrfs_queue_worker(&fs_info->endio_meta_workers,
714 &end_io_wq->work);
715 else
716 btrfs_queue_worker(&fs_info->endio_workers,
717 &end_io_wq->work);
718 }
ce9adaa5
CM		 719}
720
0cb59c99
JB		 721/*
722 * For the metadata arg you want
723 *
724 * 0 - if data
725 * 1 - if normal metadta
726 * 2 - if writing to the free space cache area
53b381b3 727 * 3 - raid parity work
0cb59c99 728 */
22c59948
CM		 729int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
730 int metadata)
0b86a832 731{
ce9adaa5 732 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 733 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
734 if (!end_io_wq)
735 return -ENOMEM;
736
737 end_io_wq->private = bio->bi_private;
738 end_io_wq->end_io = bio->bi_end_io;
22c59948 739 end_io_wq->info = info;
ce9adaa5
CM		 740 end_io_wq->error = 0;
741 end_io_wq->bio = bio;
22c59948 742 end_io_wq->metadata = metadata;
ce9adaa5
CM		 743
744 bio->bi_private = end_io_wq;
745 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 746 return 0;
747}
748
b64a2851 749unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 750{
4854ddd0
CM		 751 unsigned long limit = min_t(unsigned long,
752 info->workers.max_workers,
753 info->fs_devices->open_devices);
754 return 256 * limit;
755}
0986fe9e 756
4a69a410
CM		 757static void run_one_async_start(struct btrfs_work *work)
758{
4a69a410 759 struct async_submit_bio *async;
79787eaa 760 int ret;
4a69a410
CM		 761
762 async = container_of(work, struct async_submit_bio, work);
79787eaa
JM		 763 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
764 async->mirror_num, async->bio_flags,
765 async->bio_offset);
766 if (ret)
767 async->error = ret;
4a69a410
CM		 768}
769
770static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 771{
772 struct btrfs_fs_info *fs_info;
773 struct async_submit_bio *async;
4854ddd0 774 int limit;
8b712842
CM		 775
776 async = container_of(work, struct async_submit_bio, work);
777 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 778
b64a2851 779 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 780 limit = limit * 2 / 3;
781
66657b31 782 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
b64a2851 783 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 784 wake_up(&fs_info->async_submit_wait);
785
79787eaa
JM		 786 /* If an error occured we just want to clean up the bio and move on */
787 if (async->error) {
788 bio_endio(async->bio, async->error);
789 return;
790 }
791
4a69a410 792 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 793 async->mirror_num, async->bio_flags,
794 async->bio_offset);
4a69a410
CM		 795}
796
797static void run_one_async_free(struct btrfs_work *work)
798{
799 struct async_submit_bio *async;
800
801 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 802 kfree(async);
803}
804
44b8bd7e
CM		 805int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
806 int rw, struct bio *bio, int mirror_num,
c8b97818 807 unsigned long bio_flags,
eaf25d93 808 u64 bio_offset,
4a69a410
CM		 809 extent_submit_bio_hook_t *submit_bio_start,
810 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 811{
812 struct async_submit_bio *async;
813
814 async = kmalloc(sizeof(*async), GFP_NOFS);
815 if (!async)
816 return -ENOMEM;
817
818 async->inode = inode;
819 async->rw = rw;
820 async->bio = bio;
821 async->mirror_num = mirror_num;
4a69a410
CM		 822 async->submit_bio_start = submit_bio_start;
823 async->submit_bio_done = submit_bio_done;
824
825 async->work.func = run_one_async_start;
826 async->work.ordered_func = run_one_async_done;
827 async->work.ordered_free = run_one_async_free;
828
8b712842 829 async->work.flags = 0;
c8b97818 830 async->bio_flags = bio_flags;
eaf25d93 831 async->bio_offset = bio_offset;
8c8bee1d 832
79787eaa
JM		 833 async->error = 0;
834
cb03c743 835 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 836
7b6d91da 837 if (rw & REQ_SYNC)
d313d7a3
CM		 838 btrfs_set_work_high_prio(&async->work);
839
8b712842 840 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 841
d397712b 842 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 843 atomic_read(&fs_info->nr_async_submits)) {
844 wait_event(fs_info->async_submit_wait,
845 (atomic_read(&fs_info->nr_async_submits) == 0));
846 }
847
44b8bd7e
CM		 848 return 0;
849}
850
ce3ed71a
CM		 851static int btree_csum_one_bio(struct bio *bio)
852{
853 struct bio_vec *bvec = bio->bi_io_vec;
854 int bio_index = 0;
855 struct btrfs_root *root;
79787eaa 856 int ret = 0;
ce3ed71a
CM		 857
858 WARN_ON(bio->bi_vcnt <= 0);
d397712b 859 while (bio_index < bio->bi_vcnt) {
ce3ed71a 860 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 861 ret = csum_dirty_buffer(root, bvec->bv_page);
862 if (ret)
863 break;
ce3ed71a
CM		 864 bio_index++;
865 bvec++;
866 }
79787eaa 867 return ret;
ce3ed71a
CM		 868}
869
4a69a410
CM		 870static int __btree_submit_bio_start(struct inode *inode, int rw,
871 struct bio *bio, int mirror_num,
eaf25d93
CM		 872 unsigned long bio_flags,
873 u64 bio_offset)
22c59948 874{
8b712842
CM		 875 /*
876 * when we're called for a write, we're already in the async
5443be45 877 * submission context. Just jump into btrfs_map_bio
8b712842 878 */
79787eaa 879 return btree_csum_one_bio(bio);
4a69a410 880}
22c59948 881
4a69a410 882static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 883 int mirror_num, unsigned long bio_flags,
884 u64 bio_offset)
4a69a410 885{
61891923
SB		 886 int ret;
887
8b712842 888 /*
4a69a410
CM		 889 * when we're called for a write, we're already in the async
890 * submission context. Just jump into btrfs_map_bio
8b712842 891 */
61891923
SB		 892 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
893 if (ret)
894 bio_endio(bio, ret);
895 return ret;
0b86a832
CM		 896}
897
de0022b9
JB		 898static int check_async_write(struct inode *inode, unsigned long bio_flags)
899{
900 if (bio_flags & EXTENT_BIO_TREE_LOG)
901 return 0;
902#ifdef CONFIG_X86
903 if (cpu_has_xmm4_2)
904 return 0;
905#endif
906 return 1;
907}
908
44b8bd7e 909static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 910 int mirror_num, unsigned long bio_flags,
911 u64 bio_offset)
44b8bd7e 912{
de0022b9 913 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 914 int ret;
915
7b6d91da 916 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 917 /*
918 * called for a read, do the setup so that checksum validation
919 * can happen in the async kernel threads
920 */
f3f266ab
CM		 921 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
922 bio, 1);
1d4284bd 923 if (ret)
61891923
SB		 924 goto out_w_error;
925 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
926 mirror_num, 0);
de0022b9
JB		 927 } else if (!async) {
928 ret = btree_csum_one_bio(bio);
929 if (ret)
61891923
SB		 930 goto out_w_error;
931 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
932 mirror_num, 0);
933 } else {
934 /*
935 * kthread helpers are used to submit writes so that
936 * checksumming can happen in parallel across all CPUs
937 */
938 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
939 inode, rw, bio, mirror_num, 0,
940 bio_offset,
941 __btree_submit_bio_start,
942 __btree_submit_bio_done);
44b8bd7e 943 }
d313d7a3 944
61891923
SB		 945 if (ret) {
946out_w_error:
947 bio_endio(bio, ret);
948 }
949 return ret;
44b8bd7e
CM		 950}
951
3dd1462e 952#ifdef CONFIG_MIGRATION
784b4e29 953static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 954 struct page *newpage, struct page *page,
955 enum migrate_mode mode)
784b4e29
CM		 956{
957 /*
958 * we can't safely write a btree page from here,
959 * we haven't done the locking hook
960 */
961 if (PageDirty(page))
962 return -EAGAIN;
963 /*
964 * Buffers may be managed in a filesystem specific way.
965 * We must have no buffers or drop them.
966 */
967 if (page_has_private(page) &&
968 !try_to_release_page(page, GFP_KERNEL))
969 return -EAGAIN;
a6bc32b8 970 return migrate_page(mapping, newpage, page, mode);
784b4e29 971}
3dd1462e 972#endif
784b4e29 973
0da5468f
CM		 974
975static int btree_writepages(struct address_space *mapping,
976 struct writeback_control *wbc)
977{
d1310b2e 978 struct extent_io_tree *tree;
e2d84521
MX		 979 struct btrfs_fs_info *fs_info;
980 int ret;
981
d1310b2e 982 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 983 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 984
985 if (wbc->for_kupdate)
986 return 0;
987
e2d84521 988 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 989 /* this is a bit racy, but that's ok */
e2d84521
MX		 990 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
991 BTRFS_DIRTY_METADATA_THRESH);
992 if (ret < 0)
793955bc 993 return 0;
793955bc 994 }
0b32f4bb 995 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 996}
997
b2950863 998static int btree_readpage(struct file *file, struct page *page)
5f39d397 999{
d1310b2e
CM		 1000 struct extent_io_tree *tree;
1001 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 1002 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 1003}
22b0ebda 1004
70dec807 1005static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 1006{
98509cfc 1007 if (PageWriteback(page) || PageDirty(page))
d397712b 1008 return 0;
0c4e538b 1009
f7a52a40 1010 return try_release_extent_buffer(page);
d98237b3
CM		 1011}
1012
d47992f8
LC		 1013static void btree_invalidatepage(struct page *page, unsigned int offset,
1014 unsigned int length)
d98237b3 1015{
d1310b2e
CM		 1016 struct extent_io_tree *tree;
1017 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1018 extent_invalidatepage(tree, page, offset);
1019 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1020 if (PagePrivate(page)) {
d397712b
CM		 1021 printk(KERN_WARNING "btrfs warning page private not zero "
1022 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1023 ClearPagePrivate(page);
1024 set_page_private(page, 0);
1025 page_cache_release(page);
1026 }
d98237b3
CM		 1027}
1028
0b32f4bb
JB		 1029static int btree_set_page_dirty(struct page *page)
1030{
bb146eb2 1031#ifdef DEBUG
0b32f4bb
JB		 1032 struct extent_buffer *eb;
1033
1034 BUG_ON(!PagePrivate(page));
1035 eb = (struct extent_buffer *)page->private;
1036 BUG_ON(!eb);
1037 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 BUG_ON(!atomic_read(&eb->refs));
1039 btrfs_assert_tree_locked(eb);
bb146eb2 1040#endif
0b32f4bb
JB		 1041 return __set_page_dirty_nobuffers(page);
1042}
1043
7f09410b 1044static const struct address_space_operations btree_aops = {
d98237b3 1045 .readpage = btree_readpage,
0da5468f 1046 .writepages = btree_writepages,
5f39d397
CM		 1047 .releasepage = btree_releasepage,
1048 .invalidatepage = btree_invalidatepage,
5a92bc88 1049#ifdef CONFIG_MIGRATION
784b4e29 1050 .migratepage = btree_migratepage,
5a92bc88 1051#endif
0b32f4bb 1052 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1053};
1054
ca7a79ad
CM		 1055int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1056 u64 parent_transid)
090d1875 1057{
5f39d397
CM		 1058 struct extent_buffer *buf = NULL;
1059 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1060 int ret = 0;
090d1875 1061
db94535d 1062 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1063 if (!buf)
090d1875 1064 return 0;
d1310b2e 1065 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1066 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1067 free_extent_buffer(buf);
de428b63 1068 return ret;
090d1875
CM		 1069}
1070
ab0fff03
AJ		 1071int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1072 int mirror_num, struct extent_buffer **eb)
1073{
1074 struct extent_buffer *buf = NULL;
1075 struct inode *btree_inode = root->fs_info->btree_inode;
1076 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1077 int ret;
1078
1079 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1080 if (!buf)
1081 return 0;
1082
1083 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1084
1085 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1086 btree_get_extent, mirror_num);
1087 if (ret) {
1088 free_extent_buffer(buf);
1089 return ret;
1090 }
1091
1092 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1093 free_extent_buffer(buf);
1094 return -EIO;
0b32f4bb 1095 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1096 *eb = buf;
1097 } else {
1098 free_extent_buffer(buf);
1099 }
1100 return 0;
1101}
1102
0999df54
CM		 1103struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1104 u64 bytenr, u32 blocksize)
1105{
1106 struct inode *btree_inode = root->fs_info->btree_inode;
1107 struct extent_buffer *eb;
1108 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
f09d1f60 1109 bytenr, blocksize);
0999df54
CM		 1110 return eb;
1111}
1112
1113struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1114 u64 bytenr, u32 blocksize)
1115{
1116 struct inode *btree_inode = root->fs_info->btree_inode;
1117 struct extent_buffer *eb;
1118
1119 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
727011e0 1120 bytenr, blocksize);
0999df54
CM		 1121 return eb;
1122}
1123
1124
e02119d5
CM		 1125int btrfs_write_tree_block(struct extent_buffer *buf)
1126{
727011e0 1127 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
8aa38c31 1128 buf->start + buf->len - 1);
e02119d5
CM		 1129}
1130
1131int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1132{
727011e0 1133 return filemap_fdatawait_range(buf->pages[0]->mapping,
8aa38c31 1134 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1135}
1136
0999df54 1137struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1138 u32 blocksize, u64 parent_transid)
0999df54
CM		 1139{
1140 struct extent_buffer *buf = NULL;
0999df54
CM		 1141 int ret;
1142
0999df54
CM		 1143 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1144 if (!buf)
1145 return NULL;
0999df54 1146
ca7a79ad 1147 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
0f0fe8f7
FDBM		 1148 if (ret) {
1149 free_extent_buffer(buf);
1150 return NULL;
1151 }
5f39d397 1152 return buf;
ce9adaa5 1153
eb60ceac
CM		 1154}
1155
d5c13f92
JM		 1156void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1157 struct extent_buffer *buf)
ed2ff2cb 1158{
e2d84521
MX		 1159 struct btrfs_fs_info *fs_info = root->fs_info;
1160
55c69072 1161 if (btrfs_header_generation(buf) ==
e2d84521 1162 fs_info->running_transaction->transid) {
b9447ef8 1163 btrfs_assert_tree_locked(buf);
b4ce94de 1164
b9473439 1165 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1166 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1167 -buf->len,
1168 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1169 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1170 btrfs_set_lock_blocking(buf);
1171 clear_extent_buffer_dirty(buf);
1172 }
925baedd 1173 }
5f39d397
CM		 1174}
1175
143bede5
JM		 1176static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1177 u32 stripesize, struct btrfs_root *root,
1178 struct btrfs_fs_info *fs_info,
1179 u64 objectid)
d97e63b6 1180{
cfaa7295 1181 root->node = NULL;
a28ec197 1182 root->commit_root = NULL;
db94535d
CM		 1183 root->sectorsize = sectorsize;
1184 root->nodesize = nodesize;
1185 root->leafsize = leafsize;
87ee04eb 1186 root->stripesize = stripesize;
123abc88 1187 root->ref_cows = 0;
0b86a832 1188 root->track_dirty = 0;
c71bf099 1189 root->in_radix = 0;
d68fc57b
YZ		 1190 root->orphan_item_inserted = 0;
1191 root->orphan_cleanup_state = 0;
0b86a832 1192
0f7d52f4
CM		 1193 root->objectid = objectid;
1194 root->last_trans = 0;
13a8a7c8 1195 root->highest_objectid = 0;
eb73c1b7 1196 root->nr_delalloc_inodes = 0;
199c2a9c 1197 root->nr_ordered_extents = 0;
58176a96 1198 root->name = NULL;
6bef4d31 1199 root->inode_tree = RB_ROOT;
16cdcec7 1200 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1201 root->block_rsv = NULL;
d68fc57b 1202 root->orphan_block_rsv = NULL;
0b86a832
CM		 1203
1204 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1205 INIT_LIST_HEAD(&root->root_list);
eb73c1b7
MX		 1206 INIT_LIST_HEAD(&root->delalloc_inodes);
1207 INIT_LIST_HEAD(&root->delalloc_root);
199c2a9c
MX		 1208 INIT_LIST_HEAD(&root->ordered_extents);
1209 INIT_LIST_HEAD(&root->ordered_root);
2ab28f32
JB		 1210 INIT_LIST_HEAD(&root->logged_list[0]);
1211 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1212 spin_lock_init(&root->orphan_lock);
5d4f98a2 1213 spin_lock_init(&root->inode_lock);
eb73c1b7 1214 spin_lock_init(&root->delalloc_lock);
199c2a9c 1215 spin_lock_init(&root->ordered_extent_lock);
f0486c68 1216 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1217 spin_lock_init(&root->log_extents_lock[0]);
1218 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1219 mutex_init(&root->objectid_mutex);
e02119d5 1220 mutex_init(&root->log_mutex);
7237f183
YZ		 1221 init_waitqueue_head(&root->log_writer_wait);
1222 init_waitqueue_head(&root->log_commit_wait[0]);
1223 init_waitqueue_head(&root->log_commit_wait[1]);
1224 atomic_set(&root->log_commit[0], 0);
1225 atomic_set(&root->log_commit[1], 0);
1226 atomic_set(&root->log_writers, 0);
2ecb7923 1227 atomic_set(&root->log_batch, 0);
8a35d95f 1228 atomic_set(&root->orphan_inodes, 0);
b0feb9d9 1229 atomic_set(&root->refs, 1);
7237f183 1230 root->log_transid = 0;
257c62e1 1231 root->last_log_commit = 0;
06ea65a3
JB		 1232 if (fs_info)
1233 extent_io_tree_init(&root->dirty_log_pages,
1234 fs_info->btree_inode->i_mapping);
017e5369 1235
3768f368
CM		 1236 memset(&root->root_key, 0, sizeof(root->root_key));
1237 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1238 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1239 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
06ea65a3
JB		 1240 if (fs_info)
1241 root->defrag_trans_start = fs_info->generation;
1242 else
1243 root->defrag_trans_start = 0;
58176a96 1244 init_completion(&root->kobj_unregister);
6702ed49 1245 root->defrag_running = 0;
4d775673 1246 root->root_key.objectid = objectid;
0ee5dc67 1247 root->anon_dev = 0;
8ea05e3a 1248
5f3ab90a 1249 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1250}
1251
f84a8bd6 1252static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1253{
1254 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1255 if (root)
1256 root->fs_info = fs_info;
1257 return root;
1258}
1259
06ea65a3
JB		 1260#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1261/* Should only be used by the testing infrastructure */
1262struct btrfs_root *btrfs_alloc_dummy_root(void)
1263{
1264 struct btrfs_root *root;
1265
1266 root = btrfs_alloc_root(NULL);
1267 if (!root)
1268 return ERR_PTR(-ENOMEM);
1269 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1270 root->dummy_root = 1;
1271
1272 return root;
1273}
1274#endif
1275
20897f5c
AJ		 1276struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1277 struct btrfs_fs_info *fs_info,
1278 u64 objectid)
1279{
1280 struct extent_buffer *leaf;
1281 struct btrfs_root *tree_root = fs_info->tree_root;
1282 struct btrfs_root *root;
1283 struct btrfs_key key;
1284 int ret = 0;
1285 u64 bytenr;
6463fe58 1286 uuid_le uuid;
20897f5c
AJ		 1287
1288 root = btrfs_alloc_root(fs_info);
1289 if (!root)
1290 return ERR_PTR(-ENOMEM);
1291
1292 __setup_root(tree_root->nodesize, tree_root->leafsize,
1293 tree_root->sectorsize, tree_root->stripesize,
1294 root, fs_info, objectid);
1295 root->root_key.objectid = objectid;
1296 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1297 root->root_key.offset = 0;
1298
1299 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1300 0, objectid, NULL, 0, 0, 0);
1301 if (IS_ERR(leaf)) {
1302 ret = PTR_ERR(leaf);
1dd05682 1303 leaf = NULL;
20897f5c
AJ		 1304 goto fail;
1305 }
1306
1307 bytenr = leaf->start;
1308 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1309 btrfs_set_header_bytenr(leaf, leaf->start);
1310 btrfs_set_header_generation(leaf, trans->transid);
1311 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1312 btrfs_set_header_owner(leaf, objectid);
1313 root->node = leaf;
1314
fba6aa75 1315 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(leaf),
20897f5c
AJ		 1316 BTRFS_FSID_SIZE);
1317 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
b308bc2f 1318 btrfs_header_chunk_tree_uuid(leaf),
20897f5c
AJ		 1319 BTRFS_UUID_SIZE);
1320 btrfs_mark_buffer_dirty(leaf);
1321
1322 root->commit_root = btrfs_root_node(root);
1323 root->track_dirty = 1;
1324
1325
1326 root->root_item.flags = 0;
1327 root->root_item.byte_limit = 0;
1328 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1329 btrfs_set_root_generation(&root->root_item, trans->transid);
1330 btrfs_set_root_level(&root->root_item, 0);
1331 btrfs_set_root_refs(&root->root_item, 1);
1332 btrfs_set_root_used(&root->root_item, leaf->len);
1333 btrfs_set_root_last_snapshot(&root->root_item, 0);
1334 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1335 uuid_le_gen(&uuid);
1336 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1337 root->root_item.drop_level = 0;
1338
1339 key.objectid = objectid;
1340 key.type = BTRFS_ROOT_ITEM_KEY;
1341 key.offset = 0;
1342 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1343 if (ret)
1344 goto fail;
1345
1346 btrfs_tree_unlock(leaf);
1347
1dd05682
TI		 1348 return root;
1349
20897f5c 1350fail:
1dd05682
TI		 1351 if (leaf) {
1352 btrfs_tree_unlock(leaf);
1353 free_extent_buffer(leaf);
1354 }
1355 kfree(root);
20897f5c 1356
1dd05682 1357 return ERR_PTR(ret);
20897f5c
AJ		 1358}
1359
7237f183
YZ		 1360static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1361 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1362{
1363 struct btrfs_root *root;
1364 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1365 struct extent_buffer *leaf;
e02119d5 1366
6f07e42e 1367 root = btrfs_alloc_root(fs_info);
e02119d5 1368 if (!root)
7237f183 1369 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1370
1371 __setup_root(tree_root->nodesize, tree_root->leafsize,
1372 tree_root->sectorsize, tree_root->stripesize,
1373 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1374
1375 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1376 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1377 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1378 /*
1379 * log trees do not get reference counted because they go away
1380 * before a real commit is actually done. They do store pointers
1381 * to file data extents, and those reference counts still get
1382 * updated (along with back refs to the log tree).
1383 */
e02119d5
CM		 1384 root->ref_cows = 0;
1385
5d4f98a2 1386 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1387 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1388 0, 0, 0);
7237f183
YZ		 1389 if (IS_ERR(leaf)) {
1390 kfree(root);
1391 return ERR_CAST(leaf);
1392 }
e02119d5 1393
5d4f98a2
YZ		 1394 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1395 btrfs_set_header_bytenr(leaf, leaf->start);
1396 btrfs_set_header_generation(leaf, trans->transid);
1397 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1398 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1399 root->node = leaf;
e02119d5
CM		 1400
1401 write_extent_buffer(root->node, root->fs_info->fsid,
fba6aa75 1402 btrfs_header_fsid(root->node), BTRFS_FSID_SIZE);
e02119d5
CM		 1403 btrfs_mark_buffer_dirty(root->node);
1404 btrfs_tree_unlock(root->node);
7237f183
YZ		 1405 return root;
1406}
1407
1408int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1409 struct btrfs_fs_info *fs_info)
1410{
1411 struct btrfs_root *log_root;
1412
1413 log_root = alloc_log_tree(trans, fs_info);
1414 if (IS_ERR(log_root))
1415 return PTR_ERR(log_root);
1416 WARN_ON(fs_info->log_root_tree);
1417 fs_info->log_root_tree = log_root;
1418 return 0;
1419}
1420
1421int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1422 struct btrfs_root *root)
1423{
1424 struct btrfs_root *log_root;
1425 struct btrfs_inode_item *inode_item;
1426
1427 log_root = alloc_log_tree(trans, root->fs_info);
1428 if (IS_ERR(log_root))
1429 return PTR_ERR(log_root);
1430
1431 log_root->last_trans = trans->transid;
1432 log_root->root_key.offset = root->root_key.objectid;
1433
1434 inode_item = &log_root->root_item.inode;
3cae210f
QW		 1435 btrfs_set_stack_inode_generation(inode_item, 1);
1436 btrfs_set_stack_inode_size(inode_item, 3);
1437 btrfs_set_stack_inode_nlink(inode_item, 1);
1438 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1439 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
7237f183 1440
5d4f98a2 1441 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1442
1443 WARN_ON(root->log_root);
1444 root->log_root = log_root;
1445 root->log_transid = 0;
257c62e1 1446 root->last_log_commit = 0;
e02119d5
CM		 1447 return 0;
1448}
1449
35a3621b
SB		 1450static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1451 struct btrfs_key *key)
e02119d5
CM		 1452{
1453 struct btrfs_root *root;
1454 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1455 struct btrfs_path *path;
84234f3a 1456 u64 generation;
db94535d 1457 u32 blocksize;
cb517eab 1458 int ret;
0f7d52f4 1459
cb517eab
MX		 1460 path = btrfs_alloc_path();
1461 if (!path)
0f7d52f4 1462 return ERR_PTR(-ENOMEM);
cb517eab
MX		 1463
1464 root = btrfs_alloc_root(fs_info);
1465 if (!root) {
1466 ret = -ENOMEM;
1467 goto alloc_fail;
0f7d52f4
CM		 1468 }
1469
db94535d 1470 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb 1471 tree_root->sectorsize, tree_root->stripesize,
cb517eab 1472 root, fs_info, key->objectid);
0f7d52f4 1473
cb517eab
MX		 1474 ret = btrfs_find_root(tree_root, key, path,
1475 &root->root_item, &root->root_key);
0f7d52f4 1476 if (ret) {
13a8a7c8
YZ		 1477 if (ret > 0)
1478 ret = -ENOENT;
cb517eab 1479 goto find_fail;
0f7d52f4 1480 }
13a8a7c8 1481
84234f3a 1482 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1483 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1484 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1485 blocksize, generation);
cb517eab
MX		 1486 if (!root->node) {
1487 ret = -ENOMEM;
1488 goto find_fail;
1489 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1490 ret = -EIO;
1491 goto read_fail;
416bc658 1492 }
5d4f98a2 1493 root->commit_root = btrfs_root_node(root);
13a8a7c8 1494out:
cb517eab
MX		 1495 btrfs_free_path(path);
1496 return root;
1497
1498read_fail:
1499 free_extent_buffer(root->node);
1500find_fail:
1501 kfree(root);
1502alloc_fail:
1503 root = ERR_PTR(ret);
1504 goto out;
1505}
1506
1507struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1508 struct btrfs_key *location)
1509{
1510 struct btrfs_root *root;
1511
1512 root = btrfs_read_tree_root(tree_root, location);
1513 if (IS_ERR(root))
1514 return root;
1515
1516 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1517 root->ref_cows = 1;
08fe4db1
LZ		 1518 btrfs_check_and_init_root_item(&root->root_item);
1519 }
13a8a7c8 1520
5eda7b5e
CM		 1521 return root;
1522}
1523
cb517eab
MX		 1524int btrfs_init_fs_root(struct btrfs_root *root)
1525{
1526 int ret;
1527
1528 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1529 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1530 GFP_NOFS);
1531 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1532 ret = -ENOMEM;
1533 goto fail;
1534 }
1535
1536 btrfs_init_free_ino_ctl(root);
1537 mutex_init(&root->fs_commit_mutex);
1538 spin_lock_init(&root->cache_lock);
1539 init_waitqueue_head(&root->cache_wait);
1540
1541 ret = get_anon_bdev(&root->anon_dev);
1542 if (ret)
1543 goto fail;
1544 return 0;
1545fail:
1546 kfree(root->free_ino_ctl);
1547 kfree(root->free_ino_pinned);
1548 return ret;
1549}
1550
171170c1
ST		 1551static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1552 u64 root_id)
cb517eab
MX		 1553{
1554 struct btrfs_root *root;
1555
1556 spin_lock(&fs_info->fs_roots_radix_lock);
1557 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1558 (unsigned long)root_id);
1559 spin_unlock(&fs_info->fs_roots_radix_lock);
1560 return root;
1561}
1562
1563int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1564 struct btrfs_root *root)
1565{
1566 int ret;
1567
1568 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1569 if (ret)
1570 return ret;
1571
1572 spin_lock(&fs_info->fs_roots_radix_lock);
1573 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1574 (unsigned long)root->root_key.objectid,
1575 root);
1576 if (ret == 0)
1577 root->in_radix = 1;
1578 spin_unlock(&fs_info->fs_roots_radix_lock);
1579 radix_tree_preload_end();
1580
1581 return ret;
1582}
1583
c00869f1
MX		 1584struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1585 struct btrfs_key *location,
1586 bool check_ref)
5eda7b5e
CM		 1587{
1588 struct btrfs_root *root;
1589 int ret;
1590
edbd8d4e
CM		 1591 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1592 return fs_info->tree_root;
1593 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1594 return fs_info->extent_root;
8f18cf13
CM		 1595 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1596 return fs_info->chunk_root;
1597 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1598 return fs_info->dev_root;
0403e47e
YZ		 1599 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1600 return fs_info->csum_root;
bcef60f2
AJ		 1601 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1602 return fs_info->quota_root ? fs_info->quota_root :
1603 ERR_PTR(-ENOENT);
f7a81ea4
SB		 1604 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1605 return fs_info->uuid_root ? fs_info->uuid_root :
1606 ERR_PTR(-ENOENT);
4df27c4d 1607again:
cb517eab 1608 root = btrfs_lookup_fs_root(fs_info, location->objectid);
48475471 1609 if (root) {
c00869f1 1610 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
48475471 1611 return ERR_PTR(-ENOENT);
5eda7b5e 1612 return root;
48475471 1613 }
5eda7b5e 1614
cb517eab 1615 root = btrfs_read_fs_root(fs_info->tree_root, location);
5eda7b5e
CM		 1616 if (IS_ERR(root))
1617 return root;
3394e160 1618
c00869f1 1619 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
cb517eab 1620 ret = -ENOENT;
581bb050 1621 goto fail;
35a30d7c 1622 }
581bb050 1623
cb517eab 1624 ret = btrfs_init_fs_root(root);
ac08aedf
CM		 1625 if (ret)
1626 goto fail;
3394e160 1627
d68fc57b
YZ		 1628 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1629 if (ret < 0)
1630 goto fail;
1631 if (ret == 0)
1632 root->orphan_item_inserted = 1;
1633
cb517eab 1634 ret = btrfs_insert_fs_root(fs_info, root);
0f7d52f4 1635 if (ret) {
4df27c4d
YZ		 1636 if (ret == -EEXIST) {
1637 free_fs_root(root);
1638 goto again;
1639 }
1640 goto fail;
0f7d52f4 1641 }
edbd8d4e 1642 return root;
4df27c4d
YZ		 1643fail:
1644 free_fs_root(root);
1645 return ERR_PTR(ret);
edbd8d4e
CM		 1646}
1647
04160088
CM		 1648static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1649{
1650 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1651 int ret = 0;
04160088
CM		 1652 struct btrfs_device *device;
1653 struct backing_dev_info *bdi;
b7967db7 1654
1f78160c
XG		 1655 rcu_read_lock();
1656 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1657 if (!device->bdev)
1658 continue;
04160088
CM		 1659 bdi = blk_get_backing_dev_info(device->bdev);
1660 if (bdi && bdi_congested(bdi, bdi_bits)) {
1661 ret = 1;
1662 break;
1663 }
1664 }
1f78160c 1665 rcu_read_unlock();
04160088
CM		 1666 return ret;
1667}
1668
ad081f14
JA		 1669/*
1670 * If this fails, caller must call bdi_destroy() to get rid of the
1671 * bdi again.
1672 */
04160088
CM		 1673static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1674{
ad081f14
JA		 1675 int err;
1676
1677 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1678 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1679 if (err)
1680 return err;
1681
4575c9cc 1682 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1683 bdi->congested_fn = btrfs_congested_fn;
1684 bdi->congested_data = info;
1685 return 0;
1686}
1687
8b712842
CM		 1688/*
1689 * called by the kthread helper functions to finally call the bio end_io
1690 * functions. This is where read checksum verification actually happens
1691 */
1692static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1693{
ce9adaa5 1694 struct bio *bio;
8b712842
CM		 1695 struct end_io_wq *end_io_wq;
1696 struct btrfs_fs_info *fs_info;
ce9adaa5 1697 int error;
ce9adaa5 1698
8b712842
CM		 1699 end_io_wq = container_of(work, struct end_io_wq, work);
1700 bio = end_io_wq->bio;
1701 fs_info = end_io_wq->info;
ce9adaa5 1702
8b712842
CM		 1703 error = end_io_wq->error;
1704 bio->bi_private = end_io_wq->private;
1705 bio->bi_end_io = end_io_wq->end_io;
1706 kfree(end_io_wq);
8b712842 1707 bio_endio(bio, error);
44b8bd7e
CM		 1708}
1709
a74a4b97
CM		 1710static int cleaner_kthread(void *arg)
1711{
1712 struct btrfs_root *root = arg;
d0278245 1713 int again;
a74a4b97
CM		 1714
1715 do {
d0278245 1716 again = 0;
a74a4b97 1717
d0278245 1718 /* Make the cleaner go to sleep early. */
babbf170 1719 if (btrfs_need_cleaner_sleep(root))
d0278245
MX		 1720 goto sleep;
1721
1722 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1723 goto sleep;
1724
dc7f370c
MX		 1725 /*
1726 * Avoid the problem that we change the status of the fs
1727 * during the above check and trylock.
1728 */
babbf170 1729 if (btrfs_need_cleaner_sleep(root)) {
dc7f370c
MX		 1730 mutex_unlock(&root->fs_info->cleaner_mutex);
1731 goto sleep;
76dda93c 1732 }
a74a4b97 1733
d0278245
MX		 1734 btrfs_run_delayed_iputs(root);
1735 again = btrfs_clean_one_deleted_snapshot(root);
1736 mutex_unlock(&root->fs_info->cleaner_mutex);
1737
1738 /*
05323cd1
MX		 1739 * The defragger has dealt with the R/O remount and umount,
1740 * needn't do anything special here.
d0278245
MX		 1741 */
1742 btrfs_run_defrag_inodes(root->fs_info);
1743sleep:
9d1a2a3a 1744 if (!try_to_freeze() && !again) {
a74a4b97 1745 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1746 if (!kthread_should_stop())
1747 schedule();
a74a4b97
CM		 1748 __set_current_state(TASK_RUNNING);
1749 }
1750 } while (!kthread_should_stop());
1751 return 0;
1752}
1753
1754static int transaction_kthread(void *arg)
1755{
1756 struct btrfs_root *root = arg;
1757 struct btrfs_trans_handle *trans;
1758 struct btrfs_transaction *cur;
8929ecfa 1759 u64 transid;
a74a4b97
CM		 1760 unsigned long now;
1761 unsigned long delay;
914b2007 1762 bool cannot_commit;
a74a4b97
CM		 1763
1764 do {
914b2007 1765 cannot_commit = false;
8b87dc17 1766 delay = HZ * root->fs_info->commit_interval;
a74a4b97
CM		 1767 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1768
a4abeea4 1769 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1770 cur = root->fs_info->running_transaction;
1771 if (!cur) {
a4abeea4 1772 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1773 goto sleep;
1774 }
31153d81 1775
a74a4b97 1776 now = get_seconds();
4a9d8bde 1777 if (cur->state < TRANS_STATE_BLOCKED &&
8b87dc17
DS		 1778 (now < cur->start_time ||
1779 now - cur->start_time < root->fs_info->commit_interval)) {
a4abeea4 1780 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1781 delay = HZ * 5;
1782 goto sleep;
1783 }
8929ecfa 1784 transid = cur->transid;
a4abeea4 1785 spin_unlock(&root->fs_info->trans_lock);
56bec294 1786
79787eaa 1787 /* If the file system is aborted, this will always fail. */
354aa0fb 1788 trans = btrfs_attach_transaction(root);
914b2007 1789 if (IS_ERR(trans)) {
354aa0fb
MX		 1790 if (PTR_ERR(trans) != -ENOENT)
1791 cannot_commit = true;
79787eaa 1792 goto sleep;
914b2007 1793 }
8929ecfa 1794 if (transid == trans->transid) {
79787eaa 1795 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1796 } else {
1797 btrfs_end_transaction(trans, root);
1798 }
a74a4b97
CM		 1799sleep:
1800 wake_up_process(root->fs_info->cleaner_kthread);
1801 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1802
4e121c06
JB		 1803 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1804 &root->fs_info->fs_state)))
1805 btrfs_cleanup_transaction(root);
a0acae0e 1806 if (!try_to_freeze()) {
a74a4b97 1807 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1808 if (!kthread_should_stop() &&
914b2007
JK		 1809 (!btrfs_transaction_blocked(root->fs_info) ||
1810 cannot_commit))
8929ecfa 1811 schedule_timeout(delay);
a74a4b97
CM		 1812 __set_current_state(TASK_RUNNING);
1813 }
1814 } while (!kthread_should_stop());
1815 return 0;
1816}
1817
af31f5e5
CM		 1818/*
1819 * this will find the highest generation in the array of
1820 * root backups. The index of the highest array is returned,
1821 * or -1 if we can't find anything.
1822 *
1823 * We check to make sure the array is valid by comparing the
1824 * generation of the latest root in the array with the generation
1825 * in the super block. If they don't match we pitch it.
1826 */
1827static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1828{
1829 u64 cur;
1830 int newest_index = -1;
1831 struct btrfs_root_backup *root_backup;
1832 int i;
1833
1834 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1835 root_backup = info->super_copy->super_roots + i;
1836 cur = btrfs_backup_tree_root_gen(root_backup);
1837 if (cur == newest_gen)
1838 newest_index = i;
1839 }
1840
1841 /* check to see if we actually wrapped around */
1842 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1843 root_backup = info->super_copy->super_roots;
1844 cur = btrfs_backup_tree_root_gen(root_backup);
1845 if (cur == newest_gen)
1846 newest_index = 0;
1847 }
1848 return newest_index;
1849}
1850
1851
1852/*
1853 * find the oldest backup so we know where to store new entries
1854 * in the backup array. This will set the backup_root_index
1855 * field in the fs_info struct
1856 */
1857static void find_oldest_super_backup(struct btrfs_fs_info *info,
1858 u64 newest_gen)
1859{
1860 int newest_index = -1;
1861
1862 newest_index = find_newest_super_backup(info, newest_gen);
1863 /* if there was garbage in there, just move along */
1864 if (newest_index == -1) {
1865 info->backup_root_index = 0;
1866 } else {
1867 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1868 }
1869}
1870
1871/*
1872 * copy all the root pointers into the super backup array.
1873 * this will bump the backup pointer by one when it is
1874 * done
1875 */
1876static void backup_super_roots(struct btrfs_fs_info *info)
1877{
1878 int next_backup;
1879 struct btrfs_root_backup *root_backup;
1880 int last_backup;
1881
1882 next_backup = info->backup_root_index;
1883 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1884 BTRFS_NUM_BACKUP_ROOTS;
1885
1886 /*
1887 * just overwrite the last backup if we're at the same generation
1888 * this happens only at umount
1889 */
1890 root_backup = info->super_for_commit->super_roots + last_backup;
1891 if (btrfs_backup_tree_root_gen(root_backup) ==
1892 btrfs_header_generation(info->tree_root->node))
1893 next_backup = last_backup;
1894
1895 root_backup = info->super_for_commit->super_roots + next_backup;
1896
1897 /*
1898 * make sure all of our padding and empty slots get zero filled
1899 * regardless of which ones we use today
1900 */
1901 memset(root_backup, 0, sizeof(*root_backup));
1902
1903 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1904
1905 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1906 btrfs_set_backup_tree_root_gen(root_backup,
1907 btrfs_header_generation(info->tree_root->node));
1908
1909 btrfs_set_backup_tree_root_level(root_backup,
1910 btrfs_header_level(info->tree_root->node));
1911
1912 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1913 btrfs_set_backup_chunk_root_gen(root_backup,
1914 btrfs_header_generation(info->chunk_root->node));
1915 btrfs_set_backup_chunk_root_level(root_backup,
1916 btrfs_header_level(info->chunk_root->node));
1917
1918 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1919 btrfs_set_backup_extent_root_gen(root_backup,
1920 btrfs_header_generation(info->extent_root->node));
1921 btrfs_set_backup_extent_root_level(root_backup,
1922 btrfs_header_level(info->extent_root->node));
1923
7c7e82a7
CM		 1924 /*
1925 * we might commit during log recovery, which happens before we set
1926 * the fs_root. Make sure it is valid before we fill it in.
1927 */
1928 if (info->fs_root && info->fs_root->node) {
1929 btrfs_set_backup_fs_root(root_backup,
1930 info->fs_root->node->start);
1931 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1932 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1933 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1934 btrfs_header_level(info->fs_root->node));
7c7e82a7 1935 }
af31f5e5
CM		 1936
1937 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1938 btrfs_set_backup_dev_root_gen(root_backup,
1939 btrfs_header_generation(info->dev_root->node));
1940 btrfs_set_backup_dev_root_level(root_backup,
1941 btrfs_header_level(info->dev_root->node));
1942
1943 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1944 btrfs_set_backup_csum_root_gen(root_backup,
1945 btrfs_header_generation(info->csum_root->node));
1946 btrfs_set_backup_csum_root_level(root_backup,
1947 btrfs_header_level(info->csum_root->node));
1948
1949 btrfs_set_backup_total_bytes(root_backup,
1950 btrfs_super_total_bytes(info->super_copy));
1951 btrfs_set_backup_bytes_used(root_backup,
1952 btrfs_super_bytes_used(info->super_copy));
1953 btrfs_set_backup_num_devices(root_backup,
1954 btrfs_super_num_devices(info->super_copy));
1955
1956 /*
1957 * if we don't copy this out to the super_copy, it won't get remembered
1958 * for the next commit
1959 */
1960 memcpy(&info->super_copy->super_roots,
1961 &info->super_for_commit->super_roots,
1962 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1963}
1964
1965/*
1966 * this copies info out of the root backup array and back into
1967 * the in-memory super block. It is meant to help iterate through
1968 * the array, so you send it the number of backups you've already
1969 * tried and the last backup index you used.
1970 *
1971 * this returns -1 when it has tried all the backups
1972 */
1973static noinline int next_root_backup(struct btrfs_fs_info *info,
1974 struct btrfs_super_block *super,
1975 int *num_backups_tried, int *backup_index)
1976{
1977 struct btrfs_root_backup *root_backup;
1978 int newest = *backup_index;
1979
1980 if (*num_backups_tried == 0) {
1981 u64 gen = btrfs_super_generation(super);
1982
1983 newest = find_newest_super_backup(info, gen);
1984 if (newest == -1)
1985 return -1;
1986
1987 *backup_index = newest;
1988 *num_backups_tried = 1;
1989 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1990 /* we've tried all the backups, all done */
1991 return -1;
1992 } else {
1993 /* jump to the next oldest backup */
1994 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1995 BTRFS_NUM_BACKUP_ROOTS;
1996 *backup_index = newest;
1997 *num_backups_tried += 1;
1998 }
1999 root_backup = super->super_roots + newest;
2000
2001 btrfs_set_super_generation(super,
2002 btrfs_backup_tree_root_gen(root_backup));
2003 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2004 btrfs_set_super_root_level(super,
2005 btrfs_backup_tree_root_level(root_backup));
2006 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2007
2008 /*
2009 * fixme: the total bytes and num_devices need to match or we should
2010 * need a fsck
2011 */
2012 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2013 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2014 return 0;
2015}
2016
7abadb64
LB		 2017/* helper to cleanup workers */
2018static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2019{
2020 btrfs_stop_workers(&fs_info->generic_worker);
2021 btrfs_stop_workers(&fs_info->fixup_workers);
2022 btrfs_stop_workers(&fs_info->delalloc_workers);
2023 btrfs_stop_workers(&fs_info->workers);
2024 btrfs_stop_workers(&fs_info->endio_workers);
2025 btrfs_stop_workers(&fs_info->endio_meta_workers);
2026 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2027 btrfs_stop_workers(&fs_info->rmw_workers);
2028 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2029 btrfs_stop_workers(&fs_info->endio_write_workers);
2030 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2031 btrfs_stop_workers(&fs_info->submit_workers);
2032 btrfs_stop_workers(&fs_info->delayed_workers);
2033 btrfs_stop_workers(&fs_info->caching_workers);
2034 btrfs_stop_workers(&fs_info->readahead_workers);
2035 btrfs_stop_workers(&fs_info->flush_workers);
2f232036 2036 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
7abadb64
LB		 2037}
2038
af31f5e5
CM		 2039/* helper to cleanup tree roots */
2040static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2041{
2042 free_extent_buffer(info->tree_root->node);
2043 free_extent_buffer(info->tree_root->commit_root);
af31f5e5
CM		 2044 info->tree_root->node = NULL;
2045 info->tree_root->commit_root = NULL;
655b09fe
JB		 2046
2047 if (info->dev_root) {
2048 free_extent_buffer(info->dev_root->node);
2049 free_extent_buffer(info->dev_root->commit_root);
2050 info->dev_root->node = NULL;
2051 info->dev_root->commit_root = NULL;
2052 }
2053 if (info->extent_root) {
2054 free_extent_buffer(info->extent_root->node);
2055 free_extent_buffer(info->extent_root->commit_root);
2056 info->extent_root->node = NULL;
2057 info->extent_root->commit_root = NULL;
2058 }
2059 if (info->csum_root) {
2060 free_extent_buffer(info->csum_root->node);
2061 free_extent_buffer(info->csum_root->commit_root);
2062 info->csum_root->node = NULL;
2063 info->csum_root->commit_root = NULL;
2064 }
bcef60f2 2065 if (info->quota_root) {
655b09fe
JB		 2066 free_extent_buffer(info->quota_root->node);
2067 free_extent_buffer(info->quota_root->commit_root);
bcef60f2
AJ		 2068 info->quota_root->node = NULL;
2069 info->quota_root->commit_root = NULL;
2070 }
f7a81ea4
SB		 2071 if (info->uuid_root) {
2072 free_extent_buffer(info->uuid_root->node);
2073 free_extent_buffer(info->uuid_root->commit_root);
2074 info->uuid_root->node = NULL;
2075 info->uuid_root->commit_root = NULL;
2076 }
af31f5e5
CM		 2077 if (chunk_root) {
2078 free_extent_buffer(info->chunk_root->node);
2079 free_extent_buffer(info->chunk_root->commit_root);
2080 info->chunk_root->node = NULL;
2081 info->chunk_root->commit_root = NULL;
2082 }
2083}
2084
171f6537
JB		 2085static void del_fs_roots(struct btrfs_fs_info *fs_info)
2086{
2087 int ret;
2088 struct btrfs_root *gang[8];
2089 int i;
2090
2091 while (!list_empty(&fs_info->dead_roots)) {
2092 gang[0] = list_entry(fs_info->dead_roots.next,
2093 struct btrfs_root, root_list);
2094 list_del(&gang[0]->root_list);
2095
2096 if (gang[0]->in_radix) {
cb517eab 2097 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
171f6537
JB		 2098 } else {
2099 free_extent_buffer(gang[0]->node);
2100 free_extent_buffer(gang[0]->commit_root);
b0feb9d9 2101 btrfs_put_fs_root(gang[0]);
171f6537
JB		 2102 }
2103 }
2104
2105 while (1) {
2106 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2107 (void **)gang, 0,
2108 ARRAY_SIZE(gang));
2109 if (!ret)
2110 break;
2111 for (i = 0; i < ret; i++)
cb517eab 2112 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
171f6537
JB		 2113 }
2114}
af31f5e5 2115
ad2b2c80
AV		 2116int open_ctree(struct super_block *sb,
2117 struct btrfs_fs_devices *fs_devices,
2118 char *options)
2e635a27 2119{
db94535d
CM		 2120 u32 sectorsize;
2121 u32 nodesize;
2122 u32 leafsize;
2123 u32 blocksize;
87ee04eb 2124 u32 stripesize;
84234f3a 2125 u64 generation;
f2b636e8 2126 u64 features;
3de4586c 2127 struct btrfs_key location;
a061fc8d 2128 struct buffer_head *bh;
4d34b278 2129 struct btrfs_super_block *disk_super;
815745cf 2130 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2131 struct btrfs_root *tree_root;
4d34b278
ID		 2132 struct btrfs_root *extent_root;
2133 struct btrfs_root *csum_root;
2134 struct btrfs_root *chunk_root;
2135 struct btrfs_root *dev_root;
bcef60f2 2136 struct btrfs_root *quota_root;
f7a81ea4 2137 struct btrfs_root *uuid_root;
e02119d5 2138 struct btrfs_root *log_tree_root;
eb60ceac 2139 int ret;
e58ca020 2140 int err = -EINVAL;
af31f5e5
CM		 2141 int num_backups_tried = 0;
2142 int backup_index = 0;
70f80175
SB		 2143 bool create_uuid_tree;
2144 bool check_uuid_tree;
4543df7e 2145
f84a8bd6 2146 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e 2147 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
cb517eab 2148 if (!tree_root || !chunk_root) {
39279cc3
CM		 2149 err = -ENOMEM;
2150 goto fail;
2151 }
76dda93c
YZ		 2152
2153 ret = init_srcu_struct(&fs_info->subvol_srcu);
2154 if (ret) {
2155 err = ret;
2156 goto fail;
2157 }
2158
2159 ret = setup_bdi(fs_info, &fs_info->bdi);
2160 if (ret) {
2161 err = ret;
2162 goto fail_srcu;
2163 }
2164
e2d84521
MX		 2165 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2166 if (ret) {
2167 err = ret;
2168 goto fail_bdi;
2169 }
2170 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2171 (1 + ilog2(nr_cpu_ids));
2172
963d678b
MX		 2173 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2174 if (ret) {
2175 err = ret;
2176 goto fail_dirty_metadata_bytes;
2177 }
2178
76dda93c
YZ		 2179 fs_info->btree_inode = new_inode(sb);
2180 if (!fs_info->btree_inode) {
2181 err = -ENOMEM;
963d678b 2182 goto fail_delalloc_bytes;
76dda93c
YZ		 2183 }
2184
a6591715 2185 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2186
76dda93c 2187 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 2188 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2189 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2190 INIT_LIST_HEAD(&fs_info->delayed_iputs);
eb73c1b7 2191 INIT_LIST_HEAD(&fs_info->delalloc_roots);
11833d66 2192 INIT_LIST_HEAD(&fs_info->caching_block_groups);
eb73c1b7 2193 spin_lock_init(&fs_info->delalloc_root_lock);
a4abeea4 2194 spin_lock_init(&fs_info->trans_lock);
76dda93c 2195 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2196 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2197 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2198 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2199 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2200 spin_lock_init(&fs_info->super_lock);
f29021b2 2201 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2202 mutex_init(&fs_info->reloc_mutex);
de98ced9 2203 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2204
58176a96 2205 init_completion(&fs_info->kobj_unregister);
0b86a832 2206 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2207 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2208 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2209 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2210 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2211 BTRFS_BLOCK_RSV_GLOBAL);
2212 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2213 BTRFS_BLOCK_RSV_DELALLOC);
2214 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2215 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2216 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2217 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2218 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2219 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2220 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2221 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2222 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2223 atomic_set(&fs_info->defrag_running, 0);
fc36ed7e 2224 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2225 fs_info->sb = sb;
6f568d35 2226 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2227 fs_info->metadata_ratio = 0;
4cb5300b 2228 fs_info->defrag_inodes = RB_ROOT;
2bf64758 2229 fs_info->free_chunk_space = 0;
f29021b2 2230 fs_info->tree_mod_log = RB_ROOT;
8b87dc17 2231 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
c8b97818 2232
90519d66
AJ		 2233 /* readahead state */
2234 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2235 spin_lock_init(&fs_info->reada_lock);
c8b97818 2236
b34b086c
CM		 2237 fs_info->thread_pool_size = min_t(unsigned long,
2238 num_online_cpus() + 2, 8);
0afbaf8c 2239
199c2a9c
MX		 2240 INIT_LIST_HEAD(&fs_info->ordered_roots);
2241 spin_lock_init(&fs_info->ordered_root_lock);
16cdcec7
MX		 2242 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2243 GFP_NOFS);
2244 if (!fs_info->delayed_root) {
2245 err = -ENOMEM;
2246 goto fail_iput;
2247 }
2248 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2249
a2de733c
AJ		 2250 mutex_init(&fs_info->scrub_lock);
2251 atomic_set(&fs_info->scrubs_running, 0);
2252 atomic_set(&fs_info->scrub_pause_req, 0);
2253 atomic_set(&fs_info->scrubs_paused, 0);
2254 atomic_set(&fs_info->scrub_cancel_req, 0);
2255 init_waitqueue_head(&fs_info->scrub_pause_wait);
2256 init_rwsem(&fs_info->scrub_super_lock);
2257 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2258#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2259 fs_info->check_integrity_print_mask = 0;
2260#endif
a2de733c 2261
c9e9f97b
ID		 2262 spin_lock_init(&fs_info->balance_lock);
2263 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2264 atomic_set(&fs_info->balance_running, 0);
2265 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2266 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2267 fs_info->balance_ctl = NULL;
837d5b6e 2268 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2269
a061fc8d
CM		 2270 sb->s_blocksize = 4096;
2271 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2272 sb->s_bdi = &fs_info->bdi;
a061fc8d 2273
76dda93c 2274 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2275 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2276 /*
2277 * we set the i_size on the btree inode to the max possible int.
2278 * the real end of the address space is determined by all of
2279 * the devices in the system
2280 */
2281 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2282 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2283 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2284
5d4f98a2 2285 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2286 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2287 fs_info->btree_inode->i_mapping);
0b32f4bb 2288 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2289 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2290
2291 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2292
76dda93c
YZ		 2293 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2294 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2295 sizeof(struct btrfs_key));
72ac3c0d
JB		 2296 set_bit(BTRFS_INODE_DUMMY,
2297 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
c65ddb52 2298 insert_inode_hash(fs_info->btree_inode);
76dda93c 2299
0f9dd46c 2300 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2301 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2302 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2303
11833d66 2304 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2305 fs_info->btree_inode->i_mapping);
11833d66 2306 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2307 fs_info->btree_inode->i_mapping);
11833d66 2308 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2309 fs_info->do_barriers = 1;
e18e4809 2310
39279cc3 2311
5a3f23d5 2312 mutex_init(&fs_info->ordered_operations_mutex);
9ffba8cd 2313 mutex_init(&fs_info->ordered_extent_flush_mutex);
e02119d5 2314 mutex_init(&fs_info->tree_log_mutex);
925baedd 2315 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2316 mutex_init(&fs_info->transaction_kthread_mutex);
2317 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2318 mutex_init(&fs_info->volume_mutex);
276e680d 2319 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2320 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2321 init_rwsem(&fs_info->subvol_sem);
803b2f54 2322 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
e922e087
SB		 2323 fs_info->dev_replace.lock_owner = 0;
2324 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2325 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2326 mutex_init(&fs_info->dev_replace.lock_management_lock);
2327 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2328
416ac51d 2329 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2330 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2331 fs_info->qgroup_tree = RB_ROOT;
2332 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2333 fs_info->qgroup_seq = 1;
2334 fs_info->quota_enabled = 0;
2335 fs_info->pending_quota_state = 0;
1e8f9158 2336 fs_info->qgroup_ulist = NULL;
2f232036 2337 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2338
fa9c0d79
CM		 2339 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2340 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2341
e6dcd2dc 2342 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2343 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2344 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2345 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2346
53b381b3
DW		 2347 ret = btrfs_alloc_stripe_hash_table(fs_info);
2348 if (ret) {
83c8266a 2349 err = ret;
53b381b3
DW		 2350 goto fail_alloc;
2351 }
2352
0b86a832 2353 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2354 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2355
3c4bb26b 2356 invalidate_bdev(fs_devices->latest_bdev);
1104a885
DS		 2357
2358 /*
2359 * Read super block and check the signature bytes only
2360 */
a512bbf8 2361 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2362 if (!bh) {
2363 err = -EINVAL;
16cdcec7 2364 goto fail_alloc;
20b45077 2365 }
39279cc3 2366
1104a885
DS		 2367 /*
2368 * We want to check superblock checksum, the type is stored inside.
2369 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2370 */
2371 if (btrfs_check_super_csum(bh->b_data)) {
2372 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2373 err = -EINVAL;
2374 goto fail_alloc;
2375 }
2376
2377 /*
2378 * super_copy is zeroed at allocation time and we never touch the
2379 * following bytes up to INFO_SIZE, the checksum is calculated from
2380 * the whole block of INFO_SIZE
2381 */
6c41761f
DS		 2382 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2383 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2384 sizeof(*fs_info->super_for_commit));
a061fc8d 2385 brelse(bh);
5f39d397 2386
6c41761f 2387 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2388
1104a885
DS		 2389 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2390 if (ret) {
2391 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2392 err = -EINVAL;
2393 goto fail_alloc;
2394 }
2395
6c41761f 2396 disk_super = fs_info->super_copy;
0f7d52f4 2397 if (!btrfs_super_root(disk_super))
16cdcec7 2398 goto fail_alloc;
0f7d52f4 2399
acce952b 2400 /* check FS state, whether FS is broken. */
87533c47
MX		 2401 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2402 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2403
af31f5e5
CM		 2404 /*
2405 * run through our array of backup supers and setup
2406 * our ring pointer to the oldest one
2407 */
2408 generation = btrfs_super_generation(disk_super);
2409 find_oldest_super_backup(fs_info, generation);
2410
75e7cb7f
LB		 2411 /*
2412 * In the long term, we'll store the compression type in the super
2413 * block, and it'll be used for per file compression control.
2414 */
2415 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2416
2b82032c
YZ		 2417 ret = btrfs_parse_options(tree_root, options);
2418 if (ret) {
2419 err = ret;
16cdcec7 2420 goto fail_alloc;
2b82032c 2421 }
dfe25020 2422
f2b636e8
JB		 2423 features = btrfs_super_incompat_flags(disk_super) &
2424 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2425 if (features) {
2426 printk(KERN_ERR "BTRFS: couldn't mount because of "
2427 "unsupported optional features (%Lx).\n",
c1c9ff7c 2428 features);
f2b636e8 2429 err = -EINVAL;
16cdcec7 2430 goto fail_alloc;
f2b636e8
JB		 2431 }
2432
727011e0
CM		 2433 if (btrfs_super_leafsize(disk_super) !=
2434 btrfs_super_nodesize(disk_super)) {
2435 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2436 "blocksizes don't match. node %d leaf %d\n",
2437 btrfs_super_nodesize(disk_super),
2438 btrfs_super_leafsize(disk_super));
2439 err = -EINVAL;
2440 goto fail_alloc;
2441 }
2442 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2443 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2444 "blocksize (%d) was too large\n",
2445 btrfs_super_leafsize(disk_super));
2446 err = -EINVAL;
2447 goto fail_alloc;
2448 }
2449
5d4f98a2 2450 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2451 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2452 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2453 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2454
3173a18f
JB		 2455 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2456 printk(KERN_ERR "btrfs: has skinny extents\n");
2457
727011e0
CM		 2458 /*
2459 * flag our filesystem as having big metadata blocks if
2460 * they are bigger than the page size
2461 */
2462 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2463 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2464 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2465 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2466 }
2467
bc3f116f
CM		 2468 nodesize = btrfs_super_nodesize(disk_super);
2469 leafsize = btrfs_super_leafsize(disk_super);
2470 sectorsize = btrfs_super_sectorsize(disk_super);
2471 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2472 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2473 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2474
2475 /*
2476 * mixed block groups end up with duplicate but slightly offset
2477 * extent buffers for the same range. It leads to corruptions
2478 */
2479 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2480 (sectorsize != leafsize)) {
2481 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2482 "are not allowed for mixed block groups on %s\n",
2483 sb->s_id);
2484 goto fail_alloc;
2485 }
2486
ceda0864
MX		 2487 /*
2488 * Needn't use the lock because there is no other task which will
2489 * update the flag.
2490 */
a6fa6fae 2491 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2492
f2b636e8
JB		 2493 features = btrfs_super_compat_ro_flags(disk_super) &
2494 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2495 if (!(sb->s_flags & MS_RDONLY) && features) {
2496 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2497 "unsupported option features (%Lx).\n",
c1c9ff7c 2498 features);
f2b636e8 2499 err = -EINVAL;
16cdcec7 2500 goto fail_alloc;
f2b636e8 2501 }
61d92c32
CM		 2502
2503 btrfs_init_workers(&fs_info->generic_worker,
2504 "genwork", 1, NULL);
2505
5443be45 2506 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 2507 fs_info->thread_pool_size,
2508 &fs_info->generic_worker);
c8b97818 2509
771ed689 2510 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
45d5fd14 2511 fs_info->thread_pool_size, NULL);
771ed689 2512
8ccf6f19 2513 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
45d5fd14 2514 fs_info->thread_pool_size, NULL);
8ccf6f19 2515
5443be45 2516 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 2517 min_t(u64, fs_devices->num_devices,
45d5fd14 2518 fs_info->thread_pool_size), NULL);
61b49440 2519
bab39bf9 2520 btrfs_init_workers(&fs_info->caching_workers, "cache",
45d5fd14 2521 fs_info->thread_pool_size, NULL);
bab39bf9 2522
61b49440
CM		 2523 /* a higher idle thresh on the submit workers makes it much more
2524 * likely that bios will be send down in a sane order to the
2525 * devices
2526 */
2527 fs_info->submit_workers.idle_thresh = 64;
53863232 2528
771ed689 2529 fs_info->workers.idle_thresh = 16;
4a69a410 2530 fs_info->workers.ordered = 1;
61b49440 2531
771ed689
CM		 2532 fs_info->delalloc_workers.idle_thresh = 2;
2533 fs_info->delalloc_workers.ordered = 1;
2534
61d92c32
CM		 2535 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2536 &fs_info->generic_worker);
5443be45 2537 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2538 fs_info->thread_pool_size,
2539 &fs_info->generic_worker);
d20f7043 2540 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2541 fs_info->thread_pool_size,
2542 &fs_info->generic_worker);
cad321ad 2543 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2544 "endio-meta-write", fs_info->thread_pool_size,
2545 &fs_info->generic_worker);
53b381b3
DW		 2546 btrfs_init_workers(&fs_info->endio_raid56_workers,
2547 "endio-raid56", fs_info->thread_pool_size,
2548 &fs_info->generic_worker);
2549 btrfs_init_workers(&fs_info->rmw_workers,
2550 "rmw", fs_info->thread_pool_size,
2551 &fs_info->generic_worker);
5443be45 2552 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2553 fs_info->thread_pool_size,
2554 &fs_info->generic_worker);
0cb59c99
JB		 2555 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2556 1, &fs_info->generic_worker);
16cdcec7
MX		 2557 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2558 fs_info->thread_pool_size,
2559 &fs_info->generic_worker);
90519d66
AJ		 2560 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2561 fs_info->thread_pool_size,
2562 &fs_info->generic_worker);
2f232036
JS		 2563 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2564 &fs_info->generic_worker);
61b49440
CM		 2565
2566 /*
2567 * endios are largely parallel and should have a very
2568 * low idle thresh
2569 */
2570 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2571 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2572 fs_info->endio_raid56_workers.idle_thresh = 4;
2573 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2574
9042846b
CM		 2575 fs_info->endio_write_workers.idle_thresh = 2;
2576 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2577 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2578
0dc3b84a
JB		 2579 /*
2580 * btrfs_start_workers can really only fail because of ENOMEM so just
2581 * return -ENOMEM if any of these fail.
2582 */
2583 ret = btrfs_start_workers(&fs_info->workers);
2584 ret |= btrfs_start_workers(&fs_info->generic_worker);
2585 ret |= btrfs_start_workers(&fs_info->submit_workers);
2586 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2587 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2588 ret |= btrfs_start_workers(&fs_info->endio_workers);
2589 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2590 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2591 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2592 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2593 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2594 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2595 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2596 ret |= btrfs_start_workers(&fs_info->caching_workers);
2597 ret |= btrfs_start_workers(&fs_info->readahead_workers);
8ccf6f19 2598 ret |= btrfs_start_workers(&fs_info->flush_workers);
2f232036 2599 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
0dc3b84a 2600 if (ret) {
fed425c7 2601 err = -ENOMEM;
0dc3b84a
JB		 2602 goto fail_sb_buffer;
2603 }
4543df7e 2604
4575c9cc 2605 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2606 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2607 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2608
db94535d
CM		 2609 tree_root->nodesize = nodesize;
2610 tree_root->leafsize = leafsize;
2611 tree_root->sectorsize = sectorsize;
87ee04eb 2612 tree_root->stripesize = stripesize;
a061fc8d
CM		 2613
2614 sb->s_blocksize = sectorsize;
2615 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2616
3cae210f 2617 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
d397712b 2618 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2619 goto fail_sb_buffer;
2620 }
19c00ddc 2621
8d082fb7
LB		 2622 if (sectorsize != PAGE_SIZE) {
2623 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2624 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2625 goto fail_sb_buffer;
2626 }
2627
925baedd 2628 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2629 ret = btrfs_read_sys_array(tree_root);
925baedd 2630 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2631 if (ret) {
d397712b
CM		 2632 printk(KERN_WARNING "btrfs: failed to read the system "
2633 "array on %s\n", sb->s_id);
5d4f98a2 2634 goto fail_sb_buffer;
84eed90f 2635 }
0b86a832
CM		 2636
2637 blocksize = btrfs_level_size(tree_root,
2638 btrfs_super_chunk_root_level(disk_super));
84234f3a 2639 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2640
2641 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2642 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2643
2644 chunk_root->node = read_tree_block(chunk_root,
2645 btrfs_super_chunk_root(disk_super),
84234f3a 2646 blocksize, generation);
416bc658
JB		 2647 if (!chunk_root->node ||
2648 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
83121942
DW		 2649 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2650 sb->s_id);
af31f5e5 2651 goto fail_tree_roots;
83121942 2652 }
5d4f98a2
YZ		 2653 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2654 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2655
e17cade2 2656 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
b308bc2f 2657 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
e17cade2 2658
0b86a832 2659 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2660 if (ret) {
d397712b
CM		 2661 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2662 sb->s_id);
af31f5e5 2663 goto fail_tree_roots;
2b82032c 2664 }
0b86a832 2665
8dabb742
SB		 2666 /*
2667 * keep the device that is marked to be the target device for the
2668 * dev_replace procedure
2669 */
2670 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2671
a6b0d5c8
CM		 2672 if (!fs_devices->latest_bdev) {
2673 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2674 sb->s_id);
2675 goto fail_tree_roots;
2676 }
2677
af31f5e5 2678retry_root_backup:
db94535d
CM		 2679 blocksize = btrfs_level_size(tree_root,
2680 btrfs_super_root_level(disk_super));
84234f3a 2681 generation = btrfs_super_generation(disk_super);
0b86a832 2682
e20d96d6 2683 tree_root->node = read_tree_block(tree_root,
db94535d 2684 btrfs_super_root(disk_super),
84234f3a 2685 blocksize, generation);
af31f5e5
CM		 2686 if (!tree_root->node ||
2687 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
83121942
DW		 2688 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2689 sb->s_id);
af31f5e5
CM		 2690
2691 goto recovery_tree_root;
83121942 2692 }
af31f5e5 2693
5d4f98a2
YZ		 2694 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2695 tree_root->commit_root = btrfs_root_node(tree_root);
69e9c6c6 2696 btrfs_set_root_refs(&tree_root->root_item, 1);
db94535d 2697
cb517eab
MX		 2698 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2699 location.type = BTRFS_ROOT_ITEM_KEY;
2700 location.offset = 0;
2701
2702 extent_root = btrfs_read_tree_root(tree_root, &location);
2703 if (IS_ERR(extent_root)) {
2704 ret = PTR_ERR(extent_root);
af31f5e5 2705 goto recovery_tree_root;
cb517eab 2706 }
0b86a832 2707 extent_root->track_dirty = 1;
cb517eab 2708 fs_info->extent_root = extent_root;
0b86a832 2709
cb517eab
MX		 2710 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2711 dev_root = btrfs_read_tree_root(tree_root, &location);
2712 if (IS_ERR(dev_root)) {
2713 ret = PTR_ERR(dev_root);
af31f5e5 2714 goto recovery_tree_root;
cb517eab 2715 }
5d4f98a2 2716 dev_root->track_dirty = 1;
cb517eab
MX		 2717 fs_info->dev_root = dev_root;
2718 btrfs_init_devices_late(fs_info);
3768f368 2719
cb517eab
MX		 2720 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2721 csum_root = btrfs_read_tree_root(tree_root, &location);
2722 if (IS_ERR(csum_root)) {
2723 ret = PTR_ERR(csum_root);
af31f5e5 2724 goto recovery_tree_root;
cb517eab 2725 }
d20f7043 2726 csum_root->track_dirty = 1;
cb517eab 2727 fs_info->csum_root = csum_root;
d20f7043 2728
cb517eab
MX		 2729 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2730 quota_root = btrfs_read_tree_root(tree_root, &location);
2731 if (!IS_ERR(quota_root)) {
bcef60f2
AJ		 2732 quota_root->track_dirty = 1;
2733 fs_info->quota_enabled = 1;
2734 fs_info->pending_quota_state = 1;
cb517eab 2735 fs_info->quota_root = quota_root;
bcef60f2
AJ		 2736 }
2737
f7a81ea4
SB		 2738 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2739 uuid_root = btrfs_read_tree_root(tree_root, &location);
2740 if (IS_ERR(uuid_root)) {
2741 ret = PTR_ERR(uuid_root);
2742 if (ret != -ENOENT)
2743 goto recovery_tree_root;
2744 create_uuid_tree = true;
70f80175 2745 check_uuid_tree = false;
f7a81ea4
SB		 2746 } else {
2747 uuid_root->track_dirty = 1;
2748 fs_info->uuid_root = uuid_root;
70f80175
SB		 2749 create_uuid_tree = false;
2750 check_uuid_tree =
2751 generation != btrfs_super_uuid_tree_generation(disk_super);
f7a81ea4
SB		 2752 }
2753
8929ecfa
YZ		 2754 fs_info->generation = generation;
2755 fs_info->last_trans_committed = generation;
8929ecfa 2756
68310a5e
ID		 2757 ret = btrfs_recover_balance(fs_info);
2758 if (ret) {
2759 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2760 goto fail_block_groups;
2761 }
2762
733f4fbb
SB		 2763 ret = btrfs_init_dev_stats(fs_info);
2764 if (ret) {
2765 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2766 ret);
2767 goto fail_block_groups;
2768 }
2769
8dabb742
SB		 2770 ret = btrfs_init_dev_replace(fs_info);
2771 if (ret) {
2772 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2773 goto fail_block_groups;
2774 }
2775
2776 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2777
c59021f8 2778 ret = btrfs_init_space_info(fs_info);
2779 if (ret) {
2780 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2781 goto fail_block_groups;
2782 }
2783
1b1d1f66
JB		 2784 ret = btrfs_read_block_groups(extent_root);
2785 if (ret) {
2786 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2787 goto fail_block_groups;
2788 }
5af3e8cc
SB		 2789 fs_info->num_tolerated_disk_barrier_failures =
2790 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2791 if (fs_info->fs_devices->missing_devices >
2792 fs_info->num_tolerated_disk_barrier_failures &&
2793 !(sb->s_flags & MS_RDONLY)) {
2794 printk(KERN_WARNING
2795 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2796 goto fail_block_groups;
2797 }
9078a3e1 2798
a74a4b97
CM		 2799 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2800 "btrfs-cleaner");
57506d50 2801 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2802 goto fail_block_groups;
a74a4b97
CM		 2803
2804 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2805 tree_root,
2806 "btrfs-transaction");
57506d50 2807 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2808 goto fail_cleaner;
a74a4b97 2809
c289811c
CM		 2810 if (!btrfs_test_opt(tree_root, SSD) &&
2811 !btrfs_test_opt(tree_root, NOSSD) &&
2812 !fs_info->fs_devices->rotating) {
2813 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2814 "mode\n");
2815 btrfs_set_opt(fs_info->mount_opt, SSD);
2816 }
2817
21adbd5c
SB		 2818#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2819 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2820 ret = btrfsic_mount(tree_root, fs_devices,
2821 btrfs_test_opt(tree_root,
2822 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2823 1 : 0,
2824 fs_info->check_integrity_print_mask);
2825 if (ret)
2826 printk(KERN_WARNING "btrfs: failed to initialize"
2827 " integrity check module %s\n", sb->s_id);
2828 }
2829#endif
bcef60f2
AJ		 2830 ret = btrfs_read_qgroup_config(fs_info);
2831 if (ret)
2832 goto fail_trans_kthread;
21adbd5c 2833
acce952b 2834 /* do not make disk changes in broken FS */
68ce9682 2835 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2836 u64 bytenr = btrfs_super_log_root(disk_super);
2837
7c2ca468 2838 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2839 printk(KERN_WARNING "Btrfs log replay required "
2840 "on RO media\n");
7c2ca468 2841 err = -EIO;
bcef60f2 2842 goto fail_qgroup;
7c2ca468 2843 }
e02119d5
CM		 2844 blocksize =
2845 btrfs_level_size(tree_root,
2846 btrfs_super_log_root_level(disk_super));
d18a2c44 2847
6f07e42e 2848 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2849 if (!log_tree_root) {
2850 err = -ENOMEM;
bcef60f2 2851 goto fail_qgroup;
676e4c86 2852 }
e02119d5
CM		 2853
2854 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2855 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2856
2857 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2858 blocksize,
2859 generation + 1);
416bc658
JB		 2860 if (!log_tree_root->node ||
2861 !extent_buffer_uptodate(log_tree_root->node)) {
2862 printk(KERN_ERR "btrfs: failed to read log tree\n");
2863 free_extent_buffer(log_tree_root->node);
2864 kfree(log_tree_root);
2865 goto fail_trans_kthread;
2866 }
79787eaa 2867 /* returns with log_tree_root freed on success */
e02119d5 2868 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2869 if (ret) {
2870 btrfs_error(tree_root->fs_info, ret,
2871 "Failed to recover log tree");
2872 free_extent_buffer(log_tree_root->node);
2873 kfree(log_tree_root);
2874 goto fail_trans_kthread;
2875 }
e556ce2c
YZ		 2876
2877 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2878 ret = btrfs_commit_super(tree_root);
2879 if (ret)
2880 goto fail_trans_kthread;
e556ce2c 2881 }
e02119d5 2882 }
1a40e23b 2883
76dda93c 2884 ret = btrfs_find_orphan_roots(tree_root);
79787eaa
JM		 2885 if (ret)
2886 goto fail_trans_kthread;
76dda93c 2887
7c2ca468 2888 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2889 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2
ID		 2890 if (ret)
2891 goto fail_trans_kthread;
d68fc57b 2892
5d4f98a2 2893 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2894 if (ret < 0) {
2895 printk(KERN_WARNING
2896 "btrfs: failed to recover relocation\n");
2897 err = -EINVAL;
bcef60f2 2898 goto fail_qgroup;
d7ce5843 2899 }
7c2ca468 2900 }
1a40e23b 2901
3de4586c
CM		 2902 location.objectid = BTRFS_FS_TREE_OBJECTID;
2903 location.type = BTRFS_ROOT_ITEM_KEY;
cb517eab 2904 location.offset = 0;
3de4586c 2905
3de4586c 2906 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3140c9a3
DC		 2907 if (IS_ERR(fs_info->fs_root)) {
2908 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2909 goto fail_qgroup;
3140c9a3 2910 }
c289811c 2911
2b6ba629
ID		 2912 if (sb->s_flags & MS_RDONLY)
2913 return 0;
59641015 2914
2b6ba629
ID		 2915 down_read(&fs_info->cleanup_work_sem);
2916 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2917 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2918 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2919 close_ctree(tree_root);
2920 return ret;
2921 }
2922 up_read(&fs_info->cleanup_work_sem);
59641015 2923
2b6ba629
ID		 2924 ret = btrfs_resume_balance_async(fs_info);
2925 if (ret) {
2926 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2927 close_ctree(tree_root);
2928 return ret;
e3acc2a6
JB		 2929 }
2930
8dabb742
SB		 2931 ret = btrfs_resume_dev_replace_async(fs_info);
2932 if (ret) {
2933 pr_warn("btrfs: failed to resume dev_replace\n");
2934 close_ctree(tree_root);
2935 return ret;
2936 }
2937
b382a324
JS		 2938 btrfs_qgroup_rescan_resume(fs_info);
2939
f7a81ea4
SB		 2940 if (create_uuid_tree) {
2941 pr_info("btrfs: creating UUID tree\n");
2942 ret = btrfs_create_uuid_tree(fs_info);
2943 if (ret) {
2944 pr_warn("btrfs: failed to create the UUID tree %d\n",
2945 ret);
2946 close_ctree(tree_root);
2947 return ret;
2948 }
f420ee1e
SB		 2949 } else if (check_uuid_tree ||
2950 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
70f80175
SB		 2951 pr_info("btrfs: checking UUID tree\n");
2952 ret = btrfs_check_uuid_tree(fs_info);
2953 if (ret) {
2954 pr_warn("btrfs: failed to check the UUID tree %d\n",
2955 ret);
2956 close_ctree(tree_root);
2957 return ret;
2958 }
2959 } else {
2960 fs_info->update_uuid_tree_gen = 1;
f7a81ea4
SB		 2961 }
2962
ad2b2c80 2963 return 0;
39279cc3 2964
bcef60f2
AJ		 2965fail_qgroup:
2966 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2967fail_trans_kthread:
2968 kthread_stop(fs_info->transaction_kthread);
54067ae9 2969 btrfs_cleanup_transaction(fs_info->tree_root);
7b5ff90e 2970 del_fs_roots(fs_info);
3f157a2f 2971fail_cleaner:
a74a4b97 2972 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2973
2974 /*
2975 * make sure we're done with the btree inode before we stop our
2976 * kthreads
2977 */
2978 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2979
1b1d1f66 2980fail_block_groups:
54067ae9 2981 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2982 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2983
2984fail_tree_roots:
2985 free_root_pointers(fs_info, 1);
2b8195bb 2986 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2987
39279cc3 2988fail_sb_buffer:
7abadb64 2989 btrfs_stop_all_workers(fs_info);
16cdcec7 2990fail_alloc:
4543df7e 2991fail_iput:
586e46e2
ID		 2992 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2993
4543df7e 2994 iput(fs_info->btree_inode);
963d678b
MX		 2995fail_delalloc_bytes:
2996 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 2997fail_dirty_metadata_bytes:
2998 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 2999fail_bdi:
7e662854 3000 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 3001fail_srcu:
3002 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 3003fail:
53b381b3 3004 btrfs_free_stripe_hash_table(fs_info);
586e46e2 3005 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 3006 return err;
af31f5e5
CM		 3007
3008recovery_tree_root:
af31f5e5
CM		 3009 if (!btrfs_test_opt(tree_root, RECOVERY))
3010 goto fail_tree_roots;
3011
3012 free_root_pointers(fs_info, 0);
3013
3014 /* don't use the log in recovery mode, it won't be valid */
3015 btrfs_set_super_log_root(disk_super, 0);
3016
3017 /* we can't trust the free space cache either */
3018 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3019
3020 ret = next_root_backup(fs_info, fs_info->super_copy,
3021 &num_backups_tried, &backup_index);
3022 if (ret == -1)
3023 goto fail_block_groups;
3024 goto retry_root_backup;
eb60ceac
CM		 3025}
3026
f2984462
CM		 3027static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3028{
f2984462
CM		 3029 if (uptodate) {
3030 set_buffer_uptodate(bh);
3031 } else {
442a4f63
SB		 3032 struct btrfs_device *device = (struct btrfs_device *)
3033 bh->b_private;
3034
606686ee
JB		 3035 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3036 "I/O error on %s\n",
3037 rcu_str_deref(device->name));
1259ab75
CM		 3038 /* note, we dont' set_buffer_write_io_error because we have
3039 * our own ways of dealing with the IO errors
3040 */
f2984462 3041 clear_buffer_uptodate(bh);
442a4f63 3042 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 3043 }
3044 unlock_buffer(bh);
3045 put_bh(bh);
3046}
3047
a512bbf8
YZ		 3048struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3049{
3050 struct buffer_head *bh;
3051 struct buffer_head *latest = NULL;
3052 struct btrfs_super_block *super;
3053 int i;
3054 u64 transid = 0;
3055 u64 bytenr;
3056
3057 /* we would like to check all the supers, but that would make
3058 * a btrfs mount succeed after a mkfs from a different FS.
3059 * So, we need to add a special mount option to scan for
3060 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3061 */
3062 for (i = 0; i < 1; i++) {
3063 bytenr = btrfs_sb_offset(i);
8068a47e
AJ		 3064 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3065 i_size_read(bdev->bd_inode))
a512bbf8 3066 break;
8068a47e
AJ		 3067 bh = __bread(bdev, bytenr / 4096,
3068 BTRFS_SUPER_INFO_SIZE);
a512bbf8
YZ		 3069 if (!bh)
3070 continue;
3071
3072 super = (struct btrfs_super_block *)bh->b_data;
3073 if (btrfs_super_bytenr(super) != bytenr ||
3cae210f 3074 btrfs_super_magic(super) != BTRFS_MAGIC) {
a512bbf8
YZ		 3075 brelse(bh);
3076 continue;
3077 }
3078
3079 if (!latest || btrfs_super_generation(super) > transid) {
3080 brelse(latest);
3081 latest = bh;
3082 transid = btrfs_super_generation(super);
3083 } else {
3084 brelse(bh);
3085 }
3086 }
3087 return latest;
3088}
3089
4eedeb75
HH		 3090/*
3091 * this should be called twice, once with wait == 0 and
3092 * once with wait == 1. When wait == 0 is done, all the buffer heads
3093 * we write are pinned.
3094 *
3095 * They are released when wait == 1 is done.
3096 * max_mirrors must be the same for both runs, and it indicates how
3097 * many supers on this one device should be written.
3098 *
3099 * max_mirrors == 0 means to write them all.
3100 */
a512bbf8
YZ		 3101static int write_dev_supers(struct btrfs_device *device,
3102 struct btrfs_super_block *sb,
3103 int do_barriers, int wait, int max_mirrors)
3104{
3105 struct buffer_head *bh;
3106 int i;
3107 int ret;
3108 int errors = 0;
3109 u32 crc;
3110 u64 bytenr;
a512bbf8
YZ		 3111
3112 if (max_mirrors == 0)
3113 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3114
a512bbf8
YZ		 3115 for (i = 0; i < max_mirrors; i++) {
3116 bytenr = btrfs_sb_offset(i);
3117 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3118 break;
3119
3120 if (wait) {
3121 bh = __find_get_block(device->bdev, bytenr / 4096,
3122 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3123 if (!bh) {
3124 errors++;
3125 continue;
3126 }
a512bbf8 3127 wait_on_buffer(bh);
4eedeb75
HH		 3128 if (!buffer_uptodate(bh))
3129 errors++;
3130
3131 /* drop our reference */
3132 brelse(bh);
3133
3134 /* drop the reference from the wait == 0 run */
3135 brelse(bh);
3136 continue;
a512bbf8
YZ		 3137 } else {
3138 btrfs_set_super_bytenr(sb, bytenr);
3139
3140 crc = ~(u32)0;
b0496686 3141 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 3142 BTRFS_CSUM_SIZE, crc,
3143 BTRFS_SUPER_INFO_SIZE -
3144 BTRFS_CSUM_SIZE);
3145 btrfs_csum_final(crc, sb->csum);
3146
4eedeb75
HH		 3147 /*
3148 * one reference for us, and we leave it for the
3149 * caller
3150 */
a512bbf8
YZ		 3151 bh = __getblk(device->bdev, bytenr / 4096,
3152 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3153 if (!bh) {
3154 printk(KERN_ERR "btrfs: couldn't get super "
3155 "buffer head for bytenr %Lu\n", bytenr);
3156 errors++;
3157 continue;
3158 }
3159
a512bbf8
YZ		 3160 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3161
4eedeb75 3162 /* one reference for submit_bh */
a512bbf8 3163 get_bh(bh);
4eedeb75
HH		 3164
3165 set_buffer_uptodate(bh);
a512bbf8
YZ		 3166 lock_buffer(bh);
3167 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3168 bh->b_private = device;
a512bbf8
YZ		 3169 }
3170
387125fc
CM		 3171 /*
3172 * we fua the first super. The others we allow
3173 * to go down lazy.
3174 */
21adbd5c 3175 ret = btrfsic_submit_bh(WRITE_FUA, bh);
4eedeb75 3176 if (ret)
a512bbf8 3177 errors++;
a512bbf8
YZ		 3178 }
3179 return errors < i ? 0 : -1;
3180}
3181
387125fc
CM		 3182/*
3183 * endio for the write_dev_flush, this will wake anyone waiting
3184 * for the barrier when it is done
3185 */
3186static void btrfs_end_empty_barrier(struct bio *bio, int err)
3187{
3188 if (err) {
3189 if (err == -EOPNOTSUPP)
3190 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3191 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3192 }
3193 if (bio->bi_private)
3194 complete(bio->bi_private);
3195 bio_put(bio);
3196}
3197
3198/*
3199 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3200 * sent down. With wait == 1, it waits for the previous flush.
3201 *
3202 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3203 * capable
3204 */
3205static int write_dev_flush(struct btrfs_device *device, int wait)
3206{
3207 struct bio *bio;
3208 int ret = 0;
3209
3210 if (device->nobarriers)
3211 return 0;
3212
3213 if (wait) {
3214 bio = device->flush_bio;
3215 if (!bio)
3216 return 0;
3217
3218 wait_for_completion(&device->flush_wait);
3219
3220 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
606686ee
JB		 3221 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3222 rcu_str_deref(device->name));
387125fc 3223 device->nobarriers = 1;
5af3e8cc 3224 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3225 ret = -EIO;
5af3e8cc
SB		 3226 btrfs_dev_stat_inc_and_print(device,
3227 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3228 }
3229
3230 /* drop the reference from the wait == 0 run */
3231 bio_put(bio);
3232 device->flush_bio = NULL;
3233
3234 return ret;
3235 }
3236
3237 /*
3238 * one reference for us, and we leave it for the
3239 * caller
3240 */
9c017abc 3241 device->flush_bio = NULL;
9be3395b 3242 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
387125fc
CM		 3243 if (!bio)
3244 return -ENOMEM;
3245
3246 bio->bi_end_io = btrfs_end_empty_barrier;
3247 bio->bi_bdev = device->bdev;
3248 init_completion(&device->flush_wait);
3249 bio->bi_private = &device->flush_wait;
3250 device->flush_bio = bio;
3251
3252 bio_get(bio);
21adbd5c 3253 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3254
3255 return 0;
3256}
3257
3258/*
3259 * send an empty flush down to each device in parallel,
3260 * then wait for them
3261 */
3262static int barrier_all_devices(struct btrfs_fs_info *info)
3263{
3264 struct list_head *head;
3265 struct btrfs_device *dev;
5af3e8cc
SB		 3266 int errors_send = 0;
3267 int errors_wait = 0;
387125fc
CM		 3268 int ret;
3269
3270 /* send down all the barriers */
3271 head = &info->fs_devices->devices;
3272 list_for_each_entry_rcu(dev, head, dev_list) {
3273 if (!dev->bdev) {
5af3e8cc 3274 errors_send++;
387125fc
CM		 3275 continue;
3276 }
3277 if (!dev->in_fs_metadata || !dev->writeable)
3278 continue;
3279
3280 ret = write_dev_flush(dev, 0);
3281 if (ret)
5af3e8cc 3282 errors_send++;
387125fc
CM		 3283 }
3284
3285 /* wait for all the barriers */
3286 list_for_each_entry_rcu(dev, head, dev_list) {
3287 if (!dev->bdev) {
5af3e8cc 3288 errors_wait++;
387125fc
CM		 3289 continue;
3290 }
3291 if (!dev->in_fs_metadata || !dev->writeable)
3292 continue;
3293
3294 ret = write_dev_flush(dev, 1);
3295 if (ret)
5af3e8cc 3296 errors_wait++;
387125fc 3297 }
5af3e8cc
SB		 3298 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3299 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3300 return -EIO;
3301 return 0;
3302}
3303
5af3e8cc
SB		 3304int btrfs_calc_num_tolerated_disk_barrier_failures(
3305 struct btrfs_fs_info *fs_info)
3306{
3307 struct btrfs_ioctl_space_info space;
3308 struct btrfs_space_info *sinfo;
3309 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3310 BTRFS_BLOCK_GROUP_SYSTEM,
3311 BTRFS_BLOCK_GROUP_METADATA,
3312 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3313 int num_types = 4;
3314 int i;
3315 int c;
3316 int num_tolerated_disk_barrier_failures =
3317 (int)fs_info->fs_devices->num_devices;
3318
3319 for (i = 0; i < num_types; i++) {
3320 struct btrfs_space_info *tmp;
3321
3322 sinfo = NULL;
3323 rcu_read_lock();
3324 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3325 if (tmp->flags == types[i]) {
3326 sinfo = tmp;
3327 break;
3328 }
3329 }
3330 rcu_read_unlock();
3331
3332 if (!sinfo)
3333 continue;
3334
3335 down_read(&sinfo->groups_sem);
3336 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3337 if (!list_empty(&sinfo->block_groups[c])) {
3338 u64 flags;
3339
3340 btrfs_get_block_group_info(
3341 &sinfo->block_groups[c], &space);
3342 if (space.total_bytes == 0 ||
3343 space.used_bytes == 0)
3344 continue;
3345 flags = space.flags;
3346 /*
3347 * return
3348 * 0: if dup, single or RAID0 is configured for
3349 * any of metadata, system or data, else
3350 * 1: if RAID5 is configured, or if RAID1 or
3351 * RAID10 is configured and only two mirrors
3352 * are used, else
3353 * 2: if RAID6 is configured, else
3354 * num_mirrors - 1: if RAID1 or RAID10 is
3355 * configured and more than
3356 * 2 mirrors are used.
3357 */
3358 if (num_tolerated_disk_barrier_failures > 0 &&
3359 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3360 BTRFS_BLOCK_GROUP_RAID0)) ||
3361 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3362 == 0)))
3363 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3364 else if (num_tolerated_disk_barrier_failures > 1) {
3365 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3366 BTRFS_BLOCK_GROUP_RAID5 |
3367 BTRFS_BLOCK_GROUP_RAID10)) {
3368 num_tolerated_disk_barrier_failures = 1;
3369 } else if (flags &
15b0a89d 3370 BTRFS_BLOCK_GROUP_RAID6) {
53b381b3
DW		 3371 num_tolerated_disk_barrier_failures = 2;
3372 }
3373 }
5af3e8cc
SB		 3374 }
3375 }
3376 up_read(&sinfo->groups_sem);
3377 }
3378
3379 return num_tolerated_disk_barrier_failures;
3380}
3381
48a3b636 3382static int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3383{
e5e9a520 3384 struct list_head *head;
f2984462 3385 struct btrfs_device *dev;
a061fc8d 3386 struct btrfs_super_block *sb;
f2984462 3387 struct btrfs_dev_item *dev_item;
f2984462
CM		 3388 int ret;
3389 int do_barriers;
a236aed1
CM		 3390 int max_errors;
3391 int total_errors = 0;
a061fc8d 3392 u64 flags;
f2984462
CM		 3393
3394 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3395 backup_super_roots(root->fs_info);
f2984462 3396
6c41761f 3397 sb = root->fs_info->super_for_commit;
a061fc8d 3398 dev_item = &sb->dev_item;
e5e9a520 3399
174ba509 3400 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3401 head = &root->fs_info->fs_devices->devices;
d7306801 3402 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
387125fc 3403
5af3e8cc
SB		 3404 if (do_barriers) {
3405 ret = barrier_all_devices(root->fs_info);
3406 if (ret) {
3407 mutex_unlock(
3408 &root->fs_info->fs_devices->device_list_mutex);
3409 btrfs_error(root->fs_info, ret,
3410 "errors while submitting device barriers.");
3411 return ret;
3412 }
3413 }
387125fc 3414
1f78160c 3415 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3416 if (!dev->bdev) {
3417 total_errors++;
3418 continue;
3419 }
2b82032c 3420 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3421 continue;
3422
2b82032c 3423 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3424 btrfs_set_stack_device_type(dev_item, dev->type);
3425 btrfs_set_stack_device_id(dev_item, dev->devid);
3426 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3427 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3428 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3429 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3430 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3431 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3432 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3433
a061fc8d
CM		 3434 flags = btrfs_super_flags(sb);
3435 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3436
a512bbf8 3437 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3438 if (ret)
3439 total_errors++;
f2984462 3440 }
a236aed1 3441 if (total_errors > max_errors) {
d397712b
CM		 3442 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3443 total_errors);
a724b436 3444 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
79787eaa 3445
9d565ba4
SB		 3446 /* FUA is masked off if unsupported and can't be the reason */
3447 btrfs_error(root->fs_info, -EIO,
3448 "%d errors while writing supers", total_errors);
3449 return -EIO;
a236aed1 3450 }
f2984462 3451
a512bbf8 3452 total_errors = 0;
1f78160c 3453 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3454 if (!dev->bdev)
3455 continue;
2b82032c 3456 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3457 continue;
3458
a512bbf8
YZ		 3459 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3460 if (ret)
3461 total_errors++;
f2984462 3462 }
174ba509 3463 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3464 if (total_errors > max_errors) {
79787eaa
JM		 3465 btrfs_error(root->fs_info, -EIO,
3466 "%d errors while writing supers", total_errors);
3467 return -EIO;
a236aed1 3468 }
f2984462
CM		 3469 return 0;
3470}
3471
a512bbf8
YZ		 3472int write_ctree_super(struct btrfs_trans_handle *trans,
3473 struct btrfs_root *root, int max_mirrors)
eb60ceac 3474{
e66f709b 3475 int ret;
5f39d397 3476
a512bbf8 3477 ret = write_all_supers(root, max_mirrors);
5f39d397 3478 return ret;
cfaa7295
CM		 3479}
3480
cb517eab
MX		 3481/* Drop a fs root from the radix tree and free it. */
3482void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3483 struct btrfs_root *root)
2619ba1f 3484{
4df27c4d 3485 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3486 radix_tree_delete(&fs_info->fs_roots_radix,
3487 (unsigned long)root->root_key.objectid);
4df27c4d 3488 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3489
3490 if (btrfs_root_refs(&root->root_item) == 0)
3491 synchronize_srcu(&fs_info->subvol_srcu);
3492
d7634482 3493 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3321719e
LB		 3494 btrfs_free_log(NULL, root);
3495 btrfs_free_log_root_tree(NULL, fs_info);
3496 }
3497
581bb050
LZ		 3498 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3499 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3500 free_fs_root(root);
4df27c4d
YZ		 3501}
3502
3503static void free_fs_root(struct btrfs_root *root)
3504{
82d5902d 3505 iput(root->cache_inode);
4df27c4d 3506 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1cb048f5
FDBM		 3507 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3508 root->orphan_block_rsv = NULL;
0ee5dc67
AV		 3509 if (root->anon_dev)
3510 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 3511 free_extent_buffer(root->node);
3512 free_extent_buffer(root->commit_root);
581bb050
LZ		 3513 kfree(root->free_ino_ctl);
3514 kfree(root->free_ino_pinned);
d397712b 3515 kfree(root->name);
b0feb9d9 3516 btrfs_put_fs_root(root);
2619ba1f
CM		 3517}
3518
cb517eab
MX		 3519void btrfs_free_fs_root(struct btrfs_root *root)
3520{
3521 free_fs_root(root);
2619ba1f
CM		 3522}
3523
c146afad 3524int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3525{
c146afad
YZ		 3526 u64 root_objectid = 0;
3527 struct btrfs_root *gang[8];
3528 int i;
3768f368 3529 int ret;
e089f05c 3530
c146afad
YZ		 3531 while (1) {
3532 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3533 (void **)gang, root_objectid,
3534 ARRAY_SIZE(gang));
3535 if (!ret)
3536 break;
5d4f98a2
YZ		 3537
3538 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 3539 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 3540 int err;
3541
c146afad 3542 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3543 err = btrfs_orphan_cleanup(gang[i]);
3544 if (err)
3545 return err;
c146afad
YZ		 3546 }
3547 root_objectid++;
3548 }
3549 return 0;
3550}
a2135011 3551
c146afad
YZ		 3552int btrfs_commit_super(struct btrfs_root *root)
3553{
3554 struct btrfs_trans_handle *trans;
3555 int ret;
a74a4b97 3556
c146afad 3557 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3558 btrfs_run_delayed_iputs(root);
c146afad 3559 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3560 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3561
3562 /* wait until ongoing cleanup work done */
3563 down_write(&root->fs_info->cleanup_work_sem);
3564 up_write(&root->fs_info->cleanup_work_sem);
3565
7a7eaa40 3566 trans = btrfs_join_transaction(root);
3612b495
TI		 3567 if (IS_ERR(trans))
3568 return PTR_ERR(trans);
54aa1f4d 3569 ret = btrfs_commit_transaction(trans, root);
79787eaa
JM		 3570 if (ret)
3571 return ret;
c146afad 3572 /* run commit again to drop the original snapshot */
7a7eaa40 3573 trans = btrfs_join_transaction(root);
3612b495
TI		 3574 if (IS_ERR(trans))
3575 return PTR_ERR(trans);
79787eaa
JM		 3576 ret = btrfs_commit_transaction(trans, root);
3577 if (ret)
3578 return ret;
79154b1b 3579 ret = btrfs_write_and_wait_transaction(NULL, root);
79787eaa
JM		 3580 if (ret) {
3581 btrfs_error(root->fs_info, ret,
3582 "Failed to sync btree inode to disk.");
3583 return ret;
3584 }
d6bfde87 3585
a512bbf8 3586 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 3587 return ret;
3588}
3589
3590int close_ctree(struct btrfs_root *root)
3591{
3592 struct btrfs_fs_info *fs_info = root->fs_info;
3593 int ret;
3594
3595 fs_info->closing = 1;
3596 smp_mb();
3597
803b2f54
SB		 3598 /* wait for the uuid_scan task to finish */
3599 down(&fs_info->uuid_tree_rescan_sem);
3600 /* avoid complains from lockdep et al., set sem back to initial state */
3601 up(&fs_info->uuid_tree_rescan_sem);
3602
837d5b6e 3603 /* pause restriper - we want to resume on mount */
aa1b8cd4 3604 btrfs_pause_balance(fs_info);
837d5b6e 3605
8dabb742
SB		 3606 btrfs_dev_replace_suspend_for_unmount(fs_info);
3607
aa1b8cd4 3608 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3609
3610 /* wait for any defraggers to finish */
3611 wait_event(fs_info->transaction_wait,
3612 (atomic_read(&fs_info->defrag_running) == 0));
3613
3614 /* clear out the rbtree of defraggable inodes */
26176e7c 3615 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3616
c146afad 3617 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3618 ret = btrfs_commit_super(root);
3619 if (ret)
3620 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3621 }
3622
87533c47 3623 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3624 btrfs_error_commit_super(root);
0f7d52f4 3625
300e4f8a
JB		 3626 btrfs_put_block_group_cache(fs_info);
3627
e3029d9f
AV		 3628 kthread_stop(fs_info->transaction_kthread);
3629 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3630
f25784b3
YZ		 3631 fs_info->closing = 2;
3632 smp_mb();
3633
bcef60f2
AJ		 3634 btrfs_free_qgroup_config(root->fs_info);
3635
963d678b
MX		 3636 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3637 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3638 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3639 }
bcc63abb 3640
e3029d9f 3641 btrfs_free_block_groups(fs_info);
d10c5f31 3642
13e6c37b 3643 btrfs_stop_all_workers(fs_info);
2932505a 3644
c146afad 3645 del_fs_roots(fs_info);
d10c5f31 3646
13e6c37b 3647 free_root_pointers(fs_info, 1);
9ad6b7bc 3648
13e6c37b 3649 iput(fs_info->btree_inode);
d6bfde87 3650
21adbd5c
SB		 3651#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3652 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3653 btrfsic_unmount(root, fs_info->fs_devices);
3654#endif
3655
dfe25020 3656 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3657 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3658
e2d84521 3659 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3660 percpu_counter_destroy(&fs_info->delalloc_bytes);
04160088 3661 bdi_destroy(&fs_info->bdi);
76dda93c 3662 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3663
53b381b3
DW		 3664 btrfs_free_stripe_hash_table(fs_info);
3665
1cb048f5
FDBM		 3666 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3667 root->orphan_block_rsv = NULL;
3668
eb60ceac
CM		 3669 return 0;
3670}
3671
b9fab919
CM		 3672int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3673 int atomic)
5f39d397 3674{
1259ab75 3675 int ret;
727011e0 3676 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3677
0b32f4bb 3678 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3679 if (!ret)
3680 return ret;
3681
3682 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3683 parent_transid, atomic);
3684 if (ret == -EAGAIN)
3685 return ret;
1259ab75 3686 return !ret;
5f39d397
CM		 3687}
3688
3689int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3690{
0b32f4bb 3691 return set_extent_buffer_uptodate(buf);
5f39d397 3692}
6702ed49 3693
5f39d397
CM		 3694void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3695{
06ea65a3 3696 struct btrfs_root *root;
5f39d397 3697 u64 transid = btrfs_header_generation(buf);
b9473439 3698 int was_dirty;
b4ce94de 3699
06ea65a3
JB		 3700#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3701 /*
3702 * This is a fast path so only do this check if we have sanity tests
3703 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3704 * outside of the sanity tests.
3705 */
3706 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3707 return;
3708#endif
3709 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
b9447ef8 3710 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3711 if (transid != root->fs_info->generation)
3712 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3713 "found %llu running %llu\n",
c1c9ff7c 3714 buf->start, transid, root->fs_info->generation);
0b32f4bb 3715 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3716 if (!was_dirty)
3717 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3718 buf->len,
3719 root->fs_info->dirty_metadata_batch);
eb60ceac
CM		 3720}
3721
b53d3f5d
LB		 3722static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3723 int flush_delayed)
16cdcec7
MX		 3724{
3725 /*
3726 * looks as though older kernels can get into trouble with
3727 * this code, they end up stuck in balance_dirty_pages forever
3728 */
e2d84521 3729 int ret;
16cdcec7
MX		 3730
3731 if (current->flags & PF_MEMALLOC)
3732 return;
3733
b53d3f5d
LB		 3734 if (flush_delayed)
3735 btrfs_balance_delayed_items(root);
16cdcec7 3736
e2d84521
MX		 3737 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3738 BTRFS_DIRTY_METADATA_THRESH);
3739 if (ret > 0) {
d0e1d66b
NJ		 3740 balance_dirty_pages_ratelimited(
3741 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3742 }
3743 return;
3744}
3745
b53d3f5d 3746void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3747{
b53d3f5d
LB		 3748 __btrfs_btree_balance_dirty(root, 1);
3749}
585ad2c3 3750
b53d3f5d
LB		 3751void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3752{
3753 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3754}
6b80053d 3755
ca7a79ad 3756int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3757{
727011e0 3758 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3759 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3760}
0da5468f 3761
fcd1f065 3762static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3763 int read_only)
3764{
1104a885
DS		 3765 /*
3766 * Placeholder for checks
3767 */
fcd1f065 3768 return 0;
acce952b 3769}
3770
48a3b636 3771static void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3772{
acce952b 3773 mutex_lock(&root->fs_info->cleaner_mutex);
3774 btrfs_run_delayed_iputs(root);
3775 mutex_unlock(&root->fs_info->cleaner_mutex);
3776
3777 down_write(&root->fs_info->cleanup_work_sem);
3778 up_write(&root->fs_info->cleanup_work_sem);
3779
3780 /* cleanup FS via transaction */
3781 btrfs_cleanup_transaction(root);
acce952b 3782}
3783
569e0f35
JB		 3784static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3785 struct btrfs_root *root)
acce952b 3786{
3787 struct btrfs_inode *btrfs_inode;
3788 struct list_head splice;
3789
3790 INIT_LIST_HEAD(&splice);
3791
3792 mutex_lock(&root->fs_info->ordered_operations_mutex);
199c2a9c 3793 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3794
569e0f35 3795 list_splice_init(&t->ordered_operations, &splice);
acce952b 3796 while (!list_empty(&splice)) {
3797 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3798 ordered_operations);
3799
3800 list_del_init(&btrfs_inode->ordered_operations);
199c2a9c 3801 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3802
3803 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3804
199c2a9c 3805 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3806 }
3807
199c2a9c 3808 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3809 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3810}
3811
143bede5 3812static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3813{
acce952b 3814 struct btrfs_ordered_extent *ordered;
acce952b 3815
199c2a9c 3816 spin_lock(&root->ordered_extent_lock);
779880ef
JB		 3817 /*
3818 * This will just short circuit the ordered completion stuff which will
3819 * make sure the ordered extent gets properly cleaned up.
3820 */
199c2a9c 3821 list_for_each_entry(ordered, &root->ordered_extents,
779880ef
JB		 3822 root_extent_list)
3823 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
199c2a9c
MX		 3824 spin_unlock(&root->ordered_extent_lock);
3825}
3826
3827static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3828{
3829 struct btrfs_root *root;
3830 struct list_head splice;
3831
3832 INIT_LIST_HEAD(&splice);
3833
3834 spin_lock(&fs_info->ordered_root_lock);
3835 list_splice_init(&fs_info->ordered_roots, &splice);
3836 while (!list_empty(&splice)) {
3837 root = list_first_entry(&splice, struct btrfs_root,
3838 ordered_root);
3839 list_del_init(&root->ordered_root);
3840
3841 btrfs_destroy_ordered_extents(root);
3842
3843 cond_resched_lock(&fs_info->ordered_root_lock);
3844 }
3845 spin_unlock(&fs_info->ordered_root_lock);
acce952b 3846}
3847
35a3621b
SB		 3848static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3849 struct btrfs_root *root)
acce952b 3850{
3851 struct rb_node *node;
3852 struct btrfs_delayed_ref_root *delayed_refs;
3853 struct btrfs_delayed_ref_node *ref;
3854 int ret = 0;
3855
3856 delayed_refs = &trans->delayed_refs;
3857
3858 spin_lock(&delayed_refs->lock);
3859 if (delayed_refs->num_entries == 0) {
cfece4db 3860 spin_unlock(&delayed_refs->lock);
acce952b 3861 printk(KERN_INFO "delayed_refs has NO entry\n");
3862 return ret;
3863 }
3864
b939d1ab 3865 while ((node = rb_first(&delayed_refs->root)) != NULL) {
eb12db69 3866 struct btrfs_delayed_ref_head *head = NULL;
e78417d1 3867 bool pin_bytes = false;
acce952b 3868
eb12db69 3869 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
acce952b 3870 atomic_set(&ref->refs, 1);
3871 if (btrfs_delayed_ref_is_head(ref)) {
acce952b 3872
3873 head = btrfs_delayed_node_to_head(ref);
b939d1ab
JB		 3874 if (!mutex_trylock(&head->mutex)) {
3875 atomic_inc(&ref->refs);
3876 spin_unlock(&delayed_refs->lock);
3877
3878 /* Need to wait for the delayed ref to run */
3879 mutex_lock(&head->mutex);
3880 mutex_unlock(&head->mutex);
3881 btrfs_put_delayed_ref(ref);
3882
e18fca73 3883 spin_lock(&delayed_refs->lock);
b939d1ab
JB		 3884 continue;
3885 }
3886
54067ae9 3887 if (head->must_insert_reserved)
e78417d1 3888 pin_bytes = true;
78a6184a 3889 btrfs_free_delayed_extent_op(head->extent_op);
acce952b 3890 delayed_refs->num_heads--;
3891 if (list_empty(&head->cluster))
3892 delayed_refs->num_heads_ready--;
3893 list_del_init(&head->cluster);
acce952b 3894 }
eb12db69 3895
b939d1ab
JB		 3896 ref->in_tree = 0;
3897 rb_erase(&ref->rb_node, &delayed_refs->root);
3898 delayed_refs->num_entries--;
acce952b 3899 spin_unlock(&delayed_refs->lock);
e78417d1
JB		 3900 if (head) {
3901 if (pin_bytes)
3902 btrfs_pin_extent(root, ref->bytenr,
3903 ref->num_bytes, 1);
3904 mutex_unlock(&head->mutex);
3905 }
acce952b 3906 btrfs_put_delayed_ref(ref);
3907
3908 cond_resched();
3909 spin_lock(&delayed_refs->lock);
3910 }
3911
3912 spin_unlock(&delayed_refs->lock);
3913
3914 return ret;
3915}
3916
aec8030a 3917static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
acce952b 3918{
3919 struct btrfs_pending_snapshot *snapshot;
3920 struct list_head splice;
3921
3922 INIT_LIST_HEAD(&splice);
3923
3924 list_splice_init(&t->pending_snapshots, &splice);
3925
3926 while (!list_empty(&splice)) {
3927 snapshot = list_entry(splice.next,
3928 struct btrfs_pending_snapshot,
3929 list);
aec8030a 3930 snapshot->error = -ECANCELED;
acce952b 3931 list_del_init(&snapshot->list);
acce952b 3932 }
acce952b 3933}
3934
143bede5 3935static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3936{
3937 struct btrfs_inode *btrfs_inode;
3938 struct list_head splice;
3939
3940 INIT_LIST_HEAD(&splice);
3941
eb73c1b7
MX		 3942 spin_lock(&root->delalloc_lock);
3943 list_splice_init(&root->delalloc_inodes, &splice);
acce952b 3944
3945 while (!list_empty(&splice)) {
eb73c1b7
MX		 3946 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3947 delalloc_inodes);
acce952b 3948
3949 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3950 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3951 &btrfs_inode->runtime_flags);
eb73c1b7 3952 spin_unlock(&root->delalloc_lock);
acce952b 3953
3954 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3955
eb73c1b7 3956 spin_lock(&root->delalloc_lock);
acce952b 3957 }
3958
eb73c1b7
MX		 3959 spin_unlock(&root->delalloc_lock);
3960}
3961
3962static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3963{
3964 struct btrfs_root *root;
3965 struct list_head splice;
3966
3967 INIT_LIST_HEAD(&splice);
3968
3969 spin_lock(&fs_info->delalloc_root_lock);
3970 list_splice_init(&fs_info->delalloc_roots, &splice);
3971 while (!list_empty(&splice)) {
3972 root = list_first_entry(&splice, struct btrfs_root,
3973 delalloc_root);
3974 list_del_init(&root->delalloc_root);
3975 root = btrfs_grab_fs_root(root);
3976 BUG_ON(!root);
3977 spin_unlock(&fs_info->delalloc_root_lock);
3978
3979 btrfs_destroy_delalloc_inodes(root);
3980 btrfs_put_fs_root(root);
3981
3982 spin_lock(&fs_info->delalloc_root_lock);
3983 }
3984 spin_unlock(&fs_info->delalloc_root_lock);
acce952b 3985}
3986
3987static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3988 struct extent_io_tree *dirty_pages,
3989 int mark)
3990{
3991 int ret;
acce952b 3992 struct extent_buffer *eb;
3993 u64 start = 0;
3994 u64 end;
acce952b 3995
3996 while (1) {
3997 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3998 mark, NULL);
acce952b 3999 if (ret)
4000 break;
4001
4002 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4003 while (start <= end) {
fd8b2b61
JB		 4004 eb = btrfs_find_tree_block(root, start,
4005 root->leafsize);
69a85bd8 4006 start += root->leafsize;
fd8b2b61 4007 if (!eb)
acce952b 4008 continue;
fd8b2b61 4009 wait_on_extent_buffer_writeback(eb);
acce952b 4010
fd8b2b61
JB		 4011 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4012 &eb->bflags))
4013 clear_extent_buffer_dirty(eb);
4014 free_extent_buffer_stale(eb);
acce952b 4015 }
4016 }
4017
4018 return ret;
4019}
4020
4021static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4022 struct extent_io_tree *pinned_extents)
4023{
4024 struct extent_io_tree *unpin;
4025 u64 start;
4026 u64 end;
4027 int ret;
ed0eaa14 4028 bool loop = true;
acce952b 4029
4030 unpin = pinned_extents;
ed0eaa14 4031again:
acce952b 4032 while (1) {
4033 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 4034 EXTENT_DIRTY, NULL);
acce952b 4035 if (ret)
4036 break;
4037
4038 /* opt_discard */
5378e607
LD		 4039 if (btrfs_test_opt(root, DISCARD))
4040 ret = btrfs_error_discard_extent(root, start,
4041 end + 1 - start,
4042 NULL);
acce952b 4043
4044 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4045 btrfs_error_unpin_extent_range(root, start, end);
4046 cond_resched();
4047 }
4048
ed0eaa14
LB		 4049 if (loop) {
4050 if (unpin == &root->fs_info->freed_extents[0])
4051 unpin = &root->fs_info->freed_extents[1];
4052 else
4053 unpin = &root->fs_info->freed_extents[0];
4054 loop = false;
4055 goto again;
4056 }
4057
acce952b 4058 return 0;
4059}
4060
49b25e05
JM		 4061void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4062 struct btrfs_root *root)
4063{
4064 btrfs_destroy_delayed_refs(cur_trans, root);
4065 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
4066 cur_trans->dirty_pages.dirty_bytes);
4067
4a9d8bde 4068 cur_trans->state = TRANS_STATE_COMMIT_START;
d7096fc3 4069 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 4070
aec8030a
MX		 4071 btrfs_evict_pending_snapshots(cur_trans);
4072
4a9d8bde 4073 cur_trans->state = TRANS_STATE_UNBLOCKED;
d7096fc3 4074 wake_up(&root->fs_info->transaction_wait);
49b25e05 4075
67cde344
MX		 4076 btrfs_destroy_delayed_inodes(root);
4077 btrfs_assert_delayed_root_empty(root);
49b25e05 4078
49b25e05
JM		 4079 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4080 EXTENT_DIRTY);
6e841e32
LB		 4081 btrfs_destroy_pinned_extent(root,
4082 root->fs_info->pinned_extents);
49b25e05 4083
4a9d8bde
MX		 4084 cur_trans->state =TRANS_STATE_COMPLETED;
4085 wake_up(&cur_trans->commit_wait);
4086
49b25e05
JM		 4087 /*
4088 memset(cur_trans, 0, sizeof(*cur_trans));
4089 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4090 */
4091}
4092
48a3b636 4093static int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 4094{
4095 struct btrfs_transaction *t;
4096 LIST_HEAD(list);
4097
acce952b 4098 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4099
a4abeea4 4100 spin_lock(&root->fs_info->trans_lock);
acce952b 4101 list_splice_init(&root->fs_info->trans_list, &list);
ac673879 4102 root->fs_info->running_transaction = NULL;
a4abeea4
JB		 4103 spin_unlock(&root->fs_info->trans_lock);
4104
acce952b 4105 while (!list_empty(&list)) {
4106 t = list_entry(list.next, struct btrfs_transaction, list);
acce952b 4107
569e0f35 4108 btrfs_destroy_ordered_operations(t, root);
acce952b 4109
199c2a9c 4110 btrfs_destroy_all_ordered_extents(root->fs_info);
acce952b 4111
4112 btrfs_destroy_delayed_refs(t, root);
4113
4a9d8bde
MX		 4114 /*
4115 * FIXME: cleanup wait for commit
4116 * We needn't acquire the lock here, because we are during
4117 * the umount, there is no other task which will change it.
4118 */
4119 t->state = TRANS_STATE_COMMIT_START;
66657b31 4120 smp_mb();
acce952b 4121 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4122 wake_up(&root->fs_info->transaction_blocked_wait);
4123
aec8030a
MX		 4124 btrfs_evict_pending_snapshots(t);
4125
4a9d8bde 4126 t->state = TRANS_STATE_UNBLOCKED;
66657b31 4127 smp_mb();
acce952b 4128 if (waitqueue_active(&root->fs_info->transaction_wait))
4129 wake_up(&root->fs_info->transaction_wait);
acce952b 4130
67cde344
MX		 4131 btrfs_destroy_delayed_inodes(root);
4132 btrfs_assert_delayed_root_empty(root);
4133
eb73c1b7 4134 btrfs_destroy_all_delalloc_inodes(root->fs_info);
acce952b 4135
4136 btrfs_destroy_marked_extents(root, &t->dirty_pages,
4137 EXTENT_DIRTY);
4138
4139 btrfs_destroy_pinned_extent(root,
4140 root->fs_info->pinned_extents);
4141
4a9d8bde
MX		 4142 t->state = TRANS_STATE_COMPLETED;
4143 smp_mb();
4144 if (waitqueue_active(&t->commit_wait))
4145 wake_up(&t->commit_wait);
4146
13c5a93e 4147 atomic_set(&t->use_count, 0);
acce952b 4148 list_del_init(&t->list);
4149 memset(t, 0, sizeof(*t));
4150 kmem_cache_free(btrfs_transaction_cachep, t);
4151 }
4152
4153 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 4154
4155 return 0;
4156}
4157
d1310b2e 4158static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 4159 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 4160 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 4161 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 4162 /* note we're sharing with inode.c for the merge bio hook */
4163 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 4164};
Linux 6.x block layer and io_uring tree(s)