projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
btrfs: Replace fs_info->flush_workers with btrfs_workqueue.
[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>
5a0e3ad6 29#include <linux/slab.h>
784b4e29 30#include <linux/migrate.h>
7a36ddec 31#include <linux/ratelimit.h>
6463fe58 32#include <linux/uuid.h>
803b2f54 33#include <linux/semaphore.h>
7e75bf3f 34#include <asm/unaligned.h>
eb60ceac
CM		 35#include "ctree.h"
36#include "disk-io.h"
0b947aff 37#include "hash.h"
e089f05c 38#include "transaction.h"
0f7d52f4 39#include "btrfs_inode.h"
0b86a832 40#include "volumes.h"
db94535d 41#include "print-tree.h"
8b712842 42#include "async-thread.h"
925baedd 43#include "locking.h"
e02119d5 44#include "tree-log.h"
fa9c0d79 45#include "free-space-cache.h"
581bb050 46#include "inode-map.h"
21adbd5c 47#include "check-integrity.h"
606686ee 48#include "rcu-string.h"
8dabb742 49#include "dev-replace.h"
53b381b3 50#include "raid56.h"
5ac1d209 51#include "sysfs.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);
143bede5 67static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
acce952b 68static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 struct extent_io_tree *dirty_pages,
70 int mark);
71static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 struct extent_io_tree *pinned_extents);
48a3b636
ES		 73static int btrfs_cleanup_transaction(struct btrfs_root *root);
74static void btrfs_error_commit_super(struct btrfs_root *root);
ce9adaa5 75
d352ac68
CM		 76/*
77 * end_io_wq structs are used to do processing in task context when an IO is
78 * complete. This is used during reads to verify checksums, and it is used
79 * by writes to insert metadata for new file extents after IO is complete.
80 */
ce9adaa5
CM		 81struct end_io_wq {
82 struct bio *bio;
83 bio_end_io_t *end_io;
84 void *private;
85 struct btrfs_fs_info *info;
86 int error;
22c59948 87 int metadata;
ce9adaa5 88 struct list_head list;
8b712842 89 struct btrfs_work work;
ce9adaa5 90};
0da5468f 91
d352ac68
CM		 92/*
93 * async submit bios are used to offload expensive checksumming
94 * onto the worker threads. They checksum file and metadata bios
95 * just before they are sent down the IO stack.
96 */
44b8bd7e
CM		 97struct async_submit_bio {
98 struct inode *inode;
99 struct bio *bio;
100 struct list_head list;
4a69a410
CM		 101 extent_submit_bio_hook_t *submit_bio_start;
102 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 103 int rw;
104 int mirror_num;
c8b97818 105 unsigned long bio_flags;
eaf25d93
CM		 106 /*
107 * bio_offset is optional, can be used if the pages in the bio
108 * can't tell us where in the file the bio should go
109 */
110 u64 bio_offset;
5cdc7ad3 111 struct btrfs_work_struct work;
79787eaa 112 int error;
44b8bd7e
CM		 113};
114
85d4e461
CM		 115/*
116 * Lockdep class keys for extent_buffer->lock's in this root. For a given
117 * eb, the lockdep key is determined by the btrfs_root it belongs to and
118 * the level the eb occupies in the tree.
119 *
120 * Different roots are used for different purposes and may nest inside each
121 * other and they require separate keysets. As lockdep keys should be
122 * static, assign keysets according to the purpose of the root as indicated
123 * by btrfs_root->objectid. This ensures that all special purpose roots
124 * have separate keysets.
4008c04a 125 *
85d4e461
CM		 126 * Lock-nesting across peer nodes is always done with the immediate parent
127 * node locked thus preventing deadlock. As lockdep doesn't know this, use
128 * subclass to avoid triggering lockdep warning in such cases.
4008c04a 129 *
85d4e461
CM		 130 * The key is set by the readpage_end_io_hook after the buffer has passed
131 * csum validation but before the pages are unlocked. It is also set by
132 * btrfs_init_new_buffer on freshly allocated blocks.
4008c04a 133 *
85d4e461
CM		 134 * We also add a check to make sure the highest level of the tree is the
135 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
136 * needs update as well.
4008c04a
CM		 137 */
138#ifdef CONFIG_DEBUG_LOCK_ALLOC
139# if BTRFS_MAX_LEVEL != 8
140# error
141# endif
85d4e461
CM		 142
143static struct btrfs_lockdep_keyset {
144 u64 id; /* root objectid */
145 const char *name_stem; /* lock name stem */
146 char names[BTRFS_MAX_LEVEL + 1][20];
147 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
148} btrfs_lockdep_keysets[] = {
149 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
150 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
151 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
152 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
153 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
154 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
60b62978 155 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
85d4e461
CM		 156 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
157 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
158 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
13fd8da9 159 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
85d4e461 160 { .id = 0, .name_stem = "tree" },
4008c04a 161};
85d4e461
CM		 162
163void __init btrfs_init_lockdep(void)
164{
165 int i, j;
166
167 /* initialize lockdep class names */
168 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
169 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
170
171 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
172 snprintf(ks->names[j], sizeof(ks->names[j]),
173 "btrfs-%s-%02d", ks->name_stem, j);
174 }
175}
176
177void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
178 int level)
179{
180 struct btrfs_lockdep_keyset *ks;
181
182 BUG_ON(level >= ARRAY_SIZE(ks->keys));
183
184 /* find the matching keyset, id 0 is the default entry */
185 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
186 if (ks->id == objectid)
187 break;
188
189 lockdep_set_class_and_name(&eb->lock,
190 &ks->keys[level], ks->names[level]);
191}
192
4008c04a
CM		 193#endif
194
d352ac68
CM		 195/*
196 * extents on the btree inode are pretty simple, there's one extent
197 * that covers the entire device
198 */
b2950863 199static struct extent_map *btree_get_extent(struct inode *inode,
306e16ce 200 struct page *page, size_t pg_offset, u64 start, u64 len,
b2950863 201 int create)
7eccb903 202{
5f39d397
CM		 203 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
204 struct extent_map *em;
205 int ret;
206
890871be 207 read_lock(&em_tree->lock);
d1310b2e 208 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 209 if (em) {
210 em->bdev =
211 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 212 read_unlock(&em_tree->lock);
5f39d397 213 goto out;
a061fc8d 214 }
890871be 215 read_unlock(&em_tree->lock);
7b13b7b1 216
172ddd60 217 em = alloc_extent_map();
5f39d397
CM		 218 if (!em) {
219 em = ERR_PTR(-ENOMEM);
220 goto out;
221 }
222 em->start = 0;
0afbaf8c 223 em->len = (u64)-1;
c8b97818 224 em->block_len = (u64)-1;
5f39d397 225 em->block_start = 0;
a061fc8d 226 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 227
890871be 228 write_lock(&em_tree->lock);
09a2a8f9 229 ret = add_extent_mapping(em_tree, em, 0);
5f39d397
CM		 230 if (ret == -EEXIST) {
231 free_extent_map(em);
7b13b7b1 232 em = lookup_extent_mapping(em_tree, start, len);
b4f359ab 233 if (!em)
0433f20d 234 em = ERR_PTR(-EIO);
5f39d397 235 } else if (ret) {
7b13b7b1 236 free_extent_map(em);
0433f20d 237 em = ERR_PTR(ret);
5f39d397 238 }
890871be 239 write_unlock(&em_tree->lock);
7b13b7b1 240
5f39d397
CM		 241out:
242 return em;
7eccb903
CM		 243}
244
b0496686 245u32 btrfs_csum_data(char *data, u32 seed, size_t len)
19c00ddc 246{
0b947aff 247 return btrfs_crc32c(seed, data, len);
19c00ddc
CM		 248}
249
250void btrfs_csum_final(u32 crc, char *result)
251{
7e75bf3f 252 put_unaligned_le32(~crc, result);
19c00ddc
CM		 253}
254
d352ac68
CM		 255/*
256 * compute the csum for a btree block, and either verify it or write it
257 * into the csum field of the block.
258 */
19c00ddc
CM		 259static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
260 int verify)
261{
6c41761f 262 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
607d432d 263 char *result = NULL;
19c00ddc
CM		 264 unsigned long len;
265 unsigned long cur_len;
266 unsigned long offset = BTRFS_CSUM_SIZE;
19c00ddc
CM		 267 char *kaddr;
268 unsigned long map_start;
269 unsigned long map_len;
270 int err;
271 u32 crc = ~(u32)0;
607d432d 272 unsigned long inline_result;
19c00ddc
CM		 273
274 len = buf->len - offset;
d397712b 275 while (len > 0) {
19c00ddc 276 err = map_private_extent_buffer(buf, offset, 32,
a6591715 277 &kaddr, &map_start, &map_len);
d397712b 278 if (err)
19c00ddc 279 return 1;
19c00ddc 280 cur_len = min(len, map_len - (offset - map_start));
b0496686 281 crc = btrfs_csum_data(kaddr + offset - map_start,
19c00ddc
CM		 282 crc, cur_len);
283 len -= cur_len;
284 offset += cur_len;
19c00ddc 285 }
607d432d
JB		 286 if (csum_size > sizeof(inline_result)) {
287 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
288 if (!result)
289 return 1;
290 } else {
291 result = (char *)&inline_result;
292 }
293
19c00ddc
CM		 294 btrfs_csum_final(crc, result);
295
296 if (verify) {
607d432d 297 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 298 u32 val;
299 u32 found = 0;
607d432d 300 memcpy(&found, result, csum_size);
e4204ded 301
607d432d 302 read_extent_buffer(buf, &val, 0, csum_size);
efe120a0
FH		 303 printk_ratelimited(KERN_INFO
304 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
305 "level %d\n",
306 root->fs_info->sb->s_id, buf->start,
307 val, found, btrfs_header_level(buf));
607d432d
JB		 308 if (result != (char *)&inline_result)
309 kfree(result);
19c00ddc
CM		 310 return 1;
311 }
312 } else {
607d432d 313 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 314 }
607d432d
JB		 315 if (result != (char *)&inline_result)
316 kfree(result);
19c00ddc
CM		 317 return 0;
318}
319
d352ac68
CM		 320/*
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
325 */
1259ab75 326static int verify_parent_transid(struct extent_io_tree *io_tree,
b9fab919
CM		 327 struct extent_buffer *eb, u64 parent_transid,
328 int atomic)
1259ab75 329{
2ac55d41 330 struct extent_state *cached_state = NULL;
1259ab75
CM		 331 int ret;
332
333 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334 return 0;
335
b9fab919
CM		 336 if (atomic)
337 return -EAGAIN;
338
2ac55d41 339 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
d0082371 340 0, &cached_state);
0b32f4bb 341 if (extent_buffer_uptodate(eb) &&
1259ab75
CM		 342 btrfs_header_generation(eb) == parent_transid) {
343 ret = 0;
344 goto out;
345 }
7a36ddec 346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
193f284d 347 "found %llu\n",
c1c9ff7c 348 eb->start, parent_transid, btrfs_header_generation(eb));
1259ab75 349 ret = 1;
0b32f4bb 350 clear_extent_buffer_uptodate(eb);
33958dc6 351out:
2ac55d41
JB		 352 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
353 &cached_state, GFP_NOFS);
1259ab75 354 return ret;
1259ab75
CM		 355}
356
1104a885
DS		 357/*
358 * Return 0 if the superblock checksum type matches the checksum value of that
359 * algorithm. Pass the raw disk superblock data.
360 */
361static int btrfs_check_super_csum(char *raw_disk_sb)
362{
363 struct btrfs_super_block *disk_sb =
364 (struct btrfs_super_block *)raw_disk_sb;
365 u16 csum_type = btrfs_super_csum_type(disk_sb);
366 int ret = 0;
367
368 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
369 u32 crc = ~(u32)0;
370 const int csum_size = sizeof(crc);
371 char result[csum_size];
372
373 /*
374 * The super_block structure does not span the whole
375 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
376 * is filled with zeros and is included in the checkum.
377 */
378 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
379 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
380 btrfs_csum_final(crc, result);
381
382 if (memcmp(raw_disk_sb, result, csum_size))
383 ret = 1;
667e7d94
CM		 384
385 if (ret && btrfs_super_generation(disk_sb) < 10) {
efe120a0
FH		 386 printk(KERN_WARNING
387 "BTRFS: super block crcs don't match, older mkfs detected\n");
667e7d94
CM		 388 ret = 0;
389 }
1104a885
DS		 390 }
391
392 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
efe120a0 393 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
1104a885
DS		 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{
4eee4fa4 469 u64 start = page_offset(page);
19c00ddc 470 u64 found_start;
19c00ddc 471 struct extent_buffer *eb;
f188591e 472
4f2de97a
JB		 473 eb = (struct extent_buffer *)page->private;
474 if (page != eb->pages[0])
475 return 0;
19c00ddc 476 found_start = btrfs_header_bytenr(eb);
fae7f21c 477 if (WARN_ON(found_start != start || !PageUptodate(page)))
4f2de97a 478 return 0;
19c00ddc 479 csum_tree_block(root, eb, 0);
19c00ddc
CM		 480 return 0;
481}
482
2b82032c
YZ		 483static int check_tree_block_fsid(struct btrfs_root *root,
484 struct extent_buffer *eb)
485{
486 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
487 u8 fsid[BTRFS_UUID_SIZE];
488 int ret = 1;
489
0a4e5586 490 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
2b82032c
YZ		 491 while (fs_devices) {
492 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
493 ret = 0;
494 break;
495 }
496 fs_devices = fs_devices->seed;
497 }
498 return ret;
499}
500
a826d6dc 501#define CORRUPT(reason, eb, root, slot) \
efe120a0
FH		 502 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
503 "root=%llu, slot=%d", reason, \
c1c9ff7c 504 btrfs_header_bytenr(eb), root->objectid, slot)
a826d6dc
JB		 505
506static noinline int check_leaf(struct btrfs_root *root,
507 struct extent_buffer *leaf)
508{
509 struct btrfs_key key;
510 struct btrfs_key leaf_key;
511 u32 nritems = btrfs_header_nritems(leaf);
512 int slot;
513
514 if (nritems == 0)
515 return 0;
516
517 /* Check the 0 item */
518 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
519 BTRFS_LEAF_DATA_SIZE(root)) {
520 CORRUPT("invalid item offset size pair", leaf, root, 0);
521 return -EIO;
522 }
523
524 /*
525 * Check to make sure each items keys are in the correct order and their
526 * offsets make sense. We only have to loop through nritems-1 because
527 * we check the current slot against the next slot, which verifies the
528 * next slot's offset+size makes sense and that the current's slot
529 * offset is correct.
530 */
531 for (slot = 0; slot < nritems - 1; slot++) {
532 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
533 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
534
535 /* Make sure the keys are in the right order */
536 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
537 CORRUPT("bad key order", leaf, root, slot);
538 return -EIO;
539 }
540
541 /*
542 * Make sure the offset and ends are right, remember that the
543 * item data starts at the end of the leaf and grows towards the
544 * front.
545 */
546 if (btrfs_item_offset_nr(leaf, slot) !=
547 btrfs_item_end_nr(leaf, slot + 1)) {
548 CORRUPT("slot offset bad", leaf, root, slot);
549 return -EIO;
550 }
551
552 /*
553 * Check to make sure that we don't point outside of the leaf,
554 * just incase all the items are consistent to eachother, but
555 * all point outside of the leaf.
556 */
557 if (btrfs_item_end_nr(leaf, slot) >
558 BTRFS_LEAF_DATA_SIZE(root)) {
559 CORRUPT("slot end outside of leaf", leaf, root, slot);
560 return -EIO;
561 }
562 }
563
564 return 0;
565}
566
facc8a22
MX		 567static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
568 u64 phy_offset, struct page *page,
569 u64 start, u64 end, int mirror)
ce9adaa5 570{
ce9adaa5
CM		 571 u64 found_start;
572 int found_level;
ce9adaa5
CM		 573 struct extent_buffer *eb;
574 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 575 int ret = 0;
727011e0 576 int reads_done;
ce9adaa5 577
ce9adaa5
CM		 578 if (!page->private)
579 goto out;
d397712b 580
4f2de97a 581 eb = (struct extent_buffer *)page->private;
d397712b 582
0b32f4bb
JB		 583 /* the pending IO might have been the only thing that kept this buffer
584 * in memory. Make sure we have a ref for all this other checks
585 */
586 extent_buffer_get(eb);
587
588 reads_done = atomic_dec_and_test(&eb->io_pages);
727011e0
CM		 589 if (!reads_done)
590 goto err;
f188591e 591
5cf1ab56 592 eb->read_mirror = mirror;
ea466794
JB		 593 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
594 ret = -EIO;
595 goto err;
596 }
597
ce9adaa5 598 found_start = btrfs_header_bytenr(eb);
727011e0 599 if (found_start != eb->start) {
efe120a0 600 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
193f284d 601 "%llu %llu\n",
c1c9ff7c 602 found_start, eb->start);
f188591e 603 ret = -EIO;
ce9adaa5
CM		 604 goto err;
605 }
2b82032c 606 if (check_tree_block_fsid(root, eb)) {
efe120a0 607 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
c1c9ff7c 608 eb->start);
1259ab75
CM		 609 ret = -EIO;
610 goto err;
611 }
ce9adaa5 612 found_level = btrfs_header_level(eb);
1c24c3ce 613 if (found_level >= BTRFS_MAX_LEVEL) {
efe120a0 614 btrfs_info(root->fs_info, "bad tree block level %d",
1c24c3ce
JB		 615 (int)btrfs_header_level(eb));
616 ret = -EIO;
617 goto err;
618 }
ce9adaa5 619
85d4e461
CM		 620 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
621 eb, found_level);
4008c04a 622
ce9adaa5 623 ret = csum_tree_block(root, eb, 1);
a826d6dc 624 if (ret) {
f188591e 625 ret = -EIO;
a826d6dc
JB		 626 goto err;
627 }
628
629 /*
630 * If this is a leaf block and it is corrupt, set the corrupt bit so
631 * that we don't try and read the other copies of this block, just
632 * return -EIO.
633 */
634 if (found_level == 0 && check_leaf(root, eb)) {
635 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
636 ret = -EIO;
637 }
ce9adaa5 638
0b32f4bb
JB		 639 if (!ret)
640 set_extent_buffer_uptodate(eb);
ce9adaa5 641err:
79fb65a1
JB		 642 if (reads_done &&
643 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 644 btree_readahead_hook(root, eb, eb->start, ret);
4bb31e92 645
53b381b3
DW		 646 if (ret) {
647 /*
648 * our io error hook is going to dec the io pages
649 * again, we have to make sure it has something
650 * to decrement
651 */
652 atomic_inc(&eb->io_pages);
0b32f4bb 653 clear_extent_buffer_uptodate(eb);
53b381b3 654 }
0b32f4bb 655 free_extent_buffer(eb);
ce9adaa5 656out:
f188591e 657 return ret;
ce9adaa5
CM		 658}
659
ea466794 660static int btree_io_failed_hook(struct page *page, int failed_mirror)
4bb31e92 661{
4bb31e92
AJ		 662 struct extent_buffer *eb;
663 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
664
4f2de97a 665 eb = (struct extent_buffer *)page->private;
ea466794 666 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
5cf1ab56 667 eb->read_mirror = failed_mirror;
53b381b3 668 atomic_dec(&eb->io_pages);
ea466794 669 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 670 btree_readahead_hook(root, eb, eb->start, -EIO);
4bb31e92
AJ		 671 return -EIO; /* we fixed nothing */
672}
673
ce9adaa5 674static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 675{
676 struct end_io_wq *end_io_wq = bio->bi_private;
677 struct btrfs_fs_info *fs_info;
ce9adaa5 678
ce9adaa5 679 fs_info = end_io_wq->info;
ce9adaa5 680 end_io_wq->error = err;
8b712842
CM		 681 end_io_wq->work.func = end_workqueue_fn;
682 end_io_wq->work.flags = 0;
d20f7043 683
7b6d91da 684 if (bio->bi_rw & REQ_WRITE) {
53b381b3 685 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
cad321ad
CM		 686 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
687 &end_io_wq->work);
53b381b3 688 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
0cb59c99
JB		 689 btrfs_queue_worker(&fs_info->endio_freespace_worker,
690 &end_io_wq->work);
53b381b3
DW		 691 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
692 btrfs_queue_worker(&fs_info->endio_raid56_workers,
693 &end_io_wq->work);
cad321ad
CM		 694 else
695 btrfs_queue_worker(&fs_info->endio_write_workers,
696 &end_io_wq->work);
d20f7043 697 } else {
53b381b3
DW		 698 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
699 btrfs_queue_worker(&fs_info->endio_raid56_workers,
700 &end_io_wq->work);
701 else if (end_io_wq->metadata)
d20f7043
CM		 702 btrfs_queue_worker(&fs_info->endio_meta_workers,
703 &end_io_wq->work);
704 else
705 btrfs_queue_worker(&fs_info->endio_workers,
706 &end_io_wq->work);
707 }
ce9adaa5
CM		 708}
709
0cb59c99
JB		 710/*
711 * For the metadata arg you want
712 *
713 * 0 - if data
714 * 1 - if normal metadta
715 * 2 - if writing to the free space cache area
53b381b3 716 * 3 - raid parity work
0cb59c99 717 */
22c59948
CM		 718int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
719 int metadata)
0b86a832 720{
ce9adaa5 721 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 722 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
723 if (!end_io_wq)
724 return -ENOMEM;
725
726 end_io_wq->private = bio->bi_private;
727 end_io_wq->end_io = bio->bi_end_io;
22c59948 728 end_io_wq->info = info;
ce9adaa5
CM		 729 end_io_wq->error = 0;
730 end_io_wq->bio = bio;
22c59948 731 end_io_wq->metadata = metadata;
ce9adaa5
CM		 732
733 bio->bi_private = end_io_wq;
734 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 735 return 0;
736}
737
b64a2851 738unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 739{
4854ddd0 740 unsigned long limit = min_t(unsigned long,
5cdc7ad3 741 info->thread_pool_size,
4854ddd0
CM		 742 info->fs_devices->open_devices);
743 return 256 * limit;
744}
0986fe9e 745
5cdc7ad3 746static void run_one_async_start(struct btrfs_work_struct *work)
4a69a410 747{
4a69a410 748 struct async_submit_bio *async;
79787eaa 749 int ret;
4a69a410
CM		 750
751 async = container_of(work, struct async_submit_bio, work);
79787eaa
JM		 752 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
753 async->mirror_num, async->bio_flags,
754 async->bio_offset);
755 if (ret)
756 async->error = ret;
4a69a410
CM		 757}
758
5cdc7ad3 759static void run_one_async_done(struct btrfs_work_struct *work)
8b712842
CM		 760{
761 struct btrfs_fs_info *fs_info;
762 struct async_submit_bio *async;
4854ddd0 763 int limit;
8b712842
CM		 764
765 async = container_of(work, struct async_submit_bio, work);
766 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 767
b64a2851 768 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 769 limit = limit * 2 / 3;
770
66657b31 771 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
b64a2851 772 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 773 wake_up(&fs_info->async_submit_wait);
774
79787eaa
JM		 775 /* If an error occured we just want to clean up the bio and move on */
776 if (async->error) {
777 bio_endio(async->bio, async->error);
778 return;
779 }
780
4a69a410 781 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 782 async->mirror_num, async->bio_flags,
783 async->bio_offset);
4a69a410
CM		 784}
785
5cdc7ad3 786static void run_one_async_free(struct btrfs_work_struct *work)
4a69a410
CM		 787{
788 struct async_submit_bio *async;
789
790 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 791 kfree(async);
792}
793
44b8bd7e
CM		 794int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
795 int rw, struct bio *bio, int mirror_num,
c8b97818 796 unsigned long bio_flags,
eaf25d93 797 u64 bio_offset,
4a69a410
CM		 798 extent_submit_bio_hook_t *submit_bio_start,
799 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 800{
801 struct async_submit_bio *async;
802
803 async = kmalloc(sizeof(*async), GFP_NOFS);
804 if (!async)
805 return -ENOMEM;
806
807 async->inode = inode;
808 async->rw = rw;
809 async->bio = bio;
810 async->mirror_num = mirror_num;
4a69a410
CM		 811 async->submit_bio_start = submit_bio_start;
812 async->submit_bio_done = submit_bio_done;
813
5cdc7ad3
QW		 814 btrfs_init_work(&async->work, run_one_async_start,
815 run_one_async_done, run_one_async_free);
4a69a410 816
c8b97818 817 async->bio_flags = bio_flags;
eaf25d93 818 async->bio_offset = bio_offset;
8c8bee1d 819
79787eaa
JM		 820 async->error = 0;
821
cb03c743 822 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 823
7b6d91da 824 if (rw & REQ_SYNC)
5cdc7ad3 825 btrfs_set_work_high_priority(&async->work);
d313d7a3 826
5cdc7ad3 827 btrfs_queue_work(fs_info->workers, &async->work);
9473f16c 828
d397712b 829 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 830 atomic_read(&fs_info->nr_async_submits)) {
831 wait_event(fs_info->async_submit_wait,
832 (atomic_read(&fs_info->nr_async_submits) == 0));
833 }
834
44b8bd7e
CM		 835 return 0;
836}
837
ce3ed71a
CM		 838static int btree_csum_one_bio(struct bio *bio)
839{
840 struct bio_vec *bvec = bio->bi_io_vec;
841 int bio_index = 0;
842 struct btrfs_root *root;
79787eaa 843 int ret = 0;
ce3ed71a
CM		 844
845 WARN_ON(bio->bi_vcnt <= 0);
d397712b 846 while (bio_index < bio->bi_vcnt) {
ce3ed71a 847 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 848 ret = csum_dirty_buffer(root, bvec->bv_page);
849 if (ret)
850 break;
ce3ed71a
CM		 851 bio_index++;
852 bvec++;
853 }
79787eaa 854 return ret;
ce3ed71a
CM		 855}
856
4a69a410
CM		 857static int __btree_submit_bio_start(struct inode *inode, int rw,
858 struct bio *bio, int mirror_num,
eaf25d93
CM		 859 unsigned long bio_flags,
860 u64 bio_offset)
22c59948 861{
8b712842
CM		 862 /*
863 * when we're called for a write, we're already in the async
5443be45 864 * submission context. Just jump into btrfs_map_bio
8b712842 865 */
79787eaa 866 return btree_csum_one_bio(bio);
4a69a410 867}
22c59948 868
4a69a410 869static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 870 int mirror_num, unsigned long bio_flags,
871 u64 bio_offset)
4a69a410 872{
61891923
SB		 873 int ret;
874
8b712842 875 /*
4a69a410
CM		 876 * when we're called for a write, we're already in the async
877 * submission context. Just jump into btrfs_map_bio
8b712842 878 */
61891923
SB		 879 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
880 if (ret)
881 bio_endio(bio, ret);
882 return ret;
0b86a832
CM		 883}
884
de0022b9
JB		 885static int check_async_write(struct inode *inode, unsigned long bio_flags)
886{
887 if (bio_flags & EXTENT_BIO_TREE_LOG)
888 return 0;
889#ifdef CONFIG_X86
890 if (cpu_has_xmm4_2)
891 return 0;
892#endif
893 return 1;
894}
895
44b8bd7e 896static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 897 int mirror_num, unsigned long bio_flags,
898 u64 bio_offset)
44b8bd7e 899{
de0022b9 900 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 901 int ret;
902
7b6d91da 903 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 904 /*
905 * called for a read, do the setup so that checksum validation
906 * can happen in the async kernel threads
907 */
f3f266ab
CM		 908 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
909 bio, 1);
1d4284bd 910 if (ret)
61891923
SB		 911 goto out_w_error;
912 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
913 mirror_num, 0);
de0022b9
JB		 914 } else if (!async) {
915 ret = btree_csum_one_bio(bio);
916 if (ret)
61891923
SB		 917 goto out_w_error;
918 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
919 mirror_num, 0);
920 } else {
921 /*
922 * kthread helpers are used to submit writes so that
923 * checksumming can happen in parallel across all CPUs
924 */
925 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
926 inode, rw, bio, mirror_num, 0,
927 bio_offset,
928 __btree_submit_bio_start,
929 __btree_submit_bio_done);
44b8bd7e 930 }
d313d7a3 931
61891923
SB		 932 if (ret) {
933out_w_error:
934 bio_endio(bio, ret);
935 }
936 return ret;
44b8bd7e
CM		 937}
938
3dd1462e 939#ifdef CONFIG_MIGRATION
784b4e29 940static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 941 struct page *newpage, struct page *page,
942 enum migrate_mode mode)
784b4e29
CM		 943{
944 /*
945 * we can't safely write a btree page from here,
946 * we haven't done the locking hook
947 */
948 if (PageDirty(page))
949 return -EAGAIN;
950 /*
951 * Buffers may be managed in a filesystem specific way.
952 * We must have no buffers or drop them.
953 */
954 if (page_has_private(page) &&
955 !try_to_release_page(page, GFP_KERNEL))
956 return -EAGAIN;
a6bc32b8 957 return migrate_page(mapping, newpage, page, mode);
784b4e29 958}
3dd1462e 959#endif
784b4e29 960
0da5468f
CM		 961
962static int btree_writepages(struct address_space *mapping,
963 struct writeback_control *wbc)
964{
e2d84521
MX		 965 struct btrfs_fs_info *fs_info;
966 int ret;
967
d8d5f3e1 968 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 969
970 if (wbc->for_kupdate)
971 return 0;
972
e2d84521 973 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 974 /* this is a bit racy, but that's ok */
e2d84521
MX		 975 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
976 BTRFS_DIRTY_METADATA_THRESH);
977 if (ret < 0)
793955bc 978 return 0;
793955bc 979 }
0b32f4bb 980 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 981}
982
b2950863 983static int btree_readpage(struct file *file, struct page *page)
5f39d397 984{
d1310b2e
CM		 985 struct extent_io_tree *tree;
986 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 987 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 988}
22b0ebda 989
70dec807 990static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 991{
98509cfc 992 if (PageWriteback(page) || PageDirty(page))
d397712b 993 return 0;
0c4e538b 994
f7a52a40 995 return try_release_extent_buffer(page);
d98237b3
CM		 996}
997
d47992f8
LC		 998static void btree_invalidatepage(struct page *page, unsigned int offset,
999 unsigned int length)
d98237b3 1000{
d1310b2e
CM		 1001 struct extent_io_tree *tree;
1002 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1003 extent_invalidatepage(tree, page, offset);
1004 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1005 if (PagePrivate(page)) {
efe120a0
FH		 1006 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1007 "page private not zero on page %llu",
1008 (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1009 ClearPagePrivate(page);
1010 set_page_private(page, 0);
1011 page_cache_release(page);
1012 }
d98237b3
CM		 1013}
1014
0b32f4bb
JB		 1015static int btree_set_page_dirty(struct page *page)
1016{
bb146eb2 1017#ifdef DEBUG
0b32f4bb
JB		 1018 struct extent_buffer *eb;
1019
1020 BUG_ON(!PagePrivate(page));
1021 eb = (struct extent_buffer *)page->private;
1022 BUG_ON(!eb);
1023 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1024 BUG_ON(!atomic_read(&eb->refs));
1025 btrfs_assert_tree_locked(eb);
bb146eb2 1026#endif
0b32f4bb
JB		 1027 return __set_page_dirty_nobuffers(page);
1028}
1029
7f09410b 1030static const struct address_space_operations btree_aops = {
d98237b3 1031 .readpage = btree_readpage,
0da5468f 1032 .writepages = btree_writepages,
5f39d397
CM		 1033 .releasepage = btree_releasepage,
1034 .invalidatepage = btree_invalidatepage,
5a92bc88 1035#ifdef CONFIG_MIGRATION
784b4e29 1036 .migratepage = btree_migratepage,
5a92bc88 1037#endif
0b32f4bb 1038 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1039};
1040
ca7a79ad
CM		 1041int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1042 u64 parent_transid)
090d1875 1043{
5f39d397
CM		 1044 struct extent_buffer *buf = NULL;
1045 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1046 int ret = 0;
090d1875 1047
db94535d 1048 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1049 if (!buf)
090d1875 1050 return 0;
d1310b2e 1051 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1052 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1053 free_extent_buffer(buf);
de428b63 1054 return ret;
090d1875
CM		 1055}
1056
ab0fff03
AJ		 1057int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1058 int mirror_num, struct extent_buffer **eb)
1059{
1060 struct extent_buffer *buf = NULL;
1061 struct inode *btree_inode = root->fs_info->btree_inode;
1062 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1063 int ret;
1064
1065 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1066 if (!buf)
1067 return 0;
1068
1069 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1070
1071 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1072 btree_get_extent, mirror_num);
1073 if (ret) {
1074 free_extent_buffer(buf);
1075 return ret;
1076 }
1077
1078 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1079 free_extent_buffer(buf);
1080 return -EIO;
0b32f4bb 1081 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1082 *eb = buf;
1083 } else {
1084 free_extent_buffer(buf);
1085 }
1086 return 0;
1087}
1088
0999df54
CM		 1089struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1090 u64 bytenr, u32 blocksize)
1091{
f28491e0 1092 return find_extent_buffer(root->fs_info, bytenr);
0999df54
CM		 1093}
1094
1095struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1096 u64 bytenr, u32 blocksize)
1097{
f28491e0 1098 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
0999df54
CM		 1099}
1100
1101
e02119d5
CM		 1102int btrfs_write_tree_block(struct extent_buffer *buf)
1103{
727011e0 1104 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
8aa38c31 1105 buf->start + buf->len - 1);
e02119d5
CM		 1106}
1107
1108int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1109{
727011e0 1110 return filemap_fdatawait_range(buf->pages[0]->mapping,
8aa38c31 1111 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1112}
1113
0999df54 1114struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1115 u32 blocksize, u64 parent_transid)
0999df54
CM		 1116{
1117 struct extent_buffer *buf = NULL;
0999df54
CM		 1118 int ret;
1119
0999df54
CM		 1120 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1121 if (!buf)
1122 return NULL;
0999df54 1123
ca7a79ad 1124 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
0f0fe8f7
FDBM		 1125 if (ret) {
1126 free_extent_buffer(buf);
1127 return NULL;
1128 }
5f39d397 1129 return buf;
ce9adaa5 1130
eb60ceac
CM		 1131}
1132
d5c13f92
JM		 1133void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1134 struct extent_buffer *buf)
ed2ff2cb 1135{
e2d84521
MX		 1136 struct btrfs_fs_info *fs_info = root->fs_info;
1137
55c69072 1138 if (btrfs_header_generation(buf) ==
e2d84521 1139 fs_info->running_transaction->transid) {
b9447ef8 1140 btrfs_assert_tree_locked(buf);
b4ce94de 1141
b9473439 1142 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1143 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1144 -buf->len,
1145 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1146 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1147 btrfs_set_lock_blocking(buf);
1148 clear_extent_buffer_dirty(buf);
1149 }
925baedd 1150 }
5f39d397
CM		 1151}
1152
143bede5
JM		 1153static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1154 u32 stripesize, struct btrfs_root *root,
1155 struct btrfs_fs_info *fs_info,
1156 u64 objectid)
d97e63b6 1157{
cfaa7295 1158 root->node = NULL;
a28ec197 1159 root->commit_root = NULL;
db94535d
CM		 1160 root->sectorsize = sectorsize;
1161 root->nodesize = nodesize;
1162 root->leafsize = leafsize;
87ee04eb 1163 root->stripesize = stripesize;
123abc88 1164 root->ref_cows = 0;
0b86a832 1165 root->track_dirty = 0;
c71bf099 1166 root->in_radix = 0;
d68fc57b
YZ		 1167 root->orphan_item_inserted = 0;
1168 root->orphan_cleanup_state = 0;
0b86a832 1169
0f7d52f4
CM		 1170 root->objectid = objectid;
1171 root->last_trans = 0;
13a8a7c8 1172 root->highest_objectid = 0;
eb73c1b7 1173 root->nr_delalloc_inodes = 0;
199c2a9c 1174 root->nr_ordered_extents = 0;
58176a96 1175 root->name = NULL;
6bef4d31 1176 root->inode_tree = RB_ROOT;
16cdcec7 1177 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1178 root->block_rsv = NULL;
d68fc57b 1179 root->orphan_block_rsv = NULL;
0b86a832
CM		 1180
1181 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1182 INIT_LIST_HEAD(&root->root_list);
eb73c1b7
MX		 1183 INIT_LIST_HEAD(&root->delalloc_inodes);
1184 INIT_LIST_HEAD(&root->delalloc_root);
199c2a9c
MX		 1185 INIT_LIST_HEAD(&root->ordered_extents);
1186 INIT_LIST_HEAD(&root->ordered_root);
2ab28f32
JB		 1187 INIT_LIST_HEAD(&root->logged_list[0]);
1188 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1189 spin_lock_init(&root->orphan_lock);
5d4f98a2 1190 spin_lock_init(&root->inode_lock);
eb73c1b7 1191 spin_lock_init(&root->delalloc_lock);
199c2a9c 1192 spin_lock_init(&root->ordered_extent_lock);
f0486c68 1193 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1194 spin_lock_init(&root->log_extents_lock[0]);
1195 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1196 mutex_init(&root->objectid_mutex);
e02119d5 1197 mutex_init(&root->log_mutex);
7237f183
YZ		 1198 init_waitqueue_head(&root->log_writer_wait);
1199 init_waitqueue_head(&root->log_commit_wait[0]);
1200 init_waitqueue_head(&root->log_commit_wait[1]);
8b050d35
MX		 1201 INIT_LIST_HEAD(&root->log_ctxs[0]);
1202 INIT_LIST_HEAD(&root->log_ctxs[1]);
7237f183
YZ		 1203 atomic_set(&root->log_commit[0], 0);
1204 atomic_set(&root->log_commit[1], 0);
1205 atomic_set(&root->log_writers, 0);
2ecb7923 1206 atomic_set(&root->log_batch, 0);
8a35d95f 1207 atomic_set(&root->orphan_inodes, 0);
b0feb9d9 1208 atomic_set(&root->refs, 1);
7237f183 1209 root->log_transid = 0;
d1433deb 1210 root->log_transid_committed = -1;
257c62e1 1211 root->last_log_commit = 0;
06ea65a3
JB		 1212 if (fs_info)
1213 extent_io_tree_init(&root->dirty_log_pages,
1214 fs_info->btree_inode->i_mapping);
017e5369 1215
3768f368
CM		 1216 memset(&root->root_key, 0, sizeof(root->root_key));
1217 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1218 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1219 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
06ea65a3
JB		 1220 if (fs_info)
1221 root->defrag_trans_start = fs_info->generation;
1222 else
1223 root->defrag_trans_start = 0;
58176a96 1224 init_completion(&root->kobj_unregister);
6702ed49 1225 root->defrag_running = 0;
4d775673 1226 root->root_key.objectid = objectid;
0ee5dc67 1227 root->anon_dev = 0;
8ea05e3a 1228
5f3ab90a 1229 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1230}
1231
f84a8bd6 1232static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1233{
1234 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1235 if (root)
1236 root->fs_info = fs_info;
1237 return root;
1238}
1239
06ea65a3
JB		 1240#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1241/* Should only be used by the testing infrastructure */
1242struct btrfs_root *btrfs_alloc_dummy_root(void)
1243{
1244 struct btrfs_root *root;
1245
1246 root = btrfs_alloc_root(NULL);
1247 if (!root)
1248 return ERR_PTR(-ENOMEM);
1249 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1250 root->dummy_root = 1;
1251
1252 return root;
1253}
1254#endif
1255
20897f5c
AJ		 1256struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1257 struct btrfs_fs_info *fs_info,
1258 u64 objectid)
1259{
1260 struct extent_buffer *leaf;
1261 struct btrfs_root *tree_root = fs_info->tree_root;
1262 struct btrfs_root *root;
1263 struct btrfs_key key;
1264 int ret = 0;
6463fe58 1265 uuid_le uuid;
20897f5c
AJ		 1266
1267 root = btrfs_alloc_root(fs_info);
1268 if (!root)
1269 return ERR_PTR(-ENOMEM);
1270
1271 __setup_root(tree_root->nodesize, tree_root->leafsize,
1272 tree_root->sectorsize, tree_root->stripesize,
1273 root, fs_info, objectid);
1274 root->root_key.objectid = objectid;
1275 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1276 root->root_key.offset = 0;
1277
1278 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1279 0, objectid, NULL, 0, 0, 0);
1280 if (IS_ERR(leaf)) {
1281 ret = PTR_ERR(leaf);
1dd05682 1282 leaf = NULL;
20897f5c
AJ		 1283 goto fail;
1284 }
1285
20897f5c
AJ		 1286 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1287 btrfs_set_header_bytenr(leaf, leaf->start);
1288 btrfs_set_header_generation(leaf, trans->transid);
1289 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1290 btrfs_set_header_owner(leaf, objectid);
1291 root->node = leaf;
1292
0a4e5586 1293 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
20897f5c
AJ		 1294 BTRFS_FSID_SIZE);
1295 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
b308bc2f 1296 btrfs_header_chunk_tree_uuid(leaf),
20897f5c
AJ		 1297 BTRFS_UUID_SIZE);
1298 btrfs_mark_buffer_dirty(leaf);
1299
1300 root->commit_root = btrfs_root_node(root);
1301 root->track_dirty = 1;
1302
1303
1304 root->root_item.flags = 0;
1305 root->root_item.byte_limit = 0;
1306 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1307 btrfs_set_root_generation(&root->root_item, trans->transid);
1308 btrfs_set_root_level(&root->root_item, 0);
1309 btrfs_set_root_refs(&root->root_item, 1);
1310 btrfs_set_root_used(&root->root_item, leaf->len);
1311 btrfs_set_root_last_snapshot(&root->root_item, 0);
1312 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1313 uuid_le_gen(&uuid);
1314 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1315 root->root_item.drop_level = 0;
1316
1317 key.objectid = objectid;
1318 key.type = BTRFS_ROOT_ITEM_KEY;
1319 key.offset = 0;
1320 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1321 if (ret)
1322 goto fail;
1323
1324 btrfs_tree_unlock(leaf);
1325
1dd05682
TI		 1326 return root;
1327
20897f5c 1328fail:
1dd05682
TI		 1329 if (leaf) {
1330 btrfs_tree_unlock(leaf);
1331 free_extent_buffer(leaf);
1332 }
1333 kfree(root);
20897f5c 1334
1dd05682 1335 return ERR_PTR(ret);
20897f5c
AJ		 1336}
1337
7237f183
YZ		 1338static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1339 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1340{
1341 struct btrfs_root *root;
1342 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1343 struct extent_buffer *leaf;
e02119d5 1344
6f07e42e 1345 root = btrfs_alloc_root(fs_info);
e02119d5 1346 if (!root)
7237f183 1347 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1348
1349 __setup_root(tree_root->nodesize, tree_root->leafsize,
1350 tree_root->sectorsize, tree_root->stripesize,
1351 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1352
1353 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1354 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1355 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1356 /*
1357 * log trees do not get reference counted because they go away
1358 * before a real commit is actually done. They do store pointers
1359 * to file data extents, and those reference counts still get
1360 * updated (along with back refs to the log tree).
1361 */
e02119d5
CM		 1362 root->ref_cows = 0;
1363
5d4f98a2 1364 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1365 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1366 0, 0, 0);
7237f183
YZ		 1367 if (IS_ERR(leaf)) {
1368 kfree(root);
1369 return ERR_CAST(leaf);
1370 }
e02119d5 1371
5d4f98a2
YZ		 1372 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1373 btrfs_set_header_bytenr(leaf, leaf->start);
1374 btrfs_set_header_generation(leaf, trans->transid);
1375 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1376 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1377 root->node = leaf;
e02119d5
CM		 1378
1379 write_extent_buffer(root->node, root->fs_info->fsid,
0a4e5586 1380 btrfs_header_fsid(), BTRFS_FSID_SIZE);
e02119d5
CM		 1381 btrfs_mark_buffer_dirty(root->node);
1382 btrfs_tree_unlock(root->node);
7237f183
YZ		 1383 return root;
1384}
1385
1386int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1387 struct btrfs_fs_info *fs_info)
1388{
1389 struct btrfs_root *log_root;
1390
1391 log_root = alloc_log_tree(trans, fs_info);
1392 if (IS_ERR(log_root))
1393 return PTR_ERR(log_root);
1394 WARN_ON(fs_info->log_root_tree);
1395 fs_info->log_root_tree = log_root;
1396 return 0;
1397}
1398
1399int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1400 struct btrfs_root *root)
1401{
1402 struct btrfs_root *log_root;
1403 struct btrfs_inode_item *inode_item;
1404
1405 log_root = alloc_log_tree(trans, root->fs_info);
1406 if (IS_ERR(log_root))
1407 return PTR_ERR(log_root);
1408
1409 log_root->last_trans = trans->transid;
1410 log_root->root_key.offset = root->root_key.objectid;
1411
1412 inode_item = &log_root->root_item.inode;
3cae210f
QW		 1413 btrfs_set_stack_inode_generation(inode_item, 1);
1414 btrfs_set_stack_inode_size(inode_item, 3);
1415 btrfs_set_stack_inode_nlink(inode_item, 1);
1416 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1417 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
7237f183 1418
5d4f98a2 1419 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1420
1421 WARN_ON(root->log_root);
1422 root->log_root = log_root;
1423 root->log_transid = 0;
d1433deb 1424 root->log_transid_committed = -1;
257c62e1 1425 root->last_log_commit = 0;
e02119d5
CM		 1426 return 0;
1427}
1428
35a3621b
SB		 1429static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1430 struct btrfs_key *key)
e02119d5
CM		 1431{
1432 struct btrfs_root *root;
1433 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1434 struct btrfs_path *path;
84234f3a 1435 u64 generation;
db94535d 1436 u32 blocksize;
cb517eab 1437 int ret;
0f7d52f4 1438
cb517eab
MX		 1439 path = btrfs_alloc_path();
1440 if (!path)
0f7d52f4 1441 return ERR_PTR(-ENOMEM);
cb517eab
MX		 1442
1443 root = btrfs_alloc_root(fs_info);
1444 if (!root) {
1445 ret = -ENOMEM;
1446 goto alloc_fail;
0f7d52f4
CM		 1447 }
1448
db94535d 1449 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb 1450 tree_root->sectorsize, tree_root->stripesize,
cb517eab 1451 root, fs_info, key->objectid);
0f7d52f4 1452
cb517eab
MX		 1453 ret = btrfs_find_root(tree_root, key, path,
1454 &root->root_item, &root->root_key);
0f7d52f4 1455 if (ret) {
13a8a7c8
YZ		 1456 if (ret > 0)
1457 ret = -ENOENT;
cb517eab 1458 goto find_fail;
0f7d52f4 1459 }
13a8a7c8 1460
84234f3a 1461 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1462 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1463 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1464 blocksize, generation);
cb517eab
MX		 1465 if (!root->node) {
1466 ret = -ENOMEM;
1467 goto find_fail;
1468 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1469 ret = -EIO;
1470 goto read_fail;
416bc658 1471 }
5d4f98a2 1472 root->commit_root = btrfs_root_node(root);
13a8a7c8 1473out:
cb517eab
MX		 1474 btrfs_free_path(path);
1475 return root;
1476
1477read_fail:
1478 free_extent_buffer(root->node);
1479find_fail:
1480 kfree(root);
1481alloc_fail:
1482 root = ERR_PTR(ret);
1483 goto out;
1484}
1485
1486struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1487 struct btrfs_key *location)
1488{
1489 struct btrfs_root *root;
1490
1491 root = btrfs_read_tree_root(tree_root, location);
1492 if (IS_ERR(root))
1493 return root;
1494
1495 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1496 root->ref_cows = 1;
08fe4db1
LZ		 1497 btrfs_check_and_init_root_item(&root->root_item);
1498 }
13a8a7c8 1499
5eda7b5e
CM		 1500 return root;
1501}
1502
cb517eab
MX		 1503int btrfs_init_fs_root(struct btrfs_root *root)
1504{
1505 int ret;
1506
1507 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1508 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1509 GFP_NOFS);
1510 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1511 ret = -ENOMEM;
1512 goto fail;
1513 }
1514
1515 btrfs_init_free_ino_ctl(root);
1516 mutex_init(&root->fs_commit_mutex);
1517 spin_lock_init(&root->cache_lock);
1518 init_waitqueue_head(&root->cache_wait);
1519
1520 ret = get_anon_bdev(&root->anon_dev);
1521 if (ret)
1522 goto fail;
1523 return 0;
1524fail:
1525 kfree(root->free_ino_ctl);
1526 kfree(root->free_ino_pinned);
1527 return ret;
1528}
1529
171170c1
ST		 1530static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1531 u64 root_id)
cb517eab
MX		 1532{
1533 struct btrfs_root *root;
1534
1535 spin_lock(&fs_info->fs_roots_radix_lock);
1536 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1537 (unsigned long)root_id);
1538 spin_unlock(&fs_info->fs_roots_radix_lock);
1539 return root;
1540}
1541
1542int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1543 struct btrfs_root *root)
1544{
1545 int ret;
1546
1547 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1548 if (ret)
1549 return ret;
1550
1551 spin_lock(&fs_info->fs_roots_radix_lock);
1552 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1553 (unsigned long)root->root_key.objectid,
1554 root);
1555 if (ret == 0)
1556 root->in_radix = 1;
1557 spin_unlock(&fs_info->fs_roots_radix_lock);
1558 radix_tree_preload_end();
1559
1560 return ret;
1561}
1562
c00869f1
MX		 1563struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1564 struct btrfs_key *location,
1565 bool check_ref)
5eda7b5e
CM		 1566{
1567 struct btrfs_root *root;
1568 int ret;
1569
edbd8d4e
CM		 1570 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1571 return fs_info->tree_root;
1572 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1573 return fs_info->extent_root;
8f18cf13
CM		 1574 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1575 return fs_info->chunk_root;
1576 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1577 return fs_info->dev_root;
0403e47e
YZ		 1578 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1579 return fs_info->csum_root;
bcef60f2
AJ		 1580 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1581 return fs_info->quota_root ? fs_info->quota_root :
1582 ERR_PTR(-ENOENT);
f7a81ea4
SB		 1583 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1584 return fs_info->uuid_root ? fs_info->uuid_root :
1585 ERR_PTR(-ENOENT);
4df27c4d 1586again:
cb517eab 1587 root = btrfs_lookup_fs_root(fs_info, location->objectid);
48475471 1588 if (root) {
c00869f1 1589 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
48475471 1590 return ERR_PTR(-ENOENT);
5eda7b5e 1591 return root;
48475471 1592 }
5eda7b5e 1593
cb517eab 1594 root = btrfs_read_fs_root(fs_info->tree_root, location);
5eda7b5e
CM		 1595 if (IS_ERR(root))
1596 return root;
3394e160 1597
c00869f1 1598 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
cb517eab 1599 ret = -ENOENT;
581bb050 1600 goto fail;
35a30d7c 1601 }
581bb050 1602
cb517eab 1603 ret = btrfs_init_fs_root(root);
ac08aedf
CM		 1604 if (ret)
1605 goto fail;
3394e160 1606
3f870c28
KN		 1607 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1608 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
d68fc57b
YZ		 1609 if (ret < 0)
1610 goto fail;
1611 if (ret == 0)
1612 root->orphan_item_inserted = 1;
1613
cb517eab 1614 ret = btrfs_insert_fs_root(fs_info, root);
0f7d52f4 1615 if (ret) {
4df27c4d
YZ		 1616 if (ret == -EEXIST) {
1617 free_fs_root(root);
1618 goto again;
1619 }
1620 goto fail;
0f7d52f4 1621 }
edbd8d4e 1622 return root;
4df27c4d
YZ		 1623fail:
1624 free_fs_root(root);
1625 return ERR_PTR(ret);
edbd8d4e
CM		 1626}
1627
04160088
CM		 1628static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1629{
1630 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1631 int ret = 0;
04160088
CM		 1632 struct btrfs_device *device;
1633 struct backing_dev_info *bdi;
b7967db7 1634
1f78160c
XG		 1635 rcu_read_lock();
1636 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1637 if (!device->bdev)
1638 continue;
04160088
CM		 1639 bdi = blk_get_backing_dev_info(device->bdev);
1640 if (bdi && bdi_congested(bdi, bdi_bits)) {
1641 ret = 1;
1642 break;
1643 }
1644 }
1f78160c 1645 rcu_read_unlock();
04160088
CM		 1646 return ret;
1647}
1648
ad081f14
JA		 1649/*
1650 * If this fails, caller must call bdi_destroy() to get rid of the
1651 * bdi again.
1652 */
04160088
CM		 1653static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1654{
ad081f14
JA		 1655 int err;
1656
1657 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1658 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1659 if (err)
1660 return err;
1661
4575c9cc 1662 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1663 bdi->congested_fn = btrfs_congested_fn;
1664 bdi->congested_data = info;
1665 return 0;
1666}
1667
8b712842
CM		 1668/*
1669 * called by the kthread helper functions to finally call the bio end_io
1670 * functions. This is where read checksum verification actually happens
1671 */
1672static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1673{
ce9adaa5 1674 struct bio *bio;
8b712842 1675 struct end_io_wq *end_io_wq;
ce9adaa5 1676 int error;
ce9adaa5 1677
8b712842
CM		 1678 end_io_wq = container_of(work, struct end_io_wq, work);
1679 bio = end_io_wq->bio;
ce9adaa5 1680
8b712842
CM		 1681 error = end_io_wq->error;
1682 bio->bi_private = end_io_wq->private;
1683 bio->bi_end_io = end_io_wq->end_io;
1684 kfree(end_io_wq);
8b712842 1685 bio_endio(bio, error);
44b8bd7e
CM		 1686}
1687
a74a4b97
CM		 1688static int cleaner_kthread(void *arg)
1689{
1690 struct btrfs_root *root = arg;
d0278245 1691 int again;
a74a4b97
CM		 1692
1693 do {
d0278245 1694 again = 0;
a74a4b97 1695
d0278245 1696 /* Make the cleaner go to sleep early. */
babbf170 1697 if (btrfs_need_cleaner_sleep(root))
d0278245
MX		 1698 goto sleep;
1699
1700 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1701 goto sleep;
1702
dc7f370c
MX		 1703 /*
1704 * Avoid the problem that we change the status of the fs
1705 * during the above check and trylock.
1706 */
babbf170 1707 if (btrfs_need_cleaner_sleep(root)) {
dc7f370c
MX		 1708 mutex_unlock(&root->fs_info->cleaner_mutex);
1709 goto sleep;
76dda93c 1710 }
a74a4b97 1711
d0278245
MX		 1712 btrfs_run_delayed_iputs(root);
1713 again = btrfs_clean_one_deleted_snapshot(root);
1714 mutex_unlock(&root->fs_info->cleaner_mutex);
1715
1716 /*
05323cd1
MX		 1717 * The defragger has dealt with the R/O remount and umount,
1718 * needn't do anything special here.
d0278245
MX		 1719 */
1720 btrfs_run_defrag_inodes(root->fs_info);
1721sleep:
9d1a2a3a 1722 if (!try_to_freeze() && !again) {
a74a4b97 1723 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1724 if (!kthread_should_stop())
1725 schedule();
a74a4b97
CM		 1726 __set_current_state(TASK_RUNNING);
1727 }
1728 } while (!kthread_should_stop());
1729 return 0;
1730}
1731
1732static int transaction_kthread(void *arg)
1733{
1734 struct btrfs_root *root = arg;
1735 struct btrfs_trans_handle *trans;
1736 struct btrfs_transaction *cur;
8929ecfa 1737 u64 transid;
a74a4b97
CM		 1738 unsigned long now;
1739 unsigned long delay;
914b2007 1740 bool cannot_commit;
a74a4b97
CM		 1741
1742 do {
914b2007 1743 cannot_commit = false;
8b87dc17 1744 delay = HZ * root->fs_info->commit_interval;
a74a4b97
CM		 1745 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1746
a4abeea4 1747 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1748 cur = root->fs_info->running_transaction;
1749 if (!cur) {
a4abeea4 1750 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1751 goto sleep;
1752 }
31153d81 1753
a74a4b97 1754 now = get_seconds();
4a9d8bde 1755 if (cur->state < TRANS_STATE_BLOCKED &&
8b87dc17
DS		 1756 (now < cur->start_time ||
1757 now - cur->start_time < root->fs_info->commit_interval)) {
a4abeea4 1758 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1759 delay = HZ * 5;
1760 goto sleep;
1761 }
8929ecfa 1762 transid = cur->transid;
a4abeea4 1763 spin_unlock(&root->fs_info->trans_lock);
56bec294 1764
79787eaa 1765 /* If the file system is aborted, this will always fail. */
354aa0fb 1766 trans = btrfs_attach_transaction(root);
914b2007 1767 if (IS_ERR(trans)) {
354aa0fb
MX		 1768 if (PTR_ERR(trans) != -ENOENT)
1769 cannot_commit = true;
79787eaa 1770 goto sleep;
914b2007 1771 }
8929ecfa 1772 if (transid == trans->transid) {
79787eaa 1773 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1774 } else {
1775 btrfs_end_transaction(trans, root);
1776 }
a74a4b97
CM		 1777sleep:
1778 wake_up_process(root->fs_info->cleaner_kthread);
1779 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1780
4e121c06
JB		 1781 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1782 &root->fs_info->fs_state)))
1783 btrfs_cleanup_transaction(root);
a0acae0e 1784 if (!try_to_freeze()) {
a74a4b97 1785 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1786 if (!kthread_should_stop() &&
914b2007
JK		 1787 (!btrfs_transaction_blocked(root->fs_info) ||
1788 cannot_commit))
8929ecfa 1789 schedule_timeout(delay);
a74a4b97
CM		 1790 __set_current_state(TASK_RUNNING);
1791 }
1792 } while (!kthread_should_stop());
1793 return 0;
1794}
1795
af31f5e5
CM		 1796/*
1797 * this will find the highest generation in the array of
1798 * root backups. The index of the highest array is returned,
1799 * or -1 if we can't find anything.
1800 *
1801 * We check to make sure the array is valid by comparing the
1802 * generation of the latest root in the array with the generation
1803 * in the super block. If they don't match we pitch it.
1804 */
1805static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1806{
1807 u64 cur;
1808 int newest_index = -1;
1809 struct btrfs_root_backup *root_backup;
1810 int i;
1811
1812 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1813 root_backup = info->super_copy->super_roots + i;
1814 cur = btrfs_backup_tree_root_gen(root_backup);
1815 if (cur == newest_gen)
1816 newest_index = i;
1817 }
1818
1819 /* check to see if we actually wrapped around */
1820 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1821 root_backup = info->super_copy->super_roots;
1822 cur = btrfs_backup_tree_root_gen(root_backup);
1823 if (cur == newest_gen)
1824 newest_index = 0;
1825 }
1826 return newest_index;
1827}
1828
1829
1830/*
1831 * find the oldest backup so we know where to store new entries
1832 * in the backup array. This will set the backup_root_index
1833 * field in the fs_info struct
1834 */
1835static void find_oldest_super_backup(struct btrfs_fs_info *info,
1836 u64 newest_gen)
1837{
1838 int newest_index = -1;
1839
1840 newest_index = find_newest_super_backup(info, newest_gen);
1841 /* if there was garbage in there, just move along */
1842 if (newest_index == -1) {
1843 info->backup_root_index = 0;
1844 } else {
1845 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1846 }
1847}
1848
1849/*
1850 * copy all the root pointers into the super backup array.
1851 * this will bump the backup pointer by one when it is
1852 * done
1853 */
1854static void backup_super_roots(struct btrfs_fs_info *info)
1855{
1856 int next_backup;
1857 struct btrfs_root_backup *root_backup;
1858 int last_backup;
1859
1860 next_backup = info->backup_root_index;
1861 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1862 BTRFS_NUM_BACKUP_ROOTS;
1863
1864 /*
1865 * just overwrite the last backup if we're at the same generation
1866 * this happens only at umount
1867 */
1868 root_backup = info->super_for_commit->super_roots + last_backup;
1869 if (btrfs_backup_tree_root_gen(root_backup) ==
1870 btrfs_header_generation(info->tree_root->node))
1871 next_backup = last_backup;
1872
1873 root_backup = info->super_for_commit->super_roots + next_backup;
1874
1875 /*
1876 * make sure all of our padding and empty slots get zero filled
1877 * regardless of which ones we use today
1878 */
1879 memset(root_backup, 0, sizeof(*root_backup));
1880
1881 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1882
1883 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1884 btrfs_set_backup_tree_root_gen(root_backup,
1885 btrfs_header_generation(info->tree_root->node));
1886
1887 btrfs_set_backup_tree_root_level(root_backup,
1888 btrfs_header_level(info->tree_root->node));
1889
1890 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1891 btrfs_set_backup_chunk_root_gen(root_backup,
1892 btrfs_header_generation(info->chunk_root->node));
1893 btrfs_set_backup_chunk_root_level(root_backup,
1894 btrfs_header_level(info->chunk_root->node));
1895
1896 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1897 btrfs_set_backup_extent_root_gen(root_backup,
1898 btrfs_header_generation(info->extent_root->node));
1899 btrfs_set_backup_extent_root_level(root_backup,
1900 btrfs_header_level(info->extent_root->node));
1901
7c7e82a7
CM		 1902 /*
1903 * we might commit during log recovery, which happens before we set
1904 * the fs_root. Make sure it is valid before we fill it in.
1905 */
1906 if (info->fs_root && info->fs_root->node) {
1907 btrfs_set_backup_fs_root(root_backup,
1908 info->fs_root->node->start);
1909 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1910 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1911 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1912 btrfs_header_level(info->fs_root->node));
7c7e82a7 1913 }
af31f5e5
CM		 1914
1915 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1916 btrfs_set_backup_dev_root_gen(root_backup,
1917 btrfs_header_generation(info->dev_root->node));
1918 btrfs_set_backup_dev_root_level(root_backup,
1919 btrfs_header_level(info->dev_root->node));
1920
1921 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1922 btrfs_set_backup_csum_root_gen(root_backup,
1923 btrfs_header_generation(info->csum_root->node));
1924 btrfs_set_backup_csum_root_level(root_backup,
1925 btrfs_header_level(info->csum_root->node));
1926
1927 btrfs_set_backup_total_bytes(root_backup,
1928 btrfs_super_total_bytes(info->super_copy));
1929 btrfs_set_backup_bytes_used(root_backup,
1930 btrfs_super_bytes_used(info->super_copy));
1931 btrfs_set_backup_num_devices(root_backup,
1932 btrfs_super_num_devices(info->super_copy));
1933
1934 /*
1935 * if we don't copy this out to the super_copy, it won't get remembered
1936 * for the next commit
1937 */
1938 memcpy(&info->super_copy->super_roots,
1939 &info->super_for_commit->super_roots,
1940 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1941}
1942
1943/*
1944 * this copies info out of the root backup array and back into
1945 * the in-memory super block. It is meant to help iterate through
1946 * the array, so you send it the number of backups you've already
1947 * tried and the last backup index you used.
1948 *
1949 * this returns -1 when it has tried all the backups
1950 */
1951static noinline int next_root_backup(struct btrfs_fs_info *info,
1952 struct btrfs_super_block *super,
1953 int *num_backups_tried, int *backup_index)
1954{
1955 struct btrfs_root_backup *root_backup;
1956 int newest = *backup_index;
1957
1958 if (*num_backups_tried == 0) {
1959 u64 gen = btrfs_super_generation(super);
1960
1961 newest = find_newest_super_backup(info, gen);
1962 if (newest == -1)
1963 return -1;
1964
1965 *backup_index = newest;
1966 *num_backups_tried = 1;
1967 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1968 /* we've tried all the backups, all done */
1969 return -1;
1970 } else {
1971 /* jump to the next oldest backup */
1972 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1973 BTRFS_NUM_BACKUP_ROOTS;
1974 *backup_index = newest;
1975 *num_backups_tried += 1;
1976 }
1977 root_backup = super->super_roots + newest;
1978
1979 btrfs_set_super_generation(super,
1980 btrfs_backup_tree_root_gen(root_backup));
1981 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1982 btrfs_set_super_root_level(super,
1983 btrfs_backup_tree_root_level(root_backup));
1984 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1985
1986 /*
1987 * fixme: the total bytes and num_devices need to match or we should
1988 * need a fsck
1989 */
1990 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1991 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1992 return 0;
1993}
1994
7abadb64
LB		 1995/* helper to cleanup workers */
1996static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1997{
1998 btrfs_stop_workers(&fs_info->generic_worker);
1999 btrfs_stop_workers(&fs_info->fixup_workers);
afe3d242 2000 btrfs_destroy_workqueue(fs_info->delalloc_workers);
5cdc7ad3 2001 btrfs_destroy_workqueue(fs_info->workers);
7abadb64
LB		 2002 btrfs_stop_workers(&fs_info->endio_workers);
2003 btrfs_stop_workers(&fs_info->endio_meta_workers);
2004 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2005 btrfs_stop_workers(&fs_info->rmw_workers);
2006 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2007 btrfs_stop_workers(&fs_info->endio_write_workers);
2008 btrfs_stop_workers(&fs_info->endio_freespace_worker);
a8c93d4e 2009 btrfs_destroy_workqueue(fs_info->submit_workers);
7abadb64
LB		 2010 btrfs_stop_workers(&fs_info->delayed_workers);
2011 btrfs_stop_workers(&fs_info->caching_workers);
2012 btrfs_stop_workers(&fs_info->readahead_workers);
a44903ab 2013 btrfs_destroy_workqueue(fs_info->flush_workers);
2f232036 2014 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
7abadb64
LB		 2015}
2016
2e9f5954
R		 2017static void free_root_extent_buffers(struct btrfs_root *root)
2018{
2019 if (root) {
2020 free_extent_buffer(root->node);
2021 free_extent_buffer(root->commit_root);
2022 root->node = NULL;
2023 root->commit_root = NULL;
2024 }
2025}
2026
af31f5e5
CM		 2027/* helper to cleanup tree roots */
2028static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2029{
2e9f5954 2030 free_root_extent_buffers(info->tree_root);
655b09fe 2031
2e9f5954
R		 2032 free_root_extent_buffers(info->dev_root);
2033 free_root_extent_buffers(info->extent_root);
2034 free_root_extent_buffers(info->csum_root);
2035 free_root_extent_buffers(info->quota_root);
2036 free_root_extent_buffers(info->uuid_root);
2037 if (chunk_root)
2038 free_root_extent_buffers(info->chunk_root);
af31f5e5
CM		 2039}
2040
171f6537
JB		 2041static void del_fs_roots(struct btrfs_fs_info *fs_info)
2042{
2043 int ret;
2044 struct btrfs_root *gang[8];
2045 int i;
2046
2047 while (!list_empty(&fs_info->dead_roots)) {
2048 gang[0] = list_entry(fs_info->dead_roots.next,
2049 struct btrfs_root, root_list);
2050 list_del(&gang[0]->root_list);
2051
2052 if (gang[0]->in_radix) {
cb517eab 2053 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
171f6537
JB		 2054 } else {
2055 free_extent_buffer(gang[0]->node);
2056 free_extent_buffer(gang[0]->commit_root);
b0feb9d9 2057 btrfs_put_fs_root(gang[0]);
171f6537
JB		 2058 }
2059 }
2060
2061 while (1) {
2062 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2063 (void **)gang, 0,
2064 ARRAY_SIZE(gang));
2065 if (!ret)
2066 break;
2067 for (i = 0; i < ret; i++)
cb517eab 2068 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
171f6537 2069 }
1a4319cc
LB		 2070
2071 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2072 btrfs_free_log_root_tree(NULL, fs_info);
2073 btrfs_destroy_pinned_extent(fs_info->tree_root,
2074 fs_info->pinned_extents);
2075 }
171f6537 2076}
af31f5e5 2077
ad2b2c80
AV		 2078int open_ctree(struct super_block *sb,
2079 struct btrfs_fs_devices *fs_devices,
2080 char *options)
2e635a27 2081{
db94535d
CM		 2082 u32 sectorsize;
2083 u32 nodesize;
2084 u32 leafsize;
2085 u32 blocksize;
87ee04eb 2086 u32 stripesize;
84234f3a 2087 u64 generation;
f2b636e8 2088 u64 features;
3de4586c 2089 struct btrfs_key location;
a061fc8d 2090 struct buffer_head *bh;
4d34b278 2091 struct btrfs_super_block *disk_super;
815745cf 2092 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2093 struct btrfs_root *tree_root;
4d34b278
ID		 2094 struct btrfs_root *extent_root;
2095 struct btrfs_root *csum_root;
2096 struct btrfs_root *chunk_root;
2097 struct btrfs_root *dev_root;
bcef60f2 2098 struct btrfs_root *quota_root;
f7a81ea4 2099 struct btrfs_root *uuid_root;
e02119d5 2100 struct btrfs_root *log_tree_root;
eb60ceac 2101 int ret;
e58ca020 2102 int err = -EINVAL;
af31f5e5
CM		 2103 int num_backups_tried = 0;
2104 int backup_index = 0;
5cdc7ad3
QW		 2105 int max_active;
2106 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
70f80175
SB		 2107 bool create_uuid_tree;
2108 bool check_uuid_tree;
4543df7e 2109
f84a8bd6 2110 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e 2111 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
cb517eab 2112 if (!tree_root || !chunk_root) {
39279cc3
CM		 2113 err = -ENOMEM;
2114 goto fail;
2115 }
76dda93c
YZ		 2116
2117 ret = init_srcu_struct(&fs_info->subvol_srcu);
2118 if (ret) {
2119 err = ret;
2120 goto fail;
2121 }
2122
2123 ret = setup_bdi(fs_info, &fs_info->bdi);
2124 if (ret) {
2125 err = ret;
2126 goto fail_srcu;
2127 }
2128
e2d84521
MX		 2129 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2130 if (ret) {
2131 err = ret;
2132 goto fail_bdi;
2133 }
2134 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2135 (1 + ilog2(nr_cpu_ids));
2136
963d678b
MX		 2137 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2138 if (ret) {
2139 err = ret;
2140 goto fail_dirty_metadata_bytes;
2141 }
2142
c404e0dc
MX		 2143 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2144 if (ret) {
2145 err = ret;
2146 goto fail_delalloc_bytes;
2147 }
2148
76dda93c
YZ		 2149 fs_info->btree_inode = new_inode(sb);
2150 if (!fs_info->btree_inode) {
2151 err = -ENOMEM;
c404e0dc 2152 goto fail_bio_counter;
76dda93c
YZ		 2153 }
2154
a6591715 2155 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2156
76dda93c 2157 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
f28491e0 2158 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
8fd17795 2159 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2160 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2161 INIT_LIST_HEAD(&fs_info->delayed_iputs);
eb73c1b7 2162 INIT_LIST_HEAD(&fs_info->delalloc_roots);
11833d66 2163 INIT_LIST_HEAD(&fs_info->caching_block_groups);
eb73c1b7 2164 spin_lock_init(&fs_info->delalloc_root_lock);
a4abeea4 2165 spin_lock_init(&fs_info->trans_lock);
76dda93c 2166 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2167 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2168 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2169 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2170 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2171 spin_lock_init(&fs_info->super_lock);
f28491e0 2172 spin_lock_init(&fs_info->buffer_lock);
f29021b2 2173 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2174 mutex_init(&fs_info->reloc_mutex);
de98ced9 2175 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2176
58176a96 2177 init_completion(&fs_info->kobj_unregister);
0b86a832 2178 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2179 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2180 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2181 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2182 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2183 BTRFS_BLOCK_RSV_GLOBAL);
2184 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2185 BTRFS_BLOCK_RSV_DELALLOC);
2186 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2187 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2188 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2189 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2190 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2191 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2192 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2193 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2194 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2195 atomic_set(&fs_info->defrag_running, 0);
fc36ed7e 2196 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2197 fs_info->sb = sb;
6f568d35 2198 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2199 fs_info->metadata_ratio = 0;
4cb5300b 2200 fs_info->defrag_inodes = RB_ROOT;
2bf64758 2201 fs_info->free_chunk_space = 0;
f29021b2 2202 fs_info->tree_mod_log = RB_ROOT;
8b87dc17 2203 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
0a2b2a84 2204 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
90519d66
AJ		 2205 /* readahead state */
2206 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2207 spin_lock_init(&fs_info->reada_lock);
c8b97818 2208
b34b086c
CM		 2209 fs_info->thread_pool_size = min_t(unsigned long,
2210 num_online_cpus() + 2, 8);
0afbaf8c 2211
199c2a9c
MX		 2212 INIT_LIST_HEAD(&fs_info->ordered_roots);
2213 spin_lock_init(&fs_info->ordered_root_lock);
16cdcec7
MX		 2214 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2215 GFP_NOFS);
2216 if (!fs_info->delayed_root) {
2217 err = -ENOMEM;
2218 goto fail_iput;
2219 }
2220 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2221
a2de733c
AJ		 2222 mutex_init(&fs_info->scrub_lock);
2223 atomic_set(&fs_info->scrubs_running, 0);
2224 atomic_set(&fs_info->scrub_pause_req, 0);
2225 atomic_set(&fs_info->scrubs_paused, 0);
2226 atomic_set(&fs_info->scrub_cancel_req, 0);
c404e0dc 2227 init_waitqueue_head(&fs_info->replace_wait);
a2de733c 2228 init_waitqueue_head(&fs_info->scrub_pause_wait);
a2de733c 2229 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2230#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2231 fs_info->check_integrity_print_mask = 0;
2232#endif
a2de733c 2233
c9e9f97b
ID		 2234 spin_lock_init(&fs_info->balance_lock);
2235 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2236 atomic_set(&fs_info->balance_running, 0);
2237 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2238 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2239 fs_info->balance_ctl = NULL;
837d5b6e 2240 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2241
a061fc8d
CM		 2242 sb->s_blocksize = 4096;
2243 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2244 sb->s_bdi = &fs_info->bdi;
a061fc8d 2245
76dda93c 2246 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2247 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2248 /*
2249 * we set the i_size on the btree inode to the max possible int.
2250 * the real end of the address space is determined by all of
2251 * the devices in the system
2252 */
2253 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2254 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2255 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2256
5d4f98a2 2257 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2258 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2259 fs_info->btree_inode->i_mapping);
0b32f4bb 2260 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2261 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2262
2263 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2264
76dda93c
YZ		 2265 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2266 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2267 sizeof(struct btrfs_key));
72ac3c0d
JB		 2268 set_bit(BTRFS_INODE_DUMMY,
2269 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
778ba82b 2270 btrfs_insert_inode_hash(fs_info->btree_inode);
76dda93c 2271
0f9dd46c 2272 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2273 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2274 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2275
11833d66 2276 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2277 fs_info->btree_inode->i_mapping);
11833d66 2278 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2279 fs_info->btree_inode->i_mapping);
11833d66 2280 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2281 fs_info->do_barriers = 1;
e18e4809 2282
39279cc3 2283
5a3f23d5 2284 mutex_init(&fs_info->ordered_operations_mutex);
9ffba8cd 2285 mutex_init(&fs_info->ordered_extent_flush_mutex);
e02119d5 2286 mutex_init(&fs_info->tree_log_mutex);
925baedd 2287 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2288 mutex_init(&fs_info->transaction_kthread_mutex);
2289 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2290 mutex_init(&fs_info->volume_mutex);
276e680d 2291 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2292 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2293 init_rwsem(&fs_info->subvol_sem);
803b2f54 2294 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
e922e087
SB		 2295 fs_info->dev_replace.lock_owner = 0;
2296 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2297 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2298 mutex_init(&fs_info->dev_replace.lock_management_lock);
2299 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2300
416ac51d 2301 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2302 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2303 fs_info->qgroup_tree = RB_ROOT;
2304 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2305 fs_info->qgroup_seq = 1;
2306 fs_info->quota_enabled = 0;
2307 fs_info->pending_quota_state = 0;
1e8f9158 2308 fs_info->qgroup_ulist = NULL;
2f232036 2309 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2310
fa9c0d79
CM		 2311 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2312 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2313
e6dcd2dc 2314 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2315 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2316 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2317 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2318
53b381b3
DW		 2319 ret = btrfs_alloc_stripe_hash_table(fs_info);
2320 if (ret) {
83c8266a 2321 err = ret;
53b381b3
DW		 2322 goto fail_alloc;
2323 }
2324
0b86a832 2325 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2326 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2327
3c4bb26b 2328 invalidate_bdev(fs_devices->latest_bdev);
1104a885
DS		 2329
2330 /*
2331 * Read super block and check the signature bytes only
2332 */
a512bbf8 2333 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2334 if (!bh) {
2335 err = -EINVAL;
16cdcec7 2336 goto fail_alloc;
20b45077 2337 }
39279cc3 2338
1104a885
DS		 2339 /*
2340 * We want to check superblock checksum, the type is stored inside.
2341 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2342 */
2343 if (btrfs_check_super_csum(bh->b_data)) {
efe120a0 2344 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
1104a885
DS		 2345 err = -EINVAL;
2346 goto fail_alloc;
2347 }
2348
2349 /*
2350 * super_copy is zeroed at allocation time and we never touch the
2351 * following bytes up to INFO_SIZE, the checksum is calculated from
2352 * the whole block of INFO_SIZE
2353 */
6c41761f
DS		 2354 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2355 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2356 sizeof(*fs_info->super_for_commit));
a061fc8d 2357 brelse(bh);
5f39d397 2358
6c41761f 2359 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2360
1104a885
DS		 2361 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2362 if (ret) {
efe120a0 2363 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
1104a885
DS		 2364 err = -EINVAL;
2365 goto fail_alloc;
2366 }
2367
6c41761f 2368 disk_super = fs_info->super_copy;
0f7d52f4 2369 if (!btrfs_super_root(disk_super))
16cdcec7 2370 goto fail_alloc;
0f7d52f4 2371
acce952b 2372 /* check FS state, whether FS is broken. */
87533c47
MX		 2373 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2374 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2375
af31f5e5
CM		 2376 /*
2377 * run through our array of backup supers and setup
2378 * our ring pointer to the oldest one
2379 */
2380 generation = btrfs_super_generation(disk_super);
2381 find_oldest_super_backup(fs_info, generation);
2382
75e7cb7f
LB		 2383 /*
2384 * In the long term, we'll store the compression type in the super
2385 * block, and it'll be used for per file compression control.
2386 */
2387 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2388
2b82032c
YZ		 2389 ret = btrfs_parse_options(tree_root, options);
2390 if (ret) {
2391 err = ret;
16cdcec7 2392 goto fail_alloc;
2b82032c 2393 }
dfe25020 2394
f2b636e8
JB		 2395 features = btrfs_super_incompat_flags(disk_super) &
2396 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2397 if (features) {
2398 printk(KERN_ERR "BTRFS: couldn't mount because of "
2399 "unsupported optional features (%Lx).\n",
c1c9ff7c 2400 features);
f2b636e8 2401 err = -EINVAL;
16cdcec7 2402 goto fail_alloc;
f2b636e8
JB		 2403 }
2404
727011e0
CM		 2405 if (btrfs_super_leafsize(disk_super) !=
2406 btrfs_super_nodesize(disk_super)) {
2407 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2408 "blocksizes don't match. node %d leaf %d\n",
2409 btrfs_super_nodesize(disk_super),
2410 btrfs_super_leafsize(disk_super));
2411 err = -EINVAL;
2412 goto fail_alloc;
2413 }
2414 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2415 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2416 "blocksize (%d) was too large\n",
2417 btrfs_super_leafsize(disk_super));
2418 err = -EINVAL;
2419 goto fail_alloc;
2420 }
2421
5d4f98a2 2422 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2423 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2424 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2425 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2426
3173a18f 2427 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
efe120a0 2428 printk(KERN_ERR "BTRFS: has skinny extents\n");
3173a18f 2429
727011e0
CM		 2430 /*
2431 * flag our filesystem as having big metadata blocks if
2432 * they are bigger than the page size
2433 */
2434 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2435 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
efe120a0 2436 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
727011e0
CM		 2437 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2438 }
2439
bc3f116f
CM		 2440 nodesize = btrfs_super_nodesize(disk_super);
2441 leafsize = btrfs_super_leafsize(disk_super);
2442 sectorsize = btrfs_super_sectorsize(disk_super);
2443 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2444 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2445 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2446
2447 /*
2448 * mixed block groups end up with duplicate but slightly offset
2449 * extent buffers for the same range. It leads to corruptions
2450 */
2451 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2452 (sectorsize != leafsize)) {
efe120a0 2453 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
bc3f116f
CM		 2454 "are not allowed for mixed block groups on %s\n",
2455 sb->s_id);
2456 goto fail_alloc;
2457 }
2458
ceda0864
MX		 2459 /*
2460 * Needn't use the lock because there is no other task which will
2461 * update the flag.
2462 */
a6fa6fae 2463 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2464
f2b636e8
JB		 2465 features = btrfs_super_compat_ro_flags(disk_super) &
2466 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2467 if (!(sb->s_flags & MS_RDONLY) && features) {
2468 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2469 "unsupported option features (%Lx).\n",
c1c9ff7c 2470 features);
f2b636e8 2471 err = -EINVAL;
16cdcec7 2472 goto fail_alloc;
f2b636e8 2473 }
61d92c32 2474
5cdc7ad3 2475 max_active = fs_info->thread_pool_size;
61d92c32
CM		 2476 btrfs_init_workers(&fs_info->generic_worker,
2477 "genwork", 1, NULL);
2478
5cdc7ad3
QW		 2479 fs_info->workers =
2480 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2481 max_active, 16);
c8b97818 2482
afe3d242
QW		 2483 fs_info->delalloc_workers =
2484 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
771ed689 2485
a44903ab
QW		 2486 fs_info->flush_workers =
2487 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
61b49440 2488
bab39bf9 2489 btrfs_init_workers(&fs_info->caching_workers, "cache",
45d5fd14 2490 fs_info->thread_pool_size, NULL);
bab39bf9 2491
a8c93d4e
QW		 2492 /*
2493 * a higher idle thresh on the submit workers makes it much more
61b49440
CM		 2494 * likely that bios will be send down in a sane order to the
2495 * devices
2496 */
a8c93d4e
QW		 2497 fs_info->submit_workers =
2498 btrfs_alloc_workqueue("submit", flags,
2499 min_t(u64, fs_devices->num_devices,
2500 max_active), 64);
53863232 2501
61d92c32
CM		 2502 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2503 &fs_info->generic_worker);
5443be45 2504 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2505 fs_info->thread_pool_size,
2506 &fs_info->generic_worker);
d20f7043 2507 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2508 fs_info->thread_pool_size,
2509 &fs_info->generic_worker);
cad321ad 2510 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2511 "endio-meta-write", fs_info->thread_pool_size,
2512 &fs_info->generic_worker);
53b381b3
DW		 2513 btrfs_init_workers(&fs_info->endio_raid56_workers,
2514 "endio-raid56", fs_info->thread_pool_size,
2515 &fs_info->generic_worker);
2516 btrfs_init_workers(&fs_info->rmw_workers,
2517 "rmw", fs_info->thread_pool_size,
2518 &fs_info->generic_worker);
5443be45 2519 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2520 fs_info->thread_pool_size,
2521 &fs_info->generic_worker);
0cb59c99
JB		 2522 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2523 1, &fs_info->generic_worker);
16cdcec7
MX		 2524 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2525 fs_info->thread_pool_size,
2526 &fs_info->generic_worker);
90519d66
AJ		 2527 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2528 fs_info->thread_pool_size,
2529 &fs_info->generic_worker);
2f232036
JS		 2530 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2531 &fs_info->generic_worker);
61b49440
CM		 2532
2533 /*
2534 * endios are largely parallel and should have a very
2535 * low idle thresh
2536 */
2537 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2538 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2539 fs_info->endio_raid56_workers.idle_thresh = 4;
2540 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2541
9042846b
CM		 2542 fs_info->endio_write_workers.idle_thresh = 2;
2543 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2544 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2545
0dc3b84a
JB		 2546 /*
2547 * btrfs_start_workers can really only fail because of ENOMEM so just
2548 * return -ENOMEM if any of these fail.
2549 */
5cdc7ad3 2550 ret = btrfs_start_workers(&fs_info->generic_worker);
0dc3b84a
JB		 2551 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2552 ret |= btrfs_start_workers(&fs_info->endio_workers);
2553 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2554 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2555 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2556 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2557 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2558 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2559 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2560 ret |= btrfs_start_workers(&fs_info->caching_workers);
2561 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2f232036 2562 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
0dc3b84a 2563 if (ret) {
fed425c7 2564 err = -ENOMEM;
0dc3b84a
JB		 2565 goto fail_sb_buffer;
2566 }
a8c93d4e 2567 if (!(fs_info->workers && fs_info->delalloc_workers &&
a44903ab 2568 fs_info->submit_workers && fs_info->flush_workers)) {
5cdc7ad3
QW		 2569 err = -ENOMEM;
2570 goto fail_sb_buffer;
2571 }
4543df7e 2572
4575c9cc 2573 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2574 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2575 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2576
db94535d
CM		 2577 tree_root->nodesize = nodesize;
2578 tree_root->leafsize = leafsize;
2579 tree_root->sectorsize = sectorsize;
87ee04eb 2580 tree_root->stripesize = stripesize;
a061fc8d
CM		 2581
2582 sb->s_blocksize = sectorsize;
2583 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2584
3cae210f 2585 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
efe120a0 2586 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2587 goto fail_sb_buffer;
2588 }
19c00ddc 2589
8d082fb7 2590 if (sectorsize != PAGE_SIZE) {
efe120a0 2591 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
8d082fb7 2592 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2593 goto fail_sb_buffer;
2594 }
2595
925baedd 2596 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2597 ret = btrfs_read_sys_array(tree_root);
925baedd 2598 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2599 if (ret) {
efe120a0 2600 printk(KERN_WARNING "BTRFS: failed to read the system "
d397712b 2601 "array on %s\n", sb->s_id);
5d4f98a2 2602 goto fail_sb_buffer;
84eed90f 2603 }
0b86a832
CM		 2604
2605 blocksize = btrfs_level_size(tree_root,
2606 btrfs_super_chunk_root_level(disk_super));
84234f3a 2607 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2608
2609 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2610 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2611
2612 chunk_root->node = read_tree_block(chunk_root,
2613 btrfs_super_chunk_root(disk_super),
84234f3a 2614 blocksize, generation);
416bc658
JB		 2615 if (!chunk_root->node ||
2616 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
efe120a0 2617 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
83121942 2618 sb->s_id);
af31f5e5 2619 goto fail_tree_roots;
83121942 2620 }
5d4f98a2
YZ		 2621 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2622 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2623
e17cade2 2624 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
b308bc2f 2625 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
e17cade2 2626
0b86a832 2627 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2628 if (ret) {
efe120a0 2629 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
d397712b 2630 sb->s_id);
af31f5e5 2631 goto fail_tree_roots;
2b82032c 2632 }
0b86a832 2633
8dabb742
SB		 2634 /*
2635 * keep the device that is marked to be the target device for the
2636 * dev_replace procedure
2637 */
2638 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2639
a6b0d5c8 2640 if (!fs_devices->latest_bdev) {
efe120a0 2641 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
a6b0d5c8
CM		 2642 sb->s_id);
2643 goto fail_tree_roots;
2644 }
2645
af31f5e5 2646retry_root_backup:
db94535d
CM		 2647 blocksize = btrfs_level_size(tree_root,
2648 btrfs_super_root_level(disk_super));
84234f3a 2649 generation = btrfs_super_generation(disk_super);
0b86a832 2650
e20d96d6 2651 tree_root->node = read_tree_block(tree_root,
db94535d 2652 btrfs_super_root(disk_super),
84234f3a 2653 blocksize, generation);
af31f5e5
CM		 2654 if (!tree_root->node ||
2655 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
efe120a0 2656 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
83121942 2657 sb->s_id);
af31f5e5
CM		 2658
2659 goto recovery_tree_root;
83121942 2660 }
af31f5e5 2661
5d4f98a2
YZ		 2662 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2663 tree_root->commit_root = btrfs_root_node(tree_root);
69e9c6c6 2664 btrfs_set_root_refs(&tree_root->root_item, 1);
db94535d 2665
cb517eab
MX		 2666 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2667 location.type = BTRFS_ROOT_ITEM_KEY;
2668 location.offset = 0;
2669
2670 extent_root = btrfs_read_tree_root(tree_root, &location);
2671 if (IS_ERR(extent_root)) {
2672 ret = PTR_ERR(extent_root);
af31f5e5 2673 goto recovery_tree_root;
cb517eab 2674 }
0b86a832 2675 extent_root->track_dirty = 1;
cb517eab 2676 fs_info->extent_root = extent_root;
0b86a832 2677
cb517eab
MX		 2678 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2679 dev_root = btrfs_read_tree_root(tree_root, &location);
2680 if (IS_ERR(dev_root)) {
2681 ret = PTR_ERR(dev_root);
af31f5e5 2682 goto recovery_tree_root;
cb517eab 2683 }
5d4f98a2 2684 dev_root->track_dirty = 1;
cb517eab
MX		 2685 fs_info->dev_root = dev_root;
2686 btrfs_init_devices_late(fs_info);
3768f368 2687
cb517eab
MX		 2688 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2689 csum_root = btrfs_read_tree_root(tree_root, &location);
2690 if (IS_ERR(csum_root)) {
2691 ret = PTR_ERR(csum_root);
af31f5e5 2692 goto recovery_tree_root;
cb517eab 2693 }
d20f7043 2694 csum_root->track_dirty = 1;
cb517eab 2695 fs_info->csum_root = csum_root;
d20f7043 2696
cb517eab
MX		 2697 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2698 quota_root = btrfs_read_tree_root(tree_root, &location);
2699 if (!IS_ERR(quota_root)) {
bcef60f2
AJ		 2700 quota_root->track_dirty = 1;
2701 fs_info->quota_enabled = 1;
2702 fs_info->pending_quota_state = 1;
cb517eab 2703 fs_info->quota_root = quota_root;
bcef60f2
AJ		 2704 }
2705
f7a81ea4
SB		 2706 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2707 uuid_root = btrfs_read_tree_root(tree_root, &location);
2708 if (IS_ERR(uuid_root)) {
2709 ret = PTR_ERR(uuid_root);
2710 if (ret != -ENOENT)
2711 goto recovery_tree_root;
2712 create_uuid_tree = true;
70f80175 2713 check_uuid_tree = false;
f7a81ea4
SB		 2714 } else {
2715 uuid_root->track_dirty = 1;
2716 fs_info->uuid_root = uuid_root;
70f80175
SB		 2717 create_uuid_tree = false;
2718 check_uuid_tree =
2719 generation != btrfs_super_uuid_tree_generation(disk_super);
f7a81ea4
SB		 2720 }
2721
8929ecfa
YZ		 2722 fs_info->generation = generation;
2723 fs_info->last_trans_committed = generation;
8929ecfa 2724
68310a5e
ID		 2725 ret = btrfs_recover_balance(fs_info);
2726 if (ret) {
efe120a0 2727 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
68310a5e
ID		 2728 goto fail_block_groups;
2729 }
2730
733f4fbb
SB		 2731 ret = btrfs_init_dev_stats(fs_info);
2732 if (ret) {
efe120a0 2733 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
733f4fbb
SB		 2734 ret);
2735 goto fail_block_groups;
2736 }
2737
8dabb742
SB		 2738 ret = btrfs_init_dev_replace(fs_info);
2739 if (ret) {
efe120a0 2740 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
8dabb742
SB		 2741 goto fail_block_groups;
2742 }
2743
2744 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2745
5ac1d209
JM		 2746 ret = btrfs_sysfs_add_one(fs_info);
2747 if (ret) {
efe120a0 2748 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
5ac1d209
JM		 2749 goto fail_block_groups;
2750 }
2751
c59021f8 2752 ret = btrfs_init_space_info(fs_info);
2753 if (ret) {
efe120a0 2754 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2365dd3c 2755 goto fail_sysfs;
c59021f8 2756 }
2757
1b1d1f66
JB		 2758 ret = btrfs_read_block_groups(extent_root);
2759 if (ret) {
efe120a0 2760 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2365dd3c 2761 goto fail_sysfs;
1b1d1f66 2762 }
5af3e8cc
SB		 2763 fs_info->num_tolerated_disk_barrier_failures =
2764 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2765 if (fs_info->fs_devices->missing_devices >
2766 fs_info->num_tolerated_disk_barrier_failures &&
2767 !(sb->s_flags & MS_RDONLY)) {
efe120a0
FH		 2768 printk(KERN_WARNING "BTRFS: "
2769 "too many missing devices, writeable mount is not allowed\n");
2365dd3c 2770 goto fail_sysfs;
292fd7fc 2771 }
9078a3e1 2772
a74a4b97
CM		 2773 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2774 "btrfs-cleaner");
57506d50 2775 if (IS_ERR(fs_info->cleaner_kthread))
2365dd3c 2776 goto fail_sysfs;
a74a4b97
CM		 2777
2778 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2779 tree_root,
2780 "btrfs-transaction");
57506d50 2781 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2782 goto fail_cleaner;
a74a4b97 2783
c289811c
CM		 2784 if (!btrfs_test_opt(tree_root, SSD) &&
2785 !btrfs_test_opt(tree_root, NOSSD) &&
2786 !fs_info->fs_devices->rotating) {
efe120a0 2787 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
c289811c
CM		 2788 "mode\n");
2789 btrfs_set_opt(fs_info->mount_opt, SSD);
2790 }
2791
3818aea2
QW		 2792 /* Set the real inode map cache flag */
2793 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2794 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2795
21adbd5c
SB		 2796#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2797 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2798 ret = btrfsic_mount(tree_root, fs_devices,
2799 btrfs_test_opt(tree_root,
2800 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2801 1 : 0,
2802 fs_info->check_integrity_print_mask);
2803 if (ret)
efe120a0 2804 printk(KERN_WARNING "BTRFS: failed to initialize"
21adbd5c
SB		 2805 " integrity check module %s\n", sb->s_id);
2806 }
2807#endif
bcef60f2
AJ		 2808 ret = btrfs_read_qgroup_config(fs_info);
2809 if (ret)
2810 goto fail_trans_kthread;
21adbd5c 2811
acce952b 2812 /* do not make disk changes in broken FS */
68ce9682 2813 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2814 u64 bytenr = btrfs_super_log_root(disk_super);
2815
7c2ca468 2816 if (fs_devices->rw_devices == 0) {
efe120a0 2817 printk(KERN_WARNING "BTRFS: log replay required "
d397712b 2818 "on RO media\n");
7c2ca468 2819 err = -EIO;
bcef60f2 2820 goto fail_qgroup;
7c2ca468 2821 }
e02119d5
CM		 2822 blocksize =
2823 btrfs_level_size(tree_root,
2824 btrfs_super_log_root_level(disk_super));
d18a2c44 2825
6f07e42e 2826 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2827 if (!log_tree_root) {
2828 err = -ENOMEM;
bcef60f2 2829 goto fail_qgroup;
676e4c86 2830 }
e02119d5
CM		 2831
2832 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2833 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2834
2835 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2836 blocksize,
2837 generation + 1);
416bc658
JB		 2838 if (!log_tree_root->node ||
2839 !extent_buffer_uptodate(log_tree_root->node)) {
efe120a0 2840 printk(KERN_ERR "BTRFS: failed to read log tree\n");
416bc658
JB		 2841 free_extent_buffer(log_tree_root->node);
2842 kfree(log_tree_root);
2843 goto fail_trans_kthread;
2844 }
79787eaa 2845 /* returns with log_tree_root freed on success */
e02119d5 2846 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2847 if (ret) {
2848 btrfs_error(tree_root->fs_info, ret,
2849 "Failed to recover log tree");
2850 free_extent_buffer(log_tree_root->node);
2851 kfree(log_tree_root);
2852 goto fail_trans_kthread;
2853 }
e556ce2c
YZ		 2854
2855 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2856 ret = btrfs_commit_super(tree_root);
2857 if (ret)
2858 goto fail_trans_kthread;
e556ce2c 2859 }
e02119d5 2860 }
1a40e23b 2861
76dda93c 2862 ret = btrfs_find_orphan_roots(tree_root);
79787eaa
JM		 2863 if (ret)
2864 goto fail_trans_kthread;
76dda93c 2865
7c2ca468 2866 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2867 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2
ID		 2868 if (ret)
2869 goto fail_trans_kthread;
d68fc57b 2870
5d4f98a2 2871 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2872 if (ret < 0) {
2873 printk(KERN_WARNING
efe120a0 2874 "BTRFS: failed to recover relocation\n");
d7ce5843 2875 err = -EINVAL;
bcef60f2 2876 goto fail_qgroup;
d7ce5843 2877 }
7c2ca468 2878 }
1a40e23b 2879
3de4586c
CM		 2880 location.objectid = BTRFS_FS_TREE_OBJECTID;
2881 location.type = BTRFS_ROOT_ITEM_KEY;
cb517eab 2882 location.offset = 0;
3de4586c 2883
3de4586c 2884 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3140c9a3
DC		 2885 if (IS_ERR(fs_info->fs_root)) {
2886 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2887 goto fail_qgroup;
3140c9a3 2888 }
c289811c 2889
2b6ba629
ID		 2890 if (sb->s_flags & MS_RDONLY)
2891 return 0;
59641015 2892
2b6ba629
ID		 2893 down_read(&fs_info->cleanup_work_sem);
2894 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2895 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2896 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2897 close_ctree(tree_root);
2898 return ret;
2899 }
2900 up_read(&fs_info->cleanup_work_sem);
59641015 2901
2b6ba629
ID		 2902 ret = btrfs_resume_balance_async(fs_info);
2903 if (ret) {
efe120a0 2904 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2b6ba629
ID		 2905 close_ctree(tree_root);
2906 return ret;
e3acc2a6
JB		 2907 }
2908
8dabb742
SB		 2909 ret = btrfs_resume_dev_replace_async(fs_info);
2910 if (ret) {
efe120a0 2911 pr_warn("BTRFS: failed to resume dev_replace\n");
8dabb742
SB		 2912 close_ctree(tree_root);
2913 return ret;
2914 }
2915
b382a324
JS		 2916 btrfs_qgroup_rescan_resume(fs_info);
2917
f7a81ea4 2918 if (create_uuid_tree) {
efe120a0 2919 pr_info("BTRFS: creating UUID tree\n");
f7a81ea4
SB		 2920 ret = btrfs_create_uuid_tree(fs_info);
2921 if (ret) {
efe120a0 2922 pr_warn("BTRFS: failed to create the UUID tree %d\n",
f7a81ea4
SB		 2923 ret);
2924 close_ctree(tree_root);
2925 return ret;
2926 }
f420ee1e
SB		 2927 } else if (check_uuid_tree ||
2928 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
efe120a0 2929 pr_info("BTRFS: checking UUID tree\n");
70f80175
SB		 2930 ret = btrfs_check_uuid_tree(fs_info);
2931 if (ret) {
efe120a0 2932 pr_warn("BTRFS: failed to check the UUID tree %d\n",
70f80175
SB		 2933 ret);
2934 close_ctree(tree_root);
2935 return ret;
2936 }
2937 } else {
2938 fs_info->update_uuid_tree_gen = 1;
f7a81ea4
SB		 2939 }
2940
ad2b2c80 2941 return 0;
39279cc3 2942
bcef60f2
AJ		 2943fail_qgroup:
2944 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2945fail_trans_kthread:
2946 kthread_stop(fs_info->transaction_kthread);
54067ae9 2947 btrfs_cleanup_transaction(fs_info->tree_root);
7b5ff90e 2948 del_fs_roots(fs_info);
3f157a2f 2949fail_cleaner:
a74a4b97 2950 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2951
2952 /*
2953 * make sure we're done with the btree inode before we stop our
2954 * kthreads
2955 */
2956 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2957
2365dd3c
AJ		 2958fail_sysfs:
2959 btrfs_sysfs_remove_one(fs_info);
2960
1b1d1f66 2961fail_block_groups:
54067ae9 2962 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2963 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2964
2965fail_tree_roots:
2966 free_root_pointers(fs_info, 1);
2b8195bb 2967 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2968
39279cc3 2969fail_sb_buffer:
7abadb64 2970 btrfs_stop_all_workers(fs_info);
16cdcec7 2971fail_alloc:
4543df7e 2972fail_iput:
586e46e2
ID		 2973 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2974
4543df7e 2975 iput(fs_info->btree_inode);
c404e0dc
MX		 2976fail_bio_counter:
2977 percpu_counter_destroy(&fs_info->bio_counter);
963d678b
MX		 2978fail_delalloc_bytes:
2979 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 2980fail_dirty_metadata_bytes:
2981 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 2982fail_bdi:
7e662854 2983 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2984fail_srcu:
2985 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2986fail:
53b381b3 2987 btrfs_free_stripe_hash_table(fs_info);
586e46e2 2988 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 2989 return err;
af31f5e5
CM		 2990
2991recovery_tree_root:
af31f5e5
CM		 2992 if (!btrfs_test_opt(tree_root, RECOVERY))
2993 goto fail_tree_roots;
2994
2995 free_root_pointers(fs_info, 0);
2996
2997 /* don't use the log in recovery mode, it won't be valid */
2998 btrfs_set_super_log_root(disk_super, 0);
2999
3000 /* we can't trust the free space cache either */
3001 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3002
3003 ret = next_root_backup(fs_info, fs_info->super_copy,
3004 &num_backups_tried, &backup_index);
3005 if (ret == -1)
3006 goto fail_block_groups;
3007 goto retry_root_backup;
eb60ceac
CM		 3008}
3009
f2984462
CM		 3010static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3011{
f2984462
CM		 3012 if (uptodate) {
3013 set_buffer_uptodate(bh);
3014 } else {
442a4f63
SB		 3015 struct btrfs_device *device = (struct btrfs_device *)
3016 bh->b_private;
3017
efe120a0 3018 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
606686ee
JB		 3019 "I/O error on %s\n",
3020 rcu_str_deref(device->name));
1259ab75
CM		 3021 /* note, we dont' set_buffer_write_io_error because we have
3022 * our own ways of dealing with the IO errors
3023 */
f2984462 3024 clear_buffer_uptodate(bh);
442a4f63 3025 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 3026 }
3027 unlock_buffer(bh);
3028 put_bh(bh);
3029}
3030
a512bbf8
YZ		 3031struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3032{
3033 struct buffer_head *bh;
3034 struct buffer_head *latest = NULL;
3035 struct btrfs_super_block *super;
3036 int i;
3037 u64 transid = 0;
3038 u64 bytenr;
3039
3040 /* we would like to check all the supers, but that would make
3041 * a btrfs mount succeed after a mkfs from a different FS.
3042 * So, we need to add a special mount option to scan for
3043 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3044 */
3045 for (i = 0; i < 1; i++) {
3046 bytenr = btrfs_sb_offset(i);
8068a47e
AJ		 3047 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3048 i_size_read(bdev->bd_inode))
a512bbf8 3049 break;
8068a47e
AJ		 3050 bh = __bread(bdev, bytenr / 4096,
3051 BTRFS_SUPER_INFO_SIZE);
a512bbf8
YZ		 3052 if (!bh)
3053 continue;
3054
3055 super = (struct btrfs_super_block *)bh->b_data;
3056 if (btrfs_super_bytenr(super) != bytenr ||
3cae210f 3057 btrfs_super_magic(super) != BTRFS_MAGIC) {
a512bbf8
YZ		 3058 brelse(bh);
3059 continue;
3060 }
3061
3062 if (!latest || btrfs_super_generation(super) > transid) {
3063 brelse(latest);
3064 latest = bh;
3065 transid = btrfs_super_generation(super);
3066 } else {
3067 brelse(bh);
3068 }
3069 }
3070 return latest;
3071}
3072
4eedeb75
HH		 3073/*
3074 * this should be called twice, once with wait == 0 and
3075 * once with wait == 1. When wait == 0 is done, all the buffer heads
3076 * we write are pinned.
3077 *
3078 * They are released when wait == 1 is done.
3079 * max_mirrors must be the same for both runs, and it indicates how
3080 * many supers on this one device should be written.
3081 *
3082 * max_mirrors == 0 means to write them all.
3083 */
a512bbf8
YZ		 3084static int write_dev_supers(struct btrfs_device *device,
3085 struct btrfs_super_block *sb,
3086 int do_barriers, int wait, int max_mirrors)
3087{
3088 struct buffer_head *bh;
3089 int i;
3090 int ret;
3091 int errors = 0;
3092 u32 crc;
3093 u64 bytenr;
a512bbf8
YZ		 3094
3095 if (max_mirrors == 0)
3096 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3097
a512bbf8
YZ		 3098 for (i = 0; i < max_mirrors; i++) {
3099 bytenr = btrfs_sb_offset(i);
3100 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3101 break;
3102
3103 if (wait) {
3104 bh = __find_get_block(device->bdev, bytenr / 4096,
3105 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3106 if (!bh) {
3107 errors++;
3108 continue;
3109 }
a512bbf8 3110 wait_on_buffer(bh);
4eedeb75
HH		 3111 if (!buffer_uptodate(bh))
3112 errors++;
3113
3114 /* drop our reference */
3115 brelse(bh);
3116
3117 /* drop the reference from the wait == 0 run */
3118 brelse(bh);
3119 continue;
a512bbf8
YZ		 3120 } else {
3121 btrfs_set_super_bytenr(sb, bytenr);
3122
3123 crc = ~(u32)0;
b0496686 3124 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 3125 BTRFS_CSUM_SIZE, crc,
3126 BTRFS_SUPER_INFO_SIZE -
3127 BTRFS_CSUM_SIZE);
3128 btrfs_csum_final(crc, sb->csum);
3129
4eedeb75
HH		 3130 /*
3131 * one reference for us, and we leave it for the
3132 * caller
3133 */
a512bbf8
YZ		 3134 bh = __getblk(device->bdev, bytenr / 4096,
3135 BTRFS_SUPER_INFO_SIZE);
634554dc 3136 if (!bh) {
efe120a0 3137 printk(KERN_ERR "BTRFS: couldn't get super "
634554dc
JB		 3138 "buffer head for bytenr %Lu\n", bytenr);
3139 errors++;
3140 continue;
3141 }
3142
a512bbf8
YZ		 3143 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3144
4eedeb75 3145 /* one reference for submit_bh */
a512bbf8 3146 get_bh(bh);
4eedeb75
HH		 3147
3148 set_buffer_uptodate(bh);
a512bbf8
YZ		 3149 lock_buffer(bh);
3150 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3151 bh->b_private = device;
a512bbf8
YZ		 3152 }
3153
387125fc
CM		 3154 /*
3155 * we fua the first super. The others we allow
3156 * to go down lazy.
3157 */
e8117c26
WS		 3158 if (i == 0)
3159 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3160 else
3161 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
4eedeb75 3162 if (ret)
a512bbf8 3163 errors++;
a512bbf8
YZ		 3164 }
3165 return errors < i ? 0 : -1;
3166}
3167
387125fc
CM		 3168/*
3169 * endio for the write_dev_flush, this will wake anyone waiting
3170 * for the barrier when it is done
3171 */
3172static void btrfs_end_empty_barrier(struct bio *bio, int err)
3173{
3174 if (err) {
3175 if (err == -EOPNOTSUPP)
3176 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3177 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3178 }
3179 if (bio->bi_private)
3180 complete(bio->bi_private);
3181 bio_put(bio);
3182}
3183
3184/*
3185 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3186 * sent down. With wait == 1, it waits for the previous flush.
3187 *
3188 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3189 * capable
3190 */
3191static int write_dev_flush(struct btrfs_device *device, int wait)
3192{
3193 struct bio *bio;
3194 int ret = 0;
3195
3196 if (device->nobarriers)
3197 return 0;
3198
3199 if (wait) {
3200 bio = device->flush_bio;
3201 if (!bio)
3202 return 0;
3203
3204 wait_for_completion(&device->flush_wait);
3205
3206 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
efe120a0 3207 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
606686ee 3208 rcu_str_deref(device->name));
387125fc 3209 device->nobarriers = 1;
5af3e8cc 3210 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3211 ret = -EIO;
5af3e8cc
SB		 3212 btrfs_dev_stat_inc_and_print(device,
3213 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3214 }
3215
3216 /* drop the reference from the wait == 0 run */
3217 bio_put(bio);
3218 device->flush_bio = NULL;
3219
3220 return ret;
3221 }
3222
3223 /*
3224 * one reference for us, and we leave it for the
3225 * caller
3226 */
9c017abc 3227 device->flush_bio = NULL;
9be3395b 3228 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
387125fc
CM		 3229 if (!bio)
3230 return -ENOMEM;
3231
3232 bio->bi_end_io = btrfs_end_empty_barrier;
3233 bio->bi_bdev = device->bdev;
3234 init_completion(&device->flush_wait);
3235 bio->bi_private = &device->flush_wait;
3236 device->flush_bio = bio;
3237
3238 bio_get(bio);
21adbd5c 3239 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3240
3241 return 0;
3242}
3243
3244/*
3245 * send an empty flush down to each device in parallel,
3246 * then wait for them
3247 */
3248static int barrier_all_devices(struct btrfs_fs_info *info)
3249{
3250 struct list_head *head;
3251 struct btrfs_device *dev;
5af3e8cc
SB		 3252 int errors_send = 0;
3253 int errors_wait = 0;
387125fc
CM		 3254 int ret;
3255
3256 /* send down all the barriers */
3257 head = &info->fs_devices->devices;
3258 list_for_each_entry_rcu(dev, head, dev_list) {
f88ba6a2
HS		 3259 if (dev->missing)
3260 continue;
387125fc 3261 if (!dev->bdev) {
5af3e8cc 3262 errors_send++;
387125fc
CM		 3263 continue;
3264 }
3265 if (!dev->in_fs_metadata || !dev->writeable)
3266 continue;
3267
3268 ret = write_dev_flush(dev, 0);
3269 if (ret)
5af3e8cc 3270 errors_send++;
387125fc
CM		 3271 }
3272
3273 /* wait for all the barriers */
3274 list_for_each_entry_rcu(dev, head, dev_list) {
f88ba6a2
HS		 3275 if (dev->missing)
3276 continue;
387125fc 3277 if (!dev->bdev) {
5af3e8cc 3278 errors_wait++;
387125fc
CM		 3279 continue;
3280 }
3281 if (!dev->in_fs_metadata || !dev->writeable)
3282 continue;
3283
3284 ret = write_dev_flush(dev, 1);
3285 if (ret)
5af3e8cc 3286 errors_wait++;
387125fc 3287 }
5af3e8cc
SB		 3288 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3289 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3290 return -EIO;
3291 return 0;
3292}
3293
5af3e8cc
SB		 3294int btrfs_calc_num_tolerated_disk_barrier_failures(
3295 struct btrfs_fs_info *fs_info)
3296{
3297 struct btrfs_ioctl_space_info space;
3298 struct btrfs_space_info *sinfo;
3299 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3300 BTRFS_BLOCK_GROUP_SYSTEM,
3301 BTRFS_BLOCK_GROUP_METADATA,
3302 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3303 int num_types = 4;
3304 int i;
3305 int c;
3306 int num_tolerated_disk_barrier_failures =
3307 (int)fs_info->fs_devices->num_devices;
3308
3309 for (i = 0; i < num_types; i++) {
3310 struct btrfs_space_info *tmp;
3311
3312 sinfo = NULL;
3313 rcu_read_lock();
3314 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3315 if (tmp->flags == types[i]) {
3316 sinfo = tmp;
3317 break;
3318 }
3319 }
3320 rcu_read_unlock();
3321
3322 if (!sinfo)
3323 continue;
3324
3325 down_read(&sinfo->groups_sem);
3326 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3327 if (!list_empty(&sinfo->block_groups[c])) {
3328 u64 flags;
3329
3330 btrfs_get_block_group_info(
3331 &sinfo->block_groups[c], &space);
3332 if (space.total_bytes == 0 ||
3333 space.used_bytes == 0)
3334 continue;
3335 flags = space.flags;
3336 /*
3337 * return
3338 * 0: if dup, single or RAID0 is configured for
3339 * any of metadata, system or data, else
3340 * 1: if RAID5 is configured, or if RAID1 or
3341 * RAID10 is configured and only two mirrors
3342 * are used, else
3343 * 2: if RAID6 is configured, else
3344 * num_mirrors - 1: if RAID1 or RAID10 is
3345 * configured and more than
3346 * 2 mirrors are used.
3347 */
3348 if (num_tolerated_disk_barrier_failures > 0 &&
3349 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3350 BTRFS_BLOCK_GROUP_RAID0)) ||
3351 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3352 == 0)))
3353 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3354 else if (num_tolerated_disk_barrier_failures > 1) {
3355 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3356 BTRFS_BLOCK_GROUP_RAID5 |
3357 BTRFS_BLOCK_GROUP_RAID10)) {
3358 num_tolerated_disk_barrier_failures = 1;
3359 } else if (flags &
15b0a89d 3360 BTRFS_BLOCK_GROUP_RAID6) {
53b381b3
DW		 3361 num_tolerated_disk_barrier_failures = 2;
3362 }
3363 }
5af3e8cc
SB		 3364 }
3365 }
3366 up_read(&sinfo->groups_sem);
3367 }
3368
3369 return num_tolerated_disk_barrier_failures;
3370}
3371
48a3b636 3372static int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3373{
e5e9a520 3374 struct list_head *head;
f2984462 3375 struct btrfs_device *dev;
a061fc8d 3376 struct btrfs_super_block *sb;
f2984462 3377 struct btrfs_dev_item *dev_item;
f2984462
CM		 3378 int ret;
3379 int do_barriers;
a236aed1
CM		 3380 int max_errors;
3381 int total_errors = 0;
a061fc8d 3382 u64 flags;
f2984462
CM		 3383
3384 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3385 backup_super_roots(root->fs_info);
f2984462 3386
6c41761f 3387 sb = root->fs_info->super_for_commit;
a061fc8d 3388 dev_item = &sb->dev_item;
e5e9a520 3389
174ba509 3390 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3391 head = &root->fs_info->fs_devices->devices;
d7306801 3392 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
387125fc 3393
5af3e8cc
SB		 3394 if (do_barriers) {
3395 ret = barrier_all_devices(root->fs_info);
3396 if (ret) {
3397 mutex_unlock(
3398 &root->fs_info->fs_devices->device_list_mutex);
3399 btrfs_error(root->fs_info, ret,
3400 "errors while submitting device barriers.");
3401 return ret;
3402 }
3403 }
387125fc 3404
1f78160c 3405 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3406 if (!dev->bdev) {
3407 total_errors++;
3408 continue;
3409 }
2b82032c 3410 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3411 continue;
3412
2b82032c 3413 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3414 btrfs_set_stack_device_type(dev_item, dev->type);
3415 btrfs_set_stack_device_id(dev_item, dev->devid);
3416 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3417 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3418 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3419 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3420 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3421 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3422 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3423
a061fc8d
CM		 3424 flags = btrfs_super_flags(sb);
3425 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3426
a512bbf8 3427 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3428 if (ret)
3429 total_errors++;
f2984462 3430 }
a236aed1 3431 if (total_errors > max_errors) {
efe120a0 3432 btrfs_err(root->fs_info, "%d errors while writing supers",
d397712b 3433 total_errors);
a724b436 3434 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
79787eaa 3435
9d565ba4
SB		 3436 /* FUA is masked off if unsupported and can't be the reason */
3437 btrfs_error(root->fs_info, -EIO,
3438 "%d errors while writing supers", total_errors);
3439 return -EIO;
a236aed1 3440 }
f2984462 3441
a512bbf8 3442 total_errors = 0;
1f78160c 3443 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3444 if (!dev->bdev)
3445 continue;
2b82032c 3446 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3447 continue;
3448
a512bbf8
YZ		 3449 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3450 if (ret)
3451 total_errors++;
f2984462 3452 }
174ba509 3453 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3454 if (total_errors > max_errors) {
79787eaa
JM		 3455 btrfs_error(root->fs_info, -EIO,
3456 "%d errors while writing supers", total_errors);
3457 return -EIO;
a236aed1 3458 }
f2984462
CM		 3459 return 0;
3460}
3461
a512bbf8
YZ		 3462int write_ctree_super(struct btrfs_trans_handle *trans,
3463 struct btrfs_root *root, int max_mirrors)
eb60ceac 3464{
f570e757 3465 return write_all_supers(root, max_mirrors);
cfaa7295
CM		 3466}
3467
cb517eab
MX		 3468/* Drop a fs root from the radix tree and free it. */
3469void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3470 struct btrfs_root *root)
2619ba1f 3471{
4df27c4d 3472 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3473 radix_tree_delete(&fs_info->fs_roots_radix,
3474 (unsigned long)root->root_key.objectid);
4df27c4d 3475 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3476
3477 if (btrfs_root_refs(&root->root_item) == 0)
3478 synchronize_srcu(&fs_info->subvol_srcu);
3479
1a4319cc 3480 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3321719e 3481 btrfs_free_log(NULL, root);
3321719e 3482
581bb050
LZ		 3483 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3484 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3485 free_fs_root(root);
4df27c4d
YZ		 3486}
3487
3488static void free_fs_root(struct btrfs_root *root)
3489{
82d5902d 3490 iput(root->cache_inode);
4df27c4d 3491 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1cb048f5
FDBM		 3492 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3493 root->orphan_block_rsv = NULL;
0ee5dc67
AV		 3494 if (root->anon_dev)
3495 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 3496 free_extent_buffer(root->node);
3497 free_extent_buffer(root->commit_root);
581bb050
LZ		 3498 kfree(root->free_ino_ctl);
3499 kfree(root->free_ino_pinned);
d397712b 3500 kfree(root->name);
b0feb9d9 3501 btrfs_put_fs_root(root);
2619ba1f
CM		 3502}
3503
cb517eab
MX		 3504void btrfs_free_fs_root(struct btrfs_root *root)
3505{
3506 free_fs_root(root);
2619ba1f
CM		 3507}
3508
c146afad 3509int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3510{
c146afad
YZ		 3511 u64 root_objectid = 0;
3512 struct btrfs_root *gang[8];
3513 int i;
3768f368 3514 int ret;
e089f05c 3515
c146afad
YZ		 3516 while (1) {
3517 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3518 (void **)gang, root_objectid,
3519 ARRAY_SIZE(gang));
3520 if (!ret)
3521 break;
5d4f98a2
YZ		 3522
3523 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 3524 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 3525 int err;
3526
c146afad 3527 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3528 err = btrfs_orphan_cleanup(gang[i]);
3529 if (err)
3530 return err;
c146afad
YZ		 3531 }
3532 root_objectid++;
3533 }
3534 return 0;
3535}
a2135011 3536
c146afad
YZ		 3537int btrfs_commit_super(struct btrfs_root *root)
3538{
3539 struct btrfs_trans_handle *trans;
a74a4b97 3540
c146afad 3541 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3542 btrfs_run_delayed_iputs(root);
c146afad 3543 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3544 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3545
3546 /* wait until ongoing cleanup work done */
3547 down_write(&root->fs_info->cleanup_work_sem);
3548 up_write(&root->fs_info->cleanup_work_sem);
3549
7a7eaa40 3550 trans = btrfs_join_transaction(root);
3612b495
TI		 3551 if (IS_ERR(trans))
3552 return PTR_ERR(trans);
d52c1bcc 3553 return btrfs_commit_transaction(trans, root);
c146afad
YZ		 3554}
3555
3556int close_ctree(struct btrfs_root *root)
3557{
3558 struct btrfs_fs_info *fs_info = root->fs_info;
3559 int ret;
3560
3561 fs_info->closing = 1;
3562 smp_mb();
3563
803b2f54
SB		 3564 /* wait for the uuid_scan task to finish */
3565 down(&fs_info->uuid_tree_rescan_sem);
3566 /* avoid complains from lockdep et al., set sem back to initial state */
3567 up(&fs_info->uuid_tree_rescan_sem);
3568
837d5b6e 3569 /* pause restriper - we want to resume on mount */
aa1b8cd4 3570 btrfs_pause_balance(fs_info);
837d5b6e 3571
8dabb742
SB		 3572 btrfs_dev_replace_suspend_for_unmount(fs_info);
3573
aa1b8cd4 3574 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3575
3576 /* wait for any defraggers to finish */
3577 wait_event(fs_info->transaction_wait,
3578 (atomic_read(&fs_info->defrag_running) == 0));
3579
3580 /* clear out the rbtree of defraggable inodes */
26176e7c 3581 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3582
c146afad 3583 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3584 ret = btrfs_commit_super(root);
3585 if (ret)
efe120a0 3586 btrfs_err(root->fs_info, "commit super ret %d", ret);
acce952b 3587 }
3588
87533c47 3589 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3590 btrfs_error_commit_super(root);
0f7d52f4 3591
e3029d9f
AV		 3592 kthread_stop(fs_info->transaction_kthread);
3593 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3594
f25784b3
YZ		 3595 fs_info->closing = 2;
3596 smp_mb();
3597
bcef60f2
AJ		 3598 btrfs_free_qgroup_config(root->fs_info);
3599
963d678b 3600 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
efe120a0 3601 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
963d678b 3602 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3603 }
bcc63abb 3604
5ac1d209
JM		 3605 btrfs_sysfs_remove_one(fs_info);
3606
c146afad 3607 del_fs_roots(fs_info);
d10c5f31 3608
1a4319cc
LB		 3609 btrfs_put_block_group_cache(fs_info);
3610
2b1360da
JB		 3611 btrfs_free_block_groups(fs_info);
3612
96192499
JB		 3613 btrfs_stop_all_workers(fs_info);
3614
13e6c37b 3615 free_root_pointers(fs_info, 1);
9ad6b7bc 3616
13e6c37b 3617 iput(fs_info->btree_inode);
d6bfde87 3618
21adbd5c
SB		 3619#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3620 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3621 btrfsic_unmount(root, fs_info->fs_devices);
3622#endif
3623
dfe25020 3624 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3625 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3626
e2d84521 3627 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3628 percpu_counter_destroy(&fs_info->delalloc_bytes);
c404e0dc 3629 percpu_counter_destroy(&fs_info->bio_counter);
04160088 3630 bdi_destroy(&fs_info->bdi);
76dda93c 3631 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3632
53b381b3
DW		 3633 btrfs_free_stripe_hash_table(fs_info);
3634
1cb048f5
FDBM		 3635 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3636 root->orphan_block_rsv = NULL;
3637
eb60ceac
CM		 3638 return 0;
3639}
3640
b9fab919
CM		 3641int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3642 int atomic)
5f39d397 3643{
1259ab75 3644 int ret;
727011e0 3645 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3646
0b32f4bb 3647 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3648 if (!ret)
3649 return ret;
3650
3651 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3652 parent_transid, atomic);
3653 if (ret == -EAGAIN)
3654 return ret;
1259ab75 3655 return !ret;
5f39d397
CM		 3656}
3657
3658int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3659{
0b32f4bb 3660 return set_extent_buffer_uptodate(buf);
5f39d397 3661}
6702ed49 3662
5f39d397
CM		 3663void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3664{
06ea65a3 3665 struct btrfs_root *root;
5f39d397 3666 u64 transid = btrfs_header_generation(buf);
b9473439 3667 int was_dirty;
b4ce94de 3668
06ea65a3
JB		 3669#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3670 /*
3671 * This is a fast path so only do this check if we have sanity tests
3672 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3673 * outside of the sanity tests.
3674 */
3675 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3676 return;
3677#endif
3678 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
b9447ef8 3679 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3680 if (transid != root->fs_info->generation)
3681 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3682 "found %llu running %llu\n",
c1c9ff7c 3683 buf->start, transid, root->fs_info->generation);
0b32f4bb 3684 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3685 if (!was_dirty)
3686 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3687 buf->len,
3688 root->fs_info->dirty_metadata_batch);
eb60ceac
CM		 3689}
3690
b53d3f5d
LB		 3691static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3692 int flush_delayed)
16cdcec7
MX		 3693{
3694 /*
3695 * looks as though older kernels can get into trouble with
3696 * this code, they end up stuck in balance_dirty_pages forever
3697 */
e2d84521 3698 int ret;
16cdcec7
MX		 3699
3700 if (current->flags & PF_MEMALLOC)
3701 return;
3702
b53d3f5d
LB		 3703 if (flush_delayed)
3704 btrfs_balance_delayed_items(root);
16cdcec7 3705
e2d84521
MX		 3706 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3707 BTRFS_DIRTY_METADATA_THRESH);
3708 if (ret > 0) {
d0e1d66b
NJ		 3709 balance_dirty_pages_ratelimited(
3710 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3711 }
3712 return;
3713}
3714
b53d3f5d 3715void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3716{
b53d3f5d
LB		 3717 __btrfs_btree_balance_dirty(root, 1);
3718}
585ad2c3 3719
b53d3f5d
LB		 3720void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3721{
3722 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3723}
6b80053d 3724
ca7a79ad 3725int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3726{
727011e0 3727 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3728 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3729}
0da5468f 3730
fcd1f065 3731static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3732 int read_only)
3733{
1104a885
DS		 3734 /*
3735 * Placeholder for checks
3736 */
fcd1f065 3737 return 0;
acce952b 3738}
3739
48a3b636 3740static void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3741{
acce952b 3742 mutex_lock(&root->fs_info->cleaner_mutex);
3743 btrfs_run_delayed_iputs(root);
3744 mutex_unlock(&root->fs_info->cleaner_mutex);
3745
3746 down_write(&root->fs_info->cleanup_work_sem);
3747 up_write(&root->fs_info->cleanup_work_sem);
3748
3749 /* cleanup FS via transaction */
3750 btrfs_cleanup_transaction(root);
acce952b 3751}
3752
569e0f35
JB		 3753static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3754 struct btrfs_root *root)
acce952b 3755{
3756 struct btrfs_inode *btrfs_inode;
3757 struct list_head splice;
3758
3759 INIT_LIST_HEAD(&splice);
3760
3761 mutex_lock(&root->fs_info->ordered_operations_mutex);
199c2a9c 3762 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3763
569e0f35 3764 list_splice_init(&t->ordered_operations, &splice);
acce952b 3765 while (!list_empty(&splice)) {
3766 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3767 ordered_operations);
3768
3769 list_del_init(&btrfs_inode->ordered_operations);
199c2a9c 3770 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3771
3772 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3773
199c2a9c 3774 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3775 }
3776
199c2a9c 3777 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3778 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3779}
3780
143bede5 3781static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3782{
acce952b 3783 struct btrfs_ordered_extent *ordered;
acce952b 3784
199c2a9c 3785 spin_lock(&root->ordered_extent_lock);
779880ef
JB		 3786 /*
3787 * This will just short circuit the ordered completion stuff which will
3788 * make sure the ordered extent gets properly cleaned up.
3789 */
199c2a9c 3790 list_for_each_entry(ordered, &root->ordered_extents,
779880ef
JB		 3791 root_extent_list)
3792 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
199c2a9c
MX		 3793 spin_unlock(&root->ordered_extent_lock);
3794}
3795
3796static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3797{
3798 struct btrfs_root *root;
3799 struct list_head splice;
3800
3801 INIT_LIST_HEAD(&splice);
3802
3803 spin_lock(&fs_info->ordered_root_lock);
3804 list_splice_init(&fs_info->ordered_roots, &splice);
3805 while (!list_empty(&splice)) {
3806 root = list_first_entry(&splice, struct btrfs_root,
3807 ordered_root);
1de2cfde
JB		 3808 list_move_tail(&root->ordered_root,
3809 &fs_info->ordered_roots);
199c2a9c 3810
2a85d9ca 3811 spin_unlock(&fs_info->ordered_root_lock);
199c2a9c
MX		 3812 btrfs_destroy_ordered_extents(root);
3813
2a85d9ca
LB		 3814 cond_resched();
3815 spin_lock(&fs_info->ordered_root_lock);
199c2a9c
MX		 3816 }
3817 spin_unlock(&fs_info->ordered_root_lock);
acce952b 3818}
3819
35a3621b
SB		 3820static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3821 struct btrfs_root *root)
acce952b 3822{
3823 struct rb_node *node;
3824 struct btrfs_delayed_ref_root *delayed_refs;
3825 struct btrfs_delayed_ref_node *ref;
3826 int ret = 0;
3827
3828 delayed_refs = &trans->delayed_refs;
3829
3830 spin_lock(&delayed_refs->lock);
d7df2c79 3831 if (atomic_read(&delayed_refs->num_entries) == 0) {
cfece4db 3832 spin_unlock(&delayed_refs->lock);
efe120a0 3833 btrfs_info(root->fs_info, "delayed_refs has NO entry");
acce952b 3834 return ret;
3835 }
3836
d7df2c79
JB		 3837 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3838 struct btrfs_delayed_ref_head *head;
e78417d1 3839 bool pin_bytes = false;
acce952b 3840
d7df2c79
JB		 3841 head = rb_entry(node, struct btrfs_delayed_ref_head,
3842 href_node);
3843 if (!mutex_trylock(&head->mutex)) {
3844 atomic_inc(&head->node.refs);
3845 spin_unlock(&delayed_refs->lock);
acce952b 3846
d7df2c79 3847 mutex_lock(&head->mutex);
e78417d1 3848 mutex_unlock(&head->mutex);
d7df2c79
JB		 3849 btrfs_put_delayed_ref(&head->node);
3850 spin_lock(&delayed_refs->lock);
3851 continue;
3852 }
3853 spin_lock(&head->lock);
3854 while ((node = rb_first(&head->ref_root)) != NULL) {
3855 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3856 rb_node);
3857 ref->in_tree = 0;
3858 rb_erase(&ref->rb_node, &head->ref_root);
3859 atomic_dec(&delayed_refs->num_entries);
3860 btrfs_put_delayed_ref(ref);
e78417d1 3861 }
d7df2c79
JB		 3862 if (head->must_insert_reserved)
3863 pin_bytes = true;
3864 btrfs_free_delayed_extent_op(head->extent_op);
3865 delayed_refs->num_heads--;
3866 if (head->processing == 0)
3867 delayed_refs->num_heads_ready--;
3868 atomic_dec(&delayed_refs->num_entries);
3869 head->node.in_tree = 0;
3870 rb_erase(&head->href_node, &delayed_refs->href_root);
3871 spin_unlock(&head->lock);
3872 spin_unlock(&delayed_refs->lock);
3873 mutex_unlock(&head->mutex);
acce952b 3874
d7df2c79
JB		 3875 if (pin_bytes)
3876 btrfs_pin_extent(root, head->node.bytenr,
3877 head->node.num_bytes, 1);
3878 btrfs_put_delayed_ref(&head->node);
acce952b 3879 cond_resched();
3880 spin_lock(&delayed_refs->lock);
3881 }
3882
3883 spin_unlock(&delayed_refs->lock);
3884
3885 return ret;
3886}
3887
143bede5 3888static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3889{
3890 struct btrfs_inode *btrfs_inode;
3891 struct list_head splice;
3892
3893 INIT_LIST_HEAD(&splice);
3894
eb73c1b7
MX		 3895 spin_lock(&root->delalloc_lock);
3896 list_splice_init(&root->delalloc_inodes, &splice);
acce952b 3897
3898 while (!list_empty(&splice)) {
eb73c1b7
MX		 3899 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3900 delalloc_inodes);
acce952b 3901
3902 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3903 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3904 &btrfs_inode->runtime_flags);
eb73c1b7 3905 spin_unlock(&root->delalloc_lock);
acce952b 3906
3907 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3908
eb73c1b7 3909 spin_lock(&root->delalloc_lock);
acce952b 3910 }
3911
eb73c1b7
MX		 3912 spin_unlock(&root->delalloc_lock);
3913}
3914
3915static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3916{
3917 struct btrfs_root *root;
3918 struct list_head splice;
3919
3920 INIT_LIST_HEAD(&splice);
3921
3922 spin_lock(&fs_info->delalloc_root_lock);
3923 list_splice_init(&fs_info->delalloc_roots, &splice);
3924 while (!list_empty(&splice)) {
3925 root = list_first_entry(&splice, struct btrfs_root,
3926 delalloc_root);
3927 list_del_init(&root->delalloc_root);
3928 root = btrfs_grab_fs_root(root);
3929 BUG_ON(!root);
3930 spin_unlock(&fs_info->delalloc_root_lock);
3931
3932 btrfs_destroy_delalloc_inodes(root);
3933 btrfs_put_fs_root(root);
3934
3935 spin_lock(&fs_info->delalloc_root_lock);
3936 }
3937 spin_unlock(&fs_info->delalloc_root_lock);
acce952b 3938}
3939
3940static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3941 struct extent_io_tree *dirty_pages,
3942 int mark)
3943{
3944 int ret;
acce952b 3945 struct extent_buffer *eb;
3946 u64 start = 0;
3947 u64 end;
acce952b 3948
3949 while (1) {
3950 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3951 mark, NULL);
acce952b 3952 if (ret)
3953 break;
3954
3955 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3956 while (start <= end) {
fd8b2b61
JB		 3957 eb = btrfs_find_tree_block(root, start,
3958 root->leafsize);
69a85bd8 3959 start += root->leafsize;
fd8b2b61 3960 if (!eb)
acce952b 3961 continue;
fd8b2b61 3962 wait_on_extent_buffer_writeback(eb);
acce952b 3963
fd8b2b61
JB		 3964 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3965 &eb->bflags))
3966 clear_extent_buffer_dirty(eb);
3967 free_extent_buffer_stale(eb);
acce952b 3968 }
3969 }
3970
3971 return ret;
3972}
3973
3974static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3975 struct extent_io_tree *pinned_extents)
3976{
3977 struct extent_io_tree *unpin;
3978 u64 start;
3979 u64 end;
3980 int ret;
ed0eaa14 3981 bool loop = true;
acce952b 3982
3983 unpin = pinned_extents;
ed0eaa14 3984again:
acce952b 3985 while (1) {
3986 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 3987 EXTENT_DIRTY, NULL);
acce952b 3988 if (ret)
3989 break;
3990
3991 /* opt_discard */
5378e607
LD		 3992 if (btrfs_test_opt(root, DISCARD))
3993 ret = btrfs_error_discard_extent(root, start,
3994 end + 1 - start,
3995 NULL);
acce952b 3996
3997 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3998 btrfs_error_unpin_extent_range(root, start, end);
3999 cond_resched();
4000 }
4001
ed0eaa14
LB		 4002 if (loop) {
4003 if (unpin == &root->fs_info->freed_extents[0])
4004 unpin = &root->fs_info->freed_extents[1];
4005 else
4006 unpin = &root->fs_info->freed_extents[0];
4007 loop = false;
4008 goto again;
4009 }
4010
acce952b 4011 return 0;
4012}
4013
49b25e05
JM		 4014void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4015 struct btrfs_root *root)
4016{
724e2315
JB		 4017 btrfs_destroy_ordered_operations(cur_trans, root);
4018
49b25e05 4019 btrfs_destroy_delayed_refs(cur_trans, root);
49b25e05 4020
4a9d8bde 4021 cur_trans->state = TRANS_STATE_COMMIT_START;
d7096fc3 4022 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 4023
4a9d8bde 4024 cur_trans->state = TRANS_STATE_UNBLOCKED;
d7096fc3 4025 wake_up(&root->fs_info->transaction_wait);
49b25e05 4026
67cde344
MX		 4027 btrfs_destroy_delayed_inodes(root);
4028 btrfs_assert_delayed_root_empty(root);
49b25e05 4029
49b25e05
JM		 4030 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4031 EXTENT_DIRTY);
6e841e32
LB		 4032 btrfs_destroy_pinned_extent(root,
4033 root->fs_info->pinned_extents);
49b25e05 4034
4a9d8bde
MX		 4035 cur_trans->state =TRANS_STATE_COMPLETED;
4036 wake_up(&cur_trans->commit_wait);
4037
49b25e05
JM		 4038 /*
4039 memset(cur_trans, 0, sizeof(*cur_trans));
4040 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4041 */
4042}
4043
48a3b636 4044static int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 4045{
4046 struct btrfs_transaction *t;
acce952b 4047
acce952b 4048 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4049
a4abeea4 4050 spin_lock(&root->fs_info->trans_lock);
724e2315
JB		 4051 while (!list_empty(&root->fs_info->trans_list)) {
4052 t = list_first_entry(&root->fs_info->trans_list,
4053 struct btrfs_transaction, list);
4054 if (t->state >= TRANS_STATE_COMMIT_START) {
4055 atomic_inc(&t->use_count);
4056 spin_unlock(&root->fs_info->trans_lock);
4057 btrfs_wait_for_commit(root, t->transid);
4058 btrfs_put_transaction(t);
4059 spin_lock(&root->fs_info->trans_lock);
4060 continue;
4061 }
4062 if (t == root->fs_info->running_transaction) {
4063 t->state = TRANS_STATE_COMMIT_DOING;
4064 spin_unlock(&root->fs_info->trans_lock);
4065 /*
4066 * We wait for 0 num_writers since we don't hold a trans
4067 * handle open currently for this transaction.
4068 */
4069 wait_event(t->writer_wait,
4070 atomic_read(&t->num_writers) == 0);
4071 } else {
4072 spin_unlock(&root->fs_info->trans_lock);
4073 }
4074 btrfs_cleanup_one_transaction(t, root);
4a9d8bde 4075
724e2315
JB		 4076 spin_lock(&root->fs_info->trans_lock);
4077 if (t == root->fs_info->running_transaction)
4078 root->fs_info->running_transaction = NULL;
acce952b 4079 list_del_init(&t->list);
724e2315 4080 spin_unlock(&root->fs_info->trans_lock);
acce952b 4081
724e2315
JB		 4082 btrfs_put_transaction(t);
4083 trace_btrfs_transaction_commit(root);
4084 spin_lock(&root->fs_info->trans_lock);
4085 }
4086 spin_unlock(&root->fs_info->trans_lock);
4087 btrfs_destroy_all_ordered_extents(root->fs_info);
4088 btrfs_destroy_delayed_inodes(root);
4089 btrfs_assert_delayed_root_empty(root);
4090 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4091 btrfs_destroy_all_delalloc_inodes(root->fs_info);
acce952b 4092 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 4093
4094 return 0;
4095}
4096
d1310b2e 4097static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 4098 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 4099 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 4100 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 4101 /* note we're sharing with inode.c for the merge bio hook */
4102 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 4103};
Linux 6.x block layer and io_uring tree(s)