projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: faster/more efficient insertion of file extent items
[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;
8b712842 111 struct btrfs_work 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
CM		 740 unsigned long limit = min_t(unsigned long,
741 info->workers.max_workers,
742 info->fs_devices->open_devices);
743 return 256 * limit;
744}
0986fe9e 745
4a69a410
CM		 746static void run_one_async_start(struct btrfs_work *work)
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
759static void run_one_async_done(struct btrfs_work *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
786static void run_one_async_free(struct btrfs_work *work)
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
814 async->work.func = run_one_async_start;
815 async->work.ordered_func = run_one_async_done;
816 async->work.ordered_free = run_one_async_free;
817
8b712842 818 async->work.flags = 0;
c8b97818 819 async->bio_flags = bio_flags;
eaf25d93 820 async->bio_offset = bio_offset;
8c8bee1d 821
79787eaa
JM		 822 async->error = 0;
823
cb03c743 824 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 825
7b6d91da 826 if (rw & REQ_SYNC)
d313d7a3
CM		 827 btrfs_set_work_high_prio(&async->work);
828
8b712842 829 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 830
d397712b 831 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 832 atomic_read(&fs_info->nr_async_submits)) {
833 wait_event(fs_info->async_submit_wait,
834 (atomic_read(&fs_info->nr_async_submits) == 0));
835 }
836
44b8bd7e
CM		 837 return 0;
838}
839
ce3ed71a
CM		 840static int btree_csum_one_bio(struct bio *bio)
841{
842 struct bio_vec *bvec = bio->bi_io_vec;
843 int bio_index = 0;
844 struct btrfs_root *root;
79787eaa 845 int ret = 0;
ce3ed71a
CM		 846
847 WARN_ON(bio->bi_vcnt <= 0);
d397712b 848 while (bio_index < bio->bi_vcnt) {
ce3ed71a 849 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 850 ret = csum_dirty_buffer(root, bvec->bv_page);
851 if (ret)
852 break;
ce3ed71a
CM		 853 bio_index++;
854 bvec++;
855 }
79787eaa 856 return ret;
ce3ed71a
CM		 857}
858
4a69a410
CM		 859static int __btree_submit_bio_start(struct inode *inode, int rw,
860 struct bio *bio, int mirror_num,
eaf25d93
CM		 861 unsigned long bio_flags,
862 u64 bio_offset)
22c59948 863{
8b712842
CM		 864 /*
865 * when we're called for a write, we're already in the async
5443be45 866 * submission context. Just jump into btrfs_map_bio
8b712842 867 */
79787eaa 868 return btree_csum_one_bio(bio);
4a69a410 869}
22c59948 870
4a69a410 871static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 872 int mirror_num, unsigned long bio_flags,
873 u64 bio_offset)
4a69a410 874{
61891923
SB		 875 int ret;
876
8b712842 877 /*
4a69a410
CM		 878 * when we're called for a write, we're already in the async
879 * submission context. Just jump into btrfs_map_bio
8b712842 880 */
61891923
SB		 881 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
882 if (ret)
883 bio_endio(bio, ret);
884 return ret;
0b86a832
CM		 885}
886
de0022b9
JB		 887static int check_async_write(struct inode *inode, unsigned long bio_flags)
888{
889 if (bio_flags & EXTENT_BIO_TREE_LOG)
890 return 0;
891#ifdef CONFIG_X86
892 if (cpu_has_xmm4_2)
893 return 0;
894#endif
895 return 1;
896}
897
44b8bd7e 898static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 899 int mirror_num, unsigned long bio_flags,
900 u64 bio_offset)
44b8bd7e 901{
de0022b9 902 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 903 int ret;
904
7b6d91da 905 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 906 /*
907 * called for a read, do the setup so that checksum validation
908 * can happen in the async kernel threads
909 */
f3f266ab
CM		 910 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
911 bio, 1);
1d4284bd 912 if (ret)
61891923
SB		 913 goto out_w_error;
914 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
915 mirror_num, 0);
de0022b9
JB		 916 } else if (!async) {
917 ret = btree_csum_one_bio(bio);
918 if (ret)
61891923
SB		 919 goto out_w_error;
920 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
921 mirror_num, 0);
922 } else {
923 /*
924 * kthread helpers are used to submit writes so that
925 * checksumming can happen in parallel across all CPUs
926 */
927 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
928 inode, rw, bio, mirror_num, 0,
929 bio_offset,
930 __btree_submit_bio_start,
931 __btree_submit_bio_done);
44b8bd7e 932 }
d313d7a3 933
61891923
SB		 934 if (ret) {
935out_w_error:
936 bio_endio(bio, ret);
937 }
938 return ret;
44b8bd7e
CM		 939}
940
3dd1462e 941#ifdef CONFIG_MIGRATION
784b4e29 942static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 943 struct page *newpage, struct page *page,
944 enum migrate_mode mode)
784b4e29
CM		 945{
946 /*
947 * we can't safely write a btree page from here,
948 * we haven't done the locking hook
949 */
950 if (PageDirty(page))
951 return -EAGAIN;
952 /*
953 * Buffers may be managed in a filesystem specific way.
954 * We must have no buffers or drop them.
955 */
956 if (page_has_private(page) &&
957 !try_to_release_page(page, GFP_KERNEL))
958 return -EAGAIN;
a6bc32b8 959 return migrate_page(mapping, newpage, page, mode);
784b4e29 960}
3dd1462e 961#endif
784b4e29 962
0da5468f
CM		 963
964static int btree_writepages(struct address_space *mapping,
965 struct writeback_control *wbc)
966{
e2d84521
MX		 967 struct btrfs_fs_info *fs_info;
968 int ret;
969
d8d5f3e1 970 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 971
972 if (wbc->for_kupdate)
973 return 0;
974
e2d84521 975 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 976 /* this is a bit racy, but that's ok */
e2d84521
MX		 977 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
978 BTRFS_DIRTY_METADATA_THRESH);
979 if (ret < 0)
793955bc 980 return 0;
793955bc 981 }
0b32f4bb 982 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 983}
984
b2950863 985static int btree_readpage(struct file *file, struct page *page)
5f39d397 986{
d1310b2e
CM		 987 struct extent_io_tree *tree;
988 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 989 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 990}
22b0ebda 991
70dec807 992static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 993{
98509cfc 994 if (PageWriteback(page) || PageDirty(page))
d397712b 995 return 0;
0c4e538b 996
f7a52a40 997 return try_release_extent_buffer(page);
d98237b3
CM		 998}
999
d47992f8
LC		 1000static void btree_invalidatepage(struct page *page, unsigned int offset,
1001 unsigned int length)
d98237b3 1002{
d1310b2e
CM		 1003 struct extent_io_tree *tree;
1004 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1005 extent_invalidatepage(tree, page, offset);
1006 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1007 if (PagePrivate(page)) {
efe120a0
FH		 1008 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1009 "page private not zero on page %llu",
1010 (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1011 ClearPagePrivate(page);
1012 set_page_private(page, 0);
1013 page_cache_release(page);
1014 }
d98237b3
CM		 1015}
1016
0b32f4bb
JB		 1017static int btree_set_page_dirty(struct page *page)
1018{
bb146eb2 1019#ifdef DEBUG
0b32f4bb
JB		 1020 struct extent_buffer *eb;
1021
1022 BUG_ON(!PagePrivate(page));
1023 eb = (struct extent_buffer *)page->private;
1024 BUG_ON(!eb);
1025 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1026 BUG_ON(!atomic_read(&eb->refs));
1027 btrfs_assert_tree_locked(eb);
bb146eb2 1028#endif
0b32f4bb
JB		 1029 return __set_page_dirty_nobuffers(page);
1030}
1031
7f09410b 1032static const struct address_space_operations btree_aops = {
d98237b3 1033 .readpage = btree_readpage,
0da5468f 1034 .writepages = btree_writepages,
5f39d397
CM		 1035 .releasepage = btree_releasepage,
1036 .invalidatepage = btree_invalidatepage,
5a92bc88 1037#ifdef CONFIG_MIGRATION
784b4e29 1038 .migratepage = btree_migratepage,
5a92bc88 1039#endif
0b32f4bb 1040 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1041};
1042
ca7a79ad
CM		 1043int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1044 u64 parent_transid)
090d1875 1045{
5f39d397
CM		 1046 struct extent_buffer *buf = NULL;
1047 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1048 int ret = 0;
090d1875 1049
db94535d 1050 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1051 if (!buf)
090d1875 1052 return 0;
d1310b2e 1053 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1054 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1055 free_extent_buffer(buf);
de428b63 1056 return ret;
090d1875
CM		 1057}
1058
ab0fff03
AJ		 1059int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1060 int mirror_num, struct extent_buffer **eb)
1061{
1062 struct extent_buffer *buf = NULL;
1063 struct inode *btree_inode = root->fs_info->btree_inode;
1064 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1065 int ret;
1066
1067 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1068 if (!buf)
1069 return 0;
1070
1071 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1072
1073 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1074 btree_get_extent, mirror_num);
1075 if (ret) {
1076 free_extent_buffer(buf);
1077 return ret;
1078 }
1079
1080 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1081 free_extent_buffer(buf);
1082 return -EIO;
0b32f4bb 1083 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1084 *eb = buf;
1085 } else {
1086 free_extent_buffer(buf);
1087 }
1088 return 0;
1089}
1090
0999df54
CM		 1091struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1092 u64 bytenr, u32 blocksize)
1093{
f28491e0 1094 return find_extent_buffer(root->fs_info, bytenr);
0999df54
CM		 1095}
1096
1097struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1098 u64 bytenr, u32 blocksize)
1099{
f28491e0 1100 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
0999df54
CM		 1101}
1102
1103
e02119d5
CM		 1104int btrfs_write_tree_block(struct extent_buffer *buf)
1105{
727011e0 1106 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
8aa38c31 1107 buf->start + buf->len - 1);
e02119d5
CM		 1108}
1109
1110int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1111{
727011e0 1112 return filemap_fdatawait_range(buf->pages[0]->mapping,
8aa38c31 1113 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1114}
1115
0999df54 1116struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1117 u32 blocksize, u64 parent_transid)
0999df54
CM		 1118{
1119 struct extent_buffer *buf = NULL;
0999df54
CM		 1120 int ret;
1121
0999df54
CM		 1122 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1123 if (!buf)
1124 return NULL;
0999df54 1125
ca7a79ad 1126 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
0f0fe8f7
FDBM		 1127 if (ret) {
1128 free_extent_buffer(buf);
1129 return NULL;
1130 }
5f39d397 1131 return buf;
ce9adaa5 1132
eb60ceac
CM		 1133}
1134
d5c13f92
JM		 1135void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1136 struct extent_buffer *buf)
ed2ff2cb 1137{
e2d84521
MX		 1138 struct btrfs_fs_info *fs_info = root->fs_info;
1139
55c69072 1140 if (btrfs_header_generation(buf) ==
e2d84521 1141 fs_info->running_transaction->transid) {
b9447ef8 1142 btrfs_assert_tree_locked(buf);
b4ce94de 1143
b9473439 1144 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1145 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1146 -buf->len,
1147 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1148 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1149 btrfs_set_lock_blocking(buf);
1150 clear_extent_buffer_dirty(buf);
1151 }
925baedd 1152 }
5f39d397
CM		 1153}
1154
143bede5
JM		 1155static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1156 u32 stripesize, struct btrfs_root *root,
1157 struct btrfs_fs_info *fs_info,
1158 u64 objectid)
d97e63b6 1159{
cfaa7295 1160 root->node = NULL;
a28ec197 1161 root->commit_root = NULL;
db94535d
CM		 1162 root->sectorsize = sectorsize;
1163 root->nodesize = nodesize;
1164 root->leafsize = leafsize;
87ee04eb 1165 root->stripesize = stripesize;
123abc88 1166 root->ref_cows = 0;
0b86a832 1167 root->track_dirty = 0;
c71bf099 1168 root->in_radix = 0;
d68fc57b
YZ		 1169 root->orphan_item_inserted = 0;
1170 root->orphan_cleanup_state = 0;
0b86a832 1171
0f7d52f4
CM		 1172 root->objectid = objectid;
1173 root->last_trans = 0;
13a8a7c8 1174 root->highest_objectid = 0;
eb73c1b7 1175 root->nr_delalloc_inodes = 0;
199c2a9c 1176 root->nr_ordered_extents = 0;
58176a96 1177 root->name = NULL;
6bef4d31 1178 root->inode_tree = RB_ROOT;
16cdcec7 1179 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1180 root->block_rsv = NULL;
d68fc57b 1181 root->orphan_block_rsv = NULL;
0b86a832
CM		 1182
1183 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1184 INIT_LIST_HEAD(&root->root_list);
eb73c1b7
MX		 1185 INIT_LIST_HEAD(&root->delalloc_inodes);
1186 INIT_LIST_HEAD(&root->delalloc_root);
199c2a9c
MX		 1187 INIT_LIST_HEAD(&root->ordered_extents);
1188 INIT_LIST_HEAD(&root->ordered_root);
2ab28f32
JB		 1189 INIT_LIST_HEAD(&root->logged_list[0]);
1190 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1191 spin_lock_init(&root->orphan_lock);
5d4f98a2 1192 spin_lock_init(&root->inode_lock);
eb73c1b7 1193 spin_lock_init(&root->delalloc_lock);
199c2a9c 1194 spin_lock_init(&root->ordered_extent_lock);
f0486c68 1195 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1196 spin_lock_init(&root->log_extents_lock[0]);
1197 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1198 mutex_init(&root->objectid_mutex);
e02119d5 1199 mutex_init(&root->log_mutex);
7237f183
YZ		 1200 init_waitqueue_head(&root->log_writer_wait);
1201 init_waitqueue_head(&root->log_commit_wait[0]);
1202 init_waitqueue_head(&root->log_commit_wait[1]);
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;
257c62e1 1210 root->last_log_commit = 0;
06ea65a3
JB		 1211 if (fs_info)
1212 extent_io_tree_init(&root->dirty_log_pages,
1213 fs_info->btree_inode->i_mapping);
017e5369 1214
3768f368
CM		 1215 memset(&root->root_key, 0, sizeof(root->root_key));
1216 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1217 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1218 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
06ea65a3
JB		 1219 if (fs_info)
1220 root->defrag_trans_start = fs_info->generation;
1221 else
1222 root->defrag_trans_start = 0;
58176a96 1223 init_completion(&root->kobj_unregister);
6702ed49 1224 root->defrag_running = 0;
4d775673 1225 root->root_key.objectid = objectid;
0ee5dc67 1226 root->anon_dev = 0;
8ea05e3a 1227
5f3ab90a 1228 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1229}
1230
f84a8bd6 1231static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1232{
1233 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1234 if (root)
1235 root->fs_info = fs_info;
1236 return root;
1237}
1238
06ea65a3
JB		 1239#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1240/* Should only be used by the testing infrastructure */
1241struct btrfs_root *btrfs_alloc_dummy_root(void)
1242{
1243 struct btrfs_root *root;
1244
1245 root = btrfs_alloc_root(NULL);
1246 if (!root)
1247 return ERR_PTR(-ENOMEM);
1248 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1249 root->dummy_root = 1;
1250
1251 return root;
1252}
1253#endif
1254
20897f5c
AJ		 1255struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1256 struct btrfs_fs_info *fs_info,
1257 u64 objectid)
1258{
1259 struct extent_buffer *leaf;
1260 struct btrfs_root *tree_root = fs_info->tree_root;
1261 struct btrfs_root *root;
1262 struct btrfs_key key;
1263 int ret = 0;
6463fe58 1264 uuid_le uuid;
20897f5c
AJ		 1265
1266 root = btrfs_alloc_root(fs_info);
1267 if (!root)
1268 return ERR_PTR(-ENOMEM);
1269
1270 __setup_root(tree_root->nodesize, tree_root->leafsize,
1271 tree_root->sectorsize, tree_root->stripesize,
1272 root, fs_info, objectid);
1273 root->root_key.objectid = objectid;
1274 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1275 root->root_key.offset = 0;
1276
1277 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1278 0, objectid, NULL, 0, 0, 0);
1279 if (IS_ERR(leaf)) {
1280 ret = PTR_ERR(leaf);
1dd05682 1281 leaf = NULL;
20897f5c
AJ		 1282 goto fail;
1283 }
1284
20897f5c
AJ		 1285 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1286 btrfs_set_header_bytenr(leaf, leaf->start);
1287 btrfs_set_header_generation(leaf, trans->transid);
1288 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1289 btrfs_set_header_owner(leaf, objectid);
1290 root->node = leaf;
1291
0a4e5586 1292 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
20897f5c
AJ		 1293 BTRFS_FSID_SIZE);
1294 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
b308bc2f 1295 btrfs_header_chunk_tree_uuid(leaf),
20897f5c
AJ		 1296 BTRFS_UUID_SIZE);
1297 btrfs_mark_buffer_dirty(leaf);
1298
1299 root->commit_root = btrfs_root_node(root);
1300 root->track_dirty = 1;
1301
1302
1303 root->root_item.flags = 0;
1304 root->root_item.byte_limit = 0;
1305 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1306 btrfs_set_root_generation(&root->root_item, trans->transid);
1307 btrfs_set_root_level(&root->root_item, 0);
1308 btrfs_set_root_refs(&root->root_item, 1);
1309 btrfs_set_root_used(&root->root_item, leaf->len);
1310 btrfs_set_root_last_snapshot(&root->root_item, 0);
1311 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1312 uuid_le_gen(&uuid);
1313 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1314 root->root_item.drop_level = 0;
1315
1316 key.objectid = objectid;
1317 key.type = BTRFS_ROOT_ITEM_KEY;
1318 key.offset = 0;
1319 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1320 if (ret)
1321 goto fail;
1322
1323 btrfs_tree_unlock(leaf);
1324
1dd05682
TI		 1325 return root;
1326
20897f5c 1327fail:
1dd05682
TI		 1328 if (leaf) {
1329 btrfs_tree_unlock(leaf);
1330 free_extent_buffer(leaf);
1331 }
1332 kfree(root);
20897f5c 1333
1dd05682 1334 return ERR_PTR(ret);
20897f5c
AJ		 1335}
1336
7237f183
YZ		 1337static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1338 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1339{
1340 struct btrfs_root *root;
1341 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1342 struct extent_buffer *leaf;
e02119d5 1343
6f07e42e 1344 root = btrfs_alloc_root(fs_info);
e02119d5 1345 if (!root)
7237f183 1346 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1347
1348 __setup_root(tree_root->nodesize, tree_root->leafsize,
1349 tree_root->sectorsize, tree_root->stripesize,
1350 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1351
1352 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1353 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1354 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1355 /*
1356 * log trees do not get reference counted because they go away
1357 * before a real commit is actually done. They do store pointers
1358 * to file data extents, and those reference counts still get
1359 * updated (along with back refs to the log tree).
1360 */
e02119d5
CM		 1361 root->ref_cows = 0;
1362
5d4f98a2 1363 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1364 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1365 0, 0, 0);
7237f183
YZ		 1366 if (IS_ERR(leaf)) {
1367 kfree(root);
1368 return ERR_CAST(leaf);
1369 }
e02119d5 1370
5d4f98a2
YZ		 1371 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1372 btrfs_set_header_bytenr(leaf, leaf->start);
1373 btrfs_set_header_generation(leaf, trans->transid);
1374 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1375 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1376 root->node = leaf;
e02119d5
CM		 1377
1378 write_extent_buffer(root->node, root->fs_info->fsid,
0a4e5586 1379 btrfs_header_fsid(), BTRFS_FSID_SIZE);
e02119d5
CM		 1380 btrfs_mark_buffer_dirty(root->node);
1381 btrfs_tree_unlock(root->node);
7237f183
YZ		 1382 return root;
1383}
1384
1385int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1386 struct btrfs_fs_info *fs_info)
1387{
1388 struct btrfs_root *log_root;
1389
1390 log_root = alloc_log_tree(trans, fs_info);
1391 if (IS_ERR(log_root))
1392 return PTR_ERR(log_root);
1393 WARN_ON(fs_info->log_root_tree);
1394 fs_info->log_root_tree = log_root;
1395 return 0;
1396}
1397
1398int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1399 struct btrfs_root *root)
1400{
1401 struct btrfs_root *log_root;
1402 struct btrfs_inode_item *inode_item;
1403
1404 log_root = alloc_log_tree(trans, root->fs_info);
1405 if (IS_ERR(log_root))
1406 return PTR_ERR(log_root);
1407
1408 log_root->last_trans = trans->transid;
1409 log_root->root_key.offset = root->root_key.objectid;
1410
1411 inode_item = &log_root->root_item.inode;
3cae210f
QW		 1412 btrfs_set_stack_inode_generation(inode_item, 1);
1413 btrfs_set_stack_inode_size(inode_item, 3);
1414 btrfs_set_stack_inode_nlink(inode_item, 1);
1415 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1416 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
7237f183 1417
5d4f98a2 1418 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1419
1420 WARN_ON(root->log_root);
1421 root->log_root = log_root;
1422 root->log_transid = 0;
257c62e1 1423 root->last_log_commit = 0;
e02119d5
CM		 1424 return 0;
1425}
1426
35a3621b
SB		 1427static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1428 struct btrfs_key *key)
e02119d5
CM		 1429{
1430 struct btrfs_root *root;
1431 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1432 struct btrfs_path *path;
84234f3a 1433 u64 generation;
db94535d 1434 u32 blocksize;
cb517eab 1435 int ret;
0f7d52f4 1436
cb517eab
MX		 1437 path = btrfs_alloc_path();
1438 if (!path)
0f7d52f4 1439 return ERR_PTR(-ENOMEM);
cb517eab
MX		 1440
1441 root = btrfs_alloc_root(fs_info);
1442 if (!root) {
1443 ret = -ENOMEM;
1444 goto alloc_fail;
0f7d52f4
CM		 1445 }
1446
db94535d 1447 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb 1448 tree_root->sectorsize, tree_root->stripesize,
cb517eab 1449 root, fs_info, key->objectid);
0f7d52f4 1450
cb517eab
MX		 1451 ret = btrfs_find_root(tree_root, key, path,
1452 &root->root_item, &root->root_key);
0f7d52f4 1453 if (ret) {
13a8a7c8
YZ		 1454 if (ret > 0)
1455 ret = -ENOENT;
cb517eab 1456 goto find_fail;
0f7d52f4 1457 }
13a8a7c8 1458
84234f3a 1459 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1460 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1461 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1462 blocksize, generation);
cb517eab
MX		 1463 if (!root->node) {
1464 ret = -ENOMEM;
1465 goto find_fail;
1466 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1467 ret = -EIO;
1468 goto read_fail;
416bc658 1469 }
5d4f98a2 1470 root->commit_root = btrfs_root_node(root);
13a8a7c8 1471out:
cb517eab
MX		 1472 btrfs_free_path(path);
1473 return root;
1474
1475read_fail:
1476 free_extent_buffer(root->node);
1477find_fail:
1478 kfree(root);
1479alloc_fail:
1480 root = ERR_PTR(ret);
1481 goto out;
1482}
1483
1484struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1485 struct btrfs_key *location)
1486{
1487 struct btrfs_root *root;
1488
1489 root = btrfs_read_tree_root(tree_root, location);
1490 if (IS_ERR(root))
1491 return root;
1492
1493 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1494 root->ref_cows = 1;
08fe4db1
LZ		 1495 btrfs_check_and_init_root_item(&root->root_item);
1496 }
13a8a7c8 1497
5eda7b5e
CM		 1498 return root;
1499}
1500
cb517eab
MX		 1501int btrfs_init_fs_root(struct btrfs_root *root)
1502{
1503 int ret;
1504
1505 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1506 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1507 GFP_NOFS);
1508 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1509 ret = -ENOMEM;
1510 goto fail;
1511 }
1512
1513 btrfs_init_free_ino_ctl(root);
1514 mutex_init(&root->fs_commit_mutex);
1515 spin_lock_init(&root->cache_lock);
1516 init_waitqueue_head(&root->cache_wait);
1517
1518 ret = get_anon_bdev(&root->anon_dev);
1519 if (ret)
1520 goto fail;
1521 return 0;
1522fail:
1523 kfree(root->free_ino_ctl);
1524 kfree(root->free_ino_pinned);
1525 return ret;
1526}
1527
171170c1
ST		 1528static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1529 u64 root_id)
cb517eab
MX		 1530{
1531 struct btrfs_root *root;
1532
1533 spin_lock(&fs_info->fs_roots_radix_lock);
1534 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1535 (unsigned long)root_id);
1536 spin_unlock(&fs_info->fs_roots_radix_lock);
1537 return root;
1538}
1539
1540int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1541 struct btrfs_root *root)
1542{
1543 int ret;
1544
1545 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1546 if (ret)
1547 return ret;
1548
1549 spin_lock(&fs_info->fs_roots_radix_lock);
1550 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1551 (unsigned long)root->root_key.objectid,
1552 root);
1553 if (ret == 0)
1554 root->in_radix = 1;
1555 spin_unlock(&fs_info->fs_roots_radix_lock);
1556 radix_tree_preload_end();
1557
1558 return ret;
1559}
1560
c00869f1
MX		 1561struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1562 struct btrfs_key *location,
1563 bool check_ref)
5eda7b5e
CM		 1564{
1565 struct btrfs_root *root;
1566 int ret;
1567
edbd8d4e
CM		 1568 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1569 return fs_info->tree_root;
1570 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1571 return fs_info->extent_root;
8f18cf13
CM		 1572 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1573 return fs_info->chunk_root;
1574 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1575 return fs_info->dev_root;
0403e47e
YZ		 1576 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1577 return fs_info->csum_root;
bcef60f2
AJ		 1578 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1579 return fs_info->quota_root ? fs_info->quota_root :
1580 ERR_PTR(-ENOENT);
f7a81ea4
SB		 1581 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1582 return fs_info->uuid_root ? fs_info->uuid_root :
1583 ERR_PTR(-ENOENT);
4df27c4d 1584again:
cb517eab 1585 root = btrfs_lookup_fs_root(fs_info, location->objectid);
48475471 1586 if (root) {
c00869f1 1587 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
48475471 1588 return ERR_PTR(-ENOENT);
5eda7b5e 1589 return root;
48475471 1590 }
5eda7b5e 1591
cb517eab 1592 root = btrfs_read_fs_root(fs_info->tree_root, location);
5eda7b5e
CM		 1593 if (IS_ERR(root))
1594 return root;
3394e160 1595
c00869f1 1596 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
cb517eab 1597 ret = -ENOENT;
581bb050 1598 goto fail;
35a30d7c 1599 }
581bb050 1600
cb517eab 1601 ret = btrfs_init_fs_root(root);
ac08aedf
CM		 1602 if (ret)
1603 goto fail;
3394e160 1604
3f870c28
KN		 1605 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1606 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
d68fc57b
YZ		 1607 if (ret < 0)
1608 goto fail;
1609 if (ret == 0)
1610 root->orphan_item_inserted = 1;
1611
cb517eab 1612 ret = btrfs_insert_fs_root(fs_info, root);
0f7d52f4 1613 if (ret) {
4df27c4d
YZ		 1614 if (ret == -EEXIST) {
1615 free_fs_root(root);
1616 goto again;
1617 }
1618 goto fail;
0f7d52f4 1619 }
edbd8d4e 1620 return root;
4df27c4d
YZ		 1621fail:
1622 free_fs_root(root);
1623 return ERR_PTR(ret);
edbd8d4e
CM		 1624}
1625
04160088
CM		 1626static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1627{
1628 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1629 int ret = 0;
04160088
CM		 1630 struct btrfs_device *device;
1631 struct backing_dev_info *bdi;
b7967db7 1632
1f78160c
XG		 1633 rcu_read_lock();
1634 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1635 if (!device->bdev)
1636 continue;
04160088
CM		 1637 bdi = blk_get_backing_dev_info(device->bdev);
1638 if (bdi && bdi_congested(bdi, bdi_bits)) {
1639 ret = 1;
1640 break;
1641 }
1642 }
1f78160c 1643 rcu_read_unlock();
04160088
CM		 1644 return ret;
1645}
1646
ad081f14
JA		 1647/*
1648 * If this fails, caller must call bdi_destroy() to get rid of the
1649 * bdi again.
1650 */
04160088
CM		 1651static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1652{
ad081f14
JA		 1653 int err;
1654
1655 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1656 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1657 if (err)
1658 return err;
1659
4575c9cc 1660 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1661 bdi->congested_fn = btrfs_congested_fn;
1662 bdi->congested_data = info;
1663 return 0;
1664}
1665
8b712842
CM		 1666/*
1667 * called by the kthread helper functions to finally call the bio end_io
1668 * functions. This is where read checksum verification actually happens
1669 */
1670static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1671{
ce9adaa5 1672 struct bio *bio;
8b712842 1673 struct end_io_wq *end_io_wq;
ce9adaa5 1674 int error;
ce9adaa5 1675
8b712842
CM		 1676 end_io_wq = container_of(work, struct end_io_wq, work);
1677 bio = end_io_wq->bio;
ce9adaa5 1678
8b712842
CM		 1679 error = end_io_wq->error;
1680 bio->bi_private = end_io_wq->private;
1681 bio->bi_end_io = end_io_wq->end_io;
1682 kfree(end_io_wq);
8b712842 1683 bio_endio(bio, error);
44b8bd7e
CM		 1684}
1685
a74a4b97
CM		 1686static int cleaner_kthread(void *arg)
1687{
1688 struct btrfs_root *root = arg;
d0278245 1689 int again;
a74a4b97
CM		 1690
1691 do {
d0278245 1692 again = 0;
a74a4b97 1693
d0278245 1694 /* Make the cleaner go to sleep early. */
babbf170 1695 if (btrfs_need_cleaner_sleep(root))
d0278245
MX		 1696 goto sleep;
1697
1698 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1699 goto sleep;
1700
dc7f370c
MX		 1701 /*
1702 * Avoid the problem that we change the status of the fs
1703 * during the above check and trylock.
1704 */
babbf170 1705 if (btrfs_need_cleaner_sleep(root)) {
dc7f370c
MX		 1706 mutex_unlock(&root->fs_info->cleaner_mutex);
1707 goto sleep;
76dda93c 1708 }
a74a4b97 1709
d0278245
MX		 1710 btrfs_run_delayed_iputs(root);
1711 again = btrfs_clean_one_deleted_snapshot(root);
1712 mutex_unlock(&root->fs_info->cleaner_mutex);
1713
1714 /*
05323cd1
MX		 1715 * The defragger has dealt with the R/O remount and umount,
1716 * needn't do anything special here.
d0278245
MX		 1717 */
1718 btrfs_run_defrag_inodes(root->fs_info);
1719sleep:
9d1a2a3a 1720 if (!try_to_freeze() && !again) {
a74a4b97 1721 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1722 if (!kthread_should_stop())
1723 schedule();
a74a4b97
CM		 1724 __set_current_state(TASK_RUNNING);
1725 }
1726 } while (!kthread_should_stop());
1727 return 0;
1728}
1729
1730static int transaction_kthread(void *arg)
1731{
1732 struct btrfs_root *root = arg;
1733 struct btrfs_trans_handle *trans;
1734 struct btrfs_transaction *cur;
8929ecfa 1735 u64 transid;
a74a4b97
CM		 1736 unsigned long now;
1737 unsigned long delay;
914b2007 1738 bool cannot_commit;
a74a4b97
CM		 1739
1740 do {
914b2007 1741 cannot_commit = false;
8b87dc17 1742 delay = HZ * root->fs_info->commit_interval;
a74a4b97
CM		 1743 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1744
a4abeea4 1745 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1746 cur = root->fs_info->running_transaction;
1747 if (!cur) {
a4abeea4 1748 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1749 goto sleep;
1750 }
31153d81 1751
a74a4b97 1752 now = get_seconds();
4a9d8bde 1753 if (cur->state < TRANS_STATE_BLOCKED &&
8b87dc17
DS		 1754 (now < cur->start_time ||
1755 now - cur->start_time < root->fs_info->commit_interval)) {
a4abeea4 1756 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1757 delay = HZ * 5;
1758 goto sleep;
1759 }
8929ecfa 1760 transid = cur->transid;
a4abeea4 1761 spin_unlock(&root->fs_info->trans_lock);
56bec294 1762
79787eaa 1763 /* If the file system is aborted, this will always fail. */
354aa0fb 1764 trans = btrfs_attach_transaction(root);
914b2007 1765 if (IS_ERR(trans)) {
354aa0fb
MX		 1766 if (PTR_ERR(trans) != -ENOENT)
1767 cannot_commit = true;
79787eaa 1768 goto sleep;
914b2007 1769 }
8929ecfa 1770 if (transid == trans->transid) {
79787eaa 1771 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1772 } else {
1773 btrfs_end_transaction(trans, root);
1774 }
a74a4b97
CM		 1775sleep:
1776 wake_up_process(root->fs_info->cleaner_kthread);
1777 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1778
4e121c06
JB		 1779 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1780 &root->fs_info->fs_state)))
1781 btrfs_cleanup_transaction(root);
a0acae0e 1782 if (!try_to_freeze()) {
a74a4b97 1783 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1784 if (!kthread_should_stop() &&
914b2007
JK		 1785 (!btrfs_transaction_blocked(root->fs_info) ||
1786 cannot_commit))
8929ecfa 1787 schedule_timeout(delay);
a74a4b97
CM		 1788 __set_current_state(TASK_RUNNING);
1789 }
1790 } while (!kthread_should_stop());
1791 return 0;
1792}
1793
af31f5e5
CM		 1794/*
1795 * this will find the highest generation in the array of
1796 * root backups. The index of the highest array is returned,
1797 * or -1 if we can't find anything.
1798 *
1799 * We check to make sure the array is valid by comparing the
1800 * generation of the latest root in the array with the generation
1801 * in the super block. If they don't match we pitch it.
1802 */
1803static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1804{
1805 u64 cur;
1806 int newest_index = -1;
1807 struct btrfs_root_backup *root_backup;
1808 int i;
1809
1810 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1811 root_backup = info->super_copy->super_roots + i;
1812 cur = btrfs_backup_tree_root_gen(root_backup);
1813 if (cur == newest_gen)
1814 newest_index = i;
1815 }
1816
1817 /* check to see if we actually wrapped around */
1818 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1819 root_backup = info->super_copy->super_roots;
1820 cur = btrfs_backup_tree_root_gen(root_backup);
1821 if (cur == newest_gen)
1822 newest_index = 0;
1823 }
1824 return newest_index;
1825}
1826
1827
1828/*
1829 * find the oldest backup so we know where to store new entries
1830 * in the backup array. This will set the backup_root_index
1831 * field in the fs_info struct
1832 */
1833static void find_oldest_super_backup(struct btrfs_fs_info *info,
1834 u64 newest_gen)
1835{
1836 int newest_index = -1;
1837
1838 newest_index = find_newest_super_backup(info, newest_gen);
1839 /* if there was garbage in there, just move along */
1840 if (newest_index == -1) {
1841 info->backup_root_index = 0;
1842 } else {
1843 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1844 }
1845}
1846
1847/*
1848 * copy all the root pointers into the super backup array.
1849 * this will bump the backup pointer by one when it is
1850 * done
1851 */
1852static void backup_super_roots(struct btrfs_fs_info *info)
1853{
1854 int next_backup;
1855 struct btrfs_root_backup *root_backup;
1856 int last_backup;
1857
1858 next_backup = info->backup_root_index;
1859 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1860 BTRFS_NUM_BACKUP_ROOTS;
1861
1862 /*
1863 * just overwrite the last backup if we're at the same generation
1864 * this happens only at umount
1865 */
1866 root_backup = info->super_for_commit->super_roots + last_backup;
1867 if (btrfs_backup_tree_root_gen(root_backup) ==
1868 btrfs_header_generation(info->tree_root->node))
1869 next_backup = last_backup;
1870
1871 root_backup = info->super_for_commit->super_roots + next_backup;
1872
1873 /*
1874 * make sure all of our padding and empty slots get zero filled
1875 * regardless of which ones we use today
1876 */
1877 memset(root_backup, 0, sizeof(*root_backup));
1878
1879 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1880
1881 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1882 btrfs_set_backup_tree_root_gen(root_backup,
1883 btrfs_header_generation(info->tree_root->node));
1884
1885 btrfs_set_backup_tree_root_level(root_backup,
1886 btrfs_header_level(info->tree_root->node));
1887
1888 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1889 btrfs_set_backup_chunk_root_gen(root_backup,
1890 btrfs_header_generation(info->chunk_root->node));
1891 btrfs_set_backup_chunk_root_level(root_backup,
1892 btrfs_header_level(info->chunk_root->node));
1893
1894 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1895 btrfs_set_backup_extent_root_gen(root_backup,
1896 btrfs_header_generation(info->extent_root->node));
1897 btrfs_set_backup_extent_root_level(root_backup,
1898 btrfs_header_level(info->extent_root->node));
1899
7c7e82a7
CM		 1900 /*
1901 * we might commit during log recovery, which happens before we set
1902 * the fs_root. Make sure it is valid before we fill it in.
1903 */
1904 if (info->fs_root && info->fs_root->node) {
1905 btrfs_set_backup_fs_root(root_backup,
1906 info->fs_root->node->start);
1907 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1908 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1909 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1910 btrfs_header_level(info->fs_root->node));
7c7e82a7 1911 }
af31f5e5
CM		 1912
1913 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1914 btrfs_set_backup_dev_root_gen(root_backup,
1915 btrfs_header_generation(info->dev_root->node));
1916 btrfs_set_backup_dev_root_level(root_backup,
1917 btrfs_header_level(info->dev_root->node));
1918
1919 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1920 btrfs_set_backup_csum_root_gen(root_backup,
1921 btrfs_header_generation(info->csum_root->node));
1922 btrfs_set_backup_csum_root_level(root_backup,
1923 btrfs_header_level(info->csum_root->node));
1924
1925 btrfs_set_backup_total_bytes(root_backup,
1926 btrfs_super_total_bytes(info->super_copy));
1927 btrfs_set_backup_bytes_used(root_backup,
1928 btrfs_super_bytes_used(info->super_copy));
1929 btrfs_set_backup_num_devices(root_backup,
1930 btrfs_super_num_devices(info->super_copy));
1931
1932 /*
1933 * if we don't copy this out to the super_copy, it won't get remembered
1934 * for the next commit
1935 */
1936 memcpy(&info->super_copy->super_roots,
1937 &info->super_for_commit->super_roots,
1938 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1939}
1940
1941/*
1942 * this copies info out of the root backup array and back into
1943 * the in-memory super block. It is meant to help iterate through
1944 * the array, so you send it the number of backups you've already
1945 * tried and the last backup index you used.
1946 *
1947 * this returns -1 when it has tried all the backups
1948 */
1949static noinline int next_root_backup(struct btrfs_fs_info *info,
1950 struct btrfs_super_block *super,
1951 int *num_backups_tried, int *backup_index)
1952{
1953 struct btrfs_root_backup *root_backup;
1954 int newest = *backup_index;
1955
1956 if (*num_backups_tried == 0) {
1957 u64 gen = btrfs_super_generation(super);
1958
1959 newest = find_newest_super_backup(info, gen);
1960 if (newest == -1)
1961 return -1;
1962
1963 *backup_index = newest;
1964 *num_backups_tried = 1;
1965 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1966 /* we've tried all the backups, all done */
1967 return -1;
1968 } else {
1969 /* jump to the next oldest backup */
1970 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1971 BTRFS_NUM_BACKUP_ROOTS;
1972 *backup_index = newest;
1973 *num_backups_tried += 1;
1974 }
1975 root_backup = super->super_roots + newest;
1976
1977 btrfs_set_super_generation(super,
1978 btrfs_backup_tree_root_gen(root_backup));
1979 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1980 btrfs_set_super_root_level(super,
1981 btrfs_backup_tree_root_level(root_backup));
1982 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1983
1984 /*
1985 * fixme: the total bytes and num_devices need to match or we should
1986 * need a fsck
1987 */
1988 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1989 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1990 return 0;
1991}
1992
7abadb64
LB		 1993/* helper to cleanup workers */
1994static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1995{
1996 btrfs_stop_workers(&fs_info->generic_worker);
1997 btrfs_stop_workers(&fs_info->fixup_workers);
1998 btrfs_stop_workers(&fs_info->delalloc_workers);
1999 btrfs_stop_workers(&fs_info->workers);
2000 btrfs_stop_workers(&fs_info->endio_workers);
2001 btrfs_stop_workers(&fs_info->endio_meta_workers);
2002 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2003 btrfs_stop_workers(&fs_info->rmw_workers);
2004 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2005 btrfs_stop_workers(&fs_info->endio_write_workers);
2006 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2007 btrfs_stop_workers(&fs_info->submit_workers);
2008 btrfs_stop_workers(&fs_info->delayed_workers);
2009 btrfs_stop_workers(&fs_info->caching_workers);
2010 btrfs_stop_workers(&fs_info->readahead_workers);
2011 btrfs_stop_workers(&fs_info->flush_workers);
2f232036 2012 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
7abadb64
LB		 2013}
2014
2e9f5954
R		 2015static void free_root_extent_buffers(struct btrfs_root *root)
2016{
2017 if (root) {
2018 free_extent_buffer(root->node);
2019 free_extent_buffer(root->commit_root);
2020 root->node = NULL;
2021 root->commit_root = NULL;
2022 }
2023}
2024
af31f5e5
CM		 2025/* helper to cleanup tree roots */
2026static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2027{
2e9f5954 2028 free_root_extent_buffers(info->tree_root);
655b09fe 2029
2e9f5954
R		 2030 free_root_extent_buffers(info->dev_root);
2031 free_root_extent_buffers(info->extent_root);
2032 free_root_extent_buffers(info->csum_root);
2033 free_root_extent_buffers(info->quota_root);
2034 free_root_extent_buffers(info->uuid_root);
2035 if (chunk_root)
2036 free_root_extent_buffers(info->chunk_root);
af31f5e5
CM		 2037}
2038
171f6537
JB		 2039static void del_fs_roots(struct btrfs_fs_info *fs_info)
2040{
2041 int ret;
2042 struct btrfs_root *gang[8];
2043 int i;
2044
2045 while (!list_empty(&fs_info->dead_roots)) {
2046 gang[0] = list_entry(fs_info->dead_roots.next,
2047 struct btrfs_root, root_list);
2048 list_del(&gang[0]->root_list);
2049
2050 if (gang[0]->in_radix) {
cb517eab 2051 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
171f6537
JB		 2052 } else {
2053 free_extent_buffer(gang[0]->node);
2054 free_extent_buffer(gang[0]->commit_root);
b0feb9d9 2055 btrfs_put_fs_root(gang[0]);
171f6537
JB		 2056 }
2057 }
2058
2059 while (1) {
2060 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2061 (void **)gang, 0,
2062 ARRAY_SIZE(gang));
2063 if (!ret)
2064 break;
2065 for (i = 0; i < ret; i++)
cb517eab 2066 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
171f6537 2067 }
1a4319cc
LB		 2068
2069 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2070 btrfs_free_log_root_tree(NULL, fs_info);
2071 btrfs_destroy_pinned_extent(fs_info->tree_root,
2072 fs_info->pinned_extents);
2073 }
171f6537 2074}
af31f5e5 2075
ad2b2c80
AV		 2076int open_ctree(struct super_block *sb,
2077 struct btrfs_fs_devices *fs_devices,
2078 char *options)
2e635a27 2079{
db94535d
CM		 2080 u32 sectorsize;
2081 u32 nodesize;
2082 u32 leafsize;
2083 u32 blocksize;
87ee04eb 2084 u32 stripesize;
84234f3a 2085 u64 generation;
f2b636e8 2086 u64 features;
3de4586c 2087 struct btrfs_key location;
a061fc8d 2088 struct buffer_head *bh;
4d34b278 2089 struct btrfs_super_block *disk_super;
815745cf 2090 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2091 struct btrfs_root *tree_root;
4d34b278
ID		 2092 struct btrfs_root *extent_root;
2093 struct btrfs_root *csum_root;
2094 struct btrfs_root *chunk_root;
2095 struct btrfs_root *dev_root;
bcef60f2 2096 struct btrfs_root *quota_root;
f7a81ea4 2097 struct btrfs_root *uuid_root;
e02119d5 2098 struct btrfs_root *log_tree_root;
eb60ceac 2099 int ret;
e58ca020 2100 int err = -EINVAL;
af31f5e5
CM		 2101 int num_backups_tried = 0;
2102 int backup_index = 0;
70f80175
SB		 2103 bool create_uuid_tree;
2104 bool check_uuid_tree;
4543df7e 2105
f84a8bd6 2106 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e 2107 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
cb517eab 2108 if (!tree_root || !chunk_root) {
39279cc3
CM		 2109 err = -ENOMEM;
2110 goto fail;
2111 }
76dda93c
YZ		 2112
2113 ret = init_srcu_struct(&fs_info->subvol_srcu);
2114 if (ret) {
2115 err = ret;
2116 goto fail;
2117 }
2118
2119 ret = setup_bdi(fs_info, &fs_info->bdi);
2120 if (ret) {
2121 err = ret;
2122 goto fail_srcu;
2123 }
2124
e2d84521
MX		 2125 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2126 if (ret) {
2127 err = ret;
2128 goto fail_bdi;
2129 }
2130 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2131 (1 + ilog2(nr_cpu_ids));
2132
963d678b
MX		 2133 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2134 if (ret) {
2135 err = ret;
2136 goto fail_dirty_metadata_bytes;
2137 }
2138
c404e0dc
MX		 2139 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2140 if (ret) {
2141 err = ret;
2142 goto fail_delalloc_bytes;
2143 }
2144
76dda93c
YZ		 2145 fs_info->btree_inode = new_inode(sb);
2146 if (!fs_info->btree_inode) {
2147 err = -ENOMEM;
c404e0dc 2148 goto fail_bio_counter;
76dda93c
YZ		 2149 }
2150
a6591715 2151 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2152
76dda93c 2153 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
f28491e0 2154 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
8fd17795 2155 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2156 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2157 INIT_LIST_HEAD(&fs_info->delayed_iputs);
eb73c1b7 2158 INIT_LIST_HEAD(&fs_info->delalloc_roots);
11833d66 2159 INIT_LIST_HEAD(&fs_info->caching_block_groups);
eb73c1b7 2160 spin_lock_init(&fs_info->delalloc_root_lock);
a4abeea4 2161 spin_lock_init(&fs_info->trans_lock);
76dda93c 2162 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2163 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2164 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2165 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2166 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2167 spin_lock_init(&fs_info->super_lock);
f28491e0 2168 spin_lock_init(&fs_info->buffer_lock);
f29021b2 2169 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2170 mutex_init(&fs_info->reloc_mutex);
de98ced9 2171 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2172
58176a96 2173 init_completion(&fs_info->kobj_unregister);
0b86a832 2174 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2175 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2176 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2177 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2178 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2179 BTRFS_BLOCK_RSV_GLOBAL);
2180 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2181 BTRFS_BLOCK_RSV_DELALLOC);
2182 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2183 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2184 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2185 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2186 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2187 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2188 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2189 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2190 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2191 atomic_set(&fs_info->defrag_running, 0);
fc36ed7e 2192 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2193 fs_info->sb = sb;
6f568d35 2194 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2195 fs_info->metadata_ratio = 0;
4cb5300b 2196 fs_info->defrag_inodes = RB_ROOT;
2bf64758 2197 fs_info->free_chunk_space = 0;
f29021b2 2198 fs_info->tree_mod_log = RB_ROOT;
8b87dc17 2199 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
0a2b2a84 2200 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
90519d66
AJ		 2201 /* readahead state */
2202 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2203 spin_lock_init(&fs_info->reada_lock);
c8b97818 2204
b34b086c
CM		 2205 fs_info->thread_pool_size = min_t(unsigned long,
2206 num_online_cpus() + 2, 8);
0afbaf8c 2207
199c2a9c
MX		 2208 INIT_LIST_HEAD(&fs_info->ordered_roots);
2209 spin_lock_init(&fs_info->ordered_root_lock);
16cdcec7
MX		 2210 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2211 GFP_NOFS);
2212 if (!fs_info->delayed_root) {
2213 err = -ENOMEM;
2214 goto fail_iput;
2215 }
2216 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2217
a2de733c
AJ		 2218 mutex_init(&fs_info->scrub_lock);
2219 atomic_set(&fs_info->scrubs_running, 0);
2220 atomic_set(&fs_info->scrub_pause_req, 0);
2221 atomic_set(&fs_info->scrubs_paused, 0);
2222 atomic_set(&fs_info->scrub_cancel_req, 0);
c404e0dc 2223 init_waitqueue_head(&fs_info->replace_wait);
a2de733c 2224 init_waitqueue_head(&fs_info->scrub_pause_wait);
a2de733c 2225 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2226#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2227 fs_info->check_integrity_print_mask = 0;
2228#endif
a2de733c 2229
c9e9f97b
ID		 2230 spin_lock_init(&fs_info->balance_lock);
2231 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2232 atomic_set(&fs_info->balance_running, 0);
2233 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2234 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2235 fs_info->balance_ctl = NULL;
837d5b6e 2236 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2237
a061fc8d
CM		 2238 sb->s_blocksize = 4096;
2239 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2240 sb->s_bdi = &fs_info->bdi;
a061fc8d 2241
76dda93c 2242 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2243 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2244 /*
2245 * we set the i_size on the btree inode to the max possible int.
2246 * the real end of the address space is determined by all of
2247 * the devices in the system
2248 */
2249 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2250 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2251 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2252
5d4f98a2 2253 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2254 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2255 fs_info->btree_inode->i_mapping);
0b32f4bb 2256 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2257 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2258
2259 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2260
76dda93c
YZ		 2261 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2262 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2263 sizeof(struct btrfs_key));
72ac3c0d
JB		 2264 set_bit(BTRFS_INODE_DUMMY,
2265 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
778ba82b 2266 btrfs_insert_inode_hash(fs_info->btree_inode);
76dda93c 2267
0f9dd46c 2268 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2269 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2270 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2271
11833d66 2272 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2273 fs_info->btree_inode->i_mapping);
11833d66 2274 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2275 fs_info->btree_inode->i_mapping);
11833d66 2276 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2277 fs_info->do_barriers = 1;
e18e4809 2278
39279cc3 2279
5a3f23d5 2280 mutex_init(&fs_info->ordered_operations_mutex);
9ffba8cd 2281 mutex_init(&fs_info->ordered_extent_flush_mutex);
e02119d5 2282 mutex_init(&fs_info->tree_log_mutex);
925baedd 2283 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2284 mutex_init(&fs_info->transaction_kthread_mutex);
2285 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2286 mutex_init(&fs_info->volume_mutex);
276e680d 2287 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2288 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2289 init_rwsem(&fs_info->subvol_sem);
803b2f54 2290 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
e922e087
SB		 2291 fs_info->dev_replace.lock_owner = 0;
2292 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2293 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2294 mutex_init(&fs_info->dev_replace.lock_management_lock);
2295 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2296
416ac51d 2297 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2298 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2299 fs_info->qgroup_tree = RB_ROOT;
2300 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2301 fs_info->qgroup_seq = 1;
2302 fs_info->quota_enabled = 0;
2303 fs_info->pending_quota_state = 0;
1e8f9158 2304 fs_info->qgroup_ulist = NULL;
2f232036 2305 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2306
fa9c0d79
CM		 2307 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2308 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2309
e6dcd2dc 2310 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2311 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2312 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2313 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2314
53b381b3
DW		 2315 ret = btrfs_alloc_stripe_hash_table(fs_info);
2316 if (ret) {
83c8266a 2317 err = ret;
53b381b3
DW		 2318 goto fail_alloc;
2319 }
2320
0b86a832 2321 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2322 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2323
3c4bb26b 2324 invalidate_bdev(fs_devices->latest_bdev);
1104a885
DS		 2325
2326 /*
2327 * Read super block and check the signature bytes only
2328 */
a512bbf8 2329 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2330 if (!bh) {
2331 err = -EINVAL;
16cdcec7 2332 goto fail_alloc;
20b45077 2333 }
39279cc3 2334
1104a885
DS		 2335 /*
2336 * We want to check superblock checksum, the type is stored inside.
2337 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2338 */
2339 if (btrfs_check_super_csum(bh->b_data)) {
efe120a0 2340 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
1104a885
DS		 2341 err = -EINVAL;
2342 goto fail_alloc;
2343 }
2344
2345 /*
2346 * super_copy is zeroed at allocation time and we never touch the
2347 * following bytes up to INFO_SIZE, the checksum is calculated from
2348 * the whole block of INFO_SIZE
2349 */
6c41761f
DS		 2350 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2351 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2352 sizeof(*fs_info->super_for_commit));
a061fc8d 2353 brelse(bh);
5f39d397 2354
6c41761f 2355 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2356
1104a885
DS		 2357 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2358 if (ret) {
efe120a0 2359 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
1104a885
DS		 2360 err = -EINVAL;
2361 goto fail_alloc;
2362 }
2363
6c41761f 2364 disk_super = fs_info->super_copy;
0f7d52f4 2365 if (!btrfs_super_root(disk_super))
16cdcec7 2366 goto fail_alloc;
0f7d52f4 2367
acce952b 2368 /* check FS state, whether FS is broken. */
87533c47
MX		 2369 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2370 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2371
af31f5e5
CM		 2372 /*
2373 * run through our array of backup supers and setup
2374 * our ring pointer to the oldest one
2375 */
2376 generation = btrfs_super_generation(disk_super);
2377 find_oldest_super_backup(fs_info, generation);
2378
75e7cb7f
LB		 2379 /*
2380 * In the long term, we'll store the compression type in the super
2381 * block, and it'll be used for per file compression control.
2382 */
2383 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2384
2b82032c
YZ		 2385 ret = btrfs_parse_options(tree_root, options);
2386 if (ret) {
2387 err = ret;
16cdcec7 2388 goto fail_alloc;
2b82032c 2389 }
dfe25020 2390
f2b636e8
JB		 2391 features = btrfs_super_incompat_flags(disk_super) &
2392 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2393 if (features) {
2394 printk(KERN_ERR "BTRFS: couldn't mount because of "
2395 "unsupported optional features (%Lx).\n",
c1c9ff7c 2396 features);
f2b636e8 2397 err = -EINVAL;
16cdcec7 2398 goto fail_alloc;
f2b636e8
JB		 2399 }
2400
727011e0
CM		 2401 if (btrfs_super_leafsize(disk_super) !=
2402 btrfs_super_nodesize(disk_super)) {
2403 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2404 "blocksizes don't match. node %d leaf %d\n",
2405 btrfs_super_nodesize(disk_super),
2406 btrfs_super_leafsize(disk_super));
2407 err = -EINVAL;
2408 goto fail_alloc;
2409 }
2410 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2411 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2412 "blocksize (%d) was too large\n",
2413 btrfs_super_leafsize(disk_super));
2414 err = -EINVAL;
2415 goto fail_alloc;
2416 }
2417
5d4f98a2 2418 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2419 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2420 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2421 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2422
3173a18f 2423 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
efe120a0 2424 printk(KERN_ERR "BTRFS: has skinny extents\n");
3173a18f 2425
727011e0
CM		 2426 /*
2427 * flag our filesystem as having big metadata blocks if
2428 * they are bigger than the page size
2429 */
2430 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2431 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
efe120a0 2432 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
727011e0
CM		 2433 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2434 }
2435
bc3f116f
CM		 2436 nodesize = btrfs_super_nodesize(disk_super);
2437 leafsize = btrfs_super_leafsize(disk_super);
2438 sectorsize = btrfs_super_sectorsize(disk_super);
2439 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2440 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2441 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2442
2443 /*
2444 * mixed block groups end up with duplicate but slightly offset
2445 * extent buffers for the same range. It leads to corruptions
2446 */
2447 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2448 (sectorsize != leafsize)) {
efe120a0 2449 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
bc3f116f
CM		 2450 "are not allowed for mixed block groups on %s\n",
2451 sb->s_id);
2452 goto fail_alloc;
2453 }
2454
ceda0864
MX		 2455 /*
2456 * Needn't use the lock because there is no other task which will
2457 * update the flag.
2458 */
a6fa6fae 2459 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2460
f2b636e8
JB		 2461 features = btrfs_super_compat_ro_flags(disk_super) &
2462 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2463 if (!(sb->s_flags & MS_RDONLY) && features) {
2464 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2465 "unsupported option features (%Lx).\n",
c1c9ff7c 2466 features);
f2b636e8 2467 err = -EINVAL;
16cdcec7 2468 goto fail_alloc;
f2b636e8 2469 }
61d92c32
CM		 2470
2471 btrfs_init_workers(&fs_info->generic_worker,
2472 "genwork", 1, NULL);
2473
5443be45 2474 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 2475 fs_info->thread_pool_size,
2476 &fs_info->generic_worker);
c8b97818 2477
771ed689 2478 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
45d5fd14 2479 fs_info->thread_pool_size, NULL);
771ed689 2480
8ccf6f19 2481 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
45d5fd14 2482 fs_info->thread_pool_size, NULL);
8ccf6f19 2483
5443be45 2484 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 2485 min_t(u64, fs_devices->num_devices,
45d5fd14 2486 fs_info->thread_pool_size), NULL);
61b49440 2487
bab39bf9 2488 btrfs_init_workers(&fs_info->caching_workers, "cache",
45d5fd14 2489 fs_info->thread_pool_size, NULL);
bab39bf9 2490
61b49440
CM		 2491 /* a higher idle thresh on the submit workers makes it much more
2492 * likely that bios will be send down in a sane order to the
2493 * devices
2494 */
2495 fs_info->submit_workers.idle_thresh = 64;
53863232 2496
771ed689 2497 fs_info->workers.idle_thresh = 16;
4a69a410 2498 fs_info->workers.ordered = 1;
61b49440 2499
771ed689
CM		 2500 fs_info->delalloc_workers.idle_thresh = 2;
2501 fs_info->delalloc_workers.ordered = 1;
2502
61d92c32
CM		 2503 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2504 &fs_info->generic_worker);
5443be45 2505 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2506 fs_info->thread_pool_size,
2507 &fs_info->generic_worker);
d20f7043 2508 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2509 fs_info->thread_pool_size,
2510 &fs_info->generic_worker);
cad321ad 2511 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2512 "endio-meta-write", fs_info->thread_pool_size,
2513 &fs_info->generic_worker);
53b381b3
DW		 2514 btrfs_init_workers(&fs_info->endio_raid56_workers,
2515 "endio-raid56", fs_info->thread_pool_size,
2516 &fs_info->generic_worker);
2517 btrfs_init_workers(&fs_info->rmw_workers,
2518 "rmw", fs_info->thread_pool_size,
2519 &fs_info->generic_worker);
5443be45 2520 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2521 fs_info->thread_pool_size,
2522 &fs_info->generic_worker);
0cb59c99
JB		 2523 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2524 1, &fs_info->generic_worker);
16cdcec7
MX		 2525 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2526 fs_info->thread_pool_size,
2527 &fs_info->generic_worker);
90519d66
AJ		 2528 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2529 fs_info->thread_pool_size,
2530 &fs_info->generic_worker);
2f232036
JS		 2531 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2532 &fs_info->generic_worker);
61b49440
CM		 2533
2534 /*
2535 * endios are largely parallel and should have a very
2536 * low idle thresh
2537 */
2538 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2539 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2540 fs_info->endio_raid56_workers.idle_thresh = 4;
2541 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2542
9042846b
CM		 2543 fs_info->endio_write_workers.idle_thresh = 2;
2544 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2545 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2546
0dc3b84a
JB		 2547 /*
2548 * btrfs_start_workers can really only fail because of ENOMEM so just
2549 * return -ENOMEM if any of these fail.
2550 */
2551 ret = btrfs_start_workers(&fs_info->workers);
2552 ret |= btrfs_start_workers(&fs_info->generic_worker);
2553 ret |= btrfs_start_workers(&fs_info->submit_workers);
2554 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2555 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2556 ret |= btrfs_start_workers(&fs_info->endio_workers);
2557 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2558 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2559 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2560 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2561 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2562 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2563 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2564 ret |= btrfs_start_workers(&fs_info->caching_workers);
2565 ret |= btrfs_start_workers(&fs_info->readahead_workers);
8ccf6f19 2566 ret |= btrfs_start_workers(&fs_info->flush_workers);
2f232036 2567 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
0dc3b84a 2568 if (ret) {
fed425c7 2569 err = -ENOMEM;
0dc3b84a
JB		 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
MX		 3810
3811 btrfs_destroy_ordered_extents(root);
3812
3813 cond_resched_lock(&fs_info->ordered_root_lock);
3814 }
3815 spin_unlock(&fs_info->ordered_root_lock);
acce952b 3816}
3817
35a3621b
SB		 3818static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3819 struct btrfs_root *root)
acce952b 3820{
3821 struct rb_node *node;
3822 struct btrfs_delayed_ref_root *delayed_refs;
3823 struct btrfs_delayed_ref_node *ref;
3824 int ret = 0;
3825
3826 delayed_refs = &trans->delayed_refs;
3827
3828 spin_lock(&delayed_refs->lock);
d7df2c79 3829 if (atomic_read(&delayed_refs->num_entries) == 0) {
cfece4db 3830 spin_unlock(&delayed_refs->lock);
efe120a0 3831 btrfs_info(root->fs_info, "delayed_refs has NO entry");
acce952b 3832 return ret;
3833 }
3834
d7df2c79
JB		 3835 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3836 struct btrfs_delayed_ref_head *head;
e78417d1 3837 bool pin_bytes = false;
acce952b 3838
d7df2c79
JB		 3839 head = rb_entry(node, struct btrfs_delayed_ref_head,
3840 href_node);
3841 if (!mutex_trylock(&head->mutex)) {
3842 atomic_inc(&head->node.refs);
3843 spin_unlock(&delayed_refs->lock);
acce952b 3844
d7df2c79 3845 mutex_lock(&head->mutex);
e78417d1 3846 mutex_unlock(&head->mutex);
d7df2c79
JB		 3847 btrfs_put_delayed_ref(&head->node);
3848 spin_lock(&delayed_refs->lock);
3849 continue;
3850 }
3851 spin_lock(&head->lock);
3852 while ((node = rb_first(&head->ref_root)) != NULL) {
3853 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3854 rb_node);
3855 ref->in_tree = 0;
3856 rb_erase(&ref->rb_node, &head->ref_root);
3857 atomic_dec(&delayed_refs->num_entries);
3858 btrfs_put_delayed_ref(ref);
e78417d1 3859 }
d7df2c79
JB		 3860 if (head->must_insert_reserved)
3861 pin_bytes = true;
3862 btrfs_free_delayed_extent_op(head->extent_op);
3863 delayed_refs->num_heads--;
3864 if (head->processing == 0)
3865 delayed_refs->num_heads_ready--;
3866 atomic_dec(&delayed_refs->num_entries);
3867 head->node.in_tree = 0;
3868 rb_erase(&head->href_node, &delayed_refs->href_root);
3869 spin_unlock(&head->lock);
3870 spin_unlock(&delayed_refs->lock);
3871 mutex_unlock(&head->mutex);
acce952b 3872
d7df2c79
JB		 3873 if (pin_bytes)
3874 btrfs_pin_extent(root, head->node.bytenr,
3875 head->node.num_bytes, 1);
3876 btrfs_put_delayed_ref(&head->node);
acce952b 3877 cond_resched();
3878 spin_lock(&delayed_refs->lock);
3879 }
3880
3881 spin_unlock(&delayed_refs->lock);
3882
3883 return ret;
3884}
3885
143bede5 3886static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3887{
3888 struct btrfs_inode *btrfs_inode;
3889 struct list_head splice;
3890
3891 INIT_LIST_HEAD(&splice);
3892
eb73c1b7
MX		 3893 spin_lock(&root->delalloc_lock);
3894 list_splice_init(&root->delalloc_inodes, &splice);
acce952b 3895
3896 while (!list_empty(&splice)) {
eb73c1b7
MX		 3897 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3898 delalloc_inodes);
acce952b 3899
3900 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3901 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3902 &btrfs_inode->runtime_flags);
eb73c1b7 3903 spin_unlock(&root->delalloc_lock);
acce952b 3904
3905 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3906
eb73c1b7 3907 spin_lock(&root->delalloc_lock);
acce952b 3908 }
3909
eb73c1b7
MX		 3910 spin_unlock(&root->delalloc_lock);
3911}
3912
3913static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3914{
3915 struct btrfs_root *root;
3916 struct list_head splice;
3917
3918 INIT_LIST_HEAD(&splice);
3919
3920 spin_lock(&fs_info->delalloc_root_lock);
3921 list_splice_init(&fs_info->delalloc_roots, &splice);
3922 while (!list_empty(&splice)) {
3923 root = list_first_entry(&splice, struct btrfs_root,
3924 delalloc_root);
3925 list_del_init(&root->delalloc_root);
3926 root = btrfs_grab_fs_root(root);
3927 BUG_ON(!root);
3928 spin_unlock(&fs_info->delalloc_root_lock);
3929
3930 btrfs_destroy_delalloc_inodes(root);
3931 btrfs_put_fs_root(root);
3932
3933 spin_lock(&fs_info->delalloc_root_lock);
3934 }
3935 spin_unlock(&fs_info->delalloc_root_lock);
acce952b 3936}
3937
3938static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3939 struct extent_io_tree *dirty_pages,
3940 int mark)
3941{
3942 int ret;
acce952b 3943 struct extent_buffer *eb;
3944 u64 start = 0;
3945 u64 end;
acce952b 3946
3947 while (1) {
3948 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3949 mark, NULL);
acce952b 3950 if (ret)
3951 break;
3952
3953 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3954 while (start <= end) {
fd8b2b61
JB		 3955 eb = btrfs_find_tree_block(root, start,
3956 root->leafsize);
69a85bd8 3957 start += root->leafsize;
fd8b2b61 3958 if (!eb)
acce952b 3959 continue;
fd8b2b61 3960 wait_on_extent_buffer_writeback(eb);
acce952b 3961
fd8b2b61
JB		 3962 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3963 &eb->bflags))
3964 clear_extent_buffer_dirty(eb);
3965 free_extent_buffer_stale(eb);
acce952b 3966 }
3967 }
3968
3969 return ret;
3970}
3971
3972static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3973 struct extent_io_tree *pinned_extents)
3974{
3975 struct extent_io_tree *unpin;
3976 u64 start;
3977 u64 end;
3978 int ret;
ed0eaa14 3979 bool loop = true;
acce952b 3980
3981 unpin = pinned_extents;
ed0eaa14 3982again:
acce952b 3983 while (1) {
3984 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 3985 EXTENT_DIRTY, NULL);
acce952b 3986 if (ret)
3987 break;
3988
3989 /* opt_discard */
5378e607
LD		 3990 if (btrfs_test_opt(root, DISCARD))
3991 ret = btrfs_error_discard_extent(root, start,
3992 end + 1 - start,
3993 NULL);
acce952b 3994
3995 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3996 btrfs_error_unpin_extent_range(root, start, end);
3997 cond_resched();
3998 }
3999
ed0eaa14
LB		 4000 if (loop) {
4001 if (unpin == &root->fs_info->freed_extents[0])
4002 unpin = &root->fs_info->freed_extents[1];
4003 else
4004 unpin = &root->fs_info->freed_extents[0];
4005 loop = false;
4006 goto again;
4007 }
4008
acce952b 4009 return 0;
4010}
4011
49b25e05
JM		 4012void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4013 struct btrfs_root *root)
4014{
724e2315
JB		 4015 btrfs_destroy_ordered_operations(cur_trans, root);
4016
49b25e05 4017 btrfs_destroy_delayed_refs(cur_trans, root);
49b25e05 4018
4a9d8bde 4019 cur_trans->state = TRANS_STATE_COMMIT_START;
d7096fc3 4020 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 4021
4a9d8bde 4022 cur_trans->state = TRANS_STATE_UNBLOCKED;
d7096fc3 4023 wake_up(&root->fs_info->transaction_wait);
49b25e05 4024
67cde344
MX		 4025 btrfs_destroy_delayed_inodes(root);
4026 btrfs_assert_delayed_root_empty(root);
49b25e05 4027
49b25e05
JM		 4028 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4029 EXTENT_DIRTY);
6e841e32
LB		 4030 btrfs_destroy_pinned_extent(root,
4031 root->fs_info->pinned_extents);
49b25e05 4032
4a9d8bde
MX		 4033 cur_trans->state =TRANS_STATE_COMPLETED;
4034 wake_up(&cur_trans->commit_wait);
4035
49b25e05
JM		 4036 /*
4037 memset(cur_trans, 0, sizeof(*cur_trans));
4038 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4039 */
4040}
4041
48a3b636 4042static int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 4043{
4044 struct btrfs_transaction *t;
acce952b 4045
acce952b 4046 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4047
a4abeea4 4048 spin_lock(&root->fs_info->trans_lock);
724e2315
JB		 4049 while (!list_empty(&root->fs_info->trans_list)) {
4050 t = list_first_entry(&root->fs_info->trans_list,
4051 struct btrfs_transaction, list);
4052 if (t->state >= TRANS_STATE_COMMIT_START) {
4053 atomic_inc(&t->use_count);
4054 spin_unlock(&root->fs_info->trans_lock);
4055 btrfs_wait_for_commit(root, t->transid);
4056 btrfs_put_transaction(t);
4057 spin_lock(&root->fs_info->trans_lock);
4058 continue;
4059 }
4060 if (t == root->fs_info->running_transaction) {
4061 t->state = TRANS_STATE_COMMIT_DOING;
4062 spin_unlock(&root->fs_info->trans_lock);
4063 /*
4064 * We wait for 0 num_writers since we don't hold a trans
4065 * handle open currently for this transaction.
4066 */
4067 wait_event(t->writer_wait,
4068 atomic_read(&t->num_writers) == 0);
4069 } else {
4070 spin_unlock(&root->fs_info->trans_lock);
4071 }
4072 btrfs_cleanup_one_transaction(t, root);
4a9d8bde 4073
724e2315
JB		 4074 spin_lock(&root->fs_info->trans_lock);
4075 if (t == root->fs_info->running_transaction)
4076 root->fs_info->running_transaction = NULL;
acce952b 4077 list_del_init(&t->list);
724e2315 4078 spin_unlock(&root->fs_info->trans_lock);
acce952b 4079
724e2315
JB		 4080 btrfs_put_transaction(t);
4081 trace_btrfs_transaction_commit(root);
4082 spin_lock(&root->fs_info->trans_lock);
4083 }
4084 spin_unlock(&root->fs_info->trans_lock);
4085 btrfs_destroy_all_ordered_extents(root->fs_info);
4086 btrfs_destroy_delayed_inodes(root);
4087 btrfs_assert_delayed_root_empty(root);
4088 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4089 btrfs_destroy_all_delalloc_inodes(root->fs_info);
acce952b 4090 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 4091
4092 return 0;
4093}
4094
d1310b2e 4095static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 4096 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 4097 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 4098 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 4099 /* note we're sharing with inode.c for the merge bio hook */
4100 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 4101};
Linux 6.x block layer and io_uring tree(s)