projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: introduce per-subvolume delalloc inode list
[linux-2.6-block.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
163e783e 29#include <linux/crc32c.h>
5a0e3ad6 30#include <linux/slab.h>
784b4e29 31#include <linux/migrate.h>
7a36ddec 32#include <linux/ratelimit.h>
6463fe58 33#include <linux/uuid.h>
7e75bf3f 34#include <asm/unaligned.h>
4b4e25f2 35#include "compat.h"
eb60ceac
CM		 36#include "ctree.h"
37#include "disk-io.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"
eb60ceac 51
de0022b9
JB		 52#ifdef CONFIG_X86
53#include <asm/cpufeature.h>
54#endif
55
d1310b2e 56static struct extent_io_ops btree_extent_io_ops;
8b712842 57static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 58static void free_fs_root(struct btrfs_root *root);
fcd1f065 59static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 60 int read_only);
569e0f35
JB		 61static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
62 struct btrfs_root *root);
143bede5 63static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
acce952b 64static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
65 struct btrfs_root *root);
aec8030a 66static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
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" },
159 { .id = 0, .name_stem = "tree" },
4008c04a 160};
85d4e461
CM		 161
162void __init btrfs_init_lockdep(void)
163{
164 int i, j;
165
166 /* initialize lockdep class names */
167 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
168 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
169
170 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
171 snprintf(ks->names[j], sizeof(ks->names[j]),
172 "btrfs-%s-%02d", ks->name_stem, j);
173 }
174}
175
176void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
177 int level)
178{
179 struct btrfs_lockdep_keyset *ks;
180
181 BUG_ON(level >= ARRAY_SIZE(ks->keys));
182
183 /* find the matching keyset, id 0 is the default entry */
184 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
185 if (ks->id == objectid)
186 break;
187
188 lockdep_set_class_and_name(&eb->lock,
189 &ks->keys[level], ks->names[level]);
190}
191
4008c04a
CM		 192#endif
193
d352ac68
CM		 194/*
195 * extents on the btree inode are pretty simple, there's one extent
196 * that covers the entire device
197 */
b2950863 198static struct extent_map *btree_get_extent(struct inode *inode,
306e16ce 199 struct page *page, size_t pg_offset, u64 start, u64 len,
b2950863 200 int create)
7eccb903 201{
5f39d397
CM		 202 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
203 struct extent_map *em;
204 int ret;
205
890871be 206 read_lock(&em_tree->lock);
d1310b2e 207 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 208 if (em) {
209 em->bdev =
210 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 211 read_unlock(&em_tree->lock);
5f39d397 212 goto out;
a061fc8d 213 }
890871be 214 read_unlock(&em_tree->lock);
7b13b7b1 215
172ddd60 216 em = alloc_extent_map();
5f39d397
CM		 217 if (!em) {
218 em = ERR_PTR(-ENOMEM);
219 goto out;
220 }
221 em->start = 0;
0afbaf8c 222 em->len = (u64)-1;
c8b97818 223 em->block_len = (u64)-1;
5f39d397 224 em->block_start = 0;
a061fc8d 225 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 226
890871be 227 write_lock(&em_tree->lock);
09a2a8f9 228 ret = add_extent_mapping(em_tree, em, 0);
5f39d397
CM		 229 if (ret == -EEXIST) {
230 free_extent_map(em);
7b13b7b1 231 em = lookup_extent_mapping(em_tree, start, len);
b4f359ab 232 if (!em)
0433f20d 233 em = ERR_PTR(-EIO);
5f39d397 234 } else if (ret) {
7b13b7b1 235 free_extent_map(em);
0433f20d 236 em = ERR_PTR(ret);
5f39d397 237 }
890871be 238 write_unlock(&em_tree->lock);
7b13b7b1 239
5f39d397
CM		 240out:
241 return em;
7eccb903
CM		 242}
243
b0496686 244u32 btrfs_csum_data(char *data, u32 seed, size_t len)
19c00ddc 245{
163e783e 246 return crc32c(seed, data, len);
19c00ddc
CM		 247}
248
249void btrfs_csum_final(u32 crc, char *result)
250{
7e75bf3f 251 put_unaligned_le32(~crc, result);
19c00ddc
CM		 252}
253
d352ac68
CM		 254/*
255 * compute the csum for a btree block, and either verify it or write it
256 * into the csum field of the block.
257 */
19c00ddc
CM		 258static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
259 int verify)
260{
6c41761f 261 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
607d432d 262 char *result = NULL;
19c00ddc
CM		 263 unsigned long len;
264 unsigned long cur_len;
265 unsigned long offset = BTRFS_CSUM_SIZE;
19c00ddc
CM		 266 char *kaddr;
267 unsigned long map_start;
268 unsigned long map_len;
269 int err;
270 u32 crc = ~(u32)0;
607d432d 271 unsigned long inline_result;
19c00ddc
CM		 272
273 len = buf->len - offset;
d397712b 274 while (len > 0) {
19c00ddc 275 err = map_private_extent_buffer(buf, offset, 32,
a6591715 276 &kaddr, &map_start, &map_len);
d397712b 277 if (err)
19c00ddc 278 return 1;
19c00ddc 279 cur_len = min(len, map_len - (offset - map_start));
b0496686 280 crc = btrfs_csum_data(kaddr + offset - map_start,
19c00ddc
CM		 281 crc, cur_len);
282 len -= cur_len;
283 offset += cur_len;
19c00ddc 284 }
607d432d
JB		 285 if (csum_size > sizeof(inline_result)) {
286 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
287 if (!result)
288 return 1;
289 } else {
290 result = (char *)&inline_result;
291 }
292
19c00ddc
CM		 293 btrfs_csum_final(crc, result);
294
295 if (verify) {
607d432d 296 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 297 u32 val;
298 u32 found = 0;
607d432d 299 memcpy(&found, result, csum_size);
e4204ded 300
607d432d 301 read_extent_buffer(buf, &val, 0, csum_size);
7a36ddec 302 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
193f284d
CM		 303 "failed on %llu wanted %X found %X "
304 "level %d\n",
305 root->fs_info->sb->s_id,
306 (unsigned long long)buf->start, val, found,
307 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
CM		 347 "found %llu\n",
348 (unsigned long long)eb->start,
349 (unsigned long long)parent_transid,
350 (unsigned long long)btrfs_header_generation(eb));
1259ab75 351 ret = 1;
0b32f4bb 352 clear_extent_buffer_uptodate(eb);
33958dc6 353out:
2ac55d41
JB		 354 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
355 &cached_state, GFP_NOFS);
1259ab75 356 return ret;
1259ab75
CM		 357}
358
1104a885
DS		 359/*
360 * Return 0 if the superblock checksum type matches the checksum value of that
361 * algorithm. Pass the raw disk superblock data.
362 */
363static int btrfs_check_super_csum(char *raw_disk_sb)
364{
365 struct btrfs_super_block *disk_sb =
366 (struct btrfs_super_block *)raw_disk_sb;
367 u16 csum_type = btrfs_super_csum_type(disk_sb);
368 int ret = 0;
369
370 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
371 u32 crc = ~(u32)0;
372 const int csum_size = sizeof(crc);
373 char result[csum_size];
374
375 /*
376 * The super_block structure does not span the whole
377 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
378 * is filled with zeros and is included in the checkum.
379 */
380 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
381 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
382 btrfs_csum_final(crc, result);
383
384 if (memcmp(raw_disk_sb, result, csum_size))
385 ret = 1;
667e7d94
CM		 386
387 if (ret && btrfs_super_generation(disk_sb) < 10) {
388 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
389 ret = 0;
390 }
1104a885
DS		 391 }
392
393 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
394 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
395 csum_type);
396 ret = 1;
397 }
398
399 return ret;
400}
401
d352ac68
CM		 402/*
403 * helper to read a given tree block, doing retries as required when
404 * the checksums don't match and we have alternate mirrors to try.
405 */
f188591e
CM		 406static int btree_read_extent_buffer_pages(struct btrfs_root *root,
407 struct extent_buffer *eb,
ca7a79ad 408 u64 start, u64 parent_transid)
f188591e
CM		 409{
410 struct extent_io_tree *io_tree;
ea466794 411 int failed = 0;
f188591e
CM		 412 int ret;
413 int num_copies = 0;
414 int mirror_num = 0;
ea466794 415 int failed_mirror = 0;
f188591e 416
a826d6dc 417 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM		 418 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
419 while (1) {
bb82ab88
AJ		 420 ret = read_extent_buffer_pages(io_tree, eb, start,
421 WAIT_COMPLETE,
f188591e 422 btree_get_extent, mirror_num);
256dd1bb
SB		 423 if (!ret) {
424 if (!verify_parent_transid(io_tree, eb,
b9fab919 425 parent_transid, 0))
256dd1bb
SB		 426 break;
427 else
428 ret = -EIO;
429 }
d397712b 430
a826d6dc
JB		 431 /*
432 * This buffer's crc is fine, but its contents are corrupted, so
433 * there is no reason to read the other copies, they won't be
434 * any less wrong.
435 */
436 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
ea466794
JB		 437 break;
438
5d964051 439 num_copies = btrfs_num_copies(root->fs_info,
f188591e 440 eb->start, eb->len);
4235298e 441 if (num_copies == 1)
ea466794 442 break;
4235298e 443
5cf1ab56
JB		 444 if (!failed_mirror) {
445 failed = 1;
446 failed_mirror = eb->read_mirror;
447 }
448
f188591e 449 mirror_num++;
ea466794
JB		 450 if (mirror_num == failed_mirror)
451 mirror_num++;
452
4235298e 453 if (mirror_num > num_copies)
ea466794 454 break;
f188591e 455 }
ea466794 456
c0901581 457 if (failed && !ret && failed_mirror)
ea466794
JB		 458 repair_eb_io_failure(root, eb, failed_mirror);
459
460 return ret;
f188591e 461}
19c00ddc 462
d352ac68 463/*
d397712b
CM		 464 * checksum a dirty tree block before IO. This has extra checks to make sure
465 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 466 */
d397712b 467
b2950863 468static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 469{
d1310b2e 470 struct extent_io_tree *tree;
4eee4fa4 471 u64 start = page_offset(page);
19c00ddc 472 u64 found_start;
19c00ddc 473 struct extent_buffer *eb;
f188591e 474
d1310b2e 475 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc 476
4f2de97a
JB		 477 eb = (struct extent_buffer *)page->private;
478 if (page != eb->pages[0])
479 return 0;
19c00ddc
CM		 480 found_start = btrfs_header_bytenr(eb);
481 if (found_start != start) {
55c69072 482 WARN_ON(1);
4f2de97a 483 return 0;
55c69072 484 }
55c69072 485 if (!PageUptodate(page)) {
55c69072 486 WARN_ON(1);
4f2de97a 487 return 0;
19c00ddc 488 }
19c00ddc 489 csum_tree_block(root, eb, 0);
19c00ddc
CM		 490 return 0;
491}
492
2b82032c
YZ		 493static int check_tree_block_fsid(struct btrfs_root *root,
494 struct extent_buffer *eb)
495{
496 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
497 u8 fsid[BTRFS_UUID_SIZE];
498 int ret = 1;
499
500 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
501 BTRFS_FSID_SIZE);
502 while (fs_devices) {
503 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
504 ret = 0;
505 break;
506 }
507 fs_devices = fs_devices->seed;
508 }
509 return ret;
510}
511
a826d6dc
JB		 512#define CORRUPT(reason, eb, root, slot) \
513 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
514 "root=%llu, slot=%d\n", reason, \
515 (unsigned long long)btrfs_header_bytenr(eb), \
516 (unsigned long long)root->objectid, slot)
517
518static noinline int check_leaf(struct btrfs_root *root,
519 struct extent_buffer *leaf)
520{
521 struct btrfs_key key;
522 struct btrfs_key leaf_key;
523 u32 nritems = btrfs_header_nritems(leaf);
524 int slot;
525
526 if (nritems == 0)
527 return 0;
528
529 /* Check the 0 item */
530 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
531 BTRFS_LEAF_DATA_SIZE(root)) {
532 CORRUPT("invalid item offset size pair", leaf, root, 0);
533 return -EIO;
534 }
535
536 /*
537 * Check to make sure each items keys are in the correct order and their
538 * offsets make sense. We only have to loop through nritems-1 because
539 * we check the current slot against the next slot, which verifies the
540 * next slot's offset+size makes sense and that the current's slot
541 * offset is correct.
542 */
543 for (slot = 0; slot < nritems - 1; slot++) {
544 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
545 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
546
547 /* Make sure the keys are in the right order */
548 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
549 CORRUPT("bad key order", leaf, root, slot);
550 return -EIO;
551 }
552
553 /*
554 * Make sure the offset and ends are right, remember that the
555 * item data starts at the end of the leaf and grows towards the
556 * front.
557 */
558 if (btrfs_item_offset_nr(leaf, slot) !=
559 btrfs_item_end_nr(leaf, slot + 1)) {
560 CORRUPT("slot offset bad", leaf, root, slot);
561 return -EIO;
562 }
563
564 /*
565 * Check to make sure that we don't point outside of the leaf,
566 * just incase all the items are consistent to eachother, but
567 * all point outside of the leaf.
568 */
569 if (btrfs_item_end_nr(leaf, slot) >
570 BTRFS_LEAF_DATA_SIZE(root)) {
571 CORRUPT("slot end outside of leaf", leaf, root, slot);
572 return -EIO;
573 }
574 }
575
576 return 0;
577}
578
b2950863 579static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
5cf1ab56 580 struct extent_state *state, int mirror)
ce9adaa5
CM		 581{
582 struct extent_io_tree *tree;
583 u64 found_start;
584 int found_level;
ce9adaa5
CM		 585 struct extent_buffer *eb;
586 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 587 int ret = 0;
727011e0 588 int reads_done;
ce9adaa5 589
ce9adaa5
CM		 590 if (!page->private)
591 goto out;
d397712b 592
727011e0 593 tree = &BTRFS_I(page->mapping->host)->io_tree;
4f2de97a 594 eb = (struct extent_buffer *)page->private;
d397712b 595
0b32f4bb
JB		 596 /* the pending IO might have been the only thing that kept this buffer
597 * in memory. Make sure we have a ref for all this other checks
598 */
599 extent_buffer_get(eb);
600
601 reads_done = atomic_dec_and_test(&eb->io_pages);
727011e0
CM		 602 if (!reads_done)
603 goto err;
f188591e 604
5cf1ab56 605 eb->read_mirror = mirror;
ea466794
JB		 606 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
607 ret = -EIO;
608 goto err;
609 }
610
ce9adaa5 611 found_start = btrfs_header_bytenr(eb);
727011e0 612 if (found_start != eb->start) {
7a36ddec 613 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
193f284d
CM		 614 "%llu %llu\n",
615 (unsigned long long)found_start,
616 (unsigned long long)eb->start);
f188591e 617 ret = -EIO;
ce9adaa5
CM		 618 goto err;
619 }
2b82032c 620 if (check_tree_block_fsid(root, eb)) {
7a36ddec 621 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
193f284d 622 (unsigned long long)eb->start);
1259ab75
CM		 623 ret = -EIO;
624 goto err;
625 }
ce9adaa5 626 found_level = btrfs_header_level(eb);
1c24c3ce
JB		 627 if (found_level >= BTRFS_MAX_LEVEL) {
628 btrfs_info(root->fs_info, "bad tree block level %d\n",
629 (int)btrfs_header_level(eb));
630 ret = -EIO;
631 goto err;
632 }
ce9adaa5 633
85d4e461
CM		 634 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
635 eb, found_level);
4008c04a 636
ce9adaa5 637 ret = csum_tree_block(root, eb, 1);
a826d6dc 638 if (ret) {
f188591e 639 ret = -EIO;
a826d6dc
JB		 640 goto err;
641 }
642
643 /*
644 * If this is a leaf block and it is corrupt, set the corrupt bit so
645 * that we don't try and read the other copies of this block, just
646 * return -EIO.
647 */
648 if (found_level == 0 && check_leaf(root, eb)) {
649 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
650 ret = -EIO;
651 }
ce9adaa5 652
0b32f4bb
JB		 653 if (!ret)
654 set_extent_buffer_uptodate(eb);
ce9adaa5 655err:
79fb65a1
JB		 656 if (reads_done &&
657 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 658 btree_readahead_hook(root, eb, eb->start, ret);
4bb31e92 659
53b381b3
DW		 660 if (ret) {
661 /*
662 * our io error hook is going to dec the io pages
663 * again, we have to make sure it has something
664 * to decrement
665 */
666 atomic_inc(&eb->io_pages);
0b32f4bb 667 clear_extent_buffer_uptodate(eb);
53b381b3 668 }
0b32f4bb 669 free_extent_buffer(eb);
ce9adaa5 670out:
f188591e 671 return ret;
ce9adaa5
CM		 672}
673
ea466794 674static int btree_io_failed_hook(struct page *page, int failed_mirror)
4bb31e92 675{
4bb31e92
AJ		 676 struct extent_buffer *eb;
677 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
678
4f2de97a 679 eb = (struct extent_buffer *)page->private;
ea466794 680 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
5cf1ab56 681 eb->read_mirror = failed_mirror;
53b381b3 682 atomic_dec(&eb->io_pages);
ea466794 683 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 684 btree_readahead_hook(root, eb, eb->start, -EIO);
4bb31e92
AJ		 685 return -EIO; /* we fixed nothing */
686}
687
ce9adaa5 688static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 689{
690 struct end_io_wq *end_io_wq = bio->bi_private;
691 struct btrfs_fs_info *fs_info;
ce9adaa5 692
ce9adaa5 693 fs_info = end_io_wq->info;
ce9adaa5 694 end_io_wq->error = err;
8b712842
CM		 695 end_io_wq->work.func = end_workqueue_fn;
696 end_io_wq->work.flags = 0;
d20f7043 697
7b6d91da 698 if (bio->bi_rw & REQ_WRITE) {
53b381b3 699 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
cad321ad
CM		 700 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
701 &end_io_wq->work);
53b381b3 702 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
0cb59c99
JB		 703 btrfs_queue_worker(&fs_info->endio_freespace_worker,
704 &end_io_wq->work);
53b381b3
DW		 705 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
706 btrfs_queue_worker(&fs_info->endio_raid56_workers,
707 &end_io_wq->work);
cad321ad
CM		 708 else
709 btrfs_queue_worker(&fs_info->endio_write_workers,
710 &end_io_wq->work);
d20f7043 711 } else {
53b381b3
DW		 712 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
713 btrfs_queue_worker(&fs_info->endio_raid56_workers,
714 &end_io_wq->work);
715 else if (end_io_wq->metadata)
d20f7043
CM		 716 btrfs_queue_worker(&fs_info->endio_meta_workers,
717 &end_io_wq->work);
718 else
719 btrfs_queue_worker(&fs_info->endio_workers,
720 &end_io_wq->work);
721 }
ce9adaa5
CM		 722}
723
0cb59c99
JB		 724/*
725 * For the metadata arg you want
726 *
727 * 0 - if data
728 * 1 - if normal metadta
729 * 2 - if writing to the free space cache area
53b381b3 730 * 3 - raid parity work
0cb59c99 731 */
22c59948
CM		 732int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
733 int metadata)
0b86a832 734{
ce9adaa5 735 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 736 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
737 if (!end_io_wq)
738 return -ENOMEM;
739
740 end_io_wq->private = bio->bi_private;
741 end_io_wq->end_io = bio->bi_end_io;
22c59948 742 end_io_wq->info = info;
ce9adaa5
CM		 743 end_io_wq->error = 0;
744 end_io_wq->bio = bio;
22c59948 745 end_io_wq->metadata = metadata;
ce9adaa5
CM		 746
747 bio->bi_private = end_io_wq;
748 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 749 return 0;
750}
751
b64a2851 752unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 753{
4854ddd0
CM		 754 unsigned long limit = min_t(unsigned long,
755 info->workers.max_workers,
756 info->fs_devices->open_devices);
757 return 256 * limit;
758}
0986fe9e 759
4a69a410
CM		 760static void run_one_async_start(struct btrfs_work *work)
761{
4a69a410 762 struct async_submit_bio *async;
79787eaa 763 int ret;
4a69a410
CM		 764
765 async = container_of(work, struct async_submit_bio, work);
79787eaa
JM		 766 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
767 async->mirror_num, async->bio_flags,
768 async->bio_offset);
769 if (ret)
770 async->error = ret;
4a69a410
CM		 771}
772
773static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 774{
775 struct btrfs_fs_info *fs_info;
776 struct async_submit_bio *async;
4854ddd0 777 int limit;
8b712842
CM		 778
779 async = container_of(work, struct async_submit_bio, work);
780 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 781
b64a2851 782 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 783 limit = limit * 2 / 3;
784
66657b31 785 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
b64a2851 786 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 787 wake_up(&fs_info->async_submit_wait);
788
79787eaa
JM		 789 /* If an error occured we just want to clean up the bio and move on */
790 if (async->error) {
791 bio_endio(async->bio, async->error);
792 return;
793 }
794
4a69a410 795 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 796 async->mirror_num, async->bio_flags,
797 async->bio_offset);
4a69a410
CM		 798}
799
800static void run_one_async_free(struct btrfs_work *work)
801{
802 struct async_submit_bio *async;
803
804 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 805 kfree(async);
806}
807
44b8bd7e
CM		 808int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
809 int rw, struct bio *bio, int mirror_num,
c8b97818 810 unsigned long bio_flags,
eaf25d93 811 u64 bio_offset,
4a69a410
CM		 812 extent_submit_bio_hook_t *submit_bio_start,
813 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 814{
815 struct async_submit_bio *async;
816
817 async = kmalloc(sizeof(*async), GFP_NOFS);
818 if (!async)
819 return -ENOMEM;
820
821 async->inode = inode;
822 async->rw = rw;
823 async->bio = bio;
824 async->mirror_num = mirror_num;
4a69a410
CM		 825 async->submit_bio_start = submit_bio_start;
826 async->submit_bio_done = submit_bio_done;
827
828 async->work.func = run_one_async_start;
829 async->work.ordered_func = run_one_async_done;
830 async->work.ordered_free = run_one_async_free;
831
8b712842 832 async->work.flags = 0;
c8b97818 833 async->bio_flags = bio_flags;
eaf25d93 834 async->bio_offset = bio_offset;
8c8bee1d 835
79787eaa
JM		 836 async->error = 0;
837
cb03c743 838 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 839
7b6d91da 840 if (rw & REQ_SYNC)
d313d7a3
CM		 841 btrfs_set_work_high_prio(&async->work);
842
8b712842 843 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 844
d397712b 845 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 846 atomic_read(&fs_info->nr_async_submits)) {
847 wait_event(fs_info->async_submit_wait,
848 (atomic_read(&fs_info->nr_async_submits) == 0));
849 }
850
44b8bd7e
CM		 851 return 0;
852}
853
ce3ed71a
CM		 854static int btree_csum_one_bio(struct bio *bio)
855{
856 struct bio_vec *bvec = bio->bi_io_vec;
857 int bio_index = 0;
858 struct btrfs_root *root;
79787eaa 859 int ret = 0;
ce3ed71a
CM		 860
861 WARN_ON(bio->bi_vcnt <= 0);
d397712b 862 while (bio_index < bio->bi_vcnt) {
ce3ed71a 863 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 864 ret = csum_dirty_buffer(root, bvec->bv_page);
865 if (ret)
866 break;
ce3ed71a
CM		 867 bio_index++;
868 bvec++;
869 }
79787eaa 870 return ret;
ce3ed71a
CM		 871}
872
4a69a410
CM		 873static int __btree_submit_bio_start(struct inode *inode, int rw,
874 struct bio *bio, int mirror_num,
eaf25d93
CM		 875 unsigned long bio_flags,
876 u64 bio_offset)
22c59948 877{
8b712842
CM		 878 /*
879 * when we're called for a write, we're already in the async
5443be45 880 * submission context. Just jump into btrfs_map_bio
8b712842 881 */
79787eaa 882 return btree_csum_one_bio(bio);
4a69a410 883}
22c59948 884
4a69a410 885static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 886 int mirror_num, unsigned long bio_flags,
887 u64 bio_offset)
4a69a410 888{
61891923
SB		 889 int ret;
890
8b712842 891 /*
4a69a410
CM		 892 * when we're called for a write, we're already in the async
893 * submission context. Just jump into btrfs_map_bio
8b712842 894 */
61891923
SB		 895 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
896 if (ret)
897 bio_endio(bio, ret);
898 return ret;
0b86a832
CM		 899}
900
de0022b9
JB		 901static int check_async_write(struct inode *inode, unsigned long bio_flags)
902{
903 if (bio_flags & EXTENT_BIO_TREE_LOG)
904 return 0;
905#ifdef CONFIG_X86
906 if (cpu_has_xmm4_2)
907 return 0;
908#endif
909 return 1;
910}
911
44b8bd7e 912static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 913 int mirror_num, unsigned long bio_flags,
914 u64 bio_offset)
44b8bd7e 915{
de0022b9 916 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 917 int ret;
918
7b6d91da 919 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 920 /*
921 * called for a read, do the setup so that checksum validation
922 * can happen in the async kernel threads
923 */
f3f266ab
CM		 924 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
925 bio, 1);
1d4284bd 926 if (ret)
61891923
SB		 927 goto out_w_error;
928 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
929 mirror_num, 0);
de0022b9
JB		 930 } else if (!async) {
931 ret = btree_csum_one_bio(bio);
932 if (ret)
61891923
SB		 933 goto out_w_error;
934 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
935 mirror_num, 0);
936 } else {
937 /*
938 * kthread helpers are used to submit writes so that
939 * checksumming can happen in parallel across all CPUs
940 */
941 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
942 inode, rw, bio, mirror_num, 0,
943 bio_offset,
944 __btree_submit_bio_start,
945 __btree_submit_bio_done);
44b8bd7e 946 }
d313d7a3 947
61891923
SB		 948 if (ret) {
949out_w_error:
950 bio_endio(bio, ret);
951 }
952 return ret;
44b8bd7e
CM		 953}
954
3dd1462e 955#ifdef CONFIG_MIGRATION
784b4e29 956static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 957 struct page *newpage, struct page *page,
958 enum migrate_mode mode)
784b4e29
CM		 959{
960 /*
961 * we can't safely write a btree page from here,
962 * we haven't done the locking hook
963 */
964 if (PageDirty(page))
965 return -EAGAIN;
966 /*
967 * Buffers may be managed in a filesystem specific way.
968 * We must have no buffers or drop them.
969 */
970 if (page_has_private(page) &&
971 !try_to_release_page(page, GFP_KERNEL))
972 return -EAGAIN;
a6bc32b8 973 return migrate_page(mapping, newpage, page, mode);
784b4e29 974}
3dd1462e 975#endif
784b4e29 976
0da5468f
CM		 977
978static int btree_writepages(struct address_space *mapping,
979 struct writeback_control *wbc)
980{
d1310b2e 981 struct extent_io_tree *tree;
e2d84521
MX		 982 struct btrfs_fs_info *fs_info;
983 int ret;
984
d1310b2e 985 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 986 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 987
988 if (wbc->for_kupdate)
989 return 0;
990
e2d84521 991 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 992 /* this is a bit racy, but that's ok */
e2d84521
MX		 993 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
994 BTRFS_DIRTY_METADATA_THRESH);
995 if (ret < 0)
793955bc 996 return 0;
793955bc 997 }
0b32f4bb 998 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 999}
1000
b2950863 1001static int btree_readpage(struct file *file, struct page *page)
5f39d397 1002{
d1310b2e
CM		 1003 struct extent_io_tree *tree;
1004 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 1005 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 1006}
22b0ebda 1007
70dec807 1008static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 1009{
98509cfc 1010 if (PageWriteback(page) || PageDirty(page))
d397712b 1011 return 0;
0c4e538b 1012
f7a52a40 1013 return try_release_extent_buffer(page);
d98237b3
CM		 1014}
1015
5f39d397 1016static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 1017{
d1310b2e
CM		 1018 struct extent_io_tree *tree;
1019 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1020 extent_invalidatepage(tree, page, offset);
1021 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1022 if (PagePrivate(page)) {
d397712b
CM		 1023 printk(KERN_WARNING "btrfs warning page private not zero "
1024 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1025 ClearPagePrivate(page);
1026 set_page_private(page, 0);
1027 page_cache_release(page);
1028 }
d98237b3
CM		 1029}
1030
0b32f4bb
JB		 1031static int btree_set_page_dirty(struct page *page)
1032{
bb146eb2 1033#ifdef DEBUG
0b32f4bb
JB		 1034 struct extent_buffer *eb;
1035
1036 BUG_ON(!PagePrivate(page));
1037 eb = (struct extent_buffer *)page->private;
1038 BUG_ON(!eb);
1039 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1040 BUG_ON(!atomic_read(&eb->refs));
1041 btrfs_assert_tree_locked(eb);
bb146eb2 1042#endif
0b32f4bb
JB		 1043 return __set_page_dirty_nobuffers(page);
1044}
1045
7f09410b 1046static const struct address_space_operations btree_aops = {
d98237b3 1047 .readpage = btree_readpage,
0da5468f 1048 .writepages = btree_writepages,
5f39d397
CM		 1049 .releasepage = btree_releasepage,
1050 .invalidatepage = btree_invalidatepage,
5a92bc88 1051#ifdef CONFIG_MIGRATION
784b4e29 1052 .migratepage = btree_migratepage,
5a92bc88 1053#endif
0b32f4bb 1054 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1055};
1056
ca7a79ad
CM		 1057int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1058 u64 parent_transid)
090d1875 1059{
5f39d397
CM		 1060 struct extent_buffer *buf = NULL;
1061 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1062 int ret = 0;
090d1875 1063
db94535d 1064 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1065 if (!buf)
090d1875 1066 return 0;
d1310b2e 1067 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1068 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1069 free_extent_buffer(buf);
de428b63 1070 return ret;
090d1875
CM		 1071}
1072
ab0fff03
AJ		 1073int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1074 int mirror_num, struct extent_buffer **eb)
1075{
1076 struct extent_buffer *buf = NULL;
1077 struct inode *btree_inode = root->fs_info->btree_inode;
1078 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1079 int ret;
1080
1081 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1082 if (!buf)
1083 return 0;
1084
1085 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1086
1087 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1088 btree_get_extent, mirror_num);
1089 if (ret) {
1090 free_extent_buffer(buf);
1091 return ret;
1092 }
1093
1094 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1095 free_extent_buffer(buf);
1096 return -EIO;
0b32f4bb 1097 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1098 *eb = buf;
1099 } else {
1100 free_extent_buffer(buf);
1101 }
1102 return 0;
1103}
1104
0999df54
CM		 1105struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1106 u64 bytenr, u32 blocksize)
1107{
1108 struct inode *btree_inode = root->fs_info->btree_inode;
1109 struct extent_buffer *eb;
1110 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
f09d1f60 1111 bytenr, blocksize);
0999df54
CM		 1112 return eb;
1113}
1114
1115struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1116 u64 bytenr, u32 blocksize)
1117{
1118 struct inode *btree_inode = root->fs_info->btree_inode;
1119 struct extent_buffer *eb;
1120
1121 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
727011e0 1122 bytenr, blocksize);
0999df54
CM		 1123 return eb;
1124}
1125
1126
e02119d5
CM		 1127int btrfs_write_tree_block(struct extent_buffer *buf)
1128{
727011e0 1129 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
8aa38c31 1130 buf->start + buf->len - 1);
e02119d5
CM		 1131}
1132
1133int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1134{
727011e0 1135 return filemap_fdatawait_range(buf->pages[0]->mapping,
8aa38c31 1136 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1137}
1138
0999df54 1139struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1140 u32 blocksize, u64 parent_transid)
0999df54
CM		 1141{
1142 struct extent_buffer *buf = NULL;
0999df54
CM		 1143 int ret;
1144
0999df54
CM		 1145 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1146 if (!buf)
1147 return NULL;
0999df54 1148
ca7a79ad 1149 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
5f39d397 1150 return buf;
ce9adaa5 1151
eb60ceac
CM		 1152}
1153
d5c13f92
JM		 1154void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1155 struct extent_buffer *buf)
ed2ff2cb 1156{
e2d84521
MX		 1157 struct btrfs_fs_info *fs_info = root->fs_info;
1158
55c69072 1159 if (btrfs_header_generation(buf) ==
e2d84521 1160 fs_info->running_transaction->transid) {
b9447ef8 1161 btrfs_assert_tree_locked(buf);
b4ce94de 1162
b9473439 1163 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1164 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1165 -buf->len,
1166 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1167 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1168 btrfs_set_lock_blocking(buf);
1169 clear_extent_buffer_dirty(buf);
1170 }
925baedd 1171 }
5f39d397
CM		 1172}
1173
143bede5
JM		 1174static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1175 u32 stripesize, struct btrfs_root *root,
1176 struct btrfs_fs_info *fs_info,
1177 u64 objectid)
d97e63b6 1178{
cfaa7295 1179 root->node = NULL;
a28ec197 1180 root->commit_root = NULL;
db94535d
CM		 1181 root->sectorsize = sectorsize;
1182 root->nodesize = nodesize;
1183 root->leafsize = leafsize;
87ee04eb 1184 root->stripesize = stripesize;
123abc88 1185 root->ref_cows = 0;
0b86a832 1186 root->track_dirty = 0;
c71bf099 1187 root->in_radix = 0;
d68fc57b
YZ		 1188 root->orphan_item_inserted = 0;
1189 root->orphan_cleanup_state = 0;
0b86a832 1190
0f7d52f4
CM		 1191 root->objectid = objectid;
1192 root->last_trans = 0;
13a8a7c8 1193 root->highest_objectid = 0;
eb73c1b7 1194 root->nr_delalloc_inodes = 0;
58176a96 1195 root->name = NULL;
6bef4d31 1196 root->inode_tree = RB_ROOT;
16cdcec7 1197 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1198 root->block_rsv = NULL;
d68fc57b 1199 root->orphan_block_rsv = NULL;
0b86a832
CM		 1200
1201 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1202 INIT_LIST_HEAD(&root->root_list);
eb73c1b7
MX		 1203 INIT_LIST_HEAD(&root->delalloc_inodes);
1204 INIT_LIST_HEAD(&root->delalloc_root);
2ab28f32
JB		 1205 INIT_LIST_HEAD(&root->logged_list[0]);
1206 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1207 spin_lock_init(&root->orphan_lock);
5d4f98a2 1208 spin_lock_init(&root->inode_lock);
eb73c1b7 1209 spin_lock_init(&root->delalloc_lock);
f0486c68 1210 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1211 spin_lock_init(&root->log_extents_lock[0]);
1212 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1213 mutex_init(&root->objectid_mutex);
e02119d5 1214 mutex_init(&root->log_mutex);
7237f183
YZ		 1215 init_waitqueue_head(&root->log_writer_wait);
1216 init_waitqueue_head(&root->log_commit_wait[0]);
1217 init_waitqueue_head(&root->log_commit_wait[1]);
1218 atomic_set(&root->log_commit[0], 0);
1219 atomic_set(&root->log_commit[1], 0);
1220 atomic_set(&root->log_writers, 0);
2ecb7923 1221 atomic_set(&root->log_batch, 0);
8a35d95f 1222 atomic_set(&root->orphan_inodes, 0);
b0feb9d9 1223 atomic_set(&root->refs, 1);
7237f183 1224 root->log_transid = 0;
257c62e1 1225 root->last_log_commit = 0;
d0c803c4 1226 extent_io_tree_init(&root->dirty_log_pages,
f993c883 1227 fs_info->btree_inode->i_mapping);
017e5369 1228
3768f368
CM		 1229 memset(&root->root_key, 0, sizeof(root->root_key));
1230 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1231 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1232 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 1233 root->defrag_trans_start = fs_info->generation;
58176a96 1234 init_completion(&root->kobj_unregister);
6702ed49 1235 root->defrag_running = 0;
4d775673 1236 root->root_key.objectid = objectid;
0ee5dc67 1237 root->anon_dev = 0;
8ea05e3a 1238
5f3ab90a 1239 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1240}
1241
f84a8bd6 1242static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1243{
1244 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1245 if (root)
1246 root->fs_info = fs_info;
1247 return root;
1248}
1249
20897f5c
AJ		 1250struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1251 struct btrfs_fs_info *fs_info,
1252 u64 objectid)
1253{
1254 struct extent_buffer *leaf;
1255 struct btrfs_root *tree_root = fs_info->tree_root;
1256 struct btrfs_root *root;
1257 struct btrfs_key key;
1258 int ret = 0;
1259 u64 bytenr;
6463fe58 1260 uuid_le uuid;
20897f5c
AJ		 1261
1262 root = btrfs_alloc_root(fs_info);
1263 if (!root)
1264 return ERR_PTR(-ENOMEM);
1265
1266 __setup_root(tree_root->nodesize, tree_root->leafsize,
1267 tree_root->sectorsize, tree_root->stripesize,
1268 root, fs_info, objectid);
1269 root->root_key.objectid = objectid;
1270 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1271 root->root_key.offset = 0;
1272
1273 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1274 0, objectid, NULL, 0, 0, 0);
1275 if (IS_ERR(leaf)) {
1276 ret = PTR_ERR(leaf);
1dd05682 1277 leaf = NULL;
20897f5c
AJ		 1278 goto fail;
1279 }
1280
1281 bytenr = leaf->start;
1282 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1283 btrfs_set_header_bytenr(leaf, leaf->start);
1284 btrfs_set_header_generation(leaf, trans->transid);
1285 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1286 btrfs_set_header_owner(leaf, objectid);
1287 root->node = leaf;
1288
1289 write_extent_buffer(leaf, fs_info->fsid,
1290 (unsigned long)btrfs_header_fsid(leaf),
1291 BTRFS_FSID_SIZE);
1292 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1293 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1294 BTRFS_UUID_SIZE);
1295 btrfs_mark_buffer_dirty(leaf);
1296
1297 root->commit_root = btrfs_root_node(root);
1298 root->track_dirty = 1;
1299
1300
1301 root->root_item.flags = 0;
1302 root->root_item.byte_limit = 0;
1303 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1304 btrfs_set_root_generation(&root->root_item, trans->transid);
1305 btrfs_set_root_level(&root->root_item, 0);
1306 btrfs_set_root_refs(&root->root_item, 1);
1307 btrfs_set_root_used(&root->root_item, leaf->len);
1308 btrfs_set_root_last_snapshot(&root->root_item, 0);
1309 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1310 uuid_le_gen(&uuid);
1311 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1312 root->root_item.drop_level = 0;
1313
1314 key.objectid = objectid;
1315 key.type = BTRFS_ROOT_ITEM_KEY;
1316 key.offset = 0;
1317 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1318 if (ret)
1319 goto fail;
1320
1321 btrfs_tree_unlock(leaf);
1322
1dd05682
TI		 1323 return root;
1324
20897f5c 1325fail:
1dd05682
TI		 1326 if (leaf) {
1327 btrfs_tree_unlock(leaf);
1328 free_extent_buffer(leaf);
1329 }
1330 kfree(root);
20897f5c 1331
1dd05682 1332 return ERR_PTR(ret);
20897f5c
AJ		 1333}
1334
7237f183
YZ		 1335static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1336 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1337{
1338 struct btrfs_root *root;
1339 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1340 struct extent_buffer *leaf;
e02119d5 1341
6f07e42e 1342 root = btrfs_alloc_root(fs_info);
e02119d5 1343 if (!root)
7237f183 1344 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1345
1346 __setup_root(tree_root->nodesize, tree_root->leafsize,
1347 tree_root->sectorsize, tree_root->stripesize,
1348 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1349
1350 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1351 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1352 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1353 /*
1354 * log trees do not get reference counted because they go away
1355 * before a real commit is actually done. They do store pointers
1356 * to file data extents, and those reference counts still get
1357 * updated (along with back refs to the log tree).
1358 */
e02119d5
CM		 1359 root->ref_cows = 0;
1360
5d4f98a2 1361 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1362 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1363 0, 0, 0);
7237f183
YZ		 1364 if (IS_ERR(leaf)) {
1365 kfree(root);
1366 return ERR_CAST(leaf);
1367 }
e02119d5 1368
5d4f98a2
YZ		 1369 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1370 btrfs_set_header_bytenr(leaf, leaf->start);
1371 btrfs_set_header_generation(leaf, trans->transid);
1372 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1373 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1374 root->node = leaf;
e02119d5
CM		 1375
1376 write_extent_buffer(root->node, root->fs_info->fsid,
1377 (unsigned long)btrfs_header_fsid(root->node),
1378 BTRFS_FSID_SIZE);
1379 btrfs_mark_buffer_dirty(root->node);
1380 btrfs_tree_unlock(root->node);
7237f183
YZ		 1381 return root;
1382}
1383
1384int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1385 struct btrfs_fs_info *fs_info)
1386{
1387 struct btrfs_root *log_root;
1388
1389 log_root = alloc_log_tree(trans, fs_info);
1390 if (IS_ERR(log_root))
1391 return PTR_ERR(log_root);
1392 WARN_ON(fs_info->log_root_tree);
1393 fs_info->log_root_tree = log_root;
1394 return 0;
1395}
1396
1397int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1398 struct btrfs_root *root)
1399{
1400 struct btrfs_root *log_root;
1401 struct btrfs_inode_item *inode_item;
1402
1403 log_root = alloc_log_tree(trans, root->fs_info);
1404 if (IS_ERR(log_root))
1405 return PTR_ERR(log_root);
1406
1407 log_root->last_trans = trans->transid;
1408 log_root->root_key.offset = root->root_key.objectid;
1409
1410 inode_item = &log_root->root_item.inode;
1411 inode_item->generation = cpu_to_le64(1);
1412 inode_item->size = cpu_to_le64(3);
1413 inode_item->nlink = cpu_to_le32(1);
1414 inode_item->nbytes = cpu_to_le64(root->leafsize);
1415 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1416
5d4f98a2 1417 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1418
1419 WARN_ON(root->log_root);
1420 root->log_root = log_root;
1421 root->log_transid = 0;
257c62e1 1422 root->last_log_commit = 0;
e02119d5
CM		 1423 return 0;
1424}
1425
cb517eab
MX		 1426struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1427 struct btrfs_key *key)
e02119d5
CM		 1428{
1429 struct btrfs_root *root;
1430 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1431 struct btrfs_path *path;
84234f3a 1432 u64 generation;
db94535d 1433 u32 blocksize;
cb517eab 1434 int ret;
0f7d52f4 1435
cb517eab
MX		 1436 path = btrfs_alloc_path();
1437 if (!path)
0f7d52f4 1438 return ERR_PTR(-ENOMEM);
cb517eab
MX		 1439
1440 root = btrfs_alloc_root(fs_info);
1441 if (!root) {
1442 ret = -ENOMEM;
1443 goto alloc_fail;
0f7d52f4
CM		 1444 }
1445
db94535d 1446 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb 1447 tree_root->sectorsize, tree_root->stripesize,
cb517eab 1448 root, fs_info, key->objectid);
0f7d52f4 1449
cb517eab
MX		 1450 ret = btrfs_find_root(tree_root, key, path,
1451 &root->root_item, &root->root_key);
0f7d52f4 1452 if (ret) {
13a8a7c8
YZ		 1453 if (ret > 0)
1454 ret = -ENOENT;
cb517eab 1455 goto find_fail;
0f7d52f4 1456 }
13a8a7c8 1457
84234f3a 1458 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1459 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1460 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1461 blocksize, generation);
cb517eab
MX		 1462 if (!root->node) {
1463 ret = -ENOMEM;
1464 goto find_fail;
1465 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1466 ret = -EIO;
1467 goto read_fail;
416bc658 1468 }
5d4f98a2 1469 root->commit_root = btrfs_root_node(root);
13a8a7c8 1470out:
cb517eab
MX		 1471 btrfs_free_path(path);
1472 return root;
1473
1474read_fail:
1475 free_extent_buffer(root->node);
1476find_fail:
1477 kfree(root);
1478alloc_fail:
1479 root = ERR_PTR(ret);
1480 goto out;
1481}
1482
1483struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1484 struct btrfs_key *location)
1485{
1486 struct btrfs_root *root;
1487
1488 root = btrfs_read_tree_root(tree_root, location);
1489 if (IS_ERR(root))
1490 return root;
1491
1492 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1493 root->ref_cows = 1;
08fe4db1
LZ		 1494 btrfs_check_and_init_root_item(&root->root_item);
1495 }
13a8a7c8 1496
5eda7b5e
CM		 1497 return root;
1498}
1499
cb517eab
MX		 1500int btrfs_init_fs_root(struct btrfs_root *root)
1501{
1502 int ret;
1503
1504 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1505 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1506 GFP_NOFS);
1507 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1508 ret = -ENOMEM;
1509 goto fail;
1510 }
1511
1512 btrfs_init_free_ino_ctl(root);
1513 mutex_init(&root->fs_commit_mutex);
1514 spin_lock_init(&root->cache_lock);
1515 init_waitqueue_head(&root->cache_wait);
1516
1517 ret = get_anon_bdev(&root->anon_dev);
1518 if (ret)
1519 goto fail;
1520 return 0;
1521fail:
1522 kfree(root->free_ino_ctl);
1523 kfree(root->free_ino_pinned);
1524 return ret;
1525}
1526
1527struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1528 u64 root_id)
1529{
1530 struct btrfs_root *root;
1531
1532 spin_lock(&fs_info->fs_roots_radix_lock);
1533 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1534 (unsigned long)root_id);
1535 spin_unlock(&fs_info->fs_roots_radix_lock);
1536 return root;
1537}
1538
1539int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1540 struct btrfs_root *root)
1541{
1542 int ret;
1543
1544 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1545 if (ret)
1546 return ret;
1547
1548 spin_lock(&fs_info->fs_roots_radix_lock);
1549 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1550 (unsigned long)root->root_key.objectid,
1551 root);
1552 if (ret == 0)
1553 root->in_radix = 1;
1554 spin_unlock(&fs_info->fs_roots_radix_lock);
1555 radix_tree_preload_end();
1556
1557 return ret;
1558}
1559
edbd8d4e
CM		 1560struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1561 struct btrfs_key *location)
5eda7b5e
CM		 1562{
1563 struct btrfs_root *root;
1564 int ret;
1565
edbd8d4e
CM		 1566 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1567 return fs_info->tree_root;
1568 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1569 return fs_info->extent_root;
8f18cf13
CM		 1570 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1571 return fs_info->chunk_root;
1572 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1573 return fs_info->dev_root;
0403e47e
YZ		 1574 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1575 return fs_info->csum_root;
bcef60f2
AJ		 1576 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1577 return fs_info->quota_root ? fs_info->quota_root :
1578 ERR_PTR(-ENOENT);
4df27c4d 1579again:
cb517eab 1580 root = btrfs_lookup_fs_root(fs_info, location->objectid);
5eda7b5e
CM		 1581 if (root)
1582 return root;
1583
cb517eab 1584 root = btrfs_read_fs_root(fs_info->tree_root, location);
5eda7b5e
CM		 1585 if (IS_ERR(root))
1586 return root;
3394e160 1587
cb517eab
MX		 1588 if (btrfs_root_refs(&root->root_item) == 0) {
1589 ret = -ENOENT;
581bb050 1590 goto fail;
35a30d7c 1591 }
581bb050 1592
cb517eab 1593 ret = btrfs_init_fs_root(root);
ac08aedf
CM		 1594 if (ret)
1595 goto fail;
3394e160 1596
d68fc57b
YZ		 1597 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1598 if (ret < 0)
1599 goto fail;
1600 if (ret == 0)
1601 root->orphan_item_inserted = 1;
1602
cb517eab 1603 ret = btrfs_insert_fs_root(fs_info, root);
0f7d52f4 1604 if (ret) {
4df27c4d
YZ		 1605 if (ret == -EEXIST) {
1606 free_fs_root(root);
1607 goto again;
1608 }
1609 goto fail;
0f7d52f4 1610 }
edbd8d4e 1611 return root;
4df27c4d
YZ		 1612fail:
1613 free_fs_root(root);
1614 return ERR_PTR(ret);
edbd8d4e
CM		 1615}
1616
04160088
CM		 1617static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1618{
1619 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1620 int ret = 0;
04160088
CM		 1621 struct btrfs_device *device;
1622 struct backing_dev_info *bdi;
b7967db7 1623
1f78160c
XG		 1624 rcu_read_lock();
1625 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1626 if (!device->bdev)
1627 continue;
04160088
CM		 1628 bdi = blk_get_backing_dev_info(device->bdev);
1629 if (bdi && bdi_congested(bdi, bdi_bits)) {
1630 ret = 1;
1631 break;
1632 }
1633 }
1f78160c 1634 rcu_read_unlock();
04160088
CM		 1635 return ret;
1636}
1637
ad081f14
JA		 1638/*
1639 * If this fails, caller must call bdi_destroy() to get rid of the
1640 * bdi again.
1641 */
04160088
CM		 1642static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1643{
ad081f14
JA		 1644 int err;
1645
1646 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1647 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1648 if (err)
1649 return err;
1650
4575c9cc 1651 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1652 bdi->congested_fn = btrfs_congested_fn;
1653 bdi->congested_data = info;
1654 return 0;
1655}
1656
8b712842
CM		 1657/*
1658 * called by the kthread helper functions to finally call the bio end_io
1659 * functions. This is where read checksum verification actually happens
1660 */
1661static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1662{
ce9adaa5 1663 struct bio *bio;
8b712842
CM		 1664 struct end_io_wq *end_io_wq;
1665 struct btrfs_fs_info *fs_info;
ce9adaa5 1666 int error;
ce9adaa5 1667
8b712842
CM		 1668 end_io_wq = container_of(work, struct end_io_wq, work);
1669 bio = end_io_wq->bio;
1670 fs_info = end_io_wq->info;
ce9adaa5 1671
8b712842
CM		 1672 error = end_io_wq->error;
1673 bio->bi_private = end_io_wq->private;
1674 bio->bi_end_io = end_io_wq->end_io;
1675 kfree(end_io_wq);
8b712842 1676 bio_endio(bio, error);
44b8bd7e
CM		 1677}
1678
a74a4b97
CM		 1679static int cleaner_kthread(void *arg)
1680{
1681 struct btrfs_root *root = arg;
d0278245 1682 int again;
a74a4b97
CM		 1683
1684 do {
d0278245 1685 again = 0;
a74a4b97 1686
d0278245 1687 /* Make the cleaner go to sleep early. */
babbf170 1688 if (btrfs_need_cleaner_sleep(root))
d0278245
MX		 1689 goto sleep;
1690
1691 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1692 goto sleep;
1693
dc7f370c
MX		 1694 /*
1695 * Avoid the problem that we change the status of the fs
1696 * during the above check and trylock.
1697 */
babbf170 1698 if (btrfs_need_cleaner_sleep(root)) {
dc7f370c
MX		 1699 mutex_unlock(&root->fs_info->cleaner_mutex);
1700 goto sleep;
1701 }
1702
d0278245
MX		 1703 btrfs_run_delayed_iputs(root);
1704 again = btrfs_clean_one_deleted_snapshot(root);
1705 mutex_unlock(&root->fs_info->cleaner_mutex);
1706
1707 /*
05323cd1
MX		 1708 * The defragger has dealt with the R/O remount and umount,
1709 * needn't do anything special here.
d0278245
MX		 1710 */
1711 btrfs_run_defrag_inodes(root->fs_info);
1712sleep:
9d1a2a3a 1713 if (!try_to_freeze() && !again) {
a74a4b97 1714 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1715 if (!kthread_should_stop())
1716 schedule();
a74a4b97
CM		 1717 __set_current_state(TASK_RUNNING);
1718 }
1719 } while (!kthread_should_stop());
1720 return 0;
1721}
1722
1723static int transaction_kthread(void *arg)
1724{
1725 struct btrfs_root *root = arg;
1726 struct btrfs_trans_handle *trans;
1727 struct btrfs_transaction *cur;
8929ecfa 1728 u64 transid;
a74a4b97
CM		 1729 unsigned long now;
1730 unsigned long delay;
914b2007 1731 bool cannot_commit;
a74a4b97
CM		 1732
1733 do {
914b2007 1734 cannot_commit = false;
a74a4b97 1735 delay = HZ * 30;
a74a4b97
CM		 1736 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1737
a4abeea4 1738 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1739 cur = root->fs_info->running_transaction;
1740 if (!cur) {
a4abeea4 1741 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1742 goto sleep;
1743 }
31153d81 1744
a74a4b97 1745 now = get_seconds();
8929ecfa
YZ		 1746 if (!cur->blocked &&
1747 (now < cur->start_time || now - cur->start_time < 30)) {
a4abeea4 1748 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1749 delay = HZ * 5;
1750 goto sleep;
1751 }
8929ecfa 1752 transid = cur->transid;
a4abeea4 1753 spin_unlock(&root->fs_info->trans_lock);
56bec294 1754
79787eaa 1755 /* If the file system is aborted, this will always fail. */
354aa0fb 1756 trans = btrfs_attach_transaction(root);
914b2007 1757 if (IS_ERR(trans)) {
354aa0fb
MX		 1758 if (PTR_ERR(trans) != -ENOENT)
1759 cannot_commit = true;
79787eaa 1760 goto sleep;
914b2007 1761 }
8929ecfa 1762 if (transid == trans->transid) {
79787eaa 1763 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1764 } else {
1765 btrfs_end_transaction(trans, root);
1766 }
a74a4b97
CM		 1767sleep:
1768 wake_up_process(root->fs_info->cleaner_kthread);
1769 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1770
a0acae0e 1771 if (!try_to_freeze()) {
a74a4b97 1772 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1773 if (!kthread_should_stop() &&
914b2007
JK		 1774 (!btrfs_transaction_blocked(root->fs_info) ||
1775 cannot_commit))
8929ecfa 1776 schedule_timeout(delay);
a74a4b97
CM		 1777 __set_current_state(TASK_RUNNING);
1778 }
1779 } while (!kthread_should_stop());
1780 return 0;
1781}
1782
af31f5e5
CM		 1783/*
1784 * this will find the highest generation in the array of
1785 * root backups. The index of the highest array is returned,
1786 * or -1 if we can't find anything.
1787 *
1788 * We check to make sure the array is valid by comparing the
1789 * generation of the latest root in the array with the generation
1790 * in the super block. If they don't match we pitch it.
1791 */
1792static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1793{
1794 u64 cur;
1795 int newest_index = -1;
1796 struct btrfs_root_backup *root_backup;
1797 int i;
1798
1799 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1800 root_backup = info->super_copy->super_roots + i;
1801 cur = btrfs_backup_tree_root_gen(root_backup);
1802 if (cur == newest_gen)
1803 newest_index = i;
1804 }
1805
1806 /* check to see if we actually wrapped around */
1807 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1808 root_backup = info->super_copy->super_roots;
1809 cur = btrfs_backup_tree_root_gen(root_backup);
1810 if (cur == newest_gen)
1811 newest_index = 0;
1812 }
1813 return newest_index;
1814}
1815
1816
1817/*
1818 * find the oldest backup so we know where to store new entries
1819 * in the backup array. This will set the backup_root_index
1820 * field in the fs_info struct
1821 */
1822static void find_oldest_super_backup(struct btrfs_fs_info *info,
1823 u64 newest_gen)
1824{
1825 int newest_index = -1;
1826
1827 newest_index = find_newest_super_backup(info, newest_gen);
1828 /* if there was garbage in there, just move along */
1829 if (newest_index == -1) {
1830 info->backup_root_index = 0;
1831 } else {
1832 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1833 }
1834}
1835
1836/*
1837 * copy all the root pointers into the super backup array.
1838 * this will bump the backup pointer by one when it is
1839 * done
1840 */
1841static void backup_super_roots(struct btrfs_fs_info *info)
1842{
1843 int next_backup;
1844 struct btrfs_root_backup *root_backup;
1845 int last_backup;
1846
1847 next_backup = info->backup_root_index;
1848 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1849 BTRFS_NUM_BACKUP_ROOTS;
1850
1851 /*
1852 * just overwrite the last backup if we're at the same generation
1853 * this happens only at umount
1854 */
1855 root_backup = info->super_for_commit->super_roots + last_backup;
1856 if (btrfs_backup_tree_root_gen(root_backup) ==
1857 btrfs_header_generation(info->tree_root->node))
1858 next_backup = last_backup;
1859
1860 root_backup = info->super_for_commit->super_roots + next_backup;
1861
1862 /*
1863 * make sure all of our padding and empty slots get zero filled
1864 * regardless of which ones we use today
1865 */
1866 memset(root_backup, 0, sizeof(*root_backup));
1867
1868 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1869
1870 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1871 btrfs_set_backup_tree_root_gen(root_backup,
1872 btrfs_header_generation(info->tree_root->node));
1873
1874 btrfs_set_backup_tree_root_level(root_backup,
1875 btrfs_header_level(info->tree_root->node));
1876
1877 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1878 btrfs_set_backup_chunk_root_gen(root_backup,
1879 btrfs_header_generation(info->chunk_root->node));
1880 btrfs_set_backup_chunk_root_level(root_backup,
1881 btrfs_header_level(info->chunk_root->node));
1882
1883 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1884 btrfs_set_backup_extent_root_gen(root_backup,
1885 btrfs_header_generation(info->extent_root->node));
1886 btrfs_set_backup_extent_root_level(root_backup,
1887 btrfs_header_level(info->extent_root->node));
1888
7c7e82a7
CM		 1889 /*
1890 * we might commit during log recovery, which happens before we set
1891 * the fs_root. Make sure it is valid before we fill it in.
1892 */
1893 if (info->fs_root && info->fs_root->node) {
1894 btrfs_set_backup_fs_root(root_backup,
1895 info->fs_root->node->start);
1896 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1897 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1898 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1899 btrfs_header_level(info->fs_root->node));
7c7e82a7 1900 }
af31f5e5
CM		 1901
1902 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1903 btrfs_set_backup_dev_root_gen(root_backup,
1904 btrfs_header_generation(info->dev_root->node));
1905 btrfs_set_backup_dev_root_level(root_backup,
1906 btrfs_header_level(info->dev_root->node));
1907
1908 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1909 btrfs_set_backup_csum_root_gen(root_backup,
1910 btrfs_header_generation(info->csum_root->node));
1911 btrfs_set_backup_csum_root_level(root_backup,
1912 btrfs_header_level(info->csum_root->node));
1913
1914 btrfs_set_backup_total_bytes(root_backup,
1915 btrfs_super_total_bytes(info->super_copy));
1916 btrfs_set_backup_bytes_used(root_backup,
1917 btrfs_super_bytes_used(info->super_copy));
1918 btrfs_set_backup_num_devices(root_backup,
1919 btrfs_super_num_devices(info->super_copy));
1920
1921 /*
1922 * if we don't copy this out to the super_copy, it won't get remembered
1923 * for the next commit
1924 */
1925 memcpy(&info->super_copy->super_roots,
1926 &info->super_for_commit->super_roots,
1927 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1928}
1929
1930/*
1931 * this copies info out of the root backup array and back into
1932 * the in-memory super block. It is meant to help iterate through
1933 * the array, so you send it the number of backups you've already
1934 * tried and the last backup index you used.
1935 *
1936 * this returns -1 when it has tried all the backups
1937 */
1938static noinline int next_root_backup(struct btrfs_fs_info *info,
1939 struct btrfs_super_block *super,
1940 int *num_backups_tried, int *backup_index)
1941{
1942 struct btrfs_root_backup *root_backup;
1943 int newest = *backup_index;
1944
1945 if (*num_backups_tried == 0) {
1946 u64 gen = btrfs_super_generation(super);
1947
1948 newest = find_newest_super_backup(info, gen);
1949 if (newest == -1)
1950 return -1;
1951
1952 *backup_index = newest;
1953 *num_backups_tried = 1;
1954 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1955 /* we've tried all the backups, all done */
1956 return -1;
1957 } else {
1958 /* jump to the next oldest backup */
1959 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1960 BTRFS_NUM_BACKUP_ROOTS;
1961 *backup_index = newest;
1962 *num_backups_tried += 1;
1963 }
1964 root_backup = super->super_roots + newest;
1965
1966 btrfs_set_super_generation(super,
1967 btrfs_backup_tree_root_gen(root_backup));
1968 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1969 btrfs_set_super_root_level(super,
1970 btrfs_backup_tree_root_level(root_backup));
1971 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1972
1973 /*
1974 * fixme: the total bytes and num_devices need to match or we should
1975 * need a fsck
1976 */
1977 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1978 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1979 return 0;
1980}
1981
7abadb64
LB		 1982/* helper to cleanup workers */
1983static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1984{
1985 btrfs_stop_workers(&fs_info->generic_worker);
1986 btrfs_stop_workers(&fs_info->fixup_workers);
1987 btrfs_stop_workers(&fs_info->delalloc_workers);
1988 btrfs_stop_workers(&fs_info->workers);
1989 btrfs_stop_workers(&fs_info->endio_workers);
1990 btrfs_stop_workers(&fs_info->endio_meta_workers);
1991 btrfs_stop_workers(&fs_info->endio_raid56_workers);
1992 btrfs_stop_workers(&fs_info->rmw_workers);
1993 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1994 btrfs_stop_workers(&fs_info->endio_write_workers);
1995 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1996 btrfs_stop_workers(&fs_info->submit_workers);
1997 btrfs_stop_workers(&fs_info->delayed_workers);
1998 btrfs_stop_workers(&fs_info->caching_workers);
1999 btrfs_stop_workers(&fs_info->readahead_workers);
2000 btrfs_stop_workers(&fs_info->flush_workers);
2f232036 2001 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
7abadb64
LB		 2002}
2003
af31f5e5
CM		 2004/* helper to cleanup tree roots */
2005static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2006{
2007 free_extent_buffer(info->tree_root->node);
2008 free_extent_buffer(info->tree_root->commit_root);
af31f5e5
CM		 2009 info->tree_root->node = NULL;
2010 info->tree_root->commit_root = NULL;
655b09fe
JB		 2011
2012 if (info->dev_root) {
2013 free_extent_buffer(info->dev_root->node);
2014 free_extent_buffer(info->dev_root->commit_root);
2015 info->dev_root->node = NULL;
2016 info->dev_root->commit_root = NULL;
2017 }
2018 if (info->extent_root) {
2019 free_extent_buffer(info->extent_root->node);
2020 free_extent_buffer(info->extent_root->commit_root);
2021 info->extent_root->node = NULL;
2022 info->extent_root->commit_root = NULL;
2023 }
2024 if (info->csum_root) {
2025 free_extent_buffer(info->csum_root->node);
2026 free_extent_buffer(info->csum_root->commit_root);
2027 info->csum_root->node = NULL;
2028 info->csum_root->commit_root = NULL;
2029 }
bcef60f2 2030 if (info->quota_root) {
655b09fe
JB		 2031 free_extent_buffer(info->quota_root->node);
2032 free_extent_buffer(info->quota_root->commit_root);
bcef60f2
AJ		 2033 info->quota_root->node = NULL;
2034 info->quota_root->commit_root = NULL;
2035 }
af31f5e5
CM		 2036 if (chunk_root) {
2037 free_extent_buffer(info->chunk_root->node);
2038 free_extent_buffer(info->chunk_root->commit_root);
2039 info->chunk_root->node = NULL;
2040 info->chunk_root->commit_root = NULL;
2041 }
2042}
2043
171f6537
JB		 2044static void del_fs_roots(struct btrfs_fs_info *fs_info)
2045{
2046 int ret;
2047 struct btrfs_root *gang[8];
2048 int i;
2049
2050 while (!list_empty(&fs_info->dead_roots)) {
2051 gang[0] = list_entry(fs_info->dead_roots.next,
2052 struct btrfs_root, root_list);
2053 list_del(&gang[0]->root_list);
2054
2055 if (gang[0]->in_radix) {
cb517eab 2056 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
171f6537
JB		 2057 } else {
2058 free_extent_buffer(gang[0]->node);
2059 free_extent_buffer(gang[0]->commit_root);
b0feb9d9 2060 btrfs_put_fs_root(gang[0]);
171f6537
JB		 2061 }
2062 }
2063
2064 while (1) {
2065 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2066 (void **)gang, 0,
2067 ARRAY_SIZE(gang));
2068 if (!ret)
2069 break;
2070 for (i = 0; i < ret; i++)
cb517eab 2071 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
171f6537
JB		 2072 }
2073}
af31f5e5 2074
ad2b2c80
AV		 2075int open_ctree(struct super_block *sb,
2076 struct btrfs_fs_devices *fs_devices,
2077 char *options)
2e635a27 2078{
db94535d
CM		 2079 u32 sectorsize;
2080 u32 nodesize;
2081 u32 leafsize;
2082 u32 blocksize;
87ee04eb 2083 u32 stripesize;
84234f3a 2084 u64 generation;
f2b636e8 2085 u64 features;
3de4586c 2086 struct btrfs_key location;
a061fc8d 2087 struct buffer_head *bh;
4d34b278 2088 struct btrfs_super_block *disk_super;
815745cf 2089 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2090 struct btrfs_root *tree_root;
4d34b278
ID		 2091 struct btrfs_root *extent_root;
2092 struct btrfs_root *csum_root;
2093 struct btrfs_root *chunk_root;
2094 struct btrfs_root *dev_root;
bcef60f2 2095 struct btrfs_root *quota_root;
e02119d5 2096 struct btrfs_root *log_tree_root;
eb60ceac 2097 int ret;
e58ca020 2098 int err = -EINVAL;
af31f5e5
CM		 2099 int num_backups_tried = 0;
2100 int backup_index = 0;
4543df7e 2101
f84a8bd6 2102 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e 2103 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
cb517eab 2104 if (!tree_root || !chunk_root) {
39279cc3
CM		 2105 err = -ENOMEM;
2106 goto fail;
2107 }
76dda93c
YZ		 2108
2109 ret = init_srcu_struct(&fs_info->subvol_srcu);
2110 if (ret) {
2111 err = ret;
2112 goto fail;
2113 }
2114
2115 ret = setup_bdi(fs_info, &fs_info->bdi);
2116 if (ret) {
2117 err = ret;
2118 goto fail_srcu;
2119 }
2120
e2d84521
MX		 2121 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2122 if (ret) {
2123 err = ret;
2124 goto fail_bdi;
2125 }
2126 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2127 (1 + ilog2(nr_cpu_ids));
2128
963d678b
MX		 2129 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2130 if (ret) {
2131 err = ret;
2132 goto fail_dirty_metadata_bytes;
2133 }
2134
76dda93c
YZ		 2135 fs_info->btree_inode = new_inode(sb);
2136 if (!fs_info->btree_inode) {
2137 err = -ENOMEM;
963d678b 2138 goto fail_delalloc_bytes;
76dda93c
YZ		 2139 }
2140
a6591715 2141 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2142
76dda93c 2143 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 2144 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2145 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2146 INIT_LIST_HEAD(&fs_info->delayed_iputs);
eb73c1b7 2147 INIT_LIST_HEAD(&fs_info->delalloc_roots);
11833d66 2148 INIT_LIST_HEAD(&fs_info->caching_block_groups);
eb73c1b7 2149 spin_lock_init(&fs_info->delalloc_root_lock);
a4abeea4 2150 spin_lock_init(&fs_info->trans_lock);
76dda93c 2151 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2152 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2153 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2154 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2155 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2156 spin_lock_init(&fs_info->super_lock);
f29021b2 2157 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2158 mutex_init(&fs_info->reloc_mutex);
de98ced9 2159 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2160
58176a96 2161 init_completion(&fs_info->kobj_unregister);
0b86a832 2162 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2163 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2164 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2165 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2166 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2167 BTRFS_BLOCK_RSV_GLOBAL);
2168 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2169 BTRFS_BLOCK_RSV_DELALLOC);
2170 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2171 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2172 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2173 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2174 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2175 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2176 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2177 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2178 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2179 atomic_set(&fs_info->defrag_running, 0);
fc36ed7e 2180 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2181 fs_info->sb = sb;
6f568d35 2182 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2183 fs_info->metadata_ratio = 0;
4cb5300b 2184 fs_info->defrag_inodes = RB_ROOT;
a4abeea4 2185 fs_info->trans_no_join = 0;
2bf64758 2186 fs_info->free_chunk_space = 0;
f29021b2 2187 fs_info->tree_mod_log = RB_ROOT;
c8b97818 2188
90519d66
AJ		 2189 /* readahead state */
2190 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2191 spin_lock_init(&fs_info->reada_lock);
c8b97818 2192
b34b086c
CM		 2193 fs_info->thread_pool_size = min_t(unsigned long,
2194 num_online_cpus() + 2, 8);
0afbaf8c 2195
3eaa2885
CM		 2196 INIT_LIST_HEAD(&fs_info->ordered_extents);
2197 spin_lock_init(&fs_info->ordered_extent_lock);
16cdcec7
MX		 2198 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2199 GFP_NOFS);
2200 if (!fs_info->delayed_root) {
2201 err = -ENOMEM;
2202 goto fail_iput;
2203 }
2204 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2205
a2de733c
AJ		 2206 mutex_init(&fs_info->scrub_lock);
2207 atomic_set(&fs_info->scrubs_running, 0);
2208 atomic_set(&fs_info->scrub_pause_req, 0);
2209 atomic_set(&fs_info->scrubs_paused, 0);
2210 atomic_set(&fs_info->scrub_cancel_req, 0);
2211 init_waitqueue_head(&fs_info->scrub_pause_wait);
2212 init_rwsem(&fs_info->scrub_super_lock);
2213 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2214#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2215 fs_info->check_integrity_print_mask = 0;
2216#endif
a2de733c 2217
c9e9f97b
ID		 2218 spin_lock_init(&fs_info->balance_lock);
2219 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2220 atomic_set(&fs_info->balance_running, 0);
2221 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2222 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2223 fs_info->balance_ctl = NULL;
837d5b6e 2224 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2225
a061fc8d
CM		 2226 sb->s_blocksize = 4096;
2227 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2228 sb->s_bdi = &fs_info->bdi;
a061fc8d 2229
76dda93c 2230 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2231 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2232 /*
2233 * we set the i_size on the btree inode to the max possible int.
2234 * the real end of the address space is determined by all of
2235 * the devices in the system
2236 */
2237 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2238 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2239 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2240
5d4f98a2 2241 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2242 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2243 fs_info->btree_inode->i_mapping);
0b32f4bb 2244 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2245 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2246
2247 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2248
76dda93c
YZ		 2249 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2250 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2251 sizeof(struct btrfs_key));
72ac3c0d
JB		 2252 set_bit(BTRFS_INODE_DUMMY,
2253 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
c65ddb52 2254 insert_inode_hash(fs_info->btree_inode);
76dda93c 2255
0f9dd46c 2256 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2257 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2258 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2259
11833d66 2260 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2261 fs_info->btree_inode->i_mapping);
11833d66 2262 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2263 fs_info->btree_inode->i_mapping);
11833d66 2264 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2265 fs_info->do_barriers = 1;
e18e4809 2266
39279cc3 2267
5a3f23d5 2268 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 2269 mutex_init(&fs_info->tree_log_mutex);
925baedd 2270 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2271 mutex_init(&fs_info->transaction_kthread_mutex);
2272 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2273 mutex_init(&fs_info->volume_mutex);
276e680d 2274 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2275 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2276 init_rwsem(&fs_info->subvol_sem);
e922e087
SB		 2277 fs_info->dev_replace.lock_owner = 0;
2278 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2279 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2280 mutex_init(&fs_info->dev_replace.lock_management_lock);
2281 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2282
416ac51d 2283 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2284 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2285 fs_info->qgroup_tree = RB_ROOT;
2286 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2287 fs_info->qgroup_seq = 1;
2288 fs_info->quota_enabled = 0;
2289 fs_info->pending_quota_state = 0;
1e8f9158 2290 fs_info->qgroup_ulist = NULL;
2f232036 2291 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2292
fa9c0d79
CM		 2293 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2294 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2295
e6dcd2dc 2296 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2297 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2298 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2299 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2300
53b381b3
DW		 2301 ret = btrfs_alloc_stripe_hash_table(fs_info);
2302 if (ret) {
83c8266a 2303 err = ret;
53b381b3
DW		 2304 goto fail_alloc;
2305 }
2306
0b86a832 2307 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2308 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2309
3c4bb26b 2310 invalidate_bdev(fs_devices->latest_bdev);
1104a885
DS		 2311
2312 /*
2313 * Read super block and check the signature bytes only
2314 */
a512bbf8 2315 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2316 if (!bh) {
2317 err = -EINVAL;
16cdcec7 2318 goto fail_alloc;
20b45077 2319 }
39279cc3 2320
1104a885
DS		 2321 /*
2322 * We want to check superblock checksum, the type is stored inside.
2323 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2324 */
2325 if (btrfs_check_super_csum(bh->b_data)) {
2326 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2327 err = -EINVAL;
2328 goto fail_alloc;
2329 }
2330
2331 /*
2332 * super_copy is zeroed at allocation time and we never touch the
2333 * following bytes up to INFO_SIZE, the checksum is calculated from
2334 * the whole block of INFO_SIZE
2335 */
6c41761f
DS		 2336 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2337 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2338 sizeof(*fs_info->super_for_commit));
a061fc8d 2339 brelse(bh);
5f39d397 2340
6c41761f 2341 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2342
1104a885
DS		 2343 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2344 if (ret) {
2345 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2346 err = -EINVAL;
2347 goto fail_alloc;
2348 }
2349
6c41761f 2350 disk_super = fs_info->super_copy;
0f7d52f4 2351 if (!btrfs_super_root(disk_super))
16cdcec7 2352 goto fail_alloc;
0f7d52f4 2353
acce952b 2354 /* check FS state, whether FS is broken. */
87533c47
MX		 2355 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2356 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2357
af31f5e5
CM		 2358 /*
2359 * run through our array of backup supers and setup
2360 * our ring pointer to the oldest one
2361 */
2362 generation = btrfs_super_generation(disk_super);
2363 find_oldest_super_backup(fs_info, generation);
2364
75e7cb7f
LB		 2365 /*
2366 * In the long term, we'll store the compression type in the super
2367 * block, and it'll be used for per file compression control.
2368 */
2369 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2370
2b82032c
YZ		 2371 ret = btrfs_parse_options(tree_root, options);
2372 if (ret) {
2373 err = ret;
16cdcec7 2374 goto fail_alloc;
2b82032c 2375 }
dfe25020 2376
f2b636e8
JB		 2377 features = btrfs_super_incompat_flags(disk_super) &
2378 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2379 if (features) {
2380 printk(KERN_ERR "BTRFS: couldn't mount because of "
2381 "unsupported optional features (%Lx).\n",
21380931 2382 (unsigned long long)features);
f2b636e8 2383 err = -EINVAL;
16cdcec7 2384 goto fail_alloc;
f2b636e8
JB		 2385 }
2386
727011e0
CM		 2387 if (btrfs_super_leafsize(disk_super) !=
2388 btrfs_super_nodesize(disk_super)) {
2389 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2390 "blocksizes don't match. node %d leaf %d\n",
2391 btrfs_super_nodesize(disk_super),
2392 btrfs_super_leafsize(disk_super));
2393 err = -EINVAL;
2394 goto fail_alloc;
2395 }
2396 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2397 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2398 "blocksize (%d) was too large\n",
2399 btrfs_super_leafsize(disk_super));
2400 err = -EINVAL;
2401 goto fail_alloc;
2402 }
2403
5d4f98a2 2404 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2405 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2406 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2407 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2408
3173a18f
JB		 2409 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2410 printk(KERN_ERR "btrfs: has skinny extents\n");
2411
727011e0
CM		 2412 /*
2413 * flag our filesystem as having big metadata blocks if
2414 * they are bigger than the page size
2415 */
2416 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2417 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2418 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2419 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2420 }
2421
bc3f116f
CM		 2422 nodesize = btrfs_super_nodesize(disk_super);
2423 leafsize = btrfs_super_leafsize(disk_super);
2424 sectorsize = btrfs_super_sectorsize(disk_super);
2425 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2426 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2427 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2428
2429 /*
2430 * mixed block groups end up with duplicate but slightly offset
2431 * extent buffers for the same range. It leads to corruptions
2432 */
2433 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2434 (sectorsize != leafsize)) {
2435 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2436 "are not allowed for mixed block groups on %s\n",
2437 sb->s_id);
2438 goto fail_alloc;
2439 }
2440
ceda0864
MX		 2441 /*
2442 * Needn't use the lock because there is no other task which will
2443 * update the flag.
2444 */
a6fa6fae 2445 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2446
f2b636e8
JB		 2447 features = btrfs_super_compat_ro_flags(disk_super) &
2448 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2449 if (!(sb->s_flags & MS_RDONLY) && features) {
2450 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2451 "unsupported option features (%Lx).\n",
21380931 2452 (unsigned long long)features);
f2b636e8 2453 err = -EINVAL;
16cdcec7 2454 goto fail_alloc;
f2b636e8 2455 }
61d92c32
CM		 2456
2457 btrfs_init_workers(&fs_info->generic_worker,
2458 "genwork", 1, NULL);
2459
5443be45 2460 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 2461 fs_info->thread_pool_size,
2462 &fs_info->generic_worker);
c8b97818 2463
771ed689 2464 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 2465 fs_info->thread_pool_size,
2466 &fs_info->generic_worker);
771ed689 2467
8ccf6f19
MX		 2468 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2469 fs_info->thread_pool_size,
2470 &fs_info->generic_worker);
2471
5443be45 2472 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 2473 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 2474 fs_info->thread_pool_size),
2475 &fs_info->generic_worker);
61b49440 2476
bab39bf9
JB		 2477 btrfs_init_workers(&fs_info->caching_workers, "cache",
2478 2, &fs_info->generic_worker);
2479
61b49440
CM		 2480 /* a higher idle thresh on the submit workers makes it much more
2481 * likely that bios will be send down in a sane order to the
2482 * devices
2483 */
2484 fs_info->submit_workers.idle_thresh = 64;
53863232 2485
771ed689 2486 fs_info->workers.idle_thresh = 16;
4a69a410 2487 fs_info->workers.ordered = 1;
61b49440 2488
771ed689
CM		 2489 fs_info->delalloc_workers.idle_thresh = 2;
2490 fs_info->delalloc_workers.ordered = 1;
2491
61d92c32
CM		 2492 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2493 &fs_info->generic_worker);
5443be45 2494 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2495 fs_info->thread_pool_size,
2496 &fs_info->generic_worker);
d20f7043 2497 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2498 fs_info->thread_pool_size,
2499 &fs_info->generic_worker);
cad321ad 2500 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2501 "endio-meta-write", fs_info->thread_pool_size,
2502 &fs_info->generic_worker);
53b381b3
DW		 2503 btrfs_init_workers(&fs_info->endio_raid56_workers,
2504 "endio-raid56", fs_info->thread_pool_size,
2505 &fs_info->generic_worker);
2506 btrfs_init_workers(&fs_info->rmw_workers,
2507 "rmw", fs_info->thread_pool_size,
2508 &fs_info->generic_worker);
5443be45 2509 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2510 fs_info->thread_pool_size,
2511 &fs_info->generic_worker);
0cb59c99
JB		 2512 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2513 1, &fs_info->generic_worker);
16cdcec7
MX		 2514 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2515 fs_info->thread_pool_size,
2516 &fs_info->generic_worker);
90519d66
AJ		 2517 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2518 fs_info->thread_pool_size,
2519 &fs_info->generic_worker);
2f232036
JS		 2520 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2521 &fs_info->generic_worker);
61b49440
CM		 2522
2523 /*
2524 * endios are largely parallel and should have a very
2525 * low idle thresh
2526 */
2527 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2528 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2529 fs_info->endio_raid56_workers.idle_thresh = 4;
2530 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2531
9042846b
CM		 2532 fs_info->endio_write_workers.idle_thresh = 2;
2533 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2534 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2535
0dc3b84a
JB		 2536 /*
2537 * btrfs_start_workers can really only fail because of ENOMEM so just
2538 * return -ENOMEM if any of these fail.
2539 */
2540 ret = btrfs_start_workers(&fs_info->workers);
2541 ret |= btrfs_start_workers(&fs_info->generic_worker);
2542 ret |= btrfs_start_workers(&fs_info->submit_workers);
2543 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2544 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2545 ret |= btrfs_start_workers(&fs_info->endio_workers);
2546 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2547 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2548 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2549 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2550 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2551 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2552 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2553 ret |= btrfs_start_workers(&fs_info->caching_workers);
2554 ret |= btrfs_start_workers(&fs_info->readahead_workers);
8ccf6f19 2555 ret |= btrfs_start_workers(&fs_info->flush_workers);
2f232036 2556 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
0dc3b84a 2557 if (ret) {
fed425c7 2558 err = -ENOMEM;
0dc3b84a
JB		 2559 goto fail_sb_buffer;
2560 }
4543df7e 2561
4575c9cc 2562 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2563 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2564 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2565
db94535d
CM		 2566 tree_root->nodesize = nodesize;
2567 tree_root->leafsize = leafsize;
2568 tree_root->sectorsize = sectorsize;
87ee04eb 2569 tree_root->stripesize = stripesize;
a061fc8d
CM		 2570
2571 sb->s_blocksize = sectorsize;
2572 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2573
cdb4c574 2574 if (disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) {
d397712b 2575 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2576 goto fail_sb_buffer;
2577 }
19c00ddc 2578
8d082fb7
LB		 2579 if (sectorsize != PAGE_SIZE) {
2580 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2581 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2582 goto fail_sb_buffer;
2583 }
2584
925baedd 2585 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2586 ret = btrfs_read_sys_array(tree_root);
925baedd 2587 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2588 if (ret) {
d397712b
CM		 2589 printk(KERN_WARNING "btrfs: failed to read the system "
2590 "array on %s\n", sb->s_id);
5d4f98a2 2591 goto fail_sb_buffer;
84eed90f 2592 }
0b86a832
CM		 2593
2594 blocksize = btrfs_level_size(tree_root,
2595 btrfs_super_chunk_root_level(disk_super));
84234f3a 2596 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2597
2598 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2599 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2600
2601 chunk_root->node = read_tree_block(chunk_root,
2602 btrfs_super_chunk_root(disk_super),
84234f3a 2603 blocksize, generation);
416bc658
JB		 2604 if (!chunk_root->node ||
2605 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
83121942
DW		 2606 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2607 sb->s_id);
af31f5e5 2608 goto fail_tree_roots;
83121942 2609 }
5d4f98a2
YZ		 2610 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2611 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2612
e17cade2 2613 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 2614 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2615 BTRFS_UUID_SIZE);
e17cade2 2616
0b86a832 2617 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2618 if (ret) {
d397712b
CM		 2619 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2620 sb->s_id);
af31f5e5 2621 goto fail_tree_roots;
2b82032c 2622 }
0b86a832 2623
8dabb742
SB		 2624 /*
2625 * keep the device that is marked to be the target device for the
2626 * dev_replace procedure
2627 */
2628 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2629
a6b0d5c8
CM		 2630 if (!fs_devices->latest_bdev) {
2631 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2632 sb->s_id);
2633 goto fail_tree_roots;
2634 }
2635
af31f5e5 2636retry_root_backup:
db94535d
CM		 2637 blocksize = btrfs_level_size(tree_root,
2638 btrfs_super_root_level(disk_super));
84234f3a 2639 generation = btrfs_super_generation(disk_super);
0b86a832 2640
e20d96d6 2641 tree_root->node = read_tree_block(tree_root,
db94535d 2642 btrfs_super_root(disk_super),
84234f3a 2643 blocksize, generation);
af31f5e5
CM		 2644 if (!tree_root->node ||
2645 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
83121942
DW		 2646 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2647 sb->s_id);
af31f5e5
CM		 2648
2649 goto recovery_tree_root;
83121942 2650 }
af31f5e5 2651
5d4f98a2
YZ		 2652 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2653 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d 2654
cb517eab
MX		 2655 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2656 location.type = BTRFS_ROOT_ITEM_KEY;
2657 location.offset = 0;
2658
2659 extent_root = btrfs_read_tree_root(tree_root, &location);
2660 if (IS_ERR(extent_root)) {
2661 ret = PTR_ERR(extent_root);
af31f5e5 2662 goto recovery_tree_root;
cb517eab 2663 }
0b86a832 2664 extent_root->track_dirty = 1;
cb517eab 2665 fs_info->extent_root = extent_root;
0b86a832 2666
cb517eab
MX		 2667 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2668 dev_root = btrfs_read_tree_root(tree_root, &location);
2669 if (IS_ERR(dev_root)) {
2670 ret = PTR_ERR(dev_root);
af31f5e5 2671 goto recovery_tree_root;
cb517eab 2672 }
5d4f98a2 2673 dev_root->track_dirty = 1;
cb517eab
MX		 2674 fs_info->dev_root = dev_root;
2675 btrfs_init_devices_late(fs_info);
3768f368 2676
cb517eab
MX		 2677 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2678 csum_root = btrfs_read_tree_root(tree_root, &location);
2679 if (IS_ERR(csum_root)) {
2680 ret = PTR_ERR(csum_root);
af31f5e5 2681 goto recovery_tree_root;
cb517eab 2682 }
d20f7043 2683 csum_root->track_dirty = 1;
cb517eab 2684 fs_info->csum_root = csum_root;
d20f7043 2685
cb517eab
MX		 2686 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2687 quota_root = btrfs_read_tree_root(tree_root, &location);
2688 if (!IS_ERR(quota_root)) {
bcef60f2
AJ		 2689 quota_root->track_dirty = 1;
2690 fs_info->quota_enabled = 1;
2691 fs_info->pending_quota_state = 1;
cb517eab 2692 fs_info->quota_root = quota_root;
bcef60f2
AJ		 2693 }
2694
8929ecfa
YZ		 2695 fs_info->generation = generation;
2696 fs_info->last_trans_committed = generation;
8929ecfa 2697
68310a5e
ID		 2698 ret = btrfs_recover_balance(fs_info);
2699 if (ret) {
2700 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2701 goto fail_block_groups;
2702 }
2703
733f4fbb
SB		 2704 ret = btrfs_init_dev_stats(fs_info);
2705 if (ret) {
2706 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2707 ret);
2708 goto fail_block_groups;
2709 }
2710
8dabb742
SB		 2711 ret = btrfs_init_dev_replace(fs_info);
2712 if (ret) {
2713 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2714 goto fail_block_groups;
2715 }
2716
2717 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2718
c59021f8 2719 ret = btrfs_init_space_info(fs_info);
2720 if (ret) {
2721 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2722 goto fail_block_groups;
2723 }
2724
1b1d1f66
JB		 2725 ret = btrfs_read_block_groups(extent_root);
2726 if (ret) {
2727 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2728 goto fail_block_groups;
2729 }
5af3e8cc
SB		 2730 fs_info->num_tolerated_disk_barrier_failures =
2731 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2732 if (fs_info->fs_devices->missing_devices >
2733 fs_info->num_tolerated_disk_barrier_failures &&
2734 !(sb->s_flags & MS_RDONLY)) {
2735 printk(KERN_WARNING
2736 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2737 goto fail_block_groups;
2738 }
9078a3e1 2739
a74a4b97
CM		 2740 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2741 "btrfs-cleaner");
57506d50 2742 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2743 goto fail_block_groups;
a74a4b97
CM		 2744
2745 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2746 tree_root,
2747 "btrfs-transaction");
57506d50 2748 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2749 goto fail_cleaner;
a74a4b97 2750
c289811c
CM		 2751 if (!btrfs_test_opt(tree_root, SSD) &&
2752 !btrfs_test_opt(tree_root, NOSSD) &&
2753 !fs_info->fs_devices->rotating) {
2754 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2755 "mode\n");
2756 btrfs_set_opt(fs_info->mount_opt, SSD);
2757 }
2758
21adbd5c
SB		 2759#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2760 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2761 ret = btrfsic_mount(tree_root, fs_devices,
2762 btrfs_test_opt(tree_root,
2763 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2764 1 : 0,
2765 fs_info->check_integrity_print_mask);
2766 if (ret)
2767 printk(KERN_WARNING "btrfs: failed to initialize"
2768 " integrity check module %s\n", sb->s_id);
2769 }
2770#endif
bcef60f2
AJ		 2771 ret = btrfs_read_qgroup_config(fs_info);
2772 if (ret)
2773 goto fail_trans_kthread;
21adbd5c 2774
acce952b 2775 /* do not make disk changes in broken FS */
68ce9682 2776 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2777 u64 bytenr = btrfs_super_log_root(disk_super);
2778
7c2ca468 2779 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2780 printk(KERN_WARNING "Btrfs log replay required "
2781 "on RO media\n");
7c2ca468 2782 err = -EIO;
bcef60f2 2783 goto fail_qgroup;
7c2ca468 2784 }
e02119d5
CM		 2785 blocksize =
2786 btrfs_level_size(tree_root,
2787 btrfs_super_log_root_level(disk_super));
d18a2c44 2788
6f07e42e 2789 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2790 if (!log_tree_root) {
2791 err = -ENOMEM;
bcef60f2 2792 goto fail_qgroup;
676e4c86 2793 }
e02119d5
CM		 2794
2795 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2796 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2797
2798 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2799 blocksize,
2800 generation + 1);
416bc658
JB		 2801 if (!log_tree_root->node ||
2802 !extent_buffer_uptodate(log_tree_root->node)) {
2803 printk(KERN_ERR "btrfs: failed to read log tree\n");
2804 free_extent_buffer(log_tree_root->node);
2805 kfree(log_tree_root);
2806 goto fail_trans_kthread;
2807 }
79787eaa 2808 /* returns with log_tree_root freed on success */
e02119d5 2809 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2810 if (ret) {
2811 btrfs_error(tree_root->fs_info, ret,
2812 "Failed to recover log tree");
2813 free_extent_buffer(log_tree_root->node);
2814 kfree(log_tree_root);
2815 goto fail_trans_kthread;
2816 }
e556ce2c
YZ		 2817
2818 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2819 ret = btrfs_commit_super(tree_root);
2820 if (ret)
2821 goto fail_trans_kthread;
e556ce2c 2822 }
e02119d5 2823 }
1a40e23b 2824
76dda93c 2825 ret = btrfs_find_orphan_roots(tree_root);
79787eaa
JM		 2826 if (ret)
2827 goto fail_trans_kthread;
76dda93c 2828
7c2ca468 2829 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2830 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2
ID		 2831 if (ret)
2832 goto fail_trans_kthread;
d68fc57b 2833
5d4f98a2 2834 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2835 if (ret < 0) {
2836 printk(KERN_WARNING
2837 "btrfs: failed to recover relocation\n");
2838 err = -EINVAL;
bcef60f2 2839 goto fail_qgroup;
d7ce5843 2840 }
7c2ca468 2841 }
1a40e23b 2842
3de4586c
CM		 2843 location.objectid = BTRFS_FS_TREE_OBJECTID;
2844 location.type = BTRFS_ROOT_ITEM_KEY;
cb517eab 2845 location.offset = 0;
3de4586c 2846
3de4586c 2847 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3140c9a3
DC		 2848 if (IS_ERR(fs_info->fs_root)) {
2849 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2850 goto fail_qgroup;
3140c9a3 2851 }
c289811c 2852
2b6ba629
ID		 2853 if (sb->s_flags & MS_RDONLY)
2854 return 0;
59641015 2855
2b6ba629
ID		 2856 down_read(&fs_info->cleanup_work_sem);
2857 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2858 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2859 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2860 close_ctree(tree_root);
2861 return ret;
2862 }
2863 up_read(&fs_info->cleanup_work_sem);
59641015 2864
2b6ba629
ID		 2865 ret = btrfs_resume_balance_async(fs_info);
2866 if (ret) {
2867 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2868 close_ctree(tree_root);
2869 return ret;
e3acc2a6
JB		 2870 }
2871
8dabb742
SB		 2872 ret = btrfs_resume_dev_replace_async(fs_info);
2873 if (ret) {
2874 pr_warn("btrfs: failed to resume dev_replace\n");
2875 close_ctree(tree_root);
2876 return ret;
2877 }
2878
ad2b2c80 2879 return 0;
39279cc3 2880
bcef60f2
AJ		 2881fail_qgroup:
2882 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2883fail_trans_kthread:
2884 kthread_stop(fs_info->transaction_kthread);
54067ae9 2885 btrfs_cleanup_transaction(fs_info->tree_root);
7b5ff90e 2886 del_fs_roots(fs_info);
3f157a2f 2887fail_cleaner:
a74a4b97 2888 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2889
2890 /*
2891 * make sure we're done with the btree inode before we stop our
2892 * kthreads
2893 */
2894 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2895
1b1d1f66 2896fail_block_groups:
54067ae9 2897 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2898 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2899
2900fail_tree_roots:
2901 free_root_pointers(fs_info, 1);
2b8195bb 2902 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2903
39279cc3 2904fail_sb_buffer:
7abadb64 2905 btrfs_stop_all_workers(fs_info);
16cdcec7 2906fail_alloc:
4543df7e 2907fail_iput:
586e46e2
ID		 2908 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2909
4543df7e 2910 iput(fs_info->btree_inode);
963d678b
MX		 2911fail_delalloc_bytes:
2912 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 2913fail_dirty_metadata_bytes:
2914 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 2915fail_bdi:
7e662854 2916 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2917fail_srcu:
2918 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2919fail:
53b381b3 2920 btrfs_free_stripe_hash_table(fs_info);
586e46e2 2921 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 2922 return err;
af31f5e5
CM		 2923
2924recovery_tree_root:
af31f5e5
CM		 2925 if (!btrfs_test_opt(tree_root, RECOVERY))
2926 goto fail_tree_roots;
2927
2928 free_root_pointers(fs_info, 0);
2929
2930 /* don't use the log in recovery mode, it won't be valid */
2931 btrfs_set_super_log_root(disk_super, 0);
2932
2933 /* we can't trust the free space cache either */
2934 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2935
2936 ret = next_root_backup(fs_info, fs_info->super_copy,
2937 &num_backups_tried, &backup_index);
2938 if (ret == -1)
2939 goto fail_block_groups;
2940 goto retry_root_backup;
eb60ceac
CM		 2941}
2942
f2984462
CM		 2943static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2944{
f2984462
CM		 2945 if (uptodate) {
2946 set_buffer_uptodate(bh);
2947 } else {
442a4f63
SB		 2948 struct btrfs_device *device = (struct btrfs_device *)
2949 bh->b_private;
2950
606686ee
JB		 2951 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2952 "I/O error on %s\n",
2953 rcu_str_deref(device->name));
1259ab75
CM		 2954 /* note, we dont' set_buffer_write_io_error because we have
2955 * our own ways of dealing with the IO errors
2956 */
f2984462 2957 clear_buffer_uptodate(bh);
442a4f63 2958 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 2959 }
2960 unlock_buffer(bh);
2961 put_bh(bh);
2962}
2963
a512bbf8
YZ		 2964struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2965{
2966 struct buffer_head *bh;
2967 struct buffer_head *latest = NULL;
2968 struct btrfs_super_block *super;
2969 int i;
2970 u64 transid = 0;
2971 u64 bytenr;
2972
2973 /* we would like to check all the supers, but that would make
2974 * a btrfs mount succeed after a mkfs from a different FS.
2975 * So, we need to add a special mount option to scan for
2976 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2977 */
2978 for (i = 0; i < 1; i++) {
2979 bytenr = btrfs_sb_offset(i);
2980 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2981 break;
2982 bh = __bread(bdev, bytenr / 4096, 4096);
2983 if (!bh)
2984 continue;
2985
2986 super = (struct btrfs_super_block *)bh->b_data;
2987 if (btrfs_super_bytenr(super) != bytenr ||
cdb4c574 2988 super->magic != cpu_to_le64(BTRFS_MAGIC)) {
a512bbf8
YZ		 2989 brelse(bh);
2990 continue;
2991 }
2992
2993 if (!latest || btrfs_super_generation(super) > transid) {
2994 brelse(latest);
2995 latest = bh;
2996 transid = btrfs_super_generation(super);
2997 } else {
2998 brelse(bh);
2999 }
3000 }
3001 return latest;
3002}
3003
4eedeb75
HH		 3004/*
3005 * this should be called twice, once with wait == 0 and
3006 * once with wait == 1. When wait == 0 is done, all the buffer heads
3007 * we write are pinned.
3008 *
3009 * They are released when wait == 1 is done.
3010 * max_mirrors must be the same for both runs, and it indicates how
3011 * many supers on this one device should be written.
3012 *
3013 * max_mirrors == 0 means to write them all.
3014 */
a512bbf8
YZ		 3015static int write_dev_supers(struct btrfs_device *device,
3016 struct btrfs_super_block *sb,
3017 int do_barriers, int wait, int max_mirrors)
3018{
3019 struct buffer_head *bh;
3020 int i;
3021 int ret;
3022 int errors = 0;
3023 u32 crc;
3024 u64 bytenr;
a512bbf8
YZ		 3025
3026 if (max_mirrors == 0)
3027 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3028
a512bbf8
YZ		 3029 for (i = 0; i < max_mirrors; i++) {
3030 bytenr = btrfs_sb_offset(i);
3031 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3032 break;
3033
3034 if (wait) {
3035 bh = __find_get_block(device->bdev, bytenr / 4096,
3036 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3037 if (!bh) {
3038 errors++;
3039 continue;
3040 }
a512bbf8 3041 wait_on_buffer(bh);
4eedeb75
HH		 3042 if (!buffer_uptodate(bh))
3043 errors++;
3044
3045 /* drop our reference */
3046 brelse(bh);
3047
3048 /* drop the reference from the wait == 0 run */
3049 brelse(bh);
3050 continue;
a512bbf8
YZ		 3051 } else {
3052 btrfs_set_super_bytenr(sb, bytenr);
3053
3054 crc = ~(u32)0;
b0496686 3055 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 3056 BTRFS_CSUM_SIZE, crc,
3057 BTRFS_SUPER_INFO_SIZE -
3058 BTRFS_CSUM_SIZE);
3059 btrfs_csum_final(crc, sb->csum);
3060
4eedeb75
HH		 3061 /*
3062 * one reference for us, and we leave it for the
3063 * caller
3064 */
a512bbf8
YZ		 3065 bh = __getblk(device->bdev, bytenr / 4096,
3066 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3067 if (!bh) {
3068 printk(KERN_ERR "btrfs: couldn't get super "
3069 "buffer head for bytenr %Lu\n", bytenr);
3070 errors++;
3071 continue;
3072 }
3073
a512bbf8
YZ		 3074 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3075
4eedeb75 3076 /* one reference for submit_bh */
a512bbf8 3077 get_bh(bh);
4eedeb75
HH		 3078
3079 set_buffer_uptodate(bh);
a512bbf8
YZ		 3080 lock_buffer(bh);
3081 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3082 bh->b_private = device;
a512bbf8
YZ		 3083 }
3084
387125fc
CM		 3085 /*
3086 * we fua the first super. The others we allow
3087 * to go down lazy.
3088 */
21adbd5c 3089 ret = btrfsic_submit_bh(WRITE_FUA, bh);
4eedeb75 3090 if (ret)
a512bbf8 3091 errors++;
a512bbf8
YZ		 3092 }
3093 return errors < i ? 0 : -1;
3094}
3095
387125fc
CM		 3096/*
3097 * endio for the write_dev_flush, this will wake anyone waiting
3098 * for the barrier when it is done
3099 */
3100static void btrfs_end_empty_barrier(struct bio *bio, int err)
3101{
3102 if (err) {
3103 if (err == -EOPNOTSUPP)
3104 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3105 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3106 }
3107 if (bio->bi_private)
3108 complete(bio->bi_private);
3109 bio_put(bio);
3110}
3111
3112/*
3113 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3114 * sent down. With wait == 1, it waits for the previous flush.
3115 *
3116 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3117 * capable
3118 */
3119static int write_dev_flush(struct btrfs_device *device, int wait)
3120{
3121 struct bio *bio;
3122 int ret = 0;
3123
3124 if (device->nobarriers)
3125 return 0;
3126
3127 if (wait) {
3128 bio = device->flush_bio;
3129 if (!bio)
3130 return 0;
3131
3132 wait_for_completion(&device->flush_wait);
3133
3134 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
606686ee
JB		 3135 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3136 rcu_str_deref(device->name));
387125fc 3137 device->nobarriers = 1;
5af3e8cc 3138 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3139 ret = -EIO;
5af3e8cc
SB		 3140 btrfs_dev_stat_inc_and_print(device,
3141 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3142 }
3143
3144 /* drop the reference from the wait == 0 run */
3145 bio_put(bio);
3146 device->flush_bio = NULL;
3147
3148 return ret;
3149 }
3150
3151 /*
3152 * one reference for us, and we leave it for the
3153 * caller
3154 */
9c017abc 3155 device->flush_bio = NULL;
9be3395b 3156 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
387125fc
CM		 3157 if (!bio)
3158 return -ENOMEM;
3159
3160 bio->bi_end_io = btrfs_end_empty_barrier;
3161 bio->bi_bdev = device->bdev;
3162 init_completion(&device->flush_wait);
3163 bio->bi_private = &device->flush_wait;
3164 device->flush_bio = bio;
3165
3166 bio_get(bio);
21adbd5c 3167 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3168
3169 return 0;
3170}
3171
3172/*
3173 * send an empty flush down to each device in parallel,
3174 * then wait for them
3175 */
3176static int barrier_all_devices(struct btrfs_fs_info *info)
3177{
3178 struct list_head *head;
3179 struct btrfs_device *dev;
5af3e8cc
SB		 3180 int errors_send = 0;
3181 int errors_wait = 0;
387125fc
CM		 3182 int ret;
3183
3184 /* send down all the barriers */
3185 head = &info->fs_devices->devices;
3186 list_for_each_entry_rcu(dev, head, dev_list) {
3187 if (!dev->bdev) {
5af3e8cc 3188 errors_send++;
387125fc
CM		 3189 continue;
3190 }
3191 if (!dev->in_fs_metadata || !dev->writeable)
3192 continue;
3193
3194 ret = write_dev_flush(dev, 0);
3195 if (ret)
5af3e8cc 3196 errors_send++;
387125fc
CM		 3197 }
3198
3199 /* wait for all the barriers */
3200 list_for_each_entry_rcu(dev, head, dev_list) {
3201 if (!dev->bdev) {
5af3e8cc 3202 errors_wait++;
387125fc
CM		 3203 continue;
3204 }
3205 if (!dev->in_fs_metadata || !dev->writeable)
3206 continue;
3207
3208 ret = write_dev_flush(dev, 1);
3209 if (ret)
5af3e8cc 3210 errors_wait++;
387125fc 3211 }
5af3e8cc
SB		 3212 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3213 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3214 return -EIO;
3215 return 0;
3216}
3217
5af3e8cc
SB		 3218int btrfs_calc_num_tolerated_disk_barrier_failures(
3219 struct btrfs_fs_info *fs_info)
3220{
3221 struct btrfs_ioctl_space_info space;
3222 struct btrfs_space_info *sinfo;
3223 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3224 BTRFS_BLOCK_GROUP_SYSTEM,
3225 BTRFS_BLOCK_GROUP_METADATA,
3226 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3227 int num_types = 4;
3228 int i;
3229 int c;
3230 int num_tolerated_disk_barrier_failures =
3231 (int)fs_info->fs_devices->num_devices;
3232
3233 for (i = 0; i < num_types; i++) {
3234 struct btrfs_space_info *tmp;
3235
3236 sinfo = NULL;
3237 rcu_read_lock();
3238 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3239 if (tmp->flags == types[i]) {
3240 sinfo = tmp;
3241 break;
3242 }
3243 }
3244 rcu_read_unlock();
3245
3246 if (!sinfo)
3247 continue;
3248
3249 down_read(&sinfo->groups_sem);
3250 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3251 if (!list_empty(&sinfo->block_groups[c])) {
3252 u64 flags;
3253
3254 btrfs_get_block_group_info(
3255 &sinfo->block_groups[c], &space);
3256 if (space.total_bytes == 0 ||
3257 space.used_bytes == 0)
3258 continue;
3259 flags = space.flags;
3260 /*
3261 * return
3262 * 0: if dup, single or RAID0 is configured for
3263 * any of metadata, system or data, else
3264 * 1: if RAID5 is configured, or if RAID1 or
3265 * RAID10 is configured and only two mirrors
3266 * are used, else
3267 * 2: if RAID6 is configured, else
3268 * num_mirrors - 1: if RAID1 or RAID10 is
3269 * configured and more than
3270 * 2 mirrors are used.
3271 */
3272 if (num_tolerated_disk_barrier_failures > 0 &&
3273 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3274 BTRFS_BLOCK_GROUP_RAID0)) ||
3275 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3276 == 0)))
3277 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3278 else if (num_tolerated_disk_barrier_failures > 1) {
3279 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3280 BTRFS_BLOCK_GROUP_RAID5 |
3281 BTRFS_BLOCK_GROUP_RAID10)) {
3282 num_tolerated_disk_barrier_failures = 1;
3283 } else if (flags &
15b0a89d 3284 BTRFS_BLOCK_GROUP_RAID6) {
53b381b3
DW		 3285 num_tolerated_disk_barrier_failures = 2;
3286 }
3287 }
5af3e8cc
SB		 3288 }
3289 }
3290 up_read(&sinfo->groups_sem);
3291 }
3292
3293 return num_tolerated_disk_barrier_failures;
3294}
3295
48a3b636 3296static int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3297{
e5e9a520 3298 struct list_head *head;
f2984462 3299 struct btrfs_device *dev;
a061fc8d 3300 struct btrfs_super_block *sb;
f2984462 3301 struct btrfs_dev_item *dev_item;
f2984462
CM		 3302 int ret;
3303 int do_barriers;
a236aed1
CM		 3304 int max_errors;
3305 int total_errors = 0;
a061fc8d 3306 u64 flags;
f2984462 3307
6c41761f 3308 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
f2984462 3309 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3310 backup_super_roots(root->fs_info);
f2984462 3311
6c41761f 3312 sb = root->fs_info->super_for_commit;
a061fc8d 3313 dev_item = &sb->dev_item;
e5e9a520 3314
174ba509 3315 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3316 head = &root->fs_info->fs_devices->devices;
387125fc 3317
5af3e8cc
SB		 3318 if (do_barriers) {
3319 ret = barrier_all_devices(root->fs_info);
3320 if (ret) {
3321 mutex_unlock(
3322 &root->fs_info->fs_devices->device_list_mutex);
3323 btrfs_error(root->fs_info, ret,
3324 "errors while submitting device barriers.");
3325 return ret;
3326 }
3327 }
387125fc 3328
1f78160c 3329 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3330 if (!dev->bdev) {
3331 total_errors++;
3332 continue;
3333 }
2b82032c 3334 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3335 continue;
3336
2b82032c 3337 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3338 btrfs_set_stack_device_type(dev_item, dev->type);
3339 btrfs_set_stack_device_id(dev_item, dev->devid);
3340 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3341 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3342 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3343 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3344 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3345 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3346 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3347
a061fc8d
CM		 3348 flags = btrfs_super_flags(sb);
3349 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3350
a512bbf8 3351 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3352 if (ret)
3353 total_errors++;
f2984462 3354 }
a236aed1 3355 if (total_errors > max_errors) {
d397712b
CM		 3356 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3357 total_errors);
79787eaa
JM		 3358
3359 /* This shouldn't happen. FUA is masked off if unsupported */
a236aed1
CM		 3360 BUG();
3361 }
f2984462 3362
a512bbf8 3363 total_errors = 0;
1f78160c 3364 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3365 if (!dev->bdev)
3366 continue;
2b82032c 3367 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3368 continue;
3369
a512bbf8
YZ		 3370 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3371 if (ret)
3372 total_errors++;
f2984462 3373 }
174ba509 3374 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3375 if (total_errors > max_errors) {
79787eaa
JM		 3376 btrfs_error(root->fs_info, -EIO,
3377 "%d errors while writing supers", total_errors);
3378 return -EIO;
a236aed1 3379 }
f2984462
CM		 3380 return 0;
3381}
3382
a512bbf8
YZ		 3383int write_ctree_super(struct btrfs_trans_handle *trans,
3384 struct btrfs_root *root, int max_mirrors)
eb60ceac 3385{
e66f709b 3386 int ret;
5f39d397 3387
a512bbf8 3388 ret = write_all_supers(root, max_mirrors);
5f39d397 3389 return ret;
cfaa7295
CM		 3390}
3391
cb517eab
MX		 3392/* Drop a fs root from the radix tree and free it. */
3393void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3394 struct btrfs_root *root)
2619ba1f 3395{
4df27c4d 3396 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3397 radix_tree_delete(&fs_info->fs_roots_radix,
3398 (unsigned long)root->root_key.objectid);
4df27c4d 3399 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3400
3401 if (btrfs_root_refs(&root->root_item) == 0)
3402 synchronize_srcu(&fs_info->subvol_srcu);
3403
d7634482 3404 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3321719e
LB		 3405 btrfs_free_log(NULL, root);
3406 btrfs_free_log_root_tree(NULL, fs_info);
3407 }
3408
581bb050
LZ		 3409 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3410 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3411 free_fs_root(root);
4df27c4d
YZ		 3412}
3413
3414static void free_fs_root(struct btrfs_root *root)
3415{
82d5902d 3416 iput(root->cache_inode);
4df27c4d 3417 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
0ee5dc67
AV		 3418 if (root->anon_dev)
3419 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 3420 free_extent_buffer(root->node);
3421 free_extent_buffer(root->commit_root);
581bb050
LZ		 3422 kfree(root->free_ino_ctl);
3423 kfree(root->free_ino_pinned);
d397712b 3424 kfree(root->name);
b0feb9d9 3425 btrfs_put_fs_root(root);
2619ba1f
CM		 3426}
3427
cb517eab
MX		 3428void btrfs_free_fs_root(struct btrfs_root *root)
3429{
3430 free_fs_root(root);
3431}
3432
c146afad 3433int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3434{
c146afad
YZ		 3435 u64 root_objectid = 0;
3436 struct btrfs_root *gang[8];
3437 int i;
3768f368 3438 int ret;
e089f05c 3439
c146afad
YZ		 3440 while (1) {
3441 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3442 (void **)gang, root_objectid,
3443 ARRAY_SIZE(gang));
3444 if (!ret)
3445 break;
5d4f98a2
YZ		 3446
3447 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 3448 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 3449 int err;
3450
c146afad 3451 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3452 err = btrfs_orphan_cleanup(gang[i]);
3453 if (err)
3454 return err;
c146afad
YZ		 3455 }
3456 root_objectid++;
3457 }
3458 return 0;
3459}
a2135011 3460
c146afad
YZ		 3461int btrfs_commit_super(struct btrfs_root *root)
3462{
3463 struct btrfs_trans_handle *trans;
3464 int ret;
a74a4b97 3465
c146afad 3466 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3467 btrfs_run_delayed_iputs(root);
c146afad 3468 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3469 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3470
3471 /* wait until ongoing cleanup work done */
3472 down_write(&root->fs_info->cleanup_work_sem);
3473 up_write(&root->fs_info->cleanup_work_sem);
3474
7a7eaa40 3475 trans = btrfs_join_transaction(root);
3612b495
TI		 3476 if (IS_ERR(trans))
3477 return PTR_ERR(trans);
54aa1f4d 3478 ret = btrfs_commit_transaction(trans, root);
79787eaa
JM		 3479 if (ret)
3480 return ret;
c146afad 3481 /* run commit again to drop the original snapshot */
7a7eaa40 3482 trans = btrfs_join_transaction(root);
3612b495
TI		 3483 if (IS_ERR(trans))
3484 return PTR_ERR(trans);
79787eaa
JM		 3485 ret = btrfs_commit_transaction(trans, root);
3486 if (ret)
3487 return ret;
79154b1b 3488 ret = btrfs_write_and_wait_transaction(NULL, root);
79787eaa
JM		 3489 if (ret) {
3490 btrfs_error(root->fs_info, ret,
3491 "Failed to sync btree inode to disk.");
3492 return ret;
3493 }
d6bfde87 3494
a512bbf8 3495 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 3496 return ret;
3497}
3498
3499int close_ctree(struct btrfs_root *root)
3500{
3501 struct btrfs_fs_info *fs_info = root->fs_info;
3502 int ret;
3503
3504 fs_info->closing = 1;
3505 smp_mb();
3506
837d5b6e 3507 /* pause restriper - we want to resume on mount */
aa1b8cd4 3508 btrfs_pause_balance(fs_info);
837d5b6e 3509
8dabb742
SB		 3510 btrfs_dev_replace_suspend_for_unmount(fs_info);
3511
aa1b8cd4 3512 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3513
3514 /* wait for any defraggers to finish */
3515 wait_event(fs_info->transaction_wait,
3516 (atomic_read(&fs_info->defrag_running) == 0));
3517
3518 /* clear out the rbtree of defraggable inodes */
26176e7c 3519 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3520
c146afad 3521 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3522 ret = btrfs_commit_super(root);
3523 if (ret)
3524 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3525 }
3526
87533c47 3527 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3528 btrfs_error_commit_super(root);
0f7d52f4 3529
300e4f8a
JB		 3530 btrfs_put_block_group_cache(fs_info);
3531
e3029d9f
AV		 3532 kthread_stop(fs_info->transaction_kthread);
3533 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3534
f25784b3
YZ		 3535 fs_info->closing = 2;
3536 smp_mb();
3537
bcef60f2
AJ		 3538 btrfs_free_qgroup_config(root->fs_info);
3539
963d678b
MX		 3540 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3541 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3542 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3543 }
bcc63abb 3544
e3029d9f 3545 btrfs_free_block_groups(fs_info);
d10c5f31 3546
13e6c37b 3547 btrfs_stop_all_workers(fs_info);
2932505a 3548
c146afad 3549 del_fs_roots(fs_info);
d10c5f31 3550
13e6c37b 3551 free_root_pointers(fs_info, 1);
9ad6b7bc 3552
13e6c37b 3553 iput(fs_info->btree_inode);
d6bfde87 3554
21adbd5c
SB		 3555#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3556 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3557 btrfsic_unmount(root, fs_info->fs_devices);
3558#endif
3559
dfe25020 3560 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3561 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3562
e2d84521 3563 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3564 percpu_counter_destroy(&fs_info->delalloc_bytes);
04160088 3565 bdi_destroy(&fs_info->bdi);
76dda93c 3566 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3567
53b381b3
DW		 3568 btrfs_free_stripe_hash_table(fs_info);
3569
eb60ceac
CM		 3570 return 0;
3571}
3572
b9fab919
CM		 3573int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3574 int atomic)
5f39d397 3575{
1259ab75 3576 int ret;
727011e0 3577 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3578
0b32f4bb 3579 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3580 if (!ret)
3581 return ret;
3582
3583 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3584 parent_transid, atomic);
3585 if (ret == -EAGAIN)
3586 return ret;
1259ab75 3587 return !ret;
5f39d397
CM		 3588}
3589
3590int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3591{
0b32f4bb 3592 return set_extent_buffer_uptodate(buf);
5f39d397 3593}
6702ed49 3594
5f39d397
CM		 3595void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3596{
727011e0 3597 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
5f39d397 3598 u64 transid = btrfs_header_generation(buf);
b9473439 3599 int was_dirty;
b4ce94de 3600
b9447ef8 3601 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3602 if (transid != root->fs_info->generation)
3603 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3604 "found %llu running %llu\n",
db94535d 3605 (unsigned long long)buf->start,
d397712b
CM		 3606 (unsigned long long)transid,
3607 (unsigned long long)root->fs_info->generation);
0b32f4bb 3608 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3609 if (!was_dirty)
3610 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3611 buf->len,
3612 root->fs_info->dirty_metadata_batch);
eb60ceac
CM		 3613}
3614
b53d3f5d
LB		 3615static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3616 int flush_delayed)
16cdcec7
MX		 3617{
3618 /*
3619 * looks as though older kernels can get into trouble with
3620 * this code, they end up stuck in balance_dirty_pages forever
3621 */
e2d84521 3622 int ret;
16cdcec7
MX		 3623
3624 if (current->flags & PF_MEMALLOC)
3625 return;
3626
b53d3f5d
LB		 3627 if (flush_delayed)
3628 btrfs_balance_delayed_items(root);
16cdcec7 3629
e2d84521
MX		 3630 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3631 BTRFS_DIRTY_METADATA_THRESH);
3632 if (ret > 0) {
d0e1d66b
NJ		 3633 balance_dirty_pages_ratelimited(
3634 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3635 }
3636 return;
3637}
3638
b53d3f5d 3639void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3640{
b53d3f5d
LB		 3641 __btrfs_btree_balance_dirty(root, 1);
3642}
585ad2c3 3643
b53d3f5d
LB		 3644void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3645{
3646 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3647}
6b80053d 3648
ca7a79ad 3649int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3650{
727011e0 3651 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3652 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3653}
0da5468f 3654
fcd1f065 3655static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3656 int read_only)
3657{
1104a885
DS		 3658 /*
3659 * Placeholder for checks
3660 */
fcd1f065 3661 return 0;
acce952b 3662}
3663
48a3b636 3664static void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3665{
acce952b 3666 mutex_lock(&root->fs_info->cleaner_mutex);
3667 btrfs_run_delayed_iputs(root);
3668 mutex_unlock(&root->fs_info->cleaner_mutex);
3669
3670 down_write(&root->fs_info->cleanup_work_sem);
3671 up_write(&root->fs_info->cleanup_work_sem);
3672
3673 /* cleanup FS via transaction */
3674 btrfs_cleanup_transaction(root);
acce952b 3675}
3676
569e0f35
JB		 3677static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3678 struct btrfs_root *root)
acce952b 3679{
3680 struct btrfs_inode *btrfs_inode;
3681 struct list_head splice;
3682
3683 INIT_LIST_HEAD(&splice);
3684
3685 mutex_lock(&root->fs_info->ordered_operations_mutex);
3686 spin_lock(&root->fs_info->ordered_extent_lock);
3687
569e0f35 3688 list_splice_init(&t->ordered_operations, &splice);
acce952b 3689 while (!list_empty(&splice)) {
3690 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3691 ordered_operations);
3692
3693 list_del_init(&btrfs_inode->ordered_operations);
b216cbfb 3694 spin_unlock(&root->fs_info->ordered_extent_lock);
acce952b 3695
3696 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb
MX		 3697
3698 spin_lock(&root->fs_info->ordered_extent_lock);
acce952b 3699 }
3700
3701 spin_unlock(&root->fs_info->ordered_extent_lock);
3702 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3703}
3704
143bede5 3705static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3706{
acce952b 3707 struct btrfs_ordered_extent *ordered;
acce952b 3708
3709 spin_lock(&root->fs_info->ordered_extent_lock);
779880ef
JB		 3710 /*
3711 * This will just short circuit the ordered completion stuff which will
3712 * make sure the ordered extent gets properly cleaned up.
3713 */
3714 list_for_each_entry(ordered, &root->fs_info->ordered_extents,
3715 root_extent_list)
3716 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
acce952b 3717 spin_unlock(&root->fs_info->ordered_extent_lock);
acce952b 3718}
3719
79787eaa
JM		 3720int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3721 struct btrfs_root *root)
acce952b 3722{
3723 struct rb_node *node;
3724 struct btrfs_delayed_ref_root *delayed_refs;
3725 struct btrfs_delayed_ref_node *ref;
3726 int ret = 0;
3727
3728 delayed_refs = &trans->delayed_refs;
3729
3730 spin_lock(&delayed_refs->lock);
3731 if (delayed_refs->num_entries == 0) {
cfece4db 3732 spin_unlock(&delayed_refs->lock);
acce952b 3733 printk(KERN_INFO "delayed_refs has NO entry\n");
3734 return ret;
3735 }
3736
b939d1ab 3737 while ((node = rb_first(&delayed_refs->root)) != NULL) {
eb12db69 3738 struct btrfs_delayed_ref_head *head = NULL;
acce952b 3739
eb12db69 3740 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
acce952b 3741 atomic_set(&ref->refs, 1);
3742 if (btrfs_delayed_ref_is_head(ref)) {
acce952b 3743
3744 head = btrfs_delayed_node_to_head(ref);
b939d1ab
JB		 3745 if (!mutex_trylock(&head->mutex)) {
3746 atomic_inc(&ref->refs);
3747 spin_unlock(&delayed_refs->lock);
3748
3749 /* Need to wait for the delayed ref to run */
3750 mutex_lock(&head->mutex);
3751 mutex_unlock(&head->mutex);
3752 btrfs_put_delayed_ref(ref);
3753
e18fca73 3754 spin_lock(&delayed_refs->lock);
b939d1ab
JB		 3755 continue;
3756 }
3757
54067ae9
JB		 3758 if (head->must_insert_reserved)
3759 btrfs_pin_extent(root, ref->bytenr,
3760 ref->num_bytes, 1);
78a6184a 3761 btrfs_free_delayed_extent_op(head->extent_op);
acce952b 3762 delayed_refs->num_heads--;
3763 if (list_empty(&head->cluster))
3764 delayed_refs->num_heads_ready--;
3765 list_del_init(&head->cluster);
acce952b 3766 }
eb12db69 3767
b939d1ab
JB		 3768 ref->in_tree = 0;
3769 rb_erase(&ref->rb_node, &delayed_refs->root);
3770 delayed_refs->num_entries--;
eb12db69
JB		 3771 if (head)
3772 mutex_unlock(&head->mutex);
acce952b 3773 spin_unlock(&delayed_refs->lock);
3774 btrfs_put_delayed_ref(ref);
3775
3776 cond_resched();
3777 spin_lock(&delayed_refs->lock);
3778 }
3779
3780 spin_unlock(&delayed_refs->lock);
3781
3782 return ret;
3783}
3784
aec8030a 3785static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
acce952b 3786{
3787 struct btrfs_pending_snapshot *snapshot;
3788 struct list_head splice;
3789
3790 INIT_LIST_HEAD(&splice);
3791
3792 list_splice_init(&t->pending_snapshots, &splice);
3793
3794 while (!list_empty(&splice)) {
3795 snapshot = list_entry(splice.next,
3796 struct btrfs_pending_snapshot,
3797 list);
aec8030a 3798 snapshot->error = -ECANCELED;
acce952b 3799 list_del_init(&snapshot->list);
acce952b 3800 }
acce952b 3801}
3802
143bede5 3803static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3804{
3805 struct btrfs_inode *btrfs_inode;
3806 struct list_head splice;
3807
3808 INIT_LIST_HEAD(&splice);
3809
eb73c1b7
MX		 3810 spin_lock(&root->delalloc_lock);
3811 list_splice_init(&root->delalloc_inodes, &splice);
acce952b 3812
3813 while (!list_empty(&splice)) {
eb73c1b7
MX		 3814 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3815 delalloc_inodes);
acce952b 3816
3817 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3818 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3819 &btrfs_inode->runtime_flags);
eb73c1b7 3820 spin_unlock(&root->delalloc_lock);
acce952b 3821
3822 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3823
eb73c1b7 3824 spin_lock(&root->delalloc_lock);
acce952b 3825 }
3826
eb73c1b7
MX		 3827 spin_unlock(&root->delalloc_lock);
3828}
3829
3830static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3831{
3832 struct btrfs_root *root;
3833 struct list_head splice;
3834
3835 INIT_LIST_HEAD(&splice);
3836
3837 spin_lock(&fs_info->delalloc_root_lock);
3838 list_splice_init(&fs_info->delalloc_roots, &splice);
3839 while (!list_empty(&splice)) {
3840 root = list_first_entry(&splice, struct btrfs_root,
3841 delalloc_root);
3842 list_del_init(&root->delalloc_root);
3843 root = btrfs_grab_fs_root(root);
3844 BUG_ON(!root);
3845 spin_unlock(&fs_info->delalloc_root_lock);
3846
3847 btrfs_destroy_delalloc_inodes(root);
3848 btrfs_put_fs_root(root);
3849
3850 spin_lock(&fs_info->delalloc_root_lock);
3851 }
3852 spin_unlock(&fs_info->delalloc_root_lock);
acce952b 3853}
3854
3855static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3856 struct extent_io_tree *dirty_pages,
3857 int mark)
3858{
3859 int ret;
acce952b 3860 struct extent_buffer *eb;
3861 u64 start = 0;
3862 u64 end;
acce952b 3863
3864 while (1) {
3865 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3866 mark, NULL);
acce952b 3867 if (ret)
3868 break;
3869
3870 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3871 while (start <= end) {
fd8b2b61
JB		 3872 eb = btrfs_find_tree_block(root, start,
3873 root->leafsize);
69a85bd8 3874 start += root->leafsize;
fd8b2b61 3875 if (!eb)
acce952b 3876 continue;
fd8b2b61 3877 wait_on_extent_buffer_writeback(eb);
acce952b 3878
fd8b2b61
JB		 3879 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3880 &eb->bflags))
3881 clear_extent_buffer_dirty(eb);
3882 free_extent_buffer_stale(eb);
acce952b 3883 }
3884 }
3885
3886 return ret;
3887}
3888
3889static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3890 struct extent_io_tree *pinned_extents)
3891{
3892 struct extent_io_tree *unpin;
3893 u64 start;
3894 u64 end;
3895 int ret;
ed0eaa14 3896 bool loop = true;
acce952b 3897
3898 unpin = pinned_extents;
ed0eaa14 3899again:
acce952b 3900 while (1) {
3901 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 3902 EXTENT_DIRTY, NULL);
acce952b 3903 if (ret)
3904 break;
3905
3906 /* opt_discard */
5378e607
LD		 3907 if (btrfs_test_opt(root, DISCARD))
3908 ret = btrfs_error_discard_extent(root, start,
3909 end + 1 - start,
3910 NULL);
acce952b 3911
3912 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3913 btrfs_error_unpin_extent_range(root, start, end);
3914 cond_resched();
3915 }
3916
ed0eaa14
LB		 3917 if (loop) {
3918 if (unpin == &root->fs_info->freed_extents[0])
3919 unpin = &root->fs_info->freed_extents[1];
3920 else
3921 unpin = &root->fs_info->freed_extents[0];
3922 loop = false;
3923 goto again;
3924 }
3925
acce952b 3926 return 0;
3927}
3928
49b25e05
JM		 3929void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3930 struct btrfs_root *root)
3931{
3932 btrfs_destroy_delayed_refs(cur_trans, root);
3933 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3934 cur_trans->dirty_pages.dirty_bytes);
3935
3936 /* FIXME: cleanup wait for commit */
3937 cur_trans->in_commit = 1;
3938 cur_trans->blocked = 1;
d7096fc3 3939 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 3940
aec8030a
MX		 3941 btrfs_evict_pending_snapshots(cur_trans);
3942
49b25e05 3943 cur_trans->blocked = 0;
d7096fc3 3944 wake_up(&root->fs_info->transaction_wait);
49b25e05
JM		 3945
3946 cur_trans->commit_done = 1;
d7096fc3 3947 wake_up(&cur_trans->commit_wait);
49b25e05 3948
67cde344
MX		 3949 btrfs_destroy_delayed_inodes(root);
3950 btrfs_assert_delayed_root_empty(root);
49b25e05 3951
49b25e05
JM		 3952 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3953 EXTENT_DIRTY);
6e841e32
LB		 3954 btrfs_destroy_pinned_extent(root,
3955 root->fs_info->pinned_extents);
49b25e05
JM		 3956
3957 /*
3958 memset(cur_trans, 0, sizeof(*cur_trans));
3959 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3960 */
3961}
3962
48a3b636 3963static int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 3964{
3965 struct btrfs_transaction *t;
3966 LIST_HEAD(list);
3967
acce952b 3968 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3969
a4abeea4 3970 spin_lock(&root->fs_info->trans_lock);
acce952b 3971 list_splice_init(&root->fs_info->trans_list, &list);
a4abeea4
JB		 3972 root->fs_info->trans_no_join = 1;
3973 spin_unlock(&root->fs_info->trans_lock);
3974
acce952b 3975 while (!list_empty(&list)) {
3976 t = list_entry(list.next, struct btrfs_transaction, list);
acce952b 3977
569e0f35 3978 btrfs_destroy_ordered_operations(t, root);
acce952b 3979
3980 btrfs_destroy_ordered_extents(root);
3981
3982 btrfs_destroy_delayed_refs(t, root);
3983
acce952b 3984 /* FIXME: cleanup wait for commit */
3985 t->in_commit = 1;
3986 t->blocked = 1;
66657b31 3987 smp_mb();
acce952b 3988 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3989 wake_up(&root->fs_info->transaction_blocked_wait);
3990
aec8030a
MX		 3991 btrfs_evict_pending_snapshots(t);
3992
acce952b 3993 t->blocked = 0;
66657b31 3994 smp_mb();
acce952b 3995 if (waitqueue_active(&root->fs_info->transaction_wait))
3996 wake_up(&root->fs_info->transaction_wait);
acce952b 3997
acce952b 3998 t->commit_done = 1;
66657b31 3999 smp_mb();
acce952b 4000 if (waitqueue_active(&t->commit_wait))
4001 wake_up(&t->commit_wait);
acce952b 4002
67cde344
MX		 4003 btrfs_destroy_delayed_inodes(root);
4004 btrfs_assert_delayed_root_empty(root);
4005
eb73c1b7 4006 btrfs_destroy_all_delalloc_inodes(root->fs_info);
acce952b 4007
a4abeea4 4008 spin_lock(&root->fs_info->trans_lock);
acce952b 4009 root->fs_info->running_transaction = NULL;
a4abeea4 4010 spin_unlock(&root->fs_info->trans_lock);
acce952b 4011
4012 btrfs_destroy_marked_extents(root, &t->dirty_pages,
4013 EXTENT_DIRTY);
4014
4015 btrfs_destroy_pinned_extent(root,
4016 root->fs_info->pinned_extents);
4017
13c5a93e 4018 atomic_set(&t->use_count, 0);
acce952b 4019 list_del_init(&t->list);
4020 memset(t, 0, sizeof(*t));
4021 kmem_cache_free(btrfs_transaction_cachep, t);
4022 }
4023
a4abeea4
JB		 4024 spin_lock(&root->fs_info->trans_lock);
4025 root->fs_info->trans_no_join = 0;
4026 spin_unlock(&root->fs_info->trans_lock);
acce952b 4027 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 4028
4029 return 0;
4030}
4031
d1310b2e 4032static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 4033 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 4034 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 4035 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 4036 /* note we're sharing with inode.c for the merge bio hook */
4037 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 4038};
Linux 6.x block layer and io_uring tree(s)