projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: support printing UUID tree elements
[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
facc8a22
MX		 579static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
580 u64 phy_offset, struct page *page,
581 u64 start, u64 end, int mirror)
ce9adaa5
CM		 582{
583 struct extent_io_tree *tree;
584 u64 found_start;
585 int found_level;
ce9adaa5
CM		 586 struct extent_buffer *eb;
587 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 588 int ret = 0;
727011e0 589 int reads_done;
ce9adaa5 590
ce9adaa5
CM		 591 if (!page->private)
592 goto out;
d397712b 593
727011e0 594 tree = &BTRFS_I(page->mapping->host)->io_tree;
4f2de97a 595 eb = (struct extent_buffer *)page->private;
d397712b 596
0b32f4bb
JB		 597 /* the pending IO might have been the only thing that kept this buffer
598 * in memory. Make sure we have a ref for all this other checks
599 */
600 extent_buffer_get(eb);
601
602 reads_done = atomic_dec_and_test(&eb->io_pages);
727011e0
CM		 603 if (!reads_done)
604 goto err;
f188591e 605
5cf1ab56 606 eb->read_mirror = mirror;
ea466794
JB		 607 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
608 ret = -EIO;
609 goto err;
610 }
611
ce9adaa5 612 found_start = btrfs_header_bytenr(eb);
727011e0 613 if (found_start != eb->start) {
7a36ddec 614 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
193f284d
CM		 615 "%llu %llu\n",
616 (unsigned long long)found_start,
617 (unsigned long long)eb->start);
f188591e 618 ret = -EIO;
ce9adaa5
CM		 619 goto err;
620 }
2b82032c 621 if (check_tree_block_fsid(root, eb)) {
7a36ddec 622 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
193f284d 623 (unsigned long long)eb->start);
1259ab75
CM		 624 ret = -EIO;
625 goto err;
626 }
ce9adaa5 627 found_level = btrfs_header_level(eb);
1c24c3ce
JB		 628 if (found_level >= BTRFS_MAX_LEVEL) {
629 btrfs_info(root->fs_info, "bad tree block level %d\n",
630 (int)btrfs_header_level(eb));
631 ret = -EIO;
632 goto err;
633 }
ce9adaa5 634
85d4e461
CM		 635 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
636 eb, found_level);
4008c04a 637
ce9adaa5 638 ret = csum_tree_block(root, eb, 1);
a826d6dc 639 if (ret) {
f188591e 640 ret = -EIO;
a826d6dc
JB		 641 goto err;
642 }
643
644 /*
645 * If this is a leaf block and it is corrupt, set the corrupt bit so
646 * that we don't try and read the other copies of this block, just
647 * return -EIO.
648 */
649 if (found_level == 0 && check_leaf(root, eb)) {
650 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
651 ret = -EIO;
652 }
ce9adaa5 653
0b32f4bb
JB		 654 if (!ret)
655 set_extent_buffer_uptodate(eb);
ce9adaa5 656err:
79fb65a1
JB		 657 if (reads_done &&
658 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 659 btree_readahead_hook(root, eb, eb->start, ret);
4bb31e92 660
53b381b3
DW		 661 if (ret) {
662 /*
663 * our io error hook is going to dec the io pages
664 * again, we have to make sure it has something
665 * to decrement
666 */
667 atomic_inc(&eb->io_pages);
0b32f4bb 668 clear_extent_buffer_uptodate(eb);
53b381b3 669 }
0b32f4bb 670 free_extent_buffer(eb);
ce9adaa5 671out:
f188591e 672 return ret;
ce9adaa5
CM		 673}
674
ea466794 675static int btree_io_failed_hook(struct page *page, int failed_mirror)
4bb31e92 676{
4bb31e92
AJ		 677 struct extent_buffer *eb;
678 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
679
4f2de97a 680 eb = (struct extent_buffer *)page->private;
ea466794 681 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
5cf1ab56 682 eb->read_mirror = failed_mirror;
53b381b3 683 atomic_dec(&eb->io_pages);
ea466794 684 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 685 btree_readahead_hook(root, eb, eb->start, -EIO);
4bb31e92
AJ		 686 return -EIO; /* we fixed nothing */
687}
688
ce9adaa5 689static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 690{
691 struct end_io_wq *end_io_wq = bio->bi_private;
692 struct btrfs_fs_info *fs_info;
ce9adaa5 693
ce9adaa5 694 fs_info = end_io_wq->info;
ce9adaa5 695 end_io_wq->error = err;
8b712842
CM		 696 end_io_wq->work.func = end_workqueue_fn;
697 end_io_wq->work.flags = 0;
d20f7043 698
7b6d91da 699 if (bio->bi_rw & REQ_WRITE) {
53b381b3 700 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
cad321ad
CM		 701 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
702 &end_io_wq->work);
53b381b3 703 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
0cb59c99
JB		 704 btrfs_queue_worker(&fs_info->endio_freespace_worker,
705 &end_io_wq->work);
53b381b3
DW		 706 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
707 btrfs_queue_worker(&fs_info->endio_raid56_workers,
708 &end_io_wq->work);
cad321ad
CM		 709 else
710 btrfs_queue_worker(&fs_info->endio_write_workers,
711 &end_io_wq->work);
d20f7043 712 } else {
53b381b3
DW		 713 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
714 btrfs_queue_worker(&fs_info->endio_raid56_workers,
715 &end_io_wq->work);
716 else if (end_io_wq->metadata)
d20f7043
CM		 717 btrfs_queue_worker(&fs_info->endio_meta_workers,
718 &end_io_wq->work);
719 else
720 btrfs_queue_worker(&fs_info->endio_workers,
721 &end_io_wq->work);
722 }
ce9adaa5
CM		 723}
724
0cb59c99
JB		 725/*
726 * For the metadata arg you want
727 *
728 * 0 - if data
729 * 1 - if normal metadta
730 * 2 - if writing to the free space cache area
53b381b3 731 * 3 - raid parity work
0cb59c99 732 */
22c59948
CM		 733int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
734 int metadata)
0b86a832 735{
ce9adaa5 736 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 737 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
738 if (!end_io_wq)
739 return -ENOMEM;
740
741 end_io_wq->private = bio->bi_private;
742 end_io_wq->end_io = bio->bi_end_io;
22c59948 743 end_io_wq->info = info;
ce9adaa5
CM		 744 end_io_wq->error = 0;
745 end_io_wq->bio = bio;
22c59948 746 end_io_wq->metadata = metadata;
ce9adaa5
CM		 747
748 bio->bi_private = end_io_wq;
749 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 750 return 0;
751}
752
b64a2851 753unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 754{
4854ddd0
CM		 755 unsigned long limit = min_t(unsigned long,
756 info->workers.max_workers,
757 info->fs_devices->open_devices);
758 return 256 * limit;
759}
0986fe9e 760
4a69a410
CM		 761static void run_one_async_start(struct btrfs_work *work)
762{
4a69a410 763 struct async_submit_bio *async;
79787eaa 764 int ret;
4a69a410
CM		 765
766 async = container_of(work, struct async_submit_bio, work);
79787eaa
JM		 767 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
768 async->mirror_num, async->bio_flags,
769 async->bio_offset);
770 if (ret)
771 async->error = ret;
4a69a410
CM		 772}
773
774static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 775{
776 struct btrfs_fs_info *fs_info;
777 struct async_submit_bio *async;
4854ddd0 778 int limit;
8b712842
CM		 779
780 async = container_of(work, struct async_submit_bio, work);
781 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 782
b64a2851 783 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 784 limit = limit * 2 / 3;
785
66657b31 786 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
b64a2851 787 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 788 wake_up(&fs_info->async_submit_wait);
789
79787eaa
JM		 790 /* If an error occured we just want to clean up the bio and move on */
791 if (async->error) {
792 bio_endio(async->bio, async->error);
793 return;
794 }
795
4a69a410 796 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 797 async->mirror_num, async->bio_flags,
798 async->bio_offset);
4a69a410
CM		 799}
800
801static void run_one_async_free(struct btrfs_work *work)
802{
803 struct async_submit_bio *async;
804
805 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 806 kfree(async);
807}
808
44b8bd7e
CM		 809int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
810 int rw, struct bio *bio, int mirror_num,
c8b97818 811 unsigned long bio_flags,
eaf25d93 812 u64 bio_offset,
4a69a410
CM		 813 extent_submit_bio_hook_t *submit_bio_start,
814 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 815{
816 struct async_submit_bio *async;
817
818 async = kmalloc(sizeof(*async), GFP_NOFS);
819 if (!async)
820 return -ENOMEM;
821
822 async->inode = inode;
823 async->rw = rw;
824 async->bio = bio;
825 async->mirror_num = mirror_num;
4a69a410
CM		 826 async->submit_bio_start = submit_bio_start;
827 async->submit_bio_done = submit_bio_done;
828
829 async->work.func = run_one_async_start;
830 async->work.ordered_func = run_one_async_done;
831 async->work.ordered_free = run_one_async_free;
832
8b712842 833 async->work.flags = 0;
c8b97818 834 async->bio_flags = bio_flags;
eaf25d93 835 async->bio_offset = bio_offset;
8c8bee1d 836
79787eaa
JM		 837 async->error = 0;
838
cb03c743 839 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 840
7b6d91da 841 if (rw & REQ_SYNC)
d313d7a3
CM		 842 btrfs_set_work_high_prio(&async->work);
843
8b712842 844 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 845
d397712b 846 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 847 atomic_read(&fs_info->nr_async_submits)) {
848 wait_event(fs_info->async_submit_wait,
849 (atomic_read(&fs_info->nr_async_submits) == 0));
850 }
851
44b8bd7e
CM		 852 return 0;
853}
854
ce3ed71a
CM		 855static int btree_csum_one_bio(struct bio *bio)
856{
857 struct bio_vec *bvec = bio->bi_io_vec;
858 int bio_index = 0;
859 struct btrfs_root *root;
79787eaa 860 int ret = 0;
ce3ed71a
CM		 861
862 WARN_ON(bio->bi_vcnt <= 0);
d397712b 863 while (bio_index < bio->bi_vcnt) {
ce3ed71a 864 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 865 ret = csum_dirty_buffer(root, bvec->bv_page);
866 if (ret)
867 break;
ce3ed71a
CM		 868 bio_index++;
869 bvec++;
870 }
79787eaa 871 return ret;
ce3ed71a
CM		 872}
873
4a69a410
CM		 874static int __btree_submit_bio_start(struct inode *inode, int rw,
875 struct bio *bio, int mirror_num,
eaf25d93
CM		 876 unsigned long bio_flags,
877 u64 bio_offset)
22c59948 878{
8b712842
CM		 879 /*
880 * when we're called for a write, we're already in the async
5443be45 881 * submission context. Just jump into btrfs_map_bio
8b712842 882 */
79787eaa 883 return btree_csum_one_bio(bio);
4a69a410 884}
22c59948 885
4a69a410 886static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 887 int mirror_num, unsigned long bio_flags,
888 u64 bio_offset)
4a69a410 889{
61891923
SB		 890 int ret;
891
8b712842 892 /*
4a69a410
CM		 893 * when we're called for a write, we're already in the async
894 * submission context. Just jump into btrfs_map_bio
8b712842 895 */
61891923
SB		 896 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
897 if (ret)
898 bio_endio(bio, ret);
899 return ret;
0b86a832
CM		 900}
901
de0022b9
JB		 902static int check_async_write(struct inode *inode, unsigned long bio_flags)
903{
904 if (bio_flags & EXTENT_BIO_TREE_LOG)
905 return 0;
906#ifdef CONFIG_X86
907 if (cpu_has_xmm4_2)
908 return 0;
909#endif
910 return 1;
911}
912
44b8bd7e 913static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 914 int mirror_num, unsigned long bio_flags,
915 u64 bio_offset)
44b8bd7e 916{
de0022b9 917 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 918 int ret;
919
7b6d91da 920 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 921 /*
922 * called for a read, do the setup so that checksum validation
923 * can happen in the async kernel threads
924 */
f3f266ab
CM		 925 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
926 bio, 1);
1d4284bd 927 if (ret)
61891923
SB		 928 goto out_w_error;
929 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
930 mirror_num, 0);
de0022b9
JB		 931 } else if (!async) {
932 ret = btree_csum_one_bio(bio);
933 if (ret)
61891923
SB		 934 goto out_w_error;
935 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
936 mirror_num, 0);
937 } else {
938 /*
939 * kthread helpers are used to submit writes so that
940 * checksumming can happen in parallel across all CPUs
941 */
942 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
943 inode, rw, bio, mirror_num, 0,
944 bio_offset,
945 __btree_submit_bio_start,
946 __btree_submit_bio_done);
44b8bd7e 947 }
d313d7a3 948
61891923
SB		 949 if (ret) {
950out_w_error:
951 bio_endio(bio, ret);
952 }
953 return ret;
44b8bd7e
CM		 954}
955
3dd1462e 956#ifdef CONFIG_MIGRATION
784b4e29 957static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 958 struct page *newpage, struct page *page,
959 enum migrate_mode mode)
784b4e29
CM		 960{
961 /*
962 * we can't safely write a btree page from here,
963 * we haven't done the locking hook
964 */
965 if (PageDirty(page))
966 return -EAGAIN;
967 /*
968 * Buffers may be managed in a filesystem specific way.
969 * We must have no buffers or drop them.
970 */
971 if (page_has_private(page) &&
972 !try_to_release_page(page, GFP_KERNEL))
973 return -EAGAIN;
a6bc32b8 974 return migrate_page(mapping, newpage, page, mode);
784b4e29 975}
3dd1462e 976#endif
784b4e29 977
0da5468f
CM		 978
979static int btree_writepages(struct address_space *mapping,
980 struct writeback_control *wbc)
981{
d1310b2e 982 struct extent_io_tree *tree;
e2d84521
MX		 983 struct btrfs_fs_info *fs_info;
984 int ret;
985
d1310b2e 986 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 987 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 988
989 if (wbc->for_kupdate)
990 return 0;
991
e2d84521 992 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 993 /* this is a bit racy, but that's ok */
e2d84521
MX		 994 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
995 BTRFS_DIRTY_METADATA_THRESH);
996 if (ret < 0)
793955bc 997 return 0;
793955bc 998 }
0b32f4bb 999 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 1000}
1001
b2950863 1002static int btree_readpage(struct file *file, struct page *page)
5f39d397 1003{
d1310b2e
CM		 1004 struct extent_io_tree *tree;
1005 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 1006 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 1007}
22b0ebda 1008
70dec807 1009static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 1010{
98509cfc 1011 if (PageWriteback(page) || PageDirty(page))
d397712b 1012 return 0;
0c4e538b 1013
f7a52a40 1014 return try_release_extent_buffer(page);
d98237b3
CM		 1015}
1016
d47992f8
LC		 1017static void btree_invalidatepage(struct page *page, unsigned int offset,
1018 unsigned int length)
d98237b3 1019{
d1310b2e
CM		 1020 struct extent_io_tree *tree;
1021 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1022 extent_invalidatepage(tree, page, offset);
1023 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1024 if (PagePrivate(page)) {
d397712b
CM		 1025 printk(KERN_WARNING "btrfs warning page private not zero "
1026 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1027 ClearPagePrivate(page);
1028 set_page_private(page, 0);
1029 page_cache_release(page);
1030 }
d98237b3
CM		 1031}
1032
0b32f4bb
JB		 1033static int btree_set_page_dirty(struct page *page)
1034{
bb146eb2 1035#ifdef DEBUG
0b32f4bb
JB		 1036 struct extent_buffer *eb;
1037
1038 BUG_ON(!PagePrivate(page));
1039 eb = (struct extent_buffer *)page->private;
1040 BUG_ON(!eb);
1041 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1042 BUG_ON(!atomic_read(&eb->refs));
1043 btrfs_assert_tree_locked(eb);
bb146eb2 1044#endif
0b32f4bb
JB		 1045 return __set_page_dirty_nobuffers(page);
1046}
1047
7f09410b 1048static const struct address_space_operations btree_aops = {
d98237b3 1049 .readpage = btree_readpage,
0da5468f 1050 .writepages = btree_writepages,
5f39d397
CM		 1051 .releasepage = btree_releasepage,
1052 .invalidatepage = btree_invalidatepage,
5a92bc88 1053#ifdef CONFIG_MIGRATION
784b4e29 1054 .migratepage = btree_migratepage,
5a92bc88 1055#endif
0b32f4bb 1056 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1057};
1058
ca7a79ad
CM		 1059int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1060 u64 parent_transid)
090d1875 1061{
5f39d397
CM		 1062 struct extent_buffer *buf = NULL;
1063 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1064 int ret = 0;
090d1875 1065
db94535d 1066 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1067 if (!buf)
090d1875 1068 return 0;
d1310b2e 1069 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1070 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1071 free_extent_buffer(buf);
de428b63 1072 return ret;
090d1875
CM		 1073}
1074
ab0fff03
AJ		 1075int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1076 int mirror_num, struct extent_buffer **eb)
1077{
1078 struct extent_buffer *buf = NULL;
1079 struct inode *btree_inode = root->fs_info->btree_inode;
1080 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1081 int ret;
1082
1083 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1084 if (!buf)
1085 return 0;
1086
1087 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1088
1089 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1090 btree_get_extent, mirror_num);
1091 if (ret) {
1092 free_extent_buffer(buf);
1093 return ret;
1094 }
1095
1096 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1097 free_extent_buffer(buf);
1098 return -EIO;
0b32f4bb 1099 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1100 *eb = buf;
1101 } else {
1102 free_extent_buffer(buf);
1103 }
1104 return 0;
1105}
1106
0999df54
CM		 1107struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1108 u64 bytenr, u32 blocksize)
1109{
1110 struct inode *btree_inode = root->fs_info->btree_inode;
1111 struct extent_buffer *eb;
1112 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
f09d1f60 1113 bytenr, blocksize);
0999df54
CM		 1114 return eb;
1115}
1116
1117struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1118 u64 bytenr, u32 blocksize)
1119{
1120 struct inode *btree_inode = root->fs_info->btree_inode;
1121 struct extent_buffer *eb;
1122
1123 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
727011e0 1124 bytenr, blocksize);
0999df54
CM		 1125 return eb;
1126}
1127
1128
e02119d5
CM		 1129int btrfs_write_tree_block(struct extent_buffer *buf)
1130{
727011e0 1131 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
8aa38c31 1132 buf->start + buf->len - 1);
e02119d5
CM		 1133}
1134
1135int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1136{
727011e0 1137 return filemap_fdatawait_range(buf->pages[0]->mapping,
8aa38c31 1138 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1139}
1140
0999df54 1141struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1142 u32 blocksize, u64 parent_transid)
0999df54
CM		 1143{
1144 struct extent_buffer *buf = NULL;
0999df54
CM		 1145 int ret;
1146
0999df54
CM		 1147 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1148 if (!buf)
1149 return NULL;
0999df54 1150
ca7a79ad 1151 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
0f0fe8f7
FDBM		 1152 if (ret) {
1153 free_extent_buffer(buf);
1154 return NULL;
1155 }
5f39d397 1156 return buf;
ce9adaa5 1157
eb60ceac
CM		 1158}
1159
d5c13f92
JM		 1160void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1161 struct extent_buffer *buf)
ed2ff2cb 1162{
e2d84521
MX		 1163 struct btrfs_fs_info *fs_info = root->fs_info;
1164
55c69072 1165 if (btrfs_header_generation(buf) ==
e2d84521 1166 fs_info->running_transaction->transid) {
b9447ef8 1167 btrfs_assert_tree_locked(buf);
b4ce94de 1168
b9473439 1169 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1170 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1171 -buf->len,
1172 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1173 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1174 btrfs_set_lock_blocking(buf);
1175 clear_extent_buffer_dirty(buf);
1176 }
925baedd 1177 }
5f39d397
CM		 1178}
1179
143bede5
JM		 1180static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1181 u32 stripesize, struct btrfs_root *root,
1182 struct btrfs_fs_info *fs_info,
1183 u64 objectid)
d97e63b6 1184{
cfaa7295 1185 root->node = NULL;
a28ec197 1186 root->commit_root = NULL;
db94535d
CM		 1187 root->sectorsize = sectorsize;
1188 root->nodesize = nodesize;
1189 root->leafsize = leafsize;
87ee04eb 1190 root->stripesize = stripesize;
123abc88 1191 root->ref_cows = 0;
0b86a832 1192 root->track_dirty = 0;
c71bf099 1193 root->in_radix = 0;
d68fc57b
YZ		 1194 root->orphan_item_inserted = 0;
1195 root->orphan_cleanup_state = 0;
0b86a832 1196
0f7d52f4
CM		 1197 root->objectid = objectid;
1198 root->last_trans = 0;
13a8a7c8 1199 root->highest_objectid = 0;
eb73c1b7 1200 root->nr_delalloc_inodes = 0;
199c2a9c 1201 root->nr_ordered_extents = 0;
58176a96 1202 root->name = NULL;
6bef4d31 1203 root->inode_tree = RB_ROOT;
16cdcec7 1204 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1205 root->block_rsv = NULL;
d68fc57b 1206 root->orphan_block_rsv = NULL;
0b86a832
CM		 1207
1208 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1209 INIT_LIST_HEAD(&root->root_list);
eb73c1b7
MX		 1210 INIT_LIST_HEAD(&root->delalloc_inodes);
1211 INIT_LIST_HEAD(&root->delalloc_root);
199c2a9c
MX		 1212 INIT_LIST_HEAD(&root->ordered_extents);
1213 INIT_LIST_HEAD(&root->ordered_root);
2ab28f32
JB		 1214 INIT_LIST_HEAD(&root->logged_list[0]);
1215 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1216 spin_lock_init(&root->orphan_lock);
5d4f98a2 1217 spin_lock_init(&root->inode_lock);
eb73c1b7 1218 spin_lock_init(&root->delalloc_lock);
199c2a9c 1219 spin_lock_init(&root->ordered_extent_lock);
f0486c68 1220 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1221 spin_lock_init(&root->log_extents_lock[0]);
1222 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1223 mutex_init(&root->objectid_mutex);
e02119d5 1224 mutex_init(&root->log_mutex);
7237f183
YZ		 1225 init_waitqueue_head(&root->log_writer_wait);
1226 init_waitqueue_head(&root->log_commit_wait[0]);
1227 init_waitqueue_head(&root->log_commit_wait[1]);
1228 atomic_set(&root->log_commit[0], 0);
1229 atomic_set(&root->log_commit[1], 0);
1230 atomic_set(&root->log_writers, 0);
2ecb7923 1231 atomic_set(&root->log_batch, 0);
8a35d95f 1232 atomic_set(&root->orphan_inodes, 0);
b0feb9d9 1233 atomic_set(&root->refs, 1);
7237f183 1234 root->log_transid = 0;
257c62e1 1235 root->last_log_commit = 0;
d0c803c4 1236 extent_io_tree_init(&root->dirty_log_pages,
f993c883 1237 fs_info->btree_inode->i_mapping);
017e5369 1238
3768f368
CM		 1239 memset(&root->root_key, 0, sizeof(root->root_key));
1240 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1241 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1242 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 1243 root->defrag_trans_start = fs_info->generation;
58176a96 1244 init_completion(&root->kobj_unregister);
6702ed49 1245 root->defrag_running = 0;
4d775673 1246 root->root_key.objectid = objectid;
0ee5dc67 1247 root->anon_dev = 0;
8ea05e3a 1248
5f3ab90a 1249 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1250}
1251
f84a8bd6 1252static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1253{
1254 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1255 if (root)
1256 root->fs_info = fs_info;
1257 return root;
1258}
1259
20897f5c
AJ		 1260struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1261 struct btrfs_fs_info *fs_info,
1262 u64 objectid)
1263{
1264 struct extent_buffer *leaf;
1265 struct btrfs_root *tree_root = fs_info->tree_root;
1266 struct btrfs_root *root;
1267 struct btrfs_key key;
1268 int ret = 0;
1269 u64 bytenr;
6463fe58 1270 uuid_le uuid;
20897f5c
AJ		 1271
1272 root = btrfs_alloc_root(fs_info);
1273 if (!root)
1274 return ERR_PTR(-ENOMEM);
1275
1276 __setup_root(tree_root->nodesize, tree_root->leafsize,
1277 tree_root->sectorsize, tree_root->stripesize,
1278 root, fs_info, objectid);
1279 root->root_key.objectid = objectid;
1280 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1281 root->root_key.offset = 0;
1282
1283 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1284 0, objectid, NULL, 0, 0, 0);
1285 if (IS_ERR(leaf)) {
1286 ret = PTR_ERR(leaf);
1dd05682 1287 leaf = NULL;
20897f5c
AJ		 1288 goto fail;
1289 }
1290
1291 bytenr = leaf->start;
1292 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1293 btrfs_set_header_bytenr(leaf, leaf->start);
1294 btrfs_set_header_generation(leaf, trans->transid);
1295 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1296 btrfs_set_header_owner(leaf, objectid);
1297 root->node = leaf;
1298
1299 write_extent_buffer(leaf, fs_info->fsid,
1300 (unsigned long)btrfs_header_fsid(leaf),
1301 BTRFS_FSID_SIZE);
1302 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1303 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1304 BTRFS_UUID_SIZE);
1305 btrfs_mark_buffer_dirty(leaf);
1306
1307 root->commit_root = btrfs_root_node(root);
1308 root->track_dirty = 1;
1309
1310
1311 root->root_item.flags = 0;
1312 root->root_item.byte_limit = 0;
1313 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1314 btrfs_set_root_generation(&root->root_item, trans->transid);
1315 btrfs_set_root_level(&root->root_item, 0);
1316 btrfs_set_root_refs(&root->root_item, 1);
1317 btrfs_set_root_used(&root->root_item, leaf->len);
1318 btrfs_set_root_last_snapshot(&root->root_item, 0);
1319 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1320 uuid_le_gen(&uuid);
1321 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1322 root->root_item.drop_level = 0;
1323
1324 key.objectid = objectid;
1325 key.type = BTRFS_ROOT_ITEM_KEY;
1326 key.offset = 0;
1327 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1328 if (ret)
1329 goto fail;
1330
1331 btrfs_tree_unlock(leaf);
1332
1dd05682
TI		 1333 return root;
1334
20897f5c 1335fail:
1dd05682
TI		 1336 if (leaf) {
1337 btrfs_tree_unlock(leaf);
1338 free_extent_buffer(leaf);
1339 }
1340 kfree(root);
20897f5c 1341
1dd05682 1342 return ERR_PTR(ret);
20897f5c
AJ		 1343}
1344
7237f183
YZ		 1345static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1346 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1347{
1348 struct btrfs_root *root;
1349 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1350 struct extent_buffer *leaf;
e02119d5 1351
6f07e42e 1352 root = btrfs_alloc_root(fs_info);
e02119d5 1353 if (!root)
7237f183 1354 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1355
1356 __setup_root(tree_root->nodesize, tree_root->leafsize,
1357 tree_root->sectorsize, tree_root->stripesize,
1358 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1359
1360 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1361 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1362 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1363 /*
1364 * log trees do not get reference counted because they go away
1365 * before a real commit is actually done. They do store pointers
1366 * to file data extents, and those reference counts still get
1367 * updated (along with back refs to the log tree).
1368 */
e02119d5
CM		 1369 root->ref_cows = 0;
1370
5d4f98a2 1371 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1372 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1373 0, 0, 0);
7237f183
YZ		 1374 if (IS_ERR(leaf)) {
1375 kfree(root);
1376 return ERR_CAST(leaf);
1377 }
e02119d5 1378
5d4f98a2
YZ		 1379 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1380 btrfs_set_header_bytenr(leaf, leaf->start);
1381 btrfs_set_header_generation(leaf, trans->transid);
1382 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1383 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1384 root->node = leaf;
e02119d5
CM		 1385
1386 write_extent_buffer(root->node, root->fs_info->fsid,
1387 (unsigned long)btrfs_header_fsid(root->node),
1388 BTRFS_FSID_SIZE);
1389 btrfs_mark_buffer_dirty(root->node);
1390 btrfs_tree_unlock(root->node);
7237f183
YZ		 1391 return root;
1392}
1393
1394int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1395 struct btrfs_fs_info *fs_info)
1396{
1397 struct btrfs_root *log_root;
1398
1399 log_root = alloc_log_tree(trans, fs_info);
1400 if (IS_ERR(log_root))
1401 return PTR_ERR(log_root);
1402 WARN_ON(fs_info->log_root_tree);
1403 fs_info->log_root_tree = log_root;
1404 return 0;
1405}
1406
1407int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1408 struct btrfs_root *root)
1409{
1410 struct btrfs_root *log_root;
1411 struct btrfs_inode_item *inode_item;
1412
1413 log_root = alloc_log_tree(trans, root->fs_info);
1414 if (IS_ERR(log_root))
1415 return PTR_ERR(log_root);
1416
1417 log_root->last_trans = trans->transid;
1418 log_root->root_key.offset = root->root_key.objectid;
1419
1420 inode_item = &log_root->root_item.inode;
3cae210f
QW		 1421 btrfs_set_stack_inode_generation(inode_item, 1);
1422 btrfs_set_stack_inode_size(inode_item, 3);
1423 btrfs_set_stack_inode_nlink(inode_item, 1);
1424 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1425 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
7237f183 1426
5d4f98a2 1427 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1428
1429 WARN_ON(root->log_root);
1430 root->log_root = log_root;
1431 root->log_transid = 0;
257c62e1 1432 root->last_log_commit = 0;
e02119d5
CM		 1433 return 0;
1434}
1435
35a3621b
SB		 1436static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1437 struct btrfs_key *key)
e02119d5
CM		 1438{
1439 struct btrfs_root *root;
1440 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1441 struct btrfs_path *path;
84234f3a 1442 u64 generation;
db94535d 1443 u32 blocksize;
cb517eab 1444 int ret;
0f7d52f4 1445
cb517eab
MX		 1446 path = btrfs_alloc_path();
1447 if (!path)
0f7d52f4 1448 return ERR_PTR(-ENOMEM);
cb517eab
MX		 1449
1450 root = btrfs_alloc_root(fs_info);
1451 if (!root) {
1452 ret = -ENOMEM;
1453 goto alloc_fail;
0f7d52f4
CM		 1454 }
1455
db94535d 1456 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb 1457 tree_root->sectorsize, tree_root->stripesize,
cb517eab 1458 root, fs_info, key->objectid);
0f7d52f4 1459
cb517eab
MX		 1460 ret = btrfs_find_root(tree_root, key, path,
1461 &root->root_item, &root->root_key);
0f7d52f4 1462 if (ret) {
13a8a7c8
YZ		 1463 if (ret > 0)
1464 ret = -ENOENT;
cb517eab 1465 goto find_fail;
0f7d52f4 1466 }
13a8a7c8 1467
84234f3a 1468 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1469 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1470 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1471 blocksize, generation);
cb517eab
MX		 1472 if (!root->node) {
1473 ret = -ENOMEM;
1474 goto find_fail;
1475 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1476 ret = -EIO;
1477 goto read_fail;
416bc658 1478 }
5d4f98a2 1479 root->commit_root = btrfs_root_node(root);
13a8a7c8 1480out:
cb517eab
MX		 1481 btrfs_free_path(path);
1482 return root;
1483
1484read_fail:
1485 free_extent_buffer(root->node);
1486find_fail:
1487 kfree(root);
1488alloc_fail:
1489 root = ERR_PTR(ret);
1490 goto out;
1491}
1492
1493struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1494 struct btrfs_key *location)
1495{
1496 struct btrfs_root *root;
1497
1498 root = btrfs_read_tree_root(tree_root, location);
1499 if (IS_ERR(root))
1500 return root;
1501
1502 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1503 root->ref_cows = 1;
08fe4db1
LZ		 1504 btrfs_check_and_init_root_item(&root->root_item);
1505 }
13a8a7c8 1506
5eda7b5e
CM		 1507 return root;
1508}
1509
cb517eab
MX		 1510int btrfs_init_fs_root(struct btrfs_root *root)
1511{
1512 int ret;
1513
1514 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1515 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1516 GFP_NOFS);
1517 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1518 ret = -ENOMEM;
1519 goto fail;
1520 }
1521
1522 btrfs_init_free_ino_ctl(root);
1523 mutex_init(&root->fs_commit_mutex);
1524 spin_lock_init(&root->cache_lock);
1525 init_waitqueue_head(&root->cache_wait);
1526
1527 ret = get_anon_bdev(&root->anon_dev);
1528 if (ret)
1529 goto fail;
1530 return 0;
1531fail:
1532 kfree(root->free_ino_ctl);
1533 kfree(root->free_ino_pinned);
1534 return ret;
1535}
1536
171170c1
ST		 1537static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1538 u64 root_id)
cb517eab
MX		 1539{
1540 struct btrfs_root *root;
1541
1542 spin_lock(&fs_info->fs_roots_radix_lock);
1543 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1544 (unsigned long)root_id);
1545 spin_unlock(&fs_info->fs_roots_radix_lock);
1546 return root;
1547}
1548
1549int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1550 struct btrfs_root *root)
1551{
1552 int ret;
1553
1554 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1555 if (ret)
1556 return ret;
1557
1558 spin_lock(&fs_info->fs_roots_radix_lock);
1559 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1560 (unsigned long)root->root_key.objectid,
1561 root);
1562 if (ret == 0)
1563 root->in_radix = 1;
1564 spin_unlock(&fs_info->fs_roots_radix_lock);
1565 radix_tree_preload_end();
1566
1567 return ret;
1568}
1569
edbd8d4e
CM		 1570struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1571 struct btrfs_key *location)
5eda7b5e
CM		 1572{
1573 struct btrfs_root *root;
1574 int ret;
1575
edbd8d4e
CM		 1576 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1577 return fs_info->tree_root;
1578 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1579 return fs_info->extent_root;
8f18cf13
CM		 1580 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1581 return fs_info->chunk_root;
1582 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1583 return fs_info->dev_root;
0403e47e
YZ		 1584 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1585 return fs_info->csum_root;
bcef60f2
AJ		 1586 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1587 return fs_info->quota_root ? fs_info->quota_root :
1588 ERR_PTR(-ENOENT);
4df27c4d 1589again:
cb517eab 1590 root = btrfs_lookup_fs_root(fs_info, location->objectid);
5eda7b5e
CM		 1591 if (root)
1592 return root;
1593
cb517eab 1594 root = btrfs_read_fs_root(fs_info->tree_root, location);
5eda7b5e
CM		 1595 if (IS_ERR(root))
1596 return root;
3394e160 1597
cb517eab
MX		 1598 if (btrfs_root_refs(&root->root_item) == 0) {
1599 ret = -ENOENT;
581bb050 1600 goto fail;
35a30d7c 1601 }
581bb050 1602
cb517eab 1603 ret = btrfs_init_fs_root(root);
ac08aedf
CM		 1604 if (ret)
1605 goto fail;
3394e160 1606
d68fc57b
YZ		 1607 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1608 if (ret < 0)
1609 goto fail;
1610 if (ret == 0)
1611 root->orphan_item_inserted = 1;
1612
cb517eab 1613 ret = btrfs_insert_fs_root(fs_info, root);
0f7d52f4 1614 if (ret) {
4df27c4d
YZ		 1615 if (ret == -EEXIST) {
1616 free_fs_root(root);
1617 goto again;
1618 }
1619 goto fail;
0f7d52f4 1620 }
edbd8d4e 1621 return root;
4df27c4d
YZ		 1622fail:
1623 free_fs_root(root);
1624 return ERR_PTR(ret);
edbd8d4e
CM		 1625}
1626
04160088
CM		 1627static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1628{
1629 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1630 int ret = 0;
04160088
CM		 1631 struct btrfs_device *device;
1632 struct backing_dev_info *bdi;
b7967db7 1633
1f78160c
XG		 1634 rcu_read_lock();
1635 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1636 if (!device->bdev)
1637 continue;
04160088
CM		 1638 bdi = blk_get_backing_dev_info(device->bdev);
1639 if (bdi && bdi_congested(bdi, bdi_bits)) {
1640 ret = 1;
1641 break;
1642 }
1643 }
1f78160c 1644 rcu_read_unlock();
04160088
CM		 1645 return ret;
1646}
1647
ad081f14
JA		 1648/*
1649 * If this fails, caller must call bdi_destroy() to get rid of the
1650 * bdi again.
1651 */
04160088
CM		 1652static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1653{
ad081f14
JA		 1654 int err;
1655
1656 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1657 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1658 if (err)
1659 return err;
1660
4575c9cc 1661 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1662 bdi->congested_fn = btrfs_congested_fn;
1663 bdi->congested_data = info;
1664 return 0;
1665}
1666
8b712842
CM		 1667/*
1668 * called by the kthread helper functions to finally call the bio end_io
1669 * functions. This is where read checksum verification actually happens
1670 */
1671static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1672{
ce9adaa5 1673 struct bio *bio;
8b712842
CM		 1674 struct end_io_wq *end_io_wq;
1675 struct btrfs_fs_info *fs_info;
ce9adaa5 1676 int error;
ce9adaa5 1677
8b712842
CM		 1678 end_io_wq = container_of(work, struct end_io_wq, work);
1679 bio = end_io_wq->bio;
1680 fs_info = end_io_wq->info;
ce9adaa5 1681
8b712842
CM		 1682 error = end_io_wq->error;
1683 bio->bi_private = end_io_wq->private;
1684 bio->bi_end_io = end_io_wq->end_io;
1685 kfree(end_io_wq);
8b712842 1686 bio_endio(bio, error);
44b8bd7e
CM		 1687}
1688
a74a4b97
CM		 1689static int cleaner_kthread(void *arg)
1690{
1691 struct btrfs_root *root = arg;
d0278245 1692 int again;
a74a4b97
CM		 1693
1694 do {
d0278245 1695 again = 0;
a74a4b97 1696
d0278245 1697 /* Make the cleaner go to sleep early. */
babbf170 1698 if (btrfs_need_cleaner_sleep(root))
d0278245
MX		 1699 goto sleep;
1700
1701 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1702 goto sleep;
1703
dc7f370c
MX		 1704 /*
1705 * Avoid the problem that we change the status of the fs
1706 * during the above check and trylock.
1707 */
babbf170 1708 if (btrfs_need_cleaner_sleep(root)) {
dc7f370c
MX		 1709 mutex_unlock(&root->fs_info->cleaner_mutex);
1710 goto sleep;
76dda93c 1711 }
a74a4b97 1712
d0278245
MX		 1713 btrfs_run_delayed_iputs(root);
1714 again = btrfs_clean_one_deleted_snapshot(root);
1715 mutex_unlock(&root->fs_info->cleaner_mutex);
1716
1717 /*
05323cd1
MX		 1718 * The defragger has dealt with the R/O remount and umount,
1719 * needn't do anything special here.
d0278245
MX		 1720 */
1721 btrfs_run_defrag_inodes(root->fs_info);
1722sleep:
9d1a2a3a 1723 if (!try_to_freeze() && !again) {
a74a4b97 1724 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1725 if (!kthread_should_stop())
1726 schedule();
a74a4b97
CM		 1727 __set_current_state(TASK_RUNNING);
1728 }
1729 } while (!kthread_should_stop());
1730 return 0;
1731}
1732
1733static int transaction_kthread(void *arg)
1734{
1735 struct btrfs_root *root = arg;
1736 struct btrfs_trans_handle *trans;
1737 struct btrfs_transaction *cur;
8929ecfa 1738 u64 transid;
a74a4b97
CM		 1739 unsigned long now;
1740 unsigned long delay;
914b2007 1741 bool cannot_commit;
a74a4b97
CM		 1742
1743 do {
914b2007 1744 cannot_commit = false;
8b87dc17 1745 delay = HZ * root->fs_info->commit_interval;
a74a4b97
CM		 1746 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1747
a4abeea4 1748 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1749 cur = root->fs_info->running_transaction;
1750 if (!cur) {
a4abeea4 1751 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1752 goto sleep;
1753 }
31153d81 1754
a74a4b97 1755 now = get_seconds();
4a9d8bde 1756 if (cur->state < TRANS_STATE_BLOCKED &&
8b87dc17
DS		 1757 (now < cur->start_time ||
1758 now - cur->start_time < root->fs_info->commit_interval)) {
a4abeea4 1759 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1760 delay = HZ * 5;
1761 goto sleep;
1762 }
8929ecfa 1763 transid = cur->transid;
a4abeea4 1764 spin_unlock(&root->fs_info->trans_lock);
56bec294 1765
79787eaa 1766 /* If the file system is aborted, this will always fail. */
354aa0fb 1767 trans = btrfs_attach_transaction(root);
914b2007 1768 if (IS_ERR(trans)) {
354aa0fb
MX		 1769 if (PTR_ERR(trans) != -ENOENT)
1770 cannot_commit = true;
79787eaa 1771 goto sleep;
914b2007 1772 }
8929ecfa 1773 if (transid == trans->transid) {
79787eaa 1774 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1775 } else {
1776 btrfs_end_transaction(trans, root);
1777 }
a74a4b97
CM		 1778sleep:
1779 wake_up_process(root->fs_info->cleaner_kthread);
1780 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1781
a0acae0e 1782 if (!try_to_freeze()) {
a74a4b97 1783 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1784 if (!kthread_should_stop() &&
914b2007
JK		 1785 (!btrfs_transaction_blocked(root->fs_info) ||
1786 cannot_commit))
8929ecfa 1787 schedule_timeout(delay);
a74a4b97
CM		 1788 __set_current_state(TASK_RUNNING);
1789 }
1790 } while (!kthread_should_stop());
1791 return 0;
1792}
1793
af31f5e5
CM		 1794/*
1795 * this will find the highest generation in the array of
1796 * root backups. The index of the highest array is returned,
1797 * or -1 if we can't find anything.
1798 *
1799 * We check to make sure the array is valid by comparing the
1800 * generation of the latest root in the array with the generation
1801 * in the super block. If they don't match we pitch it.
1802 */
1803static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1804{
1805 u64 cur;
1806 int newest_index = -1;
1807 struct btrfs_root_backup *root_backup;
1808 int i;
1809
1810 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1811 root_backup = info->super_copy->super_roots + i;
1812 cur = btrfs_backup_tree_root_gen(root_backup);
1813 if (cur == newest_gen)
1814 newest_index = i;
1815 }
1816
1817 /* check to see if we actually wrapped around */
1818 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1819 root_backup = info->super_copy->super_roots;
1820 cur = btrfs_backup_tree_root_gen(root_backup);
1821 if (cur == newest_gen)
1822 newest_index = 0;
1823 }
1824 return newest_index;
1825}
1826
1827
1828/*
1829 * find the oldest backup so we know where to store new entries
1830 * in the backup array. This will set the backup_root_index
1831 * field in the fs_info struct
1832 */
1833static void find_oldest_super_backup(struct btrfs_fs_info *info,
1834 u64 newest_gen)
1835{
1836 int newest_index = -1;
1837
1838 newest_index = find_newest_super_backup(info, newest_gen);
1839 /* if there was garbage in there, just move along */
1840 if (newest_index == -1) {
1841 info->backup_root_index = 0;
1842 } else {
1843 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1844 }
1845}
1846
1847/*
1848 * copy all the root pointers into the super backup array.
1849 * this will bump the backup pointer by one when it is
1850 * done
1851 */
1852static void backup_super_roots(struct btrfs_fs_info *info)
1853{
1854 int next_backup;
1855 struct btrfs_root_backup *root_backup;
1856 int last_backup;
1857
1858 next_backup = info->backup_root_index;
1859 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1860 BTRFS_NUM_BACKUP_ROOTS;
1861
1862 /*
1863 * just overwrite the last backup if we're at the same generation
1864 * this happens only at umount
1865 */
1866 root_backup = info->super_for_commit->super_roots + last_backup;
1867 if (btrfs_backup_tree_root_gen(root_backup) ==
1868 btrfs_header_generation(info->tree_root->node))
1869 next_backup = last_backup;
1870
1871 root_backup = info->super_for_commit->super_roots + next_backup;
1872
1873 /*
1874 * make sure all of our padding and empty slots get zero filled
1875 * regardless of which ones we use today
1876 */
1877 memset(root_backup, 0, sizeof(*root_backup));
1878
1879 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1880
1881 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1882 btrfs_set_backup_tree_root_gen(root_backup,
1883 btrfs_header_generation(info->tree_root->node));
1884
1885 btrfs_set_backup_tree_root_level(root_backup,
1886 btrfs_header_level(info->tree_root->node));
1887
1888 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1889 btrfs_set_backup_chunk_root_gen(root_backup,
1890 btrfs_header_generation(info->chunk_root->node));
1891 btrfs_set_backup_chunk_root_level(root_backup,
1892 btrfs_header_level(info->chunk_root->node));
1893
1894 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1895 btrfs_set_backup_extent_root_gen(root_backup,
1896 btrfs_header_generation(info->extent_root->node));
1897 btrfs_set_backup_extent_root_level(root_backup,
1898 btrfs_header_level(info->extent_root->node));
1899
7c7e82a7
CM		 1900 /*
1901 * we might commit during log recovery, which happens before we set
1902 * the fs_root. Make sure it is valid before we fill it in.
1903 */
1904 if (info->fs_root && info->fs_root->node) {
1905 btrfs_set_backup_fs_root(root_backup,
1906 info->fs_root->node->start);
1907 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1908 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1909 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1910 btrfs_header_level(info->fs_root->node));
7c7e82a7 1911 }
af31f5e5
CM		 1912
1913 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1914 btrfs_set_backup_dev_root_gen(root_backup,
1915 btrfs_header_generation(info->dev_root->node));
1916 btrfs_set_backup_dev_root_level(root_backup,
1917 btrfs_header_level(info->dev_root->node));
1918
1919 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1920 btrfs_set_backup_csum_root_gen(root_backup,
1921 btrfs_header_generation(info->csum_root->node));
1922 btrfs_set_backup_csum_root_level(root_backup,
1923 btrfs_header_level(info->csum_root->node));
1924
1925 btrfs_set_backup_total_bytes(root_backup,
1926 btrfs_super_total_bytes(info->super_copy));
1927 btrfs_set_backup_bytes_used(root_backup,
1928 btrfs_super_bytes_used(info->super_copy));
1929 btrfs_set_backup_num_devices(root_backup,
1930 btrfs_super_num_devices(info->super_copy));
1931
1932 /*
1933 * if we don't copy this out to the super_copy, it won't get remembered
1934 * for the next commit
1935 */
1936 memcpy(&info->super_copy->super_roots,
1937 &info->super_for_commit->super_roots,
1938 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1939}
1940
1941/*
1942 * this copies info out of the root backup array and back into
1943 * the in-memory super block. It is meant to help iterate through
1944 * the array, so you send it the number of backups you've already
1945 * tried and the last backup index you used.
1946 *
1947 * this returns -1 when it has tried all the backups
1948 */
1949static noinline int next_root_backup(struct btrfs_fs_info *info,
1950 struct btrfs_super_block *super,
1951 int *num_backups_tried, int *backup_index)
1952{
1953 struct btrfs_root_backup *root_backup;
1954 int newest = *backup_index;
1955
1956 if (*num_backups_tried == 0) {
1957 u64 gen = btrfs_super_generation(super);
1958
1959 newest = find_newest_super_backup(info, gen);
1960 if (newest == -1)
1961 return -1;
1962
1963 *backup_index = newest;
1964 *num_backups_tried = 1;
1965 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1966 /* we've tried all the backups, all done */
1967 return -1;
1968 } else {
1969 /* jump to the next oldest backup */
1970 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1971 BTRFS_NUM_BACKUP_ROOTS;
1972 *backup_index = newest;
1973 *num_backups_tried += 1;
1974 }
1975 root_backup = super->super_roots + newest;
1976
1977 btrfs_set_super_generation(super,
1978 btrfs_backup_tree_root_gen(root_backup));
1979 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1980 btrfs_set_super_root_level(super,
1981 btrfs_backup_tree_root_level(root_backup));
1982 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1983
1984 /*
1985 * fixme: the total bytes and num_devices need to match or we should
1986 * need a fsck
1987 */
1988 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1989 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1990 return 0;
1991}
1992
7abadb64
LB		 1993/* helper to cleanup workers */
1994static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1995{
1996 btrfs_stop_workers(&fs_info->generic_worker);
1997 btrfs_stop_workers(&fs_info->fixup_workers);
1998 btrfs_stop_workers(&fs_info->delalloc_workers);
1999 btrfs_stop_workers(&fs_info->workers);
2000 btrfs_stop_workers(&fs_info->endio_workers);
2001 btrfs_stop_workers(&fs_info->endio_meta_workers);
2002 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2003 btrfs_stop_workers(&fs_info->rmw_workers);
2004 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2005 btrfs_stop_workers(&fs_info->endio_write_workers);
2006 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2007 btrfs_stop_workers(&fs_info->submit_workers);
2008 btrfs_stop_workers(&fs_info->delayed_workers);
2009 btrfs_stop_workers(&fs_info->caching_workers);
2010 btrfs_stop_workers(&fs_info->readahead_workers);
2011 btrfs_stop_workers(&fs_info->flush_workers);
2f232036 2012 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
7abadb64
LB		 2013}
2014
af31f5e5
CM		 2015/* helper to cleanup tree roots */
2016static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2017{
2018 free_extent_buffer(info->tree_root->node);
2019 free_extent_buffer(info->tree_root->commit_root);
af31f5e5
CM		 2020 info->tree_root->node = NULL;
2021 info->tree_root->commit_root = NULL;
655b09fe
JB		 2022
2023 if (info->dev_root) {
2024 free_extent_buffer(info->dev_root->node);
2025 free_extent_buffer(info->dev_root->commit_root);
2026 info->dev_root->node = NULL;
2027 info->dev_root->commit_root = NULL;
2028 }
2029 if (info->extent_root) {
2030 free_extent_buffer(info->extent_root->node);
2031 free_extent_buffer(info->extent_root->commit_root);
2032 info->extent_root->node = NULL;
2033 info->extent_root->commit_root = NULL;
2034 }
2035 if (info->csum_root) {
2036 free_extent_buffer(info->csum_root->node);
2037 free_extent_buffer(info->csum_root->commit_root);
2038 info->csum_root->node = NULL;
2039 info->csum_root->commit_root = NULL;
2040 }
bcef60f2 2041 if (info->quota_root) {
655b09fe
JB		 2042 free_extent_buffer(info->quota_root->node);
2043 free_extent_buffer(info->quota_root->commit_root);
bcef60f2
AJ		 2044 info->quota_root->node = NULL;
2045 info->quota_root->commit_root = NULL;
2046 }
af31f5e5
CM		 2047 if (chunk_root) {
2048 free_extent_buffer(info->chunk_root->node);
2049 free_extent_buffer(info->chunk_root->commit_root);
2050 info->chunk_root->node = NULL;
2051 info->chunk_root->commit_root = NULL;
2052 }
2053}
2054
171f6537
JB		 2055static void del_fs_roots(struct btrfs_fs_info *fs_info)
2056{
2057 int ret;
2058 struct btrfs_root *gang[8];
2059 int i;
2060
2061 while (!list_empty(&fs_info->dead_roots)) {
2062 gang[0] = list_entry(fs_info->dead_roots.next,
2063 struct btrfs_root, root_list);
2064 list_del(&gang[0]->root_list);
2065
2066 if (gang[0]->in_radix) {
cb517eab 2067 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
171f6537
JB		 2068 } else {
2069 free_extent_buffer(gang[0]->node);
2070 free_extent_buffer(gang[0]->commit_root);
b0feb9d9 2071 btrfs_put_fs_root(gang[0]);
171f6537
JB		 2072 }
2073 }
2074
2075 while (1) {
2076 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2077 (void **)gang, 0,
2078 ARRAY_SIZE(gang));
2079 if (!ret)
2080 break;
2081 for (i = 0; i < ret; i++)
cb517eab 2082 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
171f6537
JB		 2083 }
2084}
af31f5e5 2085
ad2b2c80
AV		 2086int open_ctree(struct super_block *sb,
2087 struct btrfs_fs_devices *fs_devices,
2088 char *options)
2e635a27 2089{
db94535d
CM		 2090 u32 sectorsize;
2091 u32 nodesize;
2092 u32 leafsize;
2093 u32 blocksize;
87ee04eb 2094 u32 stripesize;
84234f3a 2095 u64 generation;
f2b636e8 2096 u64 features;
3de4586c 2097 struct btrfs_key location;
a061fc8d 2098 struct buffer_head *bh;
4d34b278 2099 struct btrfs_super_block *disk_super;
815745cf 2100 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2101 struct btrfs_root *tree_root;
4d34b278
ID		 2102 struct btrfs_root *extent_root;
2103 struct btrfs_root *csum_root;
2104 struct btrfs_root *chunk_root;
2105 struct btrfs_root *dev_root;
bcef60f2 2106 struct btrfs_root *quota_root;
e02119d5 2107 struct btrfs_root *log_tree_root;
eb60ceac 2108 int ret;
e58ca020 2109 int err = -EINVAL;
af31f5e5
CM		 2110 int num_backups_tried = 0;
2111 int backup_index = 0;
4543df7e 2112
f84a8bd6 2113 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e 2114 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
cb517eab 2115 if (!tree_root || !chunk_root) {
39279cc3
CM		 2116 err = -ENOMEM;
2117 goto fail;
2118 }
76dda93c
YZ		 2119
2120 ret = init_srcu_struct(&fs_info->subvol_srcu);
2121 if (ret) {
2122 err = ret;
2123 goto fail;
2124 }
2125
2126 ret = setup_bdi(fs_info, &fs_info->bdi);
2127 if (ret) {
2128 err = ret;
2129 goto fail_srcu;
2130 }
2131
e2d84521
MX		 2132 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2133 if (ret) {
2134 err = ret;
2135 goto fail_bdi;
2136 }
2137 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2138 (1 + ilog2(nr_cpu_ids));
2139
963d678b
MX		 2140 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2141 if (ret) {
2142 err = ret;
2143 goto fail_dirty_metadata_bytes;
2144 }
2145
76dda93c
YZ		 2146 fs_info->btree_inode = new_inode(sb);
2147 if (!fs_info->btree_inode) {
2148 err = -ENOMEM;
963d678b 2149 goto fail_delalloc_bytes;
76dda93c
YZ		 2150 }
2151
a6591715 2152 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2153
76dda93c 2154 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 2155 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2156 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2157 INIT_LIST_HEAD(&fs_info->delayed_iputs);
eb73c1b7 2158 INIT_LIST_HEAD(&fs_info->delalloc_roots);
11833d66 2159 INIT_LIST_HEAD(&fs_info->caching_block_groups);
eb73c1b7 2160 spin_lock_init(&fs_info->delalloc_root_lock);
a4abeea4 2161 spin_lock_init(&fs_info->trans_lock);
76dda93c 2162 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2163 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2164 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2165 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2166 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2167 spin_lock_init(&fs_info->super_lock);
f29021b2 2168 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2169 mutex_init(&fs_info->reloc_mutex);
de98ced9 2170 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2171
58176a96 2172 init_completion(&fs_info->kobj_unregister);
0b86a832 2173 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2174 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2175 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2176 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2177 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2178 BTRFS_BLOCK_RSV_GLOBAL);
2179 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2180 BTRFS_BLOCK_RSV_DELALLOC);
2181 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2182 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2183 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2184 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2185 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2186 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2187 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2188 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2189 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2190 atomic_set(&fs_info->defrag_running, 0);
fc36ed7e 2191 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2192 fs_info->sb = sb;
6f568d35 2193 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2194 fs_info->metadata_ratio = 0;
4cb5300b 2195 fs_info->defrag_inodes = RB_ROOT;
2bf64758 2196 fs_info->free_chunk_space = 0;
f29021b2 2197 fs_info->tree_mod_log = RB_ROOT;
8b87dc17 2198 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
c8b97818 2199
90519d66
AJ		 2200 /* readahead state */
2201 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2202 spin_lock_init(&fs_info->reada_lock);
c8b97818 2203
b34b086c
CM		 2204 fs_info->thread_pool_size = min_t(unsigned long,
2205 num_online_cpus() + 2, 8);
0afbaf8c 2206
199c2a9c
MX		 2207 INIT_LIST_HEAD(&fs_info->ordered_roots);
2208 spin_lock_init(&fs_info->ordered_root_lock);
16cdcec7
MX		 2209 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2210 GFP_NOFS);
2211 if (!fs_info->delayed_root) {
2212 err = -ENOMEM;
2213 goto fail_iput;
2214 }
2215 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2216
a2de733c
AJ		 2217 mutex_init(&fs_info->scrub_lock);
2218 atomic_set(&fs_info->scrubs_running, 0);
2219 atomic_set(&fs_info->scrub_pause_req, 0);
2220 atomic_set(&fs_info->scrubs_paused, 0);
2221 atomic_set(&fs_info->scrub_cancel_req, 0);
2222 init_waitqueue_head(&fs_info->scrub_pause_wait);
2223 init_rwsem(&fs_info->scrub_super_lock);
2224 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2225#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2226 fs_info->check_integrity_print_mask = 0;
2227#endif
a2de733c 2228
c9e9f97b
ID		 2229 spin_lock_init(&fs_info->balance_lock);
2230 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2231 atomic_set(&fs_info->balance_running, 0);
2232 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2233 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2234 fs_info->balance_ctl = NULL;
837d5b6e 2235 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2236
a061fc8d
CM		 2237 sb->s_blocksize = 4096;
2238 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2239 sb->s_bdi = &fs_info->bdi;
a061fc8d 2240
76dda93c 2241 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2242 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2243 /*
2244 * we set the i_size on the btree inode to the max possible int.
2245 * the real end of the address space is determined by all of
2246 * the devices in the system
2247 */
2248 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2249 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2250 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2251
5d4f98a2 2252 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2253 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2254 fs_info->btree_inode->i_mapping);
0b32f4bb 2255 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2256 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2257
2258 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2259
76dda93c
YZ		 2260 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2261 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2262 sizeof(struct btrfs_key));
72ac3c0d
JB		 2263 set_bit(BTRFS_INODE_DUMMY,
2264 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
c65ddb52 2265 insert_inode_hash(fs_info->btree_inode);
76dda93c 2266
0f9dd46c 2267 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2268 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2269 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2270
11833d66 2271 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2272 fs_info->btree_inode->i_mapping);
11833d66 2273 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2274 fs_info->btree_inode->i_mapping);
11833d66 2275 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2276 fs_info->do_barriers = 1;
e18e4809 2277
39279cc3 2278
5a3f23d5 2279 mutex_init(&fs_info->ordered_operations_mutex);
9ffba8cd 2280 mutex_init(&fs_info->ordered_extent_flush_mutex);
e02119d5 2281 mutex_init(&fs_info->tree_log_mutex);
925baedd 2282 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2283 mutex_init(&fs_info->transaction_kthread_mutex);
2284 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2285 mutex_init(&fs_info->volume_mutex);
276e680d 2286 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2287 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2288 init_rwsem(&fs_info->subvol_sem);
e922e087
SB		 2289 fs_info->dev_replace.lock_owner = 0;
2290 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2291 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2292 mutex_init(&fs_info->dev_replace.lock_management_lock);
2293 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2294
416ac51d 2295 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2296 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2297 fs_info->qgroup_tree = RB_ROOT;
2298 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2299 fs_info->qgroup_seq = 1;
2300 fs_info->quota_enabled = 0;
2301 fs_info->pending_quota_state = 0;
1e8f9158 2302 fs_info->qgroup_ulist = NULL;
2f232036 2303 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2304
fa9c0d79
CM		 2305 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2306 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2307
e6dcd2dc 2308 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2309 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2310 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2311 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2312
53b381b3
DW		 2313 ret = btrfs_alloc_stripe_hash_table(fs_info);
2314 if (ret) {
83c8266a 2315 err = ret;
53b381b3
DW		 2316 goto fail_alloc;
2317 }
2318
0b86a832 2319 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2320 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2321
3c4bb26b 2322 invalidate_bdev(fs_devices->latest_bdev);
1104a885
DS		 2323
2324 /*
2325 * Read super block and check the signature bytes only
2326 */
a512bbf8 2327 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2328 if (!bh) {
2329 err = -EINVAL;
16cdcec7 2330 goto fail_alloc;
20b45077 2331 }
39279cc3 2332
1104a885
DS		 2333 /*
2334 * We want to check superblock checksum, the type is stored inside.
2335 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2336 */
2337 if (btrfs_check_super_csum(bh->b_data)) {
2338 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2339 err = -EINVAL;
2340 goto fail_alloc;
2341 }
2342
2343 /*
2344 * super_copy is zeroed at allocation time and we never touch the
2345 * following bytes up to INFO_SIZE, the checksum is calculated from
2346 * the whole block of INFO_SIZE
2347 */
6c41761f
DS		 2348 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2349 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2350 sizeof(*fs_info->super_for_commit));
a061fc8d 2351 brelse(bh);
5f39d397 2352
6c41761f 2353 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2354
1104a885
DS		 2355 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2356 if (ret) {
2357 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2358 err = -EINVAL;
2359 goto fail_alloc;
2360 }
2361
6c41761f 2362 disk_super = fs_info->super_copy;
0f7d52f4 2363 if (!btrfs_super_root(disk_super))
16cdcec7 2364 goto fail_alloc;
0f7d52f4 2365
acce952b 2366 /* check FS state, whether FS is broken. */
87533c47
MX		 2367 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2368 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2369
af31f5e5
CM		 2370 /*
2371 * run through our array of backup supers and setup
2372 * our ring pointer to the oldest one
2373 */
2374 generation = btrfs_super_generation(disk_super);
2375 find_oldest_super_backup(fs_info, generation);
2376
75e7cb7f
LB		 2377 /*
2378 * In the long term, we'll store the compression type in the super
2379 * block, and it'll be used for per file compression control.
2380 */
2381 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2382
2b82032c
YZ		 2383 ret = btrfs_parse_options(tree_root, options);
2384 if (ret) {
2385 err = ret;
16cdcec7 2386 goto fail_alloc;
2b82032c 2387 }
dfe25020 2388
f2b636e8
JB		 2389 features = btrfs_super_incompat_flags(disk_super) &
2390 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2391 if (features) {
2392 printk(KERN_ERR "BTRFS: couldn't mount because of "
2393 "unsupported optional features (%Lx).\n",
21380931 2394 (unsigned long long)features);
f2b636e8 2395 err = -EINVAL;
16cdcec7 2396 goto fail_alloc;
f2b636e8
JB		 2397 }
2398
727011e0
CM		 2399 if (btrfs_super_leafsize(disk_super) !=
2400 btrfs_super_nodesize(disk_super)) {
2401 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2402 "blocksizes don't match. node %d leaf %d\n",
2403 btrfs_super_nodesize(disk_super),
2404 btrfs_super_leafsize(disk_super));
2405 err = -EINVAL;
2406 goto fail_alloc;
2407 }
2408 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2409 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2410 "blocksize (%d) was too large\n",
2411 btrfs_super_leafsize(disk_super));
2412 err = -EINVAL;
2413 goto fail_alloc;
2414 }
2415
5d4f98a2 2416 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2417 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2418 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2419 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2420
3173a18f
JB		 2421 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2422 printk(KERN_ERR "btrfs: has skinny extents\n");
2423
727011e0
CM		 2424 /*
2425 * flag our filesystem as having big metadata blocks if
2426 * they are bigger than the page size
2427 */
2428 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2429 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2430 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2431 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2432 }
2433
bc3f116f
CM		 2434 nodesize = btrfs_super_nodesize(disk_super);
2435 leafsize = btrfs_super_leafsize(disk_super);
2436 sectorsize = btrfs_super_sectorsize(disk_super);
2437 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2438 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2439 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2440
2441 /*
2442 * mixed block groups end up with duplicate but slightly offset
2443 * extent buffers for the same range. It leads to corruptions
2444 */
2445 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2446 (sectorsize != leafsize)) {
2447 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2448 "are not allowed for mixed block groups on %s\n",
2449 sb->s_id);
2450 goto fail_alloc;
2451 }
2452
ceda0864
MX		 2453 /*
2454 * Needn't use the lock because there is no other task which will
2455 * update the flag.
2456 */
a6fa6fae 2457 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2458
f2b636e8
JB		 2459 features = btrfs_super_compat_ro_flags(disk_super) &
2460 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2461 if (!(sb->s_flags & MS_RDONLY) && features) {
2462 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2463 "unsupported option features (%Lx).\n",
21380931 2464 (unsigned long long)features);
f2b636e8 2465 err = -EINVAL;
16cdcec7 2466 goto fail_alloc;
f2b636e8 2467 }
61d92c32
CM		 2468
2469 btrfs_init_workers(&fs_info->generic_worker,
2470 "genwork", 1, NULL);
2471
5443be45 2472 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 2473 fs_info->thread_pool_size,
2474 &fs_info->generic_worker);
c8b97818 2475
771ed689 2476 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 2477 fs_info->thread_pool_size,
2478 &fs_info->generic_worker);
771ed689 2479
8ccf6f19
MX		 2480 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2481 fs_info->thread_pool_size,
2482 &fs_info->generic_worker);
2483
5443be45 2484 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 2485 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 2486 fs_info->thread_pool_size),
2487 &fs_info->generic_worker);
61b49440 2488
bab39bf9
JB		 2489 btrfs_init_workers(&fs_info->caching_workers, "cache",
2490 2, &fs_info->generic_worker);
2491
61b49440
CM		 2492 /* a higher idle thresh on the submit workers makes it much more
2493 * likely that bios will be send down in a sane order to the
2494 * devices
2495 */
2496 fs_info->submit_workers.idle_thresh = 64;
53863232 2497
771ed689 2498 fs_info->workers.idle_thresh = 16;
4a69a410 2499 fs_info->workers.ordered = 1;
61b49440 2500
771ed689
CM		 2501 fs_info->delalloc_workers.idle_thresh = 2;
2502 fs_info->delalloc_workers.ordered = 1;
2503
61d92c32
CM		 2504 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2505 &fs_info->generic_worker);
5443be45 2506 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2507 fs_info->thread_pool_size,
2508 &fs_info->generic_worker);
d20f7043 2509 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2510 fs_info->thread_pool_size,
2511 &fs_info->generic_worker);
cad321ad 2512 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2513 "endio-meta-write", fs_info->thread_pool_size,
2514 &fs_info->generic_worker);
53b381b3
DW		 2515 btrfs_init_workers(&fs_info->endio_raid56_workers,
2516 "endio-raid56", fs_info->thread_pool_size,
2517 &fs_info->generic_worker);
2518 btrfs_init_workers(&fs_info->rmw_workers,
2519 "rmw", fs_info->thread_pool_size,
2520 &fs_info->generic_worker);
5443be45 2521 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2522 fs_info->thread_pool_size,
2523 &fs_info->generic_worker);
0cb59c99
JB		 2524 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2525 1, &fs_info->generic_worker);
16cdcec7
MX		 2526 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2527 fs_info->thread_pool_size,
2528 &fs_info->generic_worker);
90519d66
AJ		 2529 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2530 fs_info->thread_pool_size,
2531 &fs_info->generic_worker);
2f232036
JS		 2532 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2533 &fs_info->generic_worker);
61b49440
CM		 2534
2535 /*
2536 * endios are largely parallel and should have a very
2537 * low idle thresh
2538 */
2539 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2540 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2541 fs_info->endio_raid56_workers.idle_thresh = 4;
2542 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2543
9042846b
CM		 2544 fs_info->endio_write_workers.idle_thresh = 2;
2545 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2546 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2547
0dc3b84a
JB		 2548 /*
2549 * btrfs_start_workers can really only fail because of ENOMEM so just
2550 * return -ENOMEM if any of these fail.
2551 */
2552 ret = btrfs_start_workers(&fs_info->workers);
2553 ret |= btrfs_start_workers(&fs_info->generic_worker);
2554 ret |= btrfs_start_workers(&fs_info->submit_workers);
2555 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2556 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2557 ret |= btrfs_start_workers(&fs_info->endio_workers);
2558 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2559 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2560 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2561 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2562 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2563 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2564 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2565 ret |= btrfs_start_workers(&fs_info->caching_workers);
2566 ret |= btrfs_start_workers(&fs_info->readahead_workers);
8ccf6f19 2567 ret |= btrfs_start_workers(&fs_info->flush_workers);
2f232036 2568 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
0dc3b84a 2569 if (ret) {
fed425c7 2570 err = -ENOMEM;
0dc3b84a
JB		 2571 goto fail_sb_buffer;
2572 }
4543df7e 2573
4575c9cc 2574 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2575 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2576 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2577
db94535d
CM		 2578 tree_root->nodesize = nodesize;
2579 tree_root->leafsize = leafsize;
2580 tree_root->sectorsize = sectorsize;
87ee04eb 2581 tree_root->stripesize = stripesize;
a061fc8d
CM		 2582
2583 sb->s_blocksize = sectorsize;
2584 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2585
3cae210f 2586 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
d397712b 2587 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2588 goto fail_sb_buffer;
2589 }
19c00ddc 2590
8d082fb7
LB		 2591 if (sectorsize != PAGE_SIZE) {
2592 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2593 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2594 goto fail_sb_buffer;
2595 }
2596
925baedd 2597 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2598 ret = btrfs_read_sys_array(tree_root);
925baedd 2599 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2600 if (ret) {
d397712b
CM		 2601 printk(KERN_WARNING "btrfs: failed to read the system "
2602 "array on %s\n", sb->s_id);
5d4f98a2 2603 goto fail_sb_buffer;
84eed90f 2604 }
0b86a832
CM		 2605
2606 blocksize = btrfs_level_size(tree_root,
2607 btrfs_super_chunk_root_level(disk_super));
84234f3a 2608 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2609
2610 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2611 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2612
2613 chunk_root->node = read_tree_block(chunk_root,
2614 btrfs_super_chunk_root(disk_super),
84234f3a 2615 blocksize, generation);
416bc658
JB		 2616 if (!chunk_root->node ||
2617 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
83121942
DW		 2618 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2619 sb->s_id);
af31f5e5 2620 goto fail_tree_roots;
83121942 2621 }
5d4f98a2
YZ		 2622 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2623 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2624
e17cade2 2625 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 2626 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2627 BTRFS_UUID_SIZE);
e17cade2 2628
0b86a832 2629 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2630 if (ret) {
d397712b
CM		 2631 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2632 sb->s_id);
af31f5e5 2633 goto fail_tree_roots;
2b82032c 2634 }
0b86a832 2635
8dabb742
SB		 2636 /*
2637 * keep the device that is marked to be the target device for the
2638 * dev_replace procedure
2639 */
2640 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2641
a6b0d5c8
CM		 2642 if (!fs_devices->latest_bdev) {
2643 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2644 sb->s_id);
2645 goto fail_tree_roots;
2646 }
2647
af31f5e5 2648retry_root_backup:
db94535d
CM		 2649 blocksize = btrfs_level_size(tree_root,
2650 btrfs_super_root_level(disk_super));
84234f3a 2651 generation = btrfs_super_generation(disk_super);
0b86a832 2652
e20d96d6 2653 tree_root->node = read_tree_block(tree_root,
db94535d 2654 btrfs_super_root(disk_super),
84234f3a 2655 blocksize, generation);
af31f5e5
CM		 2656 if (!tree_root->node ||
2657 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
83121942
DW		 2658 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2659 sb->s_id);
af31f5e5
CM		 2660
2661 goto recovery_tree_root;
83121942 2662 }
af31f5e5 2663
5d4f98a2
YZ		 2664 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2665 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d 2666
cb517eab
MX		 2667 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2668 location.type = BTRFS_ROOT_ITEM_KEY;
2669 location.offset = 0;
2670
2671 extent_root = btrfs_read_tree_root(tree_root, &location);
2672 if (IS_ERR(extent_root)) {
2673 ret = PTR_ERR(extent_root);
af31f5e5 2674 goto recovery_tree_root;
cb517eab 2675 }
0b86a832 2676 extent_root->track_dirty = 1;
cb517eab 2677 fs_info->extent_root = extent_root;
0b86a832 2678
cb517eab
MX		 2679 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2680 dev_root = btrfs_read_tree_root(tree_root, &location);
2681 if (IS_ERR(dev_root)) {
2682 ret = PTR_ERR(dev_root);
af31f5e5 2683 goto recovery_tree_root;
cb517eab 2684 }
5d4f98a2 2685 dev_root->track_dirty = 1;
cb517eab
MX		 2686 fs_info->dev_root = dev_root;
2687 btrfs_init_devices_late(fs_info);
3768f368 2688
cb517eab
MX		 2689 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2690 csum_root = btrfs_read_tree_root(tree_root, &location);
2691 if (IS_ERR(csum_root)) {
2692 ret = PTR_ERR(csum_root);
af31f5e5 2693 goto recovery_tree_root;
cb517eab 2694 }
d20f7043 2695 csum_root->track_dirty = 1;
cb517eab 2696 fs_info->csum_root = csum_root;
d20f7043 2697
cb517eab
MX		 2698 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2699 quota_root = btrfs_read_tree_root(tree_root, &location);
2700 if (!IS_ERR(quota_root)) {
bcef60f2
AJ		 2701 quota_root->track_dirty = 1;
2702 fs_info->quota_enabled = 1;
2703 fs_info->pending_quota_state = 1;
cb517eab 2704 fs_info->quota_root = quota_root;
bcef60f2
AJ		 2705 }
2706
8929ecfa
YZ		 2707 fs_info->generation = generation;
2708 fs_info->last_trans_committed = generation;
8929ecfa 2709
68310a5e
ID		 2710 ret = btrfs_recover_balance(fs_info);
2711 if (ret) {
2712 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2713 goto fail_block_groups;
2714 }
2715
733f4fbb
SB		 2716 ret = btrfs_init_dev_stats(fs_info);
2717 if (ret) {
2718 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2719 ret);
2720 goto fail_block_groups;
2721 }
2722
8dabb742
SB		 2723 ret = btrfs_init_dev_replace(fs_info);
2724 if (ret) {
2725 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2726 goto fail_block_groups;
2727 }
2728
2729 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2730
c59021f8 2731 ret = btrfs_init_space_info(fs_info);
2732 if (ret) {
2733 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2734 goto fail_block_groups;
2735 }
2736
1b1d1f66
JB		 2737 ret = btrfs_read_block_groups(extent_root);
2738 if (ret) {
2739 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2740 goto fail_block_groups;
2741 }
5af3e8cc
SB		 2742 fs_info->num_tolerated_disk_barrier_failures =
2743 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2744 if (fs_info->fs_devices->missing_devices >
2745 fs_info->num_tolerated_disk_barrier_failures &&
2746 !(sb->s_flags & MS_RDONLY)) {
2747 printk(KERN_WARNING
2748 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2749 goto fail_block_groups;
2750 }
9078a3e1 2751
a74a4b97
CM		 2752 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2753 "btrfs-cleaner");
57506d50 2754 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2755 goto fail_block_groups;
a74a4b97
CM		 2756
2757 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2758 tree_root,
2759 "btrfs-transaction");
57506d50 2760 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2761 goto fail_cleaner;
a74a4b97 2762
c289811c
CM		 2763 if (!btrfs_test_opt(tree_root, SSD) &&
2764 !btrfs_test_opt(tree_root, NOSSD) &&
2765 !fs_info->fs_devices->rotating) {
2766 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2767 "mode\n");
2768 btrfs_set_opt(fs_info->mount_opt, SSD);
2769 }
2770
21adbd5c
SB		 2771#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2772 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2773 ret = btrfsic_mount(tree_root, fs_devices,
2774 btrfs_test_opt(tree_root,
2775 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2776 1 : 0,
2777 fs_info->check_integrity_print_mask);
2778 if (ret)
2779 printk(KERN_WARNING "btrfs: failed to initialize"
2780 " integrity check module %s\n", sb->s_id);
2781 }
2782#endif
bcef60f2
AJ		 2783 ret = btrfs_read_qgroup_config(fs_info);
2784 if (ret)
2785 goto fail_trans_kthread;
21adbd5c 2786
acce952b 2787 /* do not make disk changes in broken FS */
68ce9682 2788 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2789 u64 bytenr = btrfs_super_log_root(disk_super);
2790
7c2ca468 2791 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2792 printk(KERN_WARNING "Btrfs log replay required "
2793 "on RO media\n");
7c2ca468 2794 err = -EIO;
bcef60f2 2795 goto fail_qgroup;
7c2ca468 2796 }
e02119d5
CM		 2797 blocksize =
2798 btrfs_level_size(tree_root,
2799 btrfs_super_log_root_level(disk_super));
d18a2c44 2800
6f07e42e 2801 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2802 if (!log_tree_root) {
2803 err = -ENOMEM;
bcef60f2 2804 goto fail_qgroup;
676e4c86 2805 }
e02119d5
CM		 2806
2807 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2808 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2809
2810 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2811 blocksize,
2812 generation + 1);
416bc658
JB		 2813 if (!log_tree_root->node ||
2814 !extent_buffer_uptodate(log_tree_root->node)) {
2815 printk(KERN_ERR "btrfs: failed to read log tree\n");
2816 free_extent_buffer(log_tree_root->node);
2817 kfree(log_tree_root);
2818 goto fail_trans_kthread;
2819 }
79787eaa 2820 /* returns with log_tree_root freed on success */
e02119d5 2821 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2822 if (ret) {
2823 btrfs_error(tree_root->fs_info, ret,
2824 "Failed to recover log tree");
2825 free_extent_buffer(log_tree_root->node);
2826 kfree(log_tree_root);
2827 goto fail_trans_kthread;
2828 }
e556ce2c
YZ		 2829
2830 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2831 ret = btrfs_commit_super(tree_root);
2832 if (ret)
2833 goto fail_trans_kthread;
e556ce2c 2834 }
e02119d5 2835 }
1a40e23b 2836
76dda93c 2837 ret = btrfs_find_orphan_roots(tree_root);
79787eaa
JM		 2838 if (ret)
2839 goto fail_trans_kthread;
76dda93c 2840
7c2ca468 2841 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2842 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2
ID		 2843 if (ret)
2844 goto fail_trans_kthread;
d68fc57b 2845
5d4f98a2 2846 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2847 if (ret < 0) {
2848 printk(KERN_WARNING
2849 "btrfs: failed to recover relocation\n");
2850 err = -EINVAL;
bcef60f2 2851 goto fail_qgroup;
d7ce5843 2852 }
7c2ca468 2853 }
1a40e23b 2854
3de4586c
CM		 2855 location.objectid = BTRFS_FS_TREE_OBJECTID;
2856 location.type = BTRFS_ROOT_ITEM_KEY;
cb517eab 2857 location.offset = 0;
3de4586c 2858
3de4586c 2859 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3140c9a3
DC		 2860 if (IS_ERR(fs_info->fs_root)) {
2861 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2862 goto fail_qgroup;
3140c9a3 2863 }
c289811c 2864
2b6ba629
ID		 2865 if (sb->s_flags & MS_RDONLY)
2866 return 0;
59641015 2867
2b6ba629
ID		 2868 down_read(&fs_info->cleanup_work_sem);
2869 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2870 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2871 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2872 close_ctree(tree_root);
2873 return ret;
2874 }
2875 up_read(&fs_info->cleanup_work_sem);
59641015 2876
2b6ba629
ID		 2877 ret = btrfs_resume_balance_async(fs_info);
2878 if (ret) {
2879 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2880 close_ctree(tree_root);
2881 return ret;
e3acc2a6
JB		 2882 }
2883
8dabb742
SB		 2884 ret = btrfs_resume_dev_replace_async(fs_info);
2885 if (ret) {
2886 pr_warn("btrfs: failed to resume dev_replace\n");
2887 close_ctree(tree_root);
2888 return ret;
2889 }
2890
b382a324
JS		 2891 btrfs_qgroup_rescan_resume(fs_info);
2892
ad2b2c80 2893 return 0;
39279cc3 2894
bcef60f2
AJ		 2895fail_qgroup:
2896 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2897fail_trans_kthread:
2898 kthread_stop(fs_info->transaction_kthread);
54067ae9 2899 btrfs_cleanup_transaction(fs_info->tree_root);
7b5ff90e 2900 del_fs_roots(fs_info);
3f157a2f 2901fail_cleaner:
a74a4b97 2902 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2903
2904 /*
2905 * make sure we're done with the btree inode before we stop our
2906 * kthreads
2907 */
2908 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2909
1b1d1f66 2910fail_block_groups:
54067ae9 2911 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2912 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2913
2914fail_tree_roots:
2915 free_root_pointers(fs_info, 1);
2b8195bb 2916 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2917
39279cc3 2918fail_sb_buffer:
7abadb64 2919 btrfs_stop_all_workers(fs_info);
16cdcec7 2920fail_alloc:
4543df7e 2921fail_iput:
586e46e2
ID		 2922 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2923
4543df7e 2924 iput(fs_info->btree_inode);
963d678b
MX		 2925fail_delalloc_bytes:
2926 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 2927fail_dirty_metadata_bytes:
2928 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 2929fail_bdi:
7e662854 2930 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2931fail_srcu:
2932 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2933fail:
53b381b3 2934 btrfs_free_stripe_hash_table(fs_info);
586e46e2 2935 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 2936 return err;
af31f5e5
CM		 2937
2938recovery_tree_root:
af31f5e5
CM		 2939 if (!btrfs_test_opt(tree_root, RECOVERY))
2940 goto fail_tree_roots;
2941
2942 free_root_pointers(fs_info, 0);
2943
2944 /* don't use the log in recovery mode, it won't be valid */
2945 btrfs_set_super_log_root(disk_super, 0);
2946
2947 /* we can't trust the free space cache either */
2948 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2949
2950 ret = next_root_backup(fs_info, fs_info->super_copy,
2951 &num_backups_tried, &backup_index);
2952 if (ret == -1)
2953 goto fail_block_groups;
2954 goto retry_root_backup;
eb60ceac
CM		 2955}
2956
f2984462
CM		 2957static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2958{
f2984462
CM		 2959 if (uptodate) {
2960 set_buffer_uptodate(bh);
2961 } else {
442a4f63
SB		 2962 struct btrfs_device *device = (struct btrfs_device *)
2963 bh->b_private;
2964
606686ee
JB		 2965 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2966 "I/O error on %s\n",
2967 rcu_str_deref(device->name));
1259ab75
CM		 2968 /* note, we dont' set_buffer_write_io_error because we have
2969 * our own ways of dealing with the IO errors
2970 */
f2984462 2971 clear_buffer_uptodate(bh);
442a4f63 2972 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 2973 }
2974 unlock_buffer(bh);
2975 put_bh(bh);
2976}
2977
a512bbf8
YZ		 2978struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2979{
2980 struct buffer_head *bh;
2981 struct buffer_head *latest = NULL;
2982 struct btrfs_super_block *super;
2983 int i;
2984 u64 transid = 0;
2985 u64 bytenr;
2986
2987 /* we would like to check all the supers, but that would make
2988 * a btrfs mount succeed after a mkfs from a different FS.
2989 * So, we need to add a special mount option to scan for
2990 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2991 */
2992 for (i = 0; i < 1; i++) {
2993 bytenr = btrfs_sb_offset(i);
8068a47e
AJ		 2994 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2995 i_size_read(bdev->bd_inode))
a512bbf8 2996 break;
8068a47e
AJ		 2997 bh = __bread(bdev, bytenr / 4096,
2998 BTRFS_SUPER_INFO_SIZE);
a512bbf8
YZ		 2999 if (!bh)
3000 continue;
3001
3002 super = (struct btrfs_super_block *)bh->b_data;
3003 if (btrfs_super_bytenr(super) != bytenr ||
3cae210f 3004 btrfs_super_magic(super) != BTRFS_MAGIC) {
a512bbf8
YZ		 3005 brelse(bh);
3006 continue;
3007 }
3008
3009 if (!latest || btrfs_super_generation(super) > transid) {
3010 brelse(latest);
3011 latest = bh;
3012 transid = btrfs_super_generation(super);
3013 } else {
3014 brelse(bh);
3015 }
3016 }
3017 return latest;
3018}
3019
4eedeb75
HH		 3020/*
3021 * this should be called twice, once with wait == 0 and
3022 * once with wait == 1. When wait == 0 is done, all the buffer heads
3023 * we write are pinned.
3024 *
3025 * They are released when wait == 1 is done.
3026 * max_mirrors must be the same for both runs, and it indicates how
3027 * many supers on this one device should be written.
3028 *
3029 * max_mirrors == 0 means to write them all.
3030 */
a512bbf8
YZ		 3031static int write_dev_supers(struct btrfs_device *device,
3032 struct btrfs_super_block *sb,
3033 int do_barriers, int wait, int max_mirrors)
3034{
3035 struct buffer_head *bh;
3036 int i;
3037 int ret;
3038 int errors = 0;
3039 u32 crc;
3040 u64 bytenr;
a512bbf8
YZ		 3041
3042 if (max_mirrors == 0)
3043 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3044
a512bbf8
YZ		 3045 for (i = 0; i < max_mirrors; i++) {
3046 bytenr = btrfs_sb_offset(i);
3047 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3048 break;
3049
3050 if (wait) {
3051 bh = __find_get_block(device->bdev, bytenr / 4096,
3052 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3053 if (!bh) {
3054 errors++;
3055 continue;
3056 }
a512bbf8 3057 wait_on_buffer(bh);
4eedeb75
HH		 3058 if (!buffer_uptodate(bh))
3059 errors++;
3060
3061 /* drop our reference */
3062 brelse(bh);
3063
3064 /* drop the reference from the wait == 0 run */
3065 brelse(bh);
3066 continue;
a512bbf8
YZ		 3067 } else {
3068 btrfs_set_super_bytenr(sb, bytenr);
3069
3070 crc = ~(u32)0;
b0496686 3071 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 3072 BTRFS_CSUM_SIZE, crc,
3073 BTRFS_SUPER_INFO_SIZE -
3074 BTRFS_CSUM_SIZE);
3075 btrfs_csum_final(crc, sb->csum);
3076
4eedeb75
HH		 3077 /*
3078 * one reference for us, and we leave it for the
3079 * caller
3080 */
a512bbf8
YZ		 3081 bh = __getblk(device->bdev, bytenr / 4096,
3082 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3083 if (!bh) {
3084 printk(KERN_ERR "btrfs: couldn't get super "
3085 "buffer head for bytenr %Lu\n", bytenr);
3086 errors++;
3087 continue;
3088 }
3089
a512bbf8
YZ		 3090 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3091
4eedeb75 3092 /* one reference for submit_bh */
a512bbf8 3093 get_bh(bh);
4eedeb75
HH		 3094
3095 set_buffer_uptodate(bh);
a512bbf8
YZ		 3096 lock_buffer(bh);
3097 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3098 bh->b_private = device;
a512bbf8
YZ		 3099 }
3100
387125fc
CM		 3101 /*
3102 * we fua the first super. The others we allow
3103 * to go down lazy.
3104 */
21adbd5c 3105 ret = btrfsic_submit_bh(WRITE_FUA, bh);
4eedeb75 3106 if (ret)
a512bbf8 3107 errors++;
a512bbf8
YZ		 3108 }
3109 return errors < i ? 0 : -1;
3110}
3111
387125fc
CM		 3112/*
3113 * endio for the write_dev_flush, this will wake anyone waiting
3114 * for the barrier when it is done
3115 */
3116static void btrfs_end_empty_barrier(struct bio *bio, int err)
3117{
3118 if (err) {
3119 if (err == -EOPNOTSUPP)
3120 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3121 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3122 }
3123 if (bio->bi_private)
3124 complete(bio->bi_private);
3125 bio_put(bio);
3126}
3127
3128/*
3129 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3130 * sent down. With wait == 1, it waits for the previous flush.
3131 *
3132 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3133 * capable
3134 */
3135static int write_dev_flush(struct btrfs_device *device, int wait)
3136{
3137 struct bio *bio;
3138 int ret = 0;
3139
3140 if (device->nobarriers)
3141 return 0;
3142
3143 if (wait) {
3144 bio = device->flush_bio;
3145 if (!bio)
3146 return 0;
3147
3148 wait_for_completion(&device->flush_wait);
3149
3150 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
606686ee
JB		 3151 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3152 rcu_str_deref(device->name));
387125fc 3153 device->nobarriers = 1;
5af3e8cc 3154 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3155 ret = -EIO;
5af3e8cc
SB		 3156 btrfs_dev_stat_inc_and_print(device,
3157 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3158 }
3159
3160 /* drop the reference from the wait == 0 run */
3161 bio_put(bio);
3162 device->flush_bio = NULL;
3163
3164 return ret;
3165 }
3166
3167 /*
3168 * one reference for us, and we leave it for the
3169 * caller
3170 */
9c017abc 3171 device->flush_bio = NULL;
9be3395b 3172 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
387125fc
CM		 3173 if (!bio)
3174 return -ENOMEM;
3175
3176 bio->bi_end_io = btrfs_end_empty_barrier;
3177 bio->bi_bdev = device->bdev;
3178 init_completion(&device->flush_wait);
3179 bio->bi_private = &device->flush_wait;
3180 device->flush_bio = bio;
3181
3182 bio_get(bio);
21adbd5c 3183 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3184
3185 return 0;
3186}
3187
3188/*
3189 * send an empty flush down to each device in parallel,
3190 * then wait for them
3191 */
3192static int barrier_all_devices(struct btrfs_fs_info *info)
3193{
3194 struct list_head *head;
3195 struct btrfs_device *dev;
5af3e8cc
SB		 3196 int errors_send = 0;
3197 int errors_wait = 0;
387125fc
CM		 3198 int ret;
3199
3200 /* send down all the barriers */
3201 head = &info->fs_devices->devices;
3202 list_for_each_entry_rcu(dev, head, dev_list) {
3203 if (!dev->bdev) {
5af3e8cc 3204 errors_send++;
387125fc
CM		 3205 continue;
3206 }
3207 if (!dev->in_fs_metadata || !dev->writeable)
3208 continue;
3209
3210 ret = write_dev_flush(dev, 0);
3211 if (ret)
5af3e8cc 3212 errors_send++;
387125fc
CM		 3213 }
3214
3215 /* wait for all the barriers */
3216 list_for_each_entry_rcu(dev, head, dev_list) {
3217 if (!dev->bdev) {
5af3e8cc 3218 errors_wait++;
387125fc
CM		 3219 continue;
3220 }
3221 if (!dev->in_fs_metadata || !dev->writeable)
3222 continue;
3223
3224 ret = write_dev_flush(dev, 1);
3225 if (ret)
5af3e8cc 3226 errors_wait++;
387125fc 3227 }
5af3e8cc
SB		 3228 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3229 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3230 return -EIO;
3231 return 0;
3232}
3233
5af3e8cc
SB		 3234int btrfs_calc_num_tolerated_disk_barrier_failures(
3235 struct btrfs_fs_info *fs_info)
3236{
3237 struct btrfs_ioctl_space_info space;
3238 struct btrfs_space_info *sinfo;
3239 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3240 BTRFS_BLOCK_GROUP_SYSTEM,
3241 BTRFS_BLOCK_GROUP_METADATA,
3242 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3243 int num_types = 4;
3244 int i;
3245 int c;
3246 int num_tolerated_disk_barrier_failures =
3247 (int)fs_info->fs_devices->num_devices;
3248
3249 for (i = 0; i < num_types; i++) {
3250 struct btrfs_space_info *tmp;
3251
3252 sinfo = NULL;
3253 rcu_read_lock();
3254 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3255 if (tmp->flags == types[i]) {
3256 sinfo = tmp;
3257 break;
3258 }
3259 }
3260 rcu_read_unlock();
3261
3262 if (!sinfo)
3263 continue;
3264
3265 down_read(&sinfo->groups_sem);
3266 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3267 if (!list_empty(&sinfo->block_groups[c])) {
3268 u64 flags;
3269
3270 btrfs_get_block_group_info(
3271 &sinfo->block_groups[c], &space);
3272 if (space.total_bytes == 0 ||
3273 space.used_bytes == 0)
3274 continue;
3275 flags = space.flags;
3276 /*
3277 * return
3278 * 0: if dup, single or RAID0 is configured for
3279 * any of metadata, system or data, else
3280 * 1: if RAID5 is configured, or if RAID1 or
3281 * RAID10 is configured and only two mirrors
3282 * are used, else
3283 * 2: if RAID6 is configured, else
3284 * num_mirrors - 1: if RAID1 or RAID10 is
3285 * configured and more than
3286 * 2 mirrors are used.
3287 */
3288 if (num_tolerated_disk_barrier_failures > 0 &&
3289 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3290 BTRFS_BLOCK_GROUP_RAID0)) ||
3291 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3292 == 0)))
3293 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3294 else if (num_tolerated_disk_barrier_failures > 1) {
3295 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3296 BTRFS_BLOCK_GROUP_RAID5 |
3297 BTRFS_BLOCK_GROUP_RAID10)) {
3298 num_tolerated_disk_barrier_failures = 1;
3299 } else if (flags &
15b0a89d 3300 BTRFS_BLOCK_GROUP_RAID6) {
53b381b3
DW		 3301 num_tolerated_disk_barrier_failures = 2;
3302 }
3303 }
5af3e8cc
SB		 3304 }
3305 }
3306 up_read(&sinfo->groups_sem);
3307 }
3308
3309 return num_tolerated_disk_barrier_failures;
3310}
3311
48a3b636 3312static int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3313{
e5e9a520 3314 struct list_head *head;
f2984462 3315 struct btrfs_device *dev;
a061fc8d 3316 struct btrfs_super_block *sb;
f2984462 3317 struct btrfs_dev_item *dev_item;
f2984462
CM		 3318 int ret;
3319 int do_barriers;
a236aed1
CM		 3320 int max_errors;
3321 int total_errors = 0;
a061fc8d 3322 u64 flags;
f2984462 3323
6c41761f 3324 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
f2984462 3325 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3326 backup_super_roots(root->fs_info);
f2984462 3327
6c41761f 3328 sb = root->fs_info->super_for_commit;
a061fc8d 3329 dev_item = &sb->dev_item;
e5e9a520 3330
174ba509 3331 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3332 head = &root->fs_info->fs_devices->devices;
387125fc 3333
5af3e8cc
SB		 3334 if (do_barriers) {
3335 ret = barrier_all_devices(root->fs_info);
3336 if (ret) {
3337 mutex_unlock(
3338 &root->fs_info->fs_devices->device_list_mutex);
3339 btrfs_error(root->fs_info, ret,
3340 "errors while submitting device barriers.");
3341 return ret;
3342 }
3343 }
387125fc 3344
1f78160c 3345 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3346 if (!dev->bdev) {
3347 total_errors++;
3348 continue;
3349 }
2b82032c 3350 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3351 continue;
3352
2b82032c 3353 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3354 btrfs_set_stack_device_type(dev_item, dev->type);
3355 btrfs_set_stack_device_id(dev_item, dev->devid);
3356 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3357 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3358 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3359 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3360 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3361 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3362 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3363
a061fc8d
CM		 3364 flags = btrfs_super_flags(sb);
3365 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3366
a512bbf8 3367 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3368 if (ret)
3369 total_errors++;
f2984462 3370 }
a236aed1 3371 if (total_errors > max_errors) {
d397712b
CM		 3372 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3373 total_errors);
79787eaa
JM		 3374
3375 /* This shouldn't happen. FUA is masked off if unsupported */
a236aed1
CM		 3376 BUG();
3377 }
f2984462 3378
a512bbf8 3379 total_errors = 0;
1f78160c 3380 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3381 if (!dev->bdev)
3382 continue;
2b82032c 3383 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3384 continue;
3385
a512bbf8
YZ		 3386 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3387 if (ret)
3388 total_errors++;
f2984462 3389 }
174ba509 3390 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3391 if (total_errors > max_errors) {
79787eaa
JM		 3392 btrfs_error(root->fs_info, -EIO,
3393 "%d errors while writing supers", total_errors);
3394 return -EIO;
a236aed1 3395 }
f2984462
CM		 3396 return 0;
3397}
3398
a512bbf8
YZ		 3399int write_ctree_super(struct btrfs_trans_handle *trans,
3400 struct btrfs_root *root, int max_mirrors)
eb60ceac 3401{
e66f709b 3402 int ret;
5f39d397 3403
a512bbf8 3404 ret = write_all_supers(root, max_mirrors);
5f39d397 3405 return ret;
cfaa7295
CM		 3406}
3407
cb517eab
MX		 3408/* Drop a fs root from the radix tree and free it. */
3409void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3410 struct btrfs_root *root)
2619ba1f 3411{
4df27c4d 3412 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3413 radix_tree_delete(&fs_info->fs_roots_radix,
3414 (unsigned long)root->root_key.objectid);
4df27c4d 3415 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3416
3417 if (btrfs_root_refs(&root->root_item) == 0)
3418 synchronize_srcu(&fs_info->subvol_srcu);
3419
d7634482 3420 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3321719e
LB		 3421 btrfs_free_log(NULL, root);
3422 btrfs_free_log_root_tree(NULL, fs_info);
3423 }
3424
581bb050
LZ		 3425 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3426 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3427 free_fs_root(root);
4df27c4d
YZ		 3428}
3429
3430static void free_fs_root(struct btrfs_root *root)
3431{
82d5902d 3432 iput(root->cache_inode);
4df27c4d 3433 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
0ee5dc67
AV		 3434 if (root->anon_dev)
3435 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 3436 free_extent_buffer(root->node);
3437 free_extent_buffer(root->commit_root);
581bb050
LZ		 3438 kfree(root->free_ino_ctl);
3439 kfree(root->free_ino_pinned);
d397712b 3440 kfree(root->name);
b0feb9d9 3441 btrfs_put_fs_root(root);
2619ba1f
CM		 3442}
3443
cb517eab
MX		 3444void btrfs_free_fs_root(struct btrfs_root *root)
3445{
3446 free_fs_root(root);
2619ba1f
CM		 3447}
3448
c146afad 3449int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3450{
c146afad
YZ		 3451 u64 root_objectid = 0;
3452 struct btrfs_root *gang[8];
3453 int i;
3768f368 3454 int ret;
e089f05c 3455
c146afad
YZ		 3456 while (1) {
3457 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3458 (void **)gang, root_objectid,
3459 ARRAY_SIZE(gang));
3460 if (!ret)
3461 break;
5d4f98a2
YZ		 3462
3463 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 3464 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 3465 int err;
3466
c146afad 3467 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3468 err = btrfs_orphan_cleanup(gang[i]);
3469 if (err)
3470 return err;
c146afad
YZ		 3471 }
3472 root_objectid++;
3473 }
3474 return 0;
3475}
a2135011 3476
c146afad
YZ		 3477int btrfs_commit_super(struct btrfs_root *root)
3478{
3479 struct btrfs_trans_handle *trans;
3480 int ret;
a74a4b97 3481
c146afad 3482 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3483 btrfs_run_delayed_iputs(root);
c146afad 3484 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3485 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3486
3487 /* wait until ongoing cleanup work done */
3488 down_write(&root->fs_info->cleanup_work_sem);
3489 up_write(&root->fs_info->cleanup_work_sem);
3490
7a7eaa40 3491 trans = btrfs_join_transaction(root);
3612b495
TI		 3492 if (IS_ERR(trans))
3493 return PTR_ERR(trans);
54aa1f4d 3494 ret = btrfs_commit_transaction(trans, root);
79787eaa
JM		 3495 if (ret)
3496 return ret;
c146afad 3497 /* run commit again to drop the original snapshot */
7a7eaa40 3498 trans = btrfs_join_transaction(root);
3612b495
TI		 3499 if (IS_ERR(trans))
3500 return PTR_ERR(trans);
79787eaa
JM		 3501 ret = btrfs_commit_transaction(trans, root);
3502 if (ret)
3503 return ret;
79154b1b 3504 ret = btrfs_write_and_wait_transaction(NULL, root);
79787eaa
JM		 3505 if (ret) {
3506 btrfs_error(root->fs_info, ret,
3507 "Failed to sync btree inode to disk.");
3508 return ret;
3509 }
d6bfde87 3510
a512bbf8 3511 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 3512 return ret;
3513}
3514
3515int close_ctree(struct btrfs_root *root)
3516{
3517 struct btrfs_fs_info *fs_info = root->fs_info;
3518 int ret;
3519
3520 fs_info->closing = 1;
3521 smp_mb();
3522
837d5b6e 3523 /* pause restriper - we want to resume on mount */
aa1b8cd4 3524 btrfs_pause_balance(fs_info);
837d5b6e 3525
8dabb742
SB		 3526 btrfs_dev_replace_suspend_for_unmount(fs_info);
3527
aa1b8cd4 3528 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3529
3530 /* wait for any defraggers to finish */
3531 wait_event(fs_info->transaction_wait,
3532 (atomic_read(&fs_info->defrag_running) == 0));
3533
3534 /* clear out the rbtree of defraggable inodes */
26176e7c 3535 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3536
c146afad 3537 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3538 ret = btrfs_commit_super(root);
3539 if (ret)
3540 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3541 }
3542
87533c47 3543 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3544 btrfs_error_commit_super(root);
0f7d52f4 3545
300e4f8a
JB		 3546 btrfs_put_block_group_cache(fs_info);
3547
e3029d9f
AV		 3548 kthread_stop(fs_info->transaction_kthread);
3549 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3550
f25784b3
YZ		 3551 fs_info->closing = 2;
3552 smp_mb();
3553
bcef60f2
AJ		 3554 btrfs_free_qgroup_config(root->fs_info);
3555
963d678b
MX		 3556 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3557 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3558 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3559 }
bcc63abb 3560
e3029d9f 3561 btrfs_free_block_groups(fs_info);
d10c5f31 3562
13e6c37b 3563 btrfs_stop_all_workers(fs_info);
2932505a 3564
c146afad 3565 del_fs_roots(fs_info);
d10c5f31 3566
13e6c37b 3567 free_root_pointers(fs_info, 1);
9ad6b7bc 3568
13e6c37b 3569 iput(fs_info->btree_inode);
d6bfde87 3570
21adbd5c
SB		 3571#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3572 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3573 btrfsic_unmount(root, fs_info->fs_devices);
3574#endif
3575
dfe25020 3576 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3577 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3578
e2d84521 3579 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3580 percpu_counter_destroy(&fs_info->delalloc_bytes);
04160088 3581 bdi_destroy(&fs_info->bdi);
76dda93c 3582 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3583
53b381b3
DW		 3584 btrfs_free_stripe_hash_table(fs_info);
3585
eb60ceac
CM		 3586 return 0;
3587}
3588
b9fab919
CM		 3589int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3590 int atomic)
5f39d397 3591{
1259ab75 3592 int ret;
727011e0 3593 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3594
0b32f4bb 3595 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3596 if (!ret)
3597 return ret;
3598
3599 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3600 parent_transid, atomic);
3601 if (ret == -EAGAIN)
3602 return ret;
1259ab75 3603 return !ret;
5f39d397
CM		 3604}
3605
3606int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3607{
0b32f4bb 3608 return set_extent_buffer_uptodate(buf);
5f39d397 3609}
6702ed49 3610
5f39d397
CM		 3611void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3612{
727011e0 3613 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
5f39d397 3614 u64 transid = btrfs_header_generation(buf);
b9473439 3615 int was_dirty;
b4ce94de 3616
b9447ef8 3617 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3618 if (transid != root->fs_info->generation)
3619 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3620 "found %llu running %llu\n",
db94535d 3621 (unsigned long long)buf->start,
d397712b
CM		 3622 (unsigned long long)transid,
3623 (unsigned long long)root->fs_info->generation);
0b32f4bb 3624 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3625 if (!was_dirty)
3626 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3627 buf->len,
3628 root->fs_info->dirty_metadata_batch);
eb60ceac
CM		 3629}
3630
b53d3f5d
LB		 3631static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3632 int flush_delayed)
16cdcec7
MX		 3633{
3634 /*
3635 * looks as though older kernels can get into trouble with
3636 * this code, they end up stuck in balance_dirty_pages forever
3637 */
e2d84521 3638 int ret;
16cdcec7
MX		 3639
3640 if (current->flags & PF_MEMALLOC)
3641 return;
3642
b53d3f5d
LB		 3643 if (flush_delayed)
3644 btrfs_balance_delayed_items(root);
16cdcec7 3645
e2d84521
MX		 3646 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3647 BTRFS_DIRTY_METADATA_THRESH);
3648 if (ret > 0) {
d0e1d66b
NJ		 3649 balance_dirty_pages_ratelimited(
3650 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3651 }
3652 return;
3653}
3654
b53d3f5d 3655void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3656{
b53d3f5d
LB		 3657 __btrfs_btree_balance_dirty(root, 1);
3658}
585ad2c3 3659
b53d3f5d
LB		 3660void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3661{
3662 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3663}
6b80053d 3664
ca7a79ad 3665int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3666{
727011e0 3667 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3668 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3669}
0da5468f 3670
fcd1f065 3671static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3672 int read_only)
3673{
1104a885
DS		 3674 /*
3675 * Placeholder for checks
3676 */
fcd1f065 3677 return 0;
acce952b 3678}
3679
48a3b636 3680static void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3681{
acce952b 3682 mutex_lock(&root->fs_info->cleaner_mutex);
3683 btrfs_run_delayed_iputs(root);
3684 mutex_unlock(&root->fs_info->cleaner_mutex);
3685
3686 down_write(&root->fs_info->cleanup_work_sem);
3687 up_write(&root->fs_info->cleanup_work_sem);
3688
3689 /* cleanup FS via transaction */
3690 btrfs_cleanup_transaction(root);
acce952b 3691}
3692
569e0f35
JB		 3693static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3694 struct btrfs_root *root)
acce952b 3695{
3696 struct btrfs_inode *btrfs_inode;
3697 struct list_head splice;
3698
3699 INIT_LIST_HEAD(&splice);
3700
3701 mutex_lock(&root->fs_info->ordered_operations_mutex);
199c2a9c 3702 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3703
569e0f35 3704 list_splice_init(&t->ordered_operations, &splice);
acce952b 3705 while (!list_empty(&splice)) {
3706 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3707 ordered_operations);
3708
3709 list_del_init(&btrfs_inode->ordered_operations);
199c2a9c 3710 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3711
3712 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3713
199c2a9c 3714 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3715 }
3716
199c2a9c 3717 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3718 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3719}
3720
143bede5 3721static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3722{
acce952b 3723 struct btrfs_ordered_extent *ordered;
acce952b 3724
199c2a9c 3725 spin_lock(&root->ordered_extent_lock);
779880ef
JB		 3726 /*
3727 * This will just short circuit the ordered completion stuff which will
3728 * make sure the ordered extent gets properly cleaned up.
3729 */
199c2a9c 3730 list_for_each_entry(ordered, &root->ordered_extents,
779880ef
JB		 3731 root_extent_list)
3732 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
199c2a9c
MX		 3733 spin_unlock(&root->ordered_extent_lock);
3734}
3735
3736static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3737{
3738 struct btrfs_root *root;
3739 struct list_head splice;
3740
3741 INIT_LIST_HEAD(&splice);
3742
3743 spin_lock(&fs_info->ordered_root_lock);
3744 list_splice_init(&fs_info->ordered_roots, &splice);
3745 while (!list_empty(&splice)) {
3746 root = list_first_entry(&splice, struct btrfs_root,
3747 ordered_root);
3748 list_del_init(&root->ordered_root);
3749
3750 btrfs_destroy_ordered_extents(root);
3751
3752 cond_resched_lock(&fs_info->ordered_root_lock);
3753 }
3754 spin_unlock(&fs_info->ordered_root_lock);
acce952b 3755}
3756
35a3621b
SB		 3757static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3758 struct btrfs_root *root)
acce952b 3759{
3760 struct rb_node *node;
3761 struct btrfs_delayed_ref_root *delayed_refs;
3762 struct btrfs_delayed_ref_node *ref;
3763 int ret = 0;
3764
3765 delayed_refs = &trans->delayed_refs;
3766
3767 spin_lock(&delayed_refs->lock);
3768 if (delayed_refs->num_entries == 0) {
cfece4db 3769 spin_unlock(&delayed_refs->lock);
acce952b 3770 printk(KERN_INFO "delayed_refs has NO entry\n");
3771 return ret;
3772 }
3773
b939d1ab 3774 while ((node = rb_first(&delayed_refs->root)) != NULL) {
eb12db69 3775 struct btrfs_delayed_ref_head *head = NULL;
e78417d1 3776 bool pin_bytes = false;
acce952b 3777
eb12db69 3778 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
acce952b 3779 atomic_set(&ref->refs, 1);
3780 if (btrfs_delayed_ref_is_head(ref)) {
acce952b 3781
3782 head = btrfs_delayed_node_to_head(ref);
b939d1ab
JB		 3783 if (!mutex_trylock(&head->mutex)) {
3784 atomic_inc(&ref->refs);
3785 spin_unlock(&delayed_refs->lock);
3786
3787 /* Need to wait for the delayed ref to run */
3788 mutex_lock(&head->mutex);
3789 mutex_unlock(&head->mutex);
3790 btrfs_put_delayed_ref(ref);
3791
e18fca73 3792 spin_lock(&delayed_refs->lock);
b939d1ab
JB		 3793 continue;
3794 }
3795
54067ae9 3796 if (head->must_insert_reserved)
e78417d1 3797 pin_bytes = true;
78a6184a 3798 btrfs_free_delayed_extent_op(head->extent_op);
acce952b 3799 delayed_refs->num_heads--;
3800 if (list_empty(&head->cluster))
3801 delayed_refs->num_heads_ready--;
3802 list_del_init(&head->cluster);
acce952b 3803 }
eb12db69 3804
b939d1ab
JB		 3805 ref->in_tree = 0;
3806 rb_erase(&ref->rb_node, &delayed_refs->root);
3807 delayed_refs->num_entries--;
acce952b 3808 spin_unlock(&delayed_refs->lock);
e78417d1
JB		 3809 if (head) {
3810 if (pin_bytes)
3811 btrfs_pin_extent(root, ref->bytenr,
3812 ref->num_bytes, 1);
3813 mutex_unlock(&head->mutex);
3814 }
acce952b 3815 btrfs_put_delayed_ref(ref);
3816
3817 cond_resched();
3818 spin_lock(&delayed_refs->lock);
3819 }
3820
3821 spin_unlock(&delayed_refs->lock);
3822
3823 return ret;
3824}
3825
aec8030a 3826static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
acce952b 3827{
3828 struct btrfs_pending_snapshot *snapshot;
3829 struct list_head splice;
3830
3831 INIT_LIST_HEAD(&splice);
3832
3833 list_splice_init(&t->pending_snapshots, &splice);
3834
3835 while (!list_empty(&splice)) {
3836 snapshot = list_entry(splice.next,
3837 struct btrfs_pending_snapshot,
3838 list);
aec8030a 3839 snapshot->error = -ECANCELED;
acce952b 3840 list_del_init(&snapshot->list);
acce952b 3841 }
acce952b 3842}
3843
143bede5 3844static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3845{
3846 struct btrfs_inode *btrfs_inode;
3847 struct list_head splice;
3848
3849 INIT_LIST_HEAD(&splice);
3850
eb73c1b7
MX		 3851 spin_lock(&root->delalloc_lock);
3852 list_splice_init(&root->delalloc_inodes, &splice);
acce952b 3853
3854 while (!list_empty(&splice)) {
eb73c1b7
MX		 3855 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3856 delalloc_inodes);
acce952b 3857
3858 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3859 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3860 &btrfs_inode->runtime_flags);
eb73c1b7 3861 spin_unlock(&root->delalloc_lock);
acce952b 3862
3863 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3864
eb73c1b7 3865 spin_lock(&root->delalloc_lock);
acce952b 3866 }
3867
eb73c1b7
MX		 3868 spin_unlock(&root->delalloc_lock);
3869}
3870
3871static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3872{
3873 struct btrfs_root *root;
3874 struct list_head splice;
3875
3876 INIT_LIST_HEAD(&splice);
3877
3878 spin_lock(&fs_info->delalloc_root_lock);
3879 list_splice_init(&fs_info->delalloc_roots, &splice);
3880 while (!list_empty(&splice)) {
3881 root = list_first_entry(&splice, struct btrfs_root,
3882 delalloc_root);
3883 list_del_init(&root->delalloc_root);
3884 root = btrfs_grab_fs_root(root);
3885 BUG_ON(!root);
3886 spin_unlock(&fs_info->delalloc_root_lock);
3887
3888 btrfs_destroy_delalloc_inodes(root);
3889 btrfs_put_fs_root(root);
3890
3891 spin_lock(&fs_info->delalloc_root_lock);
3892 }
3893 spin_unlock(&fs_info->delalloc_root_lock);
acce952b 3894}
3895
3896static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3897 struct extent_io_tree *dirty_pages,
3898 int mark)
3899{
3900 int ret;
acce952b 3901 struct extent_buffer *eb;
3902 u64 start = 0;
3903 u64 end;
acce952b 3904
3905 while (1) {
3906 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3907 mark, NULL);
acce952b 3908 if (ret)
3909 break;
3910
3911 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3912 while (start <= end) {
fd8b2b61
JB		 3913 eb = btrfs_find_tree_block(root, start,
3914 root->leafsize);
69a85bd8 3915 start += root->leafsize;
fd8b2b61 3916 if (!eb)
acce952b 3917 continue;
fd8b2b61 3918 wait_on_extent_buffer_writeback(eb);
acce952b 3919
fd8b2b61
JB		 3920 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3921 &eb->bflags))
3922 clear_extent_buffer_dirty(eb);
3923 free_extent_buffer_stale(eb);
acce952b 3924 }
3925 }
3926
3927 return ret;
3928}
3929
3930static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3931 struct extent_io_tree *pinned_extents)
3932{
3933 struct extent_io_tree *unpin;
3934 u64 start;
3935 u64 end;
3936 int ret;
ed0eaa14 3937 bool loop = true;
acce952b 3938
3939 unpin = pinned_extents;
ed0eaa14 3940again:
acce952b 3941 while (1) {
3942 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 3943 EXTENT_DIRTY, NULL);
acce952b 3944 if (ret)
3945 break;
3946
3947 /* opt_discard */
5378e607
LD		 3948 if (btrfs_test_opt(root, DISCARD))
3949 ret = btrfs_error_discard_extent(root, start,
3950 end + 1 - start,
3951 NULL);
acce952b 3952
3953 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3954 btrfs_error_unpin_extent_range(root, start, end);
3955 cond_resched();
3956 }
3957
ed0eaa14
LB		 3958 if (loop) {
3959 if (unpin == &root->fs_info->freed_extents[0])
3960 unpin = &root->fs_info->freed_extents[1];
3961 else
3962 unpin = &root->fs_info->freed_extents[0];
3963 loop = false;
3964 goto again;
3965 }
3966
acce952b 3967 return 0;
3968}
3969
49b25e05
JM		 3970void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3971 struct btrfs_root *root)
3972{
3973 btrfs_destroy_delayed_refs(cur_trans, root);
3974 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3975 cur_trans->dirty_pages.dirty_bytes);
3976
4a9d8bde 3977 cur_trans->state = TRANS_STATE_COMMIT_START;
d7096fc3 3978 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 3979
aec8030a
MX		 3980 btrfs_evict_pending_snapshots(cur_trans);
3981
4a9d8bde 3982 cur_trans->state = TRANS_STATE_UNBLOCKED;
d7096fc3 3983 wake_up(&root->fs_info->transaction_wait);
49b25e05 3984
67cde344
MX		 3985 btrfs_destroy_delayed_inodes(root);
3986 btrfs_assert_delayed_root_empty(root);
49b25e05 3987
49b25e05
JM		 3988 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3989 EXTENT_DIRTY);
6e841e32
LB		 3990 btrfs_destroy_pinned_extent(root,
3991 root->fs_info->pinned_extents);
49b25e05 3992
4a9d8bde
MX		 3993 cur_trans->state =TRANS_STATE_COMPLETED;
3994 wake_up(&cur_trans->commit_wait);
3995
49b25e05
JM		 3996 /*
3997 memset(cur_trans, 0, sizeof(*cur_trans));
3998 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3999 */
4000}
4001
48a3b636 4002static int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 4003{
4004 struct btrfs_transaction *t;
4005 LIST_HEAD(list);
4006
acce952b 4007 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4008
a4abeea4 4009 spin_lock(&root->fs_info->trans_lock);
acce952b 4010 list_splice_init(&root->fs_info->trans_list, &list);
ac673879 4011 root->fs_info->running_transaction = NULL;
a4abeea4
JB		 4012 spin_unlock(&root->fs_info->trans_lock);
4013
acce952b 4014 while (!list_empty(&list)) {
4015 t = list_entry(list.next, struct btrfs_transaction, list);
acce952b 4016
569e0f35 4017 btrfs_destroy_ordered_operations(t, root);
acce952b 4018
199c2a9c 4019 btrfs_destroy_all_ordered_extents(root->fs_info);
acce952b 4020
4021 btrfs_destroy_delayed_refs(t, root);
4022
4a9d8bde
MX		 4023 /*
4024 * FIXME: cleanup wait for commit
4025 * We needn't acquire the lock here, because we are during
4026 * the umount, there is no other task which will change it.
4027 */
4028 t->state = TRANS_STATE_COMMIT_START;
66657b31 4029 smp_mb();
acce952b 4030 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4031 wake_up(&root->fs_info->transaction_blocked_wait);
4032
aec8030a
MX		 4033 btrfs_evict_pending_snapshots(t);
4034
4a9d8bde 4035 t->state = TRANS_STATE_UNBLOCKED;
66657b31 4036 smp_mb();
acce952b 4037 if (waitqueue_active(&root->fs_info->transaction_wait))
4038 wake_up(&root->fs_info->transaction_wait);
acce952b 4039
67cde344
MX		 4040 btrfs_destroy_delayed_inodes(root);
4041 btrfs_assert_delayed_root_empty(root);
4042
eb73c1b7 4043 btrfs_destroy_all_delalloc_inodes(root->fs_info);
acce952b 4044
4045 btrfs_destroy_marked_extents(root, &t->dirty_pages,
4046 EXTENT_DIRTY);
4047
4048 btrfs_destroy_pinned_extent(root,
4049 root->fs_info->pinned_extents);
4050
4a9d8bde
MX		 4051 t->state = TRANS_STATE_COMPLETED;
4052 smp_mb();
4053 if (waitqueue_active(&t->commit_wait))
4054 wake_up(&t->commit_wait);
4055
13c5a93e 4056 atomic_set(&t->use_count, 0);
acce952b 4057 list_del_init(&t->list);
4058 memset(t, 0, sizeof(*t));
4059 kmem_cache_free(btrfs_transaction_cachep, t);
4060 }
4061
4062 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 4063
4064 return 0;
4065}
4066
d1310b2e 4067static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 4068 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 4069 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 4070 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 4071 /* note we're sharing with inode.c for the merge bio hook */
4072 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 4073};
Linux 6.x block layer and io_uring tree(s)