projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: introduce grab/put functions for the root of the fs/file tree
[linux-2.6-block.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
163e783e 29#include <linux/crc32c.h>
5a0e3ad6 30#include <linux/slab.h>
784b4e29 31#include <linux/migrate.h>
7a36ddec 32#include <linux/ratelimit.h>
6463fe58 33#include <linux/uuid.h>
7e75bf3f 34#include <asm/unaligned.h>
4b4e25f2 35#include "compat.h"
eb60ceac
CM		 36#include "ctree.h"
37#include "disk-io.h"
e089f05c 38#include "transaction.h"
0f7d52f4 39#include "btrfs_inode.h"
0b86a832 40#include "volumes.h"
db94535d 41#include "print-tree.h"
8b712842 42#include "async-thread.h"
925baedd 43#include "locking.h"
e02119d5 44#include "tree-log.h"
fa9c0d79 45#include "free-space-cache.h"
581bb050 46#include "inode-map.h"
21adbd5c 47#include "check-integrity.h"
606686ee 48#include "rcu-string.h"
8dabb742 49#include "dev-replace.h"
53b381b3 50#include "raid56.h"
eb60ceac 51
de0022b9
JB		 52#ifdef CONFIG_X86
53#include <asm/cpufeature.h>
54#endif
55
d1310b2e 56static struct extent_io_ops btree_extent_io_ops;
8b712842 57static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 58static void free_fs_root(struct btrfs_root *root);
fcd1f065 59static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 60 int read_only);
569e0f35
JB		 61static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
62 struct btrfs_root *root);
143bede5 63static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
acce952b 64static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
65 struct btrfs_root *root);
aec8030a 66static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
143bede5 67static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
acce952b 68static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 struct extent_io_tree *dirty_pages,
70 int mark);
71static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 struct extent_io_tree *pinned_extents);
48a3b636
ES		 73static int btrfs_cleanup_transaction(struct btrfs_root *root);
74static void btrfs_error_commit_super(struct btrfs_root *root);
ce9adaa5 75
d352ac68
CM		 76/*
77 * end_io_wq structs are used to do processing in task context when an IO is
78 * complete. This is used during reads to verify checksums, and it is used
79 * by writes to insert metadata for new file extents after IO is complete.
80 */
ce9adaa5
CM		 81struct end_io_wq {
82 struct bio *bio;
83 bio_end_io_t *end_io;
84 void *private;
85 struct btrfs_fs_info *info;
86 int error;
22c59948 87 int metadata;
ce9adaa5 88 struct list_head list;
8b712842 89 struct btrfs_work work;
ce9adaa5 90};
0da5468f 91
d352ac68
CM		 92/*
93 * async submit bios are used to offload expensive checksumming
94 * onto the worker threads. They checksum file and metadata bios
95 * just before they are sent down the IO stack.
96 */
44b8bd7e
CM		 97struct async_submit_bio {
98 struct inode *inode;
99 struct bio *bio;
100 struct list_head list;
4a69a410
CM		 101 extent_submit_bio_hook_t *submit_bio_start;
102 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 103 int rw;
104 int mirror_num;
c8b97818 105 unsigned long bio_flags;
eaf25d93
CM		 106 /*
107 * bio_offset is optional, can be used if the pages in the bio
108 * can't tell us where in the file the bio should go
109 */
110 u64 bio_offset;
8b712842 111 struct btrfs_work work;
79787eaa 112 int error;
44b8bd7e
CM		 113};
114
85d4e461
CM		 115/*
116 * Lockdep class keys for extent_buffer->lock's in this root. For a given
117 * eb, the lockdep key is determined by the btrfs_root it belongs to and
118 * the level the eb occupies in the tree.
119 *
120 * Different roots are used for different purposes and may nest inside each
121 * other and they require separate keysets. As lockdep keys should be
122 * static, assign keysets according to the purpose of the root as indicated
123 * by btrfs_root->objectid. This ensures that all special purpose roots
124 * have separate keysets.
4008c04a 125 *
85d4e461
CM		 126 * Lock-nesting across peer nodes is always done with the immediate parent
127 * node locked thus preventing deadlock. As lockdep doesn't know this, use
128 * subclass to avoid triggering lockdep warning in such cases.
4008c04a 129 *
85d4e461
CM		 130 * The key is set by the readpage_end_io_hook after the buffer has passed
131 * csum validation but before the pages are unlocked. It is also set by
132 * btrfs_init_new_buffer on freshly allocated blocks.
4008c04a 133 *
85d4e461
CM		 134 * We also add a check to make sure the highest level of the tree is the
135 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
136 * needs update as well.
4008c04a
CM		 137 */
138#ifdef CONFIG_DEBUG_LOCK_ALLOC
139# if BTRFS_MAX_LEVEL != 8
140# error
141# endif
85d4e461
CM		 142
143static struct btrfs_lockdep_keyset {
144 u64 id; /* root objectid */
145 const char *name_stem; /* lock name stem */
146 char names[BTRFS_MAX_LEVEL + 1][20];
147 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
148} btrfs_lockdep_keysets[] = {
149 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
150 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
151 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
152 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
153 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
154 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
60b62978 155 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
85d4e461
CM		 156 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
157 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
158 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
159 { .id = 0, .name_stem = "tree" },
4008c04a 160};
85d4e461
CM		 161
162void __init btrfs_init_lockdep(void)
163{
164 int i, j;
165
166 /* initialize lockdep class names */
167 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
168 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
169
170 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
171 snprintf(ks->names[j], sizeof(ks->names[j]),
172 "btrfs-%s-%02d", ks->name_stem, j);
173 }
174}
175
176void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
177 int level)
178{
179 struct btrfs_lockdep_keyset *ks;
180
181 BUG_ON(level >= ARRAY_SIZE(ks->keys));
182
183 /* find the matching keyset, id 0 is the default entry */
184 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
185 if (ks->id == objectid)
186 break;
187
188 lockdep_set_class_and_name(&eb->lock,
189 &ks->keys[level], ks->names[level]);
190}
191
4008c04a
CM		 192#endif
193
d352ac68
CM		 194/*
195 * extents on the btree inode are pretty simple, there's one extent
196 * that covers the entire device
197 */
b2950863 198static struct extent_map *btree_get_extent(struct inode *inode,
306e16ce 199 struct page *page, size_t pg_offset, u64 start, u64 len,
b2950863 200 int create)
7eccb903 201{
5f39d397
CM		 202 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
203 struct extent_map *em;
204 int ret;
205
890871be 206 read_lock(&em_tree->lock);
d1310b2e 207 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 208 if (em) {
209 em->bdev =
210 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 211 read_unlock(&em_tree->lock);
5f39d397 212 goto out;
a061fc8d 213 }
890871be 214 read_unlock(&em_tree->lock);
7b13b7b1 215
172ddd60 216 em = alloc_extent_map();
5f39d397
CM		 217 if (!em) {
218 em = ERR_PTR(-ENOMEM);
219 goto out;
220 }
221 em->start = 0;
0afbaf8c 222 em->len = (u64)-1;
c8b97818 223 em->block_len = (u64)-1;
5f39d397 224 em->block_start = 0;
a061fc8d 225 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 226
890871be 227 write_lock(&em_tree->lock);
09a2a8f9 228 ret = add_extent_mapping(em_tree, em, 0);
5f39d397
CM		 229 if (ret == -EEXIST) {
230 free_extent_map(em);
7b13b7b1 231 em = lookup_extent_mapping(em_tree, start, len);
b4f359ab 232 if (!em)
0433f20d 233 em = ERR_PTR(-EIO);
5f39d397 234 } else if (ret) {
7b13b7b1 235 free_extent_map(em);
0433f20d 236 em = ERR_PTR(ret);
5f39d397 237 }
890871be 238 write_unlock(&em_tree->lock);
7b13b7b1 239
5f39d397
CM		 240out:
241 return em;
7eccb903
CM		 242}
243
b0496686 244u32 btrfs_csum_data(char *data, u32 seed, size_t len)
19c00ddc 245{
163e783e 246 return crc32c(seed, data, len);
19c00ddc
CM		 247}
248
249void btrfs_csum_final(u32 crc, char *result)
250{
7e75bf3f 251 put_unaligned_le32(~crc, result);
19c00ddc
CM		 252}
253
d352ac68
CM		 254/*
255 * compute the csum for a btree block, and either verify it or write it
256 * into the csum field of the block.
257 */
19c00ddc
CM		 258static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
259 int verify)
260{
6c41761f 261 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
607d432d 262 char *result = NULL;
19c00ddc
CM		 263 unsigned long len;
264 unsigned long cur_len;
265 unsigned long offset = BTRFS_CSUM_SIZE;
19c00ddc
CM		 266 char *kaddr;
267 unsigned long map_start;
268 unsigned long map_len;
269 int err;
270 u32 crc = ~(u32)0;
607d432d 271 unsigned long inline_result;
19c00ddc
CM		 272
273 len = buf->len - offset;
d397712b 274 while (len > 0) {
19c00ddc 275 err = map_private_extent_buffer(buf, offset, 32,
a6591715 276 &kaddr, &map_start, &map_len);
d397712b 277 if (err)
19c00ddc 278 return 1;
19c00ddc 279 cur_len = min(len, map_len - (offset - map_start));
b0496686 280 crc = btrfs_csum_data(kaddr + offset - map_start,
19c00ddc
CM		 281 crc, cur_len);
282 len -= cur_len;
283 offset += cur_len;
19c00ddc 284 }
607d432d
JB		 285 if (csum_size > sizeof(inline_result)) {
286 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
287 if (!result)
288 return 1;
289 } else {
290 result = (char *)&inline_result;
291 }
292
19c00ddc
CM		 293 btrfs_csum_final(crc, result);
294
295 if (verify) {
607d432d 296 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 297 u32 val;
298 u32 found = 0;
607d432d 299 memcpy(&found, result, csum_size);
e4204ded 300
607d432d 301 read_extent_buffer(buf, &val, 0, csum_size);
7a36ddec 302 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
193f284d
CM		 303 "failed on %llu wanted %X found %X "
304 "level %d\n",
305 root->fs_info->sb->s_id,
306 (unsigned long long)buf->start, val, found,
307 btrfs_header_level(buf));
607d432d
JB		 308 if (result != (char *)&inline_result)
309 kfree(result);
19c00ddc
CM		 310 return 1;
311 }
312 } else {
607d432d 313 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 314 }
607d432d
JB		 315 if (result != (char *)&inline_result)
316 kfree(result);
19c00ddc
CM		 317 return 0;
318}
319
d352ac68
CM		 320/*
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
325 */
1259ab75 326static int verify_parent_transid(struct extent_io_tree *io_tree,
b9fab919
CM		 327 struct extent_buffer *eb, u64 parent_transid,
328 int atomic)
1259ab75 329{
2ac55d41 330 struct extent_state *cached_state = NULL;
1259ab75
CM		 331 int ret;
332
333 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334 return 0;
335
b9fab919
CM		 336 if (atomic)
337 return -EAGAIN;
338
2ac55d41 339 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
d0082371 340 0, &cached_state);
0b32f4bb 341 if (extent_buffer_uptodate(eb) &&
1259ab75
CM		 342 btrfs_header_generation(eb) == parent_transid) {
343 ret = 0;
344 goto out;
345 }
7a36ddec 346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
193f284d
CM		 347 "found %llu\n",
348 (unsigned long long)eb->start,
349 (unsigned long long)parent_transid,
350 (unsigned long long)btrfs_header_generation(eb));
1259ab75 351 ret = 1;
0b32f4bb 352 clear_extent_buffer_uptodate(eb);
33958dc6 353out:
2ac55d41
JB		 354 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
355 &cached_state, GFP_NOFS);
1259ab75 356 return ret;
1259ab75
CM		 357}
358
1104a885
DS		 359/*
360 * Return 0 if the superblock checksum type matches the checksum value of that
361 * algorithm. Pass the raw disk superblock data.
362 */
363static int btrfs_check_super_csum(char *raw_disk_sb)
364{
365 struct btrfs_super_block *disk_sb =
366 (struct btrfs_super_block *)raw_disk_sb;
367 u16 csum_type = btrfs_super_csum_type(disk_sb);
368 int ret = 0;
369
370 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
371 u32 crc = ~(u32)0;
372 const int csum_size = sizeof(crc);
373 char result[csum_size];
374
375 /*
376 * The super_block structure does not span the whole
377 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
378 * is filled with zeros and is included in the checkum.
379 */
380 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
381 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
382 btrfs_csum_final(crc, result);
383
384 if (memcmp(raw_disk_sb, result, csum_size))
385 ret = 1;
667e7d94
CM		 386
387 if (ret && btrfs_super_generation(disk_sb) < 10) {
388 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
389 ret = 0;
390 }
1104a885
DS		 391 }
392
393 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
394 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
395 csum_type);
396 ret = 1;
397 }
398
399 return ret;
400}
401
d352ac68
CM		 402/*
403 * helper to read a given tree block, doing retries as required when
404 * the checksums don't match and we have alternate mirrors to try.
405 */
f188591e
CM		 406static int btree_read_extent_buffer_pages(struct btrfs_root *root,
407 struct extent_buffer *eb,
ca7a79ad 408 u64 start, u64 parent_transid)
f188591e
CM		 409{
410 struct extent_io_tree *io_tree;
ea466794 411 int failed = 0;
f188591e
CM		 412 int ret;
413 int num_copies = 0;
414 int mirror_num = 0;
ea466794 415 int failed_mirror = 0;
f188591e 416
a826d6dc 417 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM		 418 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
419 while (1) {
bb82ab88
AJ		 420 ret = read_extent_buffer_pages(io_tree, eb, start,
421 WAIT_COMPLETE,
f188591e 422 btree_get_extent, mirror_num);
256dd1bb
SB		 423 if (!ret) {
424 if (!verify_parent_transid(io_tree, eb,
b9fab919 425 parent_transid, 0))
256dd1bb
SB		 426 break;
427 else
428 ret = -EIO;
429 }
d397712b 430
a826d6dc
JB		 431 /*
432 * This buffer's crc is fine, but its contents are corrupted, so
433 * there is no reason to read the other copies, they won't be
434 * any less wrong.
435 */
436 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
ea466794
JB		 437 break;
438
5d964051 439 num_copies = btrfs_num_copies(root->fs_info,
f188591e 440 eb->start, eb->len);
4235298e 441 if (num_copies == 1)
ea466794 442 break;
4235298e 443
5cf1ab56
JB		 444 if (!failed_mirror) {
445 failed = 1;
446 failed_mirror = eb->read_mirror;
447 }
448
f188591e 449 mirror_num++;
ea466794
JB		 450 if (mirror_num == failed_mirror)
451 mirror_num++;
452
4235298e 453 if (mirror_num > num_copies)
ea466794 454 break;
f188591e 455 }
ea466794 456
c0901581 457 if (failed && !ret && failed_mirror)
ea466794
JB		 458 repair_eb_io_failure(root, eb, failed_mirror);
459
460 return ret;
f188591e 461}
19c00ddc 462
d352ac68 463/*
d397712b
CM		 464 * checksum a dirty tree block before IO. This has extra checks to make sure
465 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 466 */
d397712b 467
b2950863 468static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 469{
d1310b2e 470 struct extent_io_tree *tree;
4eee4fa4 471 u64 start = page_offset(page);
19c00ddc 472 u64 found_start;
19c00ddc 473 struct extent_buffer *eb;
f188591e 474
d1310b2e 475 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc 476
4f2de97a
JB		 477 eb = (struct extent_buffer *)page->private;
478 if (page != eb->pages[0])
479 return 0;
19c00ddc
CM		 480 found_start = btrfs_header_bytenr(eb);
481 if (found_start != start) {
55c69072 482 WARN_ON(1);
4f2de97a 483 return 0;
55c69072 484 }
55c69072 485 if (!PageUptodate(page)) {
55c69072 486 WARN_ON(1);
4f2de97a 487 return 0;
19c00ddc 488 }
19c00ddc 489 csum_tree_block(root, eb, 0);
19c00ddc
CM		 490 return 0;
491}
492
2b82032c
YZ		 493static int check_tree_block_fsid(struct btrfs_root *root,
494 struct extent_buffer *eb)
495{
496 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
497 u8 fsid[BTRFS_UUID_SIZE];
498 int ret = 1;
499
500 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
501 BTRFS_FSID_SIZE);
502 while (fs_devices) {
503 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
504 ret = 0;
505 break;
506 }
507 fs_devices = fs_devices->seed;
508 }
509 return ret;
510}
511
a826d6dc
JB		 512#define CORRUPT(reason, eb, root, slot) \
513 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
514 "root=%llu, slot=%d\n", reason, \
515 (unsigned long long)btrfs_header_bytenr(eb), \
516 (unsigned long long)root->objectid, slot)
517
518static noinline int check_leaf(struct btrfs_root *root,
519 struct extent_buffer *leaf)
520{
521 struct btrfs_key key;
522 struct btrfs_key leaf_key;
523 u32 nritems = btrfs_header_nritems(leaf);
524 int slot;
525
526 if (nritems == 0)
527 return 0;
528
529 /* Check the 0 item */
530 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
531 BTRFS_LEAF_DATA_SIZE(root)) {
532 CORRUPT("invalid item offset size pair", leaf, root, 0);
533 return -EIO;
534 }
535
536 /*
537 * Check to make sure each items keys are in the correct order and their
538 * offsets make sense. We only have to loop through nritems-1 because
539 * we check the current slot against the next slot, which verifies the
540 * next slot's offset+size makes sense and that the current's slot
541 * offset is correct.
542 */
543 for (slot = 0; slot < nritems - 1; slot++) {
544 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
545 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
546
547 /* Make sure the keys are in the right order */
548 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
549 CORRUPT("bad key order", leaf, root, slot);
550 return -EIO;
551 }
552
553 /*
554 * Make sure the offset and ends are right, remember that the
555 * item data starts at the end of the leaf and grows towards the
556 * front.
557 */
558 if (btrfs_item_offset_nr(leaf, slot) !=
559 btrfs_item_end_nr(leaf, slot + 1)) {
560 CORRUPT("slot offset bad", leaf, root, slot);
561 return -EIO;
562 }
563
564 /*
565 * Check to make sure that we don't point outside of the leaf,
566 * just incase all the items are consistent to eachother, but
567 * all point outside of the leaf.
568 */
569 if (btrfs_item_end_nr(leaf, slot) >
570 BTRFS_LEAF_DATA_SIZE(root)) {
571 CORRUPT("slot end outside of leaf", leaf, root, slot);
572 return -EIO;
573 }
574 }
575
576 return 0;
577}
578
b2950863 579static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
5cf1ab56 580 struct extent_state *state, int mirror)
ce9adaa5
CM		 581{
582 struct extent_io_tree *tree;
583 u64 found_start;
584 int found_level;
ce9adaa5
CM		 585 struct extent_buffer *eb;
586 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 587 int ret = 0;
727011e0 588 int reads_done;
ce9adaa5 589
ce9adaa5
CM		 590 if (!page->private)
591 goto out;
d397712b 592
727011e0 593 tree = &BTRFS_I(page->mapping->host)->io_tree;
4f2de97a 594 eb = (struct extent_buffer *)page->private;
d397712b 595
0b32f4bb
JB		 596 /* the pending IO might have been the only thing that kept this buffer
597 * in memory. Make sure we have a ref for all this other checks
598 */
599 extent_buffer_get(eb);
600
601 reads_done = atomic_dec_and_test(&eb->io_pages);
727011e0
CM		 602 if (!reads_done)
603 goto err;
f188591e 604
5cf1ab56 605 eb->read_mirror = mirror;
ea466794
JB		 606 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
607 ret = -EIO;
608 goto err;
609 }
610
ce9adaa5 611 found_start = btrfs_header_bytenr(eb);
727011e0 612 if (found_start != eb->start) {
7a36ddec 613 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
193f284d
CM		 614 "%llu %llu\n",
615 (unsigned long long)found_start,
616 (unsigned long long)eb->start);
f188591e 617 ret = -EIO;
ce9adaa5
CM		 618 goto err;
619 }
2b82032c 620 if (check_tree_block_fsid(root, eb)) {
7a36ddec 621 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
193f284d 622 (unsigned long long)eb->start);
1259ab75
CM		 623 ret = -EIO;
624 goto err;
625 }
ce9adaa5 626 found_level = btrfs_header_level(eb);
1c24c3ce
JB		 627 if (found_level >= BTRFS_MAX_LEVEL) {
628 btrfs_info(root->fs_info, "bad tree block level %d\n",
629 (int)btrfs_header_level(eb));
630 ret = -EIO;
631 goto err;
632 }
ce9adaa5 633
85d4e461
CM		 634 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
635 eb, found_level);
4008c04a 636
ce9adaa5 637 ret = csum_tree_block(root, eb, 1);
a826d6dc 638 if (ret) {
f188591e 639 ret = -EIO;
a826d6dc
JB		 640 goto err;
641 }
642
643 /*
644 * If this is a leaf block and it is corrupt, set the corrupt bit so
645 * that we don't try and read the other copies of this block, just
646 * return -EIO.
647 */
648 if (found_level == 0 && check_leaf(root, eb)) {
649 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
650 ret = -EIO;
651 }
ce9adaa5 652
0b32f4bb
JB		 653 if (!ret)
654 set_extent_buffer_uptodate(eb);
ce9adaa5 655err:
79fb65a1
JB		 656 if (reads_done &&
657 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 658 btree_readahead_hook(root, eb, eb->start, ret);
4bb31e92 659
53b381b3
DW		 660 if (ret) {
661 /*
662 * our io error hook is going to dec the io pages
663 * again, we have to make sure it has something
664 * to decrement
665 */
666 atomic_inc(&eb->io_pages);
0b32f4bb 667 clear_extent_buffer_uptodate(eb);
53b381b3 668 }
0b32f4bb 669 free_extent_buffer(eb);
ce9adaa5 670out:
f188591e 671 return ret;
ce9adaa5
CM		 672}
673
ea466794 674static int btree_io_failed_hook(struct page *page, int failed_mirror)
4bb31e92 675{
4bb31e92
AJ		 676 struct extent_buffer *eb;
677 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
678
4f2de97a 679 eb = (struct extent_buffer *)page->private;
ea466794 680 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
5cf1ab56 681 eb->read_mirror = failed_mirror;
53b381b3 682 atomic_dec(&eb->io_pages);
ea466794 683 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 684 btree_readahead_hook(root, eb, eb->start, -EIO);
4bb31e92
AJ		 685 return -EIO; /* we fixed nothing */
686}
687
ce9adaa5 688static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 689{
690 struct end_io_wq *end_io_wq = bio->bi_private;
691 struct btrfs_fs_info *fs_info;
ce9adaa5 692
ce9adaa5 693 fs_info = end_io_wq->info;
ce9adaa5 694 end_io_wq->error = err;
8b712842
CM		 695 end_io_wq->work.func = end_workqueue_fn;
696 end_io_wq->work.flags = 0;
d20f7043 697
7b6d91da 698 if (bio->bi_rw & REQ_WRITE) {
53b381b3 699 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
cad321ad
CM		 700 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
701 &end_io_wq->work);
53b381b3 702 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
0cb59c99
JB		 703 btrfs_queue_worker(&fs_info->endio_freespace_worker,
704 &end_io_wq->work);
53b381b3
DW		 705 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
706 btrfs_queue_worker(&fs_info->endio_raid56_workers,
707 &end_io_wq->work);
cad321ad
CM		 708 else
709 btrfs_queue_worker(&fs_info->endio_write_workers,
710 &end_io_wq->work);
d20f7043 711 } else {
53b381b3
DW		 712 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
713 btrfs_queue_worker(&fs_info->endio_raid56_workers,
714 &end_io_wq->work);
715 else if (end_io_wq->metadata)
d20f7043
CM		 716 btrfs_queue_worker(&fs_info->endio_meta_workers,
717 &end_io_wq->work);
718 else
719 btrfs_queue_worker(&fs_info->endio_workers,
720 &end_io_wq->work);
721 }
ce9adaa5
CM		 722}
723
0cb59c99
JB		 724/*
725 * For the metadata arg you want
726 *
727 * 0 - if data
728 * 1 - if normal metadta
729 * 2 - if writing to the free space cache area
53b381b3 730 * 3 - raid parity work
0cb59c99 731 */
22c59948
CM		 732int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
733 int metadata)
0b86a832 734{
ce9adaa5 735 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 736 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
737 if (!end_io_wq)
738 return -ENOMEM;
739
740 end_io_wq->private = bio->bi_private;
741 end_io_wq->end_io = bio->bi_end_io;
22c59948 742 end_io_wq->info = info;
ce9adaa5
CM		 743 end_io_wq->error = 0;
744 end_io_wq->bio = bio;
22c59948 745 end_io_wq->metadata = metadata;
ce9adaa5
CM		 746
747 bio->bi_private = end_io_wq;
748 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 749 return 0;
750}
751
b64a2851 752unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 753{
4854ddd0
CM		 754 unsigned long limit = min_t(unsigned long,
755 info->workers.max_workers,
756 info->fs_devices->open_devices);
757 return 256 * limit;
758}
0986fe9e 759
4a69a410
CM		 760static void run_one_async_start(struct btrfs_work *work)
761{
4a69a410 762 struct async_submit_bio *async;
79787eaa 763 int ret;
4a69a410
CM		 764
765 async = container_of(work, struct async_submit_bio, work);
79787eaa
JM		 766 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
767 async->mirror_num, async->bio_flags,
768 async->bio_offset);
769 if (ret)
770 async->error = ret;
4a69a410
CM		 771}
772
773static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 774{
775 struct btrfs_fs_info *fs_info;
776 struct async_submit_bio *async;
4854ddd0 777 int limit;
8b712842
CM		 778
779 async = container_of(work, struct async_submit_bio, work);
780 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 781
b64a2851 782 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 783 limit = limit * 2 / 3;
784
66657b31 785 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
b64a2851 786 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 787 wake_up(&fs_info->async_submit_wait);
788
79787eaa
JM		 789 /* If an error occured we just want to clean up the bio and move on */
790 if (async->error) {
791 bio_endio(async->bio, async->error);
792 return;
793 }
794
4a69a410 795 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 796 async->mirror_num, async->bio_flags,
797 async->bio_offset);
4a69a410
CM		 798}
799
800static void run_one_async_free(struct btrfs_work *work)
801{
802 struct async_submit_bio *async;
803
804 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 805 kfree(async);
806}
807
44b8bd7e
CM		 808int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
809 int rw, struct bio *bio, int mirror_num,
c8b97818 810 unsigned long bio_flags,
eaf25d93 811 u64 bio_offset,
4a69a410
CM		 812 extent_submit_bio_hook_t *submit_bio_start,
813 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 814{
815 struct async_submit_bio *async;
816
817 async = kmalloc(sizeof(*async), GFP_NOFS);
818 if (!async)
819 return -ENOMEM;
820
821 async->inode = inode;
822 async->rw = rw;
823 async->bio = bio;
824 async->mirror_num = mirror_num;
4a69a410
CM		 825 async->submit_bio_start = submit_bio_start;
826 async->submit_bio_done = submit_bio_done;
827
828 async->work.func = run_one_async_start;
829 async->work.ordered_func = run_one_async_done;
830 async->work.ordered_free = run_one_async_free;
831
8b712842 832 async->work.flags = 0;
c8b97818 833 async->bio_flags = bio_flags;
eaf25d93 834 async->bio_offset = bio_offset;
8c8bee1d 835
79787eaa
JM		 836 async->error = 0;
837
cb03c743 838 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 839
7b6d91da 840 if (rw & REQ_SYNC)
d313d7a3
CM		 841 btrfs_set_work_high_prio(&async->work);
842
8b712842 843 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 844
d397712b 845 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 846 atomic_read(&fs_info->nr_async_submits)) {
847 wait_event(fs_info->async_submit_wait,
848 (atomic_read(&fs_info->nr_async_submits) == 0));
849 }
850
44b8bd7e
CM		 851 return 0;
852}
853
ce3ed71a
CM		 854static int btree_csum_one_bio(struct bio *bio)
855{
856 struct bio_vec *bvec = bio->bi_io_vec;
857 int bio_index = 0;
858 struct btrfs_root *root;
79787eaa 859 int ret = 0;
ce3ed71a
CM		 860
861 WARN_ON(bio->bi_vcnt <= 0);
d397712b 862 while (bio_index < bio->bi_vcnt) {
ce3ed71a 863 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 864 ret = csum_dirty_buffer(root, bvec->bv_page);
865 if (ret)
866 break;
ce3ed71a
CM		 867 bio_index++;
868 bvec++;
869 }
79787eaa 870 return ret;
ce3ed71a
CM		 871}
872
4a69a410
CM		 873static int __btree_submit_bio_start(struct inode *inode, int rw,
874 struct bio *bio, int mirror_num,
eaf25d93
CM		 875 unsigned long bio_flags,
876 u64 bio_offset)
22c59948 877{
8b712842
CM		 878 /*
879 * when we're called for a write, we're already in the async
5443be45 880 * submission context. Just jump into btrfs_map_bio
8b712842 881 */
79787eaa 882 return btree_csum_one_bio(bio);
4a69a410 883}
22c59948 884
4a69a410 885static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 886 int mirror_num, unsigned long bio_flags,
887 u64 bio_offset)
4a69a410 888{
61891923
SB		 889 int ret;
890
8b712842 891 /*
4a69a410
CM		 892 * when we're called for a write, we're already in the async
893 * submission context. Just jump into btrfs_map_bio
8b712842 894 */
61891923
SB		 895 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
896 if (ret)
897 bio_endio(bio, ret);
898 return ret;
0b86a832
CM		 899}
900
de0022b9
JB		 901static int check_async_write(struct inode *inode, unsigned long bio_flags)
902{
903 if (bio_flags & EXTENT_BIO_TREE_LOG)
904 return 0;
905#ifdef CONFIG_X86
906 if (cpu_has_xmm4_2)
907 return 0;
908#endif
909 return 1;
910}
911
44b8bd7e 912static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 913 int mirror_num, unsigned long bio_flags,
914 u64 bio_offset)
44b8bd7e 915{
de0022b9 916 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 917 int ret;
918
7b6d91da 919 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 920 /*
921 * called for a read, do the setup so that checksum validation
922 * can happen in the async kernel threads
923 */
f3f266ab
CM		 924 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
925 bio, 1);
1d4284bd 926 if (ret)
61891923
SB		 927 goto out_w_error;
928 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
929 mirror_num, 0);
de0022b9
JB		 930 } else if (!async) {
931 ret = btree_csum_one_bio(bio);
932 if (ret)
61891923
SB		 933 goto out_w_error;
934 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
935 mirror_num, 0);
936 } else {
937 /*
938 * kthread helpers are used to submit writes so that
939 * checksumming can happen in parallel across all CPUs
940 */
941 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
942 inode, rw, bio, mirror_num, 0,
943 bio_offset,
944 __btree_submit_bio_start,
945 __btree_submit_bio_done);
44b8bd7e 946 }
d313d7a3 947
61891923
SB		 948 if (ret) {
949out_w_error:
950 bio_endio(bio, ret);
951 }
952 return ret;
44b8bd7e
CM		 953}
954
3dd1462e 955#ifdef CONFIG_MIGRATION
784b4e29 956static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 957 struct page *newpage, struct page *page,
958 enum migrate_mode mode)
784b4e29
CM		 959{
960 /*
961 * we can't safely write a btree page from here,
962 * we haven't done the locking hook
963 */
964 if (PageDirty(page))
965 return -EAGAIN;
966 /*
967 * Buffers may be managed in a filesystem specific way.
968 * We must have no buffers or drop them.
969 */
970 if (page_has_private(page) &&
971 !try_to_release_page(page, GFP_KERNEL))
972 return -EAGAIN;
a6bc32b8 973 return migrate_page(mapping, newpage, page, mode);
784b4e29 974}
3dd1462e 975#endif
784b4e29 976
0da5468f
CM		 977
978static int btree_writepages(struct address_space *mapping,
979 struct writeback_control *wbc)
980{
d1310b2e 981 struct extent_io_tree *tree;
e2d84521
MX		 982 struct btrfs_fs_info *fs_info;
983 int ret;
984
d1310b2e 985 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 986 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 987
988 if (wbc->for_kupdate)
989 return 0;
990
e2d84521 991 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 992 /* this is a bit racy, but that's ok */
e2d84521
MX		 993 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
994 BTRFS_DIRTY_METADATA_THRESH);
995 if (ret < 0)
793955bc 996 return 0;
793955bc 997 }
0b32f4bb 998 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 999}
1000
b2950863 1001static int btree_readpage(struct file *file, struct page *page)
5f39d397 1002{
d1310b2e
CM		 1003 struct extent_io_tree *tree;
1004 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 1005 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 1006}
22b0ebda 1007
70dec807 1008static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 1009{
98509cfc 1010 if (PageWriteback(page) || PageDirty(page))
d397712b 1011 return 0;
0c4e538b 1012
f7a52a40 1013 return try_release_extent_buffer(page);
d98237b3
CM		 1014}
1015
5f39d397 1016static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 1017{
d1310b2e
CM		 1018 struct extent_io_tree *tree;
1019 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1020 extent_invalidatepage(tree, page, offset);
1021 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1022 if (PagePrivate(page)) {
d397712b
CM		 1023 printk(KERN_WARNING "btrfs warning page private not zero "
1024 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1025 ClearPagePrivate(page);
1026 set_page_private(page, 0);
1027 page_cache_release(page);
1028 }
d98237b3
CM		 1029}
1030
0b32f4bb
JB		 1031static int btree_set_page_dirty(struct page *page)
1032{
bb146eb2 1033#ifdef DEBUG
0b32f4bb
JB		 1034 struct extent_buffer *eb;
1035
1036 BUG_ON(!PagePrivate(page));
1037 eb = (struct extent_buffer *)page->private;
1038 BUG_ON(!eb);
1039 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1040 BUG_ON(!atomic_read(&eb->refs));
1041 btrfs_assert_tree_locked(eb);
bb146eb2 1042#endif
0b32f4bb
JB		 1043 return __set_page_dirty_nobuffers(page);
1044}
1045
7f09410b 1046static const struct address_space_operations btree_aops = {
d98237b3 1047 .readpage = btree_readpage,
0da5468f 1048 .writepages = btree_writepages,
5f39d397
CM		 1049 .releasepage = btree_releasepage,
1050 .invalidatepage = btree_invalidatepage,
5a92bc88 1051#ifdef CONFIG_MIGRATION
784b4e29 1052 .migratepage = btree_migratepage,
5a92bc88 1053#endif
0b32f4bb 1054 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1055};
1056
ca7a79ad
CM		 1057int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1058 u64 parent_transid)
090d1875 1059{
5f39d397
CM		 1060 struct extent_buffer *buf = NULL;
1061 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1062 int ret = 0;
090d1875 1063
db94535d 1064 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1065 if (!buf)
090d1875 1066 return 0;
d1310b2e 1067 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1068 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1069 free_extent_buffer(buf);
de428b63 1070 return ret;
090d1875
CM		 1071}
1072
ab0fff03
AJ		 1073int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1074 int mirror_num, struct extent_buffer **eb)
1075{
1076 struct extent_buffer *buf = NULL;
1077 struct inode *btree_inode = root->fs_info->btree_inode;
1078 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1079 int ret;
1080
1081 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1082 if (!buf)
1083 return 0;
1084
1085 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1086
1087 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1088 btree_get_extent, mirror_num);
1089 if (ret) {
1090 free_extent_buffer(buf);
1091 return ret;
1092 }
1093
1094 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1095 free_extent_buffer(buf);
1096 return -EIO;
0b32f4bb 1097 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1098 *eb = buf;
1099 } else {
1100 free_extent_buffer(buf);
1101 }
1102 return 0;
1103}
1104
0999df54
CM		 1105struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1106 u64 bytenr, u32 blocksize)
1107{
1108 struct inode *btree_inode = root->fs_info->btree_inode;
1109 struct extent_buffer *eb;
1110 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
f09d1f60 1111 bytenr, blocksize);
0999df54
CM		 1112 return eb;
1113}
1114
1115struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1116 u64 bytenr, u32 blocksize)
1117{
1118 struct inode *btree_inode = root->fs_info->btree_inode;
1119 struct extent_buffer *eb;
1120
1121 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
727011e0 1122 bytenr, blocksize);
0999df54
CM		 1123 return eb;
1124}
1125
1126
e02119d5
CM		 1127int btrfs_write_tree_block(struct extent_buffer *buf)
1128{
727011e0 1129 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
8aa38c31 1130 buf->start + buf->len - 1);
e02119d5
CM		 1131}
1132
1133int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1134{
727011e0 1135 return filemap_fdatawait_range(buf->pages[0]->mapping,
8aa38c31 1136 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1137}
1138
0999df54 1139struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1140 u32 blocksize, u64 parent_transid)
0999df54
CM		 1141{
1142 struct extent_buffer *buf = NULL;
0999df54
CM		 1143 int ret;
1144
0999df54
CM		 1145 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1146 if (!buf)
1147 return NULL;
0999df54 1148
ca7a79ad 1149 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
5f39d397 1150 return buf;
ce9adaa5 1151
eb60ceac
CM		 1152}
1153
d5c13f92
JM		 1154void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1155 struct extent_buffer *buf)
ed2ff2cb 1156{
e2d84521
MX		 1157 struct btrfs_fs_info *fs_info = root->fs_info;
1158
55c69072 1159 if (btrfs_header_generation(buf) ==
e2d84521 1160 fs_info->running_transaction->transid) {
b9447ef8 1161 btrfs_assert_tree_locked(buf);
b4ce94de 1162
b9473439 1163 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1164 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1165 -buf->len,
1166 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1167 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1168 btrfs_set_lock_blocking(buf);
1169 clear_extent_buffer_dirty(buf);
1170 }
925baedd 1171 }
5f39d397
CM		 1172}
1173
143bede5
JM		 1174static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1175 u32 stripesize, struct btrfs_root *root,
1176 struct btrfs_fs_info *fs_info,
1177 u64 objectid)
d97e63b6 1178{
cfaa7295 1179 root->node = NULL;
a28ec197 1180 root->commit_root = NULL;
db94535d
CM		 1181 root->sectorsize = sectorsize;
1182 root->nodesize = nodesize;
1183 root->leafsize = leafsize;
87ee04eb 1184 root->stripesize = stripesize;
123abc88 1185 root->ref_cows = 0;
0b86a832 1186 root->track_dirty = 0;
c71bf099 1187 root->in_radix = 0;
d68fc57b
YZ		 1188 root->orphan_item_inserted = 0;
1189 root->orphan_cleanup_state = 0;
0b86a832 1190
0f7d52f4
CM		 1191 root->objectid = objectid;
1192 root->last_trans = 0;
13a8a7c8 1193 root->highest_objectid = 0;
58176a96 1194 root->name = NULL;
6bef4d31 1195 root->inode_tree = RB_ROOT;
16cdcec7 1196 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1197 root->block_rsv = NULL;
d68fc57b 1198 root->orphan_block_rsv = NULL;
0b86a832
CM		 1199
1200 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1201 INIT_LIST_HEAD(&root->root_list);
2ab28f32
JB		 1202 INIT_LIST_HEAD(&root->logged_list[0]);
1203 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1204 spin_lock_init(&root->orphan_lock);
5d4f98a2 1205 spin_lock_init(&root->inode_lock);
f0486c68 1206 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1207 spin_lock_init(&root->log_extents_lock[0]);
1208 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1209 mutex_init(&root->objectid_mutex);
e02119d5 1210 mutex_init(&root->log_mutex);
7237f183
YZ		 1211 init_waitqueue_head(&root->log_writer_wait);
1212 init_waitqueue_head(&root->log_commit_wait[0]);
1213 init_waitqueue_head(&root->log_commit_wait[1]);
1214 atomic_set(&root->log_commit[0], 0);
1215 atomic_set(&root->log_commit[1], 0);
1216 atomic_set(&root->log_writers, 0);
2ecb7923 1217 atomic_set(&root->log_batch, 0);
8a35d95f 1218 atomic_set(&root->orphan_inodes, 0);
b0feb9d9 1219 atomic_set(&root->refs, 1);
7237f183 1220 root->log_transid = 0;
257c62e1 1221 root->last_log_commit = 0;
d0c803c4 1222 extent_io_tree_init(&root->dirty_log_pages,
f993c883 1223 fs_info->btree_inode->i_mapping);
017e5369 1224
3768f368
CM		 1225 memset(&root->root_key, 0, sizeof(root->root_key));
1226 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1227 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1228 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 1229 root->defrag_trans_start = fs_info->generation;
58176a96 1230 init_completion(&root->kobj_unregister);
6702ed49 1231 root->defrag_running = 0;
4d775673 1232 root->root_key.objectid = objectid;
0ee5dc67 1233 root->anon_dev = 0;
8ea05e3a 1234
5f3ab90a 1235 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1236}
1237
f84a8bd6 1238static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1239{
1240 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1241 if (root)
1242 root->fs_info = fs_info;
1243 return root;
1244}
1245
20897f5c
AJ		 1246struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1247 struct btrfs_fs_info *fs_info,
1248 u64 objectid)
1249{
1250 struct extent_buffer *leaf;
1251 struct btrfs_root *tree_root = fs_info->tree_root;
1252 struct btrfs_root *root;
1253 struct btrfs_key key;
1254 int ret = 0;
1255 u64 bytenr;
6463fe58 1256 uuid_le uuid;
20897f5c
AJ		 1257
1258 root = btrfs_alloc_root(fs_info);
1259 if (!root)
1260 return ERR_PTR(-ENOMEM);
1261
1262 __setup_root(tree_root->nodesize, tree_root->leafsize,
1263 tree_root->sectorsize, tree_root->stripesize,
1264 root, fs_info, objectid);
1265 root->root_key.objectid = objectid;
1266 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1267 root->root_key.offset = 0;
1268
1269 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1270 0, objectid, NULL, 0, 0, 0);
1271 if (IS_ERR(leaf)) {
1272 ret = PTR_ERR(leaf);
1dd05682 1273 leaf = NULL;
20897f5c
AJ		 1274 goto fail;
1275 }
1276
1277 bytenr = leaf->start;
1278 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1279 btrfs_set_header_bytenr(leaf, leaf->start);
1280 btrfs_set_header_generation(leaf, trans->transid);
1281 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1282 btrfs_set_header_owner(leaf, objectid);
1283 root->node = leaf;
1284
1285 write_extent_buffer(leaf, fs_info->fsid,
1286 (unsigned long)btrfs_header_fsid(leaf),
1287 BTRFS_FSID_SIZE);
1288 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1289 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1290 BTRFS_UUID_SIZE);
1291 btrfs_mark_buffer_dirty(leaf);
1292
1293 root->commit_root = btrfs_root_node(root);
1294 root->track_dirty = 1;
1295
1296
1297 root->root_item.flags = 0;
1298 root->root_item.byte_limit = 0;
1299 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1300 btrfs_set_root_generation(&root->root_item, trans->transid);
1301 btrfs_set_root_level(&root->root_item, 0);
1302 btrfs_set_root_refs(&root->root_item, 1);
1303 btrfs_set_root_used(&root->root_item, leaf->len);
1304 btrfs_set_root_last_snapshot(&root->root_item, 0);
1305 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1306 uuid_le_gen(&uuid);
1307 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1308 root->root_item.drop_level = 0;
1309
1310 key.objectid = objectid;
1311 key.type = BTRFS_ROOT_ITEM_KEY;
1312 key.offset = 0;
1313 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1314 if (ret)
1315 goto fail;
1316
1317 btrfs_tree_unlock(leaf);
1318
1dd05682
TI		 1319 return root;
1320
20897f5c 1321fail:
1dd05682
TI		 1322 if (leaf) {
1323 btrfs_tree_unlock(leaf);
1324 free_extent_buffer(leaf);
1325 }
1326 kfree(root);
20897f5c 1327
1dd05682 1328 return ERR_PTR(ret);
20897f5c
AJ		 1329}
1330
7237f183
YZ		 1331static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1332 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1333{
1334 struct btrfs_root *root;
1335 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1336 struct extent_buffer *leaf;
e02119d5 1337
6f07e42e 1338 root = btrfs_alloc_root(fs_info);
e02119d5 1339 if (!root)
7237f183 1340 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1341
1342 __setup_root(tree_root->nodesize, tree_root->leafsize,
1343 tree_root->sectorsize, tree_root->stripesize,
1344 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1345
1346 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1347 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1348 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1349 /*
1350 * log trees do not get reference counted because they go away
1351 * before a real commit is actually done. They do store pointers
1352 * to file data extents, and those reference counts still get
1353 * updated (along with back refs to the log tree).
1354 */
e02119d5
CM		 1355 root->ref_cows = 0;
1356
5d4f98a2 1357 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1358 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1359 0, 0, 0);
7237f183
YZ		 1360 if (IS_ERR(leaf)) {
1361 kfree(root);
1362 return ERR_CAST(leaf);
1363 }
e02119d5 1364
5d4f98a2
YZ		 1365 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1366 btrfs_set_header_bytenr(leaf, leaf->start);
1367 btrfs_set_header_generation(leaf, trans->transid);
1368 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1369 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1370 root->node = leaf;
e02119d5
CM		 1371
1372 write_extent_buffer(root->node, root->fs_info->fsid,
1373 (unsigned long)btrfs_header_fsid(root->node),
1374 BTRFS_FSID_SIZE);
1375 btrfs_mark_buffer_dirty(root->node);
1376 btrfs_tree_unlock(root->node);
7237f183
YZ		 1377 return root;
1378}
1379
1380int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1381 struct btrfs_fs_info *fs_info)
1382{
1383 struct btrfs_root *log_root;
1384
1385 log_root = alloc_log_tree(trans, fs_info);
1386 if (IS_ERR(log_root))
1387 return PTR_ERR(log_root);
1388 WARN_ON(fs_info->log_root_tree);
1389 fs_info->log_root_tree = log_root;
1390 return 0;
1391}
1392
1393int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1394 struct btrfs_root *root)
1395{
1396 struct btrfs_root *log_root;
1397 struct btrfs_inode_item *inode_item;
1398
1399 log_root = alloc_log_tree(trans, root->fs_info);
1400 if (IS_ERR(log_root))
1401 return PTR_ERR(log_root);
1402
1403 log_root->last_trans = trans->transid;
1404 log_root->root_key.offset = root->root_key.objectid;
1405
1406 inode_item = &log_root->root_item.inode;
1407 inode_item->generation = cpu_to_le64(1);
1408 inode_item->size = cpu_to_le64(3);
1409 inode_item->nlink = cpu_to_le32(1);
1410 inode_item->nbytes = cpu_to_le64(root->leafsize);
1411 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1412
5d4f98a2 1413 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1414
1415 WARN_ON(root->log_root);
1416 root->log_root = log_root;
1417 root->log_transid = 0;
257c62e1 1418 root->last_log_commit = 0;
e02119d5
CM		 1419 return 0;
1420}
1421
cb517eab
MX		 1422struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1423 struct btrfs_key *key)
e02119d5
CM		 1424{
1425 struct btrfs_root *root;
1426 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1427 struct btrfs_path *path;
84234f3a 1428 u64 generation;
db94535d 1429 u32 blocksize;
cb517eab 1430 int ret;
0f7d52f4 1431
cb517eab
MX		 1432 path = btrfs_alloc_path();
1433 if (!path)
0f7d52f4 1434 return ERR_PTR(-ENOMEM);
cb517eab
MX		 1435
1436 root = btrfs_alloc_root(fs_info);
1437 if (!root) {
1438 ret = -ENOMEM;
1439 goto alloc_fail;
0f7d52f4
CM		 1440 }
1441
db94535d 1442 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb 1443 tree_root->sectorsize, tree_root->stripesize,
cb517eab 1444 root, fs_info, key->objectid);
0f7d52f4 1445
cb517eab
MX		 1446 ret = btrfs_find_root(tree_root, key, path,
1447 &root->root_item, &root->root_key);
0f7d52f4 1448 if (ret) {
13a8a7c8
YZ		 1449 if (ret > 0)
1450 ret = -ENOENT;
cb517eab 1451 goto find_fail;
0f7d52f4 1452 }
13a8a7c8 1453
84234f3a 1454 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1455 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1456 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1457 blocksize, generation);
cb517eab
MX		 1458 if (!root->node) {
1459 ret = -ENOMEM;
1460 goto find_fail;
1461 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1462 ret = -EIO;
1463 goto read_fail;
416bc658 1464 }
5d4f98a2 1465 root->commit_root = btrfs_root_node(root);
13a8a7c8 1466out:
cb517eab
MX		 1467 btrfs_free_path(path);
1468 return root;
1469
1470read_fail:
1471 free_extent_buffer(root->node);
1472find_fail:
1473 kfree(root);
1474alloc_fail:
1475 root = ERR_PTR(ret);
1476 goto out;
1477}
1478
1479struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1480 struct btrfs_key *location)
1481{
1482 struct btrfs_root *root;
1483
1484 root = btrfs_read_tree_root(tree_root, location);
1485 if (IS_ERR(root))
1486 return root;
1487
1488 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1489 root->ref_cows = 1;
08fe4db1
LZ		 1490 btrfs_check_and_init_root_item(&root->root_item);
1491 }
13a8a7c8 1492
5eda7b5e
CM		 1493 return root;
1494}
1495
cb517eab
MX		 1496int btrfs_init_fs_root(struct btrfs_root *root)
1497{
1498 int ret;
1499
1500 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1501 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1502 GFP_NOFS);
1503 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1504 ret = -ENOMEM;
1505 goto fail;
1506 }
1507
1508 btrfs_init_free_ino_ctl(root);
1509 mutex_init(&root->fs_commit_mutex);
1510 spin_lock_init(&root->cache_lock);
1511 init_waitqueue_head(&root->cache_wait);
1512
1513 ret = get_anon_bdev(&root->anon_dev);
1514 if (ret)
1515 goto fail;
1516 return 0;
1517fail:
1518 kfree(root->free_ino_ctl);
1519 kfree(root->free_ino_pinned);
1520 return ret;
1521}
1522
1523struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1524 u64 root_id)
1525{
1526 struct btrfs_root *root;
1527
1528 spin_lock(&fs_info->fs_roots_radix_lock);
1529 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1530 (unsigned long)root_id);
1531 spin_unlock(&fs_info->fs_roots_radix_lock);
1532 return root;
1533}
1534
1535int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1536 struct btrfs_root *root)
1537{
1538 int ret;
1539
1540 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1541 if (ret)
1542 return ret;
1543
1544 spin_lock(&fs_info->fs_roots_radix_lock);
1545 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1546 (unsigned long)root->root_key.objectid,
1547 root);
1548 if (ret == 0)
1549 root->in_radix = 1;
1550 spin_unlock(&fs_info->fs_roots_radix_lock);
1551 radix_tree_preload_end();
1552
1553 return ret;
1554}
1555
edbd8d4e
CM		 1556struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1557 struct btrfs_key *location)
5eda7b5e
CM		 1558{
1559 struct btrfs_root *root;
1560 int ret;
1561
edbd8d4e
CM		 1562 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1563 return fs_info->tree_root;
1564 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1565 return fs_info->extent_root;
8f18cf13
CM		 1566 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1567 return fs_info->chunk_root;
1568 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1569 return fs_info->dev_root;
0403e47e
YZ		 1570 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1571 return fs_info->csum_root;
bcef60f2
AJ		 1572 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1573 return fs_info->quota_root ? fs_info->quota_root :
1574 ERR_PTR(-ENOENT);
4df27c4d 1575again:
cb517eab 1576 root = btrfs_lookup_fs_root(fs_info, location->objectid);
5eda7b5e
CM		 1577 if (root)
1578 return root;
1579
cb517eab 1580 root = btrfs_read_fs_root(fs_info->tree_root, location);
5eda7b5e
CM		 1581 if (IS_ERR(root))
1582 return root;
3394e160 1583
cb517eab
MX		 1584 if (btrfs_root_refs(&root->root_item) == 0) {
1585 ret = -ENOENT;
581bb050 1586 goto fail;
35a30d7c 1587 }
581bb050 1588
cb517eab 1589 ret = btrfs_init_fs_root(root);
ac08aedf
CM		 1590 if (ret)
1591 goto fail;
3394e160 1592
d68fc57b
YZ		 1593 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1594 if (ret < 0)
1595 goto fail;
1596 if (ret == 0)
1597 root->orphan_item_inserted = 1;
1598
cb517eab 1599 ret = btrfs_insert_fs_root(fs_info, root);
0f7d52f4 1600 if (ret) {
4df27c4d
YZ		 1601 if (ret == -EEXIST) {
1602 free_fs_root(root);
1603 goto again;
1604 }
1605 goto fail;
0f7d52f4 1606 }
edbd8d4e 1607 return root;
4df27c4d
YZ		 1608fail:
1609 free_fs_root(root);
1610 return ERR_PTR(ret);
edbd8d4e
CM		 1611}
1612
04160088
CM		 1613static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1614{
1615 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1616 int ret = 0;
04160088
CM		 1617 struct btrfs_device *device;
1618 struct backing_dev_info *bdi;
b7967db7 1619
1f78160c
XG		 1620 rcu_read_lock();
1621 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1622 if (!device->bdev)
1623 continue;
04160088
CM		 1624 bdi = blk_get_backing_dev_info(device->bdev);
1625 if (bdi && bdi_congested(bdi, bdi_bits)) {
1626 ret = 1;
1627 break;
1628 }
1629 }
1f78160c 1630 rcu_read_unlock();
04160088
CM		 1631 return ret;
1632}
1633
ad081f14
JA		 1634/*
1635 * If this fails, caller must call bdi_destroy() to get rid of the
1636 * bdi again.
1637 */
04160088
CM		 1638static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1639{
ad081f14
JA		 1640 int err;
1641
1642 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1643 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1644 if (err)
1645 return err;
1646
4575c9cc 1647 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1648 bdi->congested_fn = btrfs_congested_fn;
1649 bdi->congested_data = info;
1650 return 0;
1651}
1652
8b712842
CM		 1653/*
1654 * called by the kthread helper functions to finally call the bio end_io
1655 * functions. This is where read checksum verification actually happens
1656 */
1657static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1658{
ce9adaa5 1659 struct bio *bio;
8b712842
CM		 1660 struct end_io_wq *end_io_wq;
1661 struct btrfs_fs_info *fs_info;
ce9adaa5 1662 int error;
ce9adaa5 1663
8b712842
CM		 1664 end_io_wq = container_of(work, struct end_io_wq, work);
1665 bio = end_io_wq->bio;
1666 fs_info = end_io_wq->info;
ce9adaa5 1667
8b712842
CM		 1668 error = end_io_wq->error;
1669 bio->bi_private = end_io_wq->private;
1670 bio->bi_end_io = end_io_wq->end_io;
1671 kfree(end_io_wq);
8b712842 1672 bio_endio(bio, error);
44b8bd7e
CM		 1673}
1674
a74a4b97
CM		 1675static int cleaner_kthread(void *arg)
1676{
1677 struct btrfs_root *root = arg;
d0278245 1678 int again;
a74a4b97
CM		 1679
1680 do {
d0278245 1681 again = 0;
a74a4b97 1682
d0278245 1683 /* Make the cleaner go to sleep early. */
babbf170 1684 if (btrfs_need_cleaner_sleep(root))
d0278245
MX		 1685 goto sleep;
1686
1687 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1688 goto sleep;
1689
dc7f370c
MX		 1690 /*
1691 * Avoid the problem that we change the status of the fs
1692 * during the above check and trylock.
1693 */
babbf170 1694 if (btrfs_need_cleaner_sleep(root)) {
dc7f370c
MX		 1695 mutex_unlock(&root->fs_info->cleaner_mutex);
1696 goto sleep;
1697 }
1698
d0278245
MX		 1699 btrfs_run_delayed_iputs(root);
1700 again = btrfs_clean_one_deleted_snapshot(root);
1701 mutex_unlock(&root->fs_info->cleaner_mutex);
1702
1703 /*
05323cd1
MX		 1704 * The defragger has dealt with the R/O remount and umount,
1705 * needn't do anything special here.
d0278245
MX		 1706 */
1707 btrfs_run_defrag_inodes(root->fs_info);
1708sleep:
9d1a2a3a 1709 if (!try_to_freeze() && !again) {
a74a4b97 1710 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1711 if (!kthread_should_stop())
1712 schedule();
a74a4b97
CM		 1713 __set_current_state(TASK_RUNNING);
1714 }
1715 } while (!kthread_should_stop());
1716 return 0;
1717}
1718
1719static int transaction_kthread(void *arg)
1720{
1721 struct btrfs_root *root = arg;
1722 struct btrfs_trans_handle *trans;
1723 struct btrfs_transaction *cur;
8929ecfa 1724 u64 transid;
a74a4b97
CM		 1725 unsigned long now;
1726 unsigned long delay;
914b2007 1727 bool cannot_commit;
a74a4b97
CM		 1728
1729 do {
914b2007 1730 cannot_commit = false;
a74a4b97 1731 delay = HZ * 30;
a74a4b97
CM		 1732 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1733
a4abeea4 1734 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1735 cur = root->fs_info->running_transaction;
1736 if (!cur) {
a4abeea4 1737 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1738 goto sleep;
1739 }
31153d81 1740
a74a4b97 1741 now = get_seconds();
8929ecfa
YZ		 1742 if (!cur->blocked &&
1743 (now < cur->start_time || now - cur->start_time < 30)) {
a4abeea4 1744 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1745 delay = HZ * 5;
1746 goto sleep;
1747 }
8929ecfa 1748 transid = cur->transid;
a4abeea4 1749 spin_unlock(&root->fs_info->trans_lock);
56bec294 1750
79787eaa 1751 /* If the file system is aborted, this will always fail. */
354aa0fb 1752 trans = btrfs_attach_transaction(root);
914b2007 1753 if (IS_ERR(trans)) {
354aa0fb
MX		 1754 if (PTR_ERR(trans) != -ENOENT)
1755 cannot_commit = true;
79787eaa 1756 goto sleep;
914b2007 1757 }
8929ecfa 1758 if (transid == trans->transid) {
79787eaa 1759 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1760 } else {
1761 btrfs_end_transaction(trans, root);
1762 }
a74a4b97
CM		 1763sleep:
1764 wake_up_process(root->fs_info->cleaner_kthread);
1765 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1766
a0acae0e 1767 if (!try_to_freeze()) {
a74a4b97 1768 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1769 if (!kthread_should_stop() &&
914b2007
JK		 1770 (!btrfs_transaction_blocked(root->fs_info) ||
1771 cannot_commit))
8929ecfa 1772 schedule_timeout(delay);
a74a4b97
CM		 1773 __set_current_state(TASK_RUNNING);
1774 }
1775 } while (!kthread_should_stop());
1776 return 0;
1777}
1778
af31f5e5
CM		 1779/*
1780 * this will find the highest generation in the array of
1781 * root backups. The index of the highest array is returned,
1782 * or -1 if we can't find anything.
1783 *
1784 * We check to make sure the array is valid by comparing the
1785 * generation of the latest root in the array with the generation
1786 * in the super block. If they don't match we pitch it.
1787 */
1788static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1789{
1790 u64 cur;
1791 int newest_index = -1;
1792 struct btrfs_root_backup *root_backup;
1793 int i;
1794
1795 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1796 root_backup = info->super_copy->super_roots + i;
1797 cur = btrfs_backup_tree_root_gen(root_backup);
1798 if (cur == newest_gen)
1799 newest_index = i;
1800 }
1801
1802 /* check to see if we actually wrapped around */
1803 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1804 root_backup = info->super_copy->super_roots;
1805 cur = btrfs_backup_tree_root_gen(root_backup);
1806 if (cur == newest_gen)
1807 newest_index = 0;
1808 }
1809 return newest_index;
1810}
1811
1812
1813/*
1814 * find the oldest backup so we know where to store new entries
1815 * in the backup array. This will set the backup_root_index
1816 * field in the fs_info struct
1817 */
1818static void find_oldest_super_backup(struct btrfs_fs_info *info,
1819 u64 newest_gen)
1820{
1821 int newest_index = -1;
1822
1823 newest_index = find_newest_super_backup(info, newest_gen);
1824 /* if there was garbage in there, just move along */
1825 if (newest_index == -1) {
1826 info->backup_root_index = 0;
1827 } else {
1828 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1829 }
1830}
1831
1832/*
1833 * copy all the root pointers into the super backup array.
1834 * this will bump the backup pointer by one when it is
1835 * done
1836 */
1837static void backup_super_roots(struct btrfs_fs_info *info)
1838{
1839 int next_backup;
1840 struct btrfs_root_backup *root_backup;
1841 int last_backup;
1842
1843 next_backup = info->backup_root_index;
1844 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1845 BTRFS_NUM_BACKUP_ROOTS;
1846
1847 /*
1848 * just overwrite the last backup if we're at the same generation
1849 * this happens only at umount
1850 */
1851 root_backup = info->super_for_commit->super_roots + last_backup;
1852 if (btrfs_backup_tree_root_gen(root_backup) ==
1853 btrfs_header_generation(info->tree_root->node))
1854 next_backup = last_backup;
1855
1856 root_backup = info->super_for_commit->super_roots + next_backup;
1857
1858 /*
1859 * make sure all of our padding and empty slots get zero filled
1860 * regardless of which ones we use today
1861 */
1862 memset(root_backup, 0, sizeof(*root_backup));
1863
1864 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1865
1866 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1867 btrfs_set_backup_tree_root_gen(root_backup,
1868 btrfs_header_generation(info->tree_root->node));
1869
1870 btrfs_set_backup_tree_root_level(root_backup,
1871 btrfs_header_level(info->tree_root->node));
1872
1873 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1874 btrfs_set_backup_chunk_root_gen(root_backup,
1875 btrfs_header_generation(info->chunk_root->node));
1876 btrfs_set_backup_chunk_root_level(root_backup,
1877 btrfs_header_level(info->chunk_root->node));
1878
1879 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1880 btrfs_set_backup_extent_root_gen(root_backup,
1881 btrfs_header_generation(info->extent_root->node));
1882 btrfs_set_backup_extent_root_level(root_backup,
1883 btrfs_header_level(info->extent_root->node));
1884
7c7e82a7
CM		 1885 /*
1886 * we might commit during log recovery, which happens before we set
1887 * the fs_root. Make sure it is valid before we fill it in.
1888 */
1889 if (info->fs_root && info->fs_root->node) {
1890 btrfs_set_backup_fs_root(root_backup,
1891 info->fs_root->node->start);
1892 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1893 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1894 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1895 btrfs_header_level(info->fs_root->node));
7c7e82a7 1896 }
af31f5e5
CM		 1897
1898 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1899 btrfs_set_backup_dev_root_gen(root_backup,
1900 btrfs_header_generation(info->dev_root->node));
1901 btrfs_set_backup_dev_root_level(root_backup,
1902 btrfs_header_level(info->dev_root->node));
1903
1904 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1905 btrfs_set_backup_csum_root_gen(root_backup,
1906 btrfs_header_generation(info->csum_root->node));
1907 btrfs_set_backup_csum_root_level(root_backup,
1908 btrfs_header_level(info->csum_root->node));
1909
1910 btrfs_set_backup_total_bytes(root_backup,
1911 btrfs_super_total_bytes(info->super_copy));
1912 btrfs_set_backup_bytes_used(root_backup,
1913 btrfs_super_bytes_used(info->super_copy));
1914 btrfs_set_backup_num_devices(root_backup,
1915 btrfs_super_num_devices(info->super_copy));
1916
1917 /*
1918 * if we don't copy this out to the super_copy, it won't get remembered
1919 * for the next commit
1920 */
1921 memcpy(&info->super_copy->super_roots,
1922 &info->super_for_commit->super_roots,
1923 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1924}
1925
1926/*
1927 * this copies info out of the root backup array and back into
1928 * the in-memory super block. It is meant to help iterate through
1929 * the array, so you send it the number of backups you've already
1930 * tried and the last backup index you used.
1931 *
1932 * this returns -1 when it has tried all the backups
1933 */
1934static noinline int next_root_backup(struct btrfs_fs_info *info,
1935 struct btrfs_super_block *super,
1936 int *num_backups_tried, int *backup_index)
1937{
1938 struct btrfs_root_backup *root_backup;
1939 int newest = *backup_index;
1940
1941 if (*num_backups_tried == 0) {
1942 u64 gen = btrfs_super_generation(super);
1943
1944 newest = find_newest_super_backup(info, gen);
1945 if (newest == -1)
1946 return -1;
1947
1948 *backup_index = newest;
1949 *num_backups_tried = 1;
1950 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1951 /* we've tried all the backups, all done */
1952 return -1;
1953 } else {
1954 /* jump to the next oldest backup */
1955 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1956 BTRFS_NUM_BACKUP_ROOTS;
1957 *backup_index = newest;
1958 *num_backups_tried += 1;
1959 }
1960 root_backup = super->super_roots + newest;
1961
1962 btrfs_set_super_generation(super,
1963 btrfs_backup_tree_root_gen(root_backup));
1964 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1965 btrfs_set_super_root_level(super,
1966 btrfs_backup_tree_root_level(root_backup));
1967 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1968
1969 /*
1970 * fixme: the total bytes and num_devices need to match or we should
1971 * need a fsck
1972 */
1973 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1974 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1975 return 0;
1976}
1977
7abadb64
LB		 1978/* helper to cleanup workers */
1979static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1980{
1981 btrfs_stop_workers(&fs_info->generic_worker);
1982 btrfs_stop_workers(&fs_info->fixup_workers);
1983 btrfs_stop_workers(&fs_info->delalloc_workers);
1984 btrfs_stop_workers(&fs_info->workers);
1985 btrfs_stop_workers(&fs_info->endio_workers);
1986 btrfs_stop_workers(&fs_info->endio_meta_workers);
1987 btrfs_stop_workers(&fs_info->endio_raid56_workers);
1988 btrfs_stop_workers(&fs_info->rmw_workers);
1989 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1990 btrfs_stop_workers(&fs_info->endio_write_workers);
1991 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1992 btrfs_stop_workers(&fs_info->submit_workers);
1993 btrfs_stop_workers(&fs_info->delayed_workers);
1994 btrfs_stop_workers(&fs_info->caching_workers);
1995 btrfs_stop_workers(&fs_info->readahead_workers);
1996 btrfs_stop_workers(&fs_info->flush_workers);
2f232036 1997 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
7abadb64
LB		 1998}
1999
af31f5e5
CM		 2000/* helper to cleanup tree roots */
2001static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2002{
2003 free_extent_buffer(info->tree_root->node);
2004 free_extent_buffer(info->tree_root->commit_root);
af31f5e5
CM		 2005 info->tree_root->node = NULL;
2006 info->tree_root->commit_root = NULL;
655b09fe
JB		 2007
2008 if (info->dev_root) {
2009 free_extent_buffer(info->dev_root->node);
2010 free_extent_buffer(info->dev_root->commit_root);
2011 info->dev_root->node = NULL;
2012 info->dev_root->commit_root = NULL;
2013 }
2014 if (info->extent_root) {
2015 free_extent_buffer(info->extent_root->node);
2016 free_extent_buffer(info->extent_root->commit_root);
2017 info->extent_root->node = NULL;
2018 info->extent_root->commit_root = NULL;
2019 }
2020 if (info->csum_root) {
2021 free_extent_buffer(info->csum_root->node);
2022 free_extent_buffer(info->csum_root->commit_root);
2023 info->csum_root->node = NULL;
2024 info->csum_root->commit_root = NULL;
2025 }
bcef60f2 2026 if (info->quota_root) {
655b09fe
JB		 2027 free_extent_buffer(info->quota_root->node);
2028 free_extent_buffer(info->quota_root->commit_root);
bcef60f2
AJ		 2029 info->quota_root->node = NULL;
2030 info->quota_root->commit_root = NULL;
2031 }
af31f5e5
CM		 2032 if (chunk_root) {
2033 free_extent_buffer(info->chunk_root->node);
2034 free_extent_buffer(info->chunk_root->commit_root);
2035 info->chunk_root->node = NULL;
2036 info->chunk_root->commit_root = NULL;
2037 }
2038}
2039
171f6537
JB		 2040static void del_fs_roots(struct btrfs_fs_info *fs_info)
2041{
2042 int ret;
2043 struct btrfs_root *gang[8];
2044 int i;
2045
2046 while (!list_empty(&fs_info->dead_roots)) {
2047 gang[0] = list_entry(fs_info->dead_roots.next,
2048 struct btrfs_root, root_list);
2049 list_del(&gang[0]->root_list);
2050
2051 if (gang[0]->in_radix) {
cb517eab 2052 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
171f6537
JB		 2053 } else {
2054 free_extent_buffer(gang[0]->node);
2055 free_extent_buffer(gang[0]->commit_root);
b0feb9d9 2056 btrfs_put_fs_root(gang[0]);
171f6537
JB		 2057 }
2058 }
2059
2060 while (1) {
2061 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2062 (void **)gang, 0,
2063 ARRAY_SIZE(gang));
2064 if (!ret)
2065 break;
2066 for (i = 0; i < ret; i++)
cb517eab 2067 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
171f6537
JB		 2068 }
2069}
af31f5e5 2070
ad2b2c80
AV		 2071int open_ctree(struct super_block *sb,
2072 struct btrfs_fs_devices *fs_devices,
2073 char *options)
2e635a27 2074{
db94535d
CM		 2075 u32 sectorsize;
2076 u32 nodesize;
2077 u32 leafsize;
2078 u32 blocksize;
87ee04eb 2079 u32 stripesize;
84234f3a 2080 u64 generation;
f2b636e8 2081 u64 features;
3de4586c 2082 struct btrfs_key location;
a061fc8d 2083 struct buffer_head *bh;
4d34b278 2084 struct btrfs_super_block *disk_super;
815745cf 2085 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2086 struct btrfs_root *tree_root;
4d34b278
ID		 2087 struct btrfs_root *extent_root;
2088 struct btrfs_root *csum_root;
2089 struct btrfs_root *chunk_root;
2090 struct btrfs_root *dev_root;
bcef60f2 2091 struct btrfs_root *quota_root;
e02119d5 2092 struct btrfs_root *log_tree_root;
eb60ceac 2093 int ret;
e58ca020 2094 int err = -EINVAL;
af31f5e5
CM		 2095 int num_backups_tried = 0;
2096 int backup_index = 0;
4543df7e 2097
f84a8bd6 2098 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e 2099 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
cb517eab 2100 if (!tree_root || !chunk_root) {
39279cc3
CM		 2101 err = -ENOMEM;
2102 goto fail;
2103 }
76dda93c
YZ		 2104
2105 ret = init_srcu_struct(&fs_info->subvol_srcu);
2106 if (ret) {
2107 err = ret;
2108 goto fail;
2109 }
2110
2111 ret = setup_bdi(fs_info, &fs_info->bdi);
2112 if (ret) {
2113 err = ret;
2114 goto fail_srcu;
2115 }
2116
e2d84521
MX		 2117 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2118 if (ret) {
2119 err = ret;
2120 goto fail_bdi;
2121 }
2122 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2123 (1 + ilog2(nr_cpu_ids));
2124
963d678b
MX		 2125 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2126 if (ret) {
2127 err = ret;
2128 goto fail_dirty_metadata_bytes;
2129 }
2130
76dda93c
YZ		 2131 fs_info->btree_inode = new_inode(sb);
2132 if (!fs_info->btree_inode) {
2133 err = -ENOMEM;
963d678b 2134 goto fail_delalloc_bytes;
76dda93c
YZ		 2135 }
2136
a6591715 2137 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2138
76dda93c 2139 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 2140 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2141 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2142 INIT_LIST_HEAD(&fs_info->delayed_iputs);
ea8c2819 2143 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
11833d66 2144 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1832a6d5 2145 spin_lock_init(&fs_info->delalloc_lock);
a4abeea4 2146 spin_lock_init(&fs_info->trans_lock);
76dda93c 2147 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2148 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2149 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2150 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2151 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2152 spin_lock_init(&fs_info->super_lock);
f29021b2 2153 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2154 mutex_init(&fs_info->reloc_mutex);
de98ced9 2155 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2156
58176a96 2157 init_completion(&fs_info->kobj_unregister);
0b86a832 2158 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2159 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2160 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2161 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2162 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2163 BTRFS_BLOCK_RSV_GLOBAL);
2164 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2165 BTRFS_BLOCK_RSV_DELALLOC);
2166 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2167 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2168 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2169 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2170 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2171 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2172 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2173 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2174 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2175 atomic_set(&fs_info->defrag_running, 0);
fc36ed7e 2176 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2177 fs_info->sb = sb;
6f568d35 2178 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2179 fs_info->metadata_ratio = 0;
4cb5300b 2180 fs_info->defrag_inodes = RB_ROOT;
a4abeea4 2181 fs_info->trans_no_join = 0;
2bf64758 2182 fs_info->free_chunk_space = 0;
f29021b2 2183 fs_info->tree_mod_log = RB_ROOT;
c8b97818 2184
90519d66
AJ		 2185 /* readahead state */
2186 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2187 spin_lock_init(&fs_info->reada_lock);
c8b97818 2188
b34b086c
CM		 2189 fs_info->thread_pool_size = min_t(unsigned long,
2190 num_online_cpus() + 2, 8);
0afbaf8c 2191
3eaa2885
CM		 2192 INIT_LIST_HEAD(&fs_info->ordered_extents);
2193 spin_lock_init(&fs_info->ordered_extent_lock);
16cdcec7
MX		 2194 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2195 GFP_NOFS);
2196 if (!fs_info->delayed_root) {
2197 err = -ENOMEM;
2198 goto fail_iput;
2199 }
2200 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2201
a2de733c
AJ		 2202 mutex_init(&fs_info->scrub_lock);
2203 atomic_set(&fs_info->scrubs_running, 0);
2204 atomic_set(&fs_info->scrub_pause_req, 0);
2205 atomic_set(&fs_info->scrubs_paused, 0);
2206 atomic_set(&fs_info->scrub_cancel_req, 0);
2207 init_waitqueue_head(&fs_info->scrub_pause_wait);
2208 init_rwsem(&fs_info->scrub_super_lock);
2209 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2210#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2211 fs_info->check_integrity_print_mask = 0;
2212#endif
a2de733c 2213
c9e9f97b
ID		 2214 spin_lock_init(&fs_info->balance_lock);
2215 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2216 atomic_set(&fs_info->balance_running, 0);
2217 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2218 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2219 fs_info->balance_ctl = NULL;
837d5b6e 2220 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2221
a061fc8d
CM		 2222 sb->s_blocksize = 4096;
2223 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2224 sb->s_bdi = &fs_info->bdi;
a061fc8d 2225
76dda93c 2226 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2227 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2228 /*
2229 * we set the i_size on the btree inode to the max possible int.
2230 * the real end of the address space is determined by all of
2231 * the devices in the system
2232 */
2233 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2234 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2235 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2236
5d4f98a2 2237 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2238 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2239 fs_info->btree_inode->i_mapping);
0b32f4bb 2240 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2241 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2242
2243 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2244
76dda93c
YZ		 2245 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2246 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2247 sizeof(struct btrfs_key));
72ac3c0d
JB		 2248 set_bit(BTRFS_INODE_DUMMY,
2249 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
c65ddb52 2250 insert_inode_hash(fs_info->btree_inode);
76dda93c 2251
0f9dd46c 2252 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2253 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2254 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2255
11833d66 2256 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2257 fs_info->btree_inode->i_mapping);
11833d66 2258 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2259 fs_info->btree_inode->i_mapping);
11833d66 2260 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2261 fs_info->do_barriers = 1;
e18e4809 2262
39279cc3 2263
5a3f23d5 2264 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 2265 mutex_init(&fs_info->tree_log_mutex);
925baedd 2266 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2267 mutex_init(&fs_info->transaction_kthread_mutex);
2268 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2269 mutex_init(&fs_info->volume_mutex);
276e680d 2270 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2271 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2272 init_rwsem(&fs_info->subvol_sem);
e922e087
SB		 2273 fs_info->dev_replace.lock_owner = 0;
2274 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2275 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2276 mutex_init(&fs_info->dev_replace.lock_management_lock);
2277 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2278
416ac51d 2279 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2280 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2281 fs_info->qgroup_tree = RB_ROOT;
2282 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2283 fs_info->qgroup_seq = 1;
2284 fs_info->quota_enabled = 0;
2285 fs_info->pending_quota_state = 0;
1e8f9158 2286 fs_info->qgroup_ulist = NULL;
2f232036 2287 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2288
fa9c0d79
CM		 2289 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2290 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2291
e6dcd2dc 2292 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2293 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2294 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2295 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2296
53b381b3
DW		 2297 ret = btrfs_alloc_stripe_hash_table(fs_info);
2298 if (ret) {
83c8266a 2299 err = ret;
53b381b3
DW		 2300 goto fail_alloc;
2301 }
2302
0b86a832 2303 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2304 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2305
3c4bb26b 2306 invalidate_bdev(fs_devices->latest_bdev);
1104a885
DS		 2307
2308 /*
2309 * Read super block and check the signature bytes only
2310 */
a512bbf8 2311 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2312 if (!bh) {
2313 err = -EINVAL;
16cdcec7 2314 goto fail_alloc;
20b45077 2315 }
39279cc3 2316
1104a885
DS		 2317 /*
2318 * We want to check superblock checksum, the type is stored inside.
2319 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2320 */
2321 if (btrfs_check_super_csum(bh->b_data)) {
2322 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2323 err = -EINVAL;
2324 goto fail_alloc;
2325 }
2326
2327 /*
2328 * super_copy is zeroed at allocation time and we never touch the
2329 * following bytes up to INFO_SIZE, the checksum is calculated from
2330 * the whole block of INFO_SIZE
2331 */
6c41761f
DS		 2332 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2333 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2334 sizeof(*fs_info->super_for_commit));
a061fc8d 2335 brelse(bh);
5f39d397 2336
6c41761f 2337 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2338
1104a885
DS		 2339 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2340 if (ret) {
2341 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2342 err = -EINVAL;
2343 goto fail_alloc;
2344 }
2345
6c41761f 2346 disk_super = fs_info->super_copy;
0f7d52f4 2347 if (!btrfs_super_root(disk_super))
16cdcec7 2348 goto fail_alloc;
0f7d52f4 2349
acce952b 2350 /* check FS state, whether FS is broken. */
87533c47
MX		 2351 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2352 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2353
af31f5e5
CM		 2354 /*
2355 * run through our array of backup supers and setup
2356 * our ring pointer to the oldest one
2357 */
2358 generation = btrfs_super_generation(disk_super);
2359 find_oldest_super_backup(fs_info, generation);
2360
75e7cb7f
LB		 2361 /*
2362 * In the long term, we'll store the compression type in the super
2363 * block, and it'll be used for per file compression control.
2364 */
2365 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2366
2b82032c
YZ		 2367 ret = btrfs_parse_options(tree_root, options);
2368 if (ret) {
2369 err = ret;
16cdcec7 2370 goto fail_alloc;
2b82032c 2371 }
dfe25020 2372
f2b636e8
JB		 2373 features = btrfs_super_incompat_flags(disk_super) &
2374 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2375 if (features) {
2376 printk(KERN_ERR "BTRFS: couldn't mount because of "
2377 "unsupported optional features (%Lx).\n",
21380931 2378 (unsigned long long)features);
f2b636e8 2379 err = -EINVAL;
16cdcec7 2380 goto fail_alloc;
f2b636e8
JB		 2381 }
2382
727011e0
CM		 2383 if (btrfs_super_leafsize(disk_super) !=
2384 btrfs_super_nodesize(disk_super)) {
2385 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2386 "blocksizes don't match. node %d leaf %d\n",
2387 btrfs_super_nodesize(disk_super),
2388 btrfs_super_leafsize(disk_super));
2389 err = -EINVAL;
2390 goto fail_alloc;
2391 }
2392 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2393 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2394 "blocksize (%d) was too large\n",
2395 btrfs_super_leafsize(disk_super));
2396 err = -EINVAL;
2397 goto fail_alloc;
2398 }
2399
5d4f98a2 2400 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2401 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2402 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2403 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2404
3173a18f
JB		 2405 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2406 printk(KERN_ERR "btrfs: has skinny extents\n");
2407
727011e0
CM		 2408 /*
2409 * flag our filesystem as having big metadata blocks if
2410 * they are bigger than the page size
2411 */
2412 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2413 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2414 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2415 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2416 }
2417
bc3f116f
CM		 2418 nodesize = btrfs_super_nodesize(disk_super);
2419 leafsize = btrfs_super_leafsize(disk_super);
2420 sectorsize = btrfs_super_sectorsize(disk_super);
2421 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2422 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2423 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2424
2425 /*
2426 * mixed block groups end up with duplicate but slightly offset
2427 * extent buffers for the same range. It leads to corruptions
2428 */
2429 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2430 (sectorsize != leafsize)) {
2431 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2432 "are not allowed for mixed block groups on %s\n",
2433 sb->s_id);
2434 goto fail_alloc;
2435 }
2436
ceda0864
MX		 2437 /*
2438 * Needn't use the lock because there is no other task which will
2439 * update the flag.
2440 */
a6fa6fae 2441 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2442
f2b636e8
JB		 2443 features = btrfs_super_compat_ro_flags(disk_super) &
2444 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2445 if (!(sb->s_flags & MS_RDONLY) && features) {
2446 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2447 "unsupported option features (%Lx).\n",
21380931 2448 (unsigned long long)features);
f2b636e8 2449 err = -EINVAL;
16cdcec7 2450 goto fail_alloc;
f2b636e8 2451 }
61d92c32
CM		 2452
2453 btrfs_init_workers(&fs_info->generic_worker,
2454 "genwork", 1, NULL);
2455
5443be45 2456 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 2457 fs_info->thread_pool_size,
2458 &fs_info->generic_worker);
c8b97818 2459
771ed689 2460 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 2461 fs_info->thread_pool_size,
2462 &fs_info->generic_worker);
771ed689 2463
8ccf6f19
MX		 2464 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2465 fs_info->thread_pool_size,
2466 &fs_info->generic_worker);
2467
5443be45 2468 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 2469 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 2470 fs_info->thread_pool_size),
2471 &fs_info->generic_worker);
61b49440 2472
bab39bf9
JB		 2473 btrfs_init_workers(&fs_info->caching_workers, "cache",
2474 2, &fs_info->generic_worker);
2475
61b49440
CM		 2476 /* a higher idle thresh on the submit workers makes it much more
2477 * likely that bios will be send down in a sane order to the
2478 * devices
2479 */
2480 fs_info->submit_workers.idle_thresh = 64;
53863232 2481
771ed689 2482 fs_info->workers.idle_thresh = 16;
4a69a410 2483 fs_info->workers.ordered = 1;
61b49440 2484
771ed689
CM		 2485 fs_info->delalloc_workers.idle_thresh = 2;
2486 fs_info->delalloc_workers.ordered = 1;
2487
61d92c32
CM		 2488 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2489 &fs_info->generic_worker);
5443be45 2490 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2491 fs_info->thread_pool_size,
2492 &fs_info->generic_worker);
d20f7043 2493 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2494 fs_info->thread_pool_size,
2495 &fs_info->generic_worker);
cad321ad 2496 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2497 "endio-meta-write", fs_info->thread_pool_size,
2498 &fs_info->generic_worker);
53b381b3
DW		 2499 btrfs_init_workers(&fs_info->endio_raid56_workers,
2500 "endio-raid56", fs_info->thread_pool_size,
2501 &fs_info->generic_worker);
2502 btrfs_init_workers(&fs_info->rmw_workers,
2503 "rmw", fs_info->thread_pool_size,
2504 &fs_info->generic_worker);
5443be45 2505 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2506 fs_info->thread_pool_size,
2507 &fs_info->generic_worker);
0cb59c99
JB		 2508 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2509 1, &fs_info->generic_worker);
16cdcec7
MX		 2510 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2511 fs_info->thread_pool_size,
2512 &fs_info->generic_worker);
90519d66
AJ		 2513 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2514 fs_info->thread_pool_size,
2515 &fs_info->generic_worker);
2f232036
JS		 2516 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2517 &fs_info->generic_worker);
61b49440
CM		 2518
2519 /*
2520 * endios are largely parallel and should have a very
2521 * low idle thresh
2522 */
2523 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2524 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2525 fs_info->endio_raid56_workers.idle_thresh = 4;
2526 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2527
9042846b
CM		 2528 fs_info->endio_write_workers.idle_thresh = 2;
2529 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2530 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2531
0dc3b84a
JB		 2532 /*
2533 * btrfs_start_workers can really only fail because of ENOMEM so just
2534 * return -ENOMEM if any of these fail.
2535 */
2536 ret = btrfs_start_workers(&fs_info->workers);
2537 ret |= btrfs_start_workers(&fs_info->generic_worker);
2538 ret |= btrfs_start_workers(&fs_info->submit_workers);
2539 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2540 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2541 ret |= btrfs_start_workers(&fs_info->endio_workers);
2542 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2543 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2544 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2545 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2546 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2547 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2548 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2549 ret |= btrfs_start_workers(&fs_info->caching_workers);
2550 ret |= btrfs_start_workers(&fs_info->readahead_workers);
8ccf6f19 2551 ret |= btrfs_start_workers(&fs_info->flush_workers);
2f232036 2552 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
0dc3b84a 2553 if (ret) {
fed425c7 2554 err = -ENOMEM;
0dc3b84a
JB		 2555 goto fail_sb_buffer;
2556 }
4543df7e 2557
4575c9cc 2558 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2559 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2560 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2561
db94535d
CM		 2562 tree_root->nodesize = nodesize;
2563 tree_root->leafsize = leafsize;
2564 tree_root->sectorsize = sectorsize;
87ee04eb 2565 tree_root->stripesize = stripesize;
a061fc8d
CM		 2566
2567 sb->s_blocksize = sectorsize;
2568 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2569
cdb4c574 2570 if (disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) {
d397712b 2571 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2572 goto fail_sb_buffer;
2573 }
19c00ddc 2574
8d082fb7
LB		 2575 if (sectorsize != PAGE_SIZE) {
2576 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2577 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2578 goto fail_sb_buffer;
2579 }
2580
925baedd 2581 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2582 ret = btrfs_read_sys_array(tree_root);
925baedd 2583 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2584 if (ret) {
d397712b
CM		 2585 printk(KERN_WARNING "btrfs: failed to read the system "
2586 "array on %s\n", sb->s_id);
5d4f98a2 2587 goto fail_sb_buffer;
84eed90f 2588 }
0b86a832
CM		 2589
2590 blocksize = btrfs_level_size(tree_root,
2591 btrfs_super_chunk_root_level(disk_super));
84234f3a 2592 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2593
2594 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2595 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2596
2597 chunk_root->node = read_tree_block(chunk_root,
2598 btrfs_super_chunk_root(disk_super),
84234f3a 2599 blocksize, generation);
416bc658
JB		 2600 if (!chunk_root->node ||
2601 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
83121942
DW		 2602 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2603 sb->s_id);
af31f5e5 2604 goto fail_tree_roots;
83121942 2605 }
5d4f98a2
YZ		 2606 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2607 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2608
e17cade2 2609 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 2610 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2611 BTRFS_UUID_SIZE);
e17cade2 2612
0b86a832 2613 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2614 if (ret) {
d397712b
CM		 2615 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2616 sb->s_id);
af31f5e5 2617 goto fail_tree_roots;
2b82032c 2618 }
0b86a832 2619
8dabb742
SB		 2620 /*
2621 * keep the device that is marked to be the target device for the
2622 * dev_replace procedure
2623 */
2624 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2625
a6b0d5c8
CM		 2626 if (!fs_devices->latest_bdev) {
2627 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2628 sb->s_id);
2629 goto fail_tree_roots;
2630 }
2631
af31f5e5 2632retry_root_backup:
db94535d
CM		 2633 blocksize = btrfs_level_size(tree_root,
2634 btrfs_super_root_level(disk_super));
84234f3a 2635 generation = btrfs_super_generation(disk_super);
0b86a832 2636
e20d96d6 2637 tree_root->node = read_tree_block(tree_root,
db94535d 2638 btrfs_super_root(disk_super),
84234f3a 2639 blocksize, generation);
af31f5e5
CM		 2640 if (!tree_root->node ||
2641 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
83121942
DW		 2642 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2643 sb->s_id);
af31f5e5
CM		 2644
2645 goto recovery_tree_root;
83121942 2646 }
af31f5e5 2647
5d4f98a2
YZ		 2648 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2649 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d 2650
cb517eab
MX		 2651 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2652 location.type = BTRFS_ROOT_ITEM_KEY;
2653 location.offset = 0;
2654
2655 extent_root = btrfs_read_tree_root(tree_root, &location);
2656 if (IS_ERR(extent_root)) {
2657 ret = PTR_ERR(extent_root);
af31f5e5 2658 goto recovery_tree_root;
cb517eab 2659 }
0b86a832 2660 extent_root->track_dirty = 1;
cb517eab 2661 fs_info->extent_root = extent_root;
0b86a832 2662
cb517eab
MX		 2663 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2664 dev_root = btrfs_read_tree_root(tree_root, &location);
2665 if (IS_ERR(dev_root)) {
2666 ret = PTR_ERR(dev_root);
af31f5e5 2667 goto recovery_tree_root;
cb517eab 2668 }
5d4f98a2 2669 dev_root->track_dirty = 1;
cb517eab
MX		 2670 fs_info->dev_root = dev_root;
2671 btrfs_init_devices_late(fs_info);
3768f368 2672
cb517eab
MX		 2673 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2674 csum_root = btrfs_read_tree_root(tree_root, &location);
2675 if (IS_ERR(csum_root)) {
2676 ret = PTR_ERR(csum_root);
af31f5e5 2677 goto recovery_tree_root;
cb517eab 2678 }
d20f7043 2679 csum_root->track_dirty = 1;
cb517eab 2680 fs_info->csum_root = csum_root;
d20f7043 2681
cb517eab
MX		 2682 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2683 quota_root = btrfs_read_tree_root(tree_root, &location);
2684 if (!IS_ERR(quota_root)) {
bcef60f2
AJ		 2685 quota_root->track_dirty = 1;
2686 fs_info->quota_enabled = 1;
2687 fs_info->pending_quota_state = 1;
cb517eab 2688 fs_info->quota_root = quota_root;
bcef60f2
AJ		 2689 }
2690
8929ecfa
YZ		 2691 fs_info->generation = generation;
2692 fs_info->last_trans_committed = generation;
8929ecfa 2693
68310a5e
ID		 2694 ret = btrfs_recover_balance(fs_info);
2695 if (ret) {
2696 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2697 goto fail_block_groups;
2698 }
2699
733f4fbb
SB		 2700 ret = btrfs_init_dev_stats(fs_info);
2701 if (ret) {
2702 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2703 ret);
2704 goto fail_block_groups;
2705 }
2706
8dabb742
SB		 2707 ret = btrfs_init_dev_replace(fs_info);
2708 if (ret) {
2709 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2710 goto fail_block_groups;
2711 }
2712
2713 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2714
c59021f8 2715 ret = btrfs_init_space_info(fs_info);
2716 if (ret) {
2717 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2718 goto fail_block_groups;
2719 }
2720
1b1d1f66
JB		 2721 ret = btrfs_read_block_groups(extent_root);
2722 if (ret) {
2723 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2724 goto fail_block_groups;
2725 }
5af3e8cc
SB		 2726 fs_info->num_tolerated_disk_barrier_failures =
2727 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2728 if (fs_info->fs_devices->missing_devices >
2729 fs_info->num_tolerated_disk_barrier_failures &&
2730 !(sb->s_flags & MS_RDONLY)) {
2731 printk(KERN_WARNING
2732 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2733 goto fail_block_groups;
2734 }
9078a3e1 2735
a74a4b97
CM		 2736 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2737 "btrfs-cleaner");
57506d50 2738 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2739 goto fail_block_groups;
a74a4b97
CM		 2740
2741 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2742 tree_root,
2743 "btrfs-transaction");
57506d50 2744 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2745 goto fail_cleaner;
a74a4b97 2746
c289811c
CM		 2747 if (!btrfs_test_opt(tree_root, SSD) &&
2748 !btrfs_test_opt(tree_root, NOSSD) &&
2749 !fs_info->fs_devices->rotating) {
2750 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2751 "mode\n");
2752 btrfs_set_opt(fs_info->mount_opt, SSD);
2753 }
2754
21adbd5c
SB		 2755#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2756 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2757 ret = btrfsic_mount(tree_root, fs_devices,
2758 btrfs_test_opt(tree_root,
2759 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2760 1 : 0,
2761 fs_info->check_integrity_print_mask);
2762 if (ret)
2763 printk(KERN_WARNING "btrfs: failed to initialize"
2764 " integrity check module %s\n", sb->s_id);
2765 }
2766#endif
bcef60f2
AJ		 2767 ret = btrfs_read_qgroup_config(fs_info);
2768 if (ret)
2769 goto fail_trans_kthread;
21adbd5c 2770
acce952b 2771 /* do not make disk changes in broken FS */
68ce9682 2772 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2773 u64 bytenr = btrfs_super_log_root(disk_super);
2774
7c2ca468 2775 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2776 printk(KERN_WARNING "Btrfs log replay required "
2777 "on RO media\n");
7c2ca468 2778 err = -EIO;
bcef60f2 2779 goto fail_qgroup;
7c2ca468 2780 }
e02119d5
CM		 2781 blocksize =
2782 btrfs_level_size(tree_root,
2783 btrfs_super_log_root_level(disk_super));
d18a2c44 2784
6f07e42e 2785 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2786 if (!log_tree_root) {
2787 err = -ENOMEM;
bcef60f2 2788 goto fail_qgroup;
676e4c86 2789 }
e02119d5
CM		 2790
2791 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2792 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2793
2794 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2795 blocksize,
2796 generation + 1);
416bc658
JB		 2797 if (!log_tree_root->node ||
2798 !extent_buffer_uptodate(log_tree_root->node)) {
2799 printk(KERN_ERR "btrfs: failed to read log tree\n");
2800 free_extent_buffer(log_tree_root->node);
2801 kfree(log_tree_root);
2802 goto fail_trans_kthread;
2803 }
79787eaa 2804 /* returns with log_tree_root freed on success */
e02119d5 2805 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2806 if (ret) {
2807 btrfs_error(tree_root->fs_info, ret,
2808 "Failed to recover log tree");
2809 free_extent_buffer(log_tree_root->node);
2810 kfree(log_tree_root);
2811 goto fail_trans_kthread;
2812 }
e556ce2c
YZ		 2813
2814 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2815 ret = btrfs_commit_super(tree_root);
2816 if (ret)
2817 goto fail_trans_kthread;
e556ce2c 2818 }
e02119d5 2819 }
1a40e23b 2820
76dda93c 2821 ret = btrfs_find_orphan_roots(tree_root);
79787eaa
JM		 2822 if (ret)
2823 goto fail_trans_kthread;
76dda93c 2824
7c2ca468 2825 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2826 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2
ID		 2827 if (ret)
2828 goto fail_trans_kthread;
d68fc57b 2829
5d4f98a2 2830 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2831 if (ret < 0) {
2832 printk(KERN_WARNING
2833 "btrfs: failed to recover relocation\n");
2834 err = -EINVAL;
bcef60f2 2835 goto fail_qgroup;
d7ce5843 2836 }
7c2ca468 2837 }
1a40e23b 2838
3de4586c
CM		 2839 location.objectid = BTRFS_FS_TREE_OBJECTID;
2840 location.type = BTRFS_ROOT_ITEM_KEY;
cb517eab 2841 location.offset = 0;
3de4586c 2842
3de4586c 2843 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3140c9a3
DC		 2844 if (IS_ERR(fs_info->fs_root)) {
2845 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2846 goto fail_qgroup;
3140c9a3 2847 }
c289811c 2848
2b6ba629
ID		 2849 if (sb->s_flags & MS_RDONLY)
2850 return 0;
59641015 2851
2b6ba629
ID		 2852 down_read(&fs_info->cleanup_work_sem);
2853 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2854 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2855 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2856 close_ctree(tree_root);
2857 return ret;
2858 }
2859 up_read(&fs_info->cleanup_work_sem);
59641015 2860
2b6ba629
ID		 2861 ret = btrfs_resume_balance_async(fs_info);
2862 if (ret) {
2863 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2864 close_ctree(tree_root);
2865 return ret;
e3acc2a6
JB		 2866 }
2867
8dabb742
SB		 2868 ret = btrfs_resume_dev_replace_async(fs_info);
2869 if (ret) {
2870 pr_warn("btrfs: failed to resume dev_replace\n");
2871 close_ctree(tree_root);
2872 return ret;
2873 }
2874
ad2b2c80 2875 return 0;
39279cc3 2876
bcef60f2
AJ		 2877fail_qgroup:
2878 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2879fail_trans_kthread:
2880 kthread_stop(fs_info->transaction_kthread);
54067ae9 2881 btrfs_cleanup_transaction(fs_info->tree_root);
7b5ff90e 2882 del_fs_roots(fs_info);
3f157a2f 2883fail_cleaner:
a74a4b97 2884 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2885
2886 /*
2887 * make sure we're done with the btree inode before we stop our
2888 * kthreads
2889 */
2890 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2891
1b1d1f66 2892fail_block_groups:
54067ae9 2893 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2894 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2895
2896fail_tree_roots:
2897 free_root_pointers(fs_info, 1);
2b8195bb 2898 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2899
39279cc3 2900fail_sb_buffer:
7abadb64 2901 btrfs_stop_all_workers(fs_info);
16cdcec7 2902fail_alloc:
4543df7e 2903fail_iput:
586e46e2
ID		 2904 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2905
4543df7e 2906 iput(fs_info->btree_inode);
963d678b
MX		 2907fail_delalloc_bytes:
2908 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 2909fail_dirty_metadata_bytes:
2910 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 2911fail_bdi:
7e662854 2912 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2913fail_srcu:
2914 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2915fail:
53b381b3 2916 btrfs_free_stripe_hash_table(fs_info);
586e46e2 2917 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 2918 return err;
af31f5e5
CM		 2919
2920recovery_tree_root:
af31f5e5
CM		 2921 if (!btrfs_test_opt(tree_root, RECOVERY))
2922 goto fail_tree_roots;
2923
2924 free_root_pointers(fs_info, 0);
2925
2926 /* don't use the log in recovery mode, it won't be valid */
2927 btrfs_set_super_log_root(disk_super, 0);
2928
2929 /* we can't trust the free space cache either */
2930 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2931
2932 ret = next_root_backup(fs_info, fs_info->super_copy,
2933 &num_backups_tried, &backup_index);
2934 if (ret == -1)
2935 goto fail_block_groups;
2936 goto retry_root_backup;
eb60ceac
CM		 2937}
2938
f2984462
CM		 2939static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2940{
f2984462
CM		 2941 if (uptodate) {
2942 set_buffer_uptodate(bh);
2943 } else {
442a4f63
SB		 2944 struct btrfs_device *device = (struct btrfs_device *)
2945 bh->b_private;
2946
606686ee
JB		 2947 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2948 "I/O error on %s\n",
2949 rcu_str_deref(device->name));
1259ab75
CM		 2950 /* note, we dont' set_buffer_write_io_error because we have
2951 * our own ways of dealing with the IO errors
2952 */
f2984462 2953 clear_buffer_uptodate(bh);
442a4f63 2954 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 2955 }
2956 unlock_buffer(bh);
2957 put_bh(bh);
2958}
2959
a512bbf8
YZ		 2960struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2961{
2962 struct buffer_head *bh;
2963 struct buffer_head *latest = NULL;
2964 struct btrfs_super_block *super;
2965 int i;
2966 u64 transid = 0;
2967 u64 bytenr;
2968
2969 /* we would like to check all the supers, but that would make
2970 * a btrfs mount succeed after a mkfs from a different FS.
2971 * So, we need to add a special mount option to scan for
2972 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2973 */
2974 for (i = 0; i < 1; i++) {
2975 bytenr = btrfs_sb_offset(i);
2976 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2977 break;
2978 bh = __bread(bdev, bytenr / 4096, 4096);
2979 if (!bh)
2980 continue;
2981
2982 super = (struct btrfs_super_block *)bh->b_data;
2983 if (btrfs_super_bytenr(super) != bytenr ||
cdb4c574 2984 super->magic != cpu_to_le64(BTRFS_MAGIC)) {
a512bbf8
YZ		 2985 brelse(bh);
2986 continue;
2987 }
2988
2989 if (!latest || btrfs_super_generation(super) > transid) {
2990 brelse(latest);
2991 latest = bh;
2992 transid = btrfs_super_generation(super);
2993 } else {
2994 brelse(bh);
2995 }
2996 }
2997 return latest;
2998}
2999
4eedeb75
HH		 3000/*
3001 * this should be called twice, once with wait == 0 and
3002 * once with wait == 1. When wait == 0 is done, all the buffer heads
3003 * we write are pinned.
3004 *
3005 * They are released when wait == 1 is done.
3006 * max_mirrors must be the same for both runs, and it indicates how
3007 * many supers on this one device should be written.
3008 *
3009 * max_mirrors == 0 means to write them all.
3010 */
a512bbf8
YZ		 3011static int write_dev_supers(struct btrfs_device *device,
3012 struct btrfs_super_block *sb,
3013 int do_barriers, int wait, int max_mirrors)
3014{
3015 struct buffer_head *bh;
3016 int i;
3017 int ret;
3018 int errors = 0;
3019 u32 crc;
3020 u64 bytenr;
a512bbf8
YZ		 3021
3022 if (max_mirrors == 0)
3023 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3024
a512bbf8
YZ		 3025 for (i = 0; i < max_mirrors; i++) {
3026 bytenr = btrfs_sb_offset(i);
3027 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3028 break;
3029
3030 if (wait) {
3031 bh = __find_get_block(device->bdev, bytenr / 4096,
3032 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3033 if (!bh) {
3034 errors++;
3035 continue;
3036 }
a512bbf8 3037 wait_on_buffer(bh);
4eedeb75
HH		 3038 if (!buffer_uptodate(bh))
3039 errors++;
3040
3041 /* drop our reference */
3042 brelse(bh);
3043
3044 /* drop the reference from the wait == 0 run */
3045 brelse(bh);
3046 continue;
a512bbf8
YZ		 3047 } else {
3048 btrfs_set_super_bytenr(sb, bytenr);
3049
3050 crc = ~(u32)0;
b0496686 3051 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 3052 BTRFS_CSUM_SIZE, crc,
3053 BTRFS_SUPER_INFO_SIZE -
3054 BTRFS_CSUM_SIZE);
3055 btrfs_csum_final(crc, sb->csum);
3056
4eedeb75
HH		 3057 /*
3058 * one reference for us, and we leave it for the
3059 * caller
3060 */
a512bbf8
YZ		 3061 bh = __getblk(device->bdev, bytenr / 4096,
3062 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3063 if (!bh) {
3064 printk(KERN_ERR "btrfs: couldn't get super "
3065 "buffer head for bytenr %Lu\n", bytenr);
3066 errors++;
3067 continue;
3068 }
3069
a512bbf8
YZ		 3070 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3071
4eedeb75 3072 /* one reference for submit_bh */
a512bbf8 3073 get_bh(bh);
4eedeb75
HH		 3074
3075 set_buffer_uptodate(bh);
a512bbf8
YZ		 3076 lock_buffer(bh);
3077 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3078 bh->b_private = device;
a512bbf8
YZ		 3079 }
3080
387125fc
CM		 3081 /*
3082 * we fua the first super. The others we allow
3083 * to go down lazy.
3084 */
21adbd5c 3085 ret = btrfsic_submit_bh(WRITE_FUA, bh);
4eedeb75 3086 if (ret)
a512bbf8 3087 errors++;
a512bbf8
YZ		 3088 }
3089 return errors < i ? 0 : -1;
3090}
3091
387125fc
CM		 3092/*
3093 * endio for the write_dev_flush, this will wake anyone waiting
3094 * for the barrier when it is done
3095 */
3096static void btrfs_end_empty_barrier(struct bio *bio, int err)
3097{
3098 if (err) {
3099 if (err == -EOPNOTSUPP)
3100 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3101 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3102 }
3103 if (bio->bi_private)
3104 complete(bio->bi_private);
3105 bio_put(bio);
3106}
3107
3108/*
3109 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3110 * sent down. With wait == 1, it waits for the previous flush.
3111 *
3112 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3113 * capable
3114 */
3115static int write_dev_flush(struct btrfs_device *device, int wait)
3116{
3117 struct bio *bio;
3118 int ret = 0;
3119
3120 if (device->nobarriers)
3121 return 0;
3122
3123 if (wait) {
3124 bio = device->flush_bio;
3125 if (!bio)
3126 return 0;
3127
3128 wait_for_completion(&device->flush_wait);
3129
3130 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
606686ee
JB		 3131 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3132 rcu_str_deref(device->name));
387125fc 3133 device->nobarriers = 1;
5af3e8cc 3134 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3135 ret = -EIO;
5af3e8cc
SB		 3136 btrfs_dev_stat_inc_and_print(device,
3137 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3138 }
3139
3140 /* drop the reference from the wait == 0 run */
3141 bio_put(bio);
3142 device->flush_bio = NULL;
3143
3144 return ret;
3145 }
3146
3147 /*
3148 * one reference for us, and we leave it for the
3149 * caller
3150 */
9c017abc 3151 device->flush_bio = NULL;
9be3395b 3152 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
387125fc
CM		 3153 if (!bio)
3154 return -ENOMEM;
3155
3156 bio->bi_end_io = btrfs_end_empty_barrier;
3157 bio->bi_bdev = device->bdev;
3158 init_completion(&device->flush_wait);
3159 bio->bi_private = &device->flush_wait;
3160 device->flush_bio = bio;
3161
3162 bio_get(bio);
21adbd5c 3163 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3164
3165 return 0;
3166}
3167
3168/*
3169 * send an empty flush down to each device in parallel,
3170 * then wait for them
3171 */
3172static int barrier_all_devices(struct btrfs_fs_info *info)
3173{
3174 struct list_head *head;
3175 struct btrfs_device *dev;
5af3e8cc
SB		 3176 int errors_send = 0;
3177 int errors_wait = 0;
387125fc
CM		 3178 int ret;
3179
3180 /* send down all the barriers */
3181 head = &info->fs_devices->devices;
3182 list_for_each_entry_rcu(dev, head, dev_list) {
3183 if (!dev->bdev) {
5af3e8cc 3184 errors_send++;
387125fc
CM		 3185 continue;
3186 }
3187 if (!dev->in_fs_metadata || !dev->writeable)
3188 continue;
3189
3190 ret = write_dev_flush(dev, 0);
3191 if (ret)
5af3e8cc 3192 errors_send++;
387125fc
CM		 3193 }
3194
3195 /* wait for all the barriers */
3196 list_for_each_entry_rcu(dev, head, dev_list) {
3197 if (!dev->bdev) {
5af3e8cc 3198 errors_wait++;
387125fc
CM		 3199 continue;
3200 }
3201 if (!dev->in_fs_metadata || !dev->writeable)
3202 continue;
3203
3204 ret = write_dev_flush(dev, 1);
3205 if (ret)
5af3e8cc 3206 errors_wait++;
387125fc 3207 }
5af3e8cc
SB		 3208 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3209 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3210 return -EIO;
3211 return 0;
3212}
3213
5af3e8cc
SB		 3214int btrfs_calc_num_tolerated_disk_barrier_failures(
3215 struct btrfs_fs_info *fs_info)
3216{
3217 struct btrfs_ioctl_space_info space;
3218 struct btrfs_space_info *sinfo;
3219 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3220 BTRFS_BLOCK_GROUP_SYSTEM,
3221 BTRFS_BLOCK_GROUP_METADATA,
3222 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3223 int num_types = 4;
3224 int i;
3225 int c;
3226 int num_tolerated_disk_barrier_failures =
3227 (int)fs_info->fs_devices->num_devices;
3228
3229 for (i = 0; i < num_types; i++) {
3230 struct btrfs_space_info *tmp;
3231
3232 sinfo = NULL;
3233 rcu_read_lock();
3234 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3235 if (tmp->flags == types[i]) {
3236 sinfo = tmp;
3237 break;
3238 }
3239 }
3240 rcu_read_unlock();
3241
3242 if (!sinfo)
3243 continue;
3244
3245 down_read(&sinfo->groups_sem);
3246 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3247 if (!list_empty(&sinfo->block_groups[c])) {
3248 u64 flags;
3249
3250 btrfs_get_block_group_info(
3251 &sinfo->block_groups[c], &space);
3252 if (space.total_bytes == 0 ||
3253 space.used_bytes == 0)
3254 continue;
3255 flags = space.flags;
3256 /*
3257 * return
3258 * 0: if dup, single or RAID0 is configured for
3259 * any of metadata, system or data, else
3260 * 1: if RAID5 is configured, or if RAID1 or
3261 * RAID10 is configured and only two mirrors
3262 * are used, else
3263 * 2: if RAID6 is configured, else
3264 * num_mirrors - 1: if RAID1 or RAID10 is
3265 * configured and more than
3266 * 2 mirrors are used.
3267 */
3268 if (num_tolerated_disk_barrier_failures > 0 &&
3269 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3270 BTRFS_BLOCK_GROUP_RAID0)) ||
3271 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3272 == 0)))
3273 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3274 else if (num_tolerated_disk_barrier_failures > 1) {
3275 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3276 BTRFS_BLOCK_GROUP_RAID5 |
3277 BTRFS_BLOCK_GROUP_RAID10)) {
3278 num_tolerated_disk_barrier_failures = 1;
3279 } else if (flags &
15b0a89d 3280 BTRFS_BLOCK_GROUP_RAID6) {
53b381b3
DW		 3281 num_tolerated_disk_barrier_failures = 2;
3282 }
3283 }
5af3e8cc
SB		 3284 }
3285 }
3286 up_read(&sinfo->groups_sem);
3287 }
3288
3289 return num_tolerated_disk_barrier_failures;
3290}
3291
48a3b636 3292static int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3293{
e5e9a520 3294 struct list_head *head;
f2984462 3295 struct btrfs_device *dev;
a061fc8d 3296 struct btrfs_super_block *sb;
f2984462 3297 struct btrfs_dev_item *dev_item;
f2984462
CM		 3298 int ret;
3299 int do_barriers;
a236aed1
CM		 3300 int max_errors;
3301 int total_errors = 0;
a061fc8d 3302 u64 flags;
f2984462 3303
6c41761f 3304 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
f2984462 3305 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3306 backup_super_roots(root->fs_info);
f2984462 3307
6c41761f 3308 sb = root->fs_info->super_for_commit;
a061fc8d 3309 dev_item = &sb->dev_item;
e5e9a520 3310
174ba509 3311 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3312 head = &root->fs_info->fs_devices->devices;
387125fc 3313
5af3e8cc
SB		 3314 if (do_barriers) {
3315 ret = barrier_all_devices(root->fs_info);
3316 if (ret) {
3317 mutex_unlock(
3318 &root->fs_info->fs_devices->device_list_mutex);
3319 btrfs_error(root->fs_info, ret,
3320 "errors while submitting device barriers.");
3321 return ret;
3322 }
3323 }
387125fc 3324
1f78160c 3325 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3326 if (!dev->bdev) {
3327 total_errors++;
3328 continue;
3329 }
2b82032c 3330 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3331 continue;
3332
2b82032c 3333 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3334 btrfs_set_stack_device_type(dev_item, dev->type);
3335 btrfs_set_stack_device_id(dev_item, dev->devid);
3336 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3337 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3338 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3339 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3340 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3341 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3342 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3343
a061fc8d
CM		 3344 flags = btrfs_super_flags(sb);
3345 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3346
a512bbf8 3347 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3348 if (ret)
3349 total_errors++;
f2984462 3350 }
a236aed1 3351 if (total_errors > max_errors) {
d397712b
CM		 3352 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3353 total_errors);
79787eaa
JM		 3354
3355 /* This shouldn't happen. FUA is masked off if unsupported */
a236aed1
CM		 3356 BUG();
3357 }
f2984462 3358
a512bbf8 3359 total_errors = 0;
1f78160c 3360 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3361 if (!dev->bdev)
3362 continue;
2b82032c 3363 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3364 continue;
3365
a512bbf8
YZ		 3366 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3367 if (ret)
3368 total_errors++;
f2984462 3369 }
174ba509 3370 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3371 if (total_errors > max_errors) {
79787eaa
JM		 3372 btrfs_error(root->fs_info, -EIO,
3373 "%d errors while writing supers", total_errors);
3374 return -EIO;
a236aed1 3375 }
f2984462
CM		 3376 return 0;
3377}
3378
a512bbf8
YZ		 3379int write_ctree_super(struct btrfs_trans_handle *trans,
3380 struct btrfs_root *root, int max_mirrors)
eb60ceac 3381{
e66f709b 3382 int ret;
5f39d397 3383
a512bbf8 3384 ret = write_all_supers(root, max_mirrors);
5f39d397 3385 return ret;
cfaa7295
CM		 3386}
3387
cb517eab
MX		 3388/* Drop a fs root from the radix tree and free it. */
3389void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3390 struct btrfs_root *root)
2619ba1f 3391{
4df27c4d 3392 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3393 radix_tree_delete(&fs_info->fs_roots_radix,
3394 (unsigned long)root->root_key.objectid);
4df27c4d 3395 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3396
3397 if (btrfs_root_refs(&root->root_item) == 0)
3398 synchronize_srcu(&fs_info->subvol_srcu);
3399
d7634482 3400 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3321719e
LB		 3401 btrfs_free_log(NULL, root);
3402 btrfs_free_log_root_tree(NULL, fs_info);
3403 }
3404
581bb050
LZ		 3405 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3406 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3407 free_fs_root(root);
4df27c4d
YZ		 3408}
3409
3410static void free_fs_root(struct btrfs_root *root)
3411{
82d5902d 3412 iput(root->cache_inode);
4df27c4d 3413 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
0ee5dc67
AV		 3414 if (root->anon_dev)
3415 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 3416 free_extent_buffer(root->node);
3417 free_extent_buffer(root->commit_root);
581bb050
LZ		 3418 kfree(root->free_ino_ctl);
3419 kfree(root->free_ino_pinned);
d397712b 3420 kfree(root->name);
b0feb9d9 3421 btrfs_put_fs_root(root);
2619ba1f
CM		 3422}
3423
cb517eab
MX		 3424void btrfs_free_fs_root(struct btrfs_root *root)
3425{
3426 free_fs_root(root);
3427}
3428
c146afad 3429int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3430{
c146afad
YZ		 3431 u64 root_objectid = 0;
3432 struct btrfs_root *gang[8];
3433 int i;
3768f368 3434 int ret;
e089f05c 3435
c146afad
YZ		 3436 while (1) {
3437 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3438 (void **)gang, root_objectid,
3439 ARRAY_SIZE(gang));
3440 if (!ret)
3441 break;
5d4f98a2
YZ		 3442
3443 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 3444 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 3445 int err;
3446
c146afad 3447 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3448 err = btrfs_orphan_cleanup(gang[i]);
3449 if (err)
3450 return err;
c146afad
YZ		 3451 }
3452 root_objectid++;
3453 }
3454 return 0;
3455}
a2135011 3456
c146afad
YZ		 3457int btrfs_commit_super(struct btrfs_root *root)
3458{
3459 struct btrfs_trans_handle *trans;
3460 int ret;
a74a4b97 3461
c146afad 3462 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3463 btrfs_run_delayed_iputs(root);
c146afad 3464 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3465 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3466
3467 /* wait until ongoing cleanup work done */
3468 down_write(&root->fs_info->cleanup_work_sem);
3469 up_write(&root->fs_info->cleanup_work_sem);
3470
7a7eaa40 3471 trans = btrfs_join_transaction(root);
3612b495
TI		 3472 if (IS_ERR(trans))
3473 return PTR_ERR(trans);
54aa1f4d 3474 ret = btrfs_commit_transaction(trans, root);
79787eaa
JM		 3475 if (ret)
3476 return ret;
c146afad 3477 /* run commit again to drop the original snapshot */
7a7eaa40 3478 trans = btrfs_join_transaction(root);
3612b495
TI		 3479 if (IS_ERR(trans))
3480 return PTR_ERR(trans);
79787eaa
JM		 3481 ret = btrfs_commit_transaction(trans, root);
3482 if (ret)
3483 return ret;
79154b1b 3484 ret = btrfs_write_and_wait_transaction(NULL, root);
79787eaa
JM		 3485 if (ret) {
3486 btrfs_error(root->fs_info, ret,
3487 "Failed to sync btree inode to disk.");
3488 return ret;
3489 }
d6bfde87 3490
a512bbf8 3491 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 3492 return ret;
3493}
3494
3495int close_ctree(struct btrfs_root *root)
3496{
3497 struct btrfs_fs_info *fs_info = root->fs_info;
3498 int ret;
3499
3500 fs_info->closing = 1;
3501 smp_mb();
3502
837d5b6e 3503 /* pause restriper - we want to resume on mount */
aa1b8cd4 3504 btrfs_pause_balance(fs_info);
837d5b6e 3505
8dabb742
SB		 3506 btrfs_dev_replace_suspend_for_unmount(fs_info);
3507
aa1b8cd4 3508 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3509
3510 /* wait for any defraggers to finish */
3511 wait_event(fs_info->transaction_wait,
3512 (atomic_read(&fs_info->defrag_running) == 0));
3513
3514 /* clear out the rbtree of defraggable inodes */
26176e7c 3515 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3516
c146afad 3517 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3518 ret = btrfs_commit_super(root);
3519 if (ret)
3520 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3521 }
3522
87533c47 3523 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3524 btrfs_error_commit_super(root);
0f7d52f4 3525
300e4f8a
JB		 3526 btrfs_put_block_group_cache(fs_info);
3527
e3029d9f
AV		 3528 kthread_stop(fs_info->transaction_kthread);
3529 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3530
f25784b3
YZ		 3531 fs_info->closing = 2;
3532 smp_mb();
3533
bcef60f2
AJ		 3534 btrfs_free_qgroup_config(root->fs_info);
3535
963d678b
MX		 3536 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3537 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3538 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3539 }
bcc63abb 3540
e3029d9f 3541 btrfs_free_block_groups(fs_info);
d10c5f31 3542
13e6c37b 3543 btrfs_stop_all_workers(fs_info);
2932505a 3544
c146afad 3545 del_fs_roots(fs_info);
d10c5f31 3546
13e6c37b 3547 free_root_pointers(fs_info, 1);
9ad6b7bc 3548
13e6c37b 3549 iput(fs_info->btree_inode);
d6bfde87 3550
21adbd5c
SB		 3551#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3552 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3553 btrfsic_unmount(root, fs_info->fs_devices);
3554#endif
3555
dfe25020 3556 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3557 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3558
e2d84521 3559 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3560 percpu_counter_destroy(&fs_info->delalloc_bytes);
04160088 3561 bdi_destroy(&fs_info->bdi);
76dda93c 3562 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3563
53b381b3
DW		 3564 btrfs_free_stripe_hash_table(fs_info);
3565
eb60ceac
CM		 3566 return 0;
3567}
3568
b9fab919
CM		 3569int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3570 int atomic)
5f39d397 3571{
1259ab75 3572 int ret;
727011e0 3573 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3574
0b32f4bb 3575 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3576 if (!ret)
3577 return ret;
3578
3579 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3580 parent_transid, atomic);
3581 if (ret == -EAGAIN)
3582 return ret;
1259ab75 3583 return !ret;
5f39d397
CM		 3584}
3585
3586int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3587{
0b32f4bb 3588 return set_extent_buffer_uptodate(buf);
5f39d397 3589}
6702ed49 3590
5f39d397
CM		 3591void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3592{
727011e0 3593 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
5f39d397 3594 u64 transid = btrfs_header_generation(buf);
b9473439 3595 int was_dirty;
b4ce94de 3596
b9447ef8 3597 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3598 if (transid != root->fs_info->generation)
3599 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3600 "found %llu running %llu\n",
db94535d 3601 (unsigned long long)buf->start,
d397712b
CM		 3602 (unsigned long long)transid,
3603 (unsigned long long)root->fs_info->generation);
0b32f4bb 3604 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3605 if (!was_dirty)
3606 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3607 buf->len,
3608 root->fs_info->dirty_metadata_batch);
eb60ceac
CM		 3609}
3610
b53d3f5d
LB		 3611static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3612 int flush_delayed)
16cdcec7
MX		 3613{
3614 /*
3615 * looks as though older kernels can get into trouble with
3616 * this code, they end up stuck in balance_dirty_pages forever
3617 */
e2d84521 3618 int ret;
16cdcec7
MX		 3619
3620 if (current->flags & PF_MEMALLOC)
3621 return;
3622
b53d3f5d
LB		 3623 if (flush_delayed)
3624 btrfs_balance_delayed_items(root);
16cdcec7 3625
e2d84521
MX		 3626 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3627 BTRFS_DIRTY_METADATA_THRESH);
3628 if (ret > 0) {
d0e1d66b
NJ		 3629 balance_dirty_pages_ratelimited(
3630 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3631 }
3632 return;
3633}
3634
b53d3f5d 3635void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3636{
b53d3f5d
LB		 3637 __btrfs_btree_balance_dirty(root, 1);
3638}
585ad2c3 3639
b53d3f5d
LB		 3640void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3641{
3642 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3643}
6b80053d 3644
ca7a79ad 3645int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3646{
727011e0 3647 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3648 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3649}
0da5468f 3650
fcd1f065 3651static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3652 int read_only)
3653{
1104a885
DS		 3654 /*
3655 * Placeholder for checks
3656 */
fcd1f065 3657 return 0;
acce952b 3658}
3659
48a3b636 3660static void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3661{
acce952b 3662 mutex_lock(&root->fs_info->cleaner_mutex);
3663 btrfs_run_delayed_iputs(root);
3664 mutex_unlock(&root->fs_info->cleaner_mutex);
3665
3666 down_write(&root->fs_info->cleanup_work_sem);
3667 up_write(&root->fs_info->cleanup_work_sem);
3668
3669 /* cleanup FS via transaction */
3670 btrfs_cleanup_transaction(root);
acce952b 3671}
3672
569e0f35
JB		 3673static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3674 struct btrfs_root *root)
acce952b 3675{
3676 struct btrfs_inode *btrfs_inode;
3677 struct list_head splice;
3678
3679 INIT_LIST_HEAD(&splice);
3680
3681 mutex_lock(&root->fs_info->ordered_operations_mutex);
3682 spin_lock(&root->fs_info->ordered_extent_lock);
3683
569e0f35 3684 list_splice_init(&t->ordered_operations, &splice);
acce952b 3685 while (!list_empty(&splice)) {
3686 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3687 ordered_operations);
3688
3689 list_del_init(&btrfs_inode->ordered_operations);
b216cbfb 3690 spin_unlock(&root->fs_info->ordered_extent_lock);
acce952b 3691
3692 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb
MX		 3693
3694 spin_lock(&root->fs_info->ordered_extent_lock);
acce952b 3695 }
3696
3697 spin_unlock(&root->fs_info->ordered_extent_lock);
3698 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3699}
3700
143bede5 3701static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3702{
acce952b 3703 struct btrfs_ordered_extent *ordered;
acce952b 3704
3705 spin_lock(&root->fs_info->ordered_extent_lock);
779880ef
JB		 3706 /*
3707 * This will just short circuit the ordered completion stuff which will
3708 * make sure the ordered extent gets properly cleaned up.
3709 */
3710 list_for_each_entry(ordered, &root->fs_info->ordered_extents,
3711 root_extent_list)
3712 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
acce952b 3713 spin_unlock(&root->fs_info->ordered_extent_lock);
acce952b 3714}
3715
79787eaa
JM		 3716int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3717 struct btrfs_root *root)
acce952b 3718{
3719 struct rb_node *node;
3720 struct btrfs_delayed_ref_root *delayed_refs;
3721 struct btrfs_delayed_ref_node *ref;
3722 int ret = 0;
3723
3724 delayed_refs = &trans->delayed_refs;
3725
3726 spin_lock(&delayed_refs->lock);
3727 if (delayed_refs->num_entries == 0) {
cfece4db 3728 spin_unlock(&delayed_refs->lock);
acce952b 3729 printk(KERN_INFO "delayed_refs has NO entry\n");
3730 return ret;
3731 }
3732
b939d1ab 3733 while ((node = rb_first(&delayed_refs->root)) != NULL) {
eb12db69 3734 struct btrfs_delayed_ref_head *head = NULL;
acce952b 3735
eb12db69 3736 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
acce952b 3737 atomic_set(&ref->refs, 1);
3738 if (btrfs_delayed_ref_is_head(ref)) {
acce952b 3739
3740 head = btrfs_delayed_node_to_head(ref);
b939d1ab
JB		 3741 if (!mutex_trylock(&head->mutex)) {
3742 atomic_inc(&ref->refs);
3743 spin_unlock(&delayed_refs->lock);
3744
3745 /* Need to wait for the delayed ref to run */
3746 mutex_lock(&head->mutex);
3747 mutex_unlock(&head->mutex);
3748 btrfs_put_delayed_ref(ref);
3749
e18fca73 3750 spin_lock(&delayed_refs->lock);
b939d1ab
JB		 3751 continue;
3752 }
3753
54067ae9
JB		 3754 if (head->must_insert_reserved)
3755 btrfs_pin_extent(root, ref->bytenr,
3756 ref->num_bytes, 1);
78a6184a 3757 btrfs_free_delayed_extent_op(head->extent_op);
acce952b 3758 delayed_refs->num_heads--;
3759 if (list_empty(&head->cluster))
3760 delayed_refs->num_heads_ready--;
3761 list_del_init(&head->cluster);
acce952b 3762 }
eb12db69 3763
b939d1ab
JB		 3764 ref->in_tree = 0;
3765 rb_erase(&ref->rb_node, &delayed_refs->root);
3766 delayed_refs->num_entries--;
eb12db69
JB		 3767 if (head)
3768 mutex_unlock(&head->mutex);
acce952b 3769 spin_unlock(&delayed_refs->lock);
3770 btrfs_put_delayed_ref(ref);
3771
3772 cond_resched();
3773 spin_lock(&delayed_refs->lock);
3774 }
3775
3776 spin_unlock(&delayed_refs->lock);
3777
3778 return ret;
3779}
3780
aec8030a 3781static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
acce952b 3782{
3783 struct btrfs_pending_snapshot *snapshot;
3784 struct list_head splice;
3785
3786 INIT_LIST_HEAD(&splice);
3787
3788 list_splice_init(&t->pending_snapshots, &splice);
3789
3790 while (!list_empty(&splice)) {
3791 snapshot = list_entry(splice.next,
3792 struct btrfs_pending_snapshot,
3793 list);
aec8030a 3794 snapshot->error = -ECANCELED;
acce952b 3795 list_del_init(&snapshot->list);
acce952b 3796 }
acce952b 3797}
3798
143bede5 3799static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3800{
3801 struct btrfs_inode *btrfs_inode;
3802 struct list_head splice;
3803
3804 INIT_LIST_HEAD(&splice);
3805
acce952b 3806 spin_lock(&root->fs_info->delalloc_lock);
5be76758 3807 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
acce952b 3808
3809 while (!list_empty(&splice)) {
3810 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3811 delalloc_inodes);
3812
3813 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3814 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3815 &btrfs_inode->runtime_flags);
b216cbfb 3816 spin_unlock(&root->fs_info->delalloc_lock);
acce952b 3817
3818 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb
MX		 3819
3820 spin_lock(&root->fs_info->delalloc_lock);
acce952b 3821 }
3822
3823 spin_unlock(&root->fs_info->delalloc_lock);
acce952b 3824}
3825
3826static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3827 struct extent_io_tree *dirty_pages,
3828 int mark)
3829{
3830 int ret;
acce952b 3831 struct extent_buffer *eb;
3832 u64 start = 0;
3833 u64 end;
acce952b 3834
3835 while (1) {
3836 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3837 mark, NULL);
acce952b 3838 if (ret)
3839 break;
3840
3841 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3842 while (start <= end) {
fd8b2b61
JB		 3843 eb = btrfs_find_tree_block(root, start,
3844 root->leafsize);
69a85bd8 3845 start += root->leafsize;
fd8b2b61 3846 if (!eb)
acce952b 3847 continue;
fd8b2b61 3848 wait_on_extent_buffer_writeback(eb);
acce952b 3849
fd8b2b61
JB		 3850 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3851 &eb->bflags))
3852 clear_extent_buffer_dirty(eb);
3853 free_extent_buffer_stale(eb);
acce952b 3854 }
3855 }
3856
3857 return ret;
3858}
3859
3860static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3861 struct extent_io_tree *pinned_extents)
3862{
3863 struct extent_io_tree *unpin;
3864 u64 start;
3865 u64 end;
3866 int ret;
ed0eaa14 3867 bool loop = true;
acce952b 3868
3869 unpin = pinned_extents;
ed0eaa14 3870again:
acce952b 3871 while (1) {
3872 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 3873 EXTENT_DIRTY, NULL);
acce952b 3874 if (ret)
3875 break;
3876
3877 /* opt_discard */
5378e607
LD		 3878 if (btrfs_test_opt(root, DISCARD))
3879 ret = btrfs_error_discard_extent(root, start,
3880 end + 1 - start,
3881 NULL);
acce952b 3882
3883 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3884 btrfs_error_unpin_extent_range(root, start, end);
3885 cond_resched();
3886 }
3887
ed0eaa14
LB		 3888 if (loop) {
3889 if (unpin == &root->fs_info->freed_extents[0])
3890 unpin = &root->fs_info->freed_extents[1];
3891 else
3892 unpin = &root->fs_info->freed_extents[0];
3893 loop = false;
3894 goto again;
3895 }
3896
acce952b 3897 return 0;
3898}
3899
49b25e05
JM		 3900void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3901 struct btrfs_root *root)
3902{
3903 btrfs_destroy_delayed_refs(cur_trans, root);
3904 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3905 cur_trans->dirty_pages.dirty_bytes);
3906
3907 /* FIXME: cleanup wait for commit */
3908 cur_trans->in_commit = 1;
3909 cur_trans->blocked = 1;
d7096fc3 3910 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 3911
aec8030a
MX		 3912 btrfs_evict_pending_snapshots(cur_trans);
3913
49b25e05 3914 cur_trans->blocked = 0;
d7096fc3 3915 wake_up(&root->fs_info->transaction_wait);
49b25e05
JM		 3916
3917 cur_trans->commit_done = 1;
d7096fc3 3918 wake_up(&cur_trans->commit_wait);
49b25e05 3919
67cde344
MX		 3920 btrfs_destroy_delayed_inodes(root);
3921 btrfs_assert_delayed_root_empty(root);
49b25e05 3922
49b25e05
JM		 3923 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3924 EXTENT_DIRTY);
6e841e32
LB		 3925 btrfs_destroy_pinned_extent(root,
3926 root->fs_info->pinned_extents);
49b25e05
JM		 3927
3928 /*
3929 memset(cur_trans, 0, sizeof(*cur_trans));
3930 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3931 */
3932}
3933
48a3b636 3934static int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 3935{
3936 struct btrfs_transaction *t;
3937 LIST_HEAD(list);
3938
acce952b 3939 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3940
a4abeea4 3941 spin_lock(&root->fs_info->trans_lock);
acce952b 3942 list_splice_init(&root->fs_info->trans_list, &list);
a4abeea4
JB		 3943 root->fs_info->trans_no_join = 1;
3944 spin_unlock(&root->fs_info->trans_lock);
3945
acce952b 3946 while (!list_empty(&list)) {
3947 t = list_entry(list.next, struct btrfs_transaction, list);
acce952b 3948
569e0f35 3949 btrfs_destroy_ordered_operations(t, root);
acce952b 3950
3951 btrfs_destroy_ordered_extents(root);
3952
3953 btrfs_destroy_delayed_refs(t, root);
3954
acce952b 3955 /* FIXME: cleanup wait for commit */
3956 t->in_commit = 1;
3957 t->blocked = 1;
66657b31 3958 smp_mb();
acce952b 3959 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3960 wake_up(&root->fs_info->transaction_blocked_wait);
3961
aec8030a
MX		 3962 btrfs_evict_pending_snapshots(t);
3963
acce952b 3964 t->blocked = 0;
66657b31 3965 smp_mb();
acce952b 3966 if (waitqueue_active(&root->fs_info->transaction_wait))
3967 wake_up(&root->fs_info->transaction_wait);
acce952b 3968
acce952b 3969 t->commit_done = 1;
66657b31 3970 smp_mb();
acce952b 3971 if (waitqueue_active(&t->commit_wait))
3972 wake_up(&t->commit_wait);
acce952b 3973
67cde344
MX		 3974 btrfs_destroy_delayed_inodes(root);
3975 btrfs_assert_delayed_root_empty(root);
3976
acce952b 3977 btrfs_destroy_delalloc_inodes(root);
3978
a4abeea4 3979 spin_lock(&root->fs_info->trans_lock);
acce952b 3980 root->fs_info->running_transaction = NULL;
a4abeea4 3981 spin_unlock(&root->fs_info->trans_lock);
acce952b 3982
3983 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3984 EXTENT_DIRTY);
3985
3986 btrfs_destroy_pinned_extent(root,
3987 root->fs_info->pinned_extents);
3988
13c5a93e 3989 atomic_set(&t->use_count, 0);
acce952b 3990 list_del_init(&t->list);
3991 memset(t, 0, sizeof(*t));
3992 kmem_cache_free(btrfs_transaction_cachep, t);
3993 }
3994
a4abeea4
JB		 3995 spin_lock(&root->fs_info->trans_lock);
3996 root->fs_info->trans_no_join = 0;
3997 spin_unlock(&root->fs_info->trans_lock);
acce952b 3998 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 3999
4000 return 0;
4001}
4002
d1310b2e 4003static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 4004 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 4005 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 4006 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 4007 /* note we're sharing with inode.c for the merge bio hook */
4008 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 4009};
Linux 6.x block layer and io_uring tree(s)