projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: maintain subvolume items in the UUID 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
facc8a22
MX		 579static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
580 u64 phy_offset, struct page *page,
581 u64 start, u64 end, int mirror)
ce9adaa5
CM		 582{
583 struct extent_io_tree *tree;
584 u64 found_start;
585 int found_level;
ce9adaa5
CM		 586 struct extent_buffer *eb;
587 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 588 int ret = 0;
727011e0 589 int reads_done;
ce9adaa5 590
ce9adaa5
CM		 591 if (!page->private)
592 goto out;
d397712b 593
727011e0 594 tree = &BTRFS_I(page->mapping->host)->io_tree;
4f2de97a 595 eb = (struct extent_buffer *)page->private;
d397712b 596
0b32f4bb
JB		 597 /* the pending IO might have been the only thing that kept this buffer
598 * in memory. Make sure we have a ref for all this other checks
599 */
600 extent_buffer_get(eb);
601
602 reads_done = atomic_dec_and_test(&eb->io_pages);
727011e0
CM		 603 if (!reads_done)
604 goto err;
f188591e 605
5cf1ab56 606 eb->read_mirror = mirror;
ea466794
JB		 607 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
608 ret = -EIO;
609 goto err;
610 }
611
ce9adaa5 612 found_start = btrfs_header_bytenr(eb);
727011e0 613 if (found_start != eb->start) {
7a36ddec 614 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
193f284d
CM		 615 "%llu %llu\n",
616 (unsigned long long)found_start,
617 (unsigned long long)eb->start);
f188591e 618 ret = -EIO;
ce9adaa5
CM		 619 goto err;
620 }
2b82032c 621 if (check_tree_block_fsid(root, eb)) {
7a36ddec 622 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
193f284d 623 (unsigned long long)eb->start);
1259ab75
CM		 624 ret = -EIO;
625 goto err;
626 }
ce9adaa5 627 found_level = btrfs_header_level(eb);
1c24c3ce
JB		 628 if (found_level >= BTRFS_MAX_LEVEL) {
629 btrfs_info(root->fs_info, "bad tree block level %d\n",
630 (int)btrfs_header_level(eb));
631 ret = -EIO;
632 goto err;
633 }
ce9adaa5 634
85d4e461
CM		 635 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
636 eb, found_level);
4008c04a 637
ce9adaa5 638 ret = csum_tree_block(root, eb, 1);
a826d6dc 639 if (ret) {
f188591e 640 ret = -EIO;
a826d6dc
JB		 641 goto err;
642 }
643
644 /*
645 * If this is a leaf block and it is corrupt, set the corrupt bit so
646 * that we don't try and read the other copies of this block, just
647 * return -EIO.
648 */
649 if (found_level == 0 && check_leaf(root, eb)) {
650 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
651 ret = -EIO;
652 }
ce9adaa5 653
0b32f4bb
JB		 654 if (!ret)
655 set_extent_buffer_uptodate(eb);
ce9adaa5 656err:
79fb65a1
JB		 657 if (reads_done &&
658 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 659 btree_readahead_hook(root, eb, eb->start, ret);
4bb31e92 660
53b381b3
DW		 661 if (ret) {
662 /*
663 * our io error hook is going to dec the io pages
664 * again, we have to make sure it has something
665 * to decrement
666 */
667 atomic_inc(&eb->io_pages);
0b32f4bb 668 clear_extent_buffer_uptodate(eb);
53b381b3 669 }
0b32f4bb 670 free_extent_buffer(eb);
ce9adaa5 671out:
f188591e 672 return ret;
ce9adaa5
CM		 673}
674
ea466794 675static int btree_io_failed_hook(struct page *page, int failed_mirror)
4bb31e92 676{
4bb31e92
AJ		 677 struct extent_buffer *eb;
678 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
679
4f2de97a 680 eb = (struct extent_buffer *)page->private;
ea466794 681 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
5cf1ab56 682 eb->read_mirror = failed_mirror;
53b381b3 683 atomic_dec(&eb->io_pages);
ea466794 684 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 685 btree_readahead_hook(root, eb, eb->start, -EIO);
4bb31e92
AJ		 686 return -EIO; /* we fixed nothing */
687}
688
ce9adaa5 689static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 690{
691 struct end_io_wq *end_io_wq = bio->bi_private;
692 struct btrfs_fs_info *fs_info;
ce9adaa5 693
ce9adaa5 694 fs_info = end_io_wq->info;
ce9adaa5 695 end_io_wq->error = err;
8b712842
CM		 696 end_io_wq->work.func = end_workqueue_fn;
697 end_io_wq->work.flags = 0;
d20f7043 698
7b6d91da 699 if (bio->bi_rw & REQ_WRITE) {
53b381b3 700 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
cad321ad
CM		 701 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
702 &end_io_wq->work);
53b381b3 703 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
0cb59c99
JB		 704 btrfs_queue_worker(&fs_info->endio_freespace_worker,
705 &end_io_wq->work);
53b381b3
DW		 706 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
707 btrfs_queue_worker(&fs_info->endio_raid56_workers,
708 &end_io_wq->work);
cad321ad
CM		 709 else
710 btrfs_queue_worker(&fs_info->endio_write_workers,
711 &end_io_wq->work);
d20f7043 712 } else {
53b381b3
DW		 713 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
714 btrfs_queue_worker(&fs_info->endio_raid56_workers,
715 &end_io_wq->work);
716 else if (end_io_wq->metadata)
d20f7043
CM		 717 btrfs_queue_worker(&fs_info->endio_meta_workers,
718 &end_io_wq->work);
719 else
720 btrfs_queue_worker(&fs_info->endio_workers,
721 &end_io_wq->work);
722 }
ce9adaa5
CM		 723}
724
0cb59c99
JB		 725/*
726 * For the metadata arg you want
727 *
728 * 0 - if data
729 * 1 - if normal metadta
730 * 2 - if writing to the free space cache area
53b381b3 731 * 3 - raid parity work
0cb59c99 732 */
22c59948
CM		 733int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
734 int metadata)
0b86a832 735{
ce9adaa5 736 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 737 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
738 if (!end_io_wq)
739 return -ENOMEM;
740
741 end_io_wq->private = bio->bi_private;
742 end_io_wq->end_io = bio->bi_end_io;
22c59948 743 end_io_wq->info = info;
ce9adaa5
CM		 744 end_io_wq->error = 0;
745 end_io_wq->bio = bio;
22c59948 746 end_io_wq->metadata = metadata;
ce9adaa5
CM		 747
748 bio->bi_private = end_io_wq;
749 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 750 return 0;
751}
752
b64a2851 753unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 754{
4854ddd0
CM		 755 unsigned long limit = min_t(unsigned long,
756 info->workers.max_workers,
757 info->fs_devices->open_devices);
758 return 256 * limit;
759}
0986fe9e 760
4a69a410
CM		 761static void run_one_async_start(struct btrfs_work *work)
762{
4a69a410 763 struct async_submit_bio *async;
79787eaa 764 int ret;
4a69a410
CM		 765
766 async = container_of(work, struct async_submit_bio, work);
79787eaa
JM		 767 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
768 async->mirror_num, async->bio_flags,
769 async->bio_offset);
770 if (ret)
771 async->error = ret;
4a69a410
CM		 772}
773
774static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 775{
776 struct btrfs_fs_info *fs_info;
777 struct async_submit_bio *async;
4854ddd0 778 int limit;
8b712842
CM		 779
780 async = container_of(work, struct async_submit_bio, work);
781 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 782
b64a2851 783 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 784 limit = limit * 2 / 3;
785
66657b31 786 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
b64a2851 787 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 788 wake_up(&fs_info->async_submit_wait);
789
79787eaa
JM		 790 /* If an error occured we just want to clean up the bio and move on */
791 if (async->error) {
792 bio_endio(async->bio, async->error);
793 return;
794 }
795
4a69a410 796 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 797 async->mirror_num, async->bio_flags,
798 async->bio_offset);
4a69a410
CM		 799}
800
801static void run_one_async_free(struct btrfs_work *work)
802{
803 struct async_submit_bio *async;
804
805 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 806 kfree(async);
807}
808
44b8bd7e
CM		 809int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
810 int rw, struct bio *bio, int mirror_num,
c8b97818 811 unsigned long bio_flags,
eaf25d93 812 u64 bio_offset,
4a69a410
CM		 813 extent_submit_bio_hook_t *submit_bio_start,
814 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 815{
816 struct async_submit_bio *async;
817
818 async = kmalloc(sizeof(*async), GFP_NOFS);
819 if (!async)
820 return -ENOMEM;
821
822 async->inode = inode;
823 async->rw = rw;
824 async->bio = bio;
825 async->mirror_num = mirror_num;
4a69a410
CM		 826 async->submit_bio_start = submit_bio_start;
827 async->submit_bio_done = submit_bio_done;
828
829 async->work.func = run_one_async_start;
830 async->work.ordered_func = run_one_async_done;
831 async->work.ordered_free = run_one_async_free;
832
8b712842 833 async->work.flags = 0;
c8b97818 834 async->bio_flags = bio_flags;
eaf25d93 835 async->bio_offset = bio_offset;
8c8bee1d 836
79787eaa
JM		 837 async->error = 0;
838
cb03c743 839 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 840
7b6d91da 841 if (rw & REQ_SYNC)
d313d7a3
CM		 842 btrfs_set_work_high_prio(&async->work);
843
8b712842 844 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 845
d397712b 846 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 847 atomic_read(&fs_info->nr_async_submits)) {
848 wait_event(fs_info->async_submit_wait,
849 (atomic_read(&fs_info->nr_async_submits) == 0));
850 }
851
44b8bd7e
CM		 852 return 0;
853}
854
ce3ed71a
CM		 855static int btree_csum_one_bio(struct bio *bio)
856{
857 struct bio_vec *bvec = bio->bi_io_vec;
858 int bio_index = 0;
859 struct btrfs_root *root;
79787eaa 860 int ret = 0;
ce3ed71a
CM		 861
862 WARN_ON(bio->bi_vcnt <= 0);
d397712b 863 while (bio_index < bio->bi_vcnt) {
ce3ed71a 864 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 865 ret = csum_dirty_buffer(root, bvec->bv_page);
866 if (ret)
867 break;
ce3ed71a
CM		 868 bio_index++;
869 bvec++;
870 }
79787eaa 871 return ret;
ce3ed71a
CM		 872}
873
4a69a410
CM		 874static int __btree_submit_bio_start(struct inode *inode, int rw,
875 struct bio *bio, int mirror_num,
eaf25d93
CM		 876 unsigned long bio_flags,
877 u64 bio_offset)
22c59948 878{
8b712842
CM		 879 /*
880 * when we're called for a write, we're already in the async
5443be45 881 * submission context. Just jump into btrfs_map_bio
8b712842 882 */
79787eaa 883 return btree_csum_one_bio(bio);
4a69a410 884}
22c59948 885
4a69a410 886static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 887 int mirror_num, unsigned long bio_flags,
888 u64 bio_offset)
4a69a410 889{
61891923
SB		 890 int ret;
891
8b712842 892 /*
4a69a410
CM		 893 * when we're called for a write, we're already in the async
894 * submission context. Just jump into btrfs_map_bio
8b712842 895 */
61891923
SB		 896 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
897 if (ret)
898 bio_endio(bio, ret);
899 return ret;
0b86a832
CM		 900}
901
de0022b9
JB		 902static int check_async_write(struct inode *inode, unsigned long bio_flags)
903{
904 if (bio_flags & EXTENT_BIO_TREE_LOG)
905 return 0;
906#ifdef CONFIG_X86
907 if (cpu_has_xmm4_2)
908 return 0;
909#endif
910 return 1;
911}
912
44b8bd7e 913static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 914 int mirror_num, unsigned long bio_flags,
915 u64 bio_offset)
44b8bd7e 916{
de0022b9 917 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 918 int ret;
919
7b6d91da 920 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 921 /*
922 * called for a read, do the setup so that checksum validation
923 * can happen in the async kernel threads
924 */
f3f266ab
CM		 925 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
926 bio, 1);
1d4284bd 927 if (ret)
61891923
SB		 928 goto out_w_error;
929 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
930 mirror_num, 0);
de0022b9
JB		 931 } else if (!async) {
932 ret = btree_csum_one_bio(bio);
933 if (ret)
61891923
SB		 934 goto out_w_error;
935 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
936 mirror_num, 0);
937 } else {
938 /*
939 * kthread helpers are used to submit writes so that
940 * checksumming can happen in parallel across all CPUs
941 */
942 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
943 inode, rw, bio, mirror_num, 0,
944 bio_offset,
945 __btree_submit_bio_start,
946 __btree_submit_bio_done);
44b8bd7e 947 }
d313d7a3 948
61891923
SB		 949 if (ret) {
950out_w_error:
951 bio_endio(bio, ret);
952 }
953 return ret;
44b8bd7e
CM		 954}
955
3dd1462e 956#ifdef CONFIG_MIGRATION
784b4e29 957static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 958 struct page *newpage, struct page *page,
959 enum migrate_mode mode)
784b4e29
CM		 960{
961 /*
962 * we can't safely write a btree page from here,
963 * we haven't done the locking hook
964 */
965 if (PageDirty(page))
966 return -EAGAIN;
967 /*
968 * Buffers may be managed in a filesystem specific way.
969 * We must have no buffers or drop them.
970 */
971 if (page_has_private(page) &&
972 !try_to_release_page(page, GFP_KERNEL))
973 return -EAGAIN;
a6bc32b8 974 return migrate_page(mapping, newpage, page, mode);
784b4e29 975}
3dd1462e 976#endif
784b4e29 977
0da5468f
CM		 978
979static int btree_writepages(struct address_space *mapping,
980 struct writeback_control *wbc)
981{
d1310b2e 982 struct extent_io_tree *tree;
e2d84521
MX		 983 struct btrfs_fs_info *fs_info;
984 int ret;
985
d1310b2e 986 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 987 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 988
989 if (wbc->for_kupdate)
990 return 0;
991
e2d84521 992 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 993 /* this is a bit racy, but that's ok */
e2d84521
MX		 994 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
995 BTRFS_DIRTY_METADATA_THRESH);
996 if (ret < 0)
793955bc 997 return 0;
793955bc 998 }
0b32f4bb 999 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 1000}
1001
b2950863 1002static int btree_readpage(struct file *file, struct page *page)
5f39d397 1003{
d1310b2e
CM		 1004 struct extent_io_tree *tree;
1005 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 1006 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 1007}
22b0ebda 1008
70dec807 1009static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 1010{
98509cfc 1011 if (PageWriteback(page) || PageDirty(page))
d397712b 1012 return 0;
0c4e538b 1013
f7a52a40 1014 return try_release_extent_buffer(page);
d98237b3
CM		 1015}
1016
d47992f8
LC		 1017static void btree_invalidatepage(struct page *page, unsigned int offset,
1018 unsigned int length)
d98237b3 1019{
d1310b2e
CM		 1020 struct extent_io_tree *tree;
1021 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1022 extent_invalidatepage(tree, page, offset);
1023 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1024 if (PagePrivate(page)) {
d397712b
CM		 1025 printk(KERN_WARNING "btrfs warning page private not zero "
1026 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1027 ClearPagePrivate(page);
1028 set_page_private(page, 0);
1029 page_cache_release(page);
1030 }
d98237b3
CM		 1031}
1032
0b32f4bb
JB		 1033static int btree_set_page_dirty(struct page *page)
1034{
bb146eb2 1035#ifdef DEBUG
0b32f4bb
JB		 1036 struct extent_buffer *eb;
1037
1038 BUG_ON(!PagePrivate(page));
1039 eb = (struct extent_buffer *)page->private;
1040 BUG_ON(!eb);
1041 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1042 BUG_ON(!atomic_read(&eb->refs));
1043 btrfs_assert_tree_locked(eb);
bb146eb2 1044#endif
0b32f4bb
JB		 1045 return __set_page_dirty_nobuffers(page);
1046}
1047
7f09410b 1048static const struct address_space_operations btree_aops = {
d98237b3 1049 .readpage = btree_readpage,
0da5468f 1050 .writepages = btree_writepages,
5f39d397
CM		 1051 .releasepage = btree_releasepage,
1052 .invalidatepage = btree_invalidatepage,
5a92bc88 1053#ifdef CONFIG_MIGRATION
784b4e29 1054 .migratepage = btree_migratepage,
5a92bc88 1055#endif
0b32f4bb 1056 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1057};
1058
ca7a79ad
CM		 1059int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1060 u64 parent_transid)
090d1875 1061{
5f39d397
CM		 1062 struct extent_buffer *buf = NULL;
1063 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1064 int ret = 0;
090d1875 1065
db94535d 1066 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1067 if (!buf)
090d1875 1068 return 0;
d1310b2e 1069 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1070 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1071 free_extent_buffer(buf);
de428b63 1072 return ret;
090d1875
CM		 1073}
1074
ab0fff03
AJ		 1075int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1076 int mirror_num, struct extent_buffer **eb)
1077{
1078 struct extent_buffer *buf = NULL;
1079 struct inode *btree_inode = root->fs_info->btree_inode;
1080 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1081 int ret;
1082
1083 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1084 if (!buf)
1085 return 0;
1086
1087 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1088
1089 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1090 btree_get_extent, mirror_num);
1091 if (ret) {
1092 free_extent_buffer(buf);
1093 return ret;
1094 }
1095
1096 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1097 free_extent_buffer(buf);
1098 return -EIO;
0b32f4bb 1099 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1100 *eb = buf;
1101 } else {
1102 free_extent_buffer(buf);
1103 }
1104 return 0;
1105}
1106
0999df54
CM		 1107struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1108 u64 bytenr, u32 blocksize)
1109{
1110 struct inode *btree_inode = root->fs_info->btree_inode;
1111 struct extent_buffer *eb;
1112 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
f09d1f60 1113 bytenr, blocksize);
0999df54
CM		 1114 return eb;
1115}
1116
1117struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1118 u64 bytenr, u32 blocksize)
1119{
1120 struct inode *btree_inode = root->fs_info->btree_inode;
1121 struct extent_buffer *eb;
1122
1123 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
727011e0 1124 bytenr, blocksize);
0999df54
CM		 1125 return eb;
1126}
1127
1128
e02119d5
CM		 1129int btrfs_write_tree_block(struct extent_buffer *buf)
1130{
727011e0 1131 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
8aa38c31 1132 buf->start + buf->len - 1);
e02119d5
CM		 1133}
1134
1135int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1136{
727011e0 1137 return filemap_fdatawait_range(buf->pages[0]->mapping,
8aa38c31 1138 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1139}
1140
0999df54 1141struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1142 u32 blocksize, u64 parent_transid)
0999df54
CM		 1143{
1144 struct extent_buffer *buf = NULL;
0999df54
CM		 1145 int ret;
1146
0999df54
CM		 1147 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1148 if (!buf)
1149 return NULL;
0999df54 1150
ca7a79ad 1151 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
0f0fe8f7
FDBM		 1152 if (ret) {
1153 free_extent_buffer(buf);
1154 return NULL;
1155 }
5f39d397 1156 return buf;
ce9adaa5 1157
eb60ceac
CM		 1158}
1159
d5c13f92
JM		 1160void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1161 struct extent_buffer *buf)
ed2ff2cb 1162{
e2d84521
MX		 1163 struct btrfs_fs_info *fs_info = root->fs_info;
1164
55c69072 1165 if (btrfs_header_generation(buf) ==
e2d84521 1166 fs_info->running_transaction->transid) {
b9447ef8 1167 btrfs_assert_tree_locked(buf);
b4ce94de 1168
b9473439 1169 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1170 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1171 -buf->len,
1172 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1173 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1174 btrfs_set_lock_blocking(buf);
1175 clear_extent_buffer_dirty(buf);
1176 }
925baedd 1177 }
5f39d397
CM		 1178}
1179
143bede5
JM		 1180static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1181 u32 stripesize, struct btrfs_root *root,
1182 struct btrfs_fs_info *fs_info,
1183 u64 objectid)
d97e63b6 1184{
cfaa7295 1185 root->node = NULL;
a28ec197 1186 root->commit_root = NULL;
db94535d
CM		 1187 root->sectorsize = sectorsize;
1188 root->nodesize = nodesize;
1189 root->leafsize = leafsize;
87ee04eb 1190 root->stripesize = stripesize;
123abc88 1191 root->ref_cows = 0;
0b86a832 1192 root->track_dirty = 0;
c71bf099 1193 root->in_radix = 0;
d68fc57b
YZ		 1194 root->orphan_item_inserted = 0;
1195 root->orphan_cleanup_state = 0;
0b86a832 1196
0f7d52f4
CM		 1197 root->objectid = objectid;
1198 root->last_trans = 0;
13a8a7c8 1199 root->highest_objectid = 0;
eb73c1b7 1200 root->nr_delalloc_inodes = 0;
199c2a9c 1201 root->nr_ordered_extents = 0;
58176a96 1202 root->name = NULL;
6bef4d31 1203 root->inode_tree = RB_ROOT;
16cdcec7 1204 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1205 root->block_rsv = NULL;
d68fc57b 1206 root->orphan_block_rsv = NULL;
0b86a832
CM		 1207
1208 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1209 INIT_LIST_HEAD(&root->root_list);
eb73c1b7
MX		 1210 INIT_LIST_HEAD(&root->delalloc_inodes);
1211 INIT_LIST_HEAD(&root->delalloc_root);
199c2a9c
MX		 1212 INIT_LIST_HEAD(&root->ordered_extents);
1213 INIT_LIST_HEAD(&root->ordered_root);
2ab28f32
JB		 1214 INIT_LIST_HEAD(&root->logged_list[0]);
1215 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1216 spin_lock_init(&root->orphan_lock);
5d4f98a2 1217 spin_lock_init(&root->inode_lock);
eb73c1b7 1218 spin_lock_init(&root->delalloc_lock);
199c2a9c 1219 spin_lock_init(&root->ordered_extent_lock);
f0486c68 1220 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1221 spin_lock_init(&root->log_extents_lock[0]);
1222 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1223 mutex_init(&root->objectid_mutex);
e02119d5 1224 mutex_init(&root->log_mutex);
7237f183
YZ		 1225 init_waitqueue_head(&root->log_writer_wait);
1226 init_waitqueue_head(&root->log_commit_wait[0]);
1227 init_waitqueue_head(&root->log_commit_wait[1]);
1228 atomic_set(&root->log_commit[0], 0);
1229 atomic_set(&root->log_commit[1], 0);
1230 atomic_set(&root->log_writers, 0);
2ecb7923 1231 atomic_set(&root->log_batch, 0);
8a35d95f 1232 atomic_set(&root->orphan_inodes, 0);
b0feb9d9 1233 atomic_set(&root->refs, 1);
7237f183 1234 root->log_transid = 0;
257c62e1 1235 root->last_log_commit = 0;
d0c803c4 1236 extent_io_tree_init(&root->dirty_log_pages,
f993c883 1237 fs_info->btree_inode->i_mapping);
017e5369 1238
3768f368
CM		 1239 memset(&root->root_key, 0, sizeof(root->root_key));
1240 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1241 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1242 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 1243 root->defrag_trans_start = fs_info->generation;
58176a96 1244 init_completion(&root->kobj_unregister);
6702ed49 1245 root->defrag_running = 0;
4d775673 1246 root->root_key.objectid = objectid;
0ee5dc67 1247 root->anon_dev = 0;
8ea05e3a 1248
5f3ab90a 1249 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1250}
1251
f84a8bd6 1252static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1253{
1254 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1255 if (root)
1256 root->fs_info = fs_info;
1257 return root;
1258}
1259
20897f5c
AJ		 1260struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1261 struct btrfs_fs_info *fs_info,
1262 u64 objectid)
1263{
1264 struct extent_buffer *leaf;
1265 struct btrfs_root *tree_root = fs_info->tree_root;
1266 struct btrfs_root *root;
1267 struct btrfs_key key;
1268 int ret = 0;
1269 u64 bytenr;
6463fe58 1270 uuid_le uuid;
20897f5c
AJ		 1271
1272 root = btrfs_alloc_root(fs_info);
1273 if (!root)
1274 return ERR_PTR(-ENOMEM);
1275
1276 __setup_root(tree_root->nodesize, tree_root->leafsize,
1277 tree_root->sectorsize, tree_root->stripesize,
1278 root, fs_info, objectid);
1279 root->root_key.objectid = objectid;
1280 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1281 root->root_key.offset = 0;
1282
1283 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1284 0, objectid, NULL, 0, 0, 0);
1285 if (IS_ERR(leaf)) {
1286 ret = PTR_ERR(leaf);
1dd05682 1287 leaf = NULL;
20897f5c
AJ		 1288 goto fail;
1289 }
1290
1291 bytenr = leaf->start;
1292 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1293 btrfs_set_header_bytenr(leaf, leaf->start);
1294 btrfs_set_header_generation(leaf, trans->transid);
1295 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1296 btrfs_set_header_owner(leaf, objectid);
1297 root->node = leaf;
1298
1299 write_extent_buffer(leaf, fs_info->fsid,
1300 (unsigned long)btrfs_header_fsid(leaf),
1301 BTRFS_FSID_SIZE);
1302 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1303 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1304 BTRFS_UUID_SIZE);
1305 btrfs_mark_buffer_dirty(leaf);
1306
1307 root->commit_root = btrfs_root_node(root);
1308 root->track_dirty = 1;
1309
1310
1311 root->root_item.flags = 0;
1312 root->root_item.byte_limit = 0;
1313 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1314 btrfs_set_root_generation(&root->root_item, trans->transid);
1315 btrfs_set_root_level(&root->root_item, 0);
1316 btrfs_set_root_refs(&root->root_item, 1);
1317 btrfs_set_root_used(&root->root_item, leaf->len);
1318 btrfs_set_root_last_snapshot(&root->root_item, 0);
1319 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1320 uuid_le_gen(&uuid);
1321 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1322 root->root_item.drop_level = 0;
1323
1324 key.objectid = objectid;
1325 key.type = BTRFS_ROOT_ITEM_KEY;
1326 key.offset = 0;
1327 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1328 if (ret)
1329 goto fail;
1330
1331 btrfs_tree_unlock(leaf);
1332
1dd05682
TI		 1333 return root;
1334
20897f5c 1335fail:
1dd05682
TI		 1336 if (leaf) {
1337 btrfs_tree_unlock(leaf);
1338 free_extent_buffer(leaf);
1339 }
1340 kfree(root);
20897f5c 1341
1dd05682 1342 return ERR_PTR(ret);
20897f5c
AJ		 1343}
1344
7237f183
YZ		 1345static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1346 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1347{
1348 struct btrfs_root *root;
1349 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1350 struct extent_buffer *leaf;
e02119d5 1351
6f07e42e 1352 root = btrfs_alloc_root(fs_info);
e02119d5 1353 if (!root)
7237f183 1354 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1355
1356 __setup_root(tree_root->nodesize, tree_root->leafsize,
1357 tree_root->sectorsize, tree_root->stripesize,
1358 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1359
1360 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1361 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1362 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1363 /*
1364 * log trees do not get reference counted because they go away
1365 * before a real commit is actually done. They do store pointers
1366 * to file data extents, and those reference counts still get
1367 * updated (along with back refs to the log tree).
1368 */
e02119d5
CM		 1369 root->ref_cows = 0;
1370
5d4f98a2 1371 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1372 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1373 0, 0, 0);
7237f183
YZ		 1374 if (IS_ERR(leaf)) {
1375 kfree(root);
1376 return ERR_CAST(leaf);
1377 }
e02119d5 1378
5d4f98a2
YZ		 1379 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1380 btrfs_set_header_bytenr(leaf, leaf->start);
1381 btrfs_set_header_generation(leaf, trans->transid);
1382 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1383 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1384 root->node = leaf;
e02119d5
CM		 1385
1386 write_extent_buffer(root->node, root->fs_info->fsid,
1387 (unsigned long)btrfs_header_fsid(root->node),
1388 BTRFS_FSID_SIZE);
1389 btrfs_mark_buffer_dirty(root->node);
1390 btrfs_tree_unlock(root->node);
7237f183
YZ		 1391 return root;
1392}
1393
1394int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1395 struct btrfs_fs_info *fs_info)
1396{
1397 struct btrfs_root *log_root;
1398
1399 log_root = alloc_log_tree(trans, fs_info);
1400 if (IS_ERR(log_root))
1401 return PTR_ERR(log_root);
1402 WARN_ON(fs_info->log_root_tree);
1403 fs_info->log_root_tree = log_root;
1404 return 0;
1405}
1406
1407int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1408 struct btrfs_root *root)
1409{
1410 struct btrfs_root *log_root;
1411 struct btrfs_inode_item *inode_item;
1412
1413 log_root = alloc_log_tree(trans, root->fs_info);
1414 if (IS_ERR(log_root))
1415 return PTR_ERR(log_root);
1416
1417 log_root->last_trans = trans->transid;
1418 log_root->root_key.offset = root->root_key.objectid;
1419
1420 inode_item = &log_root->root_item.inode;
3cae210f
QW		 1421 btrfs_set_stack_inode_generation(inode_item, 1);
1422 btrfs_set_stack_inode_size(inode_item, 3);
1423 btrfs_set_stack_inode_nlink(inode_item, 1);
1424 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1425 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
7237f183 1426
5d4f98a2 1427 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1428
1429 WARN_ON(root->log_root);
1430 root->log_root = log_root;
1431 root->log_transid = 0;
257c62e1 1432 root->last_log_commit = 0;
e02119d5
CM		 1433 return 0;
1434}
1435
35a3621b
SB		 1436static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1437 struct btrfs_key *key)
e02119d5
CM		 1438{
1439 struct btrfs_root *root;
1440 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1441 struct btrfs_path *path;
84234f3a 1442 u64 generation;
db94535d 1443 u32 blocksize;
cb517eab 1444 int ret;
0f7d52f4 1445
cb517eab
MX		 1446 path = btrfs_alloc_path();
1447 if (!path)
0f7d52f4 1448 return ERR_PTR(-ENOMEM);
cb517eab
MX		 1449
1450 root = btrfs_alloc_root(fs_info);
1451 if (!root) {
1452 ret = -ENOMEM;
1453 goto alloc_fail;
0f7d52f4
CM		 1454 }
1455
db94535d 1456 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb 1457 tree_root->sectorsize, tree_root->stripesize,
cb517eab 1458 root, fs_info, key->objectid);
0f7d52f4 1459
cb517eab
MX		 1460 ret = btrfs_find_root(tree_root, key, path,
1461 &root->root_item, &root->root_key);
0f7d52f4 1462 if (ret) {
13a8a7c8
YZ		 1463 if (ret > 0)
1464 ret = -ENOENT;
cb517eab 1465 goto find_fail;
0f7d52f4 1466 }
13a8a7c8 1467
84234f3a 1468 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1469 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1470 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1471 blocksize, generation);
cb517eab
MX		 1472 if (!root->node) {
1473 ret = -ENOMEM;
1474 goto find_fail;
1475 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1476 ret = -EIO;
1477 goto read_fail;
416bc658 1478 }
5d4f98a2 1479 root->commit_root = btrfs_root_node(root);
13a8a7c8 1480out:
cb517eab
MX		 1481 btrfs_free_path(path);
1482 return root;
1483
1484read_fail:
1485 free_extent_buffer(root->node);
1486find_fail:
1487 kfree(root);
1488alloc_fail:
1489 root = ERR_PTR(ret);
1490 goto out;
1491}
1492
1493struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1494 struct btrfs_key *location)
1495{
1496 struct btrfs_root *root;
1497
1498 root = btrfs_read_tree_root(tree_root, location);
1499 if (IS_ERR(root))
1500 return root;
1501
1502 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1503 root->ref_cows = 1;
08fe4db1
LZ		 1504 btrfs_check_and_init_root_item(&root->root_item);
1505 }
13a8a7c8 1506
5eda7b5e
CM		 1507 return root;
1508}
1509
cb517eab
MX		 1510int btrfs_init_fs_root(struct btrfs_root *root)
1511{
1512 int ret;
1513
1514 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1515 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1516 GFP_NOFS);
1517 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1518 ret = -ENOMEM;
1519 goto fail;
1520 }
1521
1522 btrfs_init_free_ino_ctl(root);
1523 mutex_init(&root->fs_commit_mutex);
1524 spin_lock_init(&root->cache_lock);
1525 init_waitqueue_head(&root->cache_wait);
1526
1527 ret = get_anon_bdev(&root->anon_dev);
1528 if (ret)
1529 goto fail;
1530 return 0;
1531fail:
1532 kfree(root->free_ino_ctl);
1533 kfree(root->free_ino_pinned);
1534 return ret;
1535}
1536
171170c1
ST		 1537static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1538 u64 root_id)
cb517eab
MX		 1539{
1540 struct btrfs_root *root;
1541
1542 spin_lock(&fs_info->fs_roots_radix_lock);
1543 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1544 (unsigned long)root_id);
1545 spin_unlock(&fs_info->fs_roots_radix_lock);
1546 return root;
1547}
1548
1549int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1550 struct btrfs_root *root)
1551{
1552 int ret;
1553
1554 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1555 if (ret)
1556 return ret;
1557
1558 spin_lock(&fs_info->fs_roots_radix_lock);
1559 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1560 (unsigned long)root->root_key.objectid,
1561 root);
1562 if (ret == 0)
1563 root->in_radix = 1;
1564 spin_unlock(&fs_info->fs_roots_radix_lock);
1565 radix_tree_preload_end();
1566
1567 return ret;
1568}
1569
edbd8d4e
CM		 1570struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1571 struct btrfs_key *location)
5eda7b5e
CM		 1572{
1573 struct btrfs_root *root;
1574 int ret;
1575
edbd8d4e
CM		 1576 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1577 return fs_info->tree_root;
1578 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1579 return fs_info->extent_root;
8f18cf13
CM		 1580 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1581 return fs_info->chunk_root;
1582 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1583 return fs_info->dev_root;
0403e47e
YZ		 1584 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1585 return fs_info->csum_root;
bcef60f2
AJ		 1586 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1587 return fs_info->quota_root ? fs_info->quota_root :
1588 ERR_PTR(-ENOENT);
f7a81ea4
SB		 1589 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1590 return fs_info->uuid_root ? fs_info->uuid_root :
1591 ERR_PTR(-ENOENT);
4df27c4d 1592again:
cb517eab 1593 root = btrfs_lookup_fs_root(fs_info, location->objectid);
5eda7b5e
CM		 1594 if (root)
1595 return root;
1596
cb517eab 1597 root = btrfs_read_fs_root(fs_info->tree_root, location);
5eda7b5e
CM		 1598 if (IS_ERR(root))
1599 return root;
3394e160 1600
cb517eab
MX		 1601 if (btrfs_root_refs(&root->root_item) == 0) {
1602 ret = -ENOENT;
581bb050 1603 goto fail;
35a30d7c 1604 }
581bb050 1605
cb517eab 1606 ret = btrfs_init_fs_root(root);
ac08aedf
CM		 1607 if (ret)
1608 goto fail;
3394e160 1609
d68fc57b
YZ		 1610 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1611 if (ret < 0)
1612 goto fail;
1613 if (ret == 0)
1614 root->orphan_item_inserted = 1;
1615
cb517eab 1616 ret = btrfs_insert_fs_root(fs_info, root);
0f7d52f4 1617 if (ret) {
4df27c4d
YZ		 1618 if (ret == -EEXIST) {
1619 free_fs_root(root);
1620 goto again;
1621 }
1622 goto fail;
0f7d52f4 1623 }
edbd8d4e 1624 return root;
4df27c4d
YZ		 1625fail:
1626 free_fs_root(root);
1627 return ERR_PTR(ret);
edbd8d4e
CM		 1628}
1629
04160088
CM		 1630static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1631{
1632 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1633 int ret = 0;
04160088
CM		 1634 struct btrfs_device *device;
1635 struct backing_dev_info *bdi;
b7967db7 1636
1f78160c
XG		 1637 rcu_read_lock();
1638 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1639 if (!device->bdev)
1640 continue;
04160088
CM		 1641 bdi = blk_get_backing_dev_info(device->bdev);
1642 if (bdi && bdi_congested(bdi, bdi_bits)) {
1643 ret = 1;
1644 break;
1645 }
1646 }
1f78160c 1647 rcu_read_unlock();
04160088
CM		 1648 return ret;
1649}
1650
ad081f14
JA		 1651/*
1652 * If this fails, caller must call bdi_destroy() to get rid of the
1653 * bdi again.
1654 */
04160088
CM		 1655static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1656{
ad081f14
JA		 1657 int err;
1658
1659 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1660 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1661 if (err)
1662 return err;
1663
4575c9cc 1664 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1665 bdi->congested_fn = btrfs_congested_fn;
1666 bdi->congested_data = info;
1667 return 0;
1668}
1669
8b712842
CM		 1670/*
1671 * called by the kthread helper functions to finally call the bio end_io
1672 * functions. This is where read checksum verification actually happens
1673 */
1674static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1675{
ce9adaa5 1676 struct bio *bio;
8b712842
CM		 1677 struct end_io_wq *end_io_wq;
1678 struct btrfs_fs_info *fs_info;
ce9adaa5 1679 int error;
ce9adaa5 1680
8b712842
CM		 1681 end_io_wq = container_of(work, struct end_io_wq, work);
1682 bio = end_io_wq->bio;
1683 fs_info = end_io_wq->info;
ce9adaa5 1684
8b712842
CM		 1685 error = end_io_wq->error;
1686 bio->bi_private = end_io_wq->private;
1687 bio->bi_end_io = end_io_wq->end_io;
1688 kfree(end_io_wq);
8b712842 1689 bio_endio(bio, error);
44b8bd7e
CM		 1690}
1691
a74a4b97
CM		 1692static int cleaner_kthread(void *arg)
1693{
1694 struct btrfs_root *root = arg;
d0278245 1695 int again;
a74a4b97
CM		 1696
1697 do {
d0278245 1698 again = 0;
a74a4b97 1699
d0278245 1700 /* Make the cleaner go to sleep early. */
babbf170 1701 if (btrfs_need_cleaner_sleep(root))
d0278245
MX		 1702 goto sleep;
1703
1704 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1705 goto sleep;
1706
dc7f370c
MX		 1707 /*
1708 * Avoid the problem that we change the status of the fs
1709 * during the above check and trylock.
1710 */
babbf170 1711 if (btrfs_need_cleaner_sleep(root)) {
dc7f370c
MX		 1712 mutex_unlock(&root->fs_info->cleaner_mutex);
1713 goto sleep;
76dda93c 1714 }
a74a4b97 1715
d0278245
MX		 1716 btrfs_run_delayed_iputs(root);
1717 again = btrfs_clean_one_deleted_snapshot(root);
1718 mutex_unlock(&root->fs_info->cleaner_mutex);
1719
1720 /*
05323cd1
MX		 1721 * The defragger has dealt with the R/O remount and umount,
1722 * needn't do anything special here.
d0278245
MX		 1723 */
1724 btrfs_run_defrag_inodes(root->fs_info);
1725sleep:
9d1a2a3a 1726 if (!try_to_freeze() && !again) {
a74a4b97 1727 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1728 if (!kthread_should_stop())
1729 schedule();
a74a4b97
CM		 1730 __set_current_state(TASK_RUNNING);
1731 }
1732 } while (!kthread_should_stop());
1733 return 0;
1734}
1735
1736static int transaction_kthread(void *arg)
1737{
1738 struct btrfs_root *root = arg;
1739 struct btrfs_trans_handle *trans;
1740 struct btrfs_transaction *cur;
8929ecfa 1741 u64 transid;
a74a4b97
CM		 1742 unsigned long now;
1743 unsigned long delay;
914b2007 1744 bool cannot_commit;
a74a4b97
CM		 1745
1746 do {
914b2007 1747 cannot_commit = false;
8b87dc17 1748 delay = HZ * root->fs_info->commit_interval;
a74a4b97
CM		 1749 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1750
a4abeea4 1751 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1752 cur = root->fs_info->running_transaction;
1753 if (!cur) {
a4abeea4 1754 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1755 goto sleep;
1756 }
31153d81 1757
a74a4b97 1758 now = get_seconds();
4a9d8bde 1759 if (cur->state < TRANS_STATE_BLOCKED &&
8b87dc17
DS		 1760 (now < cur->start_time ||
1761 now - cur->start_time < root->fs_info->commit_interval)) {
a4abeea4 1762 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1763 delay = HZ * 5;
1764 goto sleep;
1765 }
8929ecfa 1766 transid = cur->transid;
a4abeea4 1767 spin_unlock(&root->fs_info->trans_lock);
56bec294 1768
79787eaa 1769 /* If the file system is aborted, this will always fail. */
354aa0fb 1770 trans = btrfs_attach_transaction(root);
914b2007 1771 if (IS_ERR(trans)) {
354aa0fb
MX		 1772 if (PTR_ERR(trans) != -ENOENT)
1773 cannot_commit = true;
79787eaa 1774 goto sleep;
914b2007 1775 }
8929ecfa 1776 if (transid == trans->transid) {
79787eaa 1777 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1778 } else {
1779 btrfs_end_transaction(trans, root);
1780 }
a74a4b97
CM		 1781sleep:
1782 wake_up_process(root->fs_info->cleaner_kthread);
1783 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1784
a0acae0e 1785 if (!try_to_freeze()) {
a74a4b97 1786 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1787 if (!kthread_should_stop() &&
914b2007
JK		 1788 (!btrfs_transaction_blocked(root->fs_info) ||
1789 cannot_commit))
8929ecfa 1790 schedule_timeout(delay);
a74a4b97
CM		 1791 __set_current_state(TASK_RUNNING);
1792 }
1793 } while (!kthread_should_stop());
1794 return 0;
1795}
1796
af31f5e5
CM		 1797/*
1798 * this will find the highest generation in the array of
1799 * root backups. The index of the highest array is returned,
1800 * or -1 if we can't find anything.
1801 *
1802 * We check to make sure the array is valid by comparing the
1803 * generation of the latest root in the array with the generation
1804 * in the super block. If they don't match we pitch it.
1805 */
1806static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1807{
1808 u64 cur;
1809 int newest_index = -1;
1810 struct btrfs_root_backup *root_backup;
1811 int i;
1812
1813 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1814 root_backup = info->super_copy->super_roots + i;
1815 cur = btrfs_backup_tree_root_gen(root_backup);
1816 if (cur == newest_gen)
1817 newest_index = i;
1818 }
1819
1820 /* check to see if we actually wrapped around */
1821 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1822 root_backup = info->super_copy->super_roots;
1823 cur = btrfs_backup_tree_root_gen(root_backup);
1824 if (cur == newest_gen)
1825 newest_index = 0;
1826 }
1827 return newest_index;
1828}
1829
1830
1831/*
1832 * find the oldest backup so we know where to store new entries
1833 * in the backup array. This will set the backup_root_index
1834 * field in the fs_info struct
1835 */
1836static void find_oldest_super_backup(struct btrfs_fs_info *info,
1837 u64 newest_gen)
1838{
1839 int newest_index = -1;
1840
1841 newest_index = find_newest_super_backup(info, newest_gen);
1842 /* if there was garbage in there, just move along */
1843 if (newest_index == -1) {
1844 info->backup_root_index = 0;
1845 } else {
1846 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1847 }
1848}
1849
1850/*
1851 * copy all the root pointers into the super backup array.
1852 * this will bump the backup pointer by one when it is
1853 * done
1854 */
1855static void backup_super_roots(struct btrfs_fs_info *info)
1856{
1857 int next_backup;
1858 struct btrfs_root_backup *root_backup;
1859 int last_backup;
1860
1861 next_backup = info->backup_root_index;
1862 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1863 BTRFS_NUM_BACKUP_ROOTS;
1864
1865 /*
1866 * just overwrite the last backup if we're at the same generation
1867 * this happens only at umount
1868 */
1869 root_backup = info->super_for_commit->super_roots + last_backup;
1870 if (btrfs_backup_tree_root_gen(root_backup) ==
1871 btrfs_header_generation(info->tree_root->node))
1872 next_backup = last_backup;
1873
1874 root_backup = info->super_for_commit->super_roots + next_backup;
1875
1876 /*
1877 * make sure all of our padding and empty slots get zero filled
1878 * regardless of which ones we use today
1879 */
1880 memset(root_backup, 0, sizeof(*root_backup));
1881
1882 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1883
1884 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1885 btrfs_set_backup_tree_root_gen(root_backup,
1886 btrfs_header_generation(info->tree_root->node));
1887
1888 btrfs_set_backup_tree_root_level(root_backup,
1889 btrfs_header_level(info->tree_root->node));
1890
1891 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1892 btrfs_set_backup_chunk_root_gen(root_backup,
1893 btrfs_header_generation(info->chunk_root->node));
1894 btrfs_set_backup_chunk_root_level(root_backup,
1895 btrfs_header_level(info->chunk_root->node));
1896
1897 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1898 btrfs_set_backup_extent_root_gen(root_backup,
1899 btrfs_header_generation(info->extent_root->node));
1900 btrfs_set_backup_extent_root_level(root_backup,
1901 btrfs_header_level(info->extent_root->node));
1902
7c7e82a7
CM		 1903 /*
1904 * we might commit during log recovery, which happens before we set
1905 * the fs_root. Make sure it is valid before we fill it in.
1906 */
1907 if (info->fs_root && info->fs_root->node) {
1908 btrfs_set_backup_fs_root(root_backup,
1909 info->fs_root->node->start);
1910 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1911 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1912 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1913 btrfs_header_level(info->fs_root->node));
7c7e82a7 1914 }
af31f5e5
CM		 1915
1916 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1917 btrfs_set_backup_dev_root_gen(root_backup,
1918 btrfs_header_generation(info->dev_root->node));
1919 btrfs_set_backup_dev_root_level(root_backup,
1920 btrfs_header_level(info->dev_root->node));
1921
1922 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1923 btrfs_set_backup_csum_root_gen(root_backup,
1924 btrfs_header_generation(info->csum_root->node));
1925 btrfs_set_backup_csum_root_level(root_backup,
1926 btrfs_header_level(info->csum_root->node));
1927
1928 btrfs_set_backup_total_bytes(root_backup,
1929 btrfs_super_total_bytes(info->super_copy));
1930 btrfs_set_backup_bytes_used(root_backup,
1931 btrfs_super_bytes_used(info->super_copy));
1932 btrfs_set_backup_num_devices(root_backup,
1933 btrfs_super_num_devices(info->super_copy));
1934
1935 /*
1936 * if we don't copy this out to the super_copy, it won't get remembered
1937 * for the next commit
1938 */
1939 memcpy(&info->super_copy->super_roots,
1940 &info->super_for_commit->super_roots,
1941 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1942}
1943
1944/*
1945 * this copies info out of the root backup array and back into
1946 * the in-memory super block. It is meant to help iterate through
1947 * the array, so you send it the number of backups you've already
1948 * tried and the last backup index you used.
1949 *
1950 * this returns -1 when it has tried all the backups
1951 */
1952static noinline int next_root_backup(struct btrfs_fs_info *info,
1953 struct btrfs_super_block *super,
1954 int *num_backups_tried, int *backup_index)
1955{
1956 struct btrfs_root_backup *root_backup;
1957 int newest = *backup_index;
1958
1959 if (*num_backups_tried == 0) {
1960 u64 gen = btrfs_super_generation(super);
1961
1962 newest = find_newest_super_backup(info, gen);
1963 if (newest == -1)
1964 return -1;
1965
1966 *backup_index = newest;
1967 *num_backups_tried = 1;
1968 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1969 /* we've tried all the backups, all done */
1970 return -1;
1971 } else {
1972 /* jump to the next oldest backup */
1973 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1974 BTRFS_NUM_BACKUP_ROOTS;
1975 *backup_index = newest;
1976 *num_backups_tried += 1;
1977 }
1978 root_backup = super->super_roots + newest;
1979
1980 btrfs_set_super_generation(super,
1981 btrfs_backup_tree_root_gen(root_backup));
1982 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1983 btrfs_set_super_root_level(super,
1984 btrfs_backup_tree_root_level(root_backup));
1985 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1986
1987 /*
1988 * fixme: the total bytes and num_devices need to match or we should
1989 * need a fsck
1990 */
1991 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1992 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1993 return 0;
1994}
1995
7abadb64
LB		 1996/* helper to cleanup workers */
1997static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1998{
1999 btrfs_stop_workers(&fs_info->generic_worker);
2000 btrfs_stop_workers(&fs_info->fixup_workers);
2001 btrfs_stop_workers(&fs_info->delalloc_workers);
2002 btrfs_stop_workers(&fs_info->workers);
2003 btrfs_stop_workers(&fs_info->endio_workers);
2004 btrfs_stop_workers(&fs_info->endio_meta_workers);
2005 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2006 btrfs_stop_workers(&fs_info->rmw_workers);
2007 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2008 btrfs_stop_workers(&fs_info->endio_write_workers);
2009 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2010 btrfs_stop_workers(&fs_info->submit_workers);
2011 btrfs_stop_workers(&fs_info->delayed_workers);
2012 btrfs_stop_workers(&fs_info->caching_workers);
2013 btrfs_stop_workers(&fs_info->readahead_workers);
2014 btrfs_stop_workers(&fs_info->flush_workers);
2f232036 2015 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
7abadb64
LB		 2016}
2017
af31f5e5
CM		 2018/* helper to cleanup tree roots */
2019static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2020{
2021 free_extent_buffer(info->tree_root->node);
2022 free_extent_buffer(info->tree_root->commit_root);
af31f5e5
CM		 2023 info->tree_root->node = NULL;
2024 info->tree_root->commit_root = NULL;
655b09fe
JB		 2025
2026 if (info->dev_root) {
2027 free_extent_buffer(info->dev_root->node);
2028 free_extent_buffer(info->dev_root->commit_root);
2029 info->dev_root->node = NULL;
2030 info->dev_root->commit_root = NULL;
2031 }
2032 if (info->extent_root) {
2033 free_extent_buffer(info->extent_root->node);
2034 free_extent_buffer(info->extent_root->commit_root);
2035 info->extent_root->node = NULL;
2036 info->extent_root->commit_root = NULL;
2037 }
2038 if (info->csum_root) {
2039 free_extent_buffer(info->csum_root->node);
2040 free_extent_buffer(info->csum_root->commit_root);
2041 info->csum_root->node = NULL;
2042 info->csum_root->commit_root = NULL;
2043 }
bcef60f2 2044 if (info->quota_root) {
655b09fe
JB		 2045 free_extent_buffer(info->quota_root->node);
2046 free_extent_buffer(info->quota_root->commit_root);
bcef60f2
AJ		 2047 info->quota_root->node = NULL;
2048 info->quota_root->commit_root = NULL;
2049 }
f7a81ea4
SB		 2050 if (info->uuid_root) {
2051 free_extent_buffer(info->uuid_root->node);
2052 free_extent_buffer(info->uuid_root->commit_root);
2053 info->uuid_root->node = NULL;
2054 info->uuid_root->commit_root = NULL;
2055 }
af31f5e5
CM		 2056 if (chunk_root) {
2057 free_extent_buffer(info->chunk_root->node);
2058 free_extent_buffer(info->chunk_root->commit_root);
2059 info->chunk_root->node = NULL;
2060 info->chunk_root->commit_root = NULL;
2061 }
2062}
2063
171f6537
JB		 2064static void del_fs_roots(struct btrfs_fs_info *fs_info)
2065{
2066 int ret;
2067 struct btrfs_root *gang[8];
2068 int i;
2069
2070 while (!list_empty(&fs_info->dead_roots)) {
2071 gang[0] = list_entry(fs_info->dead_roots.next,
2072 struct btrfs_root, root_list);
2073 list_del(&gang[0]->root_list);
2074
2075 if (gang[0]->in_radix) {
cb517eab 2076 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
171f6537
JB		 2077 } else {
2078 free_extent_buffer(gang[0]->node);
2079 free_extent_buffer(gang[0]->commit_root);
b0feb9d9 2080 btrfs_put_fs_root(gang[0]);
171f6537
JB		 2081 }
2082 }
2083
2084 while (1) {
2085 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2086 (void **)gang, 0,
2087 ARRAY_SIZE(gang));
2088 if (!ret)
2089 break;
2090 for (i = 0; i < ret; i++)
cb517eab 2091 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
171f6537
JB		 2092 }
2093}
af31f5e5 2094
ad2b2c80
AV		 2095int open_ctree(struct super_block *sb,
2096 struct btrfs_fs_devices *fs_devices,
2097 char *options)
2e635a27 2098{
db94535d
CM		 2099 u32 sectorsize;
2100 u32 nodesize;
2101 u32 leafsize;
2102 u32 blocksize;
87ee04eb 2103 u32 stripesize;
84234f3a 2104 u64 generation;
f2b636e8 2105 u64 features;
3de4586c 2106 struct btrfs_key location;
a061fc8d 2107 struct buffer_head *bh;
4d34b278 2108 struct btrfs_super_block *disk_super;
815745cf 2109 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2110 struct btrfs_root *tree_root;
4d34b278
ID		 2111 struct btrfs_root *extent_root;
2112 struct btrfs_root *csum_root;
2113 struct btrfs_root *chunk_root;
2114 struct btrfs_root *dev_root;
bcef60f2 2115 struct btrfs_root *quota_root;
f7a81ea4 2116 struct btrfs_root *uuid_root;
e02119d5 2117 struct btrfs_root *log_tree_root;
eb60ceac 2118 int ret;
e58ca020 2119 int err = -EINVAL;
af31f5e5
CM		 2120 int num_backups_tried = 0;
2121 int backup_index = 0;
f7a81ea4 2122 bool create_uuid_tree = false;
4543df7e 2123
f84a8bd6 2124 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e 2125 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
cb517eab 2126 if (!tree_root || !chunk_root) {
39279cc3
CM		 2127 err = -ENOMEM;
2128 goto fail;
2129 }
76dda93c
YZ		 2130
2131 ret = init_srcu_struct(&fs_info->subvol_srcu);
2132 if (ret) {
2133 err = ret;
2134 goto fail;
2135 }
2136
2137 ret = setup_bdi(fs_info, &fs_info->bdi);
2138 if (ret) {
2139 err = ret;
2140 goto fail_srcu;
2141 }
2142
e2d84521
MX		 2143 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2144 if (ret) {
2145 err = ret;
2146 goto fail_bdi;
2147 }
2148 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2149 (1 + ilog2(nr_cpu_ids));
2150
963d678b
MX		 2151 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2152 if (ret) {
2153 err = ret;
2154 goto fail_dirty_metadata_bytes;
2155 }
2156
76dda93c
YZ		 2157 fs_info->btree_inode = new_inode(sb);
2158 if (!fs_info->btree_inode) {
2159 err = -ENOMEM;
963d678b 2160 goto fail_delalloc_bytes;
76dda93c
YZ		 2161 }
2162
a6591715 2163 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2164
76dda93c 2165 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 2166 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2167 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2168 INIT_LIST_HEAD(&fs_info->delayed_iputs);
eb73c1b7 2169 INIT_LIST_HEAD(&fs_info->delalloc_roots);
11833d66 2170 INIT_LIST_HEAD(&fs_info->caching_block_groups);
eb73c1b7 2171 spin_lock_init(&fs_info->delalloc_root_lock);
a4abeea4 2172 spin_lock_init(&fs_info->trans_lock);
76dda93c 2173 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2174 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2175 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2176 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2177 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2178 spin_lock_init(&fs_info->super_lock);
f29021b2 2179 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2180 mutex_init(&fs_info->reloc_mutex);
de98ced9 2181 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2182
58176a96 2183 init_completion(&fs_info->kobj_unregister);
0b86a832 2184 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2185 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2186 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2187 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2188 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2189 BTRFS_BLOCK_RSV_GLOBAL);
2190 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2191 BTRFS_BLOCK_RSV_DELALLOC);
2192 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2193 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2194 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2195 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2196 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2197 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2198 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2199 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2200 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2201 atomic_set(&fs_info->defrag_running, 0);
fc36ed7e 2202 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2203 fs_info->sb = sb;
6f568d35 2204 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2205 fs_info->metadata_ratio = 0;
4cb5300b 2206 fs_info->defrag_inodes = RB_ROOT;
2bf64758 2207 fs_info->free_chunk_space = 0;
f29021b2 2208 fs_info->tree_mod_log = RB_ROOT;
8b87dc17 2209 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
c8b97818 2210
90519d66
AJ		 2211 /* readahead state */
2212 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2213 spin_lock_init(&fs_info->reada_lock);
c8b97818 2214
b34b086c
CM		 2215 fs_info->thread_pool_size = min_t(unsigned long,
2216 num_online_cpus() + 2, 8);
0afbaf8c 2217
199c2a9c
MX		 2218 INIT_LIST_HEAD(&fs_info->ordered_roots);
2219 spin_lock_init(&fs_info->ordered_root_lock);
16cdcec7
MX		 2220 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2221 GFP_NOFS);
2222 if (!fs_info->delayed_root) {
2223 err = -ENOMEM;
2224 goto fail_iput;
2225 }
2226 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2227
a2de733c
AJ		 2228 mutex_init(&fs_info->scrub_lock);
2229 atomic_set(&fs_info->scrubs_running, 0);
2230 atomic_set(&fs_info->scrub_pause_req, 0);
2231 atomic_set(&fs_info->scrubs_paused, 0);
2232 atomic_set(&fs_info->scrub_cancel_req, 0);
2233 init_waitqueue_head(&fs_info->scrub_pause_wait);
2234 init_rwsem(&fs_info->scrub_super_lock);
2235 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2236#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2237 fs_info->check_integrity_print_mask = 0;
2238#endif
a2de733c 2239
c9e9f97b
ID		 2240 spin_lock_init(&fs_info->balance_lock);
2241 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2242 atomic_set(&fs_info->balance_running, 0);
2243 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2244 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2245 fs_info->balance_ctl = NULL;
837d5b6e 2246 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2247
a061fc8d
CM		 2248 sb->s_blocksize = 4096;
2249 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2250 sb->s_bdi = &fs_info->bdi;
a061fc8d 2251
76dda93c 2252 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2253 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2254 /*
2255 * we set the i_size on the btree inode to the max possible int.
2256 * the real end of the address space is determined by all of
2257 * the devices in the system
2258 */
2259 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2260 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2261 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2262
5d4f98a2 2263 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2264 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2265 fs_info->btree_inode->i_mapping);
0b32f4bb 2266 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2267 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2268
2269 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2270
76dda93c
YZ		 2271 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2272 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2273 sizeof(struct btrfs_key));
72ac3c0d
JB		 2274 set_bit(BTRFS_INODE_DUMMY,
2275 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
c65ddb52 2276 insert_inode_hash(fs_info->btree_inode);
76dda93c 2277
0f9dd46c 2278 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2279 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2280 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2281
11833d66 2282 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2283 fs_info->btree_inode->i_mapping);
11833d66 2284 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2285 fs_info->btree_inode->i_mapping);
11833d66 2286 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2287 fs_info->do_barriers = 1;
e18e4809 2288
39279cc3 2289
5a3f23d5 2290 mutex_init(&fs_info->ordered_operations_mutex);
9ffba8cd 2291 mutex_init(&fs_info->ordered_extent_flush_mutex);
e02119d5 2292 mutex_init(&fs_info->tree_log_mutex);
925baedd 2293 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2294 mutex_init(&fs_info->transaction_kthread_mutex);
2295 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2296 mutex_init(&fs_info->volume_mutex);
276e680d 2297 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2298 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2299 init_rwsem(&fs_info->subvol_sem);
e922e087
SB		 2300 fs_info->dev_replace.lock_owner = 0;
2301 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2302 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2303 mutex_init(&fs_info->dev_replace.lock_management_lock);
2304 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2305
416ac51d 2306 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2307 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2308 fs_info->qgroup_tree = RB_ROOT;
2309 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2310 fs_info->qgroup_seq = 1;
2311 fs_info->quota_enabled = 0;
2312 fs_info->pending_quota_state = 0;
1e8f9158 2313 fs_info->qgroup_ulist = NULL;
2f232036 2314 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2315
fa9c0d79
CM		 2316 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2317 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2318
e6dcd2dc 2319 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2320 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2321 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2322 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2323
53b381b3
DW		 2324 ret = btrfs_alloc_stripe_hash_table(fs_info);
2325 if (ret) {
83c8266a 2326 err = ret;
53b381b3
DW		 2327 goto fail_alloc;
2328 }
2329
0b86a832 2330 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2331 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2332
3c4bb26b 2333 invalidate_bdev(fs_devices->latest_bdev);
1104a885
DS		 2334
2335 /*
2336 * Read super block and check the signature bytes only
2337 */
a512bbf8 2338 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2339 if (!bh) {
2340 err = -EINVAL;
16cdcec7 2341 goto fail_alloc;
20b45077 2342 }
39279cc3 2343
1104a885
DS		 2344 /*
2345 * We want to check superblock checksum, the type is stored inside.
2346 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2347 */
2348 if (btrfs_check_super_csum(bh->b_data)) {
2349 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2350 err = -EINVAL;
2351 goto fail_alloc;
2352 }
2353
2354 /*
2355 * super_copy is zeroed at allocation time and we never touch the
2356 * following bytes up to INFO_SIZE, the checksum is calculated from
2357 * the whole block of INFO_SIZE
2358 */
6c41761f
DS		 2359 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2360 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2361 sizeof(*fs_info->super_for_commit));
a061fc8d 2362 brelse(bh);
5f39d397 2363
6c41761f 2364 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2365
1104a885
DS		 2366 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2367 if (ret) {
2368 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2369 err = -EINVAL;
2370 goto fail_alloc;
2371 }
2372
6c41761f 2373 disk_super = fs_info->super_copy;
0f7d52f4 2374 if (!btrfs_super_root(disk_super))
16cdcec7 2375 goto fail_alloc;
0f7d52f4 2376
acce952b 2377 /* check FS state, whether FS is broken. */
87533c47
MX		 2378 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2379 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2380
af31f5e5
CM		 2381 /*
2382 * run through our array of backup supers and setup
2383 * our ring pointer to the oldest one
2384 */
2385 generation = btrfs_super_generation(disk_super);
2386 find_oldest_super_backup(fs_info, generation);
2387
75e7cb7f
LB		 2388 /*
2389 * In the long term, we'll store the compression type in the super
2390 * block, and it'll be used for per file compression control.
2391 */
2392 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2393
2b82032c
YZ		 2394 ret = btrfs_parse_options(tree_root, options);
2395 if (ret) {
2396 err = ret;
16cdcec7 2397 goto fail_alloc;
2b82032c 2398 }
dfe25020 2399
f2b636e8
JB		 2400 features = btrfs_super_incompat_flags(disk_super) &
2401 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2402 if (features) {
2403 printk(KERN_ERR "BTRFS: couldn't mount because of "
2404 "unsupported optional features (%Lx).\n",
21380931 2405 (unsigned long long)features);
f2b636e8 2406 err = -EINVAL;
16cdcec7 2407 goto fail_alloc;
f2b636e8
JB		 2408 }
2409
727011e0
CM		 2410 if (btrfs_super_leafsize(disk_super) !=
2411 btrfs_super_nodesize(disk_super)) {
2412 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2413 "blocksizes don't match. node %d leaf %d\n",
2414 btrfs_super_nodesize(disk_super),
2415 btrfs_super_leafsize(disk_super));
2416 err = -EINVAL;
2417 goto fail_alloc;
2418 }
2419 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2420 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2421 "blocksize (%d) was too large\n",
2422 btrfs_super_leafsize(disk_super));
2423 err = -EINVAL;
2424 goto fail_alloc;
2425 }
2426
5d4f98a2 2427 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2428 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2429 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2430 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2431
3173a18f
JB		 2432 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2433 printk(KERN_ERR "btrfs: has skinny extents\n");
2434
727011e0
CM		 2435 /*
2436 * flag our filesystem as having big metadata blocks if
2437 * they are bigger than the page size
2438 */
2439 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2440 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2441 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2442 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2443 }
2444
bc3f116f
CM		 2445 nodesize = btrfs_super_nodesize(disk_super);
2446 leafsize = btrfs_super_leafsize(disk_super);
2447 sectorsize = btrfs_super_sectorsize(disk_super);
2448 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2449 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2450 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2451
2452 /*
2453 * mixed block groups end up with duplicate but slightly offset
2454 * extent buffers for the same range. It leads to corruptions
2455 */
2456 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2457 (sectorsize != leafsize)) {
2458 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2459 "are not allowed for mixed block groups on %s\n",
2460 sb->s_id);
2461 goto fail_alloc;
2462 }
2463
ceda0864
MX		 2464 /*
2465 * Needn't use the lock because there is no other task which will
2466 * update the flag.
2467 */
a6fa6fae 2468 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2469
f2b636e8
JB		 2470 features = btrfs_super_compat_ro_flags(disk_super) &
2471 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2472 if (!(sb->s_flags & MS_RDONLY) && features) {
2473 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2474 "unsupported option features (%Lx).\n",
21380931 2475 (unsigned long long)features);
f2b636e8 2476 err = -EINVAL;
16cdcec7 2477 goto fail_alloc;
f2b636e8 2478 }
61d92c32
CM		 2479
2480 btrfs_init_workers(&fs_info->generic_worker,
2481 "genwork", 1, NULL);
2482
5443be45 2483 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 2484 fs_info->thread_pool_size,
2485 &fs_info->generic_worker);
c8b97818 2486
771ed689 2487 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 2488 fs_info->thread_pool_size,
2489 &fs_info->generic_worker);
771ed689 2490
8ccf6f19
MX		 2491 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2492 fs_info->thread_pool_size,
2493 &fs_info->generic_worker);
2494
5443be45 2495 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 2496 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 2497 fs_info->thread_pool_size),
2498 &fs_info->generic_worker);
61b49440 2499
bab39bf9
JB		 2500 btrfs_init_workers(&fs_info->caching_workers, "cache",
2501 2, &fs_info->generic_worker);
2502
61b49440
CM		 2503 /* a higher idle thresh on the submit workers makes it much more
2504 * likely that bios will be send down in a sane order to the
2505 * devices
2506 */
2507 fs_info->submit_workers.idle_thresh = 64;
53863232 2508
771ed689 2509 fs_info->workers.idle_thresh = 16;
4a69a410 2510 fs_info->workers.ordered = 1;
61b49440 2511
771ed689
CM		 2512 fs_info->delalloc_workers.idle_thresh = 2;
2513 fs_info->delalloc_workers.ordered = 1;
2514
61d92c32
CM		 2515 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2516 &fs_info->generic_worker);
5443be45 2517 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2518 fs_info->thread_pool_size,
2519 &fs_info->generic_worker);
d20f7043 2520 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2521 fs_info->thread_pool_size,
2522 &fs_info->generic_worker);
cad321ad 2523 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2524 "endio-meta-write", fs_info->thread_pool_size,
2525 &fs_info->generic_worker);
53b381b3
DW		 2526 btrfs_init_workers(&fs_info->endio_raid56_workers,
2527 "endio-raid56", fs_info->thread_pool_size,
2528 &fs_info->generic_worker);
2529 btrfs_init_workers(&fs_info->rmw_workers,
2530 "rmw", fs_info->thread_pool_size,
2531 &fs_info->generic_worker);
5443be45 2532 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2533 fs_info->thread_pool_size,
2534 &fs_info->generic_worker);
0cb59c99
JB		 2535 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2536 1, &fs_info->generic_worker);
16cdcec7
MX		 2537 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2538 fs_info->thread_pool_size,
2539 &fs_info->generic_worker);
90519d66
AJ		 2540 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2541 fs_info->thread_pool_size,
2542 &fs_info->generic_worker);
2f232036
JS		 2543 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2544 &fs_info->generic_worker);
61b49440
CM		 2545
2546 /*
2547 * endios are largely parallel and should have a very
2548 * low idle thresh
2549 */
2550 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2551 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2552 fs_info->endio_raid56_workers.idle_thresh = 4;
2553 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2554
9042846b
CM		 2555 fs_info->endio_write_workers.idle_thresh = 2;
2556 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2557 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2558
0dc3b84a
JB		 2559 /*
2560 * btrfs_start_workers can really only fail because of ENOMEM so just
2561 * return -ENOMEM if any of these fail.
2562 */
2563 ret = btrfs_start_workers(&fs_info->workers);
2564 ret |= btrfs_start_workers(&fs_info->generic_worker);
2565 ret |= btrfs_start_workers(&fs_info->submit_workers);
2566 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2567 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2568 ret |= btrfs_start_workers(&fs_info->endio_workers);
2569 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2570 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2571 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2572 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2573 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2574 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2575 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2576 ret |= btrfs_start_workers(&fs_info->caching_workers);
2577 ret |= btrfs_start_workers(&fs_info->readahead_workers);
8ccf6f19 2578 ret |= btrfs_start_workers(&fs_info->flush_workers);
2f232036 2579 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
0dc3b84a 2580 if (ret) {
fed425c7 2581 err = -ENOMEM;
0dc3b84a
JB		 2582 goto fail_sb_buffer;
2583 }
4543df7e 2584
4575c9cc 2585 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2586 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2587 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2588
db94535d
CM		 2589 tree_root->nodesize = nodesize;
2590 tree_root->leafsize = leafsize;
2591 tree_root->sectorsize = sectorsize;
87ee04eb 2592 tree_root->stripesize = stripesize;
a061fc8d
CM		 2593
2594 sb->s_blocksize = sectorsize;
2595 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2596
3cae210f 2597 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
d397712b 2598 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2599 goto fail_sb_buffer;
2600 }
19c00ddc 2601
8d082fb7
LB		 2602 if (sectorsize != PAGE_SIZE) {
2603 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2604 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2605 goto fail_sb_buffer;
2606 }
2607
925baedd 2608 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2609 ret = btrfs_read_sys_array(tree_root);
925baedd 2610 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2611 if (ret) {
d397712b
CM		 2612 printk(KERN_WARNING "btrfs: failed to read the system "
2613 "array on %s\n", sb->s_id);
5d4f98a2 2614 goto fail_sb_buffer;
84eed90f 2615 }
0b86a832
CM		 2616
2617 blocksize = btrfs_level_size(tree_root,
2618 btrfs_super_chunk_root_level(disk_super));
84234f3a 2619 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2620
2621 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2622 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2623
2624 chunk_root->node = read_tree_block(chunk_root,
2625 btrfs_super_chunk_root(disk_super),
84234f3a 2626 blocksize, generation);
416bc658
JB		 2627 if (!chunk_root->node ||
2628 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
83121942
DW		 2629 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2630 sb->s_id);
af31f5e5 2631 goto fail_tree_roots;
83121942 2632 }
5d4f98a2
YZ		 2633 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2634 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2635
e17cade2 2636 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 2637 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2638 BTRFS_UUID_SIZE);
e17cade2 2639
0b86a832 2640 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2641 if (ret) {
d397712b
CM		 2642 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2643 sb->s_id);
af31f5e5 2644 goto fail_tree_roots;
2b82032c 2645 }
0b86a832 2646
8dabb742
SB		 2647 /*
2648 * keep the device that is marked to be the target device for the
2649 * dev_replace procedure
2650 */
2651 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2652
a6b0d5c8
CM		 2653 if (!fs_devices->latest_bdev) {
2654 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2655 sb->s_id);
2656 goto fail_tree_roots;
2657 }
2658
af31f5e5 2659retry_root_backup:
db94535d
CM		 2660 blocksize = btrfs_level_size(tree_root,
2661 btrfs_super_root_level(disk_super));
84234f3a 2662 generation = btrfs_super_generation(disk_super);
0b86a832 2663
e20d96d6 2664 tree_root->node = read_tree_block(tree_root,
db94535d 2665 btrfs_super_root(disk_super),
84234f3a 2666 blocksize, generation);
af31f5e5
CM		 2667 if (!tree_root->node ||
2668 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
83121942
DW		 2669 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2670 sb->s_id);
af31f5e5
CM		 2671
2672 goto recovery_tree_root;
83121942 2673 }
af31f5e5 2674
5d4f98a2
YZ		 2675 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2676 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d 2677
cb517eab
MX		 2678 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2679 location.type = BTRFS_ROOT_ITEM_KEY;
2680 location.offset = 0;
2681
2682 extent_root = btrfs_read_tree_root(tree_root, &location);
2683 if (IS_ERR(extent_root)) {
2684 ret = PTR_ERR(extent_root);
af31f5e5 2685 goto recovery_tree_root;
cb517eab 2686 }
0b86a832 2687 extent_root->track_dirty = 1;
cb517eab 2688 fs_info->extent_root = extent_root;
0b86a832 2689
cb517eab
MX		 2690 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2691 dev_root = btrfs_read_tree_root(tree_root, &location);
2692 if (IS_ERR(dev_root)) {
2693 ret = PTR_ERR(dev_root);
af31f5e5 2694 goto recovery_tree_root;
cb517eab 2695 }
5d4f98a2 2696 dev_root->track_dirty = 1;
cb517eab
MX		 2697 fs_info->dev_root = dev_root;
2698 btrfs_init_devices_late(fs_info);
3768f368 2699
cb517eab
MX		 2700 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2701 csum_root = btrfs_read_tree_root(tree_root, &location);
2702 if (IS_ERR(csum_root)) {
2703 ret = PTR_ERR(csum_root);
af31f5e5 2704 goto recovery_tree_root;
cb517eab 2705 }
d20f7043 2706 csum_root->track_dirty = 1;
cb517eab 2707 fs_info->csum_root = csum_root;
d20f7043 2708
cb517eab
MX		 2709 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2710 quota_root = btrfs_read_tree_root(tree_root, &location);
2711 if (!IS_ERR(quota_root)) {
bcef60f2
AJ		 2712 quota_root->track_dirty = 1;
2713 fs_info->quota_enabled = 1;
2714 fs_info->pending_quota_state = 1;
cb517eab 2715 fs_info->quota_root = quota_root;
bcef60f2
AJ		 2716 }
2717
f7a81ea4
SB		 2718 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2719 uuid_root = btrfs_read_tree_root(tree_root, &location);
2720 if (IS_ERR(uuid_root)) {
2721 ret = PTR_ERR(uuid_root);
2722 if (ret != -ENOENT)
2723 goto recovery_tree_root;
2724 create_uuid_tree = true;
2725 } else {
2726 uuid_root->track_dirty = 1;
2727 fs_info->uuid_root = uuid_root;
2728 }
2729
8929ecfa
YZ		 2730 fs_info->generation = generation;
2731 fs_info->last_trans_committed = generation;
8929ecfa 2732
68310a5e
ID		 2733 ret = btrfs_recover_balance(fs_info);
2734 if (ret) {
2735 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2736 goto fail_block_groups;
2737 }
2738
733f4fbb
SB		 2739 ret = btrfs_init_dev_stats(fs_info);
2740 if (ret) {
2741 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2742 ret);
2743 goto fail_block_groups;
2744 }
2745
8dabb742
SB		 2746 ret = btrfs_init_dev_replace(fs_info);
2747 if (ret) {
2748 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2749 goto fail_block_groups;
2750 }
2751
2752 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2753
c59021f8 2754 ret = btrfs_init_space_info(fs_info);
2755 if (ret) {
2756 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2757 goto fail_block_groups;
2758 }
2759
1b1d1f66
JB		 2760 ret = btrfs_read_block_groups(extent_root);
2761 if (ret) {
2762 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2763 goto fail_block_groups;
2764 }
5af3e8cc
SB		 2765 fs_info->num_tolerated_disk_barrier_failures =
2766 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2767 if (fs_info->fs_devices->missing_devices >
2768 fs_info->num_tolerated_disk_barrier_failures &&
2769 !(sb->s_flags & MS_RDONLY)) {
2770 printk(KERN_WARNING
2771 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2772 goto fail_block_groups;
2773 }
9078a3e1 2774
a74a4b97
CM		 2775 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2776 "btrfs-cleaner");
57506d50 2777 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2778 goto fail_block_groups;
a74a4b97
CM		 2779
2780 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2781 tree_root,
2782 "btrfs-transaction");
57506d50 2783 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2784 goto fail_cleaner;
a74a4b97 2785
c289811c
CM		 2786 if (!btrfs_test_opt(tree_root, SSD) &&
2787 !btrfs_test_opt(tree_root, NOSSD) &&
2788 !fs_info->fs_devices->rotating) {
2789 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2790 "mode\n");
2791 btrfs_set_opt(fs_info->mount_opt, SSD);
2792 }
2793
21adbd5c
SB		 2794#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2795 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2796 ret = btrfsic_mount(tree_root, fs_devices,
2797 btrfs_test_opt(tree_root,
2798 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2799 1 : 0,
2800 fs_info->check_integrity_print_mask);
2801 if (ret)
2802 printk(KERN_WARNING "btrfs: failed to initialize"
2803 " integrity check module %s\n", sb->s_id);
2804 }
2805#endif
bcef60f2
AJ		 2806 ret = btrfs_read_qgroup_config(fs_info);
2807 if (ret)
2808 goto fail_trans_kthread;
21adbd5c 2809
acce952b 2810 /* do not make disk changes in broken FS */
68ce9682 2811 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2812 u64 bytenr = btrfs_super_log_root(disk_super);
2813
7c2ca468 2814 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2815 printk(KERN_WARNING "Btrfs log replay required "
2816 "on RO media\n");
7c2ca468 2817 err = -EIO;
bcef60f2 2818 goto fail_qgroup;
7c2ca468 2819 }
e02119d5
CM		 2820 blocksize =
2821 btrfs_level_size(tree_root,
2822 btrfs_super_log_root_level(disk_super));
d18a2c44 2823
6f07e42e 2824 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2825 if (!log_tree_root) {
2826 err = -ENOMEM;
bcef60f2 2827 goto fail_qgroup;
676e4c86 2828 }
e02119d5
CM		 2829
2830 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2831 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2832
2833 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2834 blocksize,
2835 generation + 1);
416bc658
JB		 2836 if (!log_tree_root->node ||
2837 !extent_buffer_uptodate(log_tree_root->node)) {
2838 printk(KERN_ERR "btrfs: failed to read log tree\n");
2839 free_extent_buffer(log_tree_root->node);
2840 kfree(log_tree_root);
2841 goto fail_trans_kthread;
2842 }
79787eaa 2843 /* returns with log_tree_root freed on success */
e02119d5 2844 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2845 if (ret) {
2846 btrfs_error(tree_root->fs_info, ret,
2847 "Failed to recover log tree");
2848 free_extent_buffer(log_tree_root->node);
2849 kfree(log_tree_root);
2850 goto fail_trans_kthread;
2851 }
e556ce2c
YZ		 2852
2853 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2854 ret = btrfs_commit_super(tree_root);
2855 if (ret)
2856 goto fail_trans_kthread;
e556ce2c 2857 }
e02119d5 2858 }
1a40e23b 2859
76dda93c 2860 ret = btrfs_find_orphan_roots(tree_root);
79787eaa
JM		 2861 if (ret)
2862 goto fail_trans_kthread;
76dda93c 2863
7c2ca468 2864 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2865 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2
ID		 2866 if (ret)
2867 goto fail_trans_kthread;
d68fc57b 2868
5d4f98a2 2869 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2870 if (ret < 0) {
2871 printk(KERN_WARNING
2872 "btrfs: failed to recover relocation\n");
2873 err = -EINVAL;
bcef60f2 2874 goto fail_qgroup;
d7ce5843 2875 }
7c2ca468 2876 }
1a40e23b 2877
3de4586c
CM		 2878 location.objectid = BTRFS_FS_TREE_OBJECTID;
2879 location.type = BTRFS_ROOT_ITEM_KEY;
cb517eab 2880 location.offset = 0;
3de4586c 2881
3de4586c 2882 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3140c9a3
DC		 2883 if (IS_ERR(fs_info->fs_root)) {
2884 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2885 goto fail_qgroup;
3140c9a3 2886 }
c289811c 2887
2b6ba629
ID		 2888 if (sb->s_flags & MS_RDONLY)
2889 return 0;
59641015 2890
2b6ba629
ID		 2891 down_read(&fs_info->cleanup_work_sem);
2892 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2893 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2894 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2895 close_ctree(tree_root);
2896 return ret;
2897 }
2898 up_read(&fs_info->cleanup_work_sem);
59641015 2899
2b6ba629
ID		 2900 ret = btrfs_resume_balance_async(fs_info);
2901 if (ret) {
2902 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2903 close_ctree(tree_root);
2904 return ret;
e3acc2a6
JB		 2905 }
2906
8dabb742
SB		 2907 ret = btrfs_resume_dev_replace_async(fs_info);
2908 if (ret) {
2909 pr_warn("btrfs: failed to resume dev_replace\n");
2910 close_ctree(tree_root);
2911 return ret;
2912 }
2913
b382a324
JS		 2914 btrfs_qgroup_rescan_resume(fs_info);
2915
f7a81ea4
SB		 2916 if (create_uuid_tree) {
2917 pr_info("btrfs: creating UUID tree\n");
2918 ret = btrfs_create_uuid_tree(fs_info);
2919 if (ret) {
2920 pr_warn("btrfs: failed to create the UUID tree %d\n",
2921 ret);
2922 close_ctree(tree_root);
2923 return ret;
2924 }
2925 }
2926
ad2b2c80 2927 return 0;
39279cc3 2928
bcef60f2
AJ		 2929fail_qgroup:
2930 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2931fail_trans_kthread:
2932 kthread_stop(fs_info->transaction_kthread);
54067ae9 2933 btrfs_cleanup_transaction(fs_info->tree_root);
7b5ff90e 2934 del_fs_roots(fs_info);
3f157a2f 2935fail_cleaner:
a74a4b97 2936 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2937
2938 /*
2939 * make sure we're done with the btree inode before we stop our
2940 * kthreads
2941 */
2942 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2943
1b1d1f66 2944fail_block_groups:
54067ae9 2945 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2946 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2947
2948fail_tree_roots:
2949 free_root_pointers(fs_info, 1);
2b8195bb 2950 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2951
39279cc3 2952fail_sb_buffer:
7abadb64 2953 btrfs_stop_all_workers(fs_info);
16cdcec7 2954fail_alloc:
4543df7e 2955fail_iput:
586e46e2
ID		 2956 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2957
4543df7e 2958 iput(fs_info->btree_inode);
963d678b
MX		 2959fail_delalloc_bytes:
2960 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 2961fail_dirty_metadata_bytes:
2962 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 2963fail_bdi:
7e662854 2964 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2965fail_srcu:
2966 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2967fail:
53b381b3 2968 btrfs_free_stripe_hash_table(fs_info);
586e46e2 2969 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 2970 return err;
af31f5e5
CM		 2971
2972recovery_tree_root:
af31f5e5
CM		 2973 if (!btrfs_test_opt(tree_root, RECOVERY))
2974 goto fail_tree_roots;
2975
2976 free_root_pointers(fs_info, 0);
2977
2978 /* don't use the log in recovery mode, it won't be valid */
2979 btrfs_set_super_log_root(disk_super, 0);
2980
2981 /* we can't trust the free space cache either */
2982 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2983
2984 ret = next_root_backup(fs_info, fs_info->super_copy,
2985 &num_backups_tried, &backup_index);
2986 if (ret == -1)
2987 goto fail_block_groups;
2988 goto retry_root_backup;
eb60ceac
CM		 2989}
2990
f2984462
CM		 2991static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2992{
f2984462
CM		 2993 if (uptodate) {
2994 set_buffer_uptodate(bh);
2995 } else {
442a4f63
SB		 2996 struct btrfs_device *device = (struct btrfs_device *)
2997 bh->b_private;
2998
606686ee
JB		 2999 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3000 "I/O error on %s\n",
3001 rcu_str_deref(device->name));
1259ab75
CM		 3002 /* note, we dont' set_buffer_write_io_error because we have
3003 * our own ways of dealing with the IO errors
3004 */
f2984462 3005 clear_buffer_uptodate(bh);
442a4f63 3006 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 3007 }
3008 unlock_buffer(bh);
3009 put_bh(bh);
3010}
3011
a512bbf8
YZ		 3012struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3013{
3014 struct buffer_head *bh;
3015 struct buffer_head *latest = NULL;
3016 struct btrfs_super_block *super;
3017 int i;
3018 u64 transid = 0;
3019 u64 bytenr;
3020
3021 /* we would like to check all the supers, but that would make
3022 * a btrfs mount succeed after a mkfs from a different FS.
3023 * So, we need to add a special mount option to scan for
3024 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3025 */
3026 for (i = 0; i < 1; i++) {
3027 bytenr = btrfs_sb_offset(i);
8068a47e
AJ		 3028 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3029 i_size_read(bdev->bd_inode))
a512bbf8 3030 break;
8068a47e
AJ		 3031 bh = __bread(bdev, bytenr / 4096,
3032 BTRFS_SUPER_INFO_SIZE);
a512bbf8
YZ		 3033 if (!bh)
3034 continue;
3035
3036 super = (struct btrfs_super_block *)bh->b_data;
3037 if (btrfs_super_bytenr(super) != bytenr ||
3cae210f 3038 btrfs_super_magic(super) != BTRFS_MAGIC) {
a512bbf8
YZ		 3039 brelse(bh);
3040 continue;
3041 }
3042
3043 if (!latest || btrfs_super_generation(super) > transid) {
3044 brelse(latest);
3045 latest = bh;
3046 transid = btrfs_super_generation(super);
3047 } else {
3048 brelse(bh);
3049 }
3050 }
3051 return latest;
3052}
3053
4eedeb75
HH		 3054/*
3055 * this should be called twice, once with wait == 0 and
3056 * once with wait == 1. When wait == 0 is done, all the buffer heads
3057 * we write are pinned.
3058 *
3059 * They are released when wait == 1 is done.
3060 * max_mirrors must be the same for both runs, and it indicates how
3061 * many supers on this one device should be written.
3062 *
3063 * max_mirrors == 0 means to write them all.
3064 */
a512bbf8
YZ		 3065static int write_dev_supers(struct btrfs_device *device,
3066 struct btrfs_super_block *sb,
3067 int do_barriers, int wait, int max_mirrors)
3068{
3069 struct buffer_head *bh;
3070 int i;
3071 int ret;
3072 int errors = 0;
3073 u32 crc;
3074 u64 bytenr;
a512bbf8
YZ		 3075
3076 if (max_mirrors == 0)
3077 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3078
a512bbf8
YZ		 3079 for (i = 0; i < max_mirrors; i++) {
3080 bytenr = btrfs_sb_offset(i);
3081 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3082 break;
3083
3084 if (wait) {
3085 bh = __find_get_block(device->bdev, bytenr / 4096,
3086 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3087 if (!bh) {
3088 errors++;
3089 continue;
3090 }
a512bbf8 3091 wait_on_buffer(bh);
4eedeb75
HH		 3092 if (!buffer_uptodate(bh))
3093 errors++;
3094
3095 /* drop our reference */
3096 brelse(bh);
3097
3098 /* drop the reference from the wait == 0 run */
3099 brelse(bh);
3100 continue;
a512bbf8
YZ		 3101 } else {
3102 btrfs_set_super_bytenr(sb, bytenr);
3103
3104 crc = ~(u32)0;
b0496686 3105 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 3106 BTRFS_CSUM_SIZE, crc,
3107 BTRFS_SUPER_INFO_SIZE -
3108 BTRFS_CSUM_SIZE);
3109 btrfs_csum_final(crc, sb->csum);
3110
4eedeb75
HH		 3111 /*
3112 * one reference for us, and we leave it for the
3113 * caller
3114 */
a512bbf8
YZ		 3115 bh = __getblk(device->bdev, bytenr / 4096,
3116 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3117 if (!bh) {
3118 printk(KERN_ERR "btrfs: couldn't get super "
3119 "buffer head for bytenr %Lu\n", bytenr);
3120 errors++;
3121 continue;
3122 }
3123
a512bbf8
YZ		 3124 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3125
4eedeb75 3126 /* one reference for submit_bh */
a512bbf8 3127 get_bh(bh);
4eedeb75
HH		 3128
3129 set_buffer_uptodate(bh);
a512bbf8
YZ		 3130 lock_buffer(bh);
3131 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3132 bh->b_private = device;
a512bbf8
YZ		 3133 }
3134
387125fc
CM		 3135 /*
3136 * we fua the first super. The others we allow
3137 * to go down lazy.
3138 */
21adbd5c 3139 ret = btrfsic_submit_bh(WRITE_FUA, bh);
4eedeb75 3140 if (ret)
a512bbf8 3141 errors++;
a512bbf8
YZ		 3142 }
3143 return errors < i ? 0 : -1;
3144}
3145
387125fc
CM		 3146/*
3147 * endio for the write_dev_flush, this will wake anyone waiting
3148 * for the barrier when it is done
3149 */
3150static void btrfs_end_empty_barrier(struct bio *bio, int err)
3151{
3152 if (err) {
3153 if (err == -EOPNOTSUPP)
3154 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3155 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3156 }
3157 if (bio->bi_private)
3158 complete(bio->bi_private);
3159 bio_put(bio);
3160}
3161
3162/*
3163 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3164 * sent down. With wait == 1, it waits for the previous flush.
3165 *
3166 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3167 * capable
3168 */
3169static int write_dev_flush(struct btrfs_device *device, int wait)
3170{
3171 struct bio *bio;
3172 int ret = 0;
3173
3174 if (device->nobarriers)
3175 return 0;
3176
3177 if (wait) {
3178 bio = device->flush_bio;
3179 if (!bio)
3180 return 0;
3181
3182 wait_for_completion(&device->flush_wait);
3183
3184 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
606686ee
JB		 3185 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3186 rcu_str_deref(device->name));
387125fc 3187 device->nobarriers = 1;
5af3e8cc 3188 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3189 ret = -EIO;
5af3e8cc
SB		 3190 btrfs_dev_stat_inc_and_print(device,
3191 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3192 }
3193
3194 /* drop the reference from the wait == 0 run */
3195 bio_put(bio);
3196 device->flush_bio = NULL;
3197
3198 return ret;
3199 }
3200
3201 /*
3202 * one reference for us, and we leave it for the
3203 * caller
3204 */
9c017abc 3205 device->flush_bio = NULL;
9be3395b 3206 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
387125fc
CM		 3207 if (!bio)
3208 return -ENOMEM;
3209
3210 bio->bi_end_io = btrfs_end_empty_barrier;
3211 bio->bi_bdev = device->bdev;
3212 init_completion(&device->flush_wait);
3213 bio->bi_private = &device->flush_wait;
3214 device->flush_bio = bio;
3215
3216 bio_get(bio);
21adbd5c 3217 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3218
3219 return 0;
3220}
3221
3222/*
3223 * send an empty flush down to each device in parallel,
3224 * then wait for them
3225 */
3226static int barrier_all_devices(struct btrfs_fs_info *info)
3227{
3228 struct list_head *head;
3229 struct btrfs_device *dev;
5af3e8cc
SB		 3230 int errors_send = 0;
3231 int errors_wait = 0;
387125fc
CM		 3232 int ret;
3233
3234 /* send down all the barriers */
3235 head = &info->fs_devices->devices;
3236 list_for_each_entry_rcu(dev, head, dev_list) {
3237 if (!dev->bdev) {
5af3e8cc 3238 errors_send++;
387125fc
CM		 3239 continue;
3240 }
3241 if (!dev->in_fs_metadata || !dev->writeable)
3242 continue;
3243
3244 ret = write_dev_flush(dev, 0);
3245 if (ret)
5af3e8cc 3246 errors_send++;
387125fc
CM		 3247 }
3248
3249 /* wait for all the barriers */
3250 list_for_each_entry_rcu(dev, head, dev_list) {
3251 if (!dev->bdev) {
5af3e8cc 3252 errors_wait++;
387125fc
CM		 3253 continue;
3254 }
3255 if (!dev->in_fs_metadata || !dev->writeable)
3256 continue;
3257
3258 ret = write_dev_flush(dev, 1);
3259 if (ret)
5af3e8cc 3260 errors_wait++;
387125fc 3261 }
5af3e8cc
SB		 3262 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3263 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3264 return -EIO;
3265 return 0;
3266}
3267
5af3e8cc
SB		 3268int btrfs_calc_num_tolerated_disk_barrier_failures(
3269 struct btrfs_fs_info *fs_info)
3270{
3271 struct btrfs_ioctl_space_info space;
3272 struct btrfs_space_info *sinfo;
3273 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3274 BTRFS_BLOCK_GROUP_SYSTEM,
3275 BTRFS_BLOCK_GROUP_METADATA,
3276 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3277 int num_types = 4;
3278 int i;
3279 int c;
3280 int num_tolerated_disk_barrier_failures =
3281 (int)fs_info->fs_devices->num_devices;
3282
3283 for (i = 0; i < num_types; i++) {
3284 struct btrfs_space_info *tmp;
3285
3286 sinfo = NULL;
3287 rcu_read_lock();
3288 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3289 if (tmp->flags == types[i]) {
3290 sinfo = tmp;
3291 break;
3292 }
3293 }
3294 rcu_read_unlock();
3295
3296 if (!sinfo)
3297 continue;
3298
3299 down_read(&sinfo->groups_sem);
3300 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3301 if (!list_empty(&sinfo->block_groups[c])) {
3302 u64 flags;
3303
3304 btrfs_get_block_group_info(
3305 &sinfo->block_groups[c], &space);
3306 if (space.total_bytes == 0 ||
3307 space.used_bytes == 0)
3308 continue;
3309 flags = space.flags;
3310 /*
3311 * return
3312 * 0: if dup, single or RAID0 is configured for
3313 * any of metadata, system or data, else
3314 * 1: if RAID5 is configured, or if RAID1 or
3315 * RAID10 is configured and only two mirrors
3316 * are used, else
3317 * 2: if RAID6 is configured, else
3318 * num_mirrors - 1: if RAID1 or RAID10 is
3319 * configured and more than
3320 * 2 mirrors are used.
3321 */
3322 if (num_tolerated_disk_barrier_failures > 0 &&
3323 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3324 BTRFS_BLOCK_GROUP_RAID0)) ||
3325 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3326 == 0)))
3327 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3328 else if (num_tolerated_disk_barrier_failures > 1) {
3329 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3330 BTRFS_BLOCK_GROUP_RAID5 |
3331 BTRFS_BLOCK_GROUP_RAID10)) {
3332 num_tolerated_disk_barrier_failures = 1;
3333 } else if (flags &
15b0a89d 3334 BTRFS_BLOCK_GROUP_RAID6) {
53b381b3
DW		 3335 num_tolerated_disk_barrier_failures = 2;
3336 }
3337 }
5af3e8cc
SB		 3338 }
3339 }
3340 up_read(&sinfo->groups_sem);
3341 }
3342
3343 return num_tolerated_disk_barrier_failures;
3344}
3345
48a3b636 3346static int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3347{
e5e9a520 3348 struct list_head *head;
f2984462 3349 struct btrfs_device *dev;
a061fc8d 3350 struct btrfs_super_block *sb;
f2984462 3351 struct btrfs_dev_item *dev_item;
f2984462
CM		 3352 int ret;
3353 int do_barriers;
a236aed1
CM		 3354 int max_errors;
3355 int total_errors = 0;
a061fc8d 3356 u64 flags;
f2984462 3357
6c41761f 3358 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
f2984462 3359 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3360 backup_super_roots(root->fs_info);
f2984462 3361
6c41761f 3362 sb = root->fs_info->super_for_commit;
a061fc8d 3363 dev_item = &sb->dev_item;
e5e9a520 3364
174ba509 3365 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3366 head = &root->fs_info->fs_devices->devices;
387125fc 3367
5af3e8cc
SB		 3368 if (do_barriers) {
3369 ret = barrier_all_devices(root->fs_info);
3370 if (ret) {
3371 mutex_unlock(
3372 &root->fs_info->fs_devices->device_list_mutex);
3373 btrfs_error(root->fs_info, ret,
3374 "errors while submitting device barriers.");
3375 return ret;
3376 }
3377 }
387125fc 3378
1f78160c 3379 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3380 if (!dev->bdev) {
3381 total_errors++;
3382 continue;
3383 }
2b82032c 3384 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3385 continue;
3386
2b82032c 3387 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3388 btrfs_set_stack_device_type(dev_item, dev->type);
3389 btrfs_set_stack_device_id(dev_item, dev->devid);
3390 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3391 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3392 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3393 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3394 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3395 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3396 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3397
a061fc8d
CM		 3398 flags = btrfs_super_flags(sb);
3399 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3400
a512bbf8 3401 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3402 if (ret)
3403 total_errors++;
f2984462 3404 }
a236aed1 3405 if (total_errors > max_errors) {
d397712b
CM		 3406 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3407 total_errors);
79787eaa
JM		 3408
3409 /* This shouldn't happen. FUA is masked off if unsupported */
a236aed1
CM		 3410 BUG();
3411 }
f2984462 3412
a512bbf8 3413 total_errors = 0;
1f78160c 3414 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3415 if (!dev->bdev)
3416 continue;
2b82032c 3417 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3418 continue;
3419
a512bbf8
YZ		 3420 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3421 if (ret)
3422 total_errors++;
f2984462 3423 }
174ba509 3424 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3425 if (total_errors > max_errors) {
79787eaa
JM		 3426 btrfs_error(root->fs_info, -EIO,
3427 "%d errors while writing supers", total_errors);
3428 return -EIO;
a236aed1 3429 }
f2984462
CM		 3430 return 0;
3431}
3432
a512bbf8
YZ		 3433int write_ctree_super(struct btrfs_trans_handle *trans,
3434 struct btrfs_root *root, int max_mirrors)
eb60ceac 3435{
e66f709b 3436 int ret;
5f39d397 3437
a512bbf8 3438 ret = write_all_supers(root, max_mirrors);
5f39d397 3439 return ret;
cfaa7295
CM		 3440}
3441
cb517eab
MX		 3442/* Drop a fs root from the radix tree and free it. */
3443void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3444 struct btrfs_root *root)
2619ba1f 3445{
4df27c4d 3446 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3447 radix_tree_delete(&fs_info->fs_roots_radix,
3448 (unsigned long)root->root_key.objectid);
4df27c4d 3449 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3450
3451 if (btrfs_root_refs(&root->root_item) == 0)
3452 synchronize_srcu(&fs_info->subvol_srcu);
3453
d7634482 3454 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3321719e
LB		 3455 btrfs_free_log(NULL, root);
3456 btrfs_free_log_root_tree(NULL, fs_info);
3457 }
3458
581bb050
LZ		 3459 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3460 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3461 free_fs_root(root);
4df27c4d
YZ		 3462}
3463
3464static void free_fs_root(struct btrfs_root *root)
3465{
82d5902d 3466 iput(root->cache_inode);
4df27c4d 3467 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
0ee5dc67
AV		 3468 if (root->anon_dev)
3469 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 3470 free_extent_buffer(root->node);
3471 free_extent_buffer(root->commit_root);
581bb050
LZ		 3472 kfree(root->free_ino_ctl);
3473 kfree(root->free_ino_pinned);
d397712b 3474 kfree(root->name);
b0feb9d9 3475 btrfs_put_fs_root(root);
2619ba1f
CM		 3476}
3477
cb517eab
MX		 3478void btrfs_free_fs_root(struct btrfs_root *root)
3479{
3480 free_fs_root(root);
2619ba1f
CM		 3481}
3482
c146afad 3483int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3484{
c146afad
YZ		 3485 u64 root_objectid = 0;
3486 struct btrfs_root *gang[8];
3487 int i;
3768f368 3488 int ret;
e089f05c 3489
c146afad
YZ		 3490 while (1) {
3491 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3492 (void **)gang, root_objectid,
3493 ARRAY_SIZE(gang));
3494 if (!ret)
3495 break;
5d4f98a2
YZ		 3496
3497 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 3498 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 3499 int err;
3500
c146afad 3501 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3502 err = btrfs_orphan_cleanup(gang[i]);
3503 if (err)
3504 return err;
c146afad
YZ		 3505 }
3506 root_objectid++;
3507 }
3508 return 0;
3509}
a2135011 3510
c146afad
YZ		 3511int btrfs_commit_super(struct btrfs_root *root)
3512{
3513 struct btrfs_trans_handle *trans;
3514 int ret;
a74a4b97 3515
c146afad 3516 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3517 btrfs_run_delayed_iputs(root);
c146afad 3518 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3519 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3520
3521 /* wait until ongoing cleanup work done */
3522 down_write(&root->fs_info->cleanup_work_sem);
3523 up_write(&root->fs_info->cleanup_work_sem);
3524
7a7eaa40 3525 trans = btrfs_join_transaction(root);
3612b495
TI		 3526 if (IS_ERR(trans))
3527 return PTR_ERR(trans);
54aa1f4d 3528 ret = btrfs_commit_transaction(trans, root);
79787eaa
JM		 3529 if (ret)
3530 return ret;
c146afad 3531 /* run commit again to drop the original snapshot */
7a7eaa40 3532 trans = btrfs_join_transaction(root);
3612b495
TI		 3533 if (IS_ERR(trans))
3534 return PTR_ERR(trans);
79787eaa
JM		 3535 ret = btrfs_commit_transaction(trans, root);
3536 if (ret)
3537 return ret;
79154b1b 3538 ret = btrfs_write_and_wait_transaction(NULL, root);
79787eaa
JM		 3539 if (ret) {
3540 btrfs_error(root->fs_info, ret,
3541 "Failed to sync btree inode to disk.");
3542 return ret;
3543 }
d6bfde87 3544
a512bbf8 3545 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 3546 return ret;
3547}
3548
3549int close_ctree(struct btrfs_root *root)
3550{
3551 struct btrfs_fs_info *fs_info = root->fs_info;
3552 int ret;
3553
3554 fs_info->closing = 1;
3555 smp_mb();
3556
837d5b6e 3557 /* pause restriper - we want to resume on mount */
aa1b8cd4 3558 btrfs_pause_balance(fs_info);
837d5b6e 3559
8dabb742
SB		 3560 btrfs_dev_replace_suspend_for_unmount(fs_info);
3561
aa1b8cd4 3562 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3563
3564 /* wait for any defraggers to finish */
3565 wait_event(fs_info->transaction_wait,
3566 (atomic_read(&fs_info->defrag_running) == 0));
3567
3568 /* clear out the rbtree of defraggable inodes */
26176e7c 3569 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3570
c146afad 3571 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3572 ret = btrfs_commit_super(root);
3573 if (ret)
3574 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3575 }
3576
87533c47 3577 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3578 btrfs_error_commit_super(root);
0f7d52f4 3579
300e4f8a
JB		 3580 btrfs_put_block_group_cache(fs_info);
3581
e3029d9f
AV		 3582 kthread_stop(fs_info->transaction_kthread);
3583 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3584
f25784b3
YZ		 3585 fs_info->closing = 2;
3586 smp_mb();
3587
bcef60f2
AJ		 3588 btrfs_free_qgroup_config(root->fs_info);
3589
963d678b
MX		 3590 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3591 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3592 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3593 }
bcc63abb 3594
e3029d9f 3595 btrfs_free_block_groups(fs_info);
d10c5f31 3596
13e6c37b 3597 btrfs_stop_all_workers(fs_info);
2932505a 3598
c146afad 3599 del_fs_roots(fs_info);
d10c5f31 3600
13e6c37b 3601 free_root_pointers(fs_info, 1);
9ad6b7bc 3602
13e6c37b 3603 iput(fs_info->btree_inode);
d6bfde87 3604
21adbd5c
SB		 3605#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3606 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3607 btrfsic_unmount(root, fs_info->fs_devices);
3608#endif
3609
dfe25020 3610 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3611 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3612
e2d84521 3613 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3614 percpu_counter_destroy(&fs_info->delalloc_bytes);
04160088 3615 bdi_destroy(&fs_info->bdi);
76dda93c 3616 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3617
53b381b3
DW		 3618 btrfs_free_stripe_hash_table(fs_info);
3619
eb60ceac
CM		 3620 return 0;
3621}
3622
b9fab919
CM		 3623int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3624 int atomic)
5f39d397 3625{
1259ab75 3626 int ret;
727011e0 3627 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3628
0b32f4bb 3629 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3630 if (!ret)
3631 return ret;
3632
3633 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3634 parent_transid, atomic);
3635 if (ret == -EAGAIN)
3636 return ret;
1259ab75 3637 return !ret;
5f39d397
CM		 3638}
3639
3640int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3641{
0b32f4bb 3642 return set_extent_buffer_uptodate(buf);
5f39d397 3643}
6702ed49 3644
5f39d397
CM		 3645void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3646{
727011e0 3647 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
5f39d397 3648 u64 transid = btrfs_header_generation(buf);
b9473439 3649 int was_dirty;
b4ce94de 3650
b9447ef8 3651 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3652 if (transid != root->fs_info->generation)
3653 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3654 "found %llu running %llu\n",
db94535d 3655 (unsigned long long)buf->start,
d397712b
CM		 3656 (unsigned long long)transid,
3657 (unsigned long long)root->fs_info->generation);
0b32f4bb 3658 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3659 if (!was_dirty)
3660 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3661 buf->len,
3662 root->fs_info->dirty_metadata_batch);
eb60ceac
CM		 3663}
3664
b53d3f5d
LB		 3665static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3666 int flush_delayed)
16cdcec7
MX		 3667{
3668 /*
3669 * looks as though older kernels can get into trouble with
3670 * this code, they end up stuck in balance_dirty_pages forever
3671 */
e2d84521 3672 int ret;
16cdcec7
MX		 3673
3674 if (current->flags & PF_MEMALLOC)
3675 return;
3676
b53d3f5d
LB		 3677 if (flush_delayed)
3678 btrfs_balance_delayed_items(root);
16cdcec7 3679
e2d84521
MX		 3680 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3681 BTRFS_DIRTY_METADATA_THRESH);
3682 if (ret > 0) {
d0e1d66b
NJ		 3683 balance_dirty_pages_ratelimited(
3684 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3685 }
3686 return;
3687}
3688
b53d3f5d 3689void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3690{
b53d3f5d
LB		 3691 __btrfs_btree_balance_dirty(root, 1);
3692}
585ad2c3 3693
b53d3f5d
LB		 3694void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3695{
3696 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3697}
6b80053d 3698
ca7a79ad 3699int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3700{
727011e0 3701 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3702 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3703}
0da5468f 3704
fcd1f065 3705static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3706 int read_only)
3707{
1104a885
DS		 3708 /*
3709 * Placeholder for checks
3710 */
fcd1f065 3711 return 0;
acce952b 3712}
3713
48a3b636 3714static void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3715{
acce952b 3716 mutex_lock(&root->fs_info->cleaner_mutex);
3717 btrfs_run_delayed_iputs(root);
3718 mutex_unlock(&root->fs_info->cleaner_mutex);
3719
3720 down_write(&root->fs_info->cleanup_work_sem);
3721 up_write(&root->fs_info->cleanup_work_sem);
3722
3723 /* cleanup FS via transaction */
3724 btrfs_cleanup_transaction(root);
acce952b 3725}
3726
569e0f35
JB		 3727static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3728 struct btrfs_root *root)
acce952b 3729{
3730 struct btrfs_inode *btrfs_inode;
3731 struct list_head splice;
3732
3733 INIT_LIST_HEAD(&splice);
3734
3735 mutex_lock(&root->fs_info->ordered_operations_mutex);
199c2a9c 3736 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3737
569e0f35 3738 list_splice_init(&t->ordered_operations, &splice);
acce952b 3739 while (!list_empty(&splice)) {
3740 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3741 ordered_operations);
3742
3743 list_del_init(&btrfs_inode->ordered_operations);
199c2a9c 3744 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3745
3746 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3747
199c2a9c 3748 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3749 }
3750
199c2a9c 3751 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3752 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3753}
3754
143bede5 3755static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3756{
acce952b 3757 struct btrfs_ordered_extent *ordered;
acce952b 3758
199c2a9c 3759 spin_lock(&root->ordered_extent_lock);
779880ef
JB		 3760 /*
3761 * This will just short circuit the ordered completion stuff which will
3762 * make sure the ordered extent gets properly cleaned up.
3763 */
199c2a9c 3764 list_for_each_entry(ordered, &root->ordered_extents,
779880ef
JB		 3765 root_extent_list)
3766 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
199c2a9c
MX		 3767 spin_unlock(&root->ordered_extent_lock);
3768}
3769
3770static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3771{
3772 struct btrfs_root *root;
3773 struct list_head splice;
3774
3775 INIT_LIST_HEAD(&splice);
3776
3777 spin_lock(&fs_info->ordered_root_lock);
3778 list_splice_init(&fs_info->ordered_roots, &splice);
3779 while (!list_empty(&splice)) {
3780 root = list_first_entry(&splice, struct btrfs_root,
3781 ordered_root);
3782 list_del_init(&root->ordered_root);
3783
3784 btrfs_destroy_ordered_extents(root);
3785
3786 cond_resched_lock(&fs_info->ordered_root_lock);
3787 }
3788 spin_unlock(&fs_info->ordered_root_lock);
acce952b 3789}
3790
35a3621b
SB		 3791static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3792 struct btrfs_root *root)
acce952b 3793{
3794 struct rb_node *node;
3795 struct btrfs_delayed_ref_root *delayed_refs;
3796 struct btrfs_delayed_ref_node *ref;
3797 int ret = 0;
3798
3799 delayed_refs = &trans->delayed_refs;
3800
3801 spin_lock(&delayed_refs->lock);
3802 if (delayed_refs->num_entries == 0) {
cfece4db 3803 spin_unlock(&delayed_refs->lock);
acce952b 3804 printk(KERN_INFO "delayed_refs has NO entry\n");
3805 return ret;
3806 }
3807
b939d1ab 3808 while ((node = rb_first(&delayed_refs->root)) != NULL) {
eb12db69 3809 struct btrfs_delayed_ref_head *head = NULL;
e78417d1 3810 bool pin_bytes = false;
acce952b 3811
eb12db69 3812 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
acce952b 3813 atomic_set(&ref->refs, 1);
3814 if (btrfs_delayed_ref_is_head(ref)) {
acce952b 3815
3816 head = btrfs_delayed_node_to_head(ref);
b939d1ab
JB		 3817 if (!mutex_trylock(&head->mutex)) {
3818 atomic_inc(&ref->refs);
3819 spin_unlock(&delayed_refs->lock);
3820
3821 /* Need to wait for the delayed ref to run */
3822 mutex_lock(&head->mutex);
3823 mutex_unlock(&head->mutex);
3824 btrfs_put_delayed_ref(ref);
3825
e18fca73 3826 spin_lock(&delayed_refs->lock);
b939d1ab
JB		 3827 continue;
3828 }
3829
54067ae9 3830 if (head->must_insert_reserved)
e78417d1 3831 pin_bytes = true;
78a6184a 3832 btrfs_free_delayed_extent_op(head->extent_op);
acce952b 3833 delayed_refs->num_heads--;
3834 if (list_empty(&head->cluster))
3835 delayed_refs->num_heads_ready--;
3836 list_del_init(&head->cluster);
acce952b 3837 }
eb12db69 3838
b939d1ab
JB		 3839 ref->in_tree = 0;
3840 rb_erase(&ref->rb_node, &delayed_refs->root);
3841 delayed_refs->num_entries--;
acce952b 3842 spin_unlock(&delayed_refs->lock);
e78417d1
JB		 3843 if (head) {
3844 if (pin_bytes)
3845 btrfs_pin_extent(root, ref->bytenr,
3846 ref->num_bytes, 1);
3847 mutex_unlock(&head->mutex);
3848 }
acce952b 3849 btrfs_put_delayed_ref(ref);
3850
3851 cond_resched();
3852 spin_lock(&delayed_refs->lock);
3853 }
3854
3855 spin_unlock(&delayed_refs->lock);
3856
3857 return ret;
3858}
3859
aec8030a 3860static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
acce952b 3861{
3862 struct btrfs_pending_snapshot *snapshot;
3863 struct list_head splice;
3864
3865 INIT_LIST_HEAD(&splice);
3866
3867 list_splice_init(&t->pending_snapshots, &splice);
3868
3869 while (!list_empty(&splice)) {
3870 snapshot = list_entry(splice.next,
3871 struct btrfs_pending_snapshot,
3872 list);
aec8030a 3873 snapshot->error = -ECANCELED;
acce952b 3874 list_del_init(&snapshot->list);
acce952b 3875 }
acce952b 3876}
3877
143bede5 3878static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3879{
3880 struct btrfs_inode *btrfs_inode;
3881 struct list_head splice;
3882
3883 INIT_LIST_HEAD(&splice);
3884
eb73c1b7
MX		 3885 spin_lock(&root->delalloc_lock);
3886 list_splice_init(&root->delalloc_inodes, &splice);
acce952b 3887
3888 while (!list_empty(&splice)) {
eb73c1b7
MX		 3889 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3890 delalloc_inodes);
acce952b 3891
3892 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3893 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3894 &btrfs_inode->runtime_flags);
eb73c1b7 3895 spin_unlock(&root->delalloc_lock);
acce952b 3896
3897 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3898
eb73c1b7 3899 spin_lock(&root->delalloc_lock);
acce952b 3900 }
3901
eb73c1b7
MX		 3902 spin_unlock(&root->delalloc_lock);
3903}
3904
3905static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3906{
3907 struct btrfs_root *root;
3908 struct list_head splice;
3909
3910 INIT_LIST_HEAD(&splice);
3911
3912 spin_lock(&fs_info->delalloc_root_lock);
3913 list_splice_init(&fs_info->delalloc_roots, &splice);
3914 while (!list_empty(&splice)) {
3915 root = list_first_entry(&splice, struct btrfs_root,
3916 delalloc_root);
3917 list_del_init(&root->delalloc_root);
3918 root = btrfs_grab_fs_root(root);
3919 BUG_ON(!root);
3920 spin_unlock(&fs_info->delalloc_root_lock);
3921
3922 btrfs_destroy_delalloc_inodes(root);
3923 btrfs_put_fs_root(root);
3924
3925 spin_lock(&fs_info->delalloc_root_lock);
3926 }
3927 spin_unlock(&fs_info->delalloc_root_lock);
acce952b 3928}
3929
3930static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3931 struct extent_io_tree *dirty_pages,
3932 int mark)
3933{
3934 int ret;
acce952b 3935 struct extent_buffer *eb;
3936 u64 start = 0;
3937 u64 end;
acce952b 3938
3939 while (1) {
3940 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3941 mark, NULL);
acce952b 3942 if (ret)
3943 break;
3944
3945 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3946 while (start <= end) {
fd8b2b61
JB		 3947 eb = btrfs_find_tree_block(root, start,
3948 root->leafsize);
69a85bd8 3949 start += root->leafsize;
fd8b2b61 3950 if (!eb)
acce952b 3951 continue;
fd8b2b61 3952 wait_on_extent_buffer_writeback(eb);
acce952b 3953
fd8b2b61
JB		 3954 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3955 &eb->bflags))
3956 clear_extent_buffer_dirty(eb);
3957 free_extent_buffer_stale(eb);
acce952b 3958 }
3959 }
3960
3961 return ret;
3962}
3963
3964static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3965 struct extent_io_tree *pinned_extents)
3966{
3967 struct extent_io_tree *unpin;
3968 u64 start;
3969 u64 end;
3970 int ret;
ed0eaa14 3971 bool loop = true;
acce952b 3972
3973 unpin = pinned_extents;
ed0eaa14 3974again:
acce952b 3975 while (1) {
3976 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 3977 EXTENT_DIRTY, NULL);
acce952b 3978 if (ret)
3979 break;
3980
3981 /* opt_discard */
5378e607
LD		 3982 if (btrfs_test_opt(root, DISCARD))
3983 ret = btrfs_error_discard_extent(root, start,
3984 end + 1 - start,
3985 NULL);
acce952b 3986
3987 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3988 btrfs_error_unpin_extent_range(root, start, end);
3989 cond_resched();
3990 }
3991
ed0eaa14
LB		 3992 if (loop) {
3993 if (unpin == &root->fs_info->freed_extents[0])
3994 unpin = &root->fs_info->freed_extents[1];
3995 else
3996 unpin = &root->fs_info->freed_extents[0];
3997 loop = false;
3998 goto again;
3999 }
4000
acce952b 4001 return 0;
4002}
4003
49b25e05
JM		 4004void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4005 struct btrfs_root *root)
4006{
4007 btrfs_destroy_delayed_refs(cur_trans, root);
4008 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
4009 cur_trans->dirty_pages.dirty_bytes);
4010
4a9d8bde 4011 cur_trans->state = TRANS_STATE_COMMIT_START;
d7096fc3 4012 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 4013
aec8030a
MX		 4014 btrfs_evict_pending_snapshots(cur_trans);
4015
4a9d8bde 4016 cur_trans->state = TRANS_STATE_UNBLOCKED;
d7096fc3 4017 wake_up(&root->fs_info->transaction_wait);
49b25e05 4018
67cde344
MX		 4019 btrfs_destroy_delayed_inodes(root);
4020 btrfs_assert_delayed_root_empty(root);
49b25e05 4021
49b25e05
JM		 4022 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4023 EXTENT_DIRTY);
6e841e32
LB		 4024 btrfs_destroy_pinned_extent(root,
4025 root->fs_info->pinned_extents);
49b25e05 4026
4a9d8bde
MX		 4027 cur_trans->state =TRANS_STATE_COMPLETED;
4028 wake_up(&cur_trans->commit_wait);
4029
49b25e05
JM		 4030 /*
4031 memset(cur_trans, 0, sizeof(*cur_trans));
4032 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4033 */
4034}
4035
48a3b636 4036static int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 4037{
4038 struct btrfs_transaction *t;
4039 LIST_HEAD(list);
4040
acce952b 4041 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4042
a4abeea4 4043 spin_lock(&root->fs_info->trans_lock);
acce952b 4044 list_splice_init(&root->fs_info->trans_list, &list);
ac673879 4045 root->fs_info->running_transaction = NULL;
a4abeea4
JB		 4046 spin_unlock(&root->fs_info->trans_lock);
4047
acce952b 4048 while (!list_empty(&list)) {
4049 t = list_entry(list.next, struct btrfs_transaction, list);
acce952b 4050
569e0f35 4051 btrfs_destroy_ordered_operations(t, root);
acce952b 4052
199c2a9c 4053 btrfs_destroy_all_ordered_extents(root->fs_info);
acce952b 4054
4055 btrfs_destroy_delayed_refs(t, root);
4056
4a9d8bde
MX		 4057 /*
4058 * FIXME: cleanup wait for commit
4059 * We needn't acquire the lock here, because we are during
4060 * the umount, there is no other task which will change it.
4061 */
4062 t->state = TRANS_STATE_COMMIT_START;
66657b31 4063 smp_mb();
acce952b 4064 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4065 wake_up(&root->fs_info->transaction_blocked_wait);
4066
aec8030a
MX		 4067 btrfs_evict_pending_snapshots(t);
4068
4a9d8bde 4069 t->state = TRANS_STATE_UNBLOCKED;
66657b31 4070 smp_mb();
acce952b 4071 if (waitqueue_active(&root->fs_info->transaction_wait))
4072 wake_up(&root->fs_info->transaction_wait);
acce952b 4073
67cde344
MX		 4074 btrfs_destroy_delayed_inodes(root);
4075 btrfs_assert_delayed_root_empty(root);
4076
eb73c1b7 4077 btrfs_destroy_all_delalloc_inodes(root->fs_info);
acce952b 4078
4079 btrfs_destroy_marked_extents(root, &t->dirty_pages,
4080 EXTENT_DIRTY);
4081
4082 btrfs_destroy_pinned_extent(root,
4083 root->fs_info->pinned_extents);
4084
4a9d8bde
MX		 4085 t->state = TRANS_STATE_COMPLETED;
4086 smp_mb();
4087 if (waitqueue_active(&t->commit_wait))
4088 wake_up(&t->commit_wait);
4089
13c5a93e 4090 atomic_set(&t->use_count, 0);
acce952b 4091 list_del_init(&t->list);
4092 memset(t, 0, sizeof(*t));
4093 kmem_cache_free(btrfs_transaction_cachep, t);
4094 }
4095
4096 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 4097
4098 return 0;
4099}
4100
d1310b2e 4101static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 4102 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 4103 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 4104 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 4105 /* note we're sharing with inode.c for the merge bio hook */
4106 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 4107};
Linux 6.x block layer and io_uring tree(s)