projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: split up __extent_writepage to lower stack usage
[linux-2.6-block.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
5a0e3ad6 29#include <linux/slab.h>
784b4e29 30#include <linux/migrate.h>
7a36ddec 31#include <linux/ratelimit.h>
6463fe58 32#include <linux/uuid.h>
803b2f54 33#include <linux/semaphore.h>
7e75bf3f 34#include <asm/unaligned.h>
eb60ceac
CM		 35#include "ctree.h"
36#include "disk-io.h"
0b947aff 37#include "hash.h"
e089f05c 38#include "transaction.h"
0f7d52f4 39#include "btrfs_inode.h"
0b86a832 40#include "volumes.h"
db94535d 41#include "print-tree.h"
8b712842 42#include "async-thread.h"
925baedd 43#include "locking.h"
e02119d5 44#include "tree-log.h"
fa9c0d79 45#include "free-space-cache.h"
581bb050 46#include "inode-map.h"
21adbd5c 47#include "check-integrity.h"
606686ee 48#include "rcu-string.h"
8dabb742 49#include "dev-replace.h"
53b381b3 50#include "raid56.h"
5ac1d209 51#include "sysfs.h"
fcebe456 52#include "qgroup.h"
eb60ceac 53
de0022b9
JB		 54#ifdef CONFIG_X86
55#include <asm/cpufeature.h>
56#endif
57
d1310b2e 58static struct extent_io_ops btree_extent_io_ops;
8b712842 59static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 60static void free_fs_root(struct btrfs_root *root);
fcd1f065 61static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 62 int read_only);
569e0f35
JB		 63static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
64 struct btrfs_root *root);
143bede5 65static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
acce952b 66static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
67 struct btrfs_root *root);
143bede5 68static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
acce952b 69static int btrfs_destroy_marked_extents(struct btrfs_root *root,
70 struct extent_io_tree *dirty_pages,
71 int mark);
72static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
73 struct extent_io_tree *pinned_extents);
48a3b636
ES		 74static int btrfs_cleanup_transaction(struct btrfs_root *root);
75static void btrfs_error_commit_super(struct btrfs_root *root);
ce9adaa5 76
d352ac68
CM		 77/*
78 * end_io_wq structs are used to do processing in task context when an IO is
79 * complete. This is used during reads to verify checksums, and it is used
80 * by writes to insert metadata for new file extents after IO is complete.
81 */
ce9adaa5
CM		 82struct end_io_wq {
83 struct bio *bio;
84 bio_end_io_t *end_io;
85 void *private;
86 struct btrfs_fs_info *info;
87 int error;
22c59948 88 int metadata;
ce9adaa5 89 struct list_head list;
8b712842 90 struct btrfs_work work;
ce9adaa5 91};
0da5468f 92
d352ac68
CM		 93/*
94 * async submit bios are used to offload expensive checksumming
95 * onto the worker threads. They checksum file and metadata bios
96 * just before they are sent down the IO stack.
97 */
44b8bd7e
CM		 98struct async_submit_bio {
99 struct inode *inode;
100 struct bio *bio;
101 struct list_head list;
4a69a410
CM		 102 extent_submit_bio_hook_t *submit_bio_start;
103 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 104 int rw;
105 int mirror_num;
c8b97818 106 unsigned long bio_flags;
eaf25d93
CM		 107 /*
108 * bio_offset is optional, can be used if the pages in the bio
109 * can't tell us where in the file the bio should go
110 */
111 u64 bio_offset;
8b712842 112 struct btrfs_work work;
79787eaa 113 int error;
44b8bd7e
CM		 114};
115
85d4e461
CM		 116/*
117 * Lockdep class keys for extent_buffer->lock's in this root. For a given
118 * eb, the lockdep key is determined by the btrfs_root it belongs to and
119 * the level the eb occupies in the tree.
120 *
121 * Different roots are used for different purposes and may nest inside each
122 * other and they require separate keysets. As lockdep keys should be
123 * static, assign keysets according to the purpose of the root as indicated
124 * by btrfs_root->objectid. This ensures that all special purpose roots
125 * have separate keysets.
4008c04a 126 *
85d4e461
CM		 127 * Lock-nesting across peer nodes is always done with the immediate parent
128 * node locked thus preventing deadlock. As lockdep doesn't know this, use
129 * subclass to avoid triggering lockdep warning in such cases.
4008c04a 130 *
85d4e461
CM		 131 * The key is set by the readpage_end_io_hook after the buffer has passed
132 * csum validation but before the pages are unlocked. It is also set by
133 * btrfs_init_new_buffer on freshly allocated blocks.
4008c04a 134 *
85d4e461
CM		 135 * We also add a check to make sure the highest level of the tree is the
136 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
137 * needs update as well.
4008c04a
CM		 138 */
139#ifdef CONFIG_DEBUG_LOCK_ALLOC
140# if BTRFS_MAX_LEVEL != 8
141# error
142# endif
85d4e461
CM		 143
144static struct btrfs_lockdep_keyset {
145 u64 id; /* root objectid */
146 const char *name_stem; /* lock name stem */
147 char names[BTRFS_MAX_LEVEL + 1][20];
148 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
149} btrfs_lockdep_keysets[] = {
150 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
151 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
152 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
153 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
154 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
155 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
60b62978 156 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
85d4e461
CM		 157 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
158 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
159 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
13fd8da9 160 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
85d4e461 161 { .id = 0, .name_stem = "tree" },
4008c04a 162};
85d4e461
CM		 163
164void __init btrfs_init_lockdep(void)
165{
166 int i, j;
167
168 /* initialize lockdep class names */
169 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
170 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
171
172 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
173 snprintf(ks->names[j], sizeof(ks->names[j]),
174 "btrfs-%s-%02d", ks->name_stem, j);
175 }
176}
177
178void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
179 int level)
180{
181 struct btrfs_lockdep_keyset *ks;
182
183 BUG_ON(level >= ARRAY_SIZE(ks->keys));
184
185 /* find the matching keyset, id 0 is the default entry */
186 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
187 if (ks->id == objectid)
188 break;
189
190 lockdep_set_class_and_name(&eb->lock,
191 &ks->keys[level], ks->names[level]);
192}
193
4008c04a
CM		 194#endif
195
d352ac68
CM		 196/*
197 * extents on the btree inode are pretty simple, there's one extent
198 * that covers the entire device
199 */
b2950863 200static struct extent_map *btree_get_extent(struct inode *inode,
306e16ce 201 struct page *page, size_t pg_offset, u64 start, u64 len,
b2950863 202 int create)
7eccb903 203{
5f39d397
CM		 204 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
205 struct extent_map *em;
206 int ret;
207
890871be 208 read_lock(&em_tree->lock);
d1310b2e 209 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 210 if (em) {
211 em->bdev =
212 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 213 read_unlock(&em_tree->lock);
5f39d397 214 goto out;
a061fc8d 215 }
890871be 216 read_unlock(&em_tree->lock);
7b13b7b1 217
172ddd60 218 em = alloc_extent_map();
5f39d397
CM		 219 if (!em) {
220 em = ERR_PTR(-ENOMEM);
221 goto out;
222 }
223 em->start = 0;
0afbaf8c 224 em->len = (u64)-1;
c8b97818 225 em->block_len = (u64)-1;
5f39d397 226 em->block_start = 0;
a061fc8d 227 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 228
890871be 229 write_lock(&em_tree->lock);
09a2a8f9 230 ret = add_extent_mapping(em_tree, em, 0);
5f39d397
CM		 231 if (ret == -EEXIST) {
232 free_extent_map(em);
7b13b7b1 233 em = lookup_extent_mapping(em_tree, start, len);
b4f359ab 234 if (!em)
0433f20d 235 em = ERR_PTR(-EIO);
5f39d397 236 } else if (ret) {
7b13b7b1 237 free_extent_map(em);
0433f20d 238 em = ERR_PTR(ret);
5f39d397 239 }
890871be 240 write_unlock(&em_tree->lock);
7b13b7b1 241
5f39d397
CM		 242out:
243 return em;
7eccb903
CM		 244}
245
b0496686 246u32 btrfs_csum_data(char *data, u32 seed, size_t len)
19c00ddc 247{
0b947aff 248 return btrfs_crc32c(seed, data, len);
19c00ddc
CM		 249}
250
251void btrfs_csum_final(u32 crc, char *result)
252{
7e75bf3f 253 put_unaligned_le32(~crc, result);
19c00ddc
CM		 254}
255
d352ac68
CM		 256/*
257 * compute the csum for a btree block, and either verify it or write it
258 * into the csum field of the block.
259 */
19c00ddc
CM		 260static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
261 int verify)
262{
6c41761f 263 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
607d432d 264 char *result = NULL;
19c00ddc
CM		 265 unsigned long len;
266 unsigned long cur_len;
267 unsigned long offset = BTRFS_CSUM_SIZE;
19c00ddc
CM		 268 char *kaddr;
269 unsigned long map_start;
270 unsigned long map_len;
271 int err;
272 u32 crc = ~(u32)0;
607d432d 273 unsigned long inline_result;
19c00ddc
CM		 274
275 len = buf->len - offset;
d397712b 276 while (len > 0) {
19c00ddc 277 err = map_private_extent_buffer(buf, offset, 32,
a6591715 278 &kaddr, &map_start, &map_len);
d397712b 279 if (err)
19c00ddc 280 return 1;
19c00ddc 281 cur_len = min(len, map_len - (offset - map_start));
b0496686 282 crc = btrfs_csum_data(kaddr + offset - map_start,
19c00ddc
CM		 283 crc, cur_len);
284 len -= cur_len;
285 offset += cur_len;
19c00ddc 286 }
607d432d
JB		 287 if (csum_size > sizeof(inline_result)) {
288 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
289 if (!result)
290 return 1;
291 } else {
292 result = (char *)&inline_result;
293 }
294
19c00ddc
CM		 295 btrfs_csum_final(crc, result);
296
297 if (verify) {
607d432d 298 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 299 u32 val;
300 u32 found = 0;
607d432d 301 memcpy(&found, result, csum_size);
e4204ded 302
607d432d 303 read_extent_buffer(buf, &val, 0, csum_size);
efe120a0
FH		 304 printk_ratelimited(KERN_INFO
305 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
306 "level %d\n",
307 root->fs_info->sb->s_id, buf->start,
308 val, found, btrfs_header_level(buf));
607d432d
JB		 309 if (result != (char *)&inline_result)
310 kfree(result);
19c00ddc
CM		 311 return 1;
312 }
313 } else {
607d432d 314 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 315 }
607d432d
JB		 316 if (result != (char *)&inline_result)
317 kfree(result);
19c00ddc
CM		 318 return 0;
319}
320
d352ac68
CM		 321/*
322 * we can't consider a given block up to date unless the transid of the
323 * block matches the transid in the parent node's pointer. This is how we
324 * detect blocks that either didn't get written at all or got written
325 * in the wrong place.
326 */
1259ab75 327static int verify_parent_transid(struct extent_io_tree *io_tree,
b9fab919
CM		 328 struct extent_buffer *eb, u64 parent_transid,
329 int atomic)
1259ab75 330{
2ac55d41 331 struct extent_state *cached_state = NULL;
1259ab75 332 int ret;
a26e8c9f
JB		 333 bool need_lock = (current->journal_info ==
334 (void *)BTRFS_SEND_TRANS_STUB);
1259ab75
CM		 335
336 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
337 return 0;
338
b9fab919
CM		 339 if (atomic)
340 return -EAGAIN;
341
a26e8c9f
JB		 342 if (need_lock) {
343 btrfs_tree_read_lock(eb);
344 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
345 }
346
2ac55d41 347 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
d0082371 348 0, &cached_state);
0b32f4bb 349 if (extent_buffer_uptodate(eb) &&
1259ab75
CM		 350 btrfs_header_generation(eb) == parent_transid) {
351 ret = 0;
352 goto out;
353 }
7a36ddec 354 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
193f284d 355 "found %llu\n",
c1c9ff7c 356 eb->start, parent_transid, btrfs_header_generation(eb));
1259ab75 357 ret = 1;
a26e8c9f
JB		 358
359 /*
360 * Things reading via commit roots that don't have normal protection,
361 * like send, can have a really old block in cache that may point at a
362 * block that has been free'd and re-allocated. So don't clear uptodate
363 * if we find an eb that is under IO (dirty/writeback) because we could
364 * end up reading in the stale data and then writing it back out and
365 * making everybody very sad.
366 */
367 if (!extent_buffer_under_io(eb))
368 clear_extent_buffer_uptodate(eb);
33958dc6 369out:
2ac55d41
JB		 370 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
371 &cached_state, GFP_NOFS);
a26e8c9f 372 btrfs_tree_read_unlock_blocking(eb);
1259ab75 373 return ret;
1259ab75
CM		 374}
375
1104a885
DS		 376/*
377 * Return 0 if the superblock checksum type matches the checksum value of that
378 * algorithm. Pass the raw disk superblock data.
379 */
380static int btrfs_check_super_csum(char *raw_disk_sb)
381{
382 struct btrfs_super_block *disk_sb =
383 (struct btrfs_super_block *)raw_disk_sb;
384 u16 csum_type = btrfs_super_csum_type(disk_sb);
385 int ret = 0;
386
387 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
388 u32 crc = ~(u32)0;
389 const int csum_size = sizeof(crc);
390 char result[csum_size];
391
392 /*
393 * The super_block structure does not span the whole
394 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
395 * is filled with zeros and is included in the checkum.
396 */
397 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
398 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
399 btrfs_csum_final(crc, result);
400
401 if (memcmp(raw_disk_sb, result, csum_size))
402 ret = 1;
667e7d94
CM		 403
404 if (ret && btrfs_super_generation(disk_sb) < 10) {
efe120a0
FH		 405 printk(KERN_WARNING
406 "BTRFS: super block crcs don't match, older mkfs detected\n");
667e7d94
CM		 407 ret = 0;
408 }
1104a885
DS		 409 }
410
411 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
efe120a0 412 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
1104a885
DS		 413 csum_type);
414 ret = 1;
415 }
416
417 return ret;
418}
419
d352ac68
CM		 420/*
421 * helper to read a given tree block, doing retries as required when
422 * the checksums don't match and we have alternate mirrors to try.
423 */
f188591e
CM		 424static int btree_read_extent_buffer_pages(struct btrfs_root *root,
425 struct extent_buffer *eb,
ca7a79ad 426 u64 start, u64 parent_transid)
f188591e
CM		 427{
428 struct extent_io_tree *io_tree;
ea466794 429 int failed = 0;
f188591e
CM		 430 int ret;
431 int num_copies = 0;
432 int mirror_num = 0;
ea466794 433 int failed_mirror = 0;
f188591e 434
a826d6dc 435 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM		 436 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
437 while (1) {
bb82ab88
AJ		 438 ret = read_extent_buffer_pages(io_tree, eb, start,
439 WAIT_COMPLETE,
f188591e 440 btree_get_extent, mirror_num);
256dd1bb
SB		 441 if (!ret) {
442 if (!verify_parent_transid(io_tree, eb,
b9fab919 443 parent_transid, 0))
256dd1bb
SB		 444 break;
445 else
446 ret = -EIO;
447 }
d397712b 448
a826d6dc
JB		 449 /*
450 * This buffer's crc is fine, but its contents are corrupted, so
451 * there is no reason to read the other copies, they won't be
452 * any less wrong.
453 */
454 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
ea466794
JB		 455 break;
456
5d964051 457 num_copies = btrfs_num_copies(root->fs_info,
f188591e 458 eb->start, eb->len);
4235298e 459 if (num_copies == 1)
ea466794 460 break;
4235298e 461
5cf1ab56
JB		 462 if (!failed_mirror) {
463 failed = 1;
464 failed_mirror = eb->read_mirror;
465 }
466
f188591e 467 mirror_num++;
ea466794
JB		 468 if (mirror_num == failed_mirror)
469 mirror_num++;
470
4235298e 471 if (mirror_num > num_copies)
ea466794 472 break;
f188591e 473 }
ea466794 474
c0901581 475 if (failed && !ret && failed_mirror)
ea466794
JB		 476 repair_eb_io_failure(root, eb, failed_mirror);
477
478 return ret;
f188591e 479}
19c00ddc 480
d352ac68 481/*
d397712b
CM		 482 * checksum a dirty tree block before IO. This has extra checks to make sure
483 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 484 */
d397712b 485
b2950863 486static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 487{
4eee4fa4 488 u64 start = page_offset(page);
19c00ddc 489 u64 found_start;
19c00ddc 490 struct extent_buffer *eb;
f188591e 491
4f2de97a
JB		 492 eb = (struct extent_buffer *)page->private;
493 if (page != eb->pages[0])
494 return 0;
19c00ddc 495 found_start = btrfs_header_bytenr(eb);
fae7f21c 496 if (WARN_ON(found_start != start || !PageUptodate(page)))
4f2de97a 497 return 0;
19c00ddc 498 csum_tree_block(root, eb, 0);
19c00ddc
CM		 499 return 0;
500}
501
2b82032c
YZ		 502static int check_tree_block_fsid(struct btrfs_root *root,
503 struct extent_buffer *eb)
504{
505 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
506 u8 fsid[BTRFS_UUID_SIZE];
507 int ret = 1;
508
0a4e5586 509 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
2b82032c
YZ		 510 while (fs_devices) {
511 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
512 ret = 0;
513 break;
514 }
515 fs_devices = fs_devices->seed;
516 }
517 return ret;
518}
519
a826d6dc 520#define CORRUPT(reason, eb, root, slot) \
efe120a0
FH		 521 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
522 "root=%llu, slot=%d", reason, \
c1c9ff7c 523 btrfs_header_bytenr(eb), root->objectid, slot)
a826d6dc
JB		 524
525static noinline int check_leaf(struct btrfs_root *root,
526 struct extent_buffer *leaf)
527{
528 struct btrfs_key key;
529 struct btrfs_key leaf_key;
530 u32 nritems = btrfs_header_nritems(leaf);
531 int slot;
532
533 if (nritems == 0)
534 return 0;
535
536 /* Check the 0 item */
537 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
538 BTRFS_LEAF_DATA_SIZE(root)) {
539 CORRUPT("invalid item offset size pair", leaf, root, 0);
540 return -EIO;
541 }
542
543 /*
544 * Check to make sure each items keys are in the correct order and their
545 * offsets make sense. We only have to loop through nritems-1 because
546 * we check the current slot against the next slot, which verifies the
547 * next slot's offset+size makes sense and that the current's slot
548 * offset is correct.
549 */
550 for (slot = 0; slot < nritems - 1; slot++) {
551 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
552 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
553
554 /* Make sure the keys are in the right order */
555 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
556 CORRUPT("bad key order", leaf, root, slot);
557 return -EIO;
558 }
559
560 /*
561 * Make sure the offset and ends are right, remember that the
562 * item data starts at the end of the leaf and grows towards the
563 * front.
564 */
565 if (btrfs_item_offset_nr(leaf, slot) !=
566 btrfs_item_end_nr(leaf, slot + 1)) {
567 CORRUPT("slot offset bad", leaf, root, slot);
568 return -EIO;
569 }
570
571 /*
572 * Check to make sure that we don't point outside of the leaf,
573 * just incase all the items are consistent to eachother, but
574 * all point outside of the leaf.
575 */
576 if (btrfs_item_end_nr(leaf, slot) >
577 BTRFS_LEAF_DATA_SIZE(root)) {
578 CORRUPT("slot end outside of leaf", leaf, root, slot);
579 return -EIO;
580 }
581 }
582
583 return 0;
584}
585
facc8a22
MX		 586static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
587 u64 phy_offset, struct page *page,
588 u64 start, u64 end, int mirror)
ce9adaa5 589{
ce9adaa5
CM		 590 u64 found_start;
591 int found_level;
ce9adaa5
CM		 592 struct extent_buffer *eb;
593 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 594 int ret = 0;
727011e0 595 int reads_done;
ce9adaa5 596
ce9adaa5
CM		 597 if (!page->private)
598 goto out;
d397712b 599
4f2de97a 600 eb = (struct extent_buffer *)page->private;
d397712b 601
0b32f4bb
JB		 602 /* the pending IO might have been the only thing that kept this buffer
603 * in memory. Make sure we have a ref for all this other checks
604 */
605 extent_buffer_get(eb);
606
607 reads_done = atomic_dec_and_test(&eb->io_pages);
727011e0
CM		 608 if (!reads_done)
609 goto err;
f188591e 610
5cf1ab56 611 eb->read_mirror = mirror;
ea466794
JB		 612 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
613 ret = -EIO;
614 goto err;
615 }
616
ce9adaa5 617 found_start = btrfs_header_bytenr(eb);
727011e0 618 if (found_start != eb->start) {
efe120a0 619 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
193f284d 620 "%llu %llu\n",
c1c9ff7c 621 found_start, eb->start);
f188591e 622 ret = -EIO;
ce9adaa5
CM		 623 goto err;
624 }
2b82032c 625 if (check_tree_block_fsid(root, eb)) {
efe120a0 626 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
c1c9ff7c 627 eb->start);
1259ab75
CM		 628 ret = -EIO;
629 goto err;
630 }
ce9adaa5 631 found_level = btrfs_header_level(eb);
1c24c3ce 632 if (found_level >= BTRFS_MAX_LEVEL) {
efe120a0 633 btrfs_info(root->fs_info, "bad tree block level %d",
1c24c3ce
JB		 634 (int)btrfs_header_level(eb));
635 ret = -EIO;
636 goto err;
637 }
ce9adaa5 638
85d4e461
CM		 639 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
640 eb, found_level);
4008c04a 641
ce9adaa5 642 ret = csum_tree_block(root, eb, 1);
a826d6dc 643 if (ret) {
f188591e 644 ret = -EIO;
a826d6dc
JB		 645 goto err;
646 }
647
648 /*
649 * If this is a leaf block and it is corrupt, set the corrupt bit so
650 * that we don't try and read the other copies of this block, just
651 * return -EIO.
652 */
653 if (found_level == 0 && check_leaf(root, eb)) {
654 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
655 ret = -EIO;
656 }
ce9adaa5 657
0b32f4bb
JB		 658 if (!ret)
659 set_extent_buffer_uptodate(eb);
ce9adaa5 660err:
79fb65a1
JB		 661 if (reads_done &&
662 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 663 btree_readahead_hook(root, eb, eb->start, ret);
4bb31e92 664
53b381b3
DW		 665 if (ret) {
666 /*
667 * our io error hook is going to dec the io pages
668 * again, we have to make sure it has something
669 * to decrement
670 */
671 atomic_inc(&eb->io_pages);
0b32f4bb 672 clear_extent_buffer_uptodate(eb);
53b381b3 673 }
0b32f4bb 674 free_extent_buffer(eb);
ce9adaa5 675out:
f188591e 676 return ret;
ce9adaa5
CM		 677}
678
ea466794 679static int btree_io_failed_hook(struct page *page, int failed_mirror)
4bb31e92 680{
4bb31e92
AJ		 681 struct extent_buffer *eb;
682 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
683
4f2de97a 684 eb = (struct extent_buffer *)page->private;
ea466794 685 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
5cf1ab56 686 eb->read_mirror = failed_mirror;
53b381b3 687 atomic_dec(&eb->io_pages);
ea466794 688 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 689 btree_readahead_hook(root, eb, eb->start, -EIO);
4bb31e92
AJ		 690 return -EIO; /* we fixed nothing */
691}
692
ce9adaa5 693static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 694{
695 struct end_io_wq *end_io_wq = bio->bi_private;
696 struct btrfs_fs_info *fs_info;
ce9adaa5 697
ce9adaa5 698 fs_info = end_io_wq->info;
ce9adaa5 699 end_io_wq->error = err;
fccb5d86 700 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
d20f7043 701
7b6d91da 702 if (bio->bi_rw & REQ_WRITE) {
53b381b3 703 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
fccb5d86
QW		 704 btrfs_queue_work(fs_info->endio_meta_write_workers,
705 &end_io_wq->work);
53b381b3 706 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
fccb5d86
QW		 707 btrfs_queue_work(fs_info->endio_freespace_worker,
708 &end_io_wq->work);
53b381b3 709 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
fccb5d86
QW		 710 btrfs_queue_work(fs_info->endio_raid56_workers,
711 &end_io_wq->work);
cad321ad 712 else
fccb5d86
QW		 713 btrfs_queue_work(fs_info->endio_write_workers,
714 &end_io_wq->work);
d20f7043 715 } else {
53b381b3 716 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
fccb5d86
QW		 717 btrfs_queue_work(fs_info->endio_raid56_workers,
718 &end_io_wq->work);
53b381b3 719 else if (end_io_wq->metadata)
fccb5d86
QW		 720 btrfs_queue_work(fs_info->endio_meta_workers,
721 &end_io_wq->work);
d20f7043 722 else
fccb5d86
QW		 723 btrfs_queue_work(fs_info->endio_workers,
724 &end_io_wq->work);
d20f7043 725 }
ce9adaa5
CM		 726}
727
0cb59c99
JB		 728/*
729 * For the metadata arg you want
730 *
731 * 0 - if data
732 * 1 - if normal metadta
733 * 2 - if writing to the free space cache area
53b381b3 734 * 3 - raid parity work
0cb59c99 735 */
22c59948
CM		 736int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
737 int metadata)
0b86a832 738{
ce9adaa5 739 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 740 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
741 if (!end_io_wq)
742 return -ENOMEM;
743
744 end_io_wq->private = bio->bi_private;
745 end_io_wq->end_io = bio->bi_end_io;
22c59948 746 end_io_wq->info = info;
ce9adaa5
CM		 747 end_io_wq->error = 0;
748 end_io_wq->bio = bio;
22c59948 749 end_io_wq->metadata = metadata;
ce9adaa5
CM		 750
751 bio->bi_private = end_io_wq;
752 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 753 return 0;
754}
755
b64a2851 756unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 757{
4854ddd0 758 unsigned long limit = min_t(unsigned long,
5cdc7ad3 759 info->thread_pool_size,
4854ddd0
CM		 760 info->fs_devices->open_devices);
761 return 256 * limit;
762}
0986fe9e 763
4a69a410
CM		 764static void run_one_async_start(struct btrfs_work *work)
765{
4a69a410 766 struct async_submit_bio *async;
79787eaa 767 int ret;
4a69a410
CM		 768
769 async = container_of(work, struct async_submit_bio, work);
79787eaa
JM		 770 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
771 async->mirror_num, async->bio_flags,
772 async->bio_offset);
773 if (ret)
774 async->error = ret;
4a69a410
CM		 775}
776
777static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 778{
779 struct btrfs_fs_info *fs_info;
780 struct async_submit_bio *async;
4854ddd0 781 int limit;
8b712842
CM		 782
783 async = container_of(work, struct async_submit_bio, work);
784 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 785
b64a2851 786 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 787 limit = limit * 2 / 3;
788
66657b31 789 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
b64a2851 790 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 791 wake_up(&fs_info->async_submit_wait);
792
79787eaa
JM		 793 /* If an error occured we just want to clean up the bio and move on */
794 if (async->error) {
795 bio_endio(async->bio, async->error);
796 return;
797 }
798
4a69a410 799 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 800 async->mirror_num, async->bio_flags,
801 async->bio_offset);
4a69a410
CM		 802}
803
804static void run_one_async_free(struct btrfs_work *work)
805{
806 struct async_submit_bio *async;
807
808 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 809 kfree(async);
810}
811
44b8bd7e
CM		 812int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
813 int rw, struct bio *bio, int mirror_num,
c8b97818 814 unsigned long bio_flags,
eaf25d93 815 u64 bio_offset,
4a69a410
CM		 816 extent_submit_bio_hook_t *submit_bio_start,
817 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 818{
819 struct async_submit_bio *async;
820
821 async = kmalloc(sizeof(*async), GFP_NOFS);
822 if (!async)
823 return -ENOMEM;
824
825 async->inode = inode;
826 async->rw = rw;
827 async->bio = bio;
828 async->mirror_num = mirror_num;
4a69a410
CM		 829 async->submit_bio_start = submit_bio_start;
830 async->submit_bio_done = submit_bio_done;
831
5cdc7ad3
QW		 832 btrfs_init_work(&async->work, run_one_async_start,
833 run_one_async_done, run_one_async_free);
4a69a410 834
c8b97818 835 async->bio_flags = bio_flags;
eaf25d93 836 async->bio_offset = bio_offset;
8c8bee1d 837
79787eaa
JM		 838 async->error = 0;
839
cb03c743 840 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 841
7b6d91da 842 if (rw & REQ_SYNC)
5cdc7ad3 843 btrfs_set_work_high_priority(&async->work);
d313d7a3 844
5cdc7ad3 845 btrfs_queue_work(fs_info->workers, &async->work);
9473f16c 846
d397712b 847 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 848 atomic_read(&fs_info->nr_async_submits)) {
849 wait_event(fs_info->async_submit_wait,
850 (atomic_read(&fs_info->nr_async_submits) == 0));
851 }
852
44b8bd7e
CM		 853 return 0;
854}
855
ce3ed71a
CM		 856static int btree_csum_one_bio(struct bio *bio)
857{
2c30c71b 858 struct bio_vec *bvec;
ce3ed71a 859 struct btrfs_root *root;
2c30c71b 860 int i, ret = 0;
ce3ed71a 861
2c30c71b 862 bio_for_each_segment_all(bvec, bio, i) {
ce3ed71a 863 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 864 ret = csum_dirty_buffer(root, bvec->bv_page);
865 if (ret)
866 break;
ce3ed71a 867 }
2c30c71b 868
79787eaa 869 return ret;
ce3ed71a
CM		 870}
871
4a69a410
CM		 872static int __btree_submit_bio_start(struct inode *inode, int rw,
873 struct bio *bio, int mirror_num,
eaf25d93
CM		 874 unsigned long bio_flags,
875 u64 bio_offset)
22c59948 876{
8b712842
CM		 877 /*
878 * when we're called for a write, we're already in the async
5443be45 879 * submission context. Just jump into btrfs_map_bio
8b712842 880 */
79787eaa 881 return btree_csum_one_bio(bio);
4a69a410 882}
22c59948 883
4a69a410 884static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 885 int mirror_num, unsigned long bio_flags,
886 u64 bio_offset)
4a69a410 887{
61891923
SB		 888 int ret;
889
8b712842 890 /*
4a69a410
CM		 891 * when we're called for a write, we're already in the async
892 * submission context. Just jump into btrfs_map_bio
8b712842 893 */
61891923
SB		 894 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
895 if (ret)
896 bio_endio(bio, ret);
897 return ret;
0b86a832
CM		 898}
899
de0022b9
JB		 900static int check_async_write(struct inode *inode, unsigned long bio_flags)
901{
902 if (bio_flags & EXTENT_BIO_TREE_LOG)
903 return 0;
904#ifdef CONFIG_X86
905 if (cpu_has_xmm4_2)
906 return 0;
907#endif
908 return 1;
909}
910
44b8bd7e 911static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 912 int mirror_num, unsigned long bio_flags,
913 u64 bio_offset)
44b8bd7e 914{
de0022b9 915 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 916 int ret;
917
7b6d91da 918 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 919 /*
920 * called for a read, do the setup so that checksum validation
921 * can happen in the async kernel threads
922 */
f3f266ab
CM		 923 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
924 bio, 1);
1d4284bd 925 if (ret)
61891923
SB		 926 goto out_w_error;
927 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
928 mirror_num, 0);
de0022b9
JB		 929 } else if (!async) {
930 ret = btree_csum_one_bio(bio);
931 if (ret)
61891923
SB		 932 goto out_w_error;
933 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
934 mirror_num, 0);
935 } else {
936 /*
937 * kthread helpers are used to submit writes so that
938 * checksumming can happen in parallel across all CPUs
939 */
940 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
941 inode, rw, bio, mirror_num, 0,
942 bio_offset,
943 __btree_submit_bio_start,
944 __btree_submit_bio_done);
44b8bd7e 945 }
d313d7a3 946
61891923
SB		 947 if (ret) {
948out_w_error:
949 bio_endio(bio, ret);
950 }
951 return ret;
44b8bd7e
CM		 952}
953
3dd1462e 954#ifdef CONFIG_MIGRATION
784b4e29 955static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 956 struct page *newpage, struct page *page,
957 enum migrate_mode mode)
784b4e29
CM		 958{
959 /*
960 * we can't safely write a btree page from here,
961 * we haven't done the locking hook
962 */
963 if (PageDirty(page))
964 return -EAGAIN;
965 /*
966 * Buffers may be managed in a filesystem specific way.
967 * We must have no buffers or drop them.
968 */
969 if (page_has_private(page) &&
970 !try_to_release_page(page, GFP_KERNEL))
971 return -EAGAIN;
a6bc32b8 972 return migrate_page(mapping, newpage, page, mode);
784b4e29 973}
3dd1462e 974#endif
784b4e29 975
0da5468f
CM		 976
977static int btree_writepages(struct address_space *mapping,
978 struct writeback_control *wbc)
979{
e2d84521
MX		 980 struct btrfs_fs_info *fs_info;
981 int ret;
982
d8d5f3e1 983 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 984
985 if (wbc->for_kupdate)
986 return 0;
987
e2d84521 988 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 989 /* this is a bit racy, but that's ok */
e2d84521
MX		 990 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
991 BTRFS_DIRTY_METADATA_THRESH);
992 if (ret < 0)
793955bc 993 return 0;
793955bc 994 }
0b32f4bb 995 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 996}
997
b2950863 998static int btree_readpage(struct file *file, struct page *page)
5f39d397 999{
d1310b2e
CM		 1000 struct extent_io_tree *tree;
1001 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 1002 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 1003}
22b0ebda 1004
70dec807 1005static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 1006{
98509cfc 1007 if (PageWriteback(page) || PageDirty(page))
d397712b 1008 return 0;
0c4e538b 1009
f7a52a40 1010 return try_release_extent_buffer(page);
d98237b3
CM		 1011}
1012
d47992f8
LC		 1013static void btree_invalidatepage(struct page *page, unsigned int offset,
1014 unsigned int length)
d98237b3 1015{
d1310b2e
CM		 1016 struct extent_io_tree *tree;
1017 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1018 extent_invalidatepage(tree, page, offset);
1019 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1020 if (PagePrivate(page)) {
efe120a0
FH		 1021 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1022 "page private not zero on page %llu",
1023 (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1024 ClearPagePrivate(page);
1025 set_page_private(page, 0);
1026 page_cache_release(page);
1027 }
d98237b3
CM		 1028}
1029
0b32f4bb
JB		 1030static int btree_set_page_dirty(struct page *page)
1031{
bb146eb2 1032#ifdef DEBUG
0b32f4bb
JB		 1033 struct extent_buffer *eb;
1034
1035 BUG_ON(!PagePrivate(page));
1036 eb = (struct extent_buffer *)page->private;
1037 BUG_ON(!eb);
1038 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1039 BUG_ON(!atomic_read(&eb->refs));
1040 btrfs_assert_tree_locked(eb);
bb146eb2 1041#endif
0b32f4bb
JB		 1042 return __set_page_dirty_nobuffers(page);
1043}
1044
7f09410b 1045static const struct address_space_operations btree_aops = {
d98237b3 1046 .readpage = btree_readpage,
0da5468f 1047 .writepages = btree_writepages,
5f39d397
CM		 1048 .releasepage = btree_releasepage,
1049 .invalidatepage = btree_invalidatepage,
5a92bc88 1050#ifdef CONFIG_MIGRATION
784b4e29 1051 .migratepage = btree_migratepage,
5a92bc88 1052#endif
0b32f4bb 1053 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1054};
1055
ca7a79ad
CM		 1056int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1057 u64 parent_transid)
090d1875 1058{
5f39d397
CM		 1059 struct extent_buffer *buf = NULL;
1060 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1061 int ret = 0;
090d1875 1062
db94535d 1063 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1064 if (!buf)
090d1875 1065 return 0;
d1310b2e 1066 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1067 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1068 free_extent_buffer(buf);
de428b63 1069 return ret;
090d1875
CM		 1070}
1071
ab0fff03
AJ		 1072int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1073 int mirror_num, struct extent_buffer **eb)
1074{
1075 struct extent_buffer *buf = NULL;
1076 struct inode *btree_inode = root->fs_info->btree_inode;
1077 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1078 int ret;
1079
1080 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1081 if (!buf)
1082 return 0;
1083
1084 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1085
1086 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1087 btree_get_extent, mirror_num);
1088 if (ret) {
1089 free_extent_buffer(buf);
1090 return ret;
1091 }
1092
1093 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1094 free_extent_buffer(buf);
1095 return -EIO;
0b32f4bb 1096 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1097 *eb = buf;
1098 } else {
1099 free_extent_buffer(buf);
1100 }
1101 return 0;
1102}
1103
0999df54
CM		 1104struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1105 u64 bytenr, u32 blocksize)
1106{
f28491e0 1107 return find_extent_buffer(root->fs_info, bytenr);
0999df54
CM		 1108}
1109
1110struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1111 u64 bytenr, u32 blocksize)
1112{
faa2dbf0
JB		 1113#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1114 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state)))
1115 return alloc_test_extent_buffer(root->fs_info, bytenr,
1116 blocksize);
1117#endif
f28491e0 1118 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
0999df54
CM		 1119}
1120
1121
e02119d5
CM		 1122int btrfs_write_tree_block(struct extent_buffer *buf)
1123{
727011e0 1124 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
8aa38c31 1125 buf->start + buf->len - 1);
e02119d5
CM		 1126}
1127
1128int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1129{
727011e0 1130 return filemap_fdatawait_range(buf->pages[0]->mapping,
8aa38c31 1131 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1132}
1133
0999df54 1134struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1135 u32 blocksize, u64 parent_transid)
0999df54
CM		 1136{
1137 struct extent_buffer *buf = NULL;
0999df54
CM		 1138 int ret;
1139
0999df54
CM		 1140 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1141 if (!buf)
1142 return NULL;
0999df54 1143
ca7a79ad 1144 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
0f0fe8f7
FDBM		 1145 if (ret) {
1146 free_extent_buffer(buf);
1147 return NULL;
1148 }
5f39d397 1149 return buf;
ce9adaa5 1150
eb60ceac
CM		 1151}
1152
d5c13f92
JM		 1153void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1154 struct extent_buffer *buf)
ed2ff2cb 1155{
e2d84521
MX		 1156 struct btrfs_fs_info *fs_info = root->fs_info;
1157
55c69072 1158 if (btrfs_header_generation(buf) ==
e2d84521 1159 fs_info->running_transaction->transid) {
b9447ef8 1160 btrfs_assert_tree_locked(buf);
b4ce94de 1161
b9473439 1162 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1163 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1164 -buf->len,
1165 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1166 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1167 btrfs_set_lock_blocking(buf);
1168 clear_extent_buffer_dirty(buf);
1169 }
925baedd 1170 }
5f39d397
CM		 1171}
1172
8257b2dc
MX		 1173static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1174{
1175 struct btrfs_subvolume_writers *writers;
1176 int ret;
1177
1178 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1179 if (!writers)
1180 return ERR_PTR(-ENOMEM);
1181
1182 ret = percpu_counter_init(&writers->counter, 0);
1183 if (ret < 0) {
1184 kfree(writers);
1185 return ERR_PTR(ret);
1186 }
1187
1188 init_waitqueue_head(&writers->wait);
1189 return writers;
1190}
1191
1192static void
1193btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1194{
1195 percpu_counter_destroy(&writers->counter);
1196 kfree(writers);
1197}
1198
143bede5
JM		 1199static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1200 u32 stripesize, struct btrfs_root *root,
1201 struct btrfs_fs_info *fs_info,
1202 u64 objectid)
d97e63b6 1203{
cfaa7295 1204 root->node = NULL;
a28ec197 1205 root->commit_root = NULL;
db94535d
CM		 1206 root->sectorsize = sectorsize;
1207 root->nodesize = nodesize;
1208 root->leafsize = leafsize;
87ee04eb 1209 root->stripesize = stripesize;
27cdeb70 1210 root->state = 0;
d68fc57b 1211 root->orphan_cleanup_state = 0;
0b86a832 1212
0f7d52f4
CM		 1213 root->objectid = objectid;
1214 root->last_trans = 0;
13a8a7c8 1215 root->highest_objectid = 0;
eb73c1b7 1216 root->nr_delalloc_inodes = 0;
199c2a9c 1217 root->nr_ordered_extents = 0;
58176a96 1218 root->name = NULL;
6bef4d31 1219 root->inode_tree = RB_ROOT;
16cdcec7 1220 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1221 root->block_rsv = NULL;
d68fc57b 1222 root->orphan_block_rsv = NULL;
0b86a832
CM		 1223
1224 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1225 INIT_LIST_HEAD(&root->root_list);
eb73c1b7
MX		 1226 INIT_LIST_HEAD(&root->delalloc_inodes);
1227 INIT_LIST_HEAD(&root->delalloc_root);
199c2a9c
MX		 1228 INIT_LIST_HEAD(&root->ordered_extents);
1229 INIT_LIST_HEAD(&root->ordered_root);
2ab28f32
JB		 1230 INIT_LIST_HEAD(&root->logged_list[0]);
1231 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1232 spin_lock_init(&root->orphan_lock);
5d4f98a2 1233 spin_lock_init(&root->inode_lock);
eb73c1b7 1234 spin_lock_init(&root->delalloc_lock);
199c2a9c 1235 spin_lock_init(&root->ordered_extent_lock);
f0486c68 1236 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1237 spin_lock_init(&root->log_extents_lock[0]);
1238 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1239 mutex_init(&root->objectid_mutex);
e02119d5 1240 mutex_init(&root->log_mutex);
31f3d255 1241 mutex_init(&root->ordered_extent_mutex);
573bfb72 1242 mutex_init(&root->delalloc_mutex);
7237f183
YZ		 1243 init_waitqueue_head(&root->log_writer_wait);
1244 init_waitqueue_head(&root->log_commit_wait[0]);
1245 init_waitqueue_head(&root->log_commit_wait[1]);
8b050d35
MX		 1246 INIT_LIST_HEAD(&root->log_ctxs[0]);
1247 INIT_LIST_HEAD(&root->log_ctxs[1]);
7237f183
YZ		 1248 atomic_set(&root->log_commit[0], 0);
1249 atomic_set(&root->log_commit[1], 0);
1250 atomic_set(&root->log_writers, 0);
2ecb7923 1251 atomic_set(&root->log_batch, 0);
8a35d95f 1252 atomic_set(&root->orphan_inodes, 0);
b0feb9d9 1253 atomic_set(&root->refs, 1);
8257b2dc 1254 atomic_set(&root->will_be_snapshoted, 0);
7237f183 1255 root->log_transid = 0;
d1433deb 1256 root->log_transid_committed = -1;
257c62e1 1257 root->last_log_commit = 0;
06ea65a3
JB		 1258 if (fs_info)
1259 extent_io_tree_init(&root->dirty_log_pages,
1260 fs_info->btree_inode->i_mapping);
017e5369 1261
3768f368
CM		 1262 memset(&root->root_key, 0, sizeof(root->root_key));
1263 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1264 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1265 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
06ea65a3
JB		 1266 if (fs_info)
1267 root->defrag_trans_start = fs_info->generation;
1268 else
1269 root->defrag_trans_start = 0;
58176a96 1270 init_completion(&root->kobj_unregister);
4d775673 1271 root->root_key.objectid = objectid;
0ee5dc67 1272 root->anon_dev = 0;
8ea05e3a 1273
5f3ab90a 1274 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1275}
1276
f84a8bd6 1277static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1278{
1279 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1280 if (root)
1281 root->fs_info = fs_info;
1282 return root;
1283}
1284
06ea65a3
JB		 1285#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1286/* Should only be used by the testing infrastructure */
1287struct btrfs_root *btrfs_alloc_dummy_root(void)
1288{
1289 struct btrfs_root *root;
1290
1291 root = btrfs_alloc_root(NULL);
1292 if (!root)
1293 return ERR_PTR(-ENOMEM);
1294 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
27cdeb70 1295 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
faa2dbf0 1296 root->alloc_bytenr = 0;
06ea65a3
JB		 1297
1298 return root;
1299}
1300#endif
1301
20897f5c
AJ		 1302struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1303 struct btrfs_fs_info *fs_info,
1304 u64 objectid)
1305{
1306 struct extent_buffer *leaf;
1307 struct btrfs_root *tree_root = fs_info->tree_root;
1308 struct btrfs_root *root;
1309 struct btrfs_key key;
1310 int ret = 0;
6463fe58 1311 uuid_le uuid;
20897f5c
AJ		 1312
1313 root = btrfs_alloc_root(fs_info);
1314 if (!root)
1315 return ERR_PTR(-ENOMEM);
1316
1317 __setup_root(tree_root->nodesize, tree_root->leafsize,
1318 tree_root->sectorsize, tree_root->stripesize,
1319 root, fs_info, objectid);
1320 root->root_key.objectid = objectid;
1321 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1322 root->root_key.offset = 0;
1323
1324 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1325 0, objectid, NULL, 0, 0, 0);
1326 if (IS_ERR(leaf)) {
1327 ret = PTR_ERR(leaf);
1dd05682 1328 leaf = NULL;
20897f5c
AJ		 1329 goto fail;
1330 }
1331
20897f5c
AJ		 1332 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1333 btrfs_set_header_bytenr(leaf, leaf->start);
1334 btrfs_set_header_generation(leaf, trans->transid);
1335 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1336 btrfs_set_header_owner(leaf, objectid);
1337 root->node = leaf;
1338
0a4e5586 1339 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
20897f5c
AJ		 1340 BTRFS_FSID_SIZE);
1341 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
b308bc2f 1342 btrfs_header_chunk_tree_uuid(leaf),
20897f5c
AJ		 1343 BTRFS_UUID_SIZE);
1344 btrfs_mark_buffer_dirty(leaf);
1345
1346 root->commit_root = btrfs_root_node(root);
27cdeb70 1347 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
20897f5c
AJ		 1348
1349 root->root_item.flags = 0;
1350 root->root_item.byte_limit = 0;
1351 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1352 btrfs_set_root_generation(&root->root_item, trans->transid);
1353 btrfs_set_root_level(&root->root_item, 0);
1354 btrfs_set_root_refs(&root->root_item, 1);
1355 btrfs_set_root_used(&root->root_item, leaf->len);
1356 btrfs_set_root_last_snapshot(&root->root_item, 0);
1357 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1358 uuid_le_gen(&uuid);
1359 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1360 root->root_item.drop_level = 0;
1361
1362 key.objectid = objectid;
1363 key.type = BTRFS_ROOT_ITEM_KEY;
1364 key.offset = 0;
1365 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1366 if (ret)
1367 goto fail;
1368
1369 btrfs_tree_unlock(leaf);
1370
1dd05682
TI		 1371 return root;
1372
20897f5c 1373fail:
1dd05682
TI		 1374 if (leaf) {
1375 btrfs_tree_unlock(leaf);
59885b39 1376 free_extent_buffer(root->commit_root);
1dd05682
TI		 1377 free_extent_buffer(leaf);
1378 }
1379 kfree(root);
20897f5c 1380
1dd05682 1381 return ERR_PTR(ret);
20897f5c
AJ		 1382}
1383
7237f183
YZ		 1384static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1385 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1386{
1387 struct btrfs_root *root;
1388 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1389 struct extent_buffer *leaf;
e02119d5 1390
6f07e42e 1391 root = btrfs_alloc_root(fs_info);
e02119d5 1392 if (!root)
7237f183 1393 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1394
1395 __setup_root(tree_root->nodesize, tree_root->leafsize,
1396 tree_root->sectorsize, tree_root->stripesize,
1397 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1398
1399 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1400 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1401 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
27cdeb70 1402
7237f183 1403 /*
27cdeb70
MX		 1404 * DON'T set REF_COWS for log trees
1405 *
7237f183
YZ		 1406 * log trees do not get reference counted because they go away
1407 * before a real commit is actually done. They do store pointers
1408 * to file data extents, and those reference counts still get
1409 * updated (along with back refs to the log tree).
1410 */
e02119d5 1411
5d4f98a2 1412 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1413 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1414 0, 0, 0);
7237f183
YZ		 1415 if (IS_ERR(leaf)) {
1416 kfree(root);
1417 return ERR_CAST(leaf);
1418 }
e02119d5 1419
5d4f98a2
YZ		 1420 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1421 btrfs_set_header_bytenr(leaf, leaf->start);
1422 btrfs_set_header_generation(leaf, trans->transid);
1423 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1424 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1425 root->node = leaf;
e02119d5
CM		 1426
1427 write_extent_buffer(root->node, root->fs_info->fsid,
0a4e5586 1428 btrfs_header_fsid(), BTRFS_FSID_SIZE);
e02119d5
CM		 1429 btrfs_mark_buffer_dirty(root->node);
1430 btrfs_tree_unlock(root->node);
7237f183
YZ		 1431 return root;
1432}
1433
1434int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1435 struct btrfs_fs_info *fs_info)
1436{
1437 struct btrfs_root *log_root;
1438
1439 log_root = alloc_log_tree(trans, fs_info);
1440 if (IS_ERR(log_root))
1441 return PTR_ERR(log_root);
1442 WARN_ON(fs_info->log_root_tree);
1443 fs_info->log_root_tree = log_root;
1444 return 0;
1445}
1446
1447int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1448 struct btrfs_root *root)
1449{
1450 struct btrfs_root *log_root;
1451 struct btrfs_inode_item *inode_item;
1452
1453 log_root = alloc_log_tree(trans, root->fs_info);
1454 if (IS_ERR(log_root))
1455 return PTR_ERR(log_root);
1456
1457 log_root->last_trans = trans->transid;
1458 log_root->root_key.offset = root->root_key.objectid;
1459
1460 inode_item = &log_root->root_item.inode;
3cae210f
QW		 1461 btrfs_set_stack_inode_generation(inode_item, 1);
1462 btrfs_set_stack_inode_size(inode_item, 3);
1463 btrfs_set_stack_inode_nlink(inode_item, 1);
1464 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1465 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
7237f183 1466
5d4f98a2 1467 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1468
1469 WARN_ON(root->log_root);
1470 root->log_root = log_root;
1471 root->log_transid = 0;
d1433deb 1472 root->log_transid_committed = -1;
257c62e1 1473 root->last_log_commit = 0;
e02119d5
CM		 1474 return 0;
1475}
1476
35a3621b
SB		 1477static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1478 struct btrfs_key *key)
e02119d5
CM		 1479{
1480 struct btrfs_root *root;
1481 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1482 struct btrfs_path *path;
84234f3a 1483 u64 generation;
db94535d 1484 u32 blocksize;
cb517eab 1485 int ret;
0f7d52f4 1486
cb517eab
MX		 1487 path = btrfs_alloc_path();
1488 if (!path)
0f7d52f4 1489 return ERR_PTR(-ENOMEM);
cb517eab
MX		 1490
1491 root = btrfs_alloc_root(fs_info);
1492 if (!root) {
1493 ret = -ENOMEM;
1494 goto alloc_fail;
0f7d52f4
CM		 1495 }
1496
db94535d 1497 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb 1498 tree_root->sectorsize, tree_root->stripesize,
cb517eab 1499 root, fs_info, key->objectid);
0f7d52f4 1500
cb517eab
MX		 1501 ret = btrfs_find_root(tree_root, key, path,
1502 &root->root_item, &root->root_key);
0f7d52f4 1503 if (ret) {
13a8a7c8
YZ		 1504 if (ret > 0)
1505 ret = -ENOENT;
cb517eab 1506 goto find_fail;
0f7d52f4 1507 }
13a8a7c8 1508
84234f3a 1509 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1510 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1511 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1512 blocksize, generation);
cb517eab
MX		 1513 if (!root->node) {
1514 ret = -ENOMEM;
1515 goto find_fail;
1516 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1517 ret = -EIO;
1518 goto read_fail;
416bc658 1519 }
5d4f98a2 1520 root->commit_root = btrfs_root_node(root);
13a8a7c8 1521out:
cb517eab
MX		 1522 btrfs_free_path(path);
1523 return root;
1524
1525read_fail:
1526 free_extent_buffer(root->node);
1527find_fail:
1528 kfree(root);
1529alloc_fail:
1530 root = ERR_PTR(ret);
1531 goto out;
1532}
1533
1534struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1535 struct btrfs_key *location)
1536{
1537 struct btrfs_root *root;
1538
1539 root = btrfs_read_tree_root(tree_root, location);
1540 if (IS_ERR(root))
1541 return root;
1542
1543 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
27cdeb70 1544 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
08fe4db1
LZ		 1545 btrfs_check_and_init_root_item(&root->root_item);
1546 }
13a8a7c8 1547
5eda7b5e
CM		 1548 return root;
1549}
1550
cb517eab
MX		 1551int btrfs_init_fs_root(struct btrfs_root *root)
1552{
1553 int ret;
8257b2dc 1554 struct btrfs_subvolume_writers *writers;
cb517eab
MX		 1555
1556 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1557 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1558 GFP_NOFS);
1559 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1560 ret = -ENOMEM;
1561 goto fail;
1562 }
1563
8257b2dc
MX		 1564 writers = btrfs_alloc_subvolume_writers();
1565 if (IS_ERR(writers)) {
1566 ret = PTR_ERR(writers);
1567 goto fail;
1568 }
1569 root->subv_writers = writers;
1570
cb517eab 1571 btrfs_init_free_ino_ctl(root);
cb517eab
MX		 1572 spin_lock_init(&root->cache_lock);
1573 init_waitqueue_head(&root->cache_wait);
1574
1575 ret = get_anon_bdev(&root->anon_dev);
1576 if (ret)
8257b2dc 1577 goto free_writers;
cb517eab 1578 return 0;
8257b2dc
MX		 1579
1580free_writers:
1581 btrfs_free_subvolume_writers(root->subv_writers);
cb517eab
MX		 1582fail:
1583 kfree(root->free_ino_ctl);
1584 kfree(root->free_ino_pinned);
1585 return ret;
1586}
1587
171170c1
ST		 1588static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1589 u64 root_id)
cb517eab
MX		 1590{
1591 struct btrfs_root *root;
1592
1593 spin_lock(&fs_info->fs_roots_radix_lock);
1594 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1595 (unsigned long)root_id);
1596 spin_unlock(&fs_info->fs_roots_radix_lock);
1597 return root;
1598}
1599
1600int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1601 struct btrfs_root *root)
1602{
1603 int ret;
1604
1605 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1606 if (ret)
1607 return ret;
1608
1609 spin_lock(&fs_info->fs_roots_radix_lock);
1610 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1611 (unsigned long)root->root_key.objectid,
1612 root);
1613 if (ret == 0)
27cdeb70 1614 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
cb517eab
MX		 1615 spin_unlock(&fs_info->fs_roots_radix_lock);
1616 radix_tree_preload_end();
1617
1618 return ret;
1619}
1620
c00869f1
MX		 1621struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1622 struct btrfs_key *location,
1623 bool check_ref)
5eda7b5e
CM		 1624{
1625 struct btrfs_root *root;
1626 int ret;
1627
edbd8d4e
CM		 1628 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1629 return fs_info->tree_root;
1630 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1631 return fs_info->extent_root;
8f18cf13
CM		 1632 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1633 return fs_info->chunk_root;
1634 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1635 return fs_info->dev_root;
0403e47e
YZ		 1636 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1637 return fs_info->csum_root;
bcef60f2
AJ		 1638 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1639 return fs_info->quota_root ? fs_info->quota_root :
1640 ERR_PTR(-ENOENT);
f7a81ea4
SB		 1641 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1642 return fs_info->uuid_root ? fs_info->uuid_root :
1643 ERR_PTR(-ENOENT);
4df27c4d 1644again:
cb517eab 1645 root = btrfs_lookup_fs_root(fs_info, location->objectid);
48475471 1646 if (root) {
c00869f1 1647 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
48475471 1648 return ERR_PTR(-ENOENT);
5eda7b5e 1649 return root;
48475471 1650 }
5eda7b5e 1651
cb517eab 1652 root = btrfs_read_fs_root(fs_info->tree_root, location);
5eda7b5e
CM		 1653 if (IS_ERR(root))
1654 return root;
3394e160 1655
c00869f1 1656 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
cb517eab 1657 ret = -ENOENT;
581bb050 1658 goto fail;
35a30d7c 1659 }
581bb050 1660
cb517eab 1661 ret = btrfs_init_fs_root(root);
ac08aedf
CM		 1662 if (ret)
1663 goto fail;
3394e160 1664
3f870c28
KN		 1665 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1666 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
d68fc57b
YZ		 1667 if (ret < 0)
1668 goto fail;
1669 if (ret == 0)
27cdeb70 1670 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
d68fc57b 1671
cb517eab 1672 ret = btrfs_insert_fs_root(fs_info, root);
0f7d52f4 1673 if (ret) {
4df27c4d
YZ		 1674 if (ret == -EEXIST) {
1675 free_fs_root(root);
1676 goto again;
1677 }
1678 goto fail;
0f7d52f4 1679 }
edbd8d4e 1680 return root;
4df27c4d
YZ		 1681fail:
1682 free_fs_root(root);
1683 return ERR_PTR(ret);
edbd8d4e
CM		 1684}
1685
04160088
CM		 1686static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1687{
1688 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1689 int ret = 0;
04160088
CM		 1690 struct btrfs_device *device;
1691 struct backing_dev_info *bdi;
b7967db7 1692
1f78160c
XG		 1693 rcu_read_lock();
1694 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1695 if (!device->bdev)
1696 continue;
04160088
CM		 1697 bdi = blk_get_backing_dev_info(device->bdev);
1698 if (bdi && bdi_congested(bdi, bdi_bits)) {
1699 ret = 1;
1700 break;
1701 }
1702 }
1f78160c 1703 rcu_read_unlock();
04160088
CM		 1704 return ret;
1705}
1706
ad081f14
JA		 1707/*
1708 * If this fails, caller must call bdi_destroy() to get rid of the
1709 * bdi again.
1710 */
04160088
CM		 1711static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1712{
ad081f14
JA		 1713 int err;
1714
1715 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1716 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1717 if (err)
1718 return err;
1719
4575c9cc 1720 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1721 bdi->congested_fn = btrfs_congested_fn;
1722 bdi->congested_data = info;
1723 return 0;
1724}
1725
8b712842
CM		 1726/*
1727 * called by the kthread helper functions to finally call the bio end_io
1728 * functions. This is where read checksum verification actually happens
1729 */
1730static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1731{
ce9adaa5 1732 struct bio *bio;
8b712842 1733 struct end_io_wq *end_io_wq;
ce9adaa5 1734 int error;
ce9adaa5 1735
8b712842
CM		 1736 end_io_wq = container_of(work, struct end_io_wq, work);
1737 bio = end_io_wq->bio;
ce9adaa5 1738
8b712842
CM		 1739 error = end_io_wq->error;
1740 bio->bi_private = end_io_wq->private;
1741 bio->bi_end_io = end_io_wq->end_io;
1742 kfree(end_io_wq);
bc1e79ac 1743 bio_endio_nodec(bio, error);
44b8bd7e
CM		 1744}
1745
a74a4b97
CM		 1746static int cleaner_kthread(void *arg)
1747{
1748 struct btrfs_root *root = arg;
d0278245 1749 int again;
a74a4b97
CM		 1750
1751 do {
d0278245 1752 again = 0;
a74a4b97 1753
d0278245 1754 /* Make the cleaner go to sleep early. */
babbf170 1755 if (btrfs_need_cleaner_sleep(root))
d0278245
MX		 1756 goto sleep;
1757
1758 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1759 goto sleep;
1760
dc7f370c
MX		 1761 /*
1762 * Avoid the problem that we change the status of the fs
1763 * during the above check and trylock.
1764 */
babbf170 1765 if (btrfs_need_cleaner_sleep(root)) {
dc7f370c
MX		 1766 mutex_unlock(&root->fs_info->cleaner_mutex);
1767 goto sleep;
76dda93c 1768 }
a74a4b97 1769
d0278245
MX		 1770 btrfs_run_delayed_iputs(root);
1771 again = btrfs_clean_one_deleted_snapshot(root);
1772 mutex_unlock(&root->fs_info->cleaner_mutex);
1773
1774 /*
05323cd1
MX		 1775 * The defragger has dealt with the R/O remount and umount,
1776 * needn't do anything special here.
d0278245
MX		 1777 */
1778 btrfs_run_defrag_inodes(root->fs_info);
1779sleep:
9d1a2a3a 1780 if (!try_to_freeze() && !again) {
a74a4b97 1781 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1782 if (!kthread_should_stop())
1783 schedule();
a74a4b97
CM		 1784 __set_current_state(TASK_RUNNING);
1785 }
1786 } while (!kthread_should_stop());
1787 return 0;
1788}
1789
1790static int transaction_kthread(void *arg)
1791{
1792 struct btrfs_root *root = arg;
1793 struct btrfs_trans_handle *trans;
1794 struct btrfs_transaction *cur;
8929ecfa 1795 u64 transid;
a74a4b97
CM		 1796 unsigned long now;
1797 unsigned long delay;
914b2007 1798 bool cannot_commit;
a74a4b97
CM		 1799
1800 do {
914b2007 1801 cannot_commit = false;
8b87dc17 1802 delay = HZ * root->fs_info->commit_interval;
a74a4b97
CM		 1803 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1804
a4abeea4 1805 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1806 cur = root->fs_info->running_transaction;
1807 if (!cur) {
a4abeea4 1808 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1809 goto sleep;
1810 }
31153d81 1811
a74a4b97 1812 now = get_seconds();
4a9d8bde 1813 if (cur->state < TRANS_STATE_BLOCKED &&
8b87dc17
DS		 1814 (now < cur->start_time ||
1815 now - cur->start_time < root->fs_info->commit_interval)) {
a4abeea4 1816 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1817 delay = HZ * 5;
1818 goto sleep;
1819 }
8929ecfa 1820 transid = cur->transid;
a4abeea4 1821 spin_unlock(&root->fs_info->trans_lock);
56bec294 1822
79787eaa 1823 /* If the file system is aborted, this will always fail. */
354aa0fb 1824 trans = btrfs_attach_transaction(root);
914b2007 1825 if (IS_ERR(trans)) {
354aa0fb
MX		 1826 if (PTR_ERR(trans) != -ENOENT)
1827 cannot_commit = true;
79787eaa 1828 goto sleep;
914b2007 1829 }
8929ecfa 1830 if (transid == trans->transid) {
79787eaa 1831 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1832 } else {
1833 btrfs_end_transaction(trans, root);
1834 }
a74a4b97
CM		 1835sleep:
1836 wake_up_process(root->fs_info->cleaner_kthread);
1837 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1838
4e121c06
JB		 1839 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1840 &root->fs_info->fs_state)))
1841 btrfs_cleanup_transaction(root);
a0acae0e 1842 if (!try_to_freeze()) {
a74a4b97 1843 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1844 if (!kthread_should_stop() &&
914b2007
JK		 1845 (!btrfs_transaction_blocked(root->fs_info) ||
1846 cannot_commit))
8929ecfa 1847 schedule_timeout(delay);
a74a4b97
CM		 1848 __set_current_state(TASK_RUNNING);
1849 }
1850 } while (!kthread_should_stop());
1851 return 0;
1852}
1853
af31f5e5
CM		 1854/*
1855 * this will find the highest generation in the array of
1856 * root backups. The index of the highest array is returned,
1857 * or -1 if we can't find anything.
1858 *
1859 * We check to make sure the array is valid by comparing the
1860 * generation of the latest root in the array with the generation
1861 * in the super block. If they don't match we pitch it.
1862 */
1863static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1864{
1865 u64 cur;
1866 int newest_index = -1;
1867 struct btrfs_root_backup *root_backup;
1868 int i;
1869
1870 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1871 root_backup = info->super_copy->super_roots + i;
1872 cur = btrfs_backup_tree_root_gen(root_backup);
1873 if (cur == newest_gen)
1874 newest_index = i;
1875 }
1876
1877 /* check to see if we actually wrapped around */
1878 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1879 root_backup = info->super_copy->super_roots;
1880 cur = btrfs_backup_tree_root_gen(root_backup);
1881 if (cur == newest_gen)
1882 newest_index = 0;
1883 }
1884 return newest_index;
1885}
1886
1887
1888/*
1889 * find the oldest backup so we know where to store new entries
1890 * in the backup array. This will set the backup_root_index
1891 * field in the fs_info struct
1892 */
1893static void find_oldest_super_backup(struct btrfs_fs_info *info,
1894 u64 newest_gen)
1895{
1896 int newest_index = -1;
1897
1898 newest_index = find_newest_super_backup(info, newest_gen);
1899 /* if there was garbage in there, just move along */
1900 if (newest_index == -1) {
1901 info->backup_root_index = 0;
1902 } else {
1903 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1904 }
1905}
1906
1907/*
1908 * copy all the root pointers into the super backup array.
1909 * this will bump the backup pointer by one when it is
1910 * done
1911 */
1912static void backup_super_roots(struct btrfs_fs_info *info)
1913{
1914 int next_backup;
1915 struct btrfs_root_backup *root_backup;
1916 int last_backup;
1917
1918 next_backup = info->backup_root_index;
1919 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1920 BTRFS_NUM_BACKUP_ROOTS;
1921
1922 /*
1923 * just overwrite the last backup if we're at the same generation
1924 * this happens only at umount
1925 */
1926 root_backup = info->super_for_commit->super_roots + last_backup;
1927 if (btrfs_backup_tree_root_gen(root_backup) ==
1928 btrfs_header_generation(info->tree_root->node))
1929 next_backup = last_backup;
1930
1931 root_backup = info->super_for_commit->super_roots + next_backup;
1932
1933 /*
1934 * make sure all of our padding and empty slots get zero filled
1935 * regardless of which ones we use today
1936 */
1937 memset(root_backup, 0, sizeof(*root_backup));
1938
1939 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1940
1941 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1942 btrfs_set_backup_tree_root_gen(root_backup,
1943 btrfs_header_generation(info->tree_root->node));
1944
1945 btrfs_set_backup_tree_root_level(root_backup,
1946 btrfs_header_level(info->tree_root->node));
1947
1948 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1949 btrfs_set_backup_chunk_root_gen(root_backup,
1950 btrfs_header_generation(info->chunk_root->node));
1951 btrfs_set_backup_chunk_root_level(root_backup,
1952 btrfs_header_level(info->chunk_root->node));
1953
1954 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1955 btrfs_set_backup_extent_root_gen(root_backup,
1956 btrfs_header_generation(info->extent_root->node));
1957 btrfs_set_backup_extent_root_level(root_backup,
1958 btrfs_header_level(info->extent_root->node));
1959
7c7e82a7
CM		 1960 /*
1961 * we might commit during log recovery, which happens before we set
1962 * the fs_root. Make sure it is valid before we fill it in.
1963 */
1964 if (info->fs_root && info->fs_root->node) {
1965 btrfs_set_backup_fs_root(root_backup,
1966 info->fs_root->node->start);
1967 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1968 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1969 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1970 btrfs_header_level(info->fs_root->node));
7c7e82a7 1971 }
af31f5e5
CM		 1972
1973 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1974 btrfs_set_backup_dev_root_gen(root_backup,
1975 btrfs_header_generation(info->dev_root->node));
1976 btrfs_set_backup_dev_root_level(root_backup,
1977 btrfs_header_level(info->dev_root->node));
1978
1979 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1980 btrfs_set_backup_csum_root_gen(root_backup,
1981 btrfs_header_generation(info->csum_root->node));
1982 btrfs_set_backup_csum_root_level(root_backup,
1983 btrfs_header_level(info->csum_root->node));
1984
1985 btrfs_set_backup_total_bytes(root_backup,
1986 btrfs_super_total_bytes(info->super_copy));
1987 btrfs_set_backup_bytes_used(root_backup,
1988 btrfs_super_bytes_used(info->super_copy));
1989 btrfs_set_backup_num_devices(root_backup,
1990 btrfs_super_num_devices(info->super_copy));
1991
1992 /*
1993 * if we don't copy this out to the super_copy, it won't get remembered
1994 * for the next commit
1995 */
1996 memcpy(&info->super_copy->super_roots,
1997 &info->super_for_commit->super_roots,
1998 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1999}
2000
2001/*
2002 * this copies info out of the root backup array and back into
2003 * the in-memory super block. It is meant to help iterate through
2004 * the array, so you send it the number of backups you've already
2005 * tried and the last backup index you used.
2006 *
2007 * this returns -1 when it has tried all the backups
2008 */
2009static noinline int next_root_backup(struct btrfs_fs_info *info,
2010 struct btrfs_super_block *super,
2011 int *num_backups_tried, int *backup_index)
2012{
2013 struct btrfs_root_backup *root_backup;
2014 int newest = *backup_index;
2015
2016 if (*num_backups_tried == 0) {
2017 u64 gen = btrfs_super_generation(super);
2018
2019 newest = find_newest_super_backup(info, gen);
2020 if (newest == -1)
2021 return -1;
2022
2023 *backup_index = newest;
2024 *num_backups_tried = 1;
2025 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2026 /* we've tried all the backups, all done */
2027 return -1;
2028 } else {
2029 /* jump to the next oldest backup */
2030 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2031 BTRFS_NUM_BACKUP_ROOTS;
2032 *backup_index = newest;
2033 *num_backups_tried += 1;
2034 }
2035 root_backup = super->super_roots + newest;
2036
2037 btrfs_set_super_generation(super,
2038 btrfs_backup_tree_root_gen(root_backup));
2039 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2040 btrfs_set_super_root_level(super,
2041 btrfs_backup_tree_root_level(root_backup));
2042 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2043
2044 /*
2045 * fixme: the total bytes and num_devices need to match or we should
2046 * need a fsck
2047 */
2048 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2049 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2050 return 0;
2051}
2052
7abadb64
LB		 2053/* helper to cleanup workers */
2054static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2055{
dc6e3209 2056 btrfs_destroy_workqueue(fs_info->fixup_workers);
afe3d242 2057 btrfs_destroy_workqueue(fs_info->delalloc_workers);
5cdc7ad3 2058 btrfs_destroy_workqueue(fs_info->workers);
fccb5d86
QW		 2059 btrfs_destroy_workqueue(fs_info->endio_workers);
2060 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2061 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
d05a33ac 2062 btrfs_destroy_workqueue(fs_info->rmw_workers);
fccb5d86
QW		 2063 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2064 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2065 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
a8c93d4e 2066 btrfs_destroy_workqueue(fs_info->submit_workers);
5b3bc44e 2067 btrfs_destroy_workqueue(fs_info->delayed_workers);
e66f0bb1 2068 btrfs_destroy_workqueue(fs_info->caching_workers);
736cfa15 2069 btrfs_destroy_workqueue(fs_info->readahead_workers);
a44903ab 2070 btrfs_destroy_workqueue(fs_info->flush_workers);
fc97fab0 2071 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
7abadb64
LB		 2072}
2073
2e9f5954
R		 2074static void free_root_extent_buffers(struct btrfs_root *root)
2075{
2076 if (root) {
2077 free_extent_buffer(root->node);
2078 free_extent_buffer(root->commit_root);
2079 root->node = NULL;
2080 root->commit_root = NULL;
2081 }
2082}
2083
af31f5e5
CM		 2084/* helper to cleanup tree roots */
2085static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2086{
2e9f5954 2087 free_root_extent_buffers(info->tree_root);
655b09fe 2088
2e9f5954
R		 2089 free_root_extent_buffers(info->dev_root);
2090 free_root_extent_buffers(info->extent_root);
2091 free_root_extent_buffers(info->csum_root);
2092 free_root_extent_buffers(info->quota_root);
2093 free_root_extent_buffers(info->uuid_root);
2094 if (chunk_root)
2095 free_root_extent_buffers(info->chunk_root);
af31f5e5
CM		 2096}
2097
faa2dbf0 2098void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
171f6537
JB		 2099{
2100 int ret;
2101 struct btrfs_root *gang[8];
2102 int i;
2103
2104 while (!list_empty(&fs_info->dead_roots)) {
2105 gang[0] = list_entry(fs_info->dead_roots.next,
2106 struct btrfs_root, root_list);
2107 list_del(&gang[0]->root_list);
2108
27cdeb70 2109 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
cb517eab 2110 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
171f6537
JB		 2111 } else {
2112 free_extent_buffer(gang[0]->node);
2113 free_extent_buffer(gang[0]->commit_root);
b0feb9d9 2114 btrfs_put_fs_root(gang[0]);
171f6537
JB		 2115 }
2116 }
2117
2118 while (1) {
2119 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2120 (void **)gang, 0,
2121 ARRAY_SIZE(gang));
2122 if (!ret)
2123 break;
2124 for (i = 0; i < ret; i++)
cb517eab 2125 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
171f6537 2126 }
1a4319cc
LB		 2127
2128 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2129 btrfs_free_log_root_tree(NULL, fs_info);
2130 btrfs_destroy_pinned_extent(fs_info->tree_root,
2131 fs_info->pinned_extents);
2132 }
171f6537 2133}
af31f5e5 2134
ad2b2c80
AV		 2135int open_ctree(struct super_block *sb,
2136 struct btrfs_fs_devices *fs_devices,
2137 char *options)
2e635a27 2138{
db94535d
CM		 2139 u32 sectorsize;
2140 u32 nodesize;
2141 u32 leafsize;
2142 u32 blocksize;
87ee04eb 2143 u32 stripesize;
84234f3a 2144 u64 generation;
f2b636e8 2145 u64 features;
3de4586c 2146 struct btrfs_key location;
a061fc8d 2147 struct buffer_head *bh;
4d34b278 2148 struct btrfs_super_block *disk_super;
815745cf 2149 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2150 struct btrfs_root *tree_root;
4d34b278
ID		 2151 struct btrfs_root *extent_root;
2152 struct btrfs_root *csum_root;
2153 struct btrfs_root *chunk_root;
2154 struct btrfs_root *dev_root;
bcef60f2 2155 struct btrfs_root *quota_root;
f7a81ea4 2156 struct btrfs_root *uuid_root;
e02119d5 2157 struct btrfs_root *log_tree_root;
eb60ceac 2158 int ret;
e58ca020 2159 int err = -EINVAL;
af31f5e5
CM		 2160 int num_backups_tried = 0;
2161 int backup_index = 0;
5cdc7ad3
QW		 2162 int max_active;
2163 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
70f80175
SB		 2164 bool create_uuid_tree;
2165 bool check_uuid_tree;
4543df7e 2166
f84a8bd6 2167 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e 2168 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
cb517eab 2169 if (!tree_root || !chunk_root) {
39279cc3
CM		 2170 err = -ENOMEM;
2171 goto fail;
2172 }
76dda93c
YZ		 2173
2174 ret = init_srcu_struct(&fs_info->subvol_srcu);
2175 if (ret) {
2176 err = ret;
2177 goto fail;
2178 }
2179
2180 ret = setup_bdi(fs_info, &fs_info->bdi);
2181 if (ret) {
2182 err = ret;
2183 goto fail_srcu;
2184 }
2185
e2d84521
MX		 2186 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2187 if (ret) {
2188 err = ret;
2189 goto fail_bdi;
2190 }
2191 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2192 (1 + ilog2(nr_cpu_ids));
2193
963d678b
MX		 2194 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2195 if (ret) {
2196 err = ret;
2197 goto fail_dirty_metadata_bytes;
2198 }
2199
c404e0dc
MX		 2200 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2201 if (ret) {
2202 err = ret;
2203 goto fail_delalloc_bytes;
2204 }
2205
76dda93c
YZ		 2206 fs_info->btree_inode = new_inode(sb);
2207 if (!fs_info->btree_inode) {
2208 err = -ENOMEM;
c404e0dc 2209 goto fail_bio_counter;
76dda93c
YZ		 2210 }
2211
a6591715 2212 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2213
76dda93c 2214 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
f28491e0 2215 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
8fd17795 2216 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2217 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2218 INIT_LIST_HEAD(&fs_info->delayed_iputs);
eb73c1b7 2219 INIT_LIST_HEAD(&fs_info->delalloc_roots);
11833d66 2220 INIT_LIST_HEAD(&fs_info->caching_block_groups);
eb73c1b7 2221 spin_lock_init(&fs_info->delalloc_root_lock);
a4abeea4 2222 spin_lock_init(&fs_info->trans_lock);
76dda93c 2223 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2224 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2225 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2226 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2227 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2228 spin_lock_init(&fs_info->super_lock);
fcebe456 2229 spin_lock_init(&fs_info->qgroup_op_lock);
f28491e0 2230 spin_lock_init(&fs_info->buffer_lock);
f29021b2 2231 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2232 mutex_init(&fs_info->reloc_mutex);
573bfb72 2233 mutex_init(&fs_info->delalloc_root_mutex);
de98ced9 2234 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2235
58176a96 2236 init_completion(&fs_info->kobj_unregister);
0b86a832 2237 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2238 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2239 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2240 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2241 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2242 BTRFS_BLOCK_RSV_GLOBAL);
2243 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2244 BTRFS_BLOCK_RSV_DELALLOC);
2245 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2246 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2247 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2248 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2249 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2250 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2251 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2252 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2253 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2254 atomic_set(&fs_info->defrag_running, 0);
fcebe456 2255 atomic_set(&fs_info->qgroup_op_seq, 0);
fc36ed7e 2256 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2257 fs_info->sb = sb;
6f568d35 2258 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2259 fs_info->metadata_ratio = 0;
4cb5300b 2260 fs_info->defrag_inodes = RB_ROOT;
2bf64758 2261 fs_info->free_chunk_space = 0;
f29021b2 2262 fs_info->tree_mod_log = RB_ROOT;
8b87dc17 2263 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
0a2b2a84 2264 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
90519d66
AJ		 2265 /* readahead state */
2266 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2267 spin_lock_init(&fs_info->reada_lock);
c8b97818 2268
b34b086c
CM		 2269 fs_info->thread_pool_size = min_t(unsigned long,
2270 num_online_cpus() + 2, 8);
0afbaf8c 2271
199c2a9c
MX		 2272 INIT_LIST_HEAD(&fs_info->ordered_roots);
2273 spin_lock_init(&fs_info->ordered_root_lock);
16cdcec7
MX		 2274 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2275 GFP_NOFS);
2276 if (!fs_info->delayed_root) {
2277 err = -ENOMEM;
2278 goto fail_iput;
2279 }
2280 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2281
a2de733c
AJ		 2282 mutex_init(&fs_info->scrub_lock);
2283 atomic_set(&fs_info->scrubs_running, 0);
2284 atomic_set(&fs_info->scrub_pause_req, 0);
2285 atomic_set(&fs_info->scrubs_paused, 0);
2286 atomic_set(&fs_info->scrub_cancel_req, 0);
c404e0dc 2287 init_waitqueue_head(&fs_info->replace_wait);
a2de733c 2288 init_waitqueue_head(&fs_info->scrub_pause_wait);
a2de733c 2289 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2290#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2291 fs_info->check_integrity_print_mask = 0;
2292#endif
a2de733c 2293
c9e9f97b
ID		 2294 spin_lock_init(&fs_info->balance_lock);
2295 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2296 atomic_set(&fs_info->balance_running, 0);
2297 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2298 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2299 fs_info->balance_ctl = NULL;
837d5b6e 2300 init_waitqueue_head(&fs_info->balance_wait_q);
21c7e756 2301 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
a2de733c 2302
a061fc8d
CM		 2303 sb->s_blocksize = 4096;
2304 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2305 sb->s_bdi = &fs_info->bdi;
a061fc8d 2306
76dda93c 2307 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2308 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2309 /*
2310 * we set the i_size on the btree inode to the max possible int.
2311 * the real end of the address space is determined by all of
2312 * the devices in the system
2313 */
2314 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2315 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2316 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2317
5d4f98a2 2318 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2319 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2320 fs_info->btree_inode->i_mapping);
0b32f4bb 2321 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2322 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2323
2324 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2325
76dda93c
YZ		 2326 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2327 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2328 sizeof(struct btrfs_key));
72ac3c0d
JB		 2329 set_bit(BTRFS_INODE_DUMMY,
2330 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
778ba82b 2331 btrfs_insert_inode_hash(fs_info->btree_inode);
76dda93c 2332
0f9dd46c 2333 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2334 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2335 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2336
11833d66 2337 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2338 fs_info->btree_inode->i_mapping);
11833d66 2339 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2340 fs_info->btree_inode->i_mapping);
11833d66 2341 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2342 fs_info->do_barriers = 1;
e18e4809 2343
39279cc3 2344
5a3f23d5 2345 mutex_init(&fs_info->ordered_operations_mutex);
9ffba8cd 2346 mutex_init(&fs_info->ordered_extent_flush_mutex);
e02119d5 2347 mutex_init(&fs_info->tree_log_mutex);
925baedd 2348 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2349 mutex_init(&fs_info->transaction_kthread_mutex);
2350 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2351 mutex_init(&fs_info->volume_mutex);
9e351cc8 2352 init_rwsem(&fs_info->commit_root_sem);
c71bf099 2353 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2354 init_rwsem(&fs_info->subvol_sem);
803b2f54 2355 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
e922e087
SB		 2356 fs_info->dev_replace.lock_owner = 0;
2357 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2358 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2359 mutex_init(&fs_info->dev_replace.lock_management_lock);
2360 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2361
416ac51d 2362 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2363 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d 2364 fs_info->qgroup_tree = RB_ROOT;
fcebe456 2365 fs_info->qgroup_op_tree = RB_ROOT;
416ac51d
AJ		 2366 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2367 fs_info->qgroup_seq = 1;
2368 fs_info->quota_enabled = 0;
2369 fs_info->pending_quota_state = 0;
1e8f9158 2370 fs_info->qgroup_ulist = NULL;
2f232036 2371 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2372
fa9c0d79
CM		 2373 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2374 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2375
e6dcd2dc 2376 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2377 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2378 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2379 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2380
53b381b3
DW		 2381 ret = btrfs_alloc_stripe_hash_table(fs_info);
2382 if (ret) {
83c8266a 2383 err = ret;
53b381b3
DW		 2384 goto fail_alloc;
2385 }
2386
0b86a832 2387 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2388 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2389
3c4bb26b 2390 invalidate_bdev(fs_devices->latest_bdev);
1104a885
DS		 2391
2392 /*
2393 * Read super block and check the signature bytes only
2394 */
a512bbf8 2395 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2396 if (!bh) {
2397 err = -EINVAL;
16cdcec7 2398 goto fail_alloc;
20b45077 2399 }
39279cc3 2400
1104a885
DS		 2401 /*
2402 * We want to check superblock checksum, the type is stored inside.
2403 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2404 */
2405 if (btrfs_check_super_csum(bh->b_data)) {
efe120a0 2406 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
1104a885
DS		 2407 err = -EINVAL;
2408 goto fail_alloc;
2409 }
2410
2411 /*
2412 * super_copy is zeroed at allocation time and we never touch the
2413 * following bytes up to INFO_SIZE, the checksum is calculated from
2414 * the whole block of INFO_SIZE
2415 */
6c41761f
DS		 2416 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2417 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2418 sizeof(*fs_info->super_for_commit));
a061fc8d 2419 brelse(bh);
5f39d397 2420
6c41761f 2421 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2422
1104a885
DS		 2423 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2424 if (ret) {
efe120a0 2425 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
1104a885
DS		 2426 err = -EINVAL;
2427 goto fail_alloc;
2428 }
2429
6c41761f 2430 disk_super = fs_info->super_copy;
0f7d52f4 2431 if (!btrfs_super_root(disk_super))
16cdcec7 2432 goto fail_alloc;
0f7d52f4 2433
acce952b 2434 /* check FS state, whether FS is broken. */
87533c47
MX		 2435 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2436 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2437
af31f5e5
CM		 2438 /*
2439 * run through our array of backup supers and setup
2440 * our ring pointer to the oldest one
2441 */
2442 generation = btrfs_super_generation(disk_super);
2443 find_oldest_super_backup(fs_info, generation);
2444
75e7cb7f
LB		 2445 /*
2446 * In the long term, we'll store the compression type in the super
2447 * block, and it'll be used for per file compression control.
2448 */
2449 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2450
2b82032c
YZ		 2451 ret = btrfs_parse_options(tree_root, options);
2452 if (ret) {
2453 err = ret;
16cdcec7 2454 goto fail_alloc;
2b82032c 2455 }
dfe25020 2456
f2b636e8
JB		 2457 features = btrfs_super_incompat_flags(disk_super) &
2458 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2459 if (features) {
2460 printk(KERN_ERR "BTRFS: couldn't mount because of "
2461 "unsupported optional features (%Lx).\n",
c1c9ff7c 2462 features);
f2b636e8 2463 err = -EINVAL;
16cdcec7 2464 goto fail_alloc;
f2b636e8
JB		 2465 }
2466
727011e0
CM		 2467 if (btrfs_super_leafsize(disk_super) !=
2468 btrfs_super_nodesize(disk_super)) {
2469 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2470 "blocksizes don't match. node %d leaf %d\n",
2471 btrfs_super_nodesize(disk_super),
2472 btrfs_super_leafsize(disk_super));
2473 err = -EINVAL;
2474 goto fail_alloc;
2475 }
2476 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2477 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2478 "blocksize (%d) was too large\n",
2479 btrfs_super_leafsize(disk_super));
2480 err = -EINVAL;
2481 goto fail_alloc;
2482 }
2483
5d4f98a2 2484 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2485 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2486 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2487 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2488
3173a18f 2489 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
efe120a0 2490 printk(KERN_ERR "BTRFS: has skinny extents\n");
3173a18f 2491
727011e0
CM		 2492 /*
2493 * flag our filesystem as having big metadata blocks if
2494 * they are bigger than the page size
2495 */
2496 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2497 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
efe120a0 2498 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
727011e0
CM		 2499 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2500 }
2501
bc3f116f
CM		 2502 nodesize = btrfs_super_nodesize(disk_super);
2503 leafsize = btrfs_super_leafsize(disk_super);
2504 sectorsize = btrfs_super_sectorsize(disk_super);
2505 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2506 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2507 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2508
2509 /*
2510 * mixed block groups end up with duplicate but slightly offset
2511 * extent buffers for the same range. It leads to corruptions
2512 */
2513 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2514 (sectorsize != leafsize)) {
efe120a0 2515 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
bc3f116f
CM		 2516 "are not allowed for mixed block groups on %s\n",
2517 sb->s_id);
2518 goto fail_alloc;
2519 }
2520
ceda0864
MX		 2521 /*
2522 * Needn't use the lock because there is no other task which will
2523 * update the flag.
2524 */
a6fa6fae 2525 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2526
f2b636e8
JB		 2527 features = btrfs_super_compat_ro_flags(disk_super) &
2528 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2529 if (!(sb->s_flags & MS_RDONLY) && features) {
2530 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2531 "unsupported option features (%Lx).\n",
c1c9ff7c 2532 features);
f2b636e8 2533 err = -EINVAL;
16cdcec7 2534 goto fail_alloc;
f2b636e8 2535 }
61d92c32 2536
5cdc7ad3 2537 max_active = fs_info->thread_pool_size;
61d92c32 2538
5cdc7ad3
QW		 2539 fs_info->workers =
2540 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2541 max_active, 16);
c8b97818 2542
afe3d242
QW		 2543 fs_info->delalloc_workers =
2544 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
771ed689 2545
a44903ab
QW		 2546 fs_info->flush_workers =
2547 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
8ccf6f19 2548
e66f0bb1
QW		 2549 fs_info->caching_workers =
2550 btrfs_alloc_workqueue("cache", flags, max_active, 0);
61b49440 2551
a8c93d4e
QW		 2552 /*
2553 * a higher idle thresh on the submit workers makes it much more
61b49440
CM		 2554 * likely that bios will be send down in a sane order to the
2555 * devices
2556 */
a8c93d4e
QW		 2557 fs_info->submit_workers =
2558 btrfs_alloc_workqueue("submit", flags,
2559 min_t(u64, fs_devices->num_devices,
2560 max_active), 64);
53863232 2561
dc6e3209
QW		 2562 fs_info->fixup_workers =
2563 btrfs_alloc_workqueue("fixup", flags, 1, 0);
61b49440
CM		 2564
2565 /*
2566 * endios are largely parallel and should have a very
2567 * low idle thresh
2568 */
fccb5d86
QW		 2569 fs_info->endio_workers =
2570 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2571 fs_info->endio_meta_workers =
2572 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2573 fs_info->endio_meta_write_workers =
2574 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2575 fs_info->endio_raid56_workers =
2576 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
d05a33ac
QW		 2577 fs_info->rmw_workers =
2578 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
fccb5d86
QW		 2579 fs_info->endio_write_workers =
2580 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2581 fs_info->endio_freespace_worker =
2582 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
5b3bc44e
QW		 2583 fs_info->delayed_workers =
2584 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
736cfa15
QW		 2585 fs_info->readahead_workers =
2586 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
fc97fab0
QW		 2587 fs_info->qgroup_rescan_workers =
2588 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
61b49440 2589
a8c93d4e 2590 if (!(fs_info->workers && fs_info->delalloc_workers &&
fccb5d86
QW		 2591 fs_info->submit_workers && fs_info->flush_workers &&
2592 fs_info->endio_workers && fs_info->endio_meta_workers &&
2593 fs_info->endio_meta_write_workers &&
2594 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
e66f0bb1 2595 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
dc6e3209 2596 fs_info->caching_workers && fs_info->readahead_workers &&
fc97fab0
QW		 2597 fs_info->fixup_workers && fs_info->delayed_workers &&
2598 fs_info->qgroup_rescan_workers)) {
fed425c7 2599 err = -ENOMEM;
0dc3b84a
JB		 2600 goto fail_sb_buffer;
2601 }
4543df7e 2602
4575c9cc 2603 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2604 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2605 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2606
db94535d
CM		 2607 tree_root->nodesize = nodesize;
2608 tree_root->leafsize = leafsize;
2609 tree_root->sectorsize = sectorsize;
87ee04eb 2610 tree_root->stripesize = stripesize;
a061fc8d
CM		 2611
2612 sb->s_blocksize = sectorsize;
2613 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2614
3cae210f 2615 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
efe120a0 2616 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2617 goto fail_sb_buffer;
2618 }
19c00ddc 2619
8d082fb7 2620 if (sectorsize != PAGE_SIZE) {
efe120a0 2621 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
8d082fb7 2622 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2623 goto fail_sb_buffer;
2624 }
2625
925baedd 2626 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2627 ret = btrfs_read_sys_array(tree_root);
925baedd 2628 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2629 if (ret) {
efe120a0 2630 printk(KERN_WARNING "BTRFS: failed to read the system "
d397712b 2631 "array on %s\n", sb->s_id);
5d4f98a2 2632 goto fail_sb_buffer;
84eed90f 2633 }
0b86a832
CM		 2634
2635 blocksize = btrfs_level_size(tree_root,
2636 btrfs_super_chunk_root_level(disk_super));
84234f3a 2637 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2638
2639 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2640 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2641
2642 chunk_root->node = read_tree_block(chunk_root,
2643 btrfs_super_chunk_root(disk_super),
84234f3a 2644 blocksize, generation);
416bc658
JB		 2645 if (!chunk_root->node ||
2646 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
efe120a0 2647 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
83121942 2648 sb->s_id);
af31f5e5 2649 goto fail_tree_roots;
83121942 2650 }
5d4f98a2
YZ		 2651 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2652 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2653
e17cade2 2654 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
b308bc2f 2655 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
e17cade2 2656
0b86a832 2657 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2658 if (ret) {
efe120a0 2659 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
d397712b 2660 sb->s_id);
af31f5e5 2661 goto fail_tree_roots;
2b82032c 2662 }
0b86a832 2663
8dabb742
SB		 2664 /*
2665 * keep the device that is marked to be the target device for the
2666 * dev_replace procedure
2667 */
2668 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2669
a6b0d5c8 2670 if (!fs_devices->latest_bdev) {
efe120a0 2671 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
a6b0d5c8
CM		 2672 sb->s_id);
2673 goto fail_tree_roots;
2674 }
2675
af31f5e5 2676retry_root_backup:
db94535d
CM		 2677 blocksize = btrfs_level_size(tree_root,
2678 btrfs_super_root_level(disk_super));
84234f3a 2679 generation = btrfs_super_generation(disk_super);
0b86a832 2680
e20d96d6 2681 tree_root->node = read_tree_block(tree_root,
db94535d 2682 btrfs_super_root(disk_super),
84234f3a 2683 blocksize, generation);
af31f5e5
CM		 2684 if (!tree_root->node ||
2685 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
efe120a0 2686 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
83121942 2687 sb->s_id);
af31f5e5
CM		 2688
2689 goto recovery_tree_root;
83121942 2690 }
af31f5e5 2691
5d4f98a2
YZ		 2692 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2693 tree_root->commit_root = btrfs_root_node(tree_root);
69e9c6c6 2694 btrfs_set_root_refs(&tree_root->root_item, 1);
db94535d 2695
cb517eab
MX		 2696 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2697 location.type = BTRFS_ROOT_ITEM_KEY;
2698 location.offset = 0;
2699
2700 extent_root = btrfs_read_tree_root(tree_root, &location);
2701 if (IS_ERR(extent_root)) {
2702 ret = PTR_ERR(extent_root);
af31f5e5 2703 goto recovery_tree_root;
cb517eab 2704 }
27cdeb70 2705 set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
cb517eab 2706 fs_info->extent_root = extent_root;
0b86a832 2707
cb517eab
MX		 2708 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2709 dev_root = btrfs_read_tree_root(tree_root, &location);
2710 if (IS_ERR(dev_root)) {
2711 ret = PTR_ERR(dev_root);
af31f5e5 2712 goto recovery_tree_root;
cb517eab 2713 }
27cdeb70 2714 set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
cb517eab
MX		 2715 fs_info->dev_root = dev_root;
2716 btrfs_init_devices_late(fs_info);
3768f368 2717
cb517eab
MX		 2718 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2719 csum_root = btrfs_read_tree_root(tree_root, &location);
2720 if (IS_ERR(csum_root)) {
2721 ret = PTR_ERR(csum_root);
af31f5e5 2722 goto recovery_tree_root;
cb517eab 2723 }
27cdeb70 2724 set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
cb517eab 2725 fs_info->csum_root = csum_root;
d20f7043 2726
cb517eab
MX		 2727 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2728 quota_root = btrfs_read_tree_root(tree_root, &location);
2729 if (!IS_ERR(quota_root)) {
27cdeb70 2730 set_bit(BTRFS_ROOT_TRACK_DIRTY, &quota_root->state);
bcef60f2
AJ		 2731 fs_info->quota_enabled = 1;
2732 fs_info->pending_quota_state = 1;
cb517eab 2733 fs_info->quota_root = quota_root;
bcef60f2
AJ		 2734 }
2735
f7a81ea4
SB		 2736 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2737 uuid_root = btrfs_read_tree_root(tree_root, &location);
2738 if (IS_ERR(uuid_root)) {
2739 ret = PTR_ERR(uuid_root);
2740 if (ret != -ENOENT)
2741 goto recovery_tree_root;
2742 create_uuid_tree = true;
70f80175 2743 check_uuid_tree = false;
f7a81ea4 2744 } else {
27cdeb70 2745 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
f7a81ea4 2746 fs_info->uuid_root = uuid_root;
70f80175
SB		 2747 create_uuid_tree = false;
2748 check_uuid_tree =
2749 generation != btrfs_super_uuid_tree_generation(disk_super);
f7a81ea4
SB		 2750 }
2751
8929ecfa
YZ		 2752 fs_info->generation = generation;
2753 fs_info->last_trans_committed = generation;
8929ecfa 2754
68310a5e
ID		 2755 ret = btrfs_recover_balance(fs_info);
2756 if (ret) {
efe120a0 2757 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
68310a5e
ID		 2758 goto fail_block_groups;
2759 }
2760
733f4fbb
SB		 2761 ret = btrfs_init_dev_stats(fs_info);
2762 if (ret) {
efe120a0 2763 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
733f4fbb
SB		 2764 ret);
2765 goto fail_block_groups;
2766 }
2767
8dabb742
SB		 2768 ret = btrfs_init_dev_replace(fs_info);
2769 if (ret) {
efe120a0 2770 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
8dabb742
SB		 2771 goto fail_block_groups;
2772 }
2773
2774 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2775
5ac1d209 2776 ret = btrfs_sysfs_add_one(fs_info);
c59021f8 2777 if (ret) {
efe120a0 2778 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
c59021f8 2779 goto fail_block_groups;
2780 }
2781
c59021f8 2782 ret = btrfs_init_space_info(fs_info);
2783 if (ret) {
efe120a0 2784 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2365dd3c 2785 goto fail_sysfs;
c59021f8 2786 }
2787
1b1d1f66
JB		 2788 ret = btrfs_read_block_groups(extent_root);
2789 if (ret) {
efe120a0 2790 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2365dd3c 2791 goto fail_sysfs;
1b1d1f66 2792 }
5af3e8cc
SB		 2793 fs_info->num_tolerated_disk_barrier_failures =
2794 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2795 if (fs_info->fs_devices->missing_devices >
2796 fs_info->num_tolerated_disk_barrier_failures &&
2797 !(sb->s_flags & MS_RDONLY)) {
efe120a0
FH		 2798 printk(KERN_WARNING "BTRFS: "
2799 "too many missing devices, writeable mount is not allowed\n");
2365dd3c 2800 goto fail_sysfs;
292fd7fc 2801 }
9078a3e1 2802
a74a4b97
CM		 2803 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2804 "btrfs-cleaner");
57506d50 2805 if (IS_ERR(fs_info->cleaner_kthread))
2365dd3c 2806 goto fail_sysfs;
a74a4b97
CM		 2807
2808 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2809 tree_root,
2810 "btrfs-transaction");
57506d50 2811 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2812 goto fail_cleaner;
a74a4b97 2813
c289811c
CM		 2814 if (!btrfs_test_opt(tree_root, SSD) &&
2815 !btrfs_test_opt(tree_root, NOSSD) &&
2816 !fs_info->fs_devices->rotating) {
efe120a0 2817 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
c289811c
CM		 2818 "mode\n");
2819 btrfs_set_opt(fs_info->mount_opt, SSD);
2820 }
2821
3818aea2
QW		 2822 /* Set the real inode map cache flag */
2823 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2824 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2825
21adbd5c
SB		 2826#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2827 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2828 ret = btrfsic_mount(tree_root, fs_devices,
2829 btrfs_test_opt(tree_root,
2830 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2831 1 : 0,
2832 fs_info->check_integrity_print_mask);
2833 if (ret)
efe120a0 2834 printk(KERN_WARNING "BTRFS: failed to initialize"
21adbd5c
SB		 2835 " integrity check module %s\n", sb->s_id);
2836 }
2837#endif
bcef60f2
AJ		 2838 ret = btrfs_read_qgroup_config(fs_info);
2839 if (ret)
2840 goto fail_trans_kthread;
21adbd5c 2841
acce952b 2842 /* do not make disk changes in broken FS */
68ce9682 2843 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2844 u64 bytenr = btrfs_super_log_root(disk_super);
2845
7c2ca468 2846 if (fs_devices->rw_devices == 0) {
efe120a0 2847 printk(KERN_WARNING "BTRFS: log replay required "
d397712b 2848 "on RO media\n");
7c2ca468 2849 err = -EIO;
bcef60f2 2850 goto fail_qgroup;
7c2ca468 2851 }
e02119d5
CM		 2852 blocksize =
2853 btrfs_level_size(tree_root,
2854 btrfs_super_log_root_level(disk_super));
d18a2c44 2855
6f07e42e 2856 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2857 if (!log_tree_root) {
2858 err = -ENOMEM;
bcef60f2 2859 goto fail_qgroup;
676e4c86 2860 }
e02119d5
CM		 2861
2862 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2863 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2864
2865 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2866 blocksize,
2867 generation + 1);
416bc658
JB		 2868 if (!log_tree_root->node ||
2869 !extent_buffer_uptodate(log_tree_root->node)) {
efe120a0 2870 printk(KERN_ERR "BTRFS: failed to read log tree\n");
416bc658
JB		 2871 free_extent_buffer(log_tree_root->node);
2872 kfree(log_tree_root);
28c16cbb 2873 goto fail_qgroup;
416bc658 2874 }
79787eaa 2875 /* returns with log_tree_root freed on success */
e02119d5 2876 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2877 if (ret) {
2878 btrfs_error(tree_root->fs_info, ret,
2879 "Failed to recover log tree");
2880 free_extent_buffer(log_tree_root->node);
2881 kfree(log_tree_root);
28c16cbb 2882 goto fail_qgroup;
79787eaa 2883 }
e556ce2c
YZ		 2884
2885 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2886 ret = btrfs_commit_super(tree_root);
2887 if (ret)
28c16cbb 2888 goto fail_qgroup;
e556ce2c 2889 }
e02119d5 2890 }
1a40e23b 2891
76dda93c 2892 ret = btrfs_find_orphan_roots(tree_root);
79787eaa 2893 if (ret)
28c16cbb 2894 goto fail_qgroup;
76dda93c 2895
7c2ca468 2896 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2897 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2 2898 if (ret)
28c16cbb 2899 goto fail_qgroup;
d68fc57b 2900
5d4f98a2 2901 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2902 if (ret < 0) {
2903 printk(KERN_WARNING
efe120a0 2904 "BTRFS: failed to recover relocation\n");
d7ce5843 2905 err = -EINVAL;
bcef60f2 2906 goto fail_qgroup;
d7ce5843 2907 }
7c2ca468 2908 }
1a40e23b 2909
3de4586c
CM		 2910 location.objectid = BTRFS_FS_TREE_OBJECTID;
2911 location.type = BTRFS_ROOT_ITEM_KEY;
cb517eab 2912 location.offset = 0;
3de4586c 2913
3de4586c 2914 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3140c9a3
DC		 2915 if (IS_ERR(fs_info->fs_root)) {
2916 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2917 goto fail_qgroup;
3140c9a3 2918 }
c289811c 2919
2b6ba629
ID		 2920 if (sb->s_flags & MS_RDONLY)
2921 return 0;
59641015 2922
2b6ba629
ID		 2923 down_read(&fs_info->cleanup_work_sem);
2924 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2925 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2926 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2927 close_ctree(tree_root);
2928 return ret;
2929 }
2930 up_read(&fs_info->cleanup_work_sem);
59641015 2931
2b6ba629
ID		 2932 ret = btrfs_resume_balance_async(fs_info);
2933 if (ret) {
efe120a0 2934 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2b6ba629
ID		 2935 close_ctree(tree_root);
2936 return ret;
e3acc2a6
JB		 2937 }
2938
8dabb742
SB		 2939 ret = btrfs_resume_dev_replace_async(fs_info);
2940 if (ret) {
efe120a0 2941 pr_warn("BTRFS: failed to resume dev_replace\n");
8dabb742
SB		 2942 close_ctree(tree_root);
2943 return ret;
2944 }
2945
b382a324
JS		 2946 btrfs_qgroup_rescan_resume(fs_info);
2947
f7a81ea4 2948 if (create_uuid_tree) {
efe120a0 2949 pr_info("BTRFS: creating UUID tree\n");
f7a81ea4
SB		 2950 ret = btrfs_create_uuid_tree(fs_info);
2951 if (ret) {
efe120a0 2952 pr_warn("BTRFS: failed to create the UUID tree %d\n",
f7a81ea4
SB		 2953 ret);
2954 close_ctree(tree_root);
2955 return ret;
2956 }
f420ee1e
SB		 2957 } else if (check_uuid_tree ||
2958 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
efe120a0 2959 pr_info("BTRFS: checking UUID tree\n");
70f80175
SB		 2960 ret = btrfs_check_uuid_tree(fs_info);
2961 if (ret) {
efe120a0 2962 pr_warn("BTRFS: failed to check the UUID tree %d\n",
70f80175
SB		 2963 ret);
2964 close_ctree(tree_root);
2965 return ret;
2966 }
2967 } else {
2968 fs_info->update_uuid_tree_gen = 1;
f7a81ea4
SB		 2969 }
2970
ad2b2c80 2971 return 0;
39279cc3 2972
bcef60f2
AJ		 2973fail_qgroup:
2974 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2975fail_trans_kthread:
2976 kthread_stop(fs_info->transaction_kthread);
54067ae9 2977 btrfs_cleanup_transaction(fs_info->tree_root);
faa2dbf0 2978 btrfs_free_fs_roots(fs_info);
3f157a2f 2979fail_cleaner:
a74a4b97 2980 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2981
2982 /*
2983 * make sure we're done with the btree inode before we stop our
2984 * kthreads
2985 */
2986 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2987
2365dd3c
AJ		 2988fail_sysfs:
2989 btrfs_sysfs_remove_one(fs_info);
2990
1b1d1f66 2991fail_block_groups:
54067ae9 2992 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2993 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2994
2995fail_tree_roots:
2996 free_root_pointers(fs_info, 1);
2b8195bb 2997 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2998
39279cc3 2999fail_sb_buffer:
7abadb64 3000 btrfs_stop_all_workers(fs_info);
16cdcec7 3001fail_alloc:
4543df7e 3002fail_iput:
586e46e2
ID		 3003 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3004
4543df7e 3005 iput(fs_info->btree_inode);
c404e0dc
MX		 3006fail_bio_counter:
3007 percpu_counter_destroy(&fs_info->bio_counter);
963d678b
MX		 3008fail_delalloc_bytes:
3009 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 3010fail_dirty_metadata_bytes:
3011 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 3012fail_bdi:
7e662854 3013 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 3014fail_srcu:
3015 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 3016fail:
53b381b3 3017 btrfs_free_stripe_hash_table(fs_info);
586e46e2 3018 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 3019 return err;
af31f5e5
CM		 3020
3021recovery_tree_root:
af31f5e5
CM		 3022 if (!btrfs_test_opt(tree_root, RECOVERY))
3023 goto fail_tree_roots;
3024
3025 free_root_pointers(fs_info, 0);
3026
3027 /* don't use the log in recovery mode, it won't be valid */
3028 btrfs_set_super_log_root(disk_super, 0);
3029
3030 /* we can't trust the free space cache either */
3031 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3032
3033 ret = next_root_backup(fs_info, fs_info->super_copy,
3034 &num_backups_tried, &backup_index);
3035 if (ret == -1)
3036 goto fail_block_groups;
3037 goto retry_root_backup;
eb60ceac
CM		 3038}
3039
f2984462
CM		 3040static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3041{
f2984462
CM		 3042 if (uptodate) {
3043 set_buffer_uptodate(bh);
3044 } else {
442a4f63
SB		 3045 struct btrfs_device *device = (struct btrfs_device *)
3046 bh->b_private;
3047
efe120a0 3048 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
606686ee
JB		 3049 "I/O error on %s\n",
3050 rcu_str_deref(device->name));
1259ab75
CM		 3051 /* note, we dont' set_buffer_write_io_error because we have
3052 * our own ways of dealing with the IO errors
3053 */
f2984462 3054 clear_buffer_uptodate(bh);
442a4f63 3055 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 3056 }
3057 unlock_buffer(bh);
3058 put_bh(bh);
3059}
3060
a512bbf8
YZ		 3061struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3062{
3063 struct buffer_head *bh;
3064 struct buffer_head *latest = NULL;
3065 struct btrfs_super_block *super;
3066 int i;
3067 u64 transid = 0;
3068 u64 bytenr;
3069
3070 /* we would like to check all the supers, but that would make
3071 * a btrfs mount succeed after a mkfs from a different FS.
3072 * So, we need to add a special mount option to scan for
3073 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3074 */
3075 for (i = 0; i < 1; i++) {
3076 bytenr = btrfs_sb_offset(i);
8068a47e
AJ		 3077 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3078 i_size_read(bdev->bd_inode))
a512bbf8 3079 break;
8068a47e
AJ		 3080 bh = __bread(bdev, bytenr / 4096,
3081 BTRFS_SUPER_INFO_SIZE);
a512bbf8
YZ		 3082 if (!bh)
3083 continue;
3084
3085 super = (struct btrfs_super_block *)bh->b_data;
3086 if (btrfs_super_bytenr(super) != bytenr ||
3cae210f 3087 btrfs_super_magic(super) != BTRFS_MAGIC) {
a512bbf8
YZ		 3088 brelse(bh);
3089 continue;
3090 }
3091
3092 if (!latest || btrfs_super_generation(super) > transid) {
3093 brelse(latest);
3094 latest = bh;
3095 transid = btrfs_super_generation(super);
3096 } else {
3097 brelse(bh);
3098 }
3099 }
3100 return latest;
3101}
3102
4eedeb75
HH		 3103/*
3104 * this should be called twice, once with wait == 0 and
3105 * once with wait == 1. When wait == 0 is done, all the buffer heads
3106 * we write are pinned.
3107 *
3108 * They are released when wait == 1 is done.
3109 * max_mirrors must be the same for both runs, and it indicates how
3110 * many supers on this one device should be written.
3111 *
3112 * max_mirrors == 0 means to write them all.
3113 */
a512bbf8
YZ		 3114static int write_dev_supers(struct btrfs_device *device,
3115 struct btrfs_super_block *sb,
3116 int do_barriers, int wait, int max_mirrors)
3117{
3118 struct buffer_head *bh;
3119 int i;
3120 int ret;
3121 int errors = 0;
3122 u32 crc;
3123 u64 bytenr;
a512bbf8
YZ		 3124
3125 if (max_mirrors == 0)
3126 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3127
a512bbf8
YZ		 3128 for (i = 0; i < max_mirrors; i++) {
3129 bytenr = btrfs_sb_offset(i);
3130 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3131 break;
3132
3133 if (wait) {
3134 bh = __find_get_block(device->bdev, bytenr / 4096,
3135 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3136 if (!bh) {
3137 errors++;
3138 continue;
3139 }
a512bbf8 3140 wait_on_buffer(bh);
4eedeb75
HH		 3141 if (!buffer_uptodate(bh))
3142 errors++;
3143
3144 /* drop our reference */
3145 brelse(bh);
3146
3147 /* drop the reference from the wait == 0 run */
3148 brelse(bh);
3149 continue;
a512bbf8
YZ		 3150 } else {
3151 btrfs_set_super_bytenr(sb, bytenr);
3152
3153 crc = ~(u32)0;
b0496686 3154 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 3155 BTRFS_CSUM_SIZE, crc,
3156 BTRFS_SUPER_INFO_SIZE -
3157 BTRFS_CSUM_SIZE);
3158 btrfs_csum_final(crc, sb->csum);
3159
4eedeb75
HH		 3160 /*
3161 * one reference for us, and we leave it for the
3162 * caller
3163 */
a512bbf8
YZ		 3164 bh = __getblk(device->bdev, bytenr / 4096,
3165 BTRFS_SUPER_INFO_SIZE);
634554dc 3166 if (!bh) {
efe120a0 3167 printk(KERN_ERR "BTRFS: couldn't get super "
634554dc
JB		 3168 "buffer head for bytenr %Lu\n", bytenr);
3169 errors++;
3170 continue;
3171 }
3172
a512bbf8
YZ		 3173 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3174
4eedeb75 3175 /* one reference for submit_bh */
a512bbf8 3176 get_bh(bh);
4eedeb75
HH		 3177
3178 set_buffer_uptodate(bh);
a512bbf8
YZ		 3179 lock_buffer(bh);
3180 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3181 bh->b_private = device;
a512bbf8
YZ		 3182 }
3183
387125fc
CM		 3184 /*
3185 * we fua the first super. The others we allow
3186 * to go down lazy.
3187 */
e8117c26
WS		 3188 if (i == 0)
3189 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3190 else
3191 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
4eedeb75 3192 if (ret)
a512bbf8 3193 errors++;
a512bbf8
YZ		 3194 }
3195 return errors < i ? 0 : -1;
3196}
3197
387125fc
CM		 3198/*
3199 * endio for the write_dev_flush, this will wake anyone waiting
3200 * for the barrier when it is done
3201 */
3202static void btrfs_end_empty_barrier(struct bio *bio, int err)
3203{
3204 if (err) {
3205 if (err == -EOPNOTSUPP)
3206 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3207 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3208 }
3209 if (bio->bi_private)
3210 complete(bio->bi_private);
3211 bio_put(bio);
3212}
3213
3214/*
3215 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3216 * sent down. With wait == 1, it waits for the previous flush.
3217 *
3218 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3219 * capable
3220 */
3221static int write_dev_flush(struct btrfs_device *device, int wait)
3222{
3223 struct bio *bio;
3224 int ret = 0;
3225
3226 if (device->nobarriers)
3227 return 0;
3228
3229 if (wait) {
3230 bio = device->flush_bio;
3231 if (!bio)
3232 return 0;
3233
3234 wait_for_completion(&device->flush_wait);
3235
3236 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
efe120a0 3237 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
606686ee 3238 rcu_str_deref(device->name));
387125fc 3239 device->nobarriers = 1;
5af3e8cc 3240 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3241 ret = -EIO;
5af3e8cc
SB		 3242 btrfs_dev_stat_inc_and_print(device,
3243 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3244 }
3245
3246 /* drop the reference from the wait == 0 run */
3247 bio_put(bio);
3248 device->flush_bio = NULL;
3249
3250 return ret;
3251 }
3252
3253 /*
3254 * one reference for us, and we leave it for the
3255 * caller
3256 */
9c017abc 3257 device->flush_bio = NULL;
9be3395b 3258 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
387125fc
CM		 3259 if (!bio)
3260 return -ENOMEM;
3261
3262 bio->bi_end_io = btrfs_end_empty_barrier;
3263 bio->bi_bdev = device->bdev;
3264 init_completion(&device->flush_wait);
3265 bio->bi_private = &device->flush_wait;
3266 device->flush_bio = bio;
3267
3268 bio_get(bio);
21adbd5c 3269 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3270
3271 return 0;
3272}
3273
3274/*
3275 * send an empty flush down to each device in parallel,
3276 * then wait for them
3277 */
3278static int barrier_all_devices(struct btrfs_fs_info *info)
3279{
3280 struct list_head *head;
3281 struct btrfs_device *dev;
5af3e8cc
SB		 3282 int errors_send = 0;
3283 int errors_wait = 0;
387125fc
CM		 3284 int ret;
3285
3286 /* send down all the barriers */
3287 head = &info->fs_devices->devices;
3288 list_for_each_entry_rcu(dev, head, dev_list) {
f88ba6a2
HS		 3289 if (dev->missing)
3290 continue;
387125fc 3291 if (!dev->bdev) {
5af3e8cc 3292 errors_send++;
387125fc
CM		 3293 continue;
3294 }
3295 if (!dev->in_fs_metadata || !dev->writeable)
3296 continue;
3297
3298 ret = write_dev_flush(dev, 0);
3299 if (ret)
5af3e8cc 3300 errors_send++;
387125fc
CM		 3301 }
3302
3303 /* wait for all the barriers */
3304 list_for_each_entry_rcu(dev, head, dev_list) {
f88ba6a2
HS		 3305 if (dev->missing)
3306 continue;
387125fc 3307 if (!dev->bdev) {
5af3e8cc 3308 errors_wait++;
387125fc
CM		 3309 continue;
3310 }
3311 if (!dev->in_fs_metadata || !dev->writeable)
3312 continue;
3313
3314 ret = write_dev_flush(dev, 1);
3315 if (ret)
5af3e8cc 3316 errors_wait++;
387125fc 3317 }
5af3e8cc
SB		 3318 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3319 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3320 return -EIO;
3321 return 0;
3322}
3323
5af3e8cc
SB		 3324int btrfs_calc_num_tolerated_disk_barrier_failures(
3325 struct btrfs_fs_info *fs_info)
3326{
3327 struct btrfs_ioctl_space_info space;
3328 struct btrfs_space_info *sinfo;
3329 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3330 BTRFS_BLOCK_GROUP_SYSTEM,
3331 BTRFS_BLOCK_GROUP_METADATA,
3332 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3333 int num_types = 4;
3334 int i;
3335 int c;
3336 int num_tolerated_disk_barrier_failures =
3337 (int)fs_info->fs_devices->num_devices;
3338
3339 for (i = 0; i < num_types; i++) {
3340 struct btrfs_space_info *tmp;
3341
3342 sinfo = NULL;
3343 rcu_read_lock();
3344 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3345 if (tmp->flags == types[i]) {
3346 sinfo = tmp;
3347 break;
3348 }
3349 }
3350 rcu_read_unlock();
3351
3352 if (!sinfo)
3353 continue;
3354
3355 down_read(&sinfo->groups_sem);
3356 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3357 if (!list_empty(&sinfo->block_groups[c])) {
3358 u64 flags;
3359
3360 btrfs_get_block_group_info(
3361 &sinfo->block_groups[c], &space);
3362 if (space.total_bytes == 0 ||
3363 space.used_bytes == 0)
3364 continue;
3365 flags = space.flags;
3366 /*
3367 * return
3368 * 0: if dup, single or RAID0 is configured for
3369 * any of metadata, system or data, else
3370 * 1: if RAID5 is configured, or if RAID1 or
3371 * RAID10 is configured and only two mirrors
3372 * are used, else
3373 * 2: if RAID6 is configured, else
3374 * num_mirrors - 1: if RAID1 or RAID10 is
3375 * configured and more than
3376 * 2 mirrors are used.
3377 */
3378 if (num_tolerated_disk_barrier_failures > 0 &&
3379 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3380 BTRFS_BLOCK_GROUP_RAID0)) ||
3381 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3382 == 0)))
3383 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3384 else if (num_tolerated_disk_barrier_failures > 1) {
3385 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3386 BTRFS_BLOCK_GROUP_RAID5 |
3387 BTRFS_BLOCK_GROUP_RAID10)) {
3388 num_tolerated_disk_barrier_failures = 1;
3389 } else if (flags &
15b0a89d 3390 BTRFS_BLOCK_GROUP_RAID6) {
53b381b3
DW		 3391 num_tolerated_disk_barrier_failures = 2;
3392 }
3393 }
5af3e8cc
SB		 3394 }
3395 }
3396 up_read(&sinfo->groups_sem);
3397 }
3398
3399 return num_tolerated_disk_barrier_failures;
3400}
3401
48a3b636 3402static int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3403{
e5e9a520 3404 struct list_head *head;
f2984462 3405 struct btrfs_device *dev;
a061fc8d 3406 struct btrfs_super_block *sb;
f2984462 3407 struct btrfs_dev_item *dev_item;
f2984462
CM		 3408 int ret;
3409 int do_barriers;
a236aed1
CM		 3410 int max_errors;
3411 int total_errors = 0;
a061fc8d 3412 u64 flags;
f2984462
CM		 3413
3414 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3415 backup_super_roots(root->fs_info);
f2984462 3416
6c41761f 3417 sb = root->fs_info->super_for_commit;
a061fc8d 3418 dev_item = &sb->dev_item;
e5e9a520 3419
174ba509 3420 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3421 head = &root->fs_info->fs_devices->devices;
d7306801 3422 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
387125fc 3423
5af3e8cc
SB		 3424 if (do_barriers) {
3425 ret = barrier_all_devices(root->fs_info);
3426 if (ret) {
3427 mutex_unlock(
3428 &root->fs_info->fs_devices->device_list_mutex);
3429 btrfs_error(root->fs_info, ret,
3430 "errors while submitting device barriers.");
3431 return ret;
3432 }
3433 }
387125fc 3434
1f78160c 3435 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3436 if (!dev->bdev) {
3437 total_errors++;
3438 continue;
3439 }
2b82032c 3440 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3441 continue;
3442
2b82032c 3443 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3444 btrfs_set_stack_device_type(dev_item, dev->type);
3445 btrfs_set_stack_device_id(dev_item, dev->devid);
3446 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3447 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3448 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3449 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3450 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3451 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3452 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3453
a061fc8d
CM		 3454 flags = btrfs_super_flags(sb);
3455 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3456
a512bbf8 3457 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3458 if (ret)
3459 total_errors++;
f2984462 3460 }
a236aed1 3461 if (total_errors > max_errors) {
efe120a0 3462 btrfs_err(root->fs_info, "%d errors while writing supers",
d397712b 3463 total_errors);
a724b436 3464 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
79787eaa 3465
9d565ba4
SB		 3466 /* FUA is masked off if unsupported and can't be the reason */
3467 btrfs_error(root->fs_info, -EIO,
3468 "%d errors while writing supers", total_errors);
3469 return -EIO;
a236aed1 3470 }
f2984462 3471
a512bbf8 3472 total_errors = 0;
1f78160c 3473 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3474 if (!dev->bdev)
3475 continue;
2b82032c 3476 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3477 continue;
3478
a512bbf8
YZ		 3479 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3480 if (ret)
3481 total_errors++;
f2984462 3482 }
174ba509 3483 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3484 if (total_errors > max_errors) {
79787eaa
JM		 3485 btrfs_error(root->fs_info, -EIO,
3486 "%d errors while writing supers", total_errors);
3487 return -EIO;
a236aed1 3488 }
f2984462
CM		 3489 return 0;
3490}
3491
a512bbf8
YZ		 3492int write_ctree_super(struct btrfs_trans_handle *trans,
3493 struct btrfs_root *root, int max_mirrors)
eb60ceac 3494{
f570e757 3495 return write_all_supers(root, max_mirrors);
cfaa7295
CM		 3496}
3497
cb517eab
MX		 3498/* Drop a fs root from the radix tree and free it. */
3499void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3500 struct btrfs_root *root)
2619ba1f 3501{
4df27c4d 3502 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3503 radix_tree_delete(&fs_info->fs_roots_radix,
3504 (unsigned long)root->root_key.objectid);
4df27c4d 3505 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3506
3507 if (btrfs_root_refs(&root->root_item) == 0)
3508 synchronize_srcu(&fs_info->subvol_srcu);
3509
1a4319cc 3510 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3321719e 3511 btrfs_free_log(NULL, root);
3321719e 3512
faa2dbf0
JB		 3513 if (root->free_ino_pinned)
3514 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3515 if (root->free_ino_ctl)
3516 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3517 free_fs_root(root);
4df27c4d
YZ		 3518}
3519
3520static void free_fs_root(struct btrfs_root *root)
3521{
82d5902d 3522 iput(root->cache_inode);
4df27c4d 3523 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1cb048f5
FDBM		 3524 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3525 root->orphan_block_rsv = NULL;
0ee5dc67
AV		 3526 if (root->anon_dev)
3527 free_anon_bdev(root->anon_dev);
8257b2dc
MX		 3528 if (root->subv_writers)
3529 btrfs_free_subvolume_writers(root->subv_writers);
4df27c4d
YZ		 3530 free_extent_buffer(root->node);
3531 free_extent_buffer(root->commit_root);
581bb050
LZ		 3532 kfree(root->free_ino_ctl);
3533 kfree(root->free_ino_pinned);
d397712b 3534 kfree(root->name);
b0feb9d9 3535 btrfs_put_fs_root(root);
2619ba1f
CM		 3536}
3537
cb517eab
MX		 3538void btrfs_free_fs_root(struct btrfs_root *root)
3539{
3540 free_fs_root(root);
2619ba1f
CM		 3541}
3542
c146afad 3543int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3544{
c146afad
YZ		 3545 u64 root_objectid = 0;
3546 struct btrfs_root *gang[8];
65d33fd7
QW		 3547 int i = 0;
3548 int err = 0;
3549 unsigned int ret = 0;
3550 int index;
e089f05c 3551
c146afad 3552 while (1) {
65d33fd7 3553 index = srcu_read_lock(&fs_info->subvol_srcu);
c146afad
YZ		 3554 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3555 (void **)gang, root_objectid,
3556 ARRAY_SIZE(gang));
65d33fd7
QW		 3557 if (!ret) {
3558 srcu_read_unlock(&fs_info->subvol_srcu, index);
c146afad 3559 break;
65d33fd7 3560 }
5d4f98a2 3561 root_objectid = gang[ret - 1]->root_key.objectid + 1;
65d33fd7 3562
c146afad 3563 for (i = 0; i < ret; i++) {
65d33fd7
QW		 3564 /* Avoid to grab roots in dead_roots */
3565 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3566 gang[i] = NULL;
3567 continue;
3568 }
3569 /* grab all the search result for later use */
3570 gang[i] = btrfs_grab_fs_root(gang[i]);
3571 }
3572 srcu_read_unlock(&fs_info->subvol_srcu, index);
66b4ffd1 3573
65d33fd7
QW		 3574 for (i = 0; i < ret; i++) {
3575 if (!gang[i])
3576 continue;
c146afad 3577 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3578 err = btrfs_orphan_cleanup(gang[i]);
3579 if (err)
65d33fd7
QW		 3580 break;
3581 btrfs_put_fs_root(gang[i]);
c146afad
YZ		 3582 }
3583 root_objectid++;
3584 }
65d33fd7
QW		 3585
3586 /* release the uncleaned roots due to error */
3587 for (; i < ret; i++) {
3588 if (gang[i])
3589 btrfs_put_fs_root(gang[i]);
3590 }
3591 return err;
c146afad 3592}
a2135011 3593
c146afad
YZ		 3594int btrfs_commit_super(struct btrfs_root *root)
3595{
3596 struct btrfs_trans_handle *trans;
a74a4b97 3597
c146afad 3598 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3599 btrfs_run_delayed_iputs(root);
c146afad 3600 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3601 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3602
3603 /* wait until ongoing cleanup work done */
3604 down_write(&root->fs_info->cleanup_work_sem);
3605 up_write(&root->fs_info->cleanup_work_sem);
3606
7a7eaa40 3607 trans = btrfs_join_transaction(root);
3612b495
TI		 3608 if (IS_ERR(trans))
3609 return PTR_ERR(trans);
d52c1bcc 3610 return btrfs_commit_transaction(trans, root);
c146afad
YZ		 3611}
3612
3613int close_ctree(struct btrfs_root *root)
3614{
3615 struct btrfs_fs_info *fs_info = root->fs_info;
3616 int ret;
3617
3618 fs_info->closing = 1;
3619 smp_mb();
3620
803b2f54
SB		 3621 /* wait for the uuid_scan task to finish */
3622 down(&fs_info->uuid_tree_rescan_sem);
3623 /* avoid complains from lockdep et al., set sem back to initial state */
3624 up(&fs_info->uuid_tree_rescan_sem);
3625
837d5b6e 3626 /* pause restriper - we want to resume on mount */
aa1b8cd4 3627 btrfs_pause_balance(fs_info);
837d5b6e 3628
8dabb742
SB		 3629 btrfs_dev_replace_suspend_for_unmount(fs_info);
3630
aa1b8cd4 3631 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3632
3633 /* wait for any defraggers to finish */
3634 wait_event(fs_info->transaction_wait,
3635 (atomic_read(&fs_info->defrag_running) == 0));
3636
3637 /* clear out the rbtree of defraggable inodes */
26176e7c 3638 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3639
21c7e756
MX		 3640 cancel_work_sync(&fs_info->async_reclaim_work);
3641
c146afad 3642 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3643 ret = btrfs_commit_super(root);
3644 if (ret)
efe120a0 3645 btrfs_err(root->fs_info, "commit super ret %d", ret);
acce952b 3646 }
3647
87533c47 3648 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3649 btrfs_error_commit_super(root);
0f7d52f4 3650
e3029d9f
AV		 3651 kthread_stop(fs_info->transaction_kthread);
3652 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3653
f25784b3
YZ		 3654 fs_info->closing = 2;
3655 smp_mb();
3656
bcef60f2
AJ		 3657 btrfs_free_qgroup_config(root->fs_info);
3658
963d678b 3659 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
efe120a0 3660 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
963d678b 3661 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3662 }
bcc63abb 3663
5ac1d209
JM		 3664 btrfs_sysfs_remove_one(fs_info);
3665
faa2dbf0 3666 btrfs_free_fs_roots(fs_info);
d10c5f31 3667
1a4319cc
LB		 3668 btrfs_put_block_group_cache(fs_info);
3669
2b1360da
JB		 3670 btrfs_free_block_groups(fs_info);
3671
de348ee0
WS		 3672 /*
3673 * we must make sure there is not any read request to
3674 * submit after we stopping all workers.
3675 */
3676 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
96192499
JB		 3677 btrfs_stop_all_workers(fs_info);
3678
13e6c37b 3679 free_root_pointers(fs_info, 1);
9ad6b7bc 3680
13e6c37b 3681 iput(fs_info->btree_inode);
d6bfde87 3682
21adbd5c
SB		 3683#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3684 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3685 btrfsic_unmount(root, fs_info->fs_devices);
3686#endif
3687
dfe25020 3688 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3689 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3690
e2d84521 3691 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3692 percpu_counter_destroy(&fs_info->delalloc_bytes);
c404e0dc 3693 percpu_counter_destroy(&fs_info->bio_counter);
04160088 3694 bdi_destroy(&fs_info->bdi);
76dda93c 3695 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3696
53b381b3
DW		 3697 btrfs_free_stripe_hash_table(fs_info);
3698
1cb048f5
FDBM		 3699 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3700 root->orphan_block_rsv = NULL;
3701
eb60ceac
CM		 3702 return 0;
3703}
3704
b9fab919
CM		 3705int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3706 int atomic)
5f39d397 3707{
1259ab75 3708 int ret;
727011e0 3709 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3710
0b32f4bb 3711 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3712 if (!ret)
3713 return ret;
3714
3715 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3716 parent_transid, atomic);
3717 if (ret == -EAGAIN)
3718 return ret;
1259ab75 3719 return !ret;
5f39d397
CM		 3720}
3721
3722int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3723{
0b32f4bb 3724 return set_extent_buffer_uptodate(buf);
5f39d397 3725}
6702ed49 3726
5f39d397
CM		 3727void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3728{
06ea65a3 3729 struct btrfs_root *root;
5f39d397 3730 u64 transid = btrfs_header_generation(buf);
b9473439 3731 int was_dirty;
b4ce94de 3732
06ea65a3
JB		 3733#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3734 /*
3735 * This is a fast path so only do this check if we have sanity tests
3736 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3737 * outside of the sanity tests.
3738 */
3739 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3740 return;
3741#endif
3742 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
b9447ef8 3743 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3744 if (transid != root->fs_info->generation)
3745 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3746 "found %llu running %llu\n",
c1c9ff7c 3747 buf->start, transid, root->fs_info->generation);
0b32f4bb 3748 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3749 if (!was_dirty)
3750 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3751 buf->len,
3752 root->fs_info->dirty_metadata_batch);
1f21ef0a
FM		 3753#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3754 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3755 btrfs_print_leaf(root, buf);
3756 ASSERT(0);
3757 }
3758#endif
eb60ceac
CM		 3759}
3760
b53d3f5d
LB		 3761static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3762 int flush_delayed)
16cdcec7
MX		 3763{
3764 /*
3765 * looks as though older kernels can get into trouble with
3766 * this code, they end up stuck in balance_dirty_pages forever
3767 */
e2d84521 3768 int ret;
16cdcec7
MX		 3769
3770 if (current->flags & PF_MEMALLOC)
3771 return;
3772
b53d3f5d
LB		 3773 if (flush_delayed)
3774 btrfs_balance_delayed_items(root);
16cdcec7 3775
e2d84521
MX		 3776 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3777 BTRFS_DIRTY_METADATA_THRESH);
3778 if (ret > 0) {
d0e1d66b
NJ		 3779 balance_dirty_pages_ratelimited(
3780 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3781 }
3782 return;
3783}
3784
b53d3f5d 3785void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3786{
b53d3f5d
LB		 3787 __btrfs_btree_balance_dirty(root, 1);
3788}
585ad2c3 3789
b53d3f5d
LB		 3790void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3791{
3792 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3793}
6b80053d 3794
ca7a79ad 3795int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3796{
727011e0 3797 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3798 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3799}
0da5468f 3800
fcd1f065 3801static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3802 int read_only)
3803{
1104a885
DS		 3804 /*
3805 * Placeholder for checks
3806 */
fcd1f065 3807 return 0;
acce952b 3808}
3809
48a3b636 3810static void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3811{
acce952b 3812 mutex_lock(&root->fs_info->cleaner_mutex);
3813 btrfs_run_delayed_iputs(root);
3814 mutex_unlock(&root->fs_info->cleaner_mutex);
3815
3816 down_write(&root->fs_info->cleanup_work_sem);
3817 up_write(&root->fs_info->cleanup_work_sem);
3818
3819 /* cleanup FS via transaction */
3820 btrfs_cleanup_transaction(root);
acce952b 3821}
3822
569e0f35
JB		 3823static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3824 struct btrfs_root *root)
acce952b 3825{
3826 struct btrfs_inode *btrfs_inode;
3827 struct list_head splice;
3828
3829 INIT_LIST_HEAD(&splice);
3830
3831 mutex_lock(&root->fs_info->ordered_operations_mutex);
199c2a9c 3832 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3833
569e0f35 3834 list_splice_init(&t->ordered_operations, &splice);
acce952b 3835 while (!list_empty(&splice)) {
3836 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3837 ordered_operations);
3838
3839 list_del_init(&btrfs_inode->ordered_operations);
199c2a9c 3840 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3841
3842 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3843
199c2a9c 3844 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3845 }
3846
199c2a9c 3847 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3848 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3849}
3850
143bede5 3851static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3852{
acce952b 3853 struct btrfs_ordered_extent *ordered;
acce952b 3854
199c2a9c 3855 spin_lock(&root->ordered_extent_lock);
779880ef
JB		 3856 /*
3857 * This will just short circuit the ordered completion stuff which will
3858 * make sure the ordered extent gets properly cleaned up.
3859 */
199c2a9c 3860 list_for_each_entry(ordered, &root->ordered_extents,
779880ef
JB		 3861 root_extent_list)
3862 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
199c2a9c
MX		 3863 spin_unlock(&root->ordered_extent_lock);
3864}
3865
3866static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3867{
3868 struct btrfs_root *root;
3869 struct list_head splice;
3870
3871 INIT_LIST_HEAD(&splice);
3872
3873 spin_lock(&fs_info->ordered_root_lock);
3874 list_splice_init(&fs_info->ordered_roots, &splice);
3875 while (!list_empty(&splice)) {
3876 root = list_first_entry(&splice, struct btrfs_root,
3877 ordered_root);
1de2cfde
JB		 3878 list_move_tail(&root->ordered_root,
3879 &fs_info->ordered_roots);
199c2a9c 3880
2a85d9ca 3881 spin_unlock(&fs_info->ordered_root_lock);
199c2a9c
MX		 3882 btrfs_destroy_ordered_extents(root);
3883
2a85d9ca
LB		 3884 cond_resched();
3885 spin_lock(&fs_info->ordered_root_lock);
199c2a9c
MX		 3886 }
3887 spin_unlock(&fs_info->ordered_root_lock);
acce952b 3888}
3889
35a3621b
SB		 3890static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3891 struct btrfs_root *root)
acce952b 3892{
3893 struct rb_node *node;
3894 struct btrfs_delayed_ref_root *delayed_refs;
3895 struct btrfs_delayed_ref_node *ref;
3896 int ret = 0;
3897
3898 delayed_refs = &trans->delayed_refs;
3899
3900 spin_lock(&delayed_refs->lock);
d7df2c79 3901 if (atomic_read(&delayed_refs->num_entries) == 0) {
cfece4db 3902 spin_unlock(&delayed_refs->lock);
efe120a0 3903 btrfs_info(root->fs_info, "delayed_refs has NO entry");
acce952b 3904 return ret;
3905 }
3906
d7df2c79
JB		 3907 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3908 struct btrfs_delayed_ref_head *head;
e78417d1 3909 bool pin_bytes = false;
acce952b 3910
d7df2c79
JB		 3911 head = rb_entry(node, struct btrfs_delayed_ref_head,
3912 href_node);
3913 if (!mutex_trylock(&head->mutex)) {
3914 atomic_inc(&head->node.refs);
3915 spin_unlock(&delayed_refs->lock);
eb12db69 3916
d7df2c79 3917 mutex_lock(&head->mutex);
e78417d1 3918 mutex_unlock(&head->mutex);
d7df2c79
JB		 3919 btrfs_put_delayed_ref(&head->node);
3920 spin_lock(&delayed_refs->lock);
3921 continue;
3922 }
3923 spin_lock(&head->lock);
3924 while ((node = rb_first(&head->ref_root)) != NULL) {
3925 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3926 rb_node);
3927 ref->in_tree = 0;
3928 rb_erase(&ref->rb_node, &head->ref_root);
3929 atomic_dec(&delayed_refs->num_entries);
3930 btrfs_put_delayed_ref(ref);
e78417d1 3931 }
d7df2c79
JB		 3932 if (head->must_insert_reserved)
3933 pin_bytes = true;
3934 btrfs_free_delayed_extent_op(head->extent_op);
3935 delayed_refs->num_heads--;
3936 if (head->processing == 0)
3937 delayed_refs->num_heads_ready--;
3938 atomic_dec(&delayed_refs->num_entries);
3939 head->node.in_tree = 0;
3940 rb_erase(&head->href_node, &delayed_refs->href_root);
3941 spin_unlock(&head->lock);
3942 spin_unlock(&delayed_refs->lock);
3943 mutex_unlock(&head->mutex);
acce952b 3944
d7df2c79
JB		 3945 if (pin_bytes)
3946 btrfs_pin_extent(root, head->node.bytenr,
3947 head->node.num_bytes, 1);
3948 btrfs_put_delayed_ref(&head->node);
acce952b 3949 cond_resched();
3950 spin_lock(&delayed_refs->lock);
3951 }
3952
3953 spin_unlock(&delayed_refs->lock);
3954
3955 return ret;
3956}
3957
143bede5 3958static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3959{
3960 struct btrfs_inode *btrfs_inode;
3961 struct list_head splice;
3962
3963 INIT_LIST_HEAD(&splice);
3964
eb73c1b7
MX		 3965 spin_lock(&root->delalloc_lock);
3966 list_splice_init(&root->delalloc_inodes, &splice);
acce952b 3967
3968 while (!list_empty(&splice)) {
eb73c1b7
MX		 3969 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3970 delalloc_inodes);
acce952b 3971
3972 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3973 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3974 &btrfs_inode->runtime_flags);
eb73c1b7 3975 spin_unlock(&root->delalloc_lock);
acce952b 3976
3977 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3978
eb73c1b7 3979 spin_lock(&root->delalloc_lock);
acce952b 3980 }
3981
eb73c1b7
MX		 3982 spin_unlock(&root->delalloc_lock);
3983}
3984
3985static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3986{
3987 struct btrfs_root *root;
3988 struct list_head splice;
3989
3990 INIT_LIST_HEAD(&splice);
3991
3992 spin_lock(&fs_info->delalloc_root_lock);
3993 list_splice_init(&fs_info->delalloc_roots, &splice);
3994 while (!list_empty(&splice)) {
3995 root = list_first_entry(&splice, struct btrfs_root,
3996 delalloc_root);
3997 list_del_init(&root->delalloc_root);
3998 root = btrfs_grab_fs_root(root);
3999 BUG_ON(!root);
4000 spin_unlock(&fs_info->delalloc_root_lock);
4001
4002 btrfs_destroy_delalloc_inodes(root);
4003 btrfs_put_fs_root(root);
4004
4005 spin_lock(&fs_info->delalloc_root_lock);
4006 }
4007 spin_unlock(&fs_info->delalloc_root_lock);
acce952b 4008}
4009
4010static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4011 struct extent_io_tree *dirty_pages,
4012 int mark)
4013{
4014 int ret;
acce952b 4015 struct extent_buffer *eb;
4016 u64 start = 0;
4017 u64 end;
acce952b 4018
4019 while (1) {
4020 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 4021 mark, NULL);
acce952b 4022 if (ret)
4023 break;
4024
4025 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4026 while (start <= end) {
fd8b2b61
JB		 4027 eb = btrfs_find_tree_block(root, start,
4028 root->leafsize);
69a85bd8 4029 start += root->leafsize;
fd8b2b61 4030 if (!eb)
acce952b 4031 continue;
fd8b2b61 4032 wait_on_extent_buffer_writeback(eb);
acce952b 4033
fd8b2b61
JB		 4034 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4035 &eb->bflags))
4036 clear_extent_buffer_dirty(eb);
4037 free_extent_buffer_stale(eb);
acce952b 4038 }
4039 }
4040
4041 return ret;
4042}
4043
4044static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4045 struct extent_io_tree *pinned_extents)
4046{
4047 struct extent_io_tree *unpin;
4048 u64 start;
4049 u64 end;
4050 int ret;
ed0eaa14 4051 bool loop = true;
acce952b 4052
4053 unpin = pinned_extents;
ed0eaa14 4054again:
acce952b 4055 while (1) {
4056 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 4057 EXTENT_DIRTY, NULL);
acce952b 4058 if (ret)
4059 break;
4060
4061 /* opt_discard */
5378e607
LD		 4062 if (btrfs_test_opt(root, DISCARD))
4063 ret = btrfs_error_discard_extent(root, start,
4064 end + 1 - start,
4065 NULL);
acce952b 4066
4067 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4068 btrfs_error_unpin_extent_range(root, start, end);
4069 cond_resched();
4070 }
4071
ed0eaa14
LB		 4072 if (loop) {
4073 if (unpin == &root->fs_info->freed_extents[0])
4074 unpin = &root->fs_info->freed_extents[1];
4075 else
4076 unpin = &root->fs_info->freed_extents[0];
4077 loop = false;
4078 goto again;
4079 }
4080
acce952b 4081 return 0;
4082}
4083
49b25e05
JM		 4084void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4085 struct btrfs_root *root)
4086{
724e2315
JB		 4087 btrfs_destroy_ordered_operations(cur_trans, root);
4088
49b25e05 4089 btrfs_destroy_delayed_refs(cur_trans, root);
49b25e05 4090
4a9d8bde 4091 cur_trans->state = TRANS_STATE_COMMIT_START;
d7096fc3 4092 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 4093
4a9d8bde 4094 cur_trans->state = TRANS_STATE_UNBLOCKED;
d7096fc3 4095 wake_up(&root->fs_info->transaction_wait);
49b25e05 4096
67cde344
MX		 4097 btrfs_destroy_delayed_inodes(root);
4098 btrfs_assert_delayed_root_empty(root);
49b25e05 4099
49b25e05
JM		 4100 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4101 EXTENT_DIRTY);
6e841e32
LB		 4102 btrfs_destroy_pinned_extent(root,
4103 root->fs_info->pinned_extents);
49b25e05 4104
4a9d8bde
MX		 4105 cur_trans->state =TRANS_STATE_COMPLETED;
4106 wake_up(&cur_trans->commit_wait);
4107
49b25e05
JM		 4108 /*
4109 memset(cur_trans, 0, sizeof(*cur_trans));
4110 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4111 */
4112}
4113
48a3b636 4114static int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 4115{
4116 struct btrfs_transaction *t;
acce952b 4117
acce952b 4118 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4119
a4abeea4 4120 spin_lock(&root->fs_info->trans_lock);
724e2315
JB		 4121 while (!list_empty(&root->fs_info->trans_list)) {
4122 t = list_first_entry(&root->fs_info->trans_list,
4123 struct btrfs_transaction, list);
4124 if (t->state >= TRANS_STATE_COMMIT_START) {
4125 atomic_inc(&t->use_count);
4126 spin_unlock(&root->fs_info->trans_lock);
4127 btrfs_wait_for_commit(root, t->transid);
4128 btrfs_put_transaction(t);
4129 spin_lock(&root->fs_info->trans_lock);
4130 continue;
4131 }
4132 if (t == root->fs_info->running_transaction) {
4133 t->state = TRANS_STATE_COMMIT_DOING;
4134 spin_unlock(&root->fs_info->trans_lock);
4135 /*
4136 * We wait for 0 num_writers since we don't hold a trans
4137 * handle open currently for this transaction.
4138 */
4139 wait_event(t->writer_wait,
4140 atomic_read(&t->num_writers) == 0);
4141 } else {
4142 spin_unlock(&root->fs_info->trans_lock);
4143 }
4144 btrfs_cleanup_one_transaction(t, root);
4a9d8bde 4145
724e2315
JB		 4146 spin_lock(&root->fs_info->trans_lock);
4147 if (t == root->fs_info->running_transaction)
4148 root->fs_info->running_transaction = NULL;
acce952b 4149 list_del_init(&t->list);
724e2315 4150 spin_unlock(&root->fs_info->trans_lock);
acce952b 4151
724e2315
JB		 4152 btrfs_put_transaction(t);
4153 trace_btrfs_transaction_commit(root);
4154 spin_lock(&root->fs_info->trans_lock);
4155 }
4156 spin_unlock(&root->fs_info->trans_lock);
4157 btrfs_destroy_all_ordered_extents(root->fs_info);
4158 btrfs_destroy_delayed_inodes(root);
4159 btrfs_assert_delayed_root_empty(root);
4160 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4161 btrfs_destroy_all_delalloc_inodes(root->fs_info);
acce952b 4162 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 4163
4164 return 0;
4165}
4166
d1310b2e 4167static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 4168 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 4169 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 4170 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 4171 /* note we're sharing with inode.c for the merge bio hook */
4172 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 4173};
Linux 6.x block layer and io_uring tree(s)