projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: delete unused parameter to btrfs_read_root_item()
[linux-2.6-block.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
163e783e 29#include <linux/crc32c.h>
5a0e3ad6 30#include <linux/slab.h>
784b4e29 31#include <linux/migrate.h>
7a36ddec 32#include <linux/ratelimit.h>
7e75bf3f 33#include <asm/unaligned.h>
4b4e25f2 34#include "compat.h"
eb60ceac
CM		 35#include "ctree.h"
36#include "disk-io.h"
e089f05c 37#include "transaction.h"
0f7d52f4 38#include "btrfs_inode.h"
0b86a832 39#include "volumes.h"
db94535d 40#include "print-tree.h"
8b712842 41#include "async-thread.h"
925baedd 42#include "locking.h"
e02119d5 43#include "tree-log.h"
fa9c0d79 44#include "free-space-cache.h"
581bb050 45#include "inode-map.h"
21adbd5c 46#include "check-integrity.h"
606686ee 47#include "rcu-string.h"
8dabb742 48#include "dev-replace.h"
53b381b3 49#include "raid56.h"
eb60ceac 50
de0022b9
JB		 51#ifdef CONFIG_X86
52#include <asm/cpufeature.h>
53#endif
54
d1310b2e 55static struct extent_io_ops btree_extent_io_ops;
8b712842 56static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 57static void free_fs_root(struct btrfs_root *root);
fcd1f065 58static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 59 int read_only);
569e0f35
JB		 60static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
61 struct btrfs_root *root);
143bede5 62static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
acce952b 63static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
64 struct btrfs_root *root);
aec8030a 65static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
143bede5 66static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
acce952b 67static int btrfs_destroy_marked_extents(struct btrfs_root *root,
68 struct extent_io_tree *dirty_pages,
69 int mark);
70static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
71 struct extent_io_tree *pinned_extents);
ce9adaa5 72
d352ac68
CM		 73/*
74 * end_io_wq structs are used to do processing in task context when an IO is
75 * complete. This is used during reads to verify checksums, and it is used
76 * by writes to insert metadata for new file extents after IO is complete.
77 */
ce9adaa5
CM		 78struct end_io_wq {
79 struct bio *bio;
80 bio_end_io_t *end_io;
81 void *private;
82 struct btrfs_fs_info *info;
83 int error;
22c59948 84 int metadata;
ce9adaa5 85 struct list_head list;
8b712842 86 struct btrfs_work work;
ce9adaa5 87};
0da5468f 88
d352ac68
CM		 89/*
90 * async submit bios are used to offload expensive checksumming
91 * onto the worker threads. They checksum file and metadata bios
92 * just before they are sent down the IO stack.
93 */
44b8bd7e
CM		 94struct async_submit_bio {
95 struct inode *inode;
96 struct bio *bio;
97 struct list_head list;
4a69a410
CM		 98 extent_submit_bio_hook_t *submit_bio_start;
99 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 100 int rw;
101 int mirror_num;
c8b97818 102 unsigned long bio_flags;
eaf25d93
CM		 103 /*
104 * bio_offset is optional, can be used if the pages in the bio
105 * can't tell us where in the file the bio should go
106 */
107 u64 bio_offset;
8b712842 108 struct btrfs_work work;
79787eaa 109 int error;
44b8bd7e
CM		 110};
111
85d4e461
CM		 112/*
113 * Lockdep class keys for extent_buffer->lock's in this root. For a given
114 * eb, the lockdep key is determined by the btrfs_root it belongs to and
115 * the level the eb occupies in the tree.
116 *
117 * Different roots are used for different purposes and may nest inside each
118 * other and they require separate keysets. As lockdep keys should be
119 * static, assign keysets according to the purpose of the root as indicated
120 * by btrfs_root->objectid. This ensures that all special purpose roots
121 * have separate keysets.
4008c04a 122 *
85d4e461
CM		 123 * Lock-nesting across peer nodes is always done with the immediate parent
124 * node locked thus preventing deadlock. As lockdep doesn't know this, use
125 * subclass to avoid triggering lockdep warning in such cases.
4008c04a 126 *
85d4e461
CM		 127 * The key is set by the readpage_end_io_hook after the buffer has passed
128 * csum validation but before the pages are unlocked. It is also set by
129 * btrfs_init_new_buffer on freshly allocated blocks.
4008c04a 130 *
85d4e461
CM		 131 * We also add a check to make sure the highest level of the tree is the
132 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
133 * needs update as well.
4008c04a
CM		 134 */
135#ifdef CONFIG_DEBUG_LOCK_ALLOC
136# if BTRFS_MAX_LEVEL != 8
137# error
138# endif
85d4e461
CM		 139
140static struct btrfs_lockdep_keyset {
141 u64 id; /* root objectid */
142 const char *name_stem; /* lock name stem */
143 char names[BTRFS_MAX_LEVEL + 1][20];
144 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
145} btrfs_lockdep_keysets[] = {
146 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
147 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
148 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
149 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
150 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
151 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
152 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
153 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
154 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
155 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
156 { .id = 0, .name_stem = "tree" },
4008c04a 157};
85d4e461
CM		 158
159void __init btrfs_init_lockdep(void)
160{
161 int i, j;
162
163 /* initialize lockdep class names */
164 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
165 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
166
167 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
168 snprintf(ks->names[j], sizeof(ks->names[j]),
169 "btrfs-%s-%02d", ks->name_stem, j);
170 }
171}
172
173void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
174 int level)
175{
176 struct btrfs_lockdep_keyset *ks;
177
178 BUG_ON(level >= ARRAY_SIZE(ks->keys));
179
180 /* find the matching keyset, id 0 is the default entry */
181 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
182 if (ks->id == objectid)
183 break;
184
185 lockdep_set_class_and_name(&eb->lock,
186 &ks->keys[level], ks->names[level]);
187}
188
4008c04a
CM		 189#endif
190
d352ac68
CM		 191/*
192 * extents on the btree inode are pretty simple, there's one extent
193 * that covers the entire device
194 */
b2950863 195static struct extent_map *btree_get_extent(struct inode *inode,
306e16ce 196 struct page *page, size_t pg_offset, u64 start, u64 len,
b2950863 197 int create)
7eccb903 198{
5f39d397
CM		 199 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
200 struct extent_map *em;
201 int ret;
202
890871be 203 read_lock(&em_tree->lock);
d1310b2e 204 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 205 if (em) {
206 em->bdev =
207 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 208 read_unlock(&em_tree->lock);
5f39d397 209 goto out;
a061fc8d 210 }
890871be 211 read_unlock(&em_tree->lock);
7b13b7b1 212
172ddd60 213 em = alloc_extent_map();
5f39d397
CM		 214 if (!em) {
215 em = ERR_PTR(-ENOMEM);
216 goto out;
217 }
218 em->start = 0;
0afbaf8c 219 em->len = (u64)-1;
c8b97818 220 em->block_len = (u64)-1;
5f39d397 221 em->block_start = 0;
a061fc8d 222 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 223
890871be 224 write_lock(&em_tree->lock);
09a2a8f9 225 ret = add_extent_mapping(em_tree, em, 0);
5f39d397
CM		 226 if (ret == -EEXIST) {
227 free_extent_map(em);
7b13b7b1 228 em = lookup_extent_mapping(em_tree, start, len);
b4f359ab 229 if (!em)
0433f20d 230 em = ERR_PTR(-EIO);
5f39d397 231 } else if (ret) {
7b13b7b1 232 free_extent_map(em);
0433f20d 233 em = ERR_PTR(ret);
5f39d397 234 }
890871be 235 write_unlock(&em_tree->lock);
7b13b7b1 236
5f39d397
CM		 237out:
238 return em;
7eccb903
CM		 239}
240
b0496686 241u32 btrfs_csum_data(char *data, u32 seed, size_t len)
19c00ddc 242{
163e783e 243 return crc32c(seed, data, len);
19c00ddc
CM		 244}
245
246void btrfs_csum_final(u32 crc, char *result)
247{
7e75bf3f 248 put_unaligned_le32(~crc, result);
19c00ddc
CM		 249}
250
d352ac68
CM		 251/*
252 * compute the csum for a btree block, and either verify it or write it
253 * into the csum field of the block.
254 */
19c00ddc
CM		 255static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
256 int verify)
257{
6c41761f 258 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
607d432d 259 char *result = NULL;
19c00ddc
CM		 260 unsigned long len;
261 unsigned long cur_len;
262 unsigned long offset = BTRFS_CSUM_SIZE;
19c00ddc
CM		 263 char *kaddr;
264 unsigned long map_start;
265 unsigned long map_len;
266 int err;
267 u32 crc = ~(u32)0;
607d432d 268 unsigned long inline_result;
19c00ddc
CM		 269
270 len = buf->len - offset;
d397712b 271 while (len > 0) {
19c00ddc 272 err = map_private_extent_buffer(buf, offset, 32,
a6591715 273 &kaddr, &map_start, &map_len);
d397712b 274 if (err)
19c00ddc 275 return 1;
19c00ddc 276 cur_len = min(len, map_len - (offset - map_start));
b0496686 277 crc = btrfs_csum_data(kaddr + offset - map_start,
19c00ddc
CM		 278 crc, cur_len);
279 len -= cur_len;
280 offset += cur_len;
19c00ddc 281 }
607d432d
JB		 282 if (csum_size > sizeof(inline_result)) {
283 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
284 if (!result)
285 return 1;
286 } else {
287 result = (char *)&inline_result;
288 }
289
19c00ddc
CM		 290 btrfs_csum_final(crc, result);
291
292 if (verify) {
607d432d 293 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 294 u32 val;
295 u32 found = 0;
607d432d 296 memcpy(&found, result, csum_size);
e4204ded 297
607d432d 298 read_extent_buffer(buf, &val, 0, csum_size);
7a36ddec 299 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
193f284d
CM		 300 "failed on %llu wanted %X found %X "
301 "level %d\n",
302 root->fs_info->sb->s_id,
303 (unsigned long long)buf->start, val, found,
304 btrfs_header_level(buf));
607d432d
JB		 305 if (result != (char *)&inline_result)
306 kfree(result);
19c00ddc
CM		 307 return 1;
308 }
309 } else {
607d432d 310 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 311 }
607d432d
JB		 312 if (result != (char *)&inline_result)
313 kfree(result);
19c00ddc
CM		 314 return 0;
315}
316
d352ac68
CM		 317/*
318 * we can't consider a given block up to date unless the transid of the
319 * block matches the transid in the parent node's pointer. This is how we
320 * detect blocks that either didn't get written at all or got written
321 * in the wrong place.
322 */
1259ab75 323static int verify_parent_transid(struct extent_io_tree *io_tree,
b9fab919
CM		 324 struct extent_buffer *eb, u64 parent_transid,
325 int atomic)
1259ab75 326{
2ac55d41 327 struct extent_state *cached_state = NULL;
1259ab75
CM		 328 int ret;
329
330 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
331 return 0;
332
b9fab919
CM		 333 if (atomic)
334 return -EAGAIN;
335
2ac55d41 336 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
d0082371 337 0, &cached_state);
0b32f4bb 338 if (extent_buffer_uptodate(eb) &&
1259ab75
CM		 339 btrfs_header_generation(eb) == parent_transid) {
340 ret = 0;
341 goto out;
342 }
7a36ddec 343 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
193f284d
CM		 344 "found %llu\n",
345 (unsigned long long)eb->start,
346 (unsigned long long)parent_transid,
347 (unsigned long long)btrfs_header_generation(eb));
1259ab75 348 ret = 1;
0b32f4bb 349 clear_extent_buffer_uptodate(eb);
33958dc6 350out:
2ac55d41
JB		 351 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
352 &cached_state, GFP_NOFS);
1259ab75 353 return ret;
1259ab75
CM		 354}
355
d352ac68
CM		 356/*
357 * helper to read a given tree block, doing retries as required when
358 * the checksums don't match and we have alternate mirrors to try.
359 */
f188591e
CM		 360static int btree_read_extent_buffer_pages(struct btrfs_root *root,
361 struct extent_buffer *eb,
ca7a79ad 362 u64 start, u64 parent_transid)
f188591e
CM		 363{
364 struct extent_io_tree *io_tree;
ea466794 365 int failed = 0;
f188591e
CM		 366 int ret;
367 int num_copies = 0;
368 int mirror_num = 0;
ea466794 369 int failed_mirror = 0;
f188591e 370
a826d6dc 371 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM		 372 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
373 while (1) {
bb82ab88
AJ		 374 ret = read_extent_buffer_pages(io_tree, eb, start,
375 WAIT_COMPLETE,
f188591e 376 btree_get_extent, mirror_num);
256dd1bb
SB		 377 if (!ret) {
378 if (!verify_parent_transid(io_tree, eb,
b9fab919 379 parent_transid, 0))
256dd1bb
SB		 380 break;
381 else
382 ret = -EIO;
383 }
d397712b 384
a826d6dc
JB		 385 /*
386 * This buffer's crc is fine, but its contents are corrupted, so
387 * there is no reason to read the other copies, they won't be
388 * any less wrong.
389 */
390 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
ea466794
JB		 391 break;
392
5d964051 393 num_copies = btrfs_num_copies(root->fs_info,
f188591e 394 eb->start, eb->len);
4235298e 395 if (num_copies == 1)
ea466794 396 break;
4235298e 397
5cf1ab56
JB		 398 if (!failed_mirror) {
399 failed = 1;
400 failed_mirror = eb->read_mirror;
401 }
402
f188591e 403 mirror_num++;
ea466794
JB		 404 if (mirror_num == failed_mirror)
405 mirror_num++;
406
4235298e 407 if (mirror_num > num_copies)
ea466794 408 break;
f188591e 409 }
ea466794 410
c0901581 411 if (failed && !ret && failed_mirror)
ea466794
JB		 412 repair_eb_io_failure(root, eb, failed_mirror);
413
414 return ret;
f188591e 415}
19c00ddc 416
d352ac68 417/*
d397712b
CM		 418 * checksum a dirty tree block before IO. This has extra checks to make sure
419 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 420 */
d397712b 421
b2950863 422static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 423{
d1310b2e 424 struct extent_io_tree *tree;
4eee4fa4 425 u64 start = page_offset(page);
19c00ddc 426 u64 found_start;
19c00ddc 427 struct extent_buffer *eb;
f188591e 428
d1310b2e 429 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc 430
4f2de97a
JB		 431 eb = (struct extent_buffer *)page->private;
432 if (page != eb->pages[0])
433 return 0;
19c00ddc
CM		 434 found_start = btrfs_header_bytenr(eb);
435 if (found_start != start) {
55c69072 436 WARN_ON(1);
4f2de97a 437 return 0;
55c69072 438 }
55c69072 439 if (!PageUptodate(page)) {
55c69072 440 WARN_ON(1);
4f2de97a 441 return 0;
19c00ddc 442 }
19c00ddc 443 csum_tree_block(root, eb, 0);
19c00ddc
CM		 444 return 0;
445}
446
2b82032c
YZ		 447static int check_tree_block_fsid(struct btrfs_root *root,
448 struct extent_buffer *eb)
449{
450 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
451 u8 fsid[BTRFS_UUID_SIZE];
452 int ret = 1;
453
454 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
455 BTRFS_FSID_SIZE);
456 while (fs_devices) {
457 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
458 ret = 0;
459 break;
460 }
461 fs_devices = fs_devices->seed;
462 }
463 return ret;
464}
465
a826d6dc
JB		 466#define CORRUPT(reason, eb, root, slot) \
467 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
468 "root=%llu, slot=%d\n", reason, \
469 (unsigned long long)btrfs_header_bytenr(eb), \
470 (unsigned long long)root->objectid, slot)
471
472static noinline int check_leaf(struct btrfs_root *root,
473 struct extent_buffer *leaf)
474{
475 struct btrfs_key key;
476 struct btrfs_key leaf_key;
477 u32 nritems = btrfs_header_nritems(leaf);
478 int slot;
479
480 if (nritems == 0)
481 return 0;
482
483 /* Check the 0 item */
484 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
485 BTRFS_LEAF_DATA_SIZE(root)) {
486 CORRUPT("invalid item offset size pair", leaf, root, 0);
487 return -EIO;
488 }
489
490 /*
491 * Check to make sure each items keys are in the correct order and their
492 * offsets make sense. We only have to loop through nritems-1 because
493 * we check the current slot against the next slot, which verifies the
494 * next slot's offset+size makes sense and that the current's slot
495 * offset is correct.
496 */
497 for (slot = 0; slot < nritems - 1; slot++) {
498 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
499 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
500
501 /* Make sure the keys are in the right order */
502 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
503 CORRUPT("bad key order", leaf, root, slot);
504 return -EIO;
505 }
506
507 /*
508 * Make sure the offset and ends are right, remember that the
509 * item data starts at the end of the leaf and grows towards the
510 * front.
511 */
512 if (btrfs_item_offset_nr(leaf, slot) !=
513 btrfs_item_end_nr(leaf, slot + 1)) {
514 CORRUPT("slot offset bad", leaf, root, slot);
515 return -EIO;
516 }
517
518 /*
519 * Check to make sure that we don't point outside of the leaf,
520 * just incase all the items are consistent to eachother, but
521 * all point outside of the leaf.
522 */
523 if (btrfs_item_end_nr(leaf, slot) >
524 BTRFS_LEAF_DATA_SIZE(root)) {
525 CORRUPT("slot end outside of leaf", leaf, root, slot);
526 return -EIO;
527 }
528 }
529
530 return 0;
531}
532
727011e0
CM		 533struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
534 struct page *page, int max_walk)
535{
536 struct extent_buffer *eb;
537 u64 start = page_offset(page);
538 u64 target = start;
539 u64 min_start;
540
541 if (start < max_walk)
542 min_start = 0;
543 else
544 min_start = start - max_walk;
545
546 while (start >= min_start) {
547 eb = find_extent_buffer(tree, start, 0);
548 if (eb) {
549 /*
550 * we found an extent buffer and it contains our page
551 * horray!
552 */
553 if (eb->start <= target &&
554 eb->start + eb->len > target)
555 return eb;
556
557 /* we found an extent buffer that wasn't for us */
558 free_extent_buffer(eb);
559 return NULL;
560 }
561 if (start == 0)
562 break;
563 start -= PAGE_CACHE_SIZE;
564 }
565 return NULL;
566}
567
b2950863 568static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
5cf1ab56 569 struct extent_state *state, int mirror)
ce9adaa5
CM		 570{
571 struct extent_io_tree *tree;
572 u64 found_start;
573 int found_level;
ce9adaa5
CM		 574 struct extent_buffer *eb;
575 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 576 int ret = 0;
727011e0 577 int reads_done;
ce9adaa5 578
ce9adaa5
CM		 579 if (!page->private)
580 goto out;
d397712b 581
727011e0 582 tree = &BTRFS_I(page->mapping->host)->io_tree;
4f2de97a 583 eb = (struct extent_buffer *)page->private;
d397712b 584
0b32f4bb
JB		 585 /* the pending IO might have been the only thing that kept this buffer
586 * in memory. Make sure we have a ref for all this other checks
587 */
588 extent_buffer_get(eb);
589
590 reads_done = atomic_dec_and_test(&eb->io_pages);
727011e0
CM		 591 if (!reads_done)
592 goto err;
f188591e 593
5cf1ab56 594 eb->read_mirror = mirror;
ea466794
JB		 595 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
596 ret = -EIO;
597 goto err;
598 }
599
ce9adaa5 600 found_start = btrfs_header_bytenr(eb);
727011e0 601 if (found_start != eb->start) {
7a36ddec 602 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
193f284d
CM		 603 "%llu %llu\n",
604 (unsigned long long)found_start,
605 (unsigned long long)eb->start);
f188591e 606 ret = -EIO;
ce9adaa5
CM		 607 goto err;
608 }
2b82032c 609 if (check_tree_block_fsid(root, eb)) {
7a36ddec 610 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
193f284d 611 (unsigned long long)eb->start);
1259ab75
CM		 612 ret = -EIO;
613 goto err;
614 }
ce9adaa5 615 found_level = btrfs_header_level(eb);
1c24c3ce
JB		 616 if (found_level >= BTRFS_MAX_LEVEL) {
617 btrfs_info(root->fs_info, "bad tree block level %d\n",
618 (int)btrfs_header_level(eb));
619 ret = -EIO;
620 goto err;
621 }
ce9adaa5 622
85d4e461
CM		 623 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
624 eb, found_level);
4008c04a 625
ce9adaa5 626 ret = csum_tree_block(root, eb, 1);
a826d6dc 627 if (ret) {
f188591e 628 ret = -EIO;
a826d6dc
JB		 629 goto err;
630 }
631
632 /*
633 * If this is a leaf block and it is corrupt, set the corrupt bit so
634 * that we don't try and read the other copies of this block, just
635 * return -EIO.
636 */
637 if (found_level == 0 && check_leaf(root, eb)) {
638 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
639 ret = -EIO;
640 }
ce9adaa5 641
0b32f4bb
JB		 642 if (!ret)
643 set_extent_buffer_uptodate(eb);
ce9adaa5 644err:
79fb65a1
JB		 645 if (reads_done &&
646 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 647 btree_readahead_hook(root, eb, eb->start, ret);
4bb31e92 648
53b381b3
DW		 649 if (ret) {
650 /*
651 * our io error hook is going to dec the io pages
652 * again, we have to make sure it has something
653 * to decrement
654 */
655 atomic_inc(&eb->io_pages);
0b32f4bb 656 clear_extent_buffer_uptodate(eb);
53b381b3 657 }
0b32f4bb 658 free_extent_buffer(eb);
ce9adaa5 659out:
f188591e 660 return ret;
ce9adaa5
CM		 661}
662
ea466794 663static int btree_io_failed_hook(struct page *page, int failed_mirror)
4bb31e92 664{
4bb31e92
AJ		 665 struct extent_buffer *eb;
666 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
667
4f2de97a 668 eb = (struct extent_buffer *)page->private;
ea466794 669 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
5cf1ab56 670 eb->read_mirror = failed_mirror;
53b381b3 671 atomic_dec(&eb->io_pages);
ea466794 672 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 673 btree_readahead_hook(root, eb, eb->start, -EIO);
4bb31e92
AJ		 674 return -EIO; /* we fixed nothing */
675}
676
ce9adaa5 677static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 678{
679 struct end_io_wq *end_io_wq = bio->bi_private;
680 struct btrfs_fs_info *fs_info;
ce9adaa5 681
ce9adaa5 682 fs_info = end_io_wq->info;
ce9adaa5 683 end_io_wq->error = err;
8b712842
CM		 684 end_io_wq->work.func = end_workqueue_fn;
685 end_io_wq->work.flags = 0;
d20f7043 686
7b6d91da 687 if (bio->bi_rw & REQ_WRITE) {
53b381b3 688 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
cad321ad
CM		 689 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
690 &end_io_wq->work);
53b381b3 691 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
0cb59c99
JB		 692 btrfs_queue_worker(&fs_info->endio_freespace_worker,
693 &end_io_wq->work);
53b381b3
DW		 694 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
695 btrfs_queue_worker(&fs_info->endio_raid56_workers,
696 &end_io_wq->work);
cad321ad
CM		 697 else
698 btrfs_queue_worker(&fs_info->endio_write_workers,
699 &end_io_wq->work);
d20f7043 700 } else {
53b381b3
DW		 701 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
702 btrfs_queue_worker(&fs_info->endio_raid56_workers,
703 &end_io_wq->work);
704 else if (end_io_wq->metadata)
d20f7043
CM		 705 btrfs_queue_worker(&fs_info->endio_meta_workers,
706 &end_io_wq->work);
707 else
708 btrfs_queue_worker(&fs_info->endio_workers,
709 &end_io_wq->work);
710 }
ce9adaa5
CM		 711}
712
0cb59c99
JB		 713/*
714 * For the metadata arg you want
715 *
716 * 0 - if data
717 * 1 - if normal metadta
718 * 2 - if writing to the free space cache area
53b381b3 719 * 3 - raid parity work
0cb59c99 720 */
22c59948
CM		 721int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
722 int metadata)
0b86a832 723{
ce9adaa5 724 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 725 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
726 if (!end_io_wq)
727 return -ENOMEM;
728
729 end_io_wq->private = bio->bi_private;
730 end_io_wq->end_io = bio->bi_end_io;
22c59948 731 end_io_wq->info = info;
ce9adaa5
CM		 732 end_io_wq->error = 0;
733 end_io_wq->bio = bio;
22c59948 734 end_io_wq->metadata = metadata;
ce9adaa5
CM		 735
736 bio->bi_private = end_io_wq;
737 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 738 return 0;
739}
740
b64a2851 741unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 742{
4854ddd0
CM		 743 unsigned long limit = min_t(unsigned long,
744 info->workers.max_workers,
745 info->fs_devices->open_devices);
746 return 256 * limit;
747}
0986fe9e 748
4a69a410
CM		 749static void run_one_async_start(struct btrfs_work *work)
750{
4a69a410 751 struct async_submit_bio *async;
79787eaa 752 int ret;
4a69a410
CM		 753
754 async = container_of(work, struct async_submit_bio, work);
79787eaa
JM		 755 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
756 async->mirror_num, async->bio_flags,
757 async->bio_offset);
758 if (ret)
759 async->error = ret;
4a69a410
CM		 760}
761
762static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 763{
764 struct btrfs_fs_info *fs_info;
765 struct async_submit_bio *async;
4854ddd0 766 int limit;
8b712842
CM		 767
768 async = container_of(work, struct async_submit_bio, work);
769 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 770
b64a2851 771 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 772 limit = limit * 2 / 3;
773
66657b31 774 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
b64a2851 775 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 776 wake_up(&fs_info->async_submit_wait);
777
79787eaa
JM		 778 /* If an error occured we just want to clean up the bio and move on */
779 if (async->error) {
780 bio_endio(async->bio, async->error);
781 return;
782 }
783
4a69a410 784 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 785 async->mirror_num, async->bio_flags,
786 async->bio_offset);
4a69a410
CM		 787}
788
789static void run_one_async_free(struct btrfs_work *work)
790{
791 struct async_submit_bio *async;
792
793 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 794 kfree(async);
795}
796
44b8bd7e
CM		 797int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
798 int rw, struct bio *bio, int mirror_num,
c8b97818 799 unsigned long bio_flags,
eaf25d93 800 u64 bio_offset,
4a69a410
CM		 801 extent_submit_bio_hook_t *submit_bio_start,
802 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 803{
804 struct async_submit_bio *async;
805
806 async = kmalloc(sizeof(*async), GFP_NOFS);
807 if (!async)
808 return -ENOMEM;
809
810 async->inode = inode;
811 async->rw = rw;
812 async->bio = bio;
813 async->mirror_num = mirror_num;
4a69a410
CM		 814 async->submit_bio_start = submit_bio_start;
815 async->submit_bio_done = submit_bio_done;
816
817 async->work.func = run_one_async_start;
818 async->work.ordered_func = run_one_async_done;
819 async->work.ordered_free = run_one_async_free;
820
8b712842 821 async->work.flags = 0;
c8b97818 822 async->bio_flags = bio_flags;
eaf25d93 823 async->bio_offset = bio_offset;
8c8bee1d 824
79787eaa
JM		 825 async->error = 0;
826
cb03c743 827 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 828
7b6d91da 829 if (rw & REQ_SYNC)
d313d7a3
CM		 830 btrfs_set_work_high_prio(&async->work);
831
8b712842 832 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 833
d397712b 834 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 835 atomic_read(&fs_info->nr_async_submits)) {
836 wait_event(fs_info->async_submit_wait,
837 (atomic_read(&fs_info->nr_async_submits) == 0));
838 }
839
44b8bd7e
CM		 840 return 0;
841}
842
ce3ed71a
CM		 843static int btree_csum_one_bio(struct bio *bio)
844{
845 struct bio_vec *bvec = bio->bi_io_vec;
846 int bio_index = 0;
847 struct btrfs_root *root;
79787eaa 848 int ret = 0;
ce3ed71a
CM		 849
850 WARN_ON(bio->bi_vcnt <= 0);
d397712b 851 while (bio_index < bio->bi_vcnt) {
ce3ed71a 852 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 853 ret = csum_dirty_buffer(root, bvec->bv_page);
854 if (ret)
855 break;
ce3ed71a
CM		 856 bio_index++;
857 bvec++;
858 }
79787eaa 859 return ret;
ce3ed71a
CM		 860}
861
4a69a410
CM		 862static int __btree_submit_bio_start(struct inode *inode, int rw,
863 struct bio *bio, int mirror_num,
eaf25d93
CM		 864 unsigned long bio_flags,
865 u64 bio_offset)
22c59948 866{
8b712842
CM		 867 /*
868 * when we're called for a write, we're already in the async
5443be45 869 * submission context. Just jump into btrfs_map_bio
8b712842 870 */
79787eaa 871 return btree_csum_one_bio(bio);
4a69a410 872}
22c59948 873
4a69a410 874static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 875 int mirror_num, unsigned long bio_flags,
876 u64 bio_offset)
4a69a410 877{
61891923
SB		 878 int ret;
879
8b712842 880 /*
4a69a410
CM		 881 * when we're called for a write, we're already in the async
882 * submission context. Just jump into btrfs_map_bio
8b712842 883 */
61891923
SB		 884 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
885 if (ret)
886 bio_endio(bio, ret);
887 return ret;
0b86a832
CM		 888}
889
de0022b9
JB		 890static int check_async_write(struct inode *inode, unsigned long bio_flags)
891{
892 if (bio_flags & EXTENT_BIO_TREE_LOG)
893 return 0;
894#ifdef CONFIG_X86
895 if (cpu_has_xmm4_2)
896 return 0;
897#endif
898 return 1;
899}
900
44b8bd7e 901static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 902 int mirror_num, unsigned long bio_flags,
903 u64 bio_offset)
44b8bd7e 904{
de0022b9 905 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 906 int ret;
907
7b6d91da 908 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 909 /*
910 * called for a read, do the setup so that checksum validation
911 * can happen in the async kernel threads
912 */
f3f266ab
CM		 913 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
914 bio, 1);
1d4284bd 915 if (ret)
61891923
SB		 916 goto out_w_error;
917 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
918 mirror_num, 0);
de0022b9
JB		 919 } else if (!async) {
920 ret = btree_csum_one_bio(bio);
921 if (ret)
61891923
SB		 922 goto out_w_error;
923 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
924 mirror_num, 0);
925 } else {
926 /*
927 * kthread helpers are used to submit writes so that
928 * checksumming can happen in parallel across all CPUs
929 */
930 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
931 inode, rw, bio, mirror_num, 0,
932 bio_offset,
933 __btree_submit_bio_start,
934 __btree_submit_bio_done);
44b8bd7e 935 }
d313d7a3 936
61891923
SB		 937 if (ret) {
938out_w_error:
939 bio_endio(bio, ret);
940 }
941 return ret;
44b8bd7e
CM		 942}
943
3dd1462e 944#ifdef CONFIG_MIGRATION
784b4e29 945static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 946 struct page *newpage, struct page *page,
947 enum migrate_mode mode)
784b4e29
CM		 948{
949 /*
950 * we can't safely write a btree page from here,
951 * we haven't done the locking hook
952 */
953 if (PageDirty(page))
954 return -EAGAIN;
955 /*
956 * Buffers may be managed in a filesystem specific way.
957 * We must have no buffers or drop them.
958 */
959 if (page_has_private(page) &&
960 !try_to_release_page(page, GFP_KERNEL))
961 return -EAGAIN;
a6bc32b8 962 return migrate_page(mapping, newpage, page, mode);
784b4e29 963}
3dd1462e 964#endif
784b4e29 965
0da5468f
CM		 966
967static int btree_writepages(struct address_space *mapping,
968 struct writeback_control *wbc)
969{
d1310b2e 970 struct extent_io_tree *tree;
e2d84521
MX		 971 struct btrfs_fs_info *fs_info;
972 int ret;
973
d1310b2e 974 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 975 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 976
977 if (wbc->for_kupdate)
978 return 0;
979
e2d84521 980 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 981 /* this is a bit racy, but that's ok */
e2d84521
MX		 982 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
983 BTRFS_DIRTY_METADATA_THRESH);
984 if (ret < 0)
793955bc 985 return 0;
793955bc 986 }
0b32f4bb 987 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 988}
989
b2950863 990static int btree_readpage(struct file *file, struct page *page)
5f39d397 991{
d1310b2e
CM		 992 struct extent_io_tree *tree;
993 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 994 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 995}
22b0ebda 996
70dec807 997static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 998{
98509cfc 999 if (PageWriteback(page) || PageDirty(page))
d397712b 1000 return 0;
0c4e538b
DS		 1001 /*
1002 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
1003 * slab allocation from alloc_extent_state down the callchain where
1004 * it'd hit a BUG_ON as those flags are not allowed.
1005 */
1006 gfp_flags &= ~GFP_SLAB_BUG_MASK;
1007
3083ee2e 1008 return try_release_extent_buffer(page, gfp_flags);
d98237b3
CM		 1009}
1010
5f39d397 1011static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 1012{
d1310b2e
CM		 1013 struct extent_io_tree *tree;
1014 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1015 extent_invalidatepage(tree, page, offset);
1016 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1017 if (PagePrivate(page)) {
d397712b
CM		 1018 printk(KERN_WARNING "btrfs warning page private not zero "
1019 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1020 ClearPagePrivate(page);
1021 set_page_private(page, 0);
1022 page_cache_release(page);
1023 }
d98237b3
CM		 1024}
1025
0b32f4bb
JB		 1026static int btree_set_page_dirty(struct page *page)
1027{
bb146eb2 1028#ifdef DEBUG
0b32f4bb
JB		 1029 struct extent_buffer *eb;
1030
1031 BUG_ON(!PagePrivate(page));
1032 eb = (struct extent_buffer *)page->private;
1033 BUG_ON(!eb);
1034 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1035 BUG_ON(!atomic_read(&eb->refs));
1036 btrfs_assert_tree_locked(eb);
bb146eb2 1037#endif
0b32f4bb
JB		 1038 return __set_page_dirty_nobuffers(page);
1039}
1040
7f09410b 1041static const struct address_space_operations btree_aops = {
d98237b3 1042 .readpage = btree_readpage,
0da5468f 1043 .writepages = btree_writepages,
5f39d397
CM		 1044 .releasepage = btree_releasepage,
1045 .invalidatepage = btree_invalidatepage,
5a92bc88 1046#ifdef CONFIG_MIGRATION
784b4e29 1047 .migratepage = btree_migratepage,
5a92bc88 1048#endif
0b32f4bb 1049 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1050};
1051
ca7a79ad
CM		 1052int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1053 u64 parent_transid)
090d1875 1054{
5f39d397
CM		 1055 struct extent_buffer *buf = NULL;
1056 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1057 int ret = 0;
090d1875 1058
db94535d 1059 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1060 if (!buf)
090d1875 1061 return 0;
d1310b2e 1062 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1063 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1064 free_extent_buffer(buf);
de428b63 1065 return ret;
090d1875
CM		 1066}
1067
ab0fff03
AJ		 1068int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1069 int mirror_num, struct extent_buffer **eb)
1070{
1071 struct extent_buffer *buf = NULL;
1072 struct inode *btree_inode = root->fs_info->btree_inode;
1073 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1074 int ret;
1075
1076 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1077 if (!buf)
1078 return 0;
1079
1080 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1081
1082 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1083 btree_get_extent, mirror_num);
1084 if (ret) {
1085 free_extent_buffer(buf);
1086 return ret;
1087 }
1088
1089 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1090 free_extent_buffer(buf);
1091 return -EIO;
0b32f4bb 1092 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1093 *eb = buf;
1094 } else {
1095 free_extent_buffer(buf);
1096 }
1097 return 0;
1098}
1099
0999df54
CM		 1100struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1101 u64 bytenr, u32 blocksize)
1102{
1103 struct inode *btree_inode = root->fs_info->btree_inode;
1104 struct extent_buffer *eb;
1105 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
f09d1f60 1106 bytenr, blocksize);
0999df54
CM		 1107 return eb;
1108}
1109
1110struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1111 u64 bytenr, u32 blocksize)
1112{
1113 struct inode *btree_inode = root->fs_info->btree_inode;
1114 struct extent_buffer *eb;
1115
1116 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
727011e0 1117 bytenr, blocksize);
0999df54
CM		 1118 return eb;
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);
5f39d397 1145 return buf;
ce9adaa5 1146
eb60ceac
CM		 1147}
1148
d5c13f92
JM		 1149void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1150 struct extent_buffer *buf)
ed2ff2cb 1151{
e2d84521
MX		 1152 struct btrfs_fs_info *fs_info = root->fs_info;
1153
55c69072 1154 if (btrfs_header_generation(buf) ==
e2d84521 1155 fs_info->running_transaction->transid) {
b9447ef8 1156 btrfs_assert_tree_locked(buf);
b4ce94de 1157
b9473439 1158 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1159 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1160 -buf->len,
1161 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1162 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1163 btrfs_set_lock_blocking(buf);
1164 clear_extent_buffer_dirty(buf);
1165 }
925baedd 1166 }
5f39d397
CM		 1167}
1168
143bede5
JM		 1169static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1170 u32 stripesize, struct btrfs_root *root,
1171 struct btrfs_fs_info *fs_info,
1172 u64 objectid)
d97e63b6 1173{
cfaa7295 1174 root->node = NULL;
a28ec197 1175 root->commit_root = NULL;
db94535d
CM		 1176 root->sectorsize = sectorsize;
1177 root->nodesize = nodesize;
1178 root->leafsize = leafsize;
87ee04eb 1179 root->stripesize = stripesize;
123abc88 1180 root->ref_cows = 0;
0b86a832 1181 root->track_dirty = 0;
c71bf099 1182 root->in_radix = 0;
d68fc57b
YZ		 1183 root->orphan_item_inserted = 0;
1184 root->orphan_cleanup_state = 0;
0b86a832 1185
0f7d52f4
CM		 1186 root->objectid = objectid;
1187 root->last_trans = 0;
13a8a7c8 1188 root->highest_objectid = 0;
58176a96 1189 root->name = NULL;
6bef4d31 1190 root->inode_tree = RB_ROOT;
16cdcec7 1191 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1192 root->block_rsv = NULL;
d68fc57b 1193 root->orphan_block_rsv = NULL;
0b86a832
CM		 1194
1195 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1196 INIT_LIST_HEAD(&root->root_list);
2ab28f32
JB		 1197 INIT_LIST_HEAD(&root->logged_list[0]);
1198 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1199 spin_lock_init(&root->orphan_lock);
5d4f98a2 1200 spin_lock_init(&root->inode_lock);
f0486c68 1201 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1202 spin_lock_init(&root->log_extents_lock[0]);
1203 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1204 mutex_init(&root->objectid_mutex);
e02119d5 1205 mutex_init(&root->log_mutex);
7237f183
YZ		 1206 init_waitqueue_head(&root->log_writer_wait);
1207 init_waitqueue_head(&root->log_commit_wait[0]);
1208 init_waitqueue_head(&root->log_commit_wait[1]);
1209 atomic_set(&root->log_commit[0], 0);
1210 atomic_set(&root->log_commit[1], 0);
1211 atomic_set(&root->log_writers, 0);
2ecb7923 1212 atomic_set(&root->log_batch, 0);
8a35d95f 1213 atomic_set(&root->orphan_inodes, 0);
7237f183 1214 root->log_transid = 0;
257c62e1 1215 root->last_log_commit = 0;
d0c803c4 1216 extent_io_tree_init(&root->dirty_log_pages,
f993c883 1217 fs_info->btree_inode->i_mapping);
017e5369 1218
3768f368
CM		 1219 memset(&root->root_key, 0, sizeof(root->root_key));
1220 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1221 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1222 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 1223 root->defrag_trans_start = fs_info->generation;
58176a96 1224 init_completion(&root->kobj_unregister);
6702ed49 1225 root->defrag_running = 0;
4d775673 1226 root->root_key.objectid = objectid;
0ee5dc67 1227 root->anon_dev = 0;
8ea05e3a 1228
5f3ab90a 1229 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1230}
1231
200a5c17
JM		 1232static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1233 struct btrfs_fs_info *fs_info,
1234 u64 objectid,
1235 struct btrfs_root *root)
3768f368
CM		 1236{
1237 int ret;
db94535d 1238 u32 blocksize;
84234f3a 1239 u64 generation;
3768f368 1240
db94535d 1241 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1242 tree_root->sectorsize, tree_root->stripesize,
1243 root, fs_info, objectid);
3768f368
CM		 1244 ret = btrfs_find_last_root(tree_root, objectid,
1245 &root->root_item, &root->root_key);
4df27c4d
YZ		 1246 if (ret > 0)
1247 return -ENOENT;
200a5c17
JM		 1248 else if (ret < 0)
1249 return ret;
3768f368 1250
84234f3a 1251 generation = btrfs_root_generation(&root->root_item);
db94535d 1252 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
af31f5e5 1253 root->commit_root = NULL;
db94535d 1254 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1255 blocksize, generation);
b9fab919 1256 if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
68433b73 1257 free_extent_buffer(root->node);
af31f5e5 1258 root->node = NULL;
68433b73
CM		 1259 return -EIO;
1260 }
4df27c4d 1261 root->commit_root = btrfs_root_node(root);
d97e63b6
CM		 1262 return 0;
1263}
1264
f84a8bd6 1265static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1266{
1267 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1268 if (root)
1269 root->fs_info = fs_info;
1270 return root;
1271}
1272
20897f5c
AJ		 1273struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1274 struct btrfs_fs_info *fs_info,
1275 u64 objectid)
1276{
1277 struct extent_buffer *leaf;
1278 struct btrfs_root *tree_root = fs_info->tree_root;
1279 struct btrfs_root *root;
1280 struct btrfs_key key;
1281 int ret = 0;
1282 u64 bytenr;
1283
1284 root = btrfs_alloc_root(fs_info);
1285 if (!root)
1286 return ERR_PTR(-ENOMEM);
1287
1288 __setup_root(tree_root->nodesize, tree_root->leafsize,
1289 tree_root->sectorsize, tree_root->stripesize,
1290 root, fs_info, objectid);
1291 root->root_key.objectid = objectid;
1292 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1293 root->root_key.offset = 0;
1294
1295 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1296 0, objectid, NULL, 0, 0, 0);
1297 if (IS_ERR(leaf)) {
1298 ret = PTR_ERR(leaf);
1dd05682 1299 leaf = NULL;
20897f5c
AJ		 1300 goto fail;
1301 }
1302
1303 bytenr = leaf->start;
1304 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1305 btrfs_set_header_bytenr(leaf, leaf->start);
1306 btrfs_set_header_generation(leaf, trans->transid);
1307 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1308 btrfs_set_header_owner(leaf, objectid);
1309 root->node = leaf;
1310
1311 write_extent_buffer(leaf, fs_info->fsid,
1312 (unsigned long)btrfs_header_fsid(leaf),
1313 BTRFS_FSID_SIZE);
1314 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1315 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1316 BTRFS_UUID_SIZE);
1317 btrfs_mark_buffer_dirty(leaf);
1318
1319 root->commit_root = btrfs_root_node(root);
1320 root->track_dirty = 1;
1321
1322
1323 root->root_item.flags = 0;
1324 root->root_item.byte_limit = 0;
1325 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1326 btrfs_set_root_generation(&root->root_item, trans->transid);
1327 btrfs_set_root_level(&root->root_item, 0);
1328 btrfs_set_root_refs(&root->root_item, 1);
1329 btrfs_set_root_used(&root->root_item, leaf->len);
1330 btrfs_set_root_last_snapshot(&root->root_item, 0);
1331 btrfs_set_root_dirid(&root->root_item, 0);
1332 root->root_item.drop_level = 0;
1333
1334 key.objectid = objectid;
1335 key.type = BTRFS_ROOT_ITEM_KEY;
1336 key.offset = 0;
1337 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1338 if (ret)
1339 goto fail;
1340
1341 btrfs_tree_unlock(leaf);
1342
1dd05682
TI		 1343 return root;
1344
20897f5c 1345fail:
1dd05682
TI		 1346 if (leaf) {
1347 btrfs_tree_unlock(leaf);
1348 free_extent_buffer(leaf);
1349 }
1350 kfree(root);
20897f5c 1351
1dd05682 1352 return ERR_PTR(ret);
20897f5c
AJ		 1353}
1354
7237f183
YZ		 1355static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1356 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1357{
1358 struct btrfs_root *root;
1359 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1360 struct extent_buffer *leaf;
e02119d5 1361
6f07e42e 1362 root = btrfs_alloc_root(fs_info);
e02119d5 1363 if (!root)
7237f183 1364 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1365
1366 __setup_root(tree_root->nodesize, tree_root->leafsize,
1367 tree_root->sectorsize, tree_root->stripesize,
1368 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1369
1370 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1371 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1372 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1373 /*
1374 * log trees do not get reference counted because they go away
1375 * before a real commit is actually done. They do store pointers
1376 * to file data extents, and those reference counts still get
1377 * updated (along with back refs to the log tree).
1378 */
e02119d5
CM		 1379 root->ref_cows = 0;
1380
5d4f98a2 1381 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1382 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1383 0, 0, 0);
7237f183
YZ		 1384 if (IS_ERR(leaf)) {
1385 kfree(root);
1386 return ERR_CAST(leaf);
1387 }
e02119d5 1388
5d4f98a2
YZ		 1389 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1390 btrfs_set_header_bytenr(leaf, leaf->start);
1391 btrfs_set_header_generation(leaf, trans->transid);
1392 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1393 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1394 root->node = leaf;
e02119d5
CM		 1395
1396 write_extent_buffer(root->node, root->fs_info->fsid,
1397 (unsigned long)btrfs_header_fsid(root->node),
1398 BTRFS_FSID_SIZE);
1399 btrfs_mark_buffer_dirty(root->node);
1400 btrfs_tree_unlock(root->node);
7237f183
YZ		 1401 return root;
1402}
1403
1404int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1405 struct btrfs_fs_info *fs_info)
1406{
1407 struct btrfs_root *log_root;
1408
1409 log_root = alloc_log_tree(trans, fs_info);
1410 if (IS_ERR(log_root))
1411 return PTR_ERR(log_root);
1412 WARN_ON(fs_info->log_root_tree);
1413 fs_info->log_root_tree = log_root;
1414 return 0;
1415}
1416
1417int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1418 struct btrfs_root *root)
1419{
1420 struct btrfs_root *log_root;
1421 struct btrfs_inode_item *inode_item;
1422
1423 log_root = alloc_log_tree(trans, root->fs_info);
1424 if (IS_ERR(log_root))
1425 return PTR_ERR(log_root);
1426
1427 log_root->last_trans = trans->transid;
1428 log_root->root_key.offset = root->root_key.objectid;
1429
1430 inode_item = &log_root->root_item.inode;
1431 inode_item->generation = cpu_to_le64(1);
1432 inode_item->size = cpu_to_le64(3);
1433 inode_item->nlink = cpu_to_le32(1);
1434 inode_item->nbytes = cpu_to_le64(root->leafsize);
1435 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1436
5d4f98a2 1437 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1438
1439 WARN_ON(root->log_root);
1440 root->log_root = log_root;
1441 root->log_transid = 0;
257c62e1 1442 root->last_log_commit = 0;
e02119d5
CM		 1443 return 0;
1444}
1445
1446struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1447 struct btrfs_key *location)
1448{
1449 struct btrfs_root *root;
1450 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1451 struct btrfs_path *path;
5f39d397 1452 struct extent_buffer *l;
84234f3a 1453 u64 generation;
db94535d 1454 u32 blocksize;
0f7d52f4 1455 int ret = 0;
8ea05e3a 1456 int slot;
0f7d52f4 1457
6f07e42e 1458 root = btrfs_alloc_root(fs_info);
0cf6c620 1459 if (!root)
0f7d52f4 1460 return ERR_PTR(-ENOMEM);
0f7d52f4 1461 if (location->offset == (u64)-1) {
db94535d 1462 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM		 1463 location->objectid, root);
1464 if (ret) {
0f7d52f4
CM		 1465 kfree(root);
1466 return ERR_PTR(ret);
1467 }
13a8a7c8 1468 goto out;
0f7d52f4
CM		 1469 }
1470
db94535d 1471 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1472 tree_root->sectorsize, tree_root->stripesize,
1473 root, fs_info, location->objectid);
0f7d52f4
CM		 1474
1475 path = btrfs_alloc_path();
db5b493a
TI		 1476 if (!path) {
1477 kfree(root);
1478 return ERR_PTR(-ENOMEM);
1479 }
0f7d52f4 1480 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
13a8a7c8
YZ		 1481 if (ret == 0) {
1482 l = path->nodes[0];
8ea05e3a 1483 slot = path->slots[0];
5fbf83c1 1484 btrfs_read_root_item(l, slot, &root->root_item);
13a8a7c8 1485 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4 1486 }
0f7d52f4
CM		 1487 btrfs_free_path(path);
1488 if (ret) {
5e540f77 1489 kfree(root);
13a8a7c8
YZ		 1490 if (ret > 0)
1491 ret = -ENOENT;
0f7d52f4
CM		 1492 return ERR_PTR(ret);
1493 }
13a8a7c8 1494
84234f3a 1495 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1496 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1497 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1498 blocksize, generation);
416bc658
JB		 1499 if (!root->node || !extent_buffer_uptodate(root->node)) {
1500 ret = (!root->node) ? -ENOMEM : -EIO;
1501
1502 free_extent_buffer(root->node);
1503 kfree(root);
1504 return ERR_PTR(ret);
1505 }
1506
5d4f98a2 1507 root->commit_root = btrfs_root_node(root);
79787eaa 1508 BUG_ON(!root->node); /* -ENOMEM */
13a8a7c8 1509out:
08fe4db1 1510 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1511 root->ref_cows = 1;
08fe4db1
LZ		 1512 btrfs_check_and_init_root_item(&root->root_item);
1513 }
13a8a7c8 1514
5eda7b5e
CM		 1515 return root;
1516}
1517
edbd8d4e
CM		 1518struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1519 struct btrfs_key *location)
5eda7b5e
CM		 1520{
1521 struct btrfs_root *root;
1522 int ret;
1523
edbd8d4e
CM		 1524 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1525 return fs_info->tree_root;
1526 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1527 return fs_info->extent_root;
8f18cf13
CM		 1528 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1529 return fs_info->chunk_root;
1530 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1531 return fs_info->dev_root;
0403e47e
YZ		 1532 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1533 return fs_info->csum_root;
bcef60f2
AJ		 1534 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1535 return fs_info->quota_root ? fs_info->quota_root :
1536 ERR_PTR(-ENOENT);
4df27c4d
YZ		 1537again:
1538 spin_lock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1539 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1540 (unsigned long)location->objectid);
4df27c4d 1541 spin_unlock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1542 if (root)
1543 return root;
1544
e02119d5 1545 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM		 1546 if (IS_ERR(root))
1547 return root;
3394e160 1548
581bb050 1549 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
581bb050
LZ		 1550 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1551 GFP_NOFS);
35a30d7c
DS		 1552 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1553 ret = -ENOMEM;
581bb050 1554 goto fail;
35a30d7c 1555 }
581bb050
LZ		 1556
1557 btrfs_init_free_ino_ctl(root);
1558 mutex_init(&root->fs_commit_mutex);
1559 spin_lock_init(&root->cache_lock);
1560 init_waitqueue_head(&root->cache_wait);
1561
0ee5dc67 1562 ret = get_anon_bdev(&root->anon_dev);
ac08aedf
CM		 1563 if (ret)
1564 goto fail;
3394e160 1565
d68fc57b
YZ		 1566 if (btrfs_root_refs(&root->root_item) == 0) {
1567 ret = -ENOENT;
1568 goto fail;
1569 }
1570
1571 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1572 if (ret < 0)
1573 goto fail;
1574 if (ret == 0)
1575 root->orphan_item_inserted = 1;
1576
4df27c4d
YZ		 1577 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1578 if (ret)
1579 goto fail;
1580
1581 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 1582 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1583 (unsigned long)root->root_key.objectid,
0f7d52f4 1584 root);
d68fc57b 1585 if (ret == 0)
4df27c4d 1586 root->in_radix = 1;
d68fc57b 1587
4df27c4d
YZ		 1588 spin_unlock(&fs_info->fs_roots_radix_lock);
1589 radix_tree_preload_end();
0f7d52f4 1590 if (ret) {
4df27c4d
YZ		 1591 if (ret == -EEXIST) {
1592 free_fs_root(root);
1593 goto again;
1594 }
1595 goto fail;
0f7d52f4 1596 }
4df27c4d
YZ		 1597
1598 ret = btrfs_find_dead_roots(fs_info->tree_root,
1599 root->root_key.objectid);
1600 WARN_ON(ret);
edbd8d4e 1601 return root;
4df27c4d
YZ		 1602fail:
1603 free_fs_root(root);
1604 return ERR_PTR(ret);
edbd8d4e
CM		 1605}
1606
04160088
CM		 1607static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1608{
1609 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1610 int ret = 0;
04160088
CM		 1611 struct btrfs_device *device;
1612 struct backing_dev_info *bdi;
b7967db7 1613
1f78160c
XG		 1614 rcu_read_lock();
1615 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1616 if (!device->bdev)
1617 continue;
04160088
CM		 1618 bdi = blk_get_backing_dev_info(device->bdev);
1619 if (bdi && bdi_congested(bdi, bdi_bits)) {
1620 ret = 1;
1621 break;
1622 }
1623 }
1f78160c 1624 rcu_read_unlock();
04160088
CM		 1625 return ret;
1626}
1627
ad081f14
JA		 1628/*
1629 * If this fails, caller must call bdi_destroy() to get rid of the
1630 * bdi again.
1631 */
04160088
CM		 1632static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1633{
ad081f14
JA		 1634 int err;
1635
1636 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1637 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1638 if (err)
1639 return err;
1640
4575c9cc 1641 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1642 bdi->congested_fn = btrfs_congested_fn;
1643 bdi->congested_data = info;
1644 return 0;
1645}
1646
8b712842
CM		 1647/*
1648 * called by the kthread helper functions to finally call the bio end_io
1649 * functions. This is where read checksum verification actually happens
1650 */
1651static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1652{
ce9adaa5 1653 struct bio *bio;
8b712842
CM		 1654 struct end_io_wq *end_io_wq;
1655 struct btrfs_fs_info *fs_info;
ce9adaa5 1656 int error;
ce9adaa5 1657
8b712842
CM		 1658 end_io_wq = container_of(work, struct end_io_wq, work);
1659 bio = end_io_wq->bio;
1660 fs_info = end_io_wq->info;
ce9adaa5 1661
8b712842
CM		 1662 error = end_io_wq->error;
1663 bio->bi_private = end_io_wq->private;
1664 bio->bi_end_io = end_io_wq->end_io;
1665 kfree(end_io_wq);
8b712842 1666 bio_endio(bio, error);
44b8bd7e
CM		 1667}
1668
a74a4b97
CM		 1669static int cleaner_kthread(void *arg)
1670{
1671 struct btrfs_root *root = arg;
1672
1673 do {
9d1a2a3a
DS		 1674 int again = 0;
1675
76dda93c 1676 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
9d1a2a3a
DS		 1677 down_read_trylock(&root->fs_info->sb->s_umount)) {
1678 if (mutex_trylock(&root->fs_info->cleaner_mutex)) {
1679 btrfs_run_delayed_iputs(root);
1680 again = btrfs_clean_one_deleted_snapshot(root);
1681 mutex_unlock(&root->fs_info->cleaner_mutex);
1682 }
4cb5300b 1683 btrfs_run_defrag_inodes(root->fs_info);
9d1a2a3a 1684 up_read(&root->fs_info->sb->s_umount);
76dda93c 1685 }
a74a4b97 1686
9d1a2a3a 1687 if (!try_to_freeze() && !again) {
a74a4b97 1688 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1689 if (!kthread_should_stop())
1690 schedule();
a74a4b97
CM		 1691 __set_current_state(TASK_RUNNING);
1692 }
1693 } while (!kthread_should_stop());
1694 return 0;
1695}
1696
1697static int transaction_kthread(void *arg)
1698{
1699 struct btrfs_root *root = arg;
1700 struct btrfs_trans_handle *trans;
1701 struct btrfs_transaction *cur;
8929ecfa 1702 u64 transid;
a74a4b97
CM		 1703 unsigned long now;
1704 unsigned long delay;
914b2007 1705 bool cannot_commit;
a74a4b97
CM		 1706
1707 do {
914b2007 1708 cannot_commit = false;
a74a4b97 1709 delay = HZ * 30;
a74a4b97
CM		 1710 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1711
a4abeea4 1712 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1713 cur = root->fs_info->running_transaction;
1714 if (!cur) {
a4abeea4 1715 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1716 goto sleep;
1717 }
31153d81 1718
a74a4b97 1719 now = get_seconds();
8929ecfa
YZ		 1720 if (!cur->blocked &&
1721 (now < cur->start_time || now - cur->start_time < 30)) {
a4abeea4 1722 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1723 delay = HZ * 5;
1724 goto sleep;
1725 }
8929ecfa 1726 transid = cur->transid;
a4abeea4 1727 spin_unlock(&root->fs_info->trans_lock);
56bec294 1728
79787eaa 1729 /* If the file system is aborted, this will always fail. */
354aa0fb 1730 trans = btrfs_attach_transaction(root);
914b2007 1731 if (IS_ERR(trans)) {
354aa0fb
MX		 1732 if (PTR_ERR(trans) != -ENOENT)
1733 cannot_commit = true;
79787eaa 1734 goto sleep;
914b2007 1735 }
8929ecfa 1736 if (transid == trans->transid) {
79787eaa 1737 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1738 } else {
1739 btrfs_end_transaction(trans, root);
1740 }
a74a4b97
CM		 1741sleep:
1742 wake_up_process(root->fs_info->cleaner_kthread);
1743 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1744
a0acae0e 1745 if (!try_to_freeze()) {
a74a4b97 1746 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1747 if (!kthread_should_stop() &&
914b2007
JK		 1748 (!btrfs_transaction_blocked(root->fs_info) ||
1749 cannot_commit))
8929ecfa 1750 schedule_timeout(delay);
a74a4b97
CM		 1751 __set_current_state(TASK_RUNNING);
1752 }
1753 } while (!kthread_should_stop());
1754 return 0;
1755}
1756
af31f5e5
CM		 1757/*
1758 * this will find the highest generation in the array of
1759 * root backups. The index of the highest array is returned,
1760 * or -1 if we can't find anything.
1761 *
1762 * We check to make sure the array is valid by comparing the
1763 * generation of the latest root in the array with the generation
1764 * in the super block. If they don't match we pitch it.
1765 */
1766static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1767{
1768 u64 cur;
1769 int newest_index = -1;
1770 struct btrfs_root_backup *root_backup;
1771 int i;
1772
1773 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1774 root_backup = info->super_copy->super_roots + i;
1775 cur = btrfs_backup_tree_root_gen(root_backup);
1776 if (cur == newest_gen)
1777 newest_index = i;
1778 }
1779
1780 /* check to see if we actually wrapped around */
1781 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1782 root_backup = info->super_copy->super_roots;
1783 cur = btrfs_backup_tree_root_gen(root_backup);
1784 if (cur == newest_gen)
1785 newest_index = 0;
1786 }
1787 return newest_index;
1788}
1789
1790
1791/*
1792 * find the oldest backup so we know where to store new entries
1793 * in the backup array. This will set the backup_root_index
1794 * field in the fs_info struct
1795 */
1796static void find_oldest_super_backup(struct btrfs_fs_info *info,
1797 u64 newest_gen)
1798{
1799 int newest_index = -1;
1800
1801 newest_index = find_newest_super_backup(info, newest_gen);
1802 /* if there was garbage in there, just move along */
1803 if (newest_index == -1) {
1804 info->backup_root_index = 0;
1805 } else {
1806 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1807 }
1808}
1809
1810/*
1811 * copy all the root pointers into the super backup array.
1812 * this will bump the backup pointer by one when it is
1813 * done
1814 */
1815static void backup_super_roots(struct btrfs_fs_info *info)
1816{
1817 int next_backup;
1818 struct btrfs_root_backup *root_backup;
1819 int last_backup;
1820
1821 next_backup = info->backup_root_index;
1822 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1823 BTRFS_NUM_BACKUP_ROOTS;
1824
1825 /*
1826 * just overwrite the last backup if we're at the same generation
1827 * this happens only at umount
1828 */
1829 root_backup = info->super_for_commit->super_roots + last_backup;
1830 if (btrfs_backup_tree_root_gen(root_backup) ==
1831 btrfs_header_generation(info->tree_root->node))
1832 next_backup = last_backup;
1833
1834 root_backup = info->super_for_commit->super_roots + next_backup;
1835
1836 /*
1837 * make sure all of our padding and empty slots get zero filled
1838 * regardless of which ones we use today
1839 */
1840 memset(root_backup, 0, sizeof(*root_backup));
1841
1842 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1843
1844 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1845 btrfs_set_backup_tree_root_gen(root_backup,
1846 btrfs_header_generation(info->tree_root->node));
1847
1848 btrfs_set_backup_tree_root_level(root_backup,
1849 btrfs_header_level(info->tree_root->node));
1850
1851 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1852 btrfs_set_backup_chunk_root_gen(root_backup,
1853 btrfs_header_generation(info->chunk_root->node));
1854 btrfs_set_backup_chunk_root_level(root_backup,
1855 btrfs_header_level(info->chunk_root->node));
1856
1857 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1858 btrfs_set_backup_extent_root_gen(root_backup,
1859 btrfs_header_generation(info->extent_root->node));
1860 btrfs_set_backup_extent_root_level(root_backup,
1861 btrfs_header_level(info->extent_root->node));
1862
7c7e82a7
CM		 1863 /*
1864 * we might commit during log recovery, which happens before we set
1865 * the fs_root. Make sure it is valid before we fill it in.
1866 */
1867 if (info->fs_root && info->fs_root->node) {
1868 btrfs_set_backup_fs_root(root_backup,
1869 info->fs_root->node->start);
1870 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1871 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1872 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1873 btrfs_header_level(info->fs_root->node));
7c7e82a7 1874 }
af31f5e5
CM		 1875
1876 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1877 btrfs_set_backup_dev_root_gen(root_backup,
1878 btrfs_header_generation(info->dev_root->node));
1879 btrfs_set_backup_dev_root_level(root_backup,
1880 btrfs_header_level(info->dev_root->node));
1881
1882 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1883 btrfs_set_backup_csum_root_gen(root_backup,
1884 btrfs_header_generation(info->csum_root->node));
1885 btrfs_set_backup_csum_root_level(root_backup,
1886 btrfs_header_level(info->csum_root->node));
1887
1888 btrfs_set_backup_total_bytes(root_backup,
1889 btrfs_super_total_bytes(info->super_copy));
1890 btrfs_set_backup_bytes_used(root_backup,
1891 btrfs_super_bytes_used(info->super_copy));
1892 btrfs_set_backup_num_devices(root_backup,
1893 btrfs_super_num_devices(info->super_copy));
1894
1895 /*
1896 * if we don't copy this out to the super_copy, it won't get remembered
1897 * for the next commit
1898 */
1899 memcpy(&info->super_copy->super_roots,
1900 &info->super_for_commit->super_roots,
1901 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1902}
1903
1904/*
1905 * this copies info out of the root backup array and back into
1906 * the in-memory super block. It is meant to help iterate through
1907 * the array, so you send it the number of backups you've already
1908 * tried and the last backup index you used.
1909 *
1910 * this returns -1 when it has tried all the backups
1911 */
1912static noinline int next_root_backup(struct btrfs_fs_info *info,
1913 struct btrfs_super_block *super,
1914 int *num_backups_tried, int *backup_index)
1915{
1916 struct btrfs_root_backup *root_backup;
1917 int newest = *backup_index;
1918
1919 if (*num_backups_tried == 0) {
1920 u64 gen = btrfs_super_generation(super);
1921
1922 newest = find_newest_super_backup(info, gen);
1923 if (newest == -1)
1924 return -1;
1925
1926 *backup_index = newest;
1927 *num_backups_tried = 1;
1928 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1929 /* we've tried all the backups, all done */
1930 return -1;
1931 } else {
1932 /* jump to the next oldest backup */
1933 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1934 BTRFS_NUM_BACKUP_ROOTS;
1935 *backup_index = newest;
1936 *num_backups_tried += 1;
1937 }
1938 root_backup = super->super_roots + newest;
1939
1940 btrfs_set_super_generation(super,
1941 btrfs_backup_tree_root_gen(root_backup));
1942 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1943 btrfs_set_super_root_level(super,
1944 btrfs_backup_tree_root_level(root_backup));
1945 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1946
1947 /*
1948 * fixme: the total bytes and num_devices need to match or we should
1949 * need a fsck
1950 */
1951 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1952 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1953 return 0;
1954}
1955
7abadb64
LB		 1956/* helper to cleanup workers */
1957static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1958{
1959 btrfs_stop_workers(&fs_info->generic_worker);
1960 btrfs_stop_workers(&fs_info->fixup_workers);
1961 btrfs_stop_workers(&fs_info->delalloc_workers);
1962 btrfs_stop_workers(&fs_info->workers);
1963 btrfs_stop_workers(&fs_info->endio_workers);
1964 btrfs_stop_workers(&fs_info->endio_meta_workers);
1965 btrfs_stop_workers(&fs_info->endio_raid56_workers);
1966 btrfs_stop_workers(&fs_info->rmw_workers);
1967 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1968 btrfs_stop_workers(&fs_info->endio_write_workers);
1969 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1970 btrfs_stop_workers(&fs_info->submit_workers);
1971 btrfs_stop_workers(&fs_info->delayed_workers);
1972 btrfs_stop_workers(&fs_info->caching_workers);
1973 btrfs_stop_workers(&fs_info->readahead_workers);
1974 btrfs_stop_workers(&fs_info->flush_workers);
1975}
1976
af31f5e5
CM		 1977/* helper to cleanup tree roots */
1978static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1979{
1980 free_extent_buffer(info->tree_root->node);
1981 free_extent_buffer(info->tree_root->commit_root);
1982 free_extent_buffer(info->dev_root->node);
1983 free_extent_buffer(info->dev_root->commit_root);
1984 free_extent_buffer(info->extent_root->node);
1985 free_extent_buffer(info->extent_root->commit_root);
1986 free_extent_buffer(info->csum_root->node);
1987 free_extent_buffer(info->csum_root->commit_root);
bcef60f2
AJ		 1988 if (info->quota_root) {
1989 free_extent_buffer(info->quota_root->node);
1990 free_extent_buffer(info->quota_root->commit_root);
1991 }
af31f5e5
CM		 1992
1993 info->tree_root->node = NULL;
1994 info->tree_root->commit_root = NULL;
1995 info->dev_root->node = NULL;
1996 info->dev_root->commit_root = NULL;
1997 info->extent_root->node = NULL;
1998 info->extent_root->commit_root = NULL;
1999 info->csum_root->node = NULL;
2000 info->csum_root->commit_root = NULL;
bcef60f2
AJ		 2001 if (info->quota_root) {
2002 info->quota_root->node = NULL;
2003 info->quota_root->commit_root = NULL;
2004 }
af31f5e5
CM		 2005
2006 if (chunk_root) {
2007 free_extent_buffer(info->chunk_root->node);
2008 free_extent_buffer(info->chunk_root->commit_root);
2009 info->chunk_root->node = NULL;
2010 info->chunk_root->commit_root = NULL;
2011 }
2012}
2013
171f6537
JB		 2014static void del_fs_roots(struct btrfs_fs_info *fs_info)
2015{
2016 int ret;
2017 struct btrfs_root *gang[8];
2018 int i;
2019
2020 while (!list_empty(&fs_info->dead_roots)) {
2021 gang[0] = list_entry(fs_info->dead_roots.next,
2022 struct btrfs_root, root_list);
2023 list_del(&gang[0]->root_list);
2024
2025 if (gang[0]->in_radix) {
2026 btrfs_free_fs_root(fs_info, gang[0]);
2027 } else {
2028 free_extent_buffer(gang[0]->node);
2029 free_extent_buffer(gang[0]->commit_root);
2030 kfree(gang[0]);
2031 }
2032 }
2033
2034 while (1) {
2035 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2036 (void **)gang, 0,
2037 ARRAY_SIZE(gang));
2038 if (!ret)
2039 break;
2040 for (i = 0; i < ret; i++)
2041 btrfs_free_fs_root(fs_info, gang[i]);
2042 }
2043}
af31f5e5 2044
ad2b2c80
AV		 2045int open_ctree(struct super_block *sb,
2046 struct btrfs_fs_devices *fs_devices,
2047 char *options)
2e635a27 2048{
db94535d
CM		 2049 u32 sectorsize;
2050 u32 nodesize;
2051 u32 leafsize;
2052 u32 blocksize;
87ee04eb 2053 u32 stripesize;
84234f3a 2054 u64 generation;
f2b636e8 2055 u64 features;
3de4586c 2056 struct btrfs_key location;
a061fc8d 2057 struct buffer_head *bh;
4d34b278 2058 struct btrfs_super_block *disk_super;
815745cf 2059 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2060 struct btrfs_root *tree_root;
4d34b278
ID		 2061 struct btrfs_root *extent_root;
2062 struct btrfs_root *csum_root;
2063 struct btrfs_root *chunk_root;
2064 struct btrfs_root *dev_root;
bcef60f2 2065 struct btrfs_root *quota_root;
e02119d5 2066 struct btrfs_root *log_tree_root;
eb60ceac 2067 int ret;
e58ca020 2068 int err = -EINVAL;
af31f5e5
CM		 2069 int num_backups_tried = 0;
2070 int backup_index = 0;
4543df7e 2071
f84a8bd6 2072 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e
AV		 2073 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
2074 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
2075 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2076 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
bcef60f2 2077 quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
8790d502 2078
f84a8bd6 2079 if (!tree_root || !extent_root || !csum_root ||
bcef60f2 2080 !chunk_root || !dev_root || !quota_root) {
39279cc3
CM		 2081 err = -ENOMEM;
2082 goto fail;
2083 }
76dda93c
YZ		 2084
2085 ret = init_srcu_struct(&fs_info->subvol_srcu);
2086 if (ret) {
2087 err = ret;
2088 goto fail;
2089 }
2090
2091 ret = setup_bdi(fs_info, &fs_info->bdi);
2092 if (ret) {
2093 err = ret;
2094 goto fail_srcu;
2095 }
2096
e2d84521
MX		 2097 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2098 if (ret) {
2099 err = ret;
2100 goto fail_bdi;
2101 }
2102 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2103 (1 + ilog2(nr_cpu_ids));
2104
963d678b
MX		 2105 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2106 if (ret) {
2107 err = ret;
2108 goto fail_dirty_metadata_bytes;
2109 }
2110
76dda93c
YZ		 2111 fs_info->btree_inode = new_inode(sb);
2112 if (!fs_info->btree_inode) {
2113 err = -ENOMEM;
963d678b 2114 goto fail_delalloc_bytes;
76dda93c
YZ		 2115 }
2116
a6591715 2117 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2118
76dda93c 2119 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 2120 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2121 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2122 INIT_LIST_HEAD(&fs_info->delayed_iputs);
ea8c2819 2123 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
11833d66 2124 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1832a6d5 2125 spin_lock_init(&fs_info->delalloc_lock);
a4abeea4 2126 spin_lock_init(&fs_info->trans_lock);
76dda93c 2127 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2128 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2129 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2130 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2131 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2132 spin_lock_init(&fs_info->super_lock);
f29021b2 2133 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2134 mutex_init(&fs_info->reloc_mutex);
de98ced9 2135 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2136
58176a96 2137 init_completion(&fs_info->kobj_unregister);
0b86a832 2138 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2139 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2140 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2141 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2142 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2143 BTRFS_BLOCK_RSV_GLOBAL);
2144 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2145 BTRFS_BLOCK_RSV_DELALLOC);
2146 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2147 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2148 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2149 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2150 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2151 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2152 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2153 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2154 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2155 atomic_set(&fs_info->defrag_running, 0);
f29021b2 2156 atomic_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2157 fs_info->sb = sb;
6f568d35 2158 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2159 fs_info->metadata_ratio = 0;
4cb5300b 2160 fs_info->defrag_inodes = RB_ROOT;
a4abeea4 2161 fs_info->trans_no_join = 0;
2bf64758 2162 fs_info->free_chunk_space = 0;
f29021b2 2163 fs_info->tree_mod_log = RB_ROOT;
c8b97818 2164
90519d66
AJ		 2165 /* readahead state */
2166 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2167 spin_lock_init(&fs_info->reada_lock);
c8b97818 2168
b34b086c
CM		 2169 fs_info->thread_pool_size = min_t(unsigned long,
2170 num_online_cpus() + 2, 8);
0afbaf8c 2171
3eaa2885
CM		 2172 INIT_LIST_HEAD(&fs_info->ordered_extents);
2173 spin_lock_init(&fs_info->ordered_extent_lock);
16cdcec7
MX		 2174 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2175 GFP_NOFS);
2176 if (!fs_info->delayed_root) {
2177 err = -ENOMEM;
2178 goto fail_iput;
2179 }
2180 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2181
a2de733c
AJ		 2182 mutex_init(&fs_info->scrub_lock);
2183 atomic_set(&fs_info->scrubs_running, 0);
2184 atomic_set(&fs_info->scrub_pause_req, 0);
2185 atomic_set(&fs_info->scrubs_paused, 0);
2186 atomic_set(&fs_info->scrub_cancel_req, 0);
2187 init_waitqueue_head(&fs_info->scrub_pause_wait);
2188 init_rwsem(&fs_info->scrub_super_lock);
2189 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2190#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2191 fs_info->check_integrity_print_mask = 0;
2192#endif
a2de733c 2193
c9e9f97b
ID		 2194 spin_lock_init(&fs_info->balance_lock);
2195 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2196 atomic_set(&fs_info->balance_running, 0);
2197 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2198 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2199 fs_info->balance_ctl = NULL;
837d5b6e 2200 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2201
a061fc8d
CM		 2202 sb->s_blocksize = 4096;
2203 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2204 sb->s_bdi = &fs_info->bdi;
a061fc8d 2205
76dda93c 2206 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2207 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2208 /*
2209 * we set the i_size on the btree inode to the max possible int.
2210 * the real end of the address space is determined by all of
2211 * the devices in the system
2212 */
2213 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2214 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2215 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2216
5d4f98a2 2217 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2218 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2219 fs_info->btree_inode->i_mapping);
0b32f4bb 2220 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2221 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2222
2223 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2224
76dda93c
YZ		 2225 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2226 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2227 sizeof(struct btrfs_key));
72ac3c0d
JB		 2228 set_bit(BTRFS_INODE_DUMMY,
2229 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
c65ddb52 2230 insert_inode_hash(fs_info->btree_inode);
76dda93c 2231
0f9dd46c 2232 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2233 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2234 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2235
11833d66 2236 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2237 fs_info->btree_inode->i_mapping);
11833d66 2238 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2239 fs_info->btree_inode->i_mapping);
11833d66 2240 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2241 fs_info->do_barriers = 1;
e18e4809 2242
39279cc3 2243
5a3f23d5 2244 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 2245 mutex_init(&fs_info->tree_log_mutex);
925baedd 2246 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2247 mutex_init(&fs_info->transaction_kthread_mutex);
2248 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2249 mutex_init(&fs_info->volume_mutex);
276e680d 2250 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2251 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2252 init_rwsem(&fs_info->subvol_sem);
e922e087
SB		 2253 fs_info->dev_replace.lock_owner = 0;
2254 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2255 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2256 mutex_init(&fs_info->dev_replace.lock_management_lock);
2257 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2258
416ac51d 2259 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2260 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2261 fs_info->qgroup_tree = RB_ROOT;
2262 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2263 fs_info->qgroup_seq = 1;
2264 fs_info->quota_enabled = 0;
2265 fs_info->pending_quota_state = 0;
2266
fa9c0d79
CM		 2267 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2268 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2269
e6dcd2dc 2270 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2271 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2272 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2273 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2274
53b381b3
DW		 2275 ret = btrfs_alloc_stripe_hash_table(fs_info);
2276 if (ret) {
83c8266a 2277 err = ret;
53b381b3
DW		 2278 goto fail_alloc;
2279 }
2280
0b86a832 2281 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2282 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2283
3c4bb26b 2284 invalidate_bdev(fs_devices->latest_bdev);
a512bbf8 2285 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2286 if (!bh) {
2287 err = -EINVAL;
16cdcec7 2288 goto fail_alloc;
20b45077 2289 }
39279cc3 2290
6c41761f
DS		 2291 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2292 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2293 sizeof(*fs_info->super_for_commit));
a061fc8d 2294 brelse(bh);
5f39d397 2295
6c41761f 2296 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2297
6c41761f 2298 disk_super = fs_info->super_copy;
0f7d52f4 2299 if (!btrfs_super_root(disk_super))
16cdcec7 2300 goto fail_alloc;
0f7d52f4 2301
acce952b 2302 /* check FS state, whether FS is broken. */
87533c47
MX		 2303 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2304 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2305
fcd1f065
DS		 2306 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2307 if (ret) {
2308 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2309 err = ret;
2310 goto fail_alloc;
2311 }
acce952b 2312
af31f5e5
CM		 2313 /*
2314 * run through our array of backup supers and setup
2315 * our ring pointer to the oldest one
2316 */
2317 generation = btrfs_super_generation(disk_super);
2318 find_oldest_super_backup(fs_info, generation);
2319
75e7cb7f
LB		 2320 /*
2321 * In the long term, we'll store the compression type in the super
2322 * block, and it'll be used for per file compression control.
2323 */
2324 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2325
2b82032c
YZ		 2326 ret = btrfs_parse_options(tree_root, options);
2327 if (ret) {
2328 err = ret;
16cdcec7 2329 goto fail_alloc;
2b82032c 2330 }
dfe25020 2331
f2b636e8
JB		 2332 features = btrfs_super_incompat_flags(disk_super) &
2333 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2334 if (features) {
2335 printk(KERN_ERR "BTRFS: couldn't mount because of "
2336 "unsupported optional features (%Lx).\n",
21380931 2337 (unsigned long long)features);
f2b636e8 2338 err = -EINVAL;
16cdcec7 2339 goto fail_alloc;
f2b636e8
JB		 2340 }
2341
727011e0
CM		 2342 if (btrfs_super_leafsize(disk_super) !=
2343 btrfs_super_nodesize(disk_super)) {
2344 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2345 "blocksizes don't match. node %d leaf %d\n",
2346 btrfs_super_nodesize(disk_super),
2347 btrfs_super_leafsize(disk_super));
2348 err = -EINVAL;
2349 goto fail_alloc;
2350 }
2351 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2352 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2353 "blocksize (%d) was too large\n",
2354 btrfs_super_leafsize(disk_super));
2355 err = -EINVAL;
2356 goto fail_alloc;
2357 }
2358
5d4f98a2 2359 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2360 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2361 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2362 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2363
3173a18f
JB		 2364 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2365 printk(KERN_ERR "btrfs: has skinny extents\n");
2366
727011e0
CM		 2367 /*
2368 * flag our filesystem as having big metadata blocks if
2369 * they are bigger than the page size
2370 */
2371 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2372 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2373 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2374 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2375 }
2376
bc3f116f
CM		 2377 nodesize = btrfs_super_nodesize(disk_super);
2378 leafsize = btrfs_super_leafsize(disk_super);
2379 sectorsize = btrfs_super_sectorsize(disk_super);
2380 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2381 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2382 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2383
2384 /*
2385 * mixed block groups end up with duplicate but slightly offset
2386 * extent buffers for the same range. It leads to corruptions
2387 */
2388 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2389 (sectorsize != leafsize)) {
2390 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2391 "are not allowed for mixed block groups on %s\n",
2392 sb->s_id);
2393 goto fail_alloc;
2394 }
2395
ceda0864
MX		 2396 /*
2397 * Needn't use the lock because there is no other task which will
2398 * update the flag.
2399 */
a6fa6fae 2400 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2401
f2b636e8
JB		 2402 features = btrfs_super_compat_ro_flags(disk_super) &
2403 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2404 if (!(sb->s_flags & MS_RDONLY) && features) {
2405 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2406 "unsupported option features (%Lx).\n",
21380931 2407 (unsigned long long)features);
f2b636e8 2408 err = -EINVAL;
16cdcec7 2409 goto fail_alloc;
f2b636e8 2410 }
61d92c32
CM		 2411
2412 btrfs_init_workers(&fs_info->generic_worker,
2413 "genwork", 1, NULL);
2414
5443be45 2415 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 2416 fs_info->thread_pool_size,
2417 &fs_info->generic_worker);
c8b97818 2418
771ed689 2419 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 2420 fs_info->thread_pool_size,
2421 &fs_info->generic_worker);
771ed689 2422
8ccf6f19
MX		 2423 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2424 fs_info->thread_pool_size,
2425 &fs_info->generic_worker);
2426
5443be45 2427 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 2428 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 2429 fs_info->thread_pool_size),
2430 &fs_info->generic_worker);
61b49440 2431
bab39bf9
JB		 2432 btrfs_init_workers(&fs_info->caching_workers, "cache",
2433 2, &fs_info->generic_worker);
2434
61b49440
CM		 2435 /* a higher idle thresh on the submit workers makes it much more
2436 * likely that bios will be send down in a sane order to the
2437 * devices
2438 */
2439 fs_info->submit_workers.idle_thresh = 64;
53863232 2440
771ed689 2441 fs_info->workers.idle_thresh = 16;
4a69a410 2442 fs_info->workers.ordered = 1;
61b49440 2443
771ed689
CM		 2444 fs_info->delalloc_workers.idle_thresh = 2;
2445 fs_info->delalloc_workers.ordered = 1;
2446
61d92c32
CM		 2447 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2448 &fs_info->generic_worker);
5443be45 2449 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2450 fs_info->thread_pool_size,
2451 &fs_info->generic_worker);
d20f7043 2452 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2453 fs_info->thread_pool_size,
2454 &fs_info->generic_worker);
cad321ad 2455 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2456 "endio-meta-write", fs_info->thread_pool_size,
2457 &fs_info->generic_worker);
53b381b3
DW		 2458 btrfs_init_workers(&fs_info->endio_raid56_workers,
2459 "endio-raid56", fs_info->thread_pool_size,
2460 &fs_info->generic_worker);
2461 btrfs_init_workers(&fs_info->rmw_workers,
2462 "rmw", fs_info->thread_pool_size,
2463 &fs_info->generic_worker);
5443be45 2464 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2465 fs_info->thread_pool_size,
2466 &fs_info->generic_worker);
0cb59c99
JB		 2467 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2468 1, &fs_info->generic_worker);
16cdcec7
MX		 2469 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2470 fs_info->thread_pool_size,
2471 &fs_info->generic_worker);
90519d66
AJ		 2472 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2473 fs_info->thread_pool_size,
2474 &fs_info->generic_worker);
61b49440
CM		 2475
2476 /*
2477 * endios are largely parallel and should have a very
2478 * low idle thresh
2479 */
2480 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2481 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2482 fs_info->endio_raid56_workers.idle_thresh = 4;
2483 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2484
9042846b
CM		 2485 fs_info->endio_write_workers.idle_thresh = 2;
2486 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2487 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2488
0dc3b84a
JB		 2489 /*
2490 * btrfs_start_workers can really only fail because of ENOMEM so just
2491 * return -ENOMEM if any of these fail.
2492 */
2493 ret = btrfs_start_workers(&fs_info->workers);
2494 ret |= btrfs_start_workers(&fs_info->generic_worker);
2495 ret |= btrfs_start_workers(&fs_info->submit_workers);
2496 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2497 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2498 ret |= btrfs_start_workers(&fs_info->endio_workers);
2499 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2500 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2501 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2502 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2503 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2504 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2505 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2506 ret |= btrfs_start_workers(&fs_info->caching_workers);
2507 ret |= btrfs_start_workers(&fs_info->readahead_workers);
8ccf6f19 2508 ret |= btrfs_start_workers(&fs_info->flush_workers);
0dc3b84a 2509 if (ret) {
fed425c7 2510 err = -ENOMEM;
0dc3b84a
JB		 2511 goto fail_sb_buffer;
2512 }
4543df7e 2513
4575c9cc 2514 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2515 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2516 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2517
db94535d
CM		 2518 tree_root->nodesize = nodesize;
2519 tree_root->leafsize = leafsize;
2520 tree_root->sectorsize = sectorsize;
87ee04eb 2521 tree_root->stripesize = stripesize;
a061fc8d
CM		 2522
2523 sb->s_blocksize = sectorsize;
2524 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2525
cdb4c574 2526 if (disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) {
d397712b 2527 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2528 goto fail_sb_buffer;
2529 }
19c00ddc 2530
8d082fb7
LB		 2531 if (sectorsize != PAGE_SIZE) {
2532 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2533 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2534 goto fail_sb_buffer;
2535 }
2536
925baedd 2537 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2538 ret = btrfs_read_sys_array(tree_root);
925baedd 2539 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2540 if (ret) {
d397712b
CM		 2541 printk(KERN_WARNING "btrfs: failed to read the system "
2542 "array on %s\n", sb->s_id);
5d4f98a2 2543 goto fail_sb_buffer;
84eed90f 2544 }
0b86a832
CM		 2545
2546 blocksize = btrfs_level_size(tree_root,
2547 btrfs_super_chunk_root_level(disk_super));
84234f3a 2548 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2549
2550 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2551 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2552
2553 chunk_root->node = read_tree_block(chunk_root,
2554 btrfs_super_chunk_root(disk_super),
84234f3a 2555 blocksize, generation);
416bc658
JB		 2556 if (!chunk_root->node ||
2557 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
83121942
DW		 2558 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2559 sb->s_id);
af31f5e5 2560 goto fail_tree_roots;
83121942 2561 }
5d4f98a2
YZ		 2562 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2563 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2564
e17cade2 2565 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 2566 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2567 BTRFS_UUID_SIZE);
e17cade2 2568
0b86a832 2569 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2570 if (ret) {
d397712b
CM		 2571 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2572 sb->s_id);
af31f5e5 2573 goto fail_tree_roots;
2b82032c 2574 }
0b86a832 2575
8dabb742
SB		 2576 /*
2577 * keep the device that is marked to be the target device for the
2578 * dev_replace procedure
2579 */
2580 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2581
a6b0d5c8
CM		 2582 if (!fs_devices->latest_bdev) {
2583 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2584 sb->s_id);
2585 goto fail_tree_roots;
2586 }
2587
af31f5e5 2588retry_root_backup:
db94535d
CM		 2589 blocksize = btrfs_level_size(tree_root,
2590 btrfs_super_root_level(disk_super));
84234f3a 2591 generation = btrfs_super_generation(disk_super);
0b86a832 2592
e20d96d6 2593 tree_root->node = read_tree_block(tree_root,
db94535d 2594 btrfs_super_root(disk_super),
84234f3a 2595 blocksize, generation);
af31f5e5
CM		 2596 if (!tree_root->node ||
2597 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
83121942
DW		 2598 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2599 sb->s_id);
af31f5e5
CM		 2600
2601 goto recovery_tree_root;
83121942 2602 }
af31f5e5 2603
5d4f98a2
YZ		 2604 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2605 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d
CM		 2606
2607 ret = find_and_setup_root(tree_root, fs_info,
e20d96d6 2608 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
0b86a832 2609 if (ret)
af31f5e5 2610 goto recovery_tree_root;
0b86a832
CM		 2611 extent_root->track_dirty = 1;
2612
2613 ret = find_and_setup_root(tree_root, fs_info,
2614 BTRFS_DEV_TREE_OBJECTID, dev_root);
0b86a832 2615 if (ret)
af31f5e5 2616 goto recovery_tree_root;
5d4f98a2 2617 dev_root->track_dirty = 1;
3768f368 2618
d20f7043
CM		 2619 ret = find_and_setup_root(tree_root, fs_info,
2620 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2621 if (ret)
af31f5e5 2622 goto recovery_tree_root;
d20f7043
CM		 2623 csum_root->track_dirty = 1;
2624
bcef60f2
AJ		 2625 ret = find_and_setup_root(tree_root, fs_info,
2626 BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2627 if (ret) {
2628 kfree(quota_root);
2629 quota_root = fs_info->quota_root = NULL;
2630 } else {
2631 quota_root->track_dirty = 1;
2632 fs_info->quota_enabled = 1;
2633 fs_info->pending_quota_state = 1;
2634 }
2635
8929ecfa
YZ		 2636 fs_info->generation = generation;
2637 fs_info->last_trans_committed = generation;
8929ecfa 2638
68310a5e
ID		 2639 ret = btrfs_recover_balance(fs_info);
2640 if (ret) {
2641 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2642 goto fail_block_groups;
2643 }
2644
733f4fbb
SB		 2645 ret = btrfs_init_dev_stats(fs_info);
2646 if (ret) {
2647 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2648 ret);
2649 goto fail_block_groups;
2650 }
2651
8dabb742
SB		 2652 ret = btrfs_init_dev_replace(fs_info);
2653 if (ret) {
2654 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2655 goto fail_block_groups;
2656 }
2657
2658 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2659
c59021f8 2660 ret = btrfs_init_space_info(fs_info);
2661 if (ret) {
2662 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2663 goto fail_block_groups;
2664 }
2665
1b1d1f66
JB		 2666 ret = btrfs_read_block_groups(extent_root);
2667 if (ret) {
2668 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2669 goto fail_block_groups;
2670 }
5af3e8cc
SB		 2671 fs_info->num_tolerated_disk_barrier_failures =
2672 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2673 if (fs_info->fs_devices->missing_devices >
2674 fs_info->num_tolerated_disk_barrier_failures &&
2675 !(sb->s_flags & MS_RDONLY)) {
2676 printk(KERN_WARNING
2677 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2678 goto fail_block_groups;
2679 }
9078a3e1 2680
a74a4b97
CM		 2681 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2682 "btrfs-cleaner");
57506d50 2683 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2684 goto fail_block_groups;
a74a4b97
CM		 2685
2686 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2687 tree_root,
2688 "btrfs-transaction");
57506d50 2689 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2690 goto fail_cleaner;
a74a4b97 2691
c289811c
CM		 2692 if (!btrfs_test_opt(tree_root, SSD) &&
2693 !btrfs_test_opt(tree_root, NOSSD) &&
2694 !fs_info->fs_devices->rotating) {
2695 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2696 "mode\n");
2697 btrfs_set_opt(fs_info->mount_opt, SSD);
2698 }
2699
21adbd5c
SB		 2700#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2701 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2702 ret = btrfsic_mount(tree_root, fs_devices,
2703 btrfs_test_opt(tree_root,
2704 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2705 1 : 0,
2706 fs_info->check_integrity_print_mask);
2707 if (ret)
2708 printk(KERN_WARNING "btrfs: failed to initialize"
2709 " integrity check module %s\n", sb->s_id);
2710 }
2711#endif
bcef60f2
AJ		 2712 ret = btrfs_read_qgroup_config(fs_info);
2713 if (ret)
2714 goto fail_trans_kthread;
21adbd5c 2715
acce952b 2716 /* do not make disk changes in broken FS */
68ce9682 2717 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2718 u64 bytenr = btrfs_super_log_root(disk_super);
2719
7c2ca468 2720 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2721 printk(KERN_WARNING "Btrfs log replay required "
2722 "on RO media\n");
7c2ca468 2723 err = -EIO;
bcef60f2 2724 goto fail_qgroup;
7c2ca468 2725 }
e02119d5
CM		 2726 blocksize =
2727 btrfs_level_size(tree_root,
2728 btrfs_super_log_root_level(disk_super));
d18a2c44 2729
6f07e42e 2730 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2731 if (!log_tree_root) {
2732 err = -ENOMEM;
bcef60f2 2733 goto fail_qgroup;
676e4c86 2734 }
e02119d5
CM		 2735
2736 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2737 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2738
2739 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2740 blocksize,
2741 generation + 1);
416bc658
JB		 2742 if (!log_tree_root->node ||
2743 !extent_buffer_uptodate(log_tree_root->node)) {
2744 printk(KERN_ERR "btrfs: failed to read log tree\n");
2745 free_extent_buffer(log_tree_root->node);
2746 kfree(log_tree_root);
2747 goto fail_trans_kthread;
2748 }
79787eaa 2749 /* returns with log_tree_root freed on success */
e02119d5 2750 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2751 if (ret) {
2752 btrfs_error(tree_root->fs_info, ret,
2753 "Failed to recover log tree");
2754 free_extent_buffer(log_tree_root->node);
2755 kfree(log_tree_root);
2756 goto fail_trans_kthread;
2757 }
e556ce2c
YZ		 2758
2759 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2760 ret = btrfs_commit_super(tree_root);
2761 if (ret)
2762 goto fail_trans_kthread;
e556ce2c 2763 }
e02119d5 2764 }
1a40e23b 2765
76dda93c 2766 ret = btrfs_find_orphan_roots(tree_root);
79787eaa
JM		 2767 if (ret)
2768 goto fail_trans_kthread;
76dda93c 2769
7c2ca468 2770 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2771 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2
ID		 2772 if (ret)
2773 goto fail_trans_kthread;
d68fc57b 2774
5d4f98a2 2775 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2776 if (ret < 0) {
2777 printk(KERN_WARNING
2778 "btrfs: failed to recover relocation\n");
2779 err = -EINVAL;
bcef60f2 2780 goto fail_qgroup;
d7ce5843 2781 }
7c2ca468 2782 }
1a40e23b 2783
3de4586c
CM		 2784 location.objectid = BTRFS_FS_TREE_OBJECTID;
2785 location.type = BTRFS_ROOT_ITEM_KEY;
2786 location.offset = (u64)-1;
2787
3de4586c
CM		 2788 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2789 if (!fs_info->fs_root)
bcef60f2 2790 goto fail_qgroup;
3140c9a3
DC		 2791 if (IS_ERR(fs_info->fs_root)) {
2792 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2793 goto fail_qgroup;
3140c9a3 2794 }
c289811c 2795
2b6ba629
ID		 2796 if (sb->s_flags & MS_RDONLY)
2797 return 0;
59641015 2798
2b6ba629
ID		 2799 down_read(&fs_info->cleanup_work_sem);
2800 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2801 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2802 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2803 close_ctree(tree_root);
2804 return ret;
2805 }
2806 up_read(&fs_info->cleanup_work_sem);
59641015 2807
2b6ba629
ID		 2808 ret = btrfs_resume_balance_async(fs_info);
2809 if (ret) {
2810 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2811 close_ctree(tree_root);
2812 return ret;
e3acc2a6
JB		 2813 }
2814
8dabb742
SB		 2815 ret = btrfs_resume_dev_replace_async(fs_info);
2816 if (ret) {
2817 pr_warn("btrfs: failed to resume dev_replace\n");
2818 close_ctree(tree_root);
2819 return ret;
2820 }
2821
ad2b2c80 2822 return 0;
39279cc3 2823
bcef60f2
AJ		 2824fail_qgroup:
2825 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2826fail_trans_kthread:
2827 kthread_stop(fs_info->transaction_kthread);
171f6537 2828 del_fs_roots(fs_info);
54067ae9 2829 btrfs_cleanup_transaction(fs_info->tree_root);
3f157a2f 2830fail_cleaner:
a74a4b97 2831 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2832
2833 /*
2834 * make sure we're done with the btree inode before we stop our
2835 * kthreads
2836 */
2837 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2838
1b1d1f66 2839fail_block_groups:
54067ae9 2840 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2841 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2842
2843fail_tree_roots:
2844 free_root_pointers(fs_info, 1);
2b8195bb 2845 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2846
39279cc3 2847fail_sb_buffer:
7abadb64 2848 btrfs_stop_all_workers(fs_info);
16cdcec7 2849fail_alloc:
4543df7e 2850fail_iput:
586e46e2
ID		 2851 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2852
4543df7e 2853 iput(fs_info->btree_inode);
963d678b
MX		 2854fail_delalloc_bytes:
2855 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 2856fail_dirty_metadata_bytes:
2857 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 2858fail_bdi:
7e662854 2859 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2860fail_srcu:
2861 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2862fail:
53b381b3 2863 btrfs_free_stripe_hash_table(fs_info);
586e46e2 2864 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 2865 return err;
af31f5e5
CM		 2866
2867recovery_tree_root:
af31f5e5
CM		 2868 if (!btrfs_test_opt(tree_root, RECOVERY))
2869 goto fail_tree_roots;
2870
2871 free_root_pointers(fs_info, 0);
2872
2873 /* don't use the log in recovery mode, it won't be valid */
2874 btrfs_set_super_log_root(disk_super, 0);
2875
2876 /* we can't trust the free space cache either */
2877 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2878
2879 ret = next_root_backup(fs_info, fs_info->super_copy,
2880 &num_backups_tried, &backup_index);
2881 if (ret == -1)
2882 goto fail_block_groups;
2883 goto retry_root_backup;
eb60ceac
CM		 2884}
2885
f2984462
CM		 2886static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2887{
f2984462
CM		 2888 if (uptodate) {
2889 set_buffer_uptodate(bh);
2890 } else {
442a4f63
SB		 2891 struct btrfs_device *device = (struct btrfs_device *)
2892 bh->b_private;
2893
606686ee
JB		 2894 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2895 "I/O error on %s\n",
2896 rcu_str_deref(device->name));
1259ab75
CM		 2897 /* note, we dont' set_buffer_write_io_error because we have
2898 * our own ways of dealing with the IO errors
2899 */
f2984462 2900 clear_buffer_uptodate(bh);
442a4f63 2901 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 2902 }
2903 unlock_buffer(bh);
2904 put_bh(bh);
2905}
2906
a512bbf8
YZ		 2907struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2908{
2909 struct buffer_head *bh;
2910 struct buffer_head *latest = NULL;
2911 struct btrfs_super_block *super;
2912 int i;
2913 u64 transid = 0;
2914 u64 bytenr;
2915
2916 /* we would like to check all the supers, but that would make
2917 * a btrfs mount succeed after a mkfs from a different FS.
2918 * So, we need to add a special mount option to scan for
2919 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2920 */
2921 for (i = 0; i < 1; i++) {
2922 bytenr = btrfs_sb_offset(i);
2923 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2924 break;
2925 bh = __bread(bdev, bytenr / 4096, 4096);
2926 if (!bh)
2927 continue;
2928
2929 super = (struct btrfs_super_block *)bh->b_data;
2930 if (btrfs_super_bytenr(super) != bytenr ||
cdb4c574 2931 super->magic != cpu_to_le64(BTRFS_MAGIC)) {
a512bbf8
YZ		 2932 brelse(bh);
2933 continue;
2934 }
2935
2936 if (!latest || btrfs_super_generation(super) > transid) {
2937 brelse(latest);
2938 latest = bh;
2939 transid = btrfs_super_generation(super);
2940 } else {
2941 brelse(bh);
2942 }
2943 }
2944 return latest;
2945}
2946
4eedeb75
HH		 2947/*
2948 * this should be called twice, once with wait == 0 and
2949 * once with wait == 1. When wait == 0 is done, all the buffer heads
2950 * we write are pinned.
2951 *
2952 * They are released when wait == 1 is done.
2953 * max_mirrors must be the same for both runs, and it indicates how
2954 * many supers on this one device should be written.
2955 *
2956 * max_mirrors == 0 means to write them all.
2957 */
a512bbf8
YZ		 2958static int write_dev_supers(struct btrfs_device *device,
2959 struct btrfs_super_block *sb,
2960 int do_barriers, int wait, int max_mirrors)
2961{
2962 struct buffer_head *bh;
2963 int i;
2964 int ret;
2965 int errors = 0;
2966 u32 crc;
2967 u64 bytenr;
a512bbf8
YZ		 2968
2969 if (max_mirrors == 0)
2970 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2971
a512bbf8
YZ		 2972 for (i = 0; i < max_mirrors; i++) {
2973 bytenr = btrfs_sb_offset(i);
2974 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2975 break;
2976
2977 if (wait) {
2978 bh = __find_get_block(device->bdev, bytenr / 4096,
2979 BTRFS_SUPER_INFO_SIZE);
2980 BUG_ON(!bh);
a512bbf8 2981 wait_on_buffer(bh);
4eedeb75
HH		 2982 if (!buffer_uptodate(bh))
2983 errors++;
2984
2985 /* drop our reference */
2986 brelse(bh);
2987
2988 /* drop the reference from the wait == 0 run */
2989 brelse(bh);
2990 continue;
a512bbf8
YZ		 2991 } else {
2992 btrfs_set_super_bytenr(sb, bytenr);
2993
2994 crc = ~(u32)0;
b0496686 2995 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 2996 BTRFS_CSUM_SIZE, crc,
2997 BTRFS_SUPER_INFO_SIZE -
2998 BTRFS_CSUM_SIZE);
2999 btrfs_csum_final(crc, sb->csum);
3000
4eedeb75
HH		 3001 /*
3002 * one reference for us, and we leave it for the
3003 * caller
3004 */
a512bbf8
YZ		 3005 bh = __getblk(device->bdev, bytenr / 4096,
3006 BTRFS_SUPER_INFO_SIZE);
3007 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3008
4eedeb75 3009 /* one reference for submit_bh */
a512bbf8 3010 get_bh(bh);
4eedeb75
HH		 3011
3012 set_buffer_uptodate(bh);
a512bbf8
YZ		 3013 lock_buffer(bh);
3014 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3015 bh->b_private = device;
a512bbf8
YZ		 3016 }
3017
387125fc
CM		 3018 /*
3019 * we fua the first super. The others we allow
3020 * to go down lazy.
3021 */
21adbd5c 3022 ret = btrfsic_submit_bh(WRITE_FUA, bh);
4eedeb75 3023 if (ret)
a512bbf8 3024 errors++;
a512bbf8
YZ		 3025 }
3026 return errors < i ? 0 : -1;
3027}
3028
387125fc
CM		 3029/*
3030 * endio for the write_dev_flush, this will wake anyone waiting
3031 * for the barrier when it is done
3032 */
3033static void btrfs_end_empty_barrier(struct bio *bio, int err)
3034{
3035 if (err) {
3036 if (err == -EOPNOTSUPP)
3037 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3038 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3039 }
3040 if (bio->bi_private)
3041 complete(bio->bi_private);
3042 bio_put(bio);
3043}
3044
3045/*
3046 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3047 * sent down. With wait == 1, it waits for the previous flush.
3048 *
3049 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3050 * capable
3051 */
3052static int write_dev_flush(struct btrfs_device *device, int wait)
3053{
3054 struct bio *bio;
3055 int ret = 0;
3056
3057 if (device->nobarriers)
3058 return 0;
3059
3060 if (wait) {
3061 bio = device->flush_bio;
3062 if (!bio)
3063 return 0;
3064
3065 wait_for_completion(&device->flush_wait);
3066
3067 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
606686ee
JB		 3068 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3069 rcu_str_deref(device->name));
387125fc 3070 device->nobarriers = 1;
5af3e8cc 3071 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3072 ret = -EIO;
5af3e8cc
SB		 3073 btrfs_dev_stat_inc_and_print(device,
3074 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3075 }
3076
3077 /* drop the reference from the wait == 0 run */
3078 bio_put(bio);
3079 device->flush_bio = NULL;
3080
3081 return ret;
3082 }
3083
3084 /*
3085 * one reference for us, and we leave it for the
3086 * caller
3087 */
9c017abc 3088 device->flush_bio = NULL;
387125fc
CM		 3089 bio = bio_alloc(GFP_NOFS, 0);
3090 if (!bio)
3091 return -ENOMEM;
3092
3093 bio->bi_end_io = btrfs_end_empty_barrier;
3094 bio->bi_bdev = device->bdev;
3095 init_completion(&device->flush_wait);
3096 bio->bi_private = &device->flush_wait;
3097 device->flush_bio = bio;
3098
3099 bio_get(bio);
21adbd5c 3100 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3101
3102 return 0;
3103}
3104
3105/*
3106 * send an empty flush down to each device in parallel,
3107 * then wait for them
3108 */
3109static int barrier_all_devices(struct btrfs_fs_info *info)
3110{
3111 struct list_head *head;
3112 struct btrfs_device *dev;
5af3e8cc
SB		 3113 int errors_send = 0;
3114 int errors_wait = 0;
387125fc
CM		 3115 int ret;
3116
3117 /* send down all the barriers */
3118 head = &info->fs_devices->devices;
3119 list_for_each_entry_rcu(dev, head, dev_list) {
3120 if (!dev->bdev) {
5af3e8cc 3121 errors_send++;
387125fc
CM		 3122 continue;
3123 }
3124 if (!dev->in_fs_metadata || !dev->writeable)
3125 continue;
3126
3127 ret = write_dev_flush(dev, 0);
3128 if (ret)
5af3e8cc 3129 errors_send++;
387125fc
CM		 3130 }
3131
3132 /* wait for all the barriers */
3133 list_for_each_entry_rcu(dev, head, dev_list) {
3134 if (!dev->bdev) {
5af3e8cc 3135 errors_wait++;
387125fc
CM		 3136 continue;
3137 }
3138 if (!dev->in_fs_metadata || !dev->writeable)
3139 continue;
3140
3141 ret = write_dev_flush(dev, 1);
3142 if (ret)
5af3e8cc 3143 errors_wait++;
387125fc 3144 }
5af3e8cc
SB		 3145 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3146 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3147 return -EIO;
3148 return 0;
3149}
3150
5af3e8cc
SB		 3151int btrfs_calc_num_tolerated_disk_barrier_failures(
3152 struct btrfs_fs_info *fs_info)
3153{
3154 struct btrfs_ioctl_space_info space;
3155 struct btrfs_space_info *sinfo;
3156 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3157 BTRFS_BLOCK_GROUP_SYSTEM,
3158 BTRFS_BLOCK_GROUP_METADATA,
3159 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3160 int num_types = 4;
3161 int i;
3162 int c;
3163 int num_tolerated_disk_barrier_failures =
3164 (int)fs_info->fs_devices->num_devices;
3165
3166 for (i = 0; i < num_types; i++) {
3167 struct btrfs_space_info *tmp;
3168
3169 sinfo = NULL;
3170 rcu_read_lock();
3171 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3172 if (tmp->flags == types[i]) {
3173 sinfo = tmp;
3174 break;
3175 }
3176 }
3177 rcu_read_unlock();
3178
3179 if (!sinfo)
3180 continue;
3181
3182 down_read(&sinfo->groups_sem);
3183 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3184 if (!list_empty(&sinfo->block_groups[c])) {
3185 u64 flags;
3186
3187 btrfs_get_block_group_info(
3188 &sinfo->block_groups[c], &space);
3189 if (space.total_bytes == 0 ||
3190 space.used_bytes == 0)
3191 continue;
3192 flags = space.flags;
3193 /*
3194 * return
3195 * 0: if dup, single or RAID0 is configured for
3196 * any of metadata, system or data, else
3197 * 1: if RAID5 is configured, or if RAID1 or
3198 * RAID10 is configured and only two mirrors
3199 * are used, else
3200 * 2: if RAID6 is configured, else
3201 * num_mirrors - 1: if RAID1 or RAID10 is
3202 * configured and more than
3203 * 2 mirrors are used.
3204 */
3205 if (num_tolerated_disk_barrier_failures > 0 &&
3206 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3207 BTRFS_BLOCK_GROUP_RAID0)) ||
3208 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3209 == 0)))
3210 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3211 else if (num_tolerated_disk_barrier_failures > 1) {
3212 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3213 BTRFS_BLOCK_GROUP_RAID5 |
3214 BTRFS_BLOCK_GROUP_RAID10)) {
3215 num_tolerated_disk_barrier_failures = 1;
3216 } else if (flags &
3217 BTRFS_BLOCK_GROUP_RAID5) {
3218 num_tolerated_disk_barrier_failures = 2;
3219 }
3220 }
5af3e8cc
SB		 3221 }
3222 }
3223 up_read(&sinfo->groups_sem);
3224 }
3225
3226 return num_tolerated_disk_barrier_failures;
3227}
3228
a512bbf8 3229int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3230{
e5e9a520 3231 struct list_head *head;
f2984462 3232 struct btrfs_device *dev;
a061fc8d 3233 struct btrfs_super_block *sb;
f2984462 3234 struct btrfs_dev_item *dev_item;
f2984462
CM		 3235 int ret;
3236 int do_barriers;
a236aed1
CM		 3237 int max_errors;
3238 int total_errors = 0;
a061fc8d 3239 u64 flags;
f2984462 3240
6c41761f 3241 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
f2984462 3242 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3243 backup_super_roots(root->fs_info);
f2984462 3244
6c41761f 3245 sb = root->fs_info->super_for_commit;
a061fc8d 3246 dev_item = &sb->dev_item;
e5e9a520 3247
174ba509 3248 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3249 head = &root->fs_info->fs_devices->devices;
387125fc 3250
5af3e8cc
SB		 3251 if (do_barriers) {
3252 ret = barrier_all_devices(root->fs_info);
3253 if (ret) {
3254 mutex_unlock(
3255 &root->fs_info->fs_devices->device_list_mutex);
3256 btrfs_error(root->fs_info, ret,
3257 "errors while submitting device barriers.");
3258 return ret;
3259 }
3260 }
387125fc 3261
1f78160c 3262 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3263 if (!dev->bdev) {
3264 total_errors++;
3265 continue;
3266 }
2b82032c 3267 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3268 continue;
3269
2b82032c 3270 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3271 btrfs_set_stack_device_type(dev_item, dev->type);
3272 btrfs_set_stack_device_id(dev_item, dev->devid);
3273 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3274 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3275 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3276 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3277 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3278 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3279 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3280
a061fc8d
CM		 3281 flags = btrfs_super_flags(sb);
3282 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3283
a512bbf8 3284 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3285 if (ret)
3286 total_errors++;
f2984462 3287 }
a236aed1 3288 if (total_errors > max_errors) {
d397712b
CM		 3289 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3290 total_errors);
79787eaa
JM		 3291
3292 /* This shouldn't happen. FUA is masked off if unsupported */
a236aed1
CM		 3293 BUG();
3294 }
f2984462 3295
a512bbf8 3296 total_errors = 0;
1f78160c 3297 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3298 if (!dev->bdev)
3299 continue;
2b82032c 3300 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3301 continue;
3302
a512bbf8
YZ		 3303 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3304 if (ret)
3305 total_errors++;
f2984462 3306 }
174ba509 3307 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3308 if (total_errors > max_errors) {
79787eaa
JM		 3309 btrfs_error(root->fs_info, -EIO,
3310 "%d errors while writing supers", total_errors);
3311 return -EIO;
a236aed1 3312 }
f2984462
CM		 3313 return 0;
3314}
3315
a512bbf8
YZ		 3316int write_ctree_super(struct btrfs_trans_handle *trans,
3317 struct btrfs_root *root, int max_mirrors)
eb60ceac 3318{
e66f709b 3319 int ret;
5f39d397 3320
a512bbf8 3321 ret = write_all_supers(root, max_mirrors);
5f39d397 3322 return ret;
cfaa7295
CM		 3323}
3324
143bede5 3325void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 3326{
4df27c4d 3327 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3328 radix_tree_delete(&fs_info->fs_roots_radix,
3329 (unsigned long)root->root_key.objectid);
4df27c4d 3330 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3331
3332 if (btrfs_root_refs(&root->root_item) == 0)
3333 synchronize_srcu(&fs_info->subvol_srcu);
3334
d7634482 3335 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3321719e
LB		 3336 btrfs_free_log(NULL, root);
3337 btrfs_free_log_root_tree(NULL, fs_info);
3338 }
3339
581bb050
LZ		 3340 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3341 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3342 free_fs_root(root);
4df27c4d
YZ		 3343}
3344
3345static void free_fs_root(struct btrfs_root *root)
3346{
82d5902d 3347 iput(root->cache_inode);
4df27c4d 3348 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
0ee5dc67
AV		 3349 if (root->anon_dev)
3350 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 3351 free_extent_buffer(root->node);
3352 free_extent_buffer(root->commit_root);
581bb050
LZ		 3353 kfree(root->free_ino_ctl);
3354 kfree(root->free_ino_pinned);
d397712b 3355 kfree(root->name);
2619ba1f 3356 kfree(root);
2619ba1f
CM		 3357}
3358
c146afad 3359int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3360{
c146afad
YZ		 3361 u64 root_objectid = 0;
3362 struct btrfs_root *gang[8];
3363 int i;
3768f368 3364 int ret;
e089f05c 3365
c146afad
YZ		 3366 while (1) {
3367 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3368 (void **)gang, root_objectid,
3369 ARRAY_SIZE(gang));
3370 if (!ret)
3371 break;
5d4f98a2
YZ		 3372
3373 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 3374 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 3375 int err;
3376
c146afad 3377 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3378 err = btrfs_orphan_cleanup(gang[i]);
3379 if (err)
3380 return err;
c146afad
YZ		 3381 }
3382 root_objectid++;
3383 }
3384 return 0;
3385}
a2135011 3386
c146afad
YZ		 3387int btrfs_commit_super(struct btrfs_root *root)
3388{
3389 struct btrfs_trans_handle *trans;
3390 int ret;
a74a4b97 3391
c146afad 3392 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3393 btrfs_run_delayed_iputs(root);
c146afad 3394 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3395 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3396
3397 /* wait until ongoing cleanup work done */
3398 down_write(&root->fs_info->cleanup_work_sem);
3399 up_write(&root->fs_info->cleanup_work_sem);
3400
7a7eaa40 3401 trans = btrfs_join_transaction(root);
3612b495
TI		 3402 if (IS_ERR(trans))
3403 return PTR_ERR(trans);
54aa1f4d 3404 ret = btrfs_commit_transaction(trans, root);
79787eaa
JM		 3405 if (ret)
3406 return ret;
c146afad 3407 /* run commit again to drop the original snapshot */
7a7eaa40 3408 trans = btrfs_join_transaction(root);
3612b495
TI		 3409 if (IS_ERR(trans))
3410 return PTR_ERR(trans);
79787eaa
JM		 3411 ret = btrfs_commit_transaction(trans, root);
3412 if (ret)
3413 return ret;
79154b1b 3414 ret = btrfs_write_and_wait_transaction(NULL, root);
79787eaa
JM		 3415 if (ret) {
3416 btrfs_error(root->fs_info, ret,
3417 "Failed to sync btree inode to disk.");
3418 return ret;
3419 }
d6bfde87 3420
a512bbf8 3421 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 3422 return ret;
3423}
3424
3425int close_ctree(struct btrfs_root *root)
3426{
3427 struct btrfs_fs_info *fs_info = root->fs_info;
3428 int ret;
3429
3430 fs_info->closing = 1;
3431 smp_mb();
3432
837d5b6e 3433 /* pause restriper - we want to resume on mount */
aa1b8cd4 3434 btrfs_pause_balance(fs_info);
837d5b6e 3435
8dabb742
SB		 3436 btrfs_dev_replace_suspend_for_unmount(fs_info);
3437
aa1b8cd4 3438 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3439
3440 /* wait for any defraggers to finish */
3441 wait_event(fs_info->transaction_wait,
3442 (atomic_read(&fs_info->defrag_running) == 0));
3443
3444 /* clear out the rbtree of defraggable inodes */
26176e7c 3445 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3446
c146afad 3447 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3448 ret = btrfs_commit_super(root);
3449 if (ret)
3450 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3451 }
3452
87533c47 3453 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3454 btrfs_error_commit_super(root);
0f7d52f4 3455
300e4f8a
JB		 3456 btrfs_put_block_group_cache(fs_info);
3457
e3029d9f
AV		 3458 kthread_stop(fs_info->transaction_kthread);
3459 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3460
f25784b3
YZ		 3461 fs_info->closing = 2;
3462 smp_mb();
3463
bcef60f2
AJ		 3464 btrfs_free_qgroup_config(root->fs_info);
3465
963d678b
MX		 3466 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3467 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3468 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3469 }
bcc63abb 3470
be283b2e 3471 free_root_pointers(fs_info, 1);
d20f7043 3472
e3029d9f 3473 btrfs_free_block_groups(fs_info);
d10c5f31 3474
c146afad 3475 del_fs_roots(fs_info);
d10c5f31 3476
c146afad 3477 iput(fs_info->btree_inode);
9ad6b7bc 3478
7abadb64 3479 btrfs_stop_all_workers(fs_info);
d6bfde87 3480
21adbd5c
SB		 3481#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3482 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3483 btrfsic_unmount(root, fs_info->fs_devices);
3484#endif
3485
dfe25020 3486 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3487 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3488
e2d84521 3489 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3490 percpu_counter_destroy(&fs_info->delalloc_bytes);
04160088 3491 bdi_destroy(&fs_info->bdi);
76dda93c 3492 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3493
53b381b3
DW		 3494 btrfs_free_stripe_hash_table(fs_info);
3495
eb60ceac
CM		 3496 return 0;
3497}
3498
b9fab919
CM		 3499int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3500 int atomic)
5f39d397 3501{
1259ab75 3502 int ret;
727011e0 3503 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3504
0b32f4bb 3505 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3506 if (!ret)
3507 return ret;
3508
3509 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3510 parent_transid, atomic);
3511 if (ret == -EAGAIN)
3512 return ret;
1259ab75 3513 return !ret;
5f39d397
CM		 3514}
3515
3516int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3517{
0b32f4bb 3518 return set_extent_buffer_uptodate(buf);
5f39d397 3519}
6702ed49 3520
5f39d397
CM		 3521void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3522{
727011e0 3523 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
5f39d397 3524 u64 transid = btrfs_header_generation(buf);
b9473439 3525 int was_dirty;
b4ce94de 3526
b9447ef8 3527 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3528 if (transid != root->fs_info->generation)
3529 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3530 "found %llu running %llu\n",
db94535d 3531 (unsigned long long)buf->start,
d397712b
CM		 3532 (unsigned long long)transid,
3533 (unsigned long long)root->fs_info->generation);
0b32f4bb 3534 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3535 if (!was_dirty)
3536 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3537 buf->len,
3538 root->fs_info->dirty_metadata_batch);
eb60ceac
CM		 3539}
3540
b53d3f5d
LB		 3541static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3542 int flush_delayed)
16cdcec7
MX		 3543{
3544 /*
3545 * looks as though older kernels can get into trouble with
3546 * this code, they end up stuck in balance_dirty_pages forever
3547 */
e2d84521 3548 int ret;
16cdcec7
MX		 3549
3550 if (current->flags & PF_MEMALLOC)
3551 return;
3552
b53d3f5d
LB		 3553 if (flush_delayed)
3554 btrfs_balance_delayed_items(root);
16cdcec7 3555
e2d84521
MX		 3556 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3557 BTRFS_DIRTY_METADATA_THRESH);
3558 if (ret > 0) {
d0e1d66b
NJ		 3559 balance_dirty_pages_ratelimited(
3560 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3561 }
3562 return;
3563}
3564
b53d3f5d 3565void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3566{
b53d3f5d
LB		 3567 __btrfs_btree_balance_dirty(root, 1);
3568}
585ad2c3 3569
b53d3f5d
LB		 3570void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3571{
3572 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3573}
6b80053d 3574
ca7a79ad 3575int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3576{
727011e0 3577 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3578 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3579}
0da5468f 3580
fcd1f065 3581static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3582 int read_only)
3583{
fcd1f065
DS		 3584 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3585 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3586 return -EINVAL;
3587 }
3588
acce952b 3589 if (read_only)
fcd1f065 3590 return 0;
acce952b 3591
fcd1f065 3592 return 0;
acce952b 3593}
3594
68ce9682 3595void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3596{
acce952b 3597 mutex_lock(&root->fs_info->cleaner_mutex);
3598 btrfs_run_delayed_iputs(root);
3599 mutex_unlock(&root->fs_info->cleaner_mutex);
3600
3601 down_write(&root->fs_info->cleanup_work_sem);
3602 up_write(&root->fs_info->cleanup_work_sem);
3603
3604 /* cleanup FS via transaction */
3605 btrfs_cleanup_transaction(root);
acce952b 3606}
3607
569e0f35
JB		 3608static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3609 struct btrfs_root *root)
acce952b 3610{
3611 struct btrfs_inode *btrfs_inode;
3612 struct list_head splice;
3613
3614 INIT_LIST_HEAD(&splice);
3615
3616 mutex_lock(&root->fs_info->ordered_operations_mutex);
3617 spin_lock(&root->fs_info->ordered_extent_lock);
3618
569e0f35 3619 list_splice_init(&t->ordered_operations, &splice);
acce952b 3620 while (!list_empty(&splice)) {
3621 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3622 ordered_operations);
3623
3624 list_del_init(&btrfs_inode->ordered_operations);
3625
3626 btrfs_invalidate_inodes(btrfs_inode->root);
3627 }
3628
3629 spin_unlock(&root->fs_info->ordered_extent_lock);
3630 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3631}
3632
143bede5 3633static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3634{
acce952b 3635 struct btrfs_ordered_extent *ordered;
acce952b 3636
3637 spin_lock(&root->fs_info->ordered_extent_lock);
779880ef
JB		 3638 /*
3639 * This will just short circuit the ordered completion stuff which will
3640 * make sure the ordered extent gets properly cleaned up.
3641 */
3642 list_for_each_entry(ordered, &root->fs_info->ordered_extents,
3643 root_extent_list)
3644 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
acce952b 3645 spin_unlock(&root->fs_info->ordered_extent_lock);
acce952b 3646}
3647
79787eaa
JM		 3648int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3649 struct btrfs_root *root)
acce952b 3650{
3651 struct rb_node *node;
3652 struct btrfs_delayed_ref_root *delayed_refs;
3653 struct btrfs_delayed_ref_node *ref;
3654 int ret = 0;
3655
3656 delayed_refs = &trans->delayed_refs;
3657
3658 spin_lock(&delayed_refs->lock);
3659 if (delayed_refs->num_entries == 0) {
cfece4db 3660 spin_unlock(&delayed_refs->lock);
acce952b 3661 printk(KERN_INFO "delayed_refs has NO entry\n");
3662 return ret;
3663 }
3664
b939d1ab 3665 while ((node = rb_first(&delayed_refs->root)) != NULL) {
eb12db69 3666 struct btrfs_delayed_ref_head *head = NULL;
acce952b 3667
eb12db69 3668 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
acce952b 3669 atomic_set(&ref->refs, 1);
3670 if (btrfs_delayed_ref_is_head(ref)) {
acce952b 3671
3672 head = btrfs_delayed_node_to_head(ref);
b939d1ab
JB		 3673 if (!mutex_trylock(&head->mutex)) {
3674 atomic_inc(&ref->refs);
3675 spin_unlock(&delayed_refs->lock);
3676
3677 /* Need to wait for the delayed ref to run */
3678 mutex_lock(&head->mutex);
3679 mutex_unlock(&head->mutex);
3680 btrfs_put_delayed_ref(ref);
3681
e18fca73 3682 spin_lock(&delayed_refs->lock);
b939d1ab
JB		 3683 continue;
3684 }
3685
54067ae9
JB		 3686 if (head->must_insert_reserved)
3687 btrfs_pin_extent(root, ref->bytenr,
3688 ref->num_bytes, 1);
78a6184a 3689 btrfs_free_delayed_extent_op(head->extent_op);
acce952b 3690 delayed_refs->num_heads--;
3691 if (list_empty(&head->cluster))
3692 delayed_refs->num_heads_ready--;
3693 list_del_init(&head->cluster);
acce952b 3694 }
eb12db69 3695
b939d1ab
JB		 3696 ref->in_tree = 0;
3697 rb_erase(&ref->rb_node, &delayed_refs->root);
3698 delayed_refs->num_entries--;
eb12db69
JB		 3699 if (head)
3700 mutex_unlock(&head->mutex);
acce952b 3701 spin_unlock(&delayed_refs->lock);
3702 btrfs_put_delayed_ref(ref);
3703
3704 cond_resched();
3705 spin_lock(&delayed_refs->lock);
3706 }
3707
3708 spin_unlock(&delayed_refs->lock);
3709
3710 return ret;
3711}
3712
aec8030a 3713static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
acce952b 3714{
3715 struct btrfs_pending_snapshot *snapshot;
3716 struct list_head splice;
3717
3718 INIT_LIST_HEAD(&splice);
3719
3720 list_splice_init(&t->pending_snapshots, &splice);
3721
3722 while (!list_empty(&splice)) {
3723 snapshot = list_entry(splice.next,
3724 struct btrfs_pending_snapshot,
3725 list);
aec8030a 3726 snapshot->error = -ECANCELED;
acce952b 3727 list_del_init(&snapshot->list);
acce952b 3728 }
acce952b 3729}
3730
143bede5 3731static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3732{
3733 struct btrfs_inode *btrfs_inode;
3734 struct list_head splice;
3735
3736 INIT_LIST_HEAD(&splice);
3737
acce952b 3738 spin_lock(&root->fs_info->delalloc_lock);
5be76758 3739 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
acce952b 3740
3741 while (!list_empty(&splice)) {
3742 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3743 delalloc_inodes);
3744
3745 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3746 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3747 &btrfs_inode->runtime_flags);
acce952b 3748
3749 btrfs_invalidate_inodes(btrfs_inode->root);
3750 }
3751
3752 spin_unlock(&root->fs_info->delalloc_lock);
acce952b 3753}
3754
3755static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3756 struct extent_io_tree *dirty_pages,
3757 int mark)
3758{
3759 int ret;
acce952b 3760 struct extent_buffer *eb;
3761 u64 start = 0;
3762 u64 end;
acce952b 3763
3764 while (1) {
3765 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3766 mark, NULL);
acce952b 3767 if (ret)
3768 break;
3769
3770 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3771 while (start <= end) {
fd8b2b61
JB		 3772 eb = btrfs_find_tree_block(root, start,
3773 root->leafsize);
3774 start += eb->len;
3775 if (!eb)
acce952b 3776 continue;
fd8b2b61 3777 wait_on_extent_buffer_writeback(eb);
acce952b 3778
fd8b2b61
JB		 3779 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3780 &eb->bflags))
3781 clear_extent_buffer_dirty(eb);
3782 free_extent_buffer_stale(eb);
acce952b 3783 }
3784 }
3785
3786 return ret;
3787}
3788
3789static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3790 struct extent_io_tree *pinned_extents)
3791{
3792 struct extent_io_tree *unpin;
3793 u64 start;
3794 u64 end;
3795 int ret;
ed0eaa14 3796 bool loop = true;
acce952b 3797
3798 unpin = pinned_extents;
ed0eaa14 3799again:
acce952b 3800 while (1) {
3801 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 3802 EXTENT_DIRTY, NULL);
acce952b 3803 if (ret)
3804 break;
3805
3806 /* opt_discard */
5378e607
LD		 3807 if (btrfs_test_opt(root, DISCARD))
3808 ret = btrfs_error_discard_extent(root, start,
3809 end + 1 - start,
3810 NULL);
acce952b 3811
3812 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3813 btrfs_error_unpin_extent_range(root, start, end);
3814 cond_resched();
3815 }
3816
ed0eaa14
LB		 3817 if (loop) {
3818 if (unpin == &root->fs_info->freed_extents[0])
3819 unpin = &root->fs_info->freed_extents[1];
3820 else
3821 unpin = &root->fs_info->freed_extents[0];
3822 loop = false;
3823 goto again;
3824 }
3825
acce952b 3826 return 0;
3827}
3828
49b25e05
JM		 3829void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3830 struct btrfs_root *root)
3831{
3832 btrfs_destroy_delayed_refs(cur_trans, root);
3833 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3834 cur_trans->dirty_pages.dirty_bytes);
3835
3836 /* FIXME: cleanup wait for commit */
3837 cur_trans->in_commit = 1;
3838 cur_trans->blocked = 1;
d7096fc3 3839 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 3840
aec8030a
MX		 3841 btrfs_evict_pending_snapshots(cur_trans);
3842
49b25e05 3843 cur_trans->blocked = 0;
d7096fc3 3844 wake_up(&root->fs_info->transaction_wait);
49b25e05
JM		 3845
3846 cur_trans->commit_done = 1;
d7096fc3 3847 wake_up(&cur_trans->commit_wait);
49b25e05 3848
67cde344
MX		 3849 btrfs_destroy_delayed_inodes(root);
3850 btrfs_assert_delayed_root_empty(root);
49b25e05 3851
49b25e05
JM		 3852 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3853 EXTENT_DIRTY);
6e841e32
LB		 3854 btrfs_destroy_pinned_extent(root,
3855 root->fs_info->pinned_extents);
49b25e05
JM		 3856
3857 /*
3858 memset(cur_trans, 0, sizeof(*cur_trans));
3859 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3860 */
3861}
3862
3863int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 3864{
3865 struct btrfs_transaction *t;
3866 LIST_HEAD(list);
3867
acce952b 3868 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3869
a4abeea4 3870 spin_lock(&root->fs_info->trans_lock);
acce952b 3871 list_splice_init(&root->fs_info->trans_list, &list);
a4abeea4
JB		 3872 root->fs_info->trans_no_join = 1;
3873 spin_unlock(&root->fs_info->trans_lock);
3874
acce952b 3875 while (!list_empty(&list)) {
3876 t = list_entry(list.next, struct btrfs_transaction, list);
acce952b 3877
569e0f35 3878 btrfs_destroy_ordered_operations(t, root);
acce952b 3879
3880 btrfs_destroy_ordered_extents(root);
3881
3882 btrfs_destroy_delayed_refs(t, root);
3883
acce952b 3884 /* FIXME: cleanup wait for commit */
3885 t->in_commit = 1;
3886 t->blocked = 1;
66657b31 3887 smp_mb();
acce952b 3888 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3889 wake_up(&root->fs_info->transaction_blocked_wait);
3890
aec8030a
MX		 3891 btrfs_evict_pending_snapshots(t);
3892
acce952b 3893 t->blocked = 0;
66657b31 3894 smp_mb();
acce952b 3895 if (waitqueue_active(&root->fs_info->transaction_wait))
3896 wake_up(&root->fs_info->transaction_wait);
acce952b 3897
acce952b 3898 t->commit_done = 1;
66657b31 3899 smp_mb();
acce952b 3900 if (waitqueue_active(&t->commit_wait))
3901 wake_up(&t->commit_wait);
acce952b 3902
67cde344
MX		 3903 btrfs_destroy_delayed_inodes(root);
3904 btrfs_assert_delayed_root_empty(root);
3905
acce952b 3906 btrfs_destroy_delalloc_inodes(root);
3907
a4abeea4 3908 spin_lock(&root->fs_info->trans_lock);
acce952b 3909 root->fs_info->running_transaction = NULL;
a4abeea4 3910 spin_unlock(&root->fs_info->trans_lock);
acce952b 3911
3912 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3913 EXTENT_DIRTY);
3914
3915 btrfs_destroy_pinned_extent(root,
3916 root->fs_info->pinned_extents);
3917
13c5a93e 3918 atomic_set(&t->use_count, 0);
acce952b 3919 list_del_init(&t->list);
3920 memset(t, 0, sizeof(*t));
3921 kmem_cache_free(btrfs_transaction_cachep, t);
3922 }
3923
a4abeea4
JB		 3924 spin_lock(&root->fs_info->trans_lock);
3925 root->fs_info->trans_no_join = 0;
3926 spin_unlock(&root->fs_info->trans_lock);
acce952b 3927 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 3928
3929 return 0;
3930}
3931
d1310b2e 3932static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 3933 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 3934 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 3935 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 3936 /* note we're sharing with inode.c for the merge bio hook */
3937 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 3938};
Linux 6.x block layer and io_uring tree(s)