projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: deal with errors in write_dev_supers
[linux-2.6-block.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
163e783e 29#include <linux/crc32c.h>
5a0e3ad6 30#include <linux/slab.h>
784b4e29 31#include <linux/migrate.h>
7a36ddec 32#include <linux/ratelimit.h>
6463fe58 33#include <linux/uuid.h>
7e75bf3f 34#include <asm/unaligned.h>
4b4e25f2 35#include "compat.h"
eb60ceac
CM		 36#include "ctree.h"
37#include "disk-io.h"
e089f05c 38#include "transaction.h"
0f7d52f4 39#include "btrfs_inode.h"
0b86a832 40#include "volumes.h"
db94535d 41#include "print-tree.h"
8b712842 42#include "async-thread.h"
925baedd 43#include "locking.h"
e02119d5 44#include "tree-log.h"
fa9c0d79 45#include "free-space-cache.h"
581bb050 46#include "inode-map.h"
21adbd5c 47#include "check-integrity.h"
606686ee 48#include "rcu-string.h"
8dabb742 49#include "dev-replace.h"
53b381b3 50#include "raid56.h"
eb60ceac 51
de0022b9
JB		 52#ifdef CONFIG_X86
53#include <asm/cpufeature.h>
54#endif
55
d1310b2e 56static struct extent_io_ops btree_extent_io_ops;
8b712842 57static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 58static void free_fs_root(struct btrfs_root *root);
fcd1f065 59static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 60 int read_only);
569e0f35
JB		 61static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
62 struct btrfs_root *root);
143bede5 63static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
acce952b 64static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
65 struct btrfs_root *root);
aec8030a 66static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
143bede5 67static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
acce952b 68static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 struct extent_io_tree *dirty_pages,
70 int mark);
71static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 struct extent_io_tree *pinned_extents);
ce9adaa5 73
d352ac68
CM		 74/*
75 * end_io_wq structs are used to do processing in task context when an IO is
76 * complete. This is used during reads to verify checksums, and it is used
77 * by writes to insert metadata for new file extents after IO is complete.
78 */
ce9adaa5
CM		 79struct end_io_wq {
80 struct bio *bio;
81 bio_end_io_t *end_io;
82 void *private;
83 struct btrfs_fs_info *info;
84 int error;
22c59948 85 int metadata;
ce9adaa5 86 struct list_head list;
8b712842 87 struct btrfs_work work;
ce9adaa5 88};
0da5468f 89
d352ac68
CM		 90/*
91 * async submit bios are used to offload expensive checksumming
92 * onto the worker threads. They checksum file and metadata bios
93 * just before they are sent down the IO stack.
94 */
44b8bd7e
CM		 95struct async_submit_bio {
96 struct inode *inode;
97 struct bio *bio;
98 struct list_head list;
4a69a410
CM		 99 extent_submit_bio_hook_t *submit_bio_start;
100 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 101 int rw;
102 int mirror_num;
c8b97818 103 unsigned long bio_flags;
eaf25d93
CM		 104 /*
105 * bio_offset is optional, can be used if the pages in the bio
106 * can't tell us where in the file the bio should go
107 */
108 u64 bio_offset;
8b712842 109 struct btrfs_work work;
79787eaa 110 int error;
44b8bd7e
CM		 111};
112
85d4e461
CM		 113/*
114 * Lockdep class keys for extent_buffer->lock's in this root. For a given
115 * eb, the lockdep key is determined by the btrfs_root it belongs to and
116 * the level the eb occupies in the tree.
117 *
118 * Different roots are used for different purposes and may nest inside each
119 * other and they require separate keysets. As lockdep keys should be
120 * static, assign keysets according to the purpose of the root as indicated
121 * by btrfs_root->objectid. This ensures that all special purpose roots
122 * have separate keysets.
4008c04a 123 *
85d4e461
CM		 124 * Lock-nesting across peer nodes is always done with the immediate parent
125 * node locked thus preventing deadlock. As lockdep doesn't know this, use
126 * subclass to avoid triggering lockdep warning in such cases.
4008c04a 127 *
85d4e461
CM		 128 * The key is set by the readpage_end_io_hook after the buffer has passed
129 * csum validation but before the pages are unlocked. It is also set by
130 * btrfs_init_new_buffer on freshly allocated blocks.
4008c04a 131 *
85d4e461
CM		 132 * We also add a check to make sure the highest level of the tree is the
133 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
134 * needs update as well.
4008c04a
CM		 135 */
136#ifdef CONFIG_DEBUG_LOCK_ALLOC
137# if BTRFS_MAX_LEVEL != 8
138# error
139# endif
85d4e461
CM		 140
141static struct btrfs_lockdep_keyset {
142 u64 id; /* root objectid */
143 const char *name_stem; /* lock name stem */
144 char names[BTRFS_MAX_LEVEL + 1][20];
145 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
146} btrfs_lockdep_keysets[] = {
147 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
148 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
149 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
150 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
151 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
152 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
153 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
154 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
155 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
156 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
157 { .id = 0, .name_stem = "tree" },
4008c04a 158};
85d4e461
CM		 159
160void __init btrfs_init_lockdep(void)
161{
162 int i, j;
163
164 /* initialize lockdep class names */
165 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
166 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
167
168 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
169 snprintf(ks->names[j], sizeof(ks->names[j]),
170 "btrfs-%s-%02d", ks->name_stem, j);
171 }
172}
173
174void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
175 int level)
176{
177 struct btrfs_lockdep_keyset *ks;
178
179 BUG_ON(level >= ARRAY_SIZE(ks->keys));
180
181 /* find the matching keyset, id 0 is the default entry */
182 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
183 if (ks->id == objectid)
184 break;
185
186 lockdep_set_class_and_name(&eb->lock,
187 &ks->keys[level], ks->names[level]);
188}
189
4008c04a
CM		 190#endif
191
d352ac68
CM		 192/*
193 * extents on the btree inode are pretty simple, there's one extent
194 * that covers the entire device
195 */
b2950863 196static struct extent_map *btree_get_extent(struct inode *inode,
306e16ce 197 struct page *page, size_t pg_offset, u64 start, u64 len,
b2950863 198 int create)
7eccb903 199{
5f39d397
CM		 200 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
201 struct extent_map *em;
202 int ret;
203
890871be 204 read_lock(&em_tree->lock);
d1310b2e 205 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 206 if (em) {
207 em->bdev =
208 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 209 read_unlock(&em_tree->lock);
5f39d397 210 goto out;
a061fc8d 211 }
890871be 212 read_unlock(&em_tree->lock);
7b13b7b1 213
172ddd60 214 em = alloc_extent_map();
5f39d397
CM		 215 if (!em) {
216 em = ERR_PTR(-ENOMEM);
217 goto out;
218 }
219 em->start = 0;
0afbaf8c 220 em->len = (u64)-1;
c8b97818 221 em->block_len = (u64)-1;
5f39d397 222 em->block_start = 0;
a061fc8d 223 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 224
890871be 225 write_lock(&em_tree->lock);
09a2a8f9 226 ret = add_extent_mapping(em_tree, em, 0);
5f39d397
CM		 227 if (ret == -EEXIST) {
228 free_extent_map(em);
7b13b7b1 229 em = lookup_extent_mapping(em_tree, start, len);
b4f359ab 230 if (!em)
0433f20d 231 em = ERR_PTR(-EIO);
5f39d397 232 } else if (ret) {
7b13b7b1 233 free_extent_map(em);
0433f20d 234 em = ERR_PTR(ret);
5f39d397 235 }
890871be 236 write_unlock(&em_tree->lock);
7b13b7b1 237
5f39d397
CM		 238out:
239 return em;
7eccb903
CM		 240}
241
b0496686 242u32 btrfs_csum_data(char *data, u32 seed, size_t len)
19c00ddc 243{
163e783e 244 return crc32c(seed, data, len);
19c00ddc
CM		 245}
246
247void btrfs_csum_final(u32 crc, char *result)
248{
7e75bf3f 249 put_unaligned_le32(~crc, result);
19c00ddc
CM		 250}
251
d352ac68
CM		 252/*
253 * compute the csum for a btree block, and either verify it or write it
254 * into the csum field of the block.
255 */
19c00ddc
CM		 256static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
257 int verify)
258{
6c41761f 259 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
607d432d 260 char *result = NULL;
19c00ddc
CM		 261 unsigned long len;
262 unsigned long cur_len;
263 unsigned long offset = BTRFS_CSUM_SIZE;
19c00ddc
CM		 264 char *kaddr;
265 unsigned long map_start;
266 unsigned long map_len;
267 int err;
268 u32 crc = ~(u32)0;
607d432d 269 unsigned long inline_result;
19c00ddc
CM		 270
271 len = buf->len - offset;
d397712b 272 while (len > 0) {
19c00ddc 273 err = map_private_extent_buffer(buf, offset, 32,
a6591715 274 &kaddr, &map_start, &map_len);
d397712b 275 if (err)
19c00ddc 276 return 1;
19c00ddc 277 cur_len = min(len, map_len - (offset - map_start));
b0496686 278 crc = btrfs_csum_data(kaddr + offset - map_start,
19c00ddc
CM		 279 crc, cur_len);
280 len -= cur_len;
281 offset += cur_len;
19c00ddc 282 }
607d432d
JB		 283 if (csum_size > sizeof(inline_result)) {
284 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
285 if (!result)
286 return 1;
287 } else {
288 result = (char *)&inline_result;
289 }
290
19c00ddc
CM		 291 btrfs_csum_final(crc, result);
292
293 if (verify) {
607d432d 294 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 295 u32 val;
296 u32 found = 0;
607d432d 297 memcpy(&found, result, csum_size);
e4204ded 298
607d432d 299 read_extent_buffer(buf, &val, 0, csum_size);
7a36ddec 300 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
193f284d
CM		 301 "failed on %llu wanted %X found %X "
302 "level %d\n",
303 root->fs_info->sb->s_id,
304 (unsigned long long)buf->start, val, found,
305 btrfs_header_level(buf));
607d432d
JB		 306 if (result != (char *)&inline_result)
307 kfree(result);
19c00ddc
CM		 308 return 1;
309 }
310 } else {
607d432d 311 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 312 }
607d432d
JB		 313 if (result != (char *)&inline_result)
314 kfree(result);
19c00ddc
CM		 315 return 0;
316}
317
d352ac68
CM		 318/*
319 * we can't consider a given block up to date unless the transid of the
320 * block matches the transid in the parent node's pointer. This is how we
321 * detect blocks that either didn't get written at all or got written
322 * in the wrong place.
323 */
1259ab75 324static int verify_parent_transid(struct extent_io_tree *io_tree,
b9fab919
CM		 325 struct extent_buffer *eb, u64 parent_transid,
326 int atomic)
1259ab75 327{
2ac55d41 328 struct extent_state *cached_state = NULL;
1259ab75
CM		 329 int ret;
330
331 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
332 return 0;
333
b9fab919
CM		 334 if (atomic)
335 return -EAGAIN;
336
2ac55d41 337 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
d0082371 338 0, &cached_state);
0b32f4bb 339 if (extent_buffer_uptodate(eb) &&
1259ab75
CM		 340 btrfs_header_generation(eb) == parent_transid) {
341 ret = 0;
342 goto out;
343 }
7a36ddec 344 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
193f284d
CM		 345 "found %llu\n",
346 (unsigned long long)eb->start,
347 (unsigned long long)parent_transid,
348 (unsigned long long)btrfs_header_generation(eb));
1259ab75 349 ret = 1;
0b32f4bb 350 clear_extent_buffer_uptodate(eb);
33958dc6 351out:
2ac55d41
JB		 352 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
353 &cached_state, GFP_NOFS);
1259ab75 354 return ret;
1259ab75
CM		 355}
356
d352ac68
CM		 357/*
358 * helper to read a given tree block, doing retries as required when
359 * the checksums don't match and we have alternate mirrors to try.
360 */
f188591e
CM		 361static int btree_read_extent_buffer_pages(struct btrfs_root *root,
362 struct extent_buffer *eb,
ca7a79ad 363 u64 start, u64 parent_transid)
f188591e
CM		 364{
365 struct extent_io_tree *io_tree;
ea466794 366 int failed = 0;
f188591e
CM		 367 int ret;
368 int num_copies = 0;
369 int mirror_num = 0;
ea466794 370 int failed_mirror = 0;
f188591e 371
a826d6dc 372 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM		 373 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
374 while (1) {
bb82ab88
AJ		 375 ret = read_extent_buffer_pages(io_tree, eb, start,
376 WAIT_COMPLETE,
f188591e 377 btree_get_extent, mirror_num);
256dd1bb
SB		 378 if (!ret) {
379 if (!verify_parent_transid(io_tree, eb,
b9fab919 380 parent_transid, 0))
256dd1bb
SB		 381 break;
382 else
383 ret = -EIO;
384 }
d397712b 385
a826d6dc
JB		 386 /*
387 * This buffer's crc is fine, but its contents are corrupted, so
388 * there is no reason to read the other copies, they won't be
389 * any less wrong.
390 */
391 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
ea466794
JB		 392 break;
393
5d964051 394 num_copies = btrfs_num_copies(root->fs_info,
f188591e 395 eb->start, eb->len);
4235298e 396 if (num_copies == 1)
ea466794 397 break;
4235298e 398
5cf1ab56
JB		 399 if (!failed_mirror) {
400 failed = 1;
401 failed_mirror = eb->read_mirror;
402 }
403
f188591e 404 mirror_num++;
ea466794
JB		 405 if (mirror_num == failed_mirror)
406 mirror_num++;
407
4235298e 408 if (mirror_num > num_copies)
ea466794 409 break;
f188591e 410 }
ea466794 411
c0901581 412 if (failed && !ret && failed_mirror)
ea466794
JB		 413 repair_eb_io_failure(root, eb, failed_mirror);
414
415 return ret;
f188591e 416}
19c00ddc 417
d352ac68 418/*
d397712b
CM		 419 * checksum a dirty tree block before IO. This has extra checks to make sure
420 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 421 */
d397712b 422
b2950863 423static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 424{
d1310b2e 425 struct extent_io_tree *tree;
4eee4fa4 426 u64 start = page_offset(page);
19c00ddc 427 u64 found_start;
19c00ddc 428 struct extent_buffer *eb;
f188591e 429
d1310b2e 430 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc 431
4f2de97a
JB		 432 eb = (struct extent_buffer *)page->private;
433 if (page != eb->pages[0])
434 return 0;
19c00ddc
CM		 435 found_start = btrfs_header_bytenr(eb);
436 if (found_start != start) {
55c69072 437 WARN_ON(1);
4f2de97a 438 return 0;
55c69072 439 }
55c69072 440 if (!PageUptodate(page)) {
55c69072 441 WARN_ON(1);
4f2de97a 442 return 0;
19c00ddc 443 }
19c00ddc 444 csum_tree_block(root, eb, 0);
19c00ddc
CM		 445 return 0;
446}
447
2b82032c
YZ		 448static int check_tree_block_fsid(struct btrfs_root *root,
449 struct extent_buffer *eb)
450{
451 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
452 u8 fsid[BTRFS_UUID_SIZE];
453 int ret = 1;
454
455 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
456 BTRFS_FSID_SIZE);
457 while (fs_devices) {
458 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
459 ret = 0;
460 break;
461 }
462 fs_devices = fs_devices->seed;
463 }
464 return ret;
465}
466
a826d6dc
JB		 467#define CORRUPT(reason, eb, root, slot) \
468 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
469 "root=%llu, slot=%d\n", reason, \
470 (unsigned long long)btrfs_header_bytenr(eb), \
471 (unsigned long long)root->objectid, slot)
472
473static noinline int check_leaf(struct btrfs_root *root,
474 struct extent_buffer *leaf)
475{
476 struct btrfs_key key;
477 struct btrfs_key leaf_key;
478 u32 nritems = btrfs_header_nritems(leaf);
479 int slot;
480
481 if (nritems == 0)
482 return 0;
483
484 /* Check the 0 item */
485 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
486 BTRFS_LEAF_DATA_SIZE(root)) {
487 CORRUPT("invalid item offset size pair", leaf, root, 0);
488 return -EIO;
489 }
490
491 /*
492 * Check to make sure each items keys are in the correct order and their
493 * offsets make sense. We only have to loop through nritems-1 because
494 * we check the current slot against the next slot, which verifies the
495 * next slot's offset+size makes sense and that the current's slot
496 * offset is correct.
497 */
498 for (slot = 0; slot < nritems - 1; slot++) {
499 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
500 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
501
502 /* Make sure the keys are in the right order */
503 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
504 CORRUPT("bad key order", leaf, root, slot);
505 return -EIO;
506 }
507
508 /*
509 * Make sure the offset and ends are right, remember that the
510 * item data starts at the end of the leaf and grows towards the
511 * front.
512 */
513 if (btrfs_item_offset_nr(leaf, slot) !=
514 btrfs_item_end_nr(leaf, slot + 1)) {
515 CORRUPT("slot offset bad", leaf, root, slot);
516 return -EIO;
517 }
518
519 /*
520 * Check to make sure that we don't point outside of the leaf,
521 * just incase all the items are consistent to eachother, but
522 * all point outside of the leaf.
523 */
524 if (btrfs_item_end_nr(leaf, slot) >
525 BTRFS_LEAF_DATA_SIZE(root)) {
526 CORRUPT("slot end outside of leaf", leaf, root, slot);
527 return -EIO;
528 }
529 }
530
531 return 0;
532}
533
727011e0
CM		 534struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
535 struct page *page, int max_walk)
536{
537 struct extent_buffer *eb;
538 u64 start = page_offset(page);
539 u64 target = start;
540 u64 min_start;
541
542 if (start < max_walk)
543 min_start = 0;
544 else
545 min_start = start - max_walk;
546
547 while (start >= min_start) {
548 eb = find_extent_buffer(tree, start, 0);
549 if (eb) {
550 /*
551 * we found an extent buffer and it contains our page
552 * horray!
553 */
554 if (eb->start <= target &&
555 eb->start + eb->len > target)
556 return eb;
557
558 /* we found an extent buffer that wasn't for us */
559 free_extent_buffer(eb);
560 return NULL;
561 }
562 if (start == 0)
563 break;
564 start -= PAGE_CACHE_SIZE;
565 }
566 return NULL;
567}
568
b2950863 569static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
5cf1ab56 570 struct extent_state *state, int mirror)
ce9adaa5
CM		 571{
572 struct extent_io_tree *tree;
573 u64 found_start;
574 int found_level;
ce9adaa5
CM		 575 struct extent_buffer *eb;
576 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 577 int ret = 0;
727011e0 578 int reads_done;
ce9adaa5 579
ce9adaa5
CM		 580 if (!page->private)
581 goto out;
d397712b 582
727011e0 583 tree = &BTRFS_I(page->mapping->host)->io_tree;
4f2de97a 584 eb = (struct extent_buffer *)page->private;
d397712b 585
0b32f4bb
JB		 586 /* the pending IO might have been the only thing that kept this buffer
587 * in memory. Make sure we have a ref for all this other checks
588 */
589 extent_buffer_get(eb);
590
591 reads_done = atomic_dec_and_test(&eb->io_pages);
727011e0
CM		 592 if (!reads_done)
593 goto err;
f188591e 594
5cf1ab56 595 eb->read_mirror = mirror;
ea466794
JB		 596 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
597 ret = -EIO;
598 goto err;
599 }
600
ce9adaa5 601 found_start = btrfs_header_bytenr(eb);
727011e0 602 if (found_start != eb->start) {
7a36ddec 603 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
193f284d
CM		 604 "%llu %llu\n",
605 (unsigned long long)found_start,
606 (unsigned long long)eb->start);
f188591e 607 ret = -EIO;
ce9adaa5
CM		 608 goto err;
609 }
2b82032c 610 if (check_tree_block_fsid(root, eb)) {
7a36ddec 611 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
193f284d 612 (unsigned long long)eb->start);
1259ab75
CM		 613 ret = -EIO;
614 goto err;
615 }
ce9adaa5 616 found_level = btrfs_header_level(eb);
1c24c3ce
JB		 617 if (found_level >= BTRFS_MAX_LEVEL) {
618 btrfs_info(root->fs_info, "bad tree block level %d\n",
619 (int)btrfs_header_level(eb));
620 ret = -EIO;
621 goto err;
622 }
ce9adaa5 623
85d4e461
CM		 624 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
625 eb, found_level);
4008c04a 626
ce9adaa5 627 ret = csum_tree_block(root, eb, 1);
a826d6dc 628 if (ret) {
f188591e 629 ret = -EIO;
a826d6dc
JB		 630 goto err;
631 }
632
633 /*
634 * If this is a leaf block and it is corrupt, set the corrupt bit so
635 * that we don't try and read the other copies of this block, just
636 * return -EIO.
637 */
638 if (found_level == 0 && check_leaf(root, eb)) {
639 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
640 ret = -EIO;
641 }
ce9adaa5 642
0b32f4bb
JB		 643 if (!ret)
644 set_extent_buffer_uptodate(eb);
ce9adaa5 645err:
79fb65a1
JB		 646 if (reads_done &&
647 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 648 btree_readahead_hook(root, eb, eb->start, ret);
4bb31e92 649
53b381b3
DW		 650 if (ret) {
651 /*
652 * our io error hook is going to dec the io pages
653 * again, we have to make sure it has something
654 * to decrement
655 */
656 atomic_inc(&eb->io_pages);
0b32f4bb 657 clear_extent_buffer_uptodate(eb);
53b381b3 658 }
0b32f4bb 659 free_extent_buffer(eb);
ce9adaa5 660out:
f188591e 661 return ret;
ce9adaa5
CM		 662}
663
ea466794 664static int btree_io_failed_hook(struct page *page, int failed_mirror)
4bb31e92 665{
4bb31e92
AJ		 666 struct extent_buffer *eb;
667 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
668
4f2de97a 669 eb = (struct extent_buffer *)page->private;
ea466794 670 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
5cf1ab56 671 eb->read_mirror = failed_mirror;
53b381b3 672 atomic_dec(&eb->io_pages);
ea466794 673 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 674 btree_readahead_hook(root, eb, eb->start, -EIO);
4bb31e92
AJ		 675 return -EIO; /* we fixed nothing */
676}
677
ce9adaa5 678static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 679{
680 struct end_io_wq *end_io_wq = bio->bi_private;
681 struct btrfs_fs_info *fs_info;
ce9adaa5 682
ce9adaa5 683 fs_info = end_io_wq->info;
ce9adaa5 684 end_io_wq->error = err;
8b712842
CM		 685 end_io_wq->work.func = end_workqueue_fn;
686 end_io_wq->work.flags = 0;
d20f7043 687
7b6d91da 688 if (bio->bi_rw & REQ_WRITE) {
53b381b3 689 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
cad321ad
CM		 690 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
691 &end_io_wq->work);
53b381b3 692 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
0cb59c99
JB		 693 btrfs_queue_worker(&fs_info->endio_freespace_worker,
694 &end_io_wq->work);
53b381b3
DW		 695 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
696 btrfs_queue_worker(&fs_info->endio_raid56_workers,
697 &end_io_wq->work);
cad321ad
CM		 698 else
699 btrfs_queue_worker(&fs_info->endio_write_workers,
700 &end_io_wq->work);
d20f7043 701 } else {
53b381b3
DW		 702 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
703 btrfs_queue_worker(&fs_info->endio_raid56_workers,
704 &end_io_wq->work);
705 else if (end_io_wq->metadata)
d20f7043
CM		 706 btrfs_queue_worker(&fs_info->endio_meta_workers,
707 &end_io_wq->work);
708 else
709 btrfs_queue_worker(&fs_info->endio_workers,
710 &end_io_wq->work);
711 }
ce9adaa5
CM		 712}
713
0cb59c99
JB		 714/*
715 * For the metadata arg you want
716 *
717 * 0 - if data
718 * 1 - if normal metadta
719 * 2 - if writing to the free space cache area
53b381b3 720 * 3 - raid parity work
0cb59c99 721 */
22c59948
CM		 722int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
723 int metadata)
0b86a832 724{
ce9adaa5 725 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 726 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
727 if (!end_io_wq)
728 return -ENOMEM;
729
730 end_io_wq->private = bio->bi_private;
731 end_io_wq->end_io = bio->bi_end_io;
22c59948 732 end_io_wq->info = info;
ce9adaa5
CM		 733 end_io_wq->error = 0;
734 end_io_wq->bio = bio;
22c59948 735 end_io_wq->metadata = metadata;
ce9adaa5
CM		 736
737 bio->bi_private = end_io_wq;
738 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 739 return 0;
740}
741
b64a2851 742unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 743{
4854ddd0
CM		 744 unsigned long limit = min_t(unsigned long,
745 info->workers.max_workers,
746 info->fs_devices->open_devices);
747 return 256 * limit;
748}
0986fe9e 749
4a69a410
CM		 750static void run_one_async_start(struct btrfs_work *work)
751{
4a69a410 752 struct async_submit_bio *async;
79787eaa 753 int ret;
4a69a410
CM		 754
755 async = container_of(work, struct async_submit_bio, work);
79787eaa
JM		 756 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
757 async->mirror_num, async->bio_flags,
758 async->bio_offset);
759 if (ret)
760 async->error = ret;
4a69a410
CM		 761}
762
763static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 764{
765 struct btrfs_fs_info *fs_info;
766 struct async_submit_bio *async;
4854ddd0 767 int limit;
8b712842
CM		 768
769 async = container_of(work, struct async_submit_bio, work);
770 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 771
b64a2851 772 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 773 limit = limit * 2 / 3;
774
66657b31 775 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
b64a2851 776 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 777 wake_up(&fs_info->async_submit_wait);
778
79787eaa
JM		 779 /* If an error occured we just want to clean up the bio and move on */
780 if (async->error) {
781 bio_endio(async->bio, async->error);
782 return;
783 }
784
4a69a410 785 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 786 async->mirror_num, async->bio_flags,
787 async->bio_offset);
4a69a410
CM		 788}
789
790static void run_one_async_free(struct btrfs_work *work)
791{
792 struct async_submit_bio *async;
793
794 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 795 kfree(async);
796}
797
44b8bd7e
CM		 798int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
799 int rw, struct bio *bio, int mirror_num,
c8b97818 800 unsigned long bio_flags,
eaf25d93 801 u64 bio_offset,
4a69a410
CM		 802 extent_submit_bio_hook_t *submit_bio_start,
803 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 804{
805 struct async_submit_bio *async;
806
807 async = kmalloc(sizeof(*async), GFP_NOFS);
808 if (!async)
809 return -ENOMEM;
810
811 async->inode = inode;
812 async->rw = rw;
813 async->bio = bio;
814 async->mirror_num = mirror_num;
4a69a410
CM		 815 async->submit_bio_start = submit_bio_start;
816 async->submit_bio_done = submit_bio_done;
817
818 async->work.func = run_one_async_start;
819 async->work.ordered_func = run_one_async_done;
820 async->work.ordered_free = run_one_async_free;
821
8b712842 822 async->work.flags = 0;
c8b97818 823 async->bio_flags = bio_flags;
eaf25d93 824 async->bio_offset = bio_offset;
8c8bee1d 825
79787eaa
JM		 826 async->error = 0;
827
cb03c743 828 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 829
7b6d91da 830 if (rw & REQ_SYNC)
d313d7a3
CM		 831 btrfs_set_work_high_prio(&async->work);
832
8b712842 833 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 834
d397712b 835 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 836 atomic_read(&fs_info->nr_async_submits)) {
837 wait_event(fs_info->async_submit_wait,
838 (atomic_read(&fs_info->nr_async_submits) == 0));
839 }
840
44b8bd7e
CM		 841 return 0;
842}
843
ce3ed71a
CM		 844static int btree_csum_one_bio(struct bio *bio)
845{
846 struct bio_vec *bvec = bio->bi_io_vec;
847 int bio_index = 0;
848 struct btrfs_root *root;
79787eaa 849 int ret = 0;
ce3ed71a
CM		 850
851 WARN_ON(bio->bi_vcnt <= 0);
d397712b 852 while (bio_index < bio->bi_vcnt) {
ce3ed71a 853 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 854 ret = csum_dirty_buffer(root, bvec->bv_page);
855 if (ret)
856 break;
ce3ed71a
CM		 857 bio_index++;
858 bvec++;
859 }
79787eaa 860 return ret;
ce3ed71a
CM		 861}
862
4a69a410
CM		 863static int __btree_submit_bio_start(struct inode *inode, int rw,
864 struct bio *bio, int mirror_num,
eaf25d93
CM		 865 unsigned long bio_flags,
866 u64 bio_offset)
22c59948 867{
8b712842
CM		 868 /*
869 * when we're called for a write, we're already in the async
5443be45 870 * submission context. Just jump into btrfs_map_bio
8b712842 871 */
79787eaa 872 return btree_csum_one_bio(bio);
4a69a410 873}
22c59948 874
4a69a410 875static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 876 int mirror_num, unsigned long bio_flags,
877 u64 bio_offset)
4a69a410 878{
61891923
SB		 879 int ret;
880
8b712842 881 /*
4a69a410
CM		 882 * when we're called for a write, we're already in the async
883 * submission context. Just jump into btrfs_map_bio
8b712842 884 */
61891923
SB		 885 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
886 if (ret)
887 bio_endio(bio, ret);
888 return ret;
0b86a832
CM		 889}
890
de0022b9
JB		 891static int check_async_write(struct inode *inode, unsigned long bio_flags)
892{
893 if (bio_flags & EXTENT_BIO_TREE_LOG)
894 return 0;
895#ifdef CONFIG_X86
896 if (cpu_has_xmm4_2)
897 return 0;
898#endif
899 return 1;
900}
901
44b8bd7e 902static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 903 int mirror_num, unsigned long bio_flags,
904 u64 bio_offset)
44b8bd7e 905{
de0022b9 906 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 907 int ret;
908
7b6d91da 909 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 910 /*
911 * called for a read, do the setup so that checksum validation
912 * can happen in the async kernel threads
913 */
f3f266ab
CM		 914 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
915 bio, 1);
1d4284bd 916 if (ret)
61891923
SB		 917 goto out_w_error;
918 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
919 mirror_num, 0);
de0022b9
JB		 920 } else if (!async) {
921 ret = btree_csum_one_bio(bio);
922 if (ret)
61891923
SB		 923 goto out_w_error;
924 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
925 mirror_num, 0);
926 } else {
927 /*
928 * kthread helpers are used to submit writes so that
929 * checksumming can happen in parallel across all CPUs
930 */
931 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
932 inode, rw, bio, mirror_num, 0,
933 bio_offset,
934 __btree_submit_bio_start,
935 __btree_submit_bio_done);
44b8bd7e 936 }
d313d7a3 937
61891923
SB		 938 if (ret) {
939out_w_error:
940 bio_endio(bio, ret);
941 }
942 return ret;
44b8bd7e
CM		 943}
944
3dd1462e 945#ifdef CONFIG_MIGRATION
784b4e29 946static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 947 struct page *newpage, struct page *page,
948 enum migrate_mode mode)
784b4e29
CM		 949{
950 /*
951 * we can't safely write a btree page from here,
952 * we haven't done the locking hook
953 */
954 if (PageDirty(page))
955 return -EAGAIN;
956 /*
957 * Buffers may be managed in a filesystem specific way.
958 * We must have no buffers or drop them.
959 */
960 if (page_has_private(page) &&
961 !try_to_release_page(page, GFP_KERNEL))
962 return -EAGAIN;
a6bc32b8 963 return migrate_page(mapping, newpage, page, mode);
784b4e29 964}
3dd1462e 965#endif
784b4e29 966
0da5468f
CM		 967
968static int btree_writepages(struct address_space *mapping,
969 struct writeback_control *wbc)
970{
d1310b2e 971 struct extent_io_tree *tree;
e2d84521
MX		 972 struct btrfs_fs_info *fs_info;
973 int ret;
974
d1310b2e 975 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 976 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 977
978 if (wbc->for_kupdate)
979 return 0;
980
e2d84521 981 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 982 /* this is a bit racy, but that's ok */
e2d84521
MX		 983 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
984 BTRFS_DIRTY_METADATA_THRESH);
985 if (ret < 0)
793955bc 986 return 0;
793955bc 987 }
0b32f4bb 988 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 989}
990
b2950863 991static int btree_readpage(struct file *file, struct page *page)
5f39d397 992{
d1310b2e
CM		 993 struct extent_io_tree *tree;
994 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 995 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 996}
22b0ebda 997
70dec807 998static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 999{
98509cfc 1000 if (PageWriteback(page) || PageDirty(page))
d397712b 1001 return 0;
0c4e538b
DS		 1002 /*
1003 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
1004 * slab allocation from alloc_extent_state down the callchain where
1005 * it'd hit a BUG_ON as those flags are not allowed.
1006 */
1007 gfp_flags &= ~GFP_SLAB_BUG_MASK;
1008
3083ee2e 1009 return try_release_extent_buffer(page, gfp_flags);
d98237b3
CM		 1010}
1011
5f39d397 1012static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 1013{
d1310b2e
CM		 1014 struct extent_io_tree *tree;
1015 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1016 extent_invalidatepage(tree, page, offset);
1017 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1018 if (PagePrivate(page)) {
d397712b
CM		 1019 printk(KERN_WARNING "btrfs warning page private not zero "
1020 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1021 ClearPagePrivate(page);
1022 set_page_private(page, 0);
1023 page_cache_release(page);
1024 }
d98237b3
CM		 1025}
1026
0b32f4bb
JB		 1027static int btree_set_page_dirty(struct page *page)
1028{
bb146eb2 1029#ifdef DEBUG
0b32f4bb
JB		 1030 struct extent_buffer *eb;
1031
1032 BUG_ON(!PagePrivate(page));
1033 eb = (struct extent_buffer *)page->private;
1034 BUG_ON(!eb);
1035 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1036 BUG_ON(!atomic_read(&eb->refs));
1037 btrfs_assert_tree_locked(eb);
bb146eb2 1038#endif
0b32f4bb
JB		 1039 return __set_page_dirty_nobuffers(page);
1040}
1041
7f09410b 1042static const struct address_space_operations btree_aops = {
d98237b3 1043 .readpage = btree_readpage,
0da5468f 1044 .writepages = btree_writepages,
5f39d397
CM		 1045 .releasepage = btree_releasepage,
1046 .invalidatepage = btree_invalidatepage,
5a92bc88 1047#ifdef CONFIG_MIGRATION
784b4e29 1048 .migratepage = btree_migratepage,
5a92bc88 1049#endif
0b32f4bb 1050 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1051};
1052
ca7a79ad
CM		 1053int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1054 u64 parent_transid)
090d1875 1055{
5f39d397
CM		 1056 struct extent_buffer *buf = NULL;
1057 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1058 int ret = 0;
090d1875 1059
db94535d 1060 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1061 if (!buf)
090d1875 1062 return 0;
d1310b2e 1063 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1064 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1065 free_extent_buffer(buf);
de428b63 1066 return ret;
090d1875
CM		 1067}
1068
ab0fff03
AJ		 1069int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1070 int mirror_num, struct extent_buffer **eb)
1071{
1072 struct extent_buffer *buf = NULL;
1073 struct inode *btree_inode = root->fs_info->btree_inode;
1074 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1075 int ret;
1076
1077 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1078 if (!buf)
1079 return 0;
1080
1081 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1082
1083 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1084 btree_get_extent, mirror_num);
1085 if (ret) {
1086 free_extent_buffer(buf);
1087 return ret;
1088 }
1089
1090 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1091 free_extent_buffer(buf);
1092 return -EIO;
0b32f4bb 1093 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1094 *eb = buf;
1095 } else {
1096 free_extent_buffer(buf);
1097 }
1098 return 0;
1099}
1100
0999df54
CM		 1101struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1102 u64 bytenr, u32 blocksize)
1103{
1104 struct inode *btree_inode = root->fs_info->btree_inode;
1105 struct extent_buffer *eb;
1106 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
f09d1f60 1107 bytenr, blocksize);
0999df54
CM		 1108 return eb;
1109}
1110
1111struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1112 u64 bytenr, u32 blocksize)
1113{
1114 struct inode *btree_inode = root->fs_info->btree_inode;
1115 struct extent_buffer *eb;
1116
1117 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
727011e0 1118 bytenr, blocksize);
0999df54
CM		 1119 return eb;
1120}
1121
1122
e02119d5
CM		 1123int btrfs_write_tree_block(struct extent_buffer *buf)
1124{
727011e0 1125 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
8aa38c31 1126 buf->start + buf->len - 1);
e02119d5
CM		 1127}
1128
1129int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1130{
727011e0 1131 return filemap_fdatawait_range(buf->pages[0]->mapping,
8aa38c31 1132 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1133}
1134
0999df54 1135struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1136 u32 blocksize, u64 parent_transid)
0999df54
CM		 1137{
1138 struct extent_buffer *buf = NULL;
0999df54
CM		 1139 int ret;
1140
0999df54
CM		 1141 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1142 if (!buf)
1143 return NULL;
0999df54 1144
ca7a79ad 1145 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
5f39d397 1146 return buf;
ce9adaa5 1147
eb60ceac
CM		 1148}
1149
d5c13f92
JM		 1150void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1151 struct extent_buffer *buf)
ed2ff2cb 1152{
e2d84521
MX		 1153 struct btrfs_fs_info *fs_info = root->fs_info;
1154
55c69072 1155 if (btrfs_header_generation(buf) ==
e2d84521 1156 fs_info->running_transaction->transid) {
b9447ef8 1157 btrfs_assert_tree_locked(buf);
b4ce94de 1158
b9473439 1159 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1160 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1161 -buf->len,
1162 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1163 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1164 btrfs_set_lock_blocking(buf);
1165 clear_extent_buffer_dirty(buf);
1166 }
925baedd 1167 }
5f39d397
CM		 1168}
1169
143bede5
JM		 1170static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1171 u32 stripesize, struct btrfs_root *root,
1172 struct btrfs_fs_info *fs_info,
1173 u64 objectid)
d97e63b6 1174{
cfaa7295 1175 root->node = NULL;
a28ec197 1176 root->commit_root = NULL;
db94535d
CM		 1177 root->sectorsize = sectorsize;
1178 root->nodesize = nodesize;
1179 root->leafsize = leafsize;
87ee04eb 1180 root->stripesize = stripesize;
123abc88 1181 root->ref_cows = 0;
0b86a832 1182 root->track_dirty = 0;
c71bf099 1183 root->in_radix = 0;
d68fc57b
YZ		 1184 root->orphan_item_inserted = 0;
1185 root->orphan_cleanup_state = 0;
0b86a832 1186
0f7d52f4
CM		 1187 root->objectid = objectid;
1188 root->last_trans = 0;
13a8a7c8 1189 root->highest_objectid = 0;
58176a96 1190 root->name = NULL;
6bef4d31 1191 root->inode_tree = RB_ROOT;
16cdcec7 1192 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1193 root->block_rsv = NULL;
d68fc57b 1194 root->orphan_block_rsv = NULL;
0b86a832
CM		 1195
1196 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1197 INIT_LIST_HEAD(&root->root_list);
2ab28f32
JB		 1198 INIT_LIST_HEAD(&root->logged_list[0]);
1199 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1200 spin_lock_init(&root->orphan_lock);
5d4f98a2 1201 spin_lock_init(&root->inode_lock);
f0486c68 1202 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1203 spin_lock_init(&root->log_extents_lock[0]);
1204 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1205 mutex_init(&root->objectid_mutex);
e02119d5 1206 mutex_init(&root->log_mutex);
7237f183
YZ		 1207 init_waitqueue_head(&root->log_writer_wait);
1208 init_waitqueue_head(&root->log_commit_wait[0]);
1209 init_waitqueue_head(&root->log_commit_wait[1]);
1210 atomic_set(&root->log_commit[0], 0);
1211 atomic_set(&root->log_commit[1], 0);
1212 atomic_set(&root->log_writers, 0);
2ecb7923 1213 atomic_set(&root->log_batch, 0);
8a35d95f 1214 atomic_set(&root->orphan_inodes, 0);
7237f183 1215 root->log_transid = 0;
257c62e1 1216 root->last_log_commit = 0;
d0c803c4 1217 extent_io_tree_init(&root->dirty_log_pages,
f993c883 1218 fs_info->btree_inode->i_mapping);
017e5369 1219
3768f368
CM		 1220 memset(&root->root_key, 0, sizeof(root->root_key));
1221 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1222 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1223 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 1224 root->defrag_trans_start = fs_info->generation;
58176a96 1225 init_completion(&root->kobj_unregister);
6702ed49 1226 root->defrag_running = 0;
4d775673 1227 root->root_key.objectid = objectid;
0ee5dc67 1228 root->anon_dev = 0;
8ea05e3a 1229
5f3ab90a 1230 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1231}
1232
200a5c17
JM		 1233static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1234 struct btrfs_fs_info *fs_info,
1235 u64 objectid,
1236 struct btrfs_root *root)
3768f368
CM		 1237{
1238 int ret;
db94535d 1239 u32 blocksize;
84234f3a 1240 u64 generation;
3768f368 1241
db94535d 1242 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1243 tree_root->sectorsize, tree_root->stripesize,
1244 root, fs_info, objectid);
3768f368
CM		 1245 ret = btrfs_find_last_root(tree_root, objectid,
1246 &root->root_item, &root->root_key);
4df27c4d
YZ		 1247 if (ret > 0)
1248 return -ENOENT;
200a5c17
JM		 1249 else if (ret < 0)
1250 return ret;
3768f368 1251
84234f3a 1252 generation = btrfs_root_generation(&root->root_item);
db94535d 1253 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
af31f5e5 1254 root->commit_root = NULL;
db94535d 1255 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1256 blocksize, generation);
b9fab919 1257 if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
68433b73 1258 free_extent_buffer(root->node);
af31f5e5 1259 root->node = NULL;
68433b73
CM		 1260 return -EIO;
1261 }
4df27c4d 1262 root->commit_root = btrfs_root_node(root);
d97e63b6
CM		 1263 return 0;
1264}
1265
f84a8bd6 1266static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1267{
1268 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1269 if (root)
1270 root->fs_info = fs_info;
1271 return root;
1272}
1273
20897f5c
AJ		 1274struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1275 struct btrfs_fs_info *fs_info,
1276 u64 objectid)
1277{
1278 struct extent_buffer *leaf;
1279 struct btrfs_root *tree_root = fs_info->tree_root;
1280 struct btrfs_root *root;
1281 struct btrfs_key key;
1282 int ret = 0;
1283 u64 bytenr;
6463fe58 1284 uuid_le uuid;
20897f5c
AJ		 1285
1286 root = btrfs_alloc_root(fs_info);
1287 if (!root)
1288 return ERR_PTR(-ENOMEM);
1289
1290 __setup_root(tree_root->nodesize, tree_root->leafsize,
1291 tree_root->sectorsize, tree_root->stripesize,
1292 root, fs_info, objectid);
1293 root->root_key.objectid = objectid;
1294 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1295 root->root_key.offset = 0;
1296
1297 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1298 0, objectid, NULL, 0, 0, 0);
1299 if (IS_ERR(leaf)) {
1300 ret = PTR_ERR(leaf);
1dd05682 1301 leaf = NULL;
20897f5c
AJ		 1302 goto fail;
1303 }
1304
1305 bytenr = leaf->start;
1306 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1307 btrfs_set_header_bytenr(leaf, leaf->start);
1308 btrfs_set_header_generation(leaf, trans->transid);
1309 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1310 btrfs_set_header_owner(leaf, objectid);
1311 root->node = leaf;
1312
1313 write_extent_buffer(leaf, fs_info->fsid,
1314 (unsigned long)btrfs_header_fsid(leaf),
1315 BTRFS_FSID_SIZE);
1316 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1317 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1318 BTRFS_UUID_SIZE);
1319 btrfs_mark_buffer_dirty(leaf);
1320
1321 root->commit_root = btrfs_root_node(root);
1322 root->track_dirty = 1;
1323
1324
1325 root->root_item.flags = 0;
1326 root->root_item.byte_limit = 0;
1327 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1328 btrfs_set_root_generation(&root->root_item, trans->transid);
1329 btrfs_set_root_level(&root->root_item, 0);
1330 btrfs_set_root_refs(&root->root_item, 1);
1331 btrfs_set_root_used(&root->root_item, leaf->len);
1332 btrfs_set_root_last_snapshot(&root->root_item, 0);
1333 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1334 uuid_le_gen(&uuid);
1335 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1336 root->root_item.drop_level = 0;
1337
1338 key.objectid = objectid;
1339 key.type = BTRFS_ROOT_ITEM_KEY;
1340 key.offset = 0;
1341 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1342 if (ret)
1343 goto fail;
1344
1345 btrfs_tree_unlock(leaf);
1346
1dd05682
TI		 1347 return root;
1348
20897f5c 1349fail:
1dd05682
TI		 1350 if (leaf) {
1351 btrfs_tree_unlock(leaf);
1352 free_extent_buffer(leaf);
1353 }
1354 kfree(root);
20897f5c 1355
1dd05682 1356 return ERR_PTR(ret);
20897f5c
AJ		 1357}
1358
7237f183
YZ		 1359static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1360 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1361{
1362 struct btrfs_root *root;
1363 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1364 struct extent_buffer *leaf;
e02119d5 1365
6f07e42e 1366 root = btrfs_alloc_root(fs_info);
e02119d5 1367 if (!root)
7237f183 1368 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1369
1370 __setup_root(tree_root->nodesize, tree_root->leafsize,
1371 tree_root->sectorsize, tree_root->stripesize,
1372 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1373
1374 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1375 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1376 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1377 /*
1378 * log trees do not get reference counted because they go away
1379 * before a real commit is actually done. They do store pointers
1380 * to file data extents, and those reference counts still get
1381 * updated (along with back refs to the log tree).
1382 */
e02119d5
CM		 1383 root->ref_cows = 0;
1384
5d4f98a2 1385 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1386 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1387 0, 0, 0);
7237f183
YZ		 1388 if (IS_ERR(leaf)) {
1389 kfree(root);
1390 return ERR_CAST(leaf);
1391 }
e02119d5 1392
5d4f98a2
YZ		 1393 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1394 btrfs_set_header_bytenr(leaf, leaf->start);
1395 btrfs_set_header_generation(leaf, trans->transid);
1396 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1397 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1398 root->node = leaf;
e02119d5
CM		 1399
1400 write_extent_buffer(root->node, root->fs_info->fsid,
1401 (unsigned long)btrfs_header_fsid(root->node),
1402 BTRFS_FSID_SIZE);
1403 btrfs_mark_buffer_dirty(root->node);
1404 btrfs_tree_unlock(root->node);
7237f183
YZ		 1405 return root;
1406}
1407
1408int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1409 struct btrfs_fs_info *fs_info)
1410{
1411 struct btrfs_root *log_root;
1412
1413 log_root = alloc_log_tree(trans, fs_info);
1414 if (IS_ERR(log_root))
1415 return PTR_ERR(log_root);
1416 WARN_ON(fs_info->log_root_tree);
1417 fs_info->log_root_tree = log_root;
1418 return 0;
1419}
1420
1421int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1422 struct btrfs_root *root)
1423{
1424 struct btrfs_root *log_root;
1425 struct btrfs_inode_item *inode_item;
1426
1427 log_root = alloc_log_tree(trans, root->fs_info);
1428 if (IS_ERR(log_root))
1429 return PTR_ERR(log_root);
1430
1431 log_root->last_trans = trans->transid;
1432 log_root->root_key.offset = root->root_key.objectid;
1433
1434 inode_item = &log_root->root_item.inode;
1435 inode_item->generation = cpu_to_le64(1);
1436 inode_item->size = cpu_to_le64(3);
1437 inode_item->nlink = cpu_to_le32(1);
1438 inode_item->nbytes = cpu_to_le64(root->leafsize);
1439 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1440
5d4f98a2 1441 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1442
1443 WARN_ON(root->log_root);
1444 root->log_root = log_root;
1445 root->log_transid = 0;
257c62e1 1446 root->last_log_commit = 0;
e02119d5
CM		 1447 return 0;
1448}
1449
1450struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1451 struct btrfs_key *location)
1452{
1453 struct btrfs_root *root;
1454 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1455 struct btrfs_path *path;
5f39d397 1456 struct extent_buffer *l;
84234f3a 1457 u64 generation;
db94535d 1458 u32 blocksize;
0f7d52f4 1459 int ret = 0;
8ea05e3a 1460 int slot;
0f7d52f4 1461
6f07e42e 1462 root = btrfs_alloc_root(fs_info);
0cf6c620 1463 if (!root)
0f7d52f4 1464 return ERR_PTR(-ENOMEM);
0f7d52f4 1465 if (location->offset == (u64)-1) {
db94535d 1466 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM		 1467 location->objectid, root);
1468 if (ret) {
0f7d52f4
CM		 1469 kfree(root);
1470 return ERR_PTR(ret);
1471 }
13a8a7c8 1472 goto out;
0f7d52f4
CM		 1473 }
1474
db94535d 1475 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1476 tree_root->sectorsize, tree_root->stripesize,
1477 root, fs_info, location->objectid);
0f7d52f4
CM		 1478
1479 path = btrfs_alloc_path();
db5b493a
TI		 1480 if (!path) {
1481 kfree(root);
1482 return ERR_PTR(-ENOMEM);
1483 }
0f7d52f4 1484 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
13a8a7c8
YZ		 1485 if (ret == 0) {
1486 l = path->nodes[0];
8ea05e3a 1487 slot = path->slots[0];
5fbf83c1 1488 btrfs_read_root_item(l, slot, &root->root_item);
13a8a7c8 1489 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4 1490 }
0f7d52f4
CM		 1491 btrfs_free_path(path);
1492 if (ret) {
5e540f77 1493 kfree(root);
13a8a7c8
YZ		 1494 if (ret > 0)
1495 ret = -ENOENT;
0f7d52f4
CM		 1496 return ERR_PTR(ret);
1497 }
13a8a7c8 1498
84234f3a 1499 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1500 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1501 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1502 blocksize, generation);
416bc658
JB		 1503 if (!root->node || !extent_buffer_uptodate(root->node)) {
1504 ret = (!root->node) ? -ENOMEM : -EIO;
1505
1506 free_extent_buffer(root->node);
1507 kfree(root);
1508 return ERR_PTR(ret);
1509 }
1510
5d4f98a2 1511 root->commit_root = btrfs_root_node(root);
79787eaa 1512 BUG_ON(!root->node); /* -ENOMEM */
13a8a7c8 1513out:
08fe4db1 1514 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1515 root->ref_cows = 1;
08fe4db1
LZ		 1516 btrfs_check_and_init_root_item(&root->root_item);
1517 }
13a8a7c8 1518
5eda7b5e
CM		 1519 return root;
1520}
1521
edbd8d4e
CM		 1522struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1523 struct btrfs_key *location)
5eda7b5e
CM		 1524{
1525 struct btrfs_root *root;
1526 int ret;
1527
edbd8d4e
CM		 1528 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1529 return fs_info->tree_root;
1530 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1531 return fs_info->extent_root;
8f18cf13
CM		 1532 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1533 return fs_info->chunk_root;
1534 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1535 return fs_info->dev_root;
0403e47e
YZ		 1536 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1537 return fs_info->csum_root;
bcef60f2
AJ		 1538 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1539 return fs_info->quota_root ? fs_info->quota_root :
1540 ERR_PTR(-ENOENT);
4df27c4d
YZ		 1541again:
1542 spin_lock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1543 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1544 (unsigned long)location->objectid);
4df27c4d 1545 spin_unlock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1546 if (root)
1547 return root;
1548
e02119d5 1549 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM		 1550 if (IS_ERR(root))
1551 return root;
3394e160 1552
581bb050 1553 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
581bb050
LZ		 1554 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1555 GFP_NOFS);
35a30d7c
DS		 1556 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1557 ret = -ENOMEM;
581bb050 1558 goto fail;
35a30d7c 1559 }
581bb050
LZ		 1560
1561 btrfs_init_free_ino_ctl(root);
1562 mutex_init(&root->fs_commit_mutex);
1563 spin_lock_init(&root->cache_lock);
1564 init_waitqueue_head(&root->cache_wait);
1565
0ee5dc67 1566 ret = get_anon_bdev(&root->anon_dev);
ac08aedf
CM		 1567 if (ret)
1568 goto fail;
3394e160 1569
d68fc57b
YZ		 1570 if (btrfs_root_refs(&root->root_item) == 0) {
1571 ret = -ENOENT;
1572 goto fail;
1573 }
1574
1575 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1576 if (ret < 0)
1577 goto fail;
1578 if (ret == 0)
1579 root->orphan_item_inserted = 1;
1580
4df27c4d
YZ		 1581 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1582 if (ret)
1583 goto fail;
1584
1585 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 1586 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1587 (unsigned long)root->root_key.objectid,
0f7d52f4 1588 root);
d68fc57b 1589 if (ret == 0)
4df27c4d 1590 root->in_radix = 1;
d68fc57b 1591
4df27c4d
YZ		 1592 spin_unlock(&fs_info->fs_roots_radix_lock);
1593 radix_tree_preload_end();
0f7d52f4 1594 if (ret) {
4df27c4d
YZ		 1595 if (ret == -EEXIST) {
1596 free_fs_root(root);
1597 goto again;
1598 }
1599 goto fail;
0f7d52f4 1600 }
4df27c4d
YZ		 1601
1602 ret = btrfs_find_dead_roots(fs_info->tree_root,
1603 root->root_key.objectid);
1604 WARN_ON(ret);
edbd8d4e 1605 return root;
4df27c4d
YZ		 1606fail:
1607 free_fs_root(root);
1608 return ERR_PTR(ret);
edbd8d4e
CM		 1609}
1610
04160088
CM		 1611static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1612{
1613 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1614 int ret = 0;
04160088
CM		 1615 struct btrfs_device *device;
1616 struct backing_dev_info *bdi;
b7967db7 1617
1f78160c
XG		 1618 rcu_read_lock();
1619 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1620 if (!device->bdev)
1621 continue;
04160088
CM		 1622 bdi = blk_get_backing_dev_info(device->bdev);
1623 if (bdi && bdi_congested(bdi, bdi_bits)) {
1624 ret = 1;
1625 break;
1626 }
1627 }
1f78160c 1628 rcu_read_unlock();
04160088
CM		 1629 return ret;
1630}
1631
ad081f14
JA		 1632/*
1633 * If this fails, caller must call bdi_destroy() to get rid of the
1634 * bdi again.
1635 */
04160088
CM		 1636static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1637{
ad081f14
JA		 1638 int err;
1639
1640 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1641 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1642 if (err)
1643 return err;
1644
4575c9cc 1645 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1646 bdi->congested_fn = btrfs_congested_fn;
1647 bdi->congested_data = info;
1648 return 0;
1649}
1650
8b712842
CM		 1651/*
1652 * called by the kthread helper functions to finally call the bio end_io
1653 * functions. This is where read checksum verification actually happens
1654 */
1655static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1656{
ce9adaa5 1657 struct bio *bio;
8b712842
CM		 1658 struct end_io_wq *end_io_wq;
1659 struct btrfs_fs_info *fs_info;
ce9adaa5 1660 int error;
ce9adaa5 1661
8b712842
CM		 1662 end_io_wq = container_of(work, struct end_io_wq, work);
1663 bio = end_io_wq->bio;
1664 fs_info = end_io_wq->info;
ce9adaa5 1665
8b712842
CM		 1666 error = end_io_wq->error;
1667 bio->bi_private = end_io_wq->private;
1668 bio->bi_end_io = end_io_wq->end_io;
1669 kfree(end_io_wq);
8b712842 1670 bio_endio(bio, error);
44b8bd7e
CM		 1671}
1672
a74a4b97
CM		 1673static int cleaner_kthread(void *arg)
1674{
1675 struct btrfs_root *root = arg;
1676
1677 do {
9d1a2a3a
DS		 1678 int again = 0;
1679
76dda93c 1680 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
9d1a2a3a
DS		 1681 down_read_trylock(&root->fs_info->sb->s_umount)) {
1682 if (mutex_trylock(&root->fs_info->cleaner_mutex)) {
1683 btrfs_run_delayed_iputs(root);
1684 again = btrfs_clean_one_deleted_snapshot(root);
1685 mutex_unlock(&root->fs_info->cleaner_mutex);
1686 }
4cb5300b 1687 btrfs_run_defrag_inodes(root->fs_info);
9d1a2a3a 1688 up_read(&root->fs_info->sb->s_umount);
76dda93c 1689 }
a74a4b97 1690
9d1a2a3a 1691 if (!try_to_freeze() && !again) {
a74a4b97 1692 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1693 if (!kthread_should_stop())
1694 schedule();
a74a4b97
CM		 1695 __set_current_state(TASK_RUNNING);
1696 }
1697 } while (!kthread_should_stop());
1698 return 0;
1699}
1700
1701static int transaction_kthread(void *arg)
1702{
1703 struct btrfs_root *root = arg;
1704 struct btrfs_trans_handle *trans;
1705 struct btrfs_transaction *cur;
8929ecfa 1706 u64 transid;
a74a4b97
CM		 1707 unsigned long now;
1708 unsigned long delay;
914b2007 1709 bool cannot_commit;
a74a4b97
CM		 1710
1711 do {
914b2007 1712 cannot_commit = false;
a74a4b97 1713 delay = HZ * 30;
a74a4b97
CM		 1714 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1715
a4abeea4 1716 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1717 cur = root->fs_info->running_transaction;
1718 if (!cur) {
a4abeea4 1719 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1720 goto sleep;
1721 }
31153d81 1722
a74a4b97 1723 now = get_seconds();
8929ecfa
YZ		 1724 if (!cur->blocked &&
1725 (now < cur->start_time || now - cur->start_time < 30)) {
a4abeea4 1726 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1727 delay = HZ * 5;
1728 goto sleep;
1729 }
8929ecfa 1730 transid = cur->transid;
a4abeea4 1731 spin_unlock(&root->fs_info->trans_lock);
56bec294 1732
79787eaa 1733 /* If the file system is aborted, this will always fail. */
354aa0fb 1734 trans = btrfs_attach_transaction(root);
914b2007 1735 if (IS_ERR(trans)) {
354aa0fb
MX		 1736 if (PTR_ERR(trans) != -ENOENT)
1737 cannot_commit = true;
79787eaa 1738 goto sleep;
914b2007 1739 }
8929ecfa 1740 if (transid == trans->transid) {
79787eaa 1741 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1742 } else {
1743 btrfs_end_transaction(trans, root);
1744 }
a74a4b97
CM		 1745sleep:
1746 wake_up_process(root->fs_info->cleaner_kthread);
1747 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1748
a0acae0e 1749 if (!try_to_freeze()) {
a74a4b97 1750 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1751 if (!kthread_should_stop() &&
914b2007
JK		 1752 (!btrfs_transaction_blocked(root->fs_info) ||
1753 cannot_commit))
8929ecfa 1754 schedule_timeout(delay);
a74a4b97
CM		 1755 __set_current_state(TASK_RUNNING);
1756 }
1757 } while (!kthread_should_stop());
1758 return 0;
1759}
1760
af31f5e5
CM		 1761/*
1762 * this will find the highest generation in the array of
1763 * root backups. The index of the highest array is returned,
1764 * or -1 if we can't find anything.
1765 *
1766 * We check to make sure the array is valid by comparing the
1767 * generation of the latest root in the array with the generation
1768 * in the super block. If they don't match we pitch it.
1769 */
1770static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1771{
1772 u64 cur;
1773 int newest_index = -1;
1774 struct btrfs_root_backup *root_backup;
1775 int i;
1776
1777 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1778 root_backup = info->super_copy->super_roots + i;
1779 cur = btrfs_backup_tree_root_gen(root_backup);
1780 if (cur == newest_gen)
1781 newest_index = i;
1782 }
1783
1784 /* check to see if we actually wrapped around */
1785 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1786 root_backup = info->super_copy->super_roots;
1787 cur = btrfs_backup_tree_root_gen(root_backup);
1788 if (cur == newest_gen)
1789 newest_index = 0;
1790 }
1791 return newest_index;
1792}
1793
1794
1795/*
1796 * find the oldest backup so we know where to store new entries
1797 * in the backup array. This will set the backup_root_index
1798 * field in the fs_info struct
1799 */
1800static void find_oldest_super_backup(struct btrfs_fs_info *info,
1801 u64 newest_gen)
1802{
1803 int newest_index = -1;
1804
1805 newest_index = find_newest_super_backup(info, newest_gen);
1806 /* if there was garbage in there, just move along */
1807 if (newest_index == -1) {
1808 info->backup_root_index = 0;
1809 } else {
1810 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1811 }
1812}
1813
1814/*
1815 * copy all the root pointers into the super backup array.
1816 * this will bump the backup pointer by one when it is
1817 * done
1818 */
1819static void backup_super_roots(struct btrfs_fs_info *info)
1820{
1821 int next_backup;
1822 struct btrfs_root_backup *root_backup;
1823 int last_backup;
1824
1825 next_backup = info->backup_root_index;
1826 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1827 BTRFS_NUM_BACKUP_ROOTS;
1828
1829 /*
1830 * just overwrite the last backup if we're at the same generation
1831 * this happens only at umount
1832 */
1833 root_backup = info->super_for_commit->super_roots + last_backup;
1834 if (btrfs_backup_tree_root_gen(root_backup) ==
1835 btrfs_header_generation(info->tree_root->node))
1836 next_backup = last_backup;
1837
1838 root_backup = info->super_for_commit->super_roots + next_backup;
1839
1840 /*
1841 * make sure all of our padding and empty slots get zero filled
1842 * regardless of which ones we use today
1843 */
1844 memset(root_backup, 0, sizeof(*root_backup));
1845
1846 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1847
1848 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1849 btrfs_set_backup_tree_root_gen(root_backup,
1850 btrfs_header_generation(info->tree_root->node));
1851
1852 btrfs_set_backup_tree_root_level(root_backup,
1853 btrfs_header_level(info->tree_root->node));
1854
1855 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1856 btrfs_set_backup_chunk_root_gen(root_backup,
1857 btrfs_header_generation(info->chunk_root->node));
1858 btrfs_set_backup_chunk_root_level(root_backup,
1859 btrfs_header_level(info->chunk_root->node));
1860
1861 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1862 btrfs_set_backup_extent_root_gen(root_backup,
1863 btrfs_header_generation(info->extent_root->node));
1864 btrfs_set_backup_extent_root_level(root_backup,
1865 btrfs_header_level(info->extent_root->node));
1866
7c7e82a7
CM		 1867 /*
1868 * we might commit during log recovery, which happens before we set
1869 * the fs_root. Make sure it is valid before we fill it in.
1870 */
1871 if (info->fs_root && info->fs_root->node) {
1872 btrfs_set_backup_fs_root(root_backup,
1873 info->fs_root->node->start);
1874 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1875 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1876 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1877 btrfs_header_level(info->fs_root->node));
7c7e82a7 1878 }
af31f5e5
CM		 1879
1880 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1881 btrfs_set_backup_dev_root_gen(root_backup,
1882 btrfs_header_generation(info->dev_root->node));
1883 btrfs_set_backup_dev_root_level(root_backup,
1884 btrfs_header_level(info->dev_root->node));
1885
1886 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1887 btrfs_set_backup_csum_root_gen(root_backup,
1888 btrfs_header_generation(info->csum_root->node));
1889 btrfs_set_backup_csum_root_level(root_backup,
1890 btrfs_header_level(info->csum_root->node));
1891
1892 btrfs_set_backup_total_bytes(root_backup,
1893 btrfs_super_total_bytes(info->super_copy));
1894 btrfs_set_backup_bytes_used(root_backup,
1895 btrfs_super_bytes_used(info->super_copy));
1896 btrfs_set_backup_num_devices(root_backup,
1897 btrfs_super_num_devices(info->super_copy));
1898
1899 /*
1900 * if we don't copy this out to the super_copy, it won't get remembered
1901 * for the next commit
1902 */
1903 memcpy(&info->super_copy->super_roots,
1904 &info->super_for_commit->super_roots,
1905 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1906}
1907
1908/*
1909 * this copies info out of the root backup array and back into
1910 * the in-memory super block. It is meant to help iterate through
1911 * the array, so you send it the number of backups you've already
1912 * tried and the last backup index you used.
1913 *
1914 * this returns -1 when it has tried all the backups
1915 */
1916static noinline int next_root_backup(struct btrfs_fs_info *info,
1917 struct btrfs_super_block *super,
1918 int *num_backups_tried, int *backup_index)
1919{
1920 struct btrfs_root_backup *root_backup;
1921 int newest = *backup_index;
1922
1923 if (*num_backups_tried == 0) {
1924 u64 gen = btrfs_super_generation(super);
1925
1926 newest = find_newest_super_backup(info, gen);
1927 if (newest == -1)
1928 return -1;
1929
1930 *backup_index = newest;
1931 *num_backups_tried = 1;
1932 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1933 /* we've tried all the backups, all done */
1934 return -1;
1935 } else {
1936 /* jump to the next oldest backup */
1937 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1938 BTRFS_NUM_BACKUP_ROOTS;
1939 *backup_index = newest;
1940 *num_backups_tried += 1;
1941 }
1942 root_backup = super->super_roots + newest;
1943
1944 btrfs_set_super_generation(super,
1945 btrfs_backup_tree_root_gen(root_backup));
1946 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1947 btrfs_set_super_root_level(super,
1948 btrfs_backup_tree_root_level(root_backup));
1949 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1950
1951 /*
1952 * fixme: the total bytes and num_devices need to match or we should
1953 * need a fsck
1954 */
1955 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1956 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1957 return 0;
1958}
1959
7abadb64
LB		 1960/* helper to cleanup workers */
1961static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1962{
1963 btrfs_stop_workers(&fs_info->generic_worker);
1964 btrfs_stop_workers(&fs_info->fixup_workers);
1965 btrfs_stop_workers(&fs_info->delalloc_workers);
1966 btrfs_stop_workers(&fs_info->workers);
1967 btrfs_stop_workers(&fs_info->endio_workers);
1968 btrfs_stop_workers(&fs_info->endio_meta_workers);
1969 btrfs_stop_workers(&fs_info->endio_raid56_workers);
1970 btrfs_stop_workers(&fs_info->rmw_workers);
1971 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1972 btrfs_stop_workers(&fs_info->endio_write_workers);
1973 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1974 btrfs_stop_workers(&fs_info->submit_workers);
1975 btrfs_stop_workers(&fs_info->delayed_workers);
1976 btrfs_stop_workers(&fs_info->caching_workers);
1977 btrfs_stop_workers(&fs_info->readahead_workers);
1978 btrfs_stop_workers(&fs_info->flush_workers);
2f232036 1979 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
7abadb64
LB		 1980}
1981
af31f5e5
CM		 1982/* helper to cleanup tree roots */
1983static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1984{
1985 free_extent_buffer(info->tree_root->node);
1986 free_extent_buffer(info->tree_root->commit_root);
1987 free_extent_buffer(info->dev_root->node);
1988 free_extent_buffer(info->dev_root->commit_root);
1989 free_extent_buffer(info->extent_root->node);
1990 free_extent_buffer(info->extent_root->commit_root);
1991 free_extent_buffer(info->csum_root->node);
1992 free_extent_buffer(info->csum_root->commit_root);
bcef60f2
AJ		 1993 if (info->quota_root) {
1994 free_extent_buffer(info->quota_root->node);
1995 free_extent_buffer(info->quota_root->commit_root);
1996 }
af31f5e5
CM		 1997
1998 info->tree_root->node = NULL;
1999 info->tree_root->commit_root = NULL;
2000 info->dev_root->node = NULL;
2001 info->dev_root->commit_root = NULL;
2002 info->extent_root->node = NULL;
2003 info->extent_root->commit_root = NULL;
2004 info->csum_root->node = NULL;
2005 info->csum_root->commit_root = NULL;
bcef60f2
AJ		 2006 if (info->quota_root) {
2007 info->quota_root->node = NULL;
2008 info->quota_root->commit_root = NULL;
2009 }
af31f5e5
CM		 2010
2011 if (chunk_root) {
2012 free_extent_buffer(info->chunk_root->node);
2013 free_extent_buffer(info->chunk_root->commit_root);
2014 info->chunk_root->node = NULL;
2015 info->chunk_root->commit_root = NULL;
2016 }
2017}
2018
171f6537
JB		 2019static void del_fs_roots(struct btrfs_fs_info *fs_info)
2020{
2021 int ret;
2022 struct btrfs_root *gang[8];
2023 int i;
2024
2025 while (!list_empty(&fs_info->dead_roots)) {
2026 gang[0] = list_entry(fs_info->dead_roots.next,
2027 struct btrfs_root, root_list);
2028 list_del(&gang[0]->root_list);
2029
2030 if (gang[0]->in_radix) {
2031 btrfs_free_fs_root(fs_info, gang[0]);
2032 } else {
2033 free_extent_buffer(gang[0]->node);
2034 free_extent_buffer(gang[0]->commit_root);
2035 kfree(gang[0]);
2036 }
2037 }
2038
2039 while (1) {
2040 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2041 (void **)gang, 0,
2042 ARRAY_SIZE(gang));
2043 if (!ret)
2044 break;
2045 for (i = 0; i < ret; i++)
2046 btrfs_free_fs_root(fs_info, gang[i]);
2047 }
2048}
af31f5e5 2049
ad2b2c80
AV		 2050int open_ctree(struct super_block *sb,
2051 struct btrfs_fs_devices *fs_devices,
2052 char *options)
2e635a27 2053{
db94535d
CM		 2054 u32 sectorsize;
2055 u32 nodesize;
2056 u32 leafsize;
2057 u32 blocksize;
87ee04eb 2058 u32 stripesize;
84234f3a 2059 u64 generation;
f2b636e8 2060 u64 features;
3de4586c 2061 struct btrfs_key location;
a061fc8d 2062 struct buffer_head *bh;
4d34b278 2063 struct btrfs_super_block *disk_super;
815745cf 2064 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2065 struct btrfs_root *tree_root;
4d34b278
ID		 2066 struct btrfs_root *extent_root;
2067 struct btrfs_root *csum_root;
2068 struct btrfs_root *chunk_root;
2069 struct btrfs_root *dev_root;
bcef60f2 2070 struct btrfs_root *quota_root;
e02119d5 2071 struct btrfs_root *log_tree_root;
eb60ceac 2072 int ret;
e58ca020 2073 int err = -EINVAL;
af31f5e5
CM		 2074 int num_backups_tried = 0;
2075 int backup_index = 0;
4543df7e 2076
f84a8bd6 2077 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e
AV		 2078 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
2079 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
2080 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2081 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
bcef60f2 2082 quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
8790d502 2083
f84a8bd6 2084 if (!tree_root || !extent_root || !csum_root ||
bcef60f2 2085 !chunk_root || !dev_root || !quota_root) {
39279cc3
CM		 2086 err = -ENOMEM;
2087 goto fail;
2088 }
76dda93c
YZ		 2089
2090 ret = init_srcu_struct(&fs_info->subvol_srcu);
2091 if (ret) {
2092 err = ret;
2093 goto fail;
2094 }
2095
2096 ret = setup_bdi(fs_info, &fs_info->bdi);
2097 if (ret) {
2098 err = ret;
2099 goto fail_srcu;
2100 }
2101
e2d84521
MX		 2102 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2103 if (ret) {
2104 err = ret;
2105 goto fail_bdi;
2106 }
2107 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2108 (1 + ilog2(nr_cpu_ids));
2109
963d678b
MX		 2110 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2111 if (ret) {
2112 err = ret;
2113 goto fail_dirty_metadata_bytes;
2114 }
2115
76dda93c
YZ		 2116 fs_info->btree_inode = new_inode(sb);
2117 if (!fs_info->btree_inode) {
2118 err = -ENOMEM;
963d678b 2119 goto fail_delalloc_bytes;
76dda93c
YZ		 2120 }
2121
a6591715 2122 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2123
76dda93c 2124 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 2125 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2126 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2127 INIT_LIST_HEAD(&fs_info->delayed_iputs);
ea8c2819 2128 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
11833d66 2129 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1832a6d5 2130 spin_lock_init(&fs_info->delalloc_lock);
a4abeea4 2131 spin_lock_init(&fs_info->trans_lock);
76dda93c 2132 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2133 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2134 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2135 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2136 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2137 spin_lock_init(&fs_info->super_lock);
f29021b2 2138 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2139 mutex_init(&fs_info->reloc_mutex);
de98ced9 2140 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2141
58176a96 2142 init_completion(&fs_info->kobj_unregister);
0b86a832 2143 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2144 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2145 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2146 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2147 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2148 BTRFS_BLOCK_RSV_GLOBAL);
2149 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2150 BTRFS_BLOCK_RSV_DELALLOC);
2151 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2152 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2153 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2154 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2155 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2156 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2157 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2158 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2159 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2160 atomic_set(&fs_info->defrag_running, 0);
fc36ed7e 2161 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2162 fs_info->sb = sb;
6f568d35 2163 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2164 fs_info->metadata_ratio = 0;
4cb5300b 2165 fs_info->defrag_inodes = RB_ROOT;
a4abeea4 2166 fs_info->trans_no_join = 0;
2bf64758 2167 fs_info->free_chunk_space = 0;
f29021b2 2168 fs_info->tree_mod_log = RB_ROOT;
c8b97818 2169
90519d66
AJ		 2170 /* readahead state */
2171 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2172 spin_lock_init(&fs_info->reada_lock);
c8b97818 2173
b34b086c
CM		 2174 fs_info->thread_pool_size = min_t(unsigned long,
2175 num_online_cpus() + 2, 8);
0afbaf8c 2176
3eaa2885
CM		 2177 INIT_LIST_HEAD(&fs_info->ordered_extents);
2178 spin_lock_init(&fs_info->ordered_extent_lock);
16cdcec7
MX		 2179 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2180 GFP_NOFS);
2181 if (!fs_info->delayed_root) {
2182 err = -ENOMEM;
2183 goto fail_iput;
2184 }
2185 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2186
a2de733c
AJ		 2187 mutex_init(&fs_info->scrub_lock);
2188 atomic_set(&fs_info->scrubs_running, 0);
2189 atomic_set(&fs_info->scrub_pause_req, 0);
2190 atomic_set(&fs_info->scrubs_paused, 0);
2191 atomic_set(&fs_info->scrub_cancel_req, 0);
2192 init_waitqueue_head(&fs_info->scrub_pause_wait);
2193 init_rwsem(&fs_info->scrub_super_lock);
2194 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2195#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2196 fs_info->check_integrity_print_mask = 0;
2197#endif
a2de733c 2198
c9e9f97b
ID		 2199 spin_lock_init(&fs_info->balance_lock);
2200 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2201 atomic_set(&fs_info->balance_running, 0);
2202 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2203 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2204 fs_info->balance_ctl = NULL;
837d5b6e 2205 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2206
a061fc8d
CM		 2207 sb->s_blocksize = 4096;
2208 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2209 sb->s_bdi = &fs_info->bdi;
a061fc8d 2210
76dda93c 2211 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2212 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2213 /*
2214 * we set the i_size on the btree inode to the max possible int.
2215 * the real end of the address space is determined by all of
2216 * the devices in the system
2217 */
2218 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2219 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2220 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2221
5d4f98a2 2222 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2223 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2224 fs_info->btree_inode->i_mapping);
0b32f4bb 2225 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2226 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2227
2228 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2229
76dda93c
YZ		 2230 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2231 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2232 sizeof(struct btrfs_key));
72ac3c0d
JB		 2233 set_bit(BTRFS_INODE_DUMMY,
2234 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
c65ddb52 2235 insert_inode_hash(fs_info->btree_inode);
76dda93c 2236
0f9dd46c 2237 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2238 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2239 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2240
11833d66 2241 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2242 fs_info->btree_inode->i_mapping);
11833d66 2243 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2244 fs_info->btree_inode->i_mapping);
11833d66 2245 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2246 fs_info->do_barriers = 1;
e18e4809 2247
39279cc3 2248
5a3f23d5 2249 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 2250 mutex_init(&fs_info->tree_log_mutex);
925baedd 2251 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2252 mutex_init(&fs_info->transaction_kthread_mutex);
2253 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2254 mutex_init(&fs_info->volume_mutex);
276e680d 2255 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2256 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2257 init_rwsem(&fs_info->subvol_sem);
e922e087
SB		 2258 fs_info->dev_replace.lock_owner = 0;
2259 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2260 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2261 mutex_init(&fs_info->dev_replace.lock_management_lock);
2262 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2263
416ac51d 2264 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2265 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2266 fs_info->qgroup_tree = RB_ROOT;
2267 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2268 fs_info->qgroup_seq = 1;
2269 fs_info->quota_enabled = 0;
2270 fs_info->pending_quota_state = 0;
2f232036 2271 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2272
fa9c0d79
CM		 2273 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2274 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2275
e6dcd2dc 2276 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2277 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2278 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2279 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2280
53b381b3
DW		 2281 ret = btrfs_alloc_stripe_hash_table(fs_info);
2282 if (ret) {
83c8266a 2283 err = ret;
53b381b3
DW		 2284 goto fail_alloc;
2285 }
2286
0b86a832 2287 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2288 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2289
3c4bb26b 2290 invalidate_bdev(fs_devices->latest_bdev);
a512bbf8 2291 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2292 if (!bh) {
2293 err = -EINVAL;
16cdcec7 2294 goto fail_alloc;
20b45077 2295 }
39279cc3 2296
6c41761f
DS		 2297 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2298 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2299 sizeof(*fs_info->super_for_commit));
a061fc8d 2300 brelse(bh);
5f39d397 2301
6c41761f 2302 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2303
6c41761f 2304 disk_super = fs_info->super_copy;
0f7d52f4 2305 if (!btrfs_super_root(disk_super))
16cdcec7 2306 goto fail_alloc;
0f7d52f4 2307
acce952b 2308 /* check FS state, whether FS is broken. */
87533c47
MX		 2309 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2310 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2311
fcd1f065
DS		 2312 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2313 if (ret) {
2314 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2315 err = ret;
2316 goto fail_alloc;
2317 }
acce952b 2318
af31f5e5
CM		 2319 /*
2320 * run through our array of backup supers and setup
2321 * our ring pointer to the oldest one
2322 */
2323 generation = btrfs_super_generation(disk_super);
2324 find_oldest_super_backup(fs_info, generation);
2325
75e7cb7f
LB		 2326 /*
2327 * In the long term, we'll store the compression type in the super
2328 * block, and it'll be used for per file compression control.
2329 */
2330 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2331
2b82032c
YZ		 2332 ret = btrfs_parse_options(tree_root, options);
2333 if (ret) {
2334 err = ret;
16cdcec7 2335 goto fail_alloc;
2b82032c 2336 }
dfe25020 2337
f2b636e8
JB		 2338 features = btrfs_super_incompat_flags(disk_super) &
2339 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2340 if (features) {
2341 printk(KERN_ERR "BTRFS: couldn't mount because of "
2342 "unsupported optional features (%Lx).\n",
21380931 2343 (unsigned long long)features);
f2b636e8 2344 err = -EINVAL;
16cdcec7 2345 goto fail_alloc;
f2b636e8
JB		 2346 }
2347
727011e0
CM		 2348 if (btrfs_super_leafsize(disk_super) !=
2349 btrfs_super_nodesize(disk_super)) {
2350 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2351 "blocksizes don't match. node %d leaf %d\n",
2352 btrfs_super_nodesize(disk_super),
2353 btrfs_super_leafsize(disk_super));
2354 err = -EINVAL;
2355 goto fail_alloc;
2356 }
2357 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2358 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2359 "blocksize (%d) was too large\n",
2360 btrfs_super_leafsize(disk_super));
2361 err = -EINVAL;
2362 goto fail_alloc;
2363 }
2364
5d4f98a2 2365 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2366 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2367 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2368 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2369
3173a18f
JB		 2370 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2371 printk(KERN_ERR "btrfs: has skinny extents\n");
2372
727011e0
CM		 2373 /*
2374 * flag our filesystem as having big metadata blocks if
2375 * they are bigger than the page size
2376 */
2377 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2378 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2379 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2380 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2381 }
2382
bc3f116f
CM		 2383 nodesize = btrfs_super_nodesize(disk_super);
2384 leafsize = btrfs_super_leafsize(disk_super);
2385 sectorsize = btrfs_super_sectorsize(disk_super);
2386 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2387 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2388 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2389
2390 /*
2391 * mixed block groups end up with duplicate but slightly offset
2392 * extent buffers for the same range. It leads to corruptions
2393 */
2394 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2395 (sectorsize != leafsize)) {
2396 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2397 "are not allowed for mixed block groups on %s\n",
2398 sb->s_id);
2399 goto fail_alloc;
2400 }
2401
ceda0864
MX		 2402 /*
2403 * Needn't use the lock because there is no other task which will
2404 * update the flag.
2405 */
a6fa6fae 2406 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2407
f2b636e8
JB		 2408 features = btrfs_super_compat_ro_flags(disk_super) &
2409 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2410 if (!(sb->s_flags & MS_RDONLY) && features) {
2411 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2412 "unsupported option features (%Lx).\n",
21380931 2413 (unsigned long long)features);
f2b636e8 2414 err = -EINVAL;
16cdcec7 2415 goto fail_alloc;
f2b636e8 2416 }
61d92c32
CM		 2417
2418 btrfs_init_workers(&fs_info->generic_worker,
2419 "genwork", 1, NULL);
2420
5443be45 2421 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 2422 fs_info->thread_pool_size,
2423 &fs_info->generic_worker);
c8b97818 2424
771ed689 2425 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 2426 fs_info->thread_pool_size,
2427 &fs_info->generic_worker);
771ed689 2428
8ccf6f19
MX		 2429 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2430 fs_info->thread_pool_size,
2431 &fs_info->generic_worker);
2432
5443be45 2433 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 2434 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 2435 fs_info->thread_pool_size),
2436 &fs_info->generic_worker);
61b49440 2437
bab39bf9
JB		 2438 btrfs_init_workers(&fs_info->caching_workers, "cache",
2439 2, &fs_info->generic_worker);
2440
61b49440
CM		 2441 /* a higher idle thresh on the submit workers makes it much more
2442 * likely that bios will be send down in a sane order to the
2443 * devices
2444 */
2445 fs_info->submit_workers.idle_thresh = 64;
53863232 2446
771ed689 2447 fs_info->workers.idle_thresh = 16;
4a69a410 2448 fs_info->workers.ordered = 1;
61b49440 2449
771ed689
CM		 2450 fs_info->delalloc_workers.idle_thresh = 2;
2451 fs_info->delalloc_workers.ordered = 1;
2452
61d92c32
CM		 2453 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2454 &fs_info->generic_worker);
5443be45 2455 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2456 fs_info->thread_pool_size,
2457 &fs_info->generic_worker);
d20f7043 2458 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2459 fs_info->thread_pool_size,
2460 &fs_info->generic_worker);
cad321ad 2461 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2462 "endio-meta-write", fs_info->thread_pool_size,
2463 &fs_info->generic_worker);
53b381b3
DW		 2464 btrfs_init_workers(&fs_info->endio_raid56_workers,
2465 "endio-raid56", fs_info->thread_pool_size,
2466 &fs_info->generic_worker);
2467 btrfs_init_workers(&fs_info->rmw_workers,
2468 "rmw", fs_info->thread_pool_size,
2469 &fs_info->generic_worker);
5443be45 2470 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2471 fs_info->thread_pool_size,
2472 &fs_info->generic_worker);
0cb59c99
JB		 2473 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2474 1, &fs_info->generic_worker);
16cdcec7
MX		 2475 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2476 fs_info->thread_pool_size,
2477 &fs_info->generic_worker);
90519d66
AJ		 2478 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2479 fs_info->thread_pool_size,
2480 &fs_info->generic_worker);
2f232036
JS		 2481 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2482 &fs_info->generic_worker);
61b49440
CM		 2483
2484 /*
2485 * endios are largely parallel and should have a very
2486 * low idle thresh
2487 */
2488 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2489 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2490 fs_info->endio_raid56_workers.idle_thresh = 4;
2491 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2492
9042846b
CM		 2493 fs_info->endio_write_workers.idle_thresh = 2;
2494 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2495 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2496
0dc3b84a
JB		 2497 /*
2498 * btrfs_start_workers can really only fail because of ENOMEM so just
2499 * return -ENOMEM if any of these fail.
2500 */
2501 ret = btrfs_start_workers(&fs_info->workers);
2502 ret |= btrfs_start_workers(&fs_info->generic_worker);
2503 ret |= btrfs_start_workers(&fs_info->submit_workers);
2504 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2505 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2506 ret |= btrfs_start_workers(&fs_info->endio_workers);
2507 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2508 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2509 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2510 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2511 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2512 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2513 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2514 ret |= btrfs_start_workers(&fs_info->caching_workers);
2515 ret |= btrfs_start_workers(&fs_info->readahead_workers);
8ccf6f19 2516 ret |= btrfs_start_workers(&fs_info->flush_workers);
2f232036 2517 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
0dc3b84a 2518 if (ret) {
fed425c7 2519 err = -ENOMEM;
0dc3b84a
JB		 2520 goto fail_sb_buffer;
2521 }
4543df7e 2522
4575c9cc 2523 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2524 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2525 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2526
db94535d
CM		 2527 tree_root->nodesize = nodesize;
2528 tree_root->leafsize = leafsize;
2529 tree_root->sectorsize = sectorsize;
87ee04eb 2530 tree_root->stripesize = stripesize;
a061fc8d
CM		 2531
2532 sb->s_blocksize = sectorsize;
2533 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2534
cdb4c574 2535 if (disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) {
d397712b 2536 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2537 goto fail_sb_buffer;
2538 }
19c00ddc 2539
8d082fb7
LB		 2540 if (sectorsize != PAGE_SIZE) {
2541 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2542 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2543 goto fail_sb_buffer;
2544 }
2545
925baedd 2546 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2547 ret = btrfs_read_sys_array(tree_root);
925baedd 2548 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2549 if (ret) {
d397712b
CM		 2550 printk(KERN_WARNING "btrfs: failed to read the system "
2551 "array on %s\n", sb->s_id);
5d4f98a2 2552 goto fail_sb_buffer;
84eed90f 2553 }
0b86a832
CM		 2554
2555 blocksize = btrfs_level_size(tree_root,
2556 btrfs_super_chunk_root_level(disk_super));
84234f3a 2557 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2558
2559 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2560 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2561
2562 chunk_root->node = read_tree_block(chunk_root,
2563 btrfs_super_chunk_root(disk_super),
84234f3a 2564 blocksize, generation);
416bc658
JB		 2565 if (!chunk_root->node ||
2566 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
83121942
DW		 2567 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2568 sb->s_id);
af31f5e5 2569 goto fail_tree_roots;
83121942 2570 }
5d4f98a2
YZ		 2571 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2572 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2573
e17cade2 2574 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 2575 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2576 BTRFS_UUID_SIZE);
e17cade2 2577
0b86a832 2578 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2579 if (ret) {
d397712b
CM		 2580 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2581 sb->s_id);
af31f5e5 2582 goto fail_tree_roots;
2b82032c 2583 }
0b86a832 2584
8dabb742
SB		 2585 /*
2586 * keep the device that is marked to be the target device for the
2587 * dev_replace procedure
2588 */
2589 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2590
a6b0d5c8
CM		 2591 if (!fs_devices->latest_bdev) {
2592 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2593 sb->s_id);
2594 goto fail_tree_roots;
2595 }
2596
af31f5e5 2597retry_root_backup:
db94535d
CM		 2598 blocksize = btrfs_level_size(tree_root,
2599 btrfs_super_root_level(disk_super));
84234f3a 2600 generation = btrfs_super_generation(disk_super);
0b86a832 2601
e20d96d6 2602 tree_root->node = read_tree_block(tree_root,
db94535d 2603 btrfs_super_root(disk_super),
84234f3a 2604 blocksize, generation);
af31f5e5
CM		 2605 if (!tree_root->node ||
2606 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
83121942
DW		 2607 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2608 sb->s_id);
af31f5e5
CM		 2609
2610 goto recovery_tree_root;
83121942 2611 }
af31f5e5 2612
5d4f98a2
YZ		 2613 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2614 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d
CM		 2615
2616 ret = find_and_setup_root(tree_root, fs_info,
e20d96d6 2617 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
0b86a832 2618 if (ret)
af31f5e5 2619 goto recovery_tree_root;
0b86a832
CM		 2620 extent_root->track_dirty = 1;
2621
2622 ret = find_and_setup_root(tree_root, fs_info,
2623 BTRFS_DEV_TREE_OBJECTID, dev_root);
0b86a832 2624 if (ret)
af31f5e5 2625 goto recovery_tree_root;
5d4f98a2 2626 dev_root->track_dirty = 1;
3768f368 2627
d20f7043
CM		 2628 ret = find_and_setup_root(tree_root, fs_info,
2629 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2630 if (ret)
af31f5e5 2631 goto recovery_tree_root;
d20f7043
CM		 2632 csum_root->track_dirty = 1;
2633
bcef60f2
AJ		 2634 ret = find_and_setup_root(tree_root, fs_info,
2635 BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2636 if (ret) {
2637 kfree(quota_root);
2638 quota_root = fs_info->quota_root = NULL;
2639 } else {
2640 quota_root->track_dirty = 1;
2641 fs_info->quota_enabled = 1;
2642 fs_info->pending_quota_state = 1;
2643 }
2644
8929ecfa
YZ		 2645 fs_info->generation = generation;
2646 fs_info->last_trans_committed = generation;
8929ecfa 2647
68310a5e
ID		 2648 ret = btrfs_recover_balance(fs_info);
2649 if (ret) {
2650 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2651 goto fail_block_groups;
2652 }
2653
733f4fbb
SB		 2654 ret = btrfs_init_dev_stats(fs_info);
2655 if (ret) {
2656 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2657 ret);
2658 goto fail_block_groups;
2659 }
2660
8dabb742
SB		 2661 ret = btrfs_init_dev_replace(fs_info);
2662 if (ret) {
2663 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2664 goto fail_block_groups;
2665 }
2666
2667 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2668
c59021f8 2669 ret = btrfs_init_space_info(fs_info);
2670 if (ret) {
2671 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2672 goto fail_block_groups;
2673 }
2674
1b1d1f66
JB		 2675 ret = btrfs_read_block_groups(extent_root);
2676 if (ret) {
2677 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2678 goto fail_block_groups;
2679 }
5af3e8cc
SB		 2680 fs_info->num_tolerated_disk_barrier_failures =
2681 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2682 if (fs_info->fs_devices->missing_devices >
2683 fs_info->num_tolerated_disk_barrier_failures &&
2684 !(sb->s_flags & MS_RDONLY)) {
2685 printk(KERN_WARNING
2686 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2687 goto fail_block_groups;
2688 }
9078a3e1 2689
a74a4b97
CM		 2690 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2691 "btrfs-cleaner");
57506d50 2692 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2693 goto fail_block_groups;
a74a4b97
CM		 2694
2695 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2696 tree_root,
2697 "btrfs-transaction");
57506d50 2698 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2699 goto fail_cleaner;
a74a4b97 2700
c289811c
CM		 2701 if (!btrfs_test_opt(tree_root, SSD) &&
2702 !btrfs_test_opt(tree_root, NOSSD) &&
2703 !fs_info->fs_devices->rotating) {
2704 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2705 "mode\n");
2706 btrfs_set_opt(fs_info->mount_opt, SSD);
2707 }
2708
21adbd5c
SB		 2709#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2710 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2711 ret = btrfsic_mount(tree_root, fs_devices,
2712 btrfs_test_opt(tree_root,
2713 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2714 1 : 0,
2715 fs_info->check_integrity_print_mask);
2716 if (ret)
2717 printk(KERN_WARNING "btrfs: failed to initialize"
2718 " integrity check module %s\n", sb->s_id);
2719 }
2720#endif
bcef60f2
AJ		 2721 ret = btrfs_read_qgroup_config(fs_info);
2722 if (ret)
2723 goto fail_trans_kthread;
21adbd5c 2724
acce952b 2725 /* do not make disk changes in broken FS */
68ce9682 2726 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2727 u64 bytenr = btrfs_super_log_root(disk_super);
2728
7c2ca468 2729 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2730 printk(KERN_WARNING "Btrfs log replay required "
2731 "on RO media\n");
7c2ca468 2732 err = -EIO;
bcef60f2 2733 goto fail_qgroup;
7c2ca468 2734 }
e02119d5
CM		 2735 blocksize =
2736 btrfs_level_size(tree_root,
2737 btrfs_super_log_root_level(disk_super));
d18a2c44 2738
6f07e42e 2739 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2740 if (!log_tree_root) {
2741 err = -ENOMEM;
bcef60f2 2742 goto fail_qgroup;
676e4c86 2743 }
e02119d5
CM		 2744
2745 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2746 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2747
2748 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2749 blocksize,
2750 generation + 1);
416bc658
JB		 2751 if (!log_tree_root->node ||
2752 !extent_buffer_uptodate(log_tree_root->node)) {
2753 printk(KERN_ERR "btrfs: failed to read log tree\n");
2754 free_extent_buffer(log_tree_root->node);
2755 kfree(log_tree_root);
2756 goto fail_trans_kthread;
2757 }
79787eaa 2758 /* returns with log_tree_root freed on success */
e02119d5 2759 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2760 if (ret) {
2761 btrfs_error(tree_root->fs_info, ret,
2762 "Failed to recover log tree");
2763 free_extent_buffer(log_tree_root->node);
2764 kfree(log_tree_root);
2765 goto fail_trans_kthread;
2766 }
e556ce2c
YZ		 2767
2768 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2769 ret = btrfs_commit_super(tree_root);
2770 if (ret)
2771 goto fail_trans_kthread;
e556ce2c 2772 }
e02119d5 2773 }
1a40e23b 2774
76dda93c 2775 ret = btrfs_find_orphan_roots(tree_root);
79787eaa
JM		 2776 if (ret)
2777 goto fail_trans_kthread;
76dda93c 2778
7c2ca468 2779 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2780 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2
ID		 2781 if (ret)
2782 goto fail_trans_kthread;
d68fc57b 2783
5d4f98a2 2784 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2785 if (ret < 0) {
2786 printk(KERN_WARNING
2787 "btrfs: failed to recover relocation\n");
2788 err = -EINVAL;
bcef60f2 2789 goto fail_qgroup;
d7ce5843 2790 }
7c2ca468 2791 }
1a40e23b 2792
3de4586c
CM		 2793 location.objectid = BTRFS_FS_TREE_OBJECTID;
2794 location.type = BTRFS_ROOT_ITEM_KEY;
2795 location.offset = (u64)-1;
2796
3de4586c
CM		 2797 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2798 if (!fs_info->fs_root)
bcef60f2 2799 goto fail_qgroup;
3140c9a3
DC		 2800 if (IS_ERR(fs_info->fs_root)) {
2801 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2802 goto fail_qgroup;
3140c9a3 2803 }
c289811c 2804
2b6ba629
ID		 2805 if (sb->s_flags & MS_RDONLY)
2806 return 0;
59641015 2807
2b6ba629
ID		 2808 down_read(&fs_info->cleanup_work_sem);
2809 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2810 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2811 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2812 close_ctree(tree_root);
2813 return ret;
2814 }
2815 up_read(&fs_info->cleanup_work_sem);
59641015 2816
2b6ba629
ID		 2817 ret = btrfs_resume_balance_async(fs_info);
2818 if (ret) {
2819 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2820 close_ctree(tree_root);
2821 return ret;
e3acc2a6
JB		 2822 }
2823
8dabb742
SB		 2824 ret = btrfs_resume_dev_replace_async(fs_info);
2825 if (ret) {
2826 pr_warn("btrfs: failed to resume dev_replace\n");
2827 close_ctree(tree_root);
2828 return ret;
2829 }
2830
ad2b2c80 2831 return 0;
39279cc3 2832
bcef60f2
AJ		 2833fail_qgroup:
2834 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2835fail_trans_kthread:
2836 kthread_stop(fs_info->transaction_kthread);
171f6537 2837 del_fs_roots(fs_info);
54067ae9 2838 btrfs_cleanup_transaction(fs_info->tree_root);
3f157a2f 2839fail_cleaner:
a74a4b97 2840 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2841
2842 /*
2843 * make sure we're done with the btree inode before we stop our
2844 * kthreads
2845 */
2846 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2847
1b1d1f66 2848fail_block_groups:
54067ae9 2849 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2850 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2851
2852fail_tree_roots:
2853 free_root_pointers(fs_info, 1);
2b8195bb 2854 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2855
39279cc3 2856fail_sb_buffer:
7abadb64 2857 btrfs_stop_all_workers(fs_info);
16cdcec7 2858fail_alloc:
4543df7e 2859fail_iput:
586e46e2
ID		 2860 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2861
4543df7e 2862 iput(fs_info->btree_inode);
963d678b
MX		 2863fail_delalloc_bytes:
2864 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 2865fail_dirty_metadata_bytes:
2866 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 2867fail_bdi:
7e662854 2868 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2869fail_srcu:
2870 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2871fail:
53b381b3 2872 btrfs_free_stripe_hash_table(fs_info);
586e46e2 2873 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 2874 return err;
af31f5e5
CM		 2875
2876recovery_tree_root:
af31f5e5
CM		 2877 if (!btrfs_test_opt(tree_root, RECOVERY))
2878 goto fail_tree_roots;
2879
2880 free_root_pointers(fs_info, 0);
2881
2882 /* don't use the log in recovery mode, it won't be valid */
2883 btrfs_set_super_log_root(disk_super, 0);
2884
2885 /* we can't trust the free space cache either */
2886 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2887
2888 ret = next_root_backup(fs_info, fs_info->super_copy,
2889 &num_backups_tried, &backup_index);
2890 if (ret == -1)
2891 goto fail_block_groups;
2892 goto retry_root_backup;
eb60ceac
CM		 2893}
2894
f2984462
CM		 2895static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2896{
f2984462
CM		 2897 if (uptodate) {
2898 set_buffer_uptodate(bh);
2899 } else {
442a4f63
SB		 2900 struct btrfs_device *device = (struct btrfs_device *)
2901 bh->b_private;
2902
606686ee
JB		 2903 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2904 "I/O error on %s\n",
2905 rcu_str_deref(device->name));
1259ab75
CM		 2906 /* note, we dont' set_buffer_write_io_error because we have
2907 * our own ways of dealing with the IO errors
2908 */
f2984462 2909 clear_buffer_uptodate(bh);
442a4f63 2910 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 2911 }
2912 unlock_buffer(bh);
2913 put_bh(bh);
2914}
2915
a512bbf8
YZ		 2916struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2917{
2918 struct buffer_head *bh;
2919 struct buffer_head *latest = NULL;
2920 struct btrfs_super_block *super;
2921 int i;
2922 u64 transid = 0;
2923 u64 bytenr;
2924
2925 /* we would like to check all the supers, but that would make
2926 * a btrfs mount succeed after a mkfs from a different FS.
2927 * So, we need to add a special mount option to scan for
2928 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2929 */
2930 for (i = 0; i < 1; i++) {
2931 bytenr = btrfs_sb_offset(i);
2932 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2933 break;
2934 bh = __bread(bdev, bytenr / 4096, 4096);
2935 if (!bh)
2936 continue;
2937
2938 super = (struct btrfs_super_block *)bh->b_data;
2939 if (btrfs_super_bytenr(super) != bytenr ||
cdb4c574 2940 super->magic != cpu_to_le64(BTRFS_MAGIC)) {
a512bbf8
YZ		 2941 brelse(bh);
2942 continue;
2943 }
2944
2945 if (!latest || btrfs_super_generation(super) > transid) {
2946 brelse(latest);
2947 latest = bh;
2948 transid = btrfs_super_generation(super);
2949 } else {
2950 brelse(bh);
2951 }
2952 }
2953 return latest;
2954}
2955
4eedeb75
HH		 2956/*
2957 * this should be called twice, once with wait == 0 and
2958 * once with wait == 1. When wait == 0 is done, all the buffer heads
2959 * we write are pinned.
2960 *
2961 * They are released when wait == 1 is done.
2962 * max_mirrors must be the same for both runs, and it indicates how
2963 * many supers on this one device should be written.
2964 *
2965 * max_mirrors == 0 means to write them all.
2966 */
a512bbf8
YZ		 2967static int write_dev_supers(struct btrfs_device *device,
2968 struct btrfs_super_block *sb,
2969 int do_barriers, int wait, int max_mirrors)
2970{
2971 struct buffer_head *bh;
2972 int i;
2973 int ret;
2974 int errors = 0;
2975 u32 crc;
2976 u64 bytenr;
a512bbf8
YZ		 2977
2978 if (max_mirrors == 0)
2979 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2980
a512bbf8
YZ		 2981 for (i = 0; i < max_mirrors; i++) {
2982 bytenr = btrfs_sb_offset(i);
2983 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2984 break;
2985
2986 if (wait) {
2987 bh = __find_get_block(device->bdev, bytenr / 4096,
2988 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 2989 if (!bh) {
2990 errors++;
2991 continue;
2992 }
a512bbf8 2993 wait_on_buffer(bh);
4eedeb75
HH		 2994 if (!buffer_uptodate(bh))
2995 errors++;
2996
2997 /* drop our reference */
2998 brelse(bh);
2999
3000 /* drop the reference from the wait == 0 run */
3001 brelse(bh);
3002 continue;
a512bbf8
YZ		 3003 } else {
3004 btrfs_set_super_bytenr(sb, bytenr);
3005
3006 crc = ~(u32)0;
b0496686 3007 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 3008 BTRFS_CSUM_SIZE, crc,
3009 BTRFS_SUPER_INFO_SIZE -
3010 BTRFS_CSUM_SIZE);
3011 btrfs_csum_final(crc, sb->csum);
3012
4eedeb75
HH		 3013 /*
3014 * one reference for us, and we leave it for the
3015 * caller
3016 */
a512bbf8
YZ		 3017 bh = __getblk(device->bdev, bytenr / 4096,
3018 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3019 if (!bh) {
3020 printk(KERN_ERR "btrfs: couldn't get super "
3021 "buffer head for bytenr %Lu\n", bytenr);
3022 errors++;
3023 continue;
3024 }
3025
a512bbf8
YZ		 3026 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3027
4eedeb75 3028 /* one reference for submit_bh */
a512bbf8 3029 get_bh(bh);
4eedeb75
HH		 3030
3031 set_buffer_uptodate(bh);
a512bbf8
YZ		 3032 lock_buffer(bh);
3033 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3034 bh->b_private = device;
a512bbf8
YZ		 3035 }
3036
387125fc
CM		 3037 /*
3038 * we fua the first super. The others we allow
3039 * to go down lazy.
3040 */
21adbd5c 3041 ret = btrfsic_submit_bh(WRITE_FUA, bh);
4eedeb75 3042 if (ret)
a512bbf8 3043 errors++;
a512bbf8
YZ		 3044 }
3045 return errors < i ? 0 : -1;
3046}
3047
387125fc
CM		 3048/*
3049 * endio for the write_dev_flush, this will wake anyone waiting
3050 * for the barrier when it is done
3051 */
3052static void btrfs_end_empty_barrier(struct bio *bio, int err)
3053{
3054 if (err) {
3055 if (err == -EOPNOTSUPP)
3056 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3057 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3058 }
3059 if (bio->bi_private)
3060 complete(bio->bi_private);
3061 bio_put(bio);
3062}
3063
3064/*
3065 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3066 * sent down. With wait == 1, it waits for the previous flush.
3067 *
3068 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3069 * capable
3070 */
3071static int write_dev_flush(struct btrfs_device *device, int wait)
3072{
3073 struct bio *bio;
3074 int ret = 0;
3075
3076 if (device->nobarriers)
3077 return 0;
3078
3079 if (wait) {
3080 bio = device->flush_bio;
3081 if (!bio)
3082 return 0;
3083
3084 wait_for_completion(&device->flush_wait);
3085
3086 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
606686ee
JB		 3087 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3088 rcu_str_deref(device->name));
387125fc 3089 device->nobarriers = 1;
5af3e8cc 3090 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3091 ret = -EIO;
5af3e8cc
SB		 3092 btrfs_dev_stat_inc_and_print(device,
3093 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3094 }
3095
3096 /* drop the reference from the wait == 0 run */
3097 bio_put(bio);
3098 device->flush_bio = NULL;
3099
3100 return ret;
3101 }
3102
3103 /*
3104 * one reference for us, and we leave it for the
3105 * caller
3106 */
9c017abc 3107 device->flush_bio = NULL;
387125fc
CM		 3108 bio = bio_alloc(GFP_NOFS, 0);
3109 if (!bio)
3110 return -ENOMEM;
3111
3112 bio->bi_end_io = btrfs_end_empty_barrier;
3113 bio->bi_bdev = device->bdev;
3114 init_completion(&device->flush_wait);
3115 bio->bi_private = &device->flush_wait;
3116 device->flush_bio = bio;
3117
3118 bio_get(bio);
21adbd5c 3119 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3120
3121 return 0;
3122}
3123
3124/*
3125 * send an empty flush down to each device in parallel,
3126 * then wait for them
3127 */
3128static int barrier_all_devices(struct btrfs_fs_info *info)
3129{
3130 struct list_head *head;
3131 struct btrfs_device *dev;
5af3e8cc
SB		 3132 int errors_send = 0;
3133 int errors_wait = 0;
387125fc
CM		 3134 int ret;
3135
3136 /* send down all the barriers */
3137 head = &info->fs_devices->devices;
3138 list_for_each_entry_rcu(dev, head, dev_list) {
3139 if (!dev->bdev) {
5af3e8cc 3140 errors_send++;
387125fc
CM		 3141 continue;
3142 }
3143 if (!dev->in_fs_metadata || !dev->writeable)
3144 continue;
3145
3146 ret = write_dev_flush(dev, 0);
3147 if (ret)
5af3e8cc 3148 errors_send++;
387125fc
CM		 3149 }
3150
3151 /* wait for all the barriers */
3152 list_for_each_entry_rcu(dev, head, dev_list) {
3153 if (!dev->bdev) {
5af3e8cc 3154 errors_wait++;
387125fc
CM		 3155 continue;
3156 }
3157 if (!dev->in_fs_metadata || !dev->writeable)
3158 continue;
3159
3160 ret = write_dev_flush(dev, 1);
3161 if (ret)
5af3e8cc 3162 errors_wait++;
387125fc 3163 }
5af3e8cc
SB		 3164 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3165 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3166 return -EIO;
3167 return 0;
3168}
3169
5af3e8cc
SB		 3170int btrfs_calc_num_tolerated_disk_barrier_failures(
3171 struct btrfs_fs_info *fs_info)
3172{
3173 struct btrfs_ioctl_space_info space;
3174 struct btrfs_space_info *sinfo;
3175 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3176 BTRFS_BLOCK_GROUP_SYSTEM,
3177 BTRFS_BLOCK_GROUP_METADATA,
3178 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3179 int num_types = 4;
3180 int i;
3181 int c;
3182 int num_tolerated_disk_barrier_failures =
3183 (int)fs_info->fs_devices->num_devices;
3184
3185 for (i = 0; i < num_types; i++) {
3186 struct btrfs_space_info *tmp;
3187
3188 sinfo = NULL;
3189 rcu_read_lock();
3190 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3191 if (tmp->flags == types[i]) {
3192 sinfo = tmp;
3193 break;
3194 }
3195 }
3196 rcu_read_unlock();
3197
3198 if (!sinfo)
3199 continue;
3200
3201 down_read(&sinfo->groups_sem);
3202 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3203 if (!list_empty(&sinfo->block_groups[c])) {
3204 u64 flags;
3205
3206 btrfs_get_block_group_info(
3207 &sinfo->block_groups[c], &space);
3208 if (space.total_bytes == 0 ||
3209 space.used_bytes == 0)
3210 continue;
3211 flags = space.flags;
3212 /*
3213 * return
3214 * 0: if dup, single or RAID0 is configured for
3215 * any of metadata, system or data, else
3216 * 1: if RAID5 is configured, or if RAID1 or
3217 * RAID10 is configured and only two mirrors
3218 * are used, else
3219 * 2: if RAID6 is configured, else
3220 * num_mirrors - 1: if RAID1 or RAID10 is
3221 * configured and more than
3222 * 2 mirrors are used.
3223 */
3224 if (num_tolerated_disk_barrier_failures > 0 &&
3225 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3226 BTRFS_BLOCK_GROUP_RAID0)) ||
3227 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3228 == 0)))
3229 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3230 else if (num_tolerated_disk_barrier_failures > 1) {
3231 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3232 BTRFS_BLOCK_GROUP_RAID5 |
3233 BTRFS_BLOCK_GROUP_RAID10)) {
3234 num_tolerated_disk_barrier_failures = 1;
3235 } else if (flags &
3236 BTRFS_BLOCK_GROUP_RAID5) {
3237 num_tolerated_disk_barrier_failures = 2;
3238 }
3239 }
5af3e8cc
SB		 3240 }
3241 }
3242 up_read(&sinfo->groups_sem);
3243 }
3244
3245 return num_tolerated_disk_barrier_failures;
3246}
3247
a512bbf8 3248int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3249{
e5e9a520 3250 struct list_head *head;
f2984462 3251 struct btrfs_device *dev;
a061fc8d 3252 struct btrfs_super_block *sb;
f2984462 3253 struct btrfs_dev_item *dev_item;
f2984462
CM		 3254 int ret;
3255 int do_barriers;
a236aed1
CM		 3256 int max_errors;
3257 int total_errors = 0;
a061fc8d 3258 u64 flags;
f2984462 3259
6c41761f 3260 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
f2984462 3261 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3262 backup_super_roots(root->fs_info);
f2984462 3263
6c41761f 3264 sb = root->fs_info->super_for_commit;
a061fc8d 3265 dev_item = &sb->dev_item;
e5e9a520 3266
174ba509 3267 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3268 head = &root->fs_info->fs_devices->devices;
387125fc 3269
5af3e8cc
SB		 3270 if (do_barriers) {
3271 ret = barrier_all_devices(root->fs_info);
3272 if (ret) {
3273 mutex_unlock(
3274 &root->fs_info->fs_devices->device_list_mutex);
3275 btrfs_error(root->fs_info, ret,
3276 "errors while submitting device barriers.");
3277 return ret;
3278 }
3279 }
387125fc 3280
1f78160c 3281 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3282 if (!dev->bdev) {
3283 total_errors++;
3284 continue;
3285 }
2b82032c 3286 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3287 continue;
3288
2b82032c 3289 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3290 btrfs_set_stack_device_type(dev_item, dev->type);
3291 btrfs_set_stack_device_id(dev_item, dev->devid);
3292 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3293 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3294 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3295 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3296 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3297 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3298 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3299
a061fc8d
CM		 3300 flags = btrfs_super_flags(sb);
3301 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3302
a512bbf8 3303 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3304 if (ret)
3305 total_errors++;
f2984462 3306 }
a236aed1 3307 if (total_errors > max_errors) {
d397712b
CM		 3308 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3309 total_errors);
79787eaa
JM		 3310
3311 /* This shouldn't happen. FUA is masked off if unsupported */
a236aed1
CM		 3312 BUG();
3313 }
f2984462 3314
a512bbf8 3315 total_errors = 0;
1f78160c 3316 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3317 if (!dev->bdev)
3318 continue;
2b82032c 3319 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3320 continue;
3321
a512bbf8
YZ		 3322 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3323 if (ret)
3324 total_errors++;
f2984462 3325 }
174ba509 3326 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3327 if (total_errors > max_errors) {
79787eaa
JM		 3328 btrfs_error(root->fs_info, -EIO,
3329 "%d errors while writing supers", total_errors);
3330 return -EIO;
a236aed1 3331 }
f2984462
CM		 3332 return 0;
3333}
3334
a512bbf8
YZ		 3335int write_ctree_super(struct btrfs_trans_handle *trans,
3336 struct btrfs_root *root, int max_mirrors)
eb60ceac 3337{
e66f709b 3338 int ret;
5f39d397 3339
a512bbf8 3340 ret = write_all_supers(root, max_mirrors);
5f39d397 3341 return ret;
cfaa7295
CM		 3342}
3343
143bede5 3344void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 3345{
4df27c4d 3346 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3347 radix_tree_delete(&fs_info->fs_roots_radix,
3348 (unsigned long)root->root_key.objectid);
4df27c4d 3349 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3350
3351 if (btrfs_root_refs(&root->root_item) == 0)
3352 synchronize_srcu(&fs_info->subvol_srcu);
3353
d7634482 3354 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3321719e
LB		 3355 btrfs_free_log(NULL, root);
3356 btrfs_free_log_root_tree(NULL, fs_info);
3357 }
3358
581bb050
LZ		 3359 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3360 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3361 free_fs_root(root);
4df27c4d
YZ		 3362}
3363
3364static void free_fs_root(struct btrfs_root *root)
3365{
82d5902d 3366 iput(root->cache_inode);
4df27c4d 3367 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
0ee5dc67
AV		 3368 if (root->anon_dev)
3369 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 3370 free_extent_buffer(root->node);
3371 free_extent_buffer(root->commit_root);
581bb050
LZ		 3372 kfree(root->free_ino_ctl);
3373 kfree(root->free_ino_pinned);
d397712b 3374 kfree(root->name);
2619ba1f 3375 kfree(root);
2619ba1f
CM		 3376}
3377
c146afad 3378int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3379{
c146afad
YZ		 3380 u64 root_objectid = 0;
3381 struct btrfs_root *gang[8];
3382 int i;
3768f368 3383 int ret;
e089f05c 3384
c146afad
YZ		 3385 while (1) {
3386 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3387 (void **)gang, root_objectid,
3388 ARRAY_SIZE(gang));
3389 if (!ret)
3390 break;
5d4f98a2
YZ		 3391
3392 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 3393 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 3394 int err;
3395
c146afad 3396 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3397 err = btrfs_orphan_cleanup(gang[i]);
3398 if (err)
3399 return err;
c146afad
YZ		 3400 }
3401 root_objectid++;
3402 }
3403 return 0;
3404}
a2135011 3405
c146afad
YZ		 3406int btrfs_commit_super(struct btrfs_root *root)
3407{
3408 struct btrfs_trans_handle *trans;
3409 int ret;
a74a4b97 3410
c146afad 3411 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3412 btrfs_run_delayed_iputs(root);
c146afad 3413 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3414 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3415
3416 /* wait until ongoing cleanup work done */
3417 down_write(&root->fs_info->cleanup_work_sem);
3418 up_write(&root->fs_info->cleanup_work_sem);
3419
7a7eaa40 3420 trans = btrfs_join_transaction(root);
3612b495
TI		 3421 if (IS_ERR(trans))
3422 return PTR_ERR(trans);
54aa1f4d 3423 ret = btrfs_commit_transaction(trans, root);
79787eaa
JM		 3424 if (ret)
3425 return ret;
c146afad 3426 /* run commit again to drop the original snapshot */
7a7eaa40 3427 trans = btrfs_join_transaction(root);
3612b495
TI		 3428 if (IS_ERR(trans))
3429 return PTR_ERR(trans);
79787eaa
JM		 3430 ret = btrfs_commit_transaction(trans, root);
3431 if (ret)
3432 return ret;
79154b1b 3433 ret = btrfs_write_and_wait_transaction(NULL, root);
79787eaa
JM		 3434 if (ret) {
3435 btrfs_error(root->fs_info, ret,
3436 "Failed to sync btree inode to disk.");
3437 return ret;
3438 }
d6bfde87 3439
a512bbf8 3440 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 3441 return ret;
3442}
3443
3444int close_ctree(struct btrfs_root *root)
3445{
3446 struct btrfs_fs_info *fs_info = root->fs_info;
3447 int ret;
3448
3449 fs_info->closing = 1;
3450 smp_mb();
3451
837d5b6e 3452 /* pause restriper - we want to resume on mount */
aa1b8cd4 3453 btrfs_pause_balance(fs_info);
837d5b6e 3454
8dabb742
SB		 3455 btrfs_dev_replace_suspend_for_unmount(fs_info);
3456
aa1b8cd4 3457 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3458
3459 /* wait for any defraggers to finish */
3460 wait_event(fs_info->transaction_wait,
3461 (atomic_read(&fs_info->defrag_running) == 0));
3462
3463 /* clear out the rbtree of defraggable inodes */
26176e7c 3464 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3465
c146afad 3466 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3467 ret = btrfs_commit_super(root);
3468 if (ret)
3469 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3470 }
3471
87533c47 3472 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3473 btrfs_error_commit_super(root);
0f7d52f4 3474
300e4f8a
JB		 3475 btrfs_put_block_group_cache(fs_info);
3476
e3029d9f
AV		 3477 kthread_stop(fs_info->transaction_kthread);
3478 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3479
f25784b3
YZ		 3480 fs_info->closing = 2;
3481 smp_mb();
3482
bcef60f2
AJ		 3483 btrfs_free_qgroup_config(root->fs_info);
3484
963d678b
MX		 3485 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3486 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3487 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3488 }
bcc63abb 3489
be283b2e 3490 free_root_pointers(fs_info, 1);
d20f7043 3491
e3029d9f 3492 btrfs_free_block_groups(fs_info);
d10c5f31 3493
c146afad 3494 del_fs_roots(fs_info);
d10c5f31 3495
c146afad 3496 iput(fs_info->btree_inode);
9ad6b7bc 3497
7abadb64 3498 btrfs_stop_all_workers(fs_info);
d6bfde87 3499
21adbd5c
SB		 3500#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3501 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3502 btrfsic_unmount(root, fs_info->fs_devices);
3503#endif
3504
dfe25020 3505 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3506 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3507
e2d84521 3508 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3509 percpu_counter_destroy(&fs_info->delalloc_bytes);
04160088 3510 bdi_destroy(&fs_info->bdi);
76dda93c 3511 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3512
53b381b3
DW		 3513 btrfs_free_stripe_hash_table(fs_info);
3514
eb60ceac
CM		 3515 return 0;
3516}
3517
b9fab919
CM		 3518int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3519 int atomic)
5f39d397 3520{
1259ab75 3521 int ret;
727011e0 3522 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3523
0b32f4bb 3524 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3525 if (!ret)
3526 return ret;
3527
3528 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3529 parent_transid, atomic);
3530 if (ret == -EAGAIN)
3531 return ret;
1259ab75 3532 return !ret;
5f39d397
CM		 3533}
3534
3535int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3536{
0b32f4bb 3537 return set_extent_buffer_uptodate(buf);
5f39d397 3538}
6702ed49 3539
5f39d397
CM		 3540void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3541{
727011e0 3542 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
5f39d397 3543 u64 transid = btrfs_header_generation(buf);
b9473439 3544 int was_dirty;
b4ce94de 3545
b9447ef8 3546 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3547 if (transid != root->fs_info->generation)
3548 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3549 "found %llu running %llu\n",
db94535d 3550 (unsigned long long)buf->start,
d397712b
CM		 3551 (unsigned long long)transid,
3552 (unsigned long long)root->fs_info->generation);
0b32f4bb 3553 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3554 if (!was_dirty)
3555 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3556 buf->len,
3557 root->fs_info->dirty_metadata_batch);
eb60ceac
CM		 3558}
3559
b53d3f5d
LB		 3560static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3561 int flush_delayed)
16cdcec7
MX		 3562{
3563 /*
3564 * looks as though older kernels can get into trouble with
3565 * this code, they end up stuck in balance_dirty_pages forever
3566 */
e2d84521 3567 int ret;
16cdcec7
MX		 3568
3569 if (current->flags & PF_MEMALLOC)
3570 return;
3571
b53d3f5d
LB		 3572 if (flush_delayed)
3573 btrfs_balance_delayed_items(root);
16cdcec7 3574
e2d84521
MX		 3575 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3576 BTRFS_DIRTY_METADATA_THRESH);
3577 if (ret > 0) {
d0e1d66b
NJ		 3578 balance_dirty_pages_ratelimited(
3579 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3580 }
3581 return;
3582}
3583
b53d3f5d 3584void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3585{
b53d3f5d
LB		 3586 __btrfs_btree_balance_dirty(root, 1);
3587}
585ad2c3 3588
b53d3f5d
LB		 3589void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3590{
3591 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3592}
6b80053d 3593
ca7a79ad 3594int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3595{
727011e0 3596 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3597 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3598}
0da5468f 3599
fcd1f065 3600static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3601 int read_only)
3602{
fcd1f065
DS		 3603 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3604 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3605 return -EINVAL;
3606 }
3607
acce952b 3608 if (read_only)
fcd1f065 3609 return 0;
acce952b 3610
fcd1f065 3611 return 0;
acce952b 3612}
3613
68ce9682 3614void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3615{
acce952b 3616 mutex_lock(&root->fs_info->cleaner_mutex);
3617 btrfs_run_delayed_iputs(root);
3618 mutex_unlock(&root->fs_info->cleaner_mutex);
3619
3620 down_write(&root->fs_info->cleanup_work_sem);
3621 up_write(&root->fs_info->cleanup_work_sem);
3622
3623 /* cleanup FS via transaction */
3624 btrfs_cleanup_transaction(root);
acce952b 3625}
3626
569e0f35
JB		 3627static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3628 struct btrfs_root *root)
acce952b 3629{
3630 struct btrfs_inode *btrfs_inode;
3631 struct list_head splice;
3632
3633 INIT_LIST_HEAD(&splice);
3634
3635 mutex_lock(&root->fs_info->ordered_operations_mutex);
3636 spin_lock(&root->fs_info->ordered_extent_lock);
3637
569e0f35 3638 list_splice_init(&t->ordered_operations, &splice);
acce952b 3639 while (!list_empty(&splice)) {
3640 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3641 ordered_operations);
3642
3643 list_del_init(&btrfs_inode->ordered_operations);
3644
3645 btrfs_invalidate_inodes(btrfs_inode->root);
3646 }
3647
3648 spin_unlock(&root->fs_info->ordered_extent_lock);
3649 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3650}
3651
143bede5 3652static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3653{
acce952b 3654 struct btrfs_ordered_extent *ordered;
acce952b 3655
3656 spin_lock(&root->fs_info->ordered_extent_lock);
779880ef
JB		 3657 /*
3658 * This will just short circuit the ordered completion stuff which will
3659 * make sure the ordered extent gets properly cleaned up.
3660 */
3661 list_for_each_entry(ordered, &root->fs_info->ordered_extents,
3662 root_extent_list)
3663 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
acce952b 3664 spin_unlock(&root->fs_info->ordered_extent_lock);
acce952b 3665}
3666
79787eaa
JM		 3667int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3668 struct btrfs_root *root)
acce952b 3669{
3670 struct rb_node *node;
3671 struct btrfs_delayed_ref_root *delayed_refs;
3672 struct btrfs_delayed_ref_node *ref;
3673 int ret = 0;
3674
3675 delayed_refs = &trans->delayed_refs;
3676
3677 spin_lock(&delayed_refs->lock);
3678 if (delayed_refs->num_entries == 0) {
cfece4db 3679 spin_unlock(&delayed_refs->lock);
acce952b 3680 printk(KERN_INFO "delayed_refs has NO entry\n");
3681 return ret;
3682 }
3683
b939d1ab 3684 while ((node = rb_first(&delayed_refs->root)) != NULL) {
eb12db69 3685 struct btrfs_delayed_ref_head *head = NULL;
acce952b 3686
eb12db69 3687 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
acce952b 3688 atomic_set(&ref->refs, 1);
3689 if (btrfs_delayed_ref_is_head(ref)) {
acce952b 3690
3691 head = btrfs_delayed_node_to_head(ref);
b939d1ab
JB		 3692 if (!mutex_trylock(&head->mutex)) {
3693 atomic_inc(&ref->refs);
3694 spin_unlock(&delayed_refs->lock);
3695
3696 /* Need to wait for the delayed ref to run */
3697 mutex_lock(&head->mutex);
3698 mutex_unlock(&head->mutex);
3699 btrfs_put_delayed_ref(ref);
3700
e18fca73 3701 spin_lock(&delayed_refs->lock);
b939d1ab
JB		 3702 continue;
3703 }
3704
54067ae9
JB		 3705 if (head->must_insert_reserved)
3706 btrfs_pin_extent(root, ref->bytenr,
3707 ref->num_bytes, 1);
78a6184a 3708 btrfs_free_delayed_extent_op(head->extent_op);
acce952b 3709 delayed_refs->num_heads--;
3710 if (list_empty(&head->cluster))
3711 delayed_refs->num_heads_ready--;
3712 list_del_init(&head->cluster);
acce952b 3713 }
eb12db69 3714
b939d1ab
JB		 3715 ref->in_tree = 0;
3716 rb_erase(&ref->rb_node, &delayed_refs->root);
3717 delayed_refs->num_entries--;
eb12db69
JB		 3718 if (head)
3719 mutex_unlock(&head->mutex);
acce952b 3720 spin_unlock(&delayed_refs->lock);
3721 btrfs_put_delayed_ref(ref);
3722
3723 cond_resched();
3724 spin_lock(&delayed_refs->lock);
3725 }
3726
3727 spin_unlock(&delayed_refs->lock);
3728
3729 return ret;
3730}
3731
aec8030a 3732static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
acce952b 3733{
3734 struct btrfs_pending_snapshot *snapshot;
3735 struct list_head splice;
3736
3737 INIT_LIST_HEAD(&splice);
3738
3739 list_splice_init(&t->pending_snapshots, &splice);
3740
3741 while (!list_empty(&splice)) {
3742 snapshot = list_entry(splice.next,
3743 struct btrfs_pending_snapshot,
3744 list);
aec8030a 3745 snapshot->error = -ECANCELED;
acce952b 3746 list_del_init(&snapshot->list);
acce952b 3747 }
acce952b 3748}
3749
143bede5 3750static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3751{
3752 struct btrfs_inode *btrfs_inode;
3753 struct list_head splice;
3754
3755 INIT_LIST_HEAD(&splice);
3756
acce952b 3757 spin_lock(&root->fs_info->delalloc_lock);
5be76758 3758 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
acce952b 3759
3760 while (!list_empty(&splice)) {
3761 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3762 delalloc_inodes);
3763
3764 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3765 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3766 &btrfs_inode->runtime_flags);
acce952b 3767
3768 btrfs_invalidate_inodes(btrfs_inode->root);
3769 }
3770
3771 spin_unlock(&root->fs_info->delalloc_lock);
acce952b 3772}
3773
3774static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3775 struct extent_io_tree *dirty_pages,
3776 int mark)
3777{
3778 int ret;
acce952b 3779 struct extent_buffer *eb;
3780 u64 start = 0;
3781 u64 end;
acce952b 3782
3783 while (1) {
3784 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3785 mark, NULL);
acce952b 3786 if (ret)
3787 break;
3788
3789 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3790 while (start <= end) {
fd8b2b61
JB		 3791 eb = btrfs_find_tree_block(root, start,
3792 root->leafsize);
3793 start += eb->len;
3794 if (!eb)
acce952b 3795 continue;
fd8b2b61 3796 wait_on_extent_buffer_writeback(eb);
acce952b 3797
fd8b2b61
JB		 3798 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3799 &eb->bflags))
3800 clear_extent_buffer_dirty(eb);
3801 free_extent_buffer_stale(eb);
acce952b 3802 }
3803 }
3804
3805 return ret;
3806}
3807
3808static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3809 struct extent_io_tree *pinned_extents)
3810{
3811 struct extent_io_tree *unpin;
3812 u64 start;
3813 u64 end;
3814 int ret;
ed0eaa14 3815 bool loop = true;
acce952b 3816
3817 unpin = pinned_extents;
ed0eaa14 3818again:
acce952b 3819 while (1) {
3820 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 3821 EXTENT_DIRTY, NULL);
acce952b 3822 if (ret)
3823 break;
3824
3825 /* opt_discard */
5378e607
LD		 3826 if (btrfs_test_opt(root, DISCARD))
3827 ret = btrfs_error_discard_extent(root, start,
3828 end + 1 - start,
3829 NULL);
acce952b 3830
3831 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3832 btrfs_error_unpin_extent_range(root, start, end);
3833 cond_resched();
3834 }
3835
ed0eaa14
LB		 3836 if (loop) {
3837 if (unpin == &root->fs_info->freed_extents[0])
3838 unpin = &root->fs_info->freed_extents[1];
3839 else
3840 unpin = &root->fs_info->freed_extents[0];
3841 loop = false;
3842 goto again;
3843 }
3844
acce952b 3845 return 0;
3846}
3847
49b25e05
JM		 3848void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3849 struct btrfs_root *root)
3850{
3851 btrfs_destroy_delayed_refs(cur_trans, root);
3852 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3853 cur_trans->dirty_pages.dirty_bytes);
3854
3855 /* FIXME: cleanup wait for commit */
3856 cur_trans->in_commit = 1;
3857 cur_trans->blocked = 1;
d7096fc3 3858 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 3859
aec8030a
MX		 3860 btrfs_evict_pending_snapshots(cur_trans);
3861
49b25e05 3862 cur_trans->blocked = 0;
d7096fc3 3863 wake_up(&root->fs_info->transaction_wait);
49b25e05
JM		 3864
3865 cur_trans->commit_done = 1;
d7096fc3 3866 wake_up(&cur_trans->commit_wait);
49b25e05 3867
67cde344
MX		 3868 btrfs_destroy_delayed_inodes(root);
3869 btrfs_assert_delayed_root_empty(root);
49b25e05 3870
49b25e05
JM		 3871 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3872 EXTENT_DIRTY);
6e841e32
LB		 3873 btrfs_destroy_pinned_extent(root,
3874 root->fs_info->pinned_extents);
49b25e05
JM		 3875
3876 /*
3877 memset(cur_trans, 0, sizeof(*cur_trans));
3878 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3879 */
3880}
3881
3882int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 3883{
3884 struct btrfs_transaction *t;
3885 LIST_HEAD(list);
3886
acce952b 3887 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3888
a4abeea4 3889 spin_lock(&root->fs_info->trans_lock);
acce952b 3890 list_splice_init(&root->fs_info->trans_list, &list);
a4abeea4
JB		 3891 root->fs_info->trans_no_join = 1;
3892 spin_unlock(&root->fs_info->trans_lock);
3893
acce952b 3894 while (!list_empty(&list)) {
3895 t = list_entry(list.next, struct btrfs_transaction, list);
acce952b 3896
569e0f35 3897 btrfs_destroy_ordered_operations(t, root);
acce952b 3898
3899 btrfs_destroy_ordered_extents(root);
3900
3901 btrfs_destroy_delayed_refs(t, root);
3902
acce952b 3903 /* FIXME: cleanup wait for commit */
3904 t->in_commit = 1;
3905 t->blocked = 1;
66657b31 3906 smp_mb();
acce952b 3907 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3908 wake_up(&root->fs_info->transaction_blocked_wait);
3909
aec8030a
MX		 3910 btrfs_evict_pending_snapshots(t);
3911
acce952b 3912 t->blocked = 0;
66657b31 3913 smp_mb();
acce952b 3914 if (waitqueue_active(&root->fs_info->transaction_wait))
3915 wake_up(&root->fs_info->transaction_wait);
acce952b 3916
acce952b 3917 t->commit_done = 1;
66657b31 3918 smp_mb();
acce952b 3919 if (waitqueue_active(&t->commit_wait))
3920 wake_up(&t->commit_wait);
acce952b 3921
67cde344
MX		 3922 btrfs_destroy_delayed_inodes(root);
3923 btrfs_assert_delayed_root_empty(root);
3924
acce952b 3925 btrfs_destroy_delalloc_inodes(root);
3926
a4abeea4 3927 spin_lock(&root->fs_info->trans_lock);
acce952b 3928 root->fs_info->running_transaction = NULL;
a4abeea4 3929 spin_unlock(&root->fs_info->trans_lock);
acce952b 3930
3931 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3932 EXTENT_DIRTY);
3933
3934 btrfs_destroy_pinned_extent(root,
3935 root->fs_info->pinned_extents);
3936
13c5a93e 3937 atomic_set(&t->use_count, 0);
acce952b 3938 list_del_init(&t->list);
3939 memset(t, 0, sizeof(*t));
3940 kmem_cache_free(btrfs_transaction_cachep, t);
3941 }
3942
a4abeea4
JB		 3943 spin_lock(&root->fs_info->trans_lock);
3944 root->fs_info->trans_no_join = 0;
3945 spin_unlock(&root->fs_info->trans_lock);
acce952b 3946 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 3947
3948 return 0;
3949}
3950
d1310b2e 3951static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 3952 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 3953 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 3954 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 3955 /* note we're sharing with inode.c for the merge bio hook */
3956 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 3957};
Linux 6.x block layer and io_uring tree(s)