projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: introduce uuid-tree-gen field
[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>
803b2f54 34#include <linux/semaphore.h>
7e75bf3f 35#include <asm/unaligned.h>
4b4e25f2 36#include "compat.h"
eb60ceac
CM		 37#include "ctree.h"
38#include "disk-io.h"
e089f05c 39#include "transaction.h"
0f7d52f4 40#include "btrfs_inode.h"
0b86a832 41#include "volumes.h"
db94535d 42#include "print-tree.h"
8b712842 43#include "async-thread.h"
925baedd 44#include "locking.h"
e02119d5 45#include "tree-log.h"
fa9c0d79 46#include "free-space-cache.h"
581bb050 47#include "inode-map.h"
21adbd5c 48#include "check-integrity.h"
606686ee 49#include "rcu-string.h"
8dabb742 50#include "dev-replace.h"
53b381b3 51#include "raid56.h"
eb60ceac 52
de0022b9
JB		 53#ifdef CONFIG_X86
54#include <asm/cpufeature.h>
55#endif
56
d1310b2e 57static struct extent_io_ops btree_extent_io_ops;
8b712842 58static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 59static void free_fs_root(struct btrfs_root *root);
fcd1f065 60static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 61 int read_only);
569e0f35
JB		 62static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
63 struct btrfs_root *root);
143bede5 64static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
acce952b 65static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
66 struct btrfs_root *root);
aec8030a 67static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
143bede5 68static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
acce952b 69static int btrfs_destroy_marked_extents(struct btrfs_root *root,
70 struct extent_io_tree *dirty_pages,
71 int mark);
72static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
73 struct extent_io_tree *pinned_extents);
48a3b636
ES		 74static int btrfs_cleanup_transaction(struct btrfs_root *root);
75static void btrfs_error_commit_super(struct btrfs_root *root);
ce9adaa5 76
d352ac68
CM		 77/*
78 * end_io_wq structs are used to do processing in task context when an IO is
79 * complete. This is used during reads to verify checksums, and it is used
80 * by writes to insert metadata for new file extents after IO is complete.
81 */
ce9adaa5
CM		 82struct end_io_wq {
83 struct bio *bio;
84 bio_end_io_t *end_io;
85 void *private;
86 struct btrfs_fs_info *info;
87 int error;
22c59948 88 int metadata;
ce9adaa5 89 struct list_head list;
8b712842 90 struct btrfs_work work;
ce9adaa5 91};
0da5468f 92
d352ac68
CM		 93/*
94 * async submit bios are used to offload expensive checksumming
95 * onto the worker threads. They checksum file and metadata bios
96 * just before they are sent down the IO stack.
97 */
44b8bd7e
CM		 98struct async_submit_bio {
99 struct inode *inode;
100 struct bio *bio;
101 struct list_head list;
4a69a410
CM		 102 extent_submit_bio_hook_t *submit_bio_start;
103 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 104 int rw;
105 int mirror_num;
c8b97818 106 unsigned long bio_flags;
eaf25d93
CM		 107 /*
108 * bio_offset is optional, can be used if the pages in the bio
109 * can't tell us where in the file the bio should go
110 */
111 u64 bio_offset;
8b712842 112 struct btrfs_work work;
79787eaa 113 int error;
44b8bd7e
CM		 114};
115
85d4e461
CM		 116/*
117 * Lockdep class keys for extent_buffer->lock's in this root. For a given
118 * eb, the lockdep key is determined by the btrfs_root it belongs to and
119 * the level the eb occupies in the tree.
120 *
121 * Different roots are used for different purposes and may nest inside each
122 * other and they require separate keysets. As lockdep keys should be
123 * static, assign keysets according to the purpose of the root as indicated
124 * by btrfs_root->objectid. This ensures that all special purpose roots
125 * have separate keysets.
4008c04a 126 *
85d4e461
CM		 127 * Lock-nesting across peer nodes is always done with the immediate parent
128 * node locked thus preventing deadlock. As lockdep doesn't know this, use
129 * subclass to avoid triggering lockdep warning in such cases.
4008c04a 130 *
85d4e461
CM		 131 * The key is set by the readpage_end_io_hook after the buffer has passed
132 * csum validation but before the pages are unlocked. It is also set by
133 * btrfs_init_new_buffer on freshly allocated blocks.
4008c04a 134 *
85d4e461
CM		 135 * We also add a check to make sure the highest level of the tree is the
136 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
137 * needs update as well.
4008c04a
CM		 138 */
139#ifdef CONFIG_DEBUG_LOCK_ALLOC
140# if BTRFS_MAX_LEVEL != 8
141# error
142# endif
85d4e461
CM		 143
144static struct btrfs_lockdep_keyset {
145 u64 id; /* root objectid */
146 const char *name_stem; /* lock name stem */
147 char names[BTRFS_MAX_LEVEL + 1][20];
148 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
149} btrfs_lockdep_keysets[] = {
150 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
151 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
152 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
153 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
154 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
155 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
60b62978 156 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
85d4e461
CM		 157 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
158 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
159 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
160 { .id = 0, .name_stem = "tree" },
4008c04a 161};
85d4e461
CM		 162
163void __init btrfs_init_lockdep(void)
164{
165 int i, j;
166
167 /* initialize lockdep class names */
168 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
169 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
170
171 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
172 snprintf(ks->names[j], sizeof(ks->names[j]),
173 "btrfs-%s-%02d", ks->name_stem, j);
174 }
175}
176
177void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
178 int level)
179{
180 struct btrfs_lockdep_keyset *ks;
181
182 BUG_ON(level >= ARRAY_SIZE(ks->keys));
183
184 /* find the matching keyset, id 0 is the default entry */
185 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
186 if (ks->id == objectid)
187 break;
188
189 lockdep_set_class_and_name(&eb->lock,
190 &ks->keys[level], ks->names[level]);
191}
192
4008c04a
CM		 193#endif
194
d352ac68
CM		 195/*
196 * extents on the btree inode are pretty simple, there's one extent
197 * that covers the entire device
198 */
b2950863 199static struct extent_map *btree_get_extent(struct inode *inode,
306e16ce 200 struct page *page, size_t pg_offset, u64 start, u64 len,
b2950863 201 int create)
7eccb903 202{
5f39d397
CM		 203 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
204 struct extent_map *em;
205 int ret;
206
890871be 207 read_lock(&em_tree->lock);
d1310b2e 208 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 209 if (em) {
210 em->bdev =
211 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 212 read_unlock(&em_tree->lock);
5f39d397 213 goto out;
a061fc8d 214 }
890871be 215 read_unlock(&em_tree->lock);
7b13b7b1 216
172ddd60 217 em = alloc_extent_map();
5f39d397
CM		 218 if (!em) {
219 em = ERR_PTR(-ENOMEM);
220 goto out;
221 }
222 em->start = 0;
0afbaf8c 223 em->len = (u64)-1;
c8b97818 224 em->block_len = (u64)-1;
5f39d397 225 em->block_start = 0;
a061fc8d 226 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 227
890871be 228 write_lock(&em_tree->lock);
09a2a8f9 229 ret = add_extent_mapping(em_tree, em, 0);
5f39d397
CM		 230 if (ret == -EEXIST) {
231 free_extent_map(em);
7b13b7b1 232 em = lookup_extent_mapping(em_tree, start, len);
b4f359ab 233 if (!em)
0433f20d 234 em = ERR_PTR(-EIO);
5f39d397 235 } else if (ret) {
7b13b7b1 236 free_extent_map(em);
0433f20d 237 em = ERR_PTR(ret);
5f39d397 238 }
890871be 239 write_unlock(&em_tree->lock);
7b13b7b1 240
5f39d397
CM		 241out:
242 return em;
7eccb903
CM		 243}
244
b0496686 245u32 btrfs_csum_data(char *data, u32 seed, size_t len)
19c00ddc 246{
163e783e 247 return crc32c(seed, data, len);
19c00ddc
CM		 248}
249
250void btrfs_csum_final(u32 crc, char *result)
251{
7e75bf3f 252 put_unaligned_le32(~crc, result);
19c00ddc
CM		 253}
254
d352ac68
CM		 255/*
256 * compute the csum for a btree block, and either verify it or write it
257 * into the csum field of the block.
258 */
19c00ddc
CM		 259static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
260 int verify)
261{
6c41761f 262 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
607d432d 263 char *result = NULL;
19c00ddc
CM		 264 unsigned long len;
265 unsigned long cur_len;
266 unsigned long offset = BTRFS_CSUM_SIZE;
19c00ddc
CM		 267 char *kaddr;
268 unsigned long map_start;
269 unsigned long map_len;
270 int err;
271 u32 crc = ~(u32)0;
607d432d 272 unsigned long inline_result;
19c00ddc
CM		 273
274 len = buf->len - offset;
d397712b 275 while (len > 0) {
19c00ddc 276 err = map_private_extent_buffer(buf, offset, 32,
a6591715 277 &kaddr, &map_start, &map_len);
d397712b 278 if (err)
19c00ddc 279 return 1;
19c00ddc 280 cur_len = min(len, map_len - (offset - map_start));
b0496686 281 crc = btrfs_csum_data(kaddr + offset - map_start,
19c00ddc
CM		 282 crc, cur_len);
283 len -= cur_len;
284 offset += cur_len;
19c00ddc 285 }
607d432d
JB		 286 if (csum_size > sizeof(inline_result)) {
287 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
288 if (!result)
289 return 1;
290 } else {
291 result = (char *)&inline_result;
292 }
293
19c00ddc
CM		 294 btrfs_csum_final(crc, result);
295
296 if (verify) {
607d432d 297 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 298 u32 val;
299 u32 found = 0;
607d432d 300 memcpy(&found, result, csum_size);
e4204ded 301
607d432d 302 read_extent_buffer(buf, &val, 0, csum_size);
7a36ddec 303 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
193f284d
CM		 304 "failed on %llu wanted %X found %X "
305 "level %d\n",
306 root->fs_info->sb->s_id,
307 (unsigned long long)buf->start, val, found,
308 btrfs_header_level(buf));
607d432d
JB		 309 if (result != (char *)&inline_result)
310 kfree(result);
19c00ddc
CM		 311 return 1;
312 }
313 } else {
607d432d 314 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 315 }
607d432d
JB		 316 if (result != (char *)&inline_result)
317 kfree(result);
19c00ddc
CM		 318 return 0;
319}
320
d352ac68
CM		 321/*
322 * we can't consider a given block up to date unless the transid of the
323 * block matches the transid in the parent node's pointer. This is how we
324 * detect blocks that either didn't get written at all or got written
325 * in the wrong place.
326 */
1259ab75 327static int verify_parent_transid(struct extent_io_tree *io_tree,
b9fab919
CM		 328 struct extent_buffer *eb, u64 parent_transid,
329 int atomic)
1259ab75 330{
2ac55d41 331 struct extent_state *cached_state = NULL;
1259ab75
CM		 332 int ret;
333
334 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
335 return 0;
336
b9fab919
CM		 337 if (atomic)
338 return -EAGAIN;
339
2ac55d41 340 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
d0082371 341 0, &cached_state);
0b32f4bb 342 if (extent_buffer_uptodate(eb) &&
1259ab75
CM		 343 btrfs_header_generation(eb) == parent_transid) {
344 ret = 0;
345 goto out;
346 }
7a36ddec 347 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
193f284d
CM		 348 "found %llu\n",
349 (unsigned long long)eb->start,
350 (unsigned long long)parent_transid,
351 (unsigned long long)btrfs_header_generation(eb));
1259ab75 352 ret = 1;
0b32f4bb 353 clear_extent_buffer_uptodate(eb);
33958dc6 354out:
2ac55d41
JB		 355 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
356 &cached_state, GFP_NOFS);
1259ab75 357 return ret;
1259ab75
CM		 358}
359
1104a885
DS		 360/*
361 * Return 0 if the superblock checksum type matches the checksum value of that
362 * algorithm. Pass the raw disk superblock data.
363 */
364static int btrfs_check_super_csum(char *raw_disk_sb)
365{
366 struct btrfs_super_block *disk_sb =
367 (struct btrfs_super_block *)raw_disk_sb;
368 u16 csum_type = btrfs_super_csum_type(disk_sb);
369 int ret = 0;
370
371 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
372 u32 crc = ~(u32)0;
373 const int csum_size = sizeof(crc);
374 char result[csum_size];
375
376 /*
377 * The super_block structure does not span the whole
378 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
379 * is filled with zeros and is included in the checkum.
380 */
381 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
382 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
383 btrfs_csum_final(crc, result);
384
385 if (memcmp(raw_disk_sb, result, csum_size))
386 ret = 1;
667e7d94
CM		 387
388 if (ret && btrfs_super_generation(disk_sb) < 10) {
389 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
390 ret = 0;
391 }
1104a885
DS		 392 }
393
394 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
395 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
396 csum_type);
397 ret = 1;
398 }
399
400 return ret;
401}
402
d352ac68
CM		 403/*
404 * helper to read a given tree block, doing retries as required when
405 * the checksums don't match and we have alternate mirrors to try.
406 */
f188591e
CM		 407static int btree_read_extent_buffer_pages(struct btrfs_root *root,
408 struct extent_buffer *eb,
ca7a79ad 409 u64 start, u64 parent_transid)
f188591e
CM		 410{
411 struct extent_io_tree *io_tree;
ea466794 412 int failed = 0;
f188591e
CM		 413 int ret;
414 int num_copies = 0;
415 int mirror_num = 0;
ea466794 416 int failed_mirror = 0;
f188591e 417
a826d6dc 418 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM		 419 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
420 while (1) {
bb82ab88
AJ		 421 ret = read_extent_buffer_pages(io_tree, eb, start,
422 WAIT_COMPLETE,
f188591e 423 btree_get_extent, mirror_num);
256dd1bb
SB		 424 if (!ret) {
425 if (!verify_parent_transid(io_tree, eb,
b9fab919 426 parent_transid, 0))
256dd1bb
SB		 427 break;
428 else
429 ret = -EIO;
430 }
d397712b 431
a826d6dc
JB		 432 /*
433 * This buffer's crc is fine, but its contents are corrupted, so
434 * there is no reason to read the other copies, they won't be
435 * any less wrong.
436 */
437 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
ea466794
JB		 438 break;
439
5d964051 440 num_copies = btrfs_num_copies(root->fs_info,
f188591e 441 eb->start, eb->len);
4235298e 442 if (num_copies == 1)
ea466794 443 break;
4235298e 444
5cf1ab56
JB		 445 if (!failed_mirror) {
446 failed = 1;
447 failed_mirror = eb->read_mirror;
448 }
449
f188591e 450 mirror_num++;
ea466794
JB		 451 if (mirror_num == failed_mirror)
452 mirror_num++;
453
4235298e 454 if (mirror_num > num_copies)
ea466794 455 break;
f188591e 456 }
ea466794 457
c0901581 458 if (failed && !ret && failed_mirror)
ea466794
JB		 459 repair_eb_io_failure(root, eb, failed_mirror);
460
461 return ret;
f188591e 462}
19c00ddc 463
d352ac68 464/*
d397712b
CM		 465 * checksum a dirty tree block before IO. This has extra checks to make sure
466 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 467 */
d397712b 468
b2950863 469static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 470{
d1310b2e 471 struct extent_io_tree *tree;
4eee4fa4 472 u64 start = page_offset(page);
19c00ddc 473 u64 found_start;
19c00ddc 474 struct extent_buffer *eb;
f188591e 475
d1310b2e 476 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc 477
4f2de97a
JB		 478 eb = (struct extent_buffer *)page->private;
479 if (page != eb->pages[0])
480 return 0;
19c00ddc
CM		 481 found_start = btrfs_header_bytenr(eb);
482 if (found_start != start) {
55c69072 483 WARN_ON(1);
4f2de97a 484 return 0;
55c69072 485 }
55c69072 486 if (!PageUptodate(page)) {
55c69072 487 WARN_ON(1);
4f2de97a 488 return 0;
19c00ddc 489 }
19c00ddc 490 csum_tree_block(root, eb, 0);
19c00ddc
CM		 491 return 0;
492}
493
2b82032c
YZ		 494static int check_tree_block_fsid(struct btrfs_root *root,
495 struct extent_buffer *eb)
496{
497 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
498 u8 fsid[BTRFS_UUID_SIZE];
499 int ret = 1;
500
501 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
502 BTRFS_FSID_SIZE);
503 while (fs_devices) {
504 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
505 ret = 0;
506 break;
507 }
508 fs_devices = fs_devices->seed;
509 }
510 return ret;
511}
512
a826d6dc
JB		 513#define CORRUPT(reason, eb, root, slot) \
514 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
515 "root=%llu, slot=%d\n", reason, \
516 (unsigned long long)btrfs_header_bytenr(eb), \
517 (unsigned long long)root->objectid, slot)
518
519static noinline int check_leaf(struct btrfs_root *root,
520 struct extent_buffer *leaf)
521{
522 struct btrfs_key key;
523 struct btrfs_key leaf_key;
524 u32 nritems = btrfs_header_nritems(leaf);
525 int slot;
526
527 if (nritems == 0)
528 return 0;
529
530 /* Check the 0 item */
531 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
532 BTRFS_LEAF_DATA_SIZE(root)) {
533 CORRUPT("invalid item offset size pair", leaf, root, 0);
534 return -EIO;
535 }
536
537 /*
538 * Check to make sure each items keys are in the correct order and their
539 * offsets make sense. We only have to loop through nritems-1 because
540 * we check the current slot against the next slot, which verifies the
541 * next slot's offset+size makes sense and that the current's slot
542 * offset is correct.
543 */
544 for (slot = 0; slot < nritems - 1; slot++) {
545 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
546 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
547
548 /* Make sure the keys are in the right order */
549 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
550 CORRUPT("bad key order", leaf, root, slot);
551 return -EIO;
552 }
553
554 /*
555 * Make sure the offset and ends are right, remember that the
556 * item data starts at the end of the leaf and grows towards the
557 * front.
558 */
559 if (btrfs_item_offset_nr(leaf, slot) !=
560 btrfs_item_end_nr(leaf, slot + 1)) {
561 CORRUPT("slot offset bad", leaf, root, slot);
562 return -EIO;
563 }
564
565 /*
566 * Check to make sure that we don't point outside of the leaf,
567 * just incase all the items are consistent to eachother, but
568 * all point outside of the leaf.
569 */
570 if (btrfs_item_end_nr(leaf, slot) >
571 BTRFS_LEAF_DATA_SIZE(root)) {
572 CORRUPT("slot end outside of leaf", leaf, root, slot);
573 return -EIO;
574 }
575 }
576
577 return 0;
578}
579
facc8a22
MX		 580static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
581 u64 phy_offset, struct page *page,
582 u64 start, u64 end, int mirror)
ce9adaa5
CM		 583{
584 struct extent_io_tree *tree;
585 u64 found_start;
586 int found_level;
ce9adaa5
CM		 587 struct extent_buffer *eb;
588 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 589 int ret = 0;
727011e0 590 int reads_done;
ce9adaa5 591
ce9adaa5
CM		 592 if (!page->private)
593 goto out;
d397712b 594
727011e0 595 tree = &BTRFS_I(page->mapping->host)->io_tree;
4f2de97a 596 eb = (struct extent_buffer *)page->private;
d397712b 597
0b32f4bb
JB		 598 /* the pending IO might have been the only thing that kept this buffer
599 * in memory. Make sure we have a ref for all this other checks
600 */
601 extent_buffer_get(eb);
602
603 reads_done = atomic_dec_and_test(&eb->io_pages);
727011e0
CM		 604 if (!reads_done)
605 goto err;
f188591e 606
5cf1ab56 607 eb->read_mirror = mirror;
ea466794
JB		 608 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
609 ret = -EIO;
610 goto err;
611 }
612
ce9adaa5 613 found_start = btrfs_header_bytenr(eb);
727011e0 614 if (found_start != eb->start) {
7a36ddec 615 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
193f284d
CM		 616 "%llu %llu\n",
617 (unsigned long long)found_start,
618 (unsigned long long)eb->start);
f188591e 619 ret = -EIO;
ce9adaa5
CM		 620 goto err;
621 }
2b82032c 622 if (check_tree_block_fsid(root, eb)) {
7a36ddec 623 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
193f284d 624 (unsigned long long)eb->start);
1259ab75
CM		 625 ret = -EIO;
626 goto err;
627 }
ce9adaa5 628 found_level = btrfs_header_level(eb);
1c24c3ce
JB		 629 if (found_level >= BTRFS_MAX_LEVEL) {
630 btrfs_info(root->fs_info, "bad tree block level %d\n",
631 (int)btrfs_header_level(eb));
632 ret = -EIO;
633 goto err;
634 }
ce9adaa5 635
85d4e461
CM		 636 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
637 eb, found_level);
4008c04a 638
ce9adaa5 639 ret = csum_tree_block(root, eb, 1);
a826d6dc 640 if (ret) {
f188591e 641 ret = -EIO;
a826d6dc
JB		 642 goto err;
643 }
644
645 /*
646 * If this is a leaf block and it is corrupt, set the corrupt bit so
647 * that we don't try and read the other copies of this block, just
648 * return -EIO.
649 */
650 if (found_level == 0 && check_leaf(root, eb)) {
651 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
652 ret = -EIO;
653 }
ce9adaa5 654
0b32f4bb
JB		 655 if (!ret)
656 set_extent_buffer_uptodate(eb);
ce9adaa5 657err:
79fb65a1
JB		 658 if (reads_done &&
659 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 660 btree_readahead_hook(root, eb, eb->start, ret);
4bb31e92 661
53b381b3
DW		 662 if (ret) {
663 /*
664 * our io error hook is going to dec the io pages
665 * again, we have to make sure it has something
666 * to decrement
667 */
668 atomic_inc(&eb->io_pages);
0b32f4bb 669 clear_extent_buffer_uptodate(eb);
53b381b3 670 }
0b32f4bb 671 free_extent_buffer(eb);
ce9adaa5 672out:
f188591e 673 return ret;
ce9adaa5
CM		 674}
675
ea466794 676static int btree_io_failed_hook(struct page *page, int failed_mirror)
4bb31e92 677{
4bb31e92
AJ		 678 struct extent_buffer *eb;
679 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
680
4f2de97a 681 eb = (struct extent_buffer *)page->private;
ea466794 682 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
5cf1ab56 683 eb->read_mirror = failed_mirror;
53b381b3 684 atomic_dec(&eb->io_pages);
ea466794 685 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 686 btree_readahead_hook(root, eb, eb->start, -EIO);
4bb31e92
AJ		 687 return -EIO; /* we fixed nothing */
688}
689
ce9adaa5 690static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 691{
692 struct end_io_wq *end_io_wq = bio->bi_private;
693 struct btrfs_fs_info *fs_info;
ce9adaa5 694
ce9adaa5 695 fs_info = end_io_wq->info;
ce9adaa5 696 end_io_wq->error = err;
8b712842
CM		 697 end_io_wq->work.func = end_workqueue_fn;
698 end_io_wq->work.flags = 0;
d20f7043 699
7b6d91da 700 if (bio->bi_rw & REQ_WRITE) {
53b381b3 701 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
cad321ad
CM		 702 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
703 &end_io_wq->work);
53b381b3 704 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
0cb59c99
JB		 705 btrfs_queue_worker(&fs_info->endio_freespace_worker,
706 &end_io_wq->work);
53b381b3
DW		 707 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
708 btrfs_queue_worker(&fs_info->endio_raid56_workers,
709 &end_io_wq->work);
cad321ad
CM		 710 else
711 btrfs_queue_worker(&fs_info->endio_write_workers,
712 &end_io_wq->work);
d20f7043 713 } else {
53b381b3
DW		 714 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
715 btrfs_queue_worker(&fs_info->endio_raid56_workers,
716 &end_io_wq->work);
717 else if (end_io_wq->metadata)
d20f7043
CM		 718 btrfs_queue_worker(&fs_info->endio_meta_workers,
719 &end_io_wq->work);
720 else
721 btrfs_queue_worker(&fs_info->endio_workers,
722 &end_io_wq->work);
723 }
ce9adaa5
CM		 724}
725
0cb59c99
JB		 726/*
727 * For the metadata arg you want
728 *
729 * 0 - if data
730 * 1 - if normal metadta
731 * 2 - if writing to the free space cache area
53b381b3 732 * 3 - raid parity work
0cb59c99 733 */
22c59948
CM		 734int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
735 int metadata)
0b86a832 736{
ce9adaa5 737 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 738 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
739 if (!end_io_wq)
740 return -ENOMEM;
741
742 end_io_wq->private = bio->bi_private;
743 end_io_wq->end_io = bio->bi_end_io;
22c59948 744 end_io_wq->info = info;
ce9adaa5
CM		 745 end_io_wq->error = 0;
746 end_io_wq->bio = bio;
22c59948 747 end_io_wq->metadata = metadata;
ce9adaa5
CM		 748
749 bio->bi_private = end_io_wq;
750 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 751 return 0;
752}
753
b64a2851 754unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 755{
4854ddd0
CM		 756 unsigned long limit = min_t(unsigned long,
757 info->workers.max_workers,
758 info->fs_devices->open_devices);
759 return 256 * limit;
760}
0986fe9e 761
4a69a410
CM		 762static void run_one_async_start(struct btrfs_work *work)
763{
4a69a410 764 struct async_submit_bio *async;
79787eaa 765 int ret;
4a69a410
CM		 766
767 async = container_of(work, struct async_submit_bio, work);
79787eaa
JM		 768 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
769 async->mirror_num, async->bio_flags,
770 async->bio_offset);
771 if (ret)
772 async->error = ret;
4a69a410
CM		 773}
774
775static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 776{
777 struct btrfs_fs_info *fs_info;
778 struct async_submit_bio *async;
4854ddd0 779 int limit;
8b712842
CM		 780
781 async = container_of(work, struct async_submit_bio, work);
782 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 783
b64a2851 784 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 785 limit = limit * 2 / 3;
786
66657b31 787 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
b64a2851 788 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 789 wake_up(&fs_info->async_submit_wait);
790
79787eaa
JM		 791 /* If an error occured we just want to clean up the bio and move on */
792 if (async->error) {
793 bio_endio(async->bio, async->error);
794 return;
795 }
796
4a69a410 797 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 798 async->mirror_num, async->bio_flags,
799 async->bio_offset);
4a69a410
CM		 800}
801
802static void run_one_async_free(struct btrfs_work *work)
803{
804 struct async_submit_bio *async;
805
806 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 807 kfree(async);
808}
809
44b8bd7e
CM		 810int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
811 int rw, struct bio *bio, int mirror_num,
c8b97818 812 unsigned long bio_flags,
eaf25d93 813 u64 bio_offset,
4a69a410
CM		 814 extent_submit_bio_hook_t *submit_bio_start,
815 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 816{
817 struct async_submit_bio *async;
818
819 async = kmalloc(sizeof(*async), GFP_NOFS);
820 if (!async)
821 return -ENOMEM;
822
823 async->inode = inode;
824 async->rw = rw;
825 async->bio = bio;
826 async->mirror_num = mirror_num;
4a69a410
CM		 827 async->submit_bio_start = submit_bio_start;
828 async->submit_bio_done = submit_bio_done;
829
830 async->work.func = run_one_async_start;
831 async->work.ordered_func = run_one_async_done;
832 async->work.ordered_free = run_one_async_free;
833
8b712842 834 async->work.flags = 0;
c8b97818 835 async->bio_flags = bio_flags;
eaf25d93 836 async->bio_offset = bio_offset;
8c8bee1d 837
79787eaa
JM		 838 async->error = 0;
839
cb03c743 840 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 841
7b6d91da 842 if (rw & REQ_SYNC)
d313d7a3
CM		 843 btrfs_set_work_high_prio(&async->work);
844
8b712842 845 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 846
d397712b 847 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 848 atomic_read(&fs_info->nr_async_submits)) {
849 wait_event(fs_info->async_submit_wait,
850 (atomic_read(&fs_info->nr_async_submits) == 0));
851 }
852
44b8bd7e
CM		 853 return 0;
854}
855
ce3ed71a
CM		 856static int btree_csum_one_bio(struct bio *bio)
857{
858 struct bio_vec *bvec = bio->bi_io_vec;
859 int bio_index = 0;
860 struct btrfs_root *root;
79787eaa 861 int ret = 0;
ce3ed71a
CM		 862
863 WARN_ON(bio->bi_vcnt <= 0);
d397712b 864 while (bio_index < bio->bi_vcnt) {
ce3ed71a 865 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 866 ret = csum_dirty_buffer(root, bvec->bv_page);
867 if (ret)
868 break;
ce3ed71a
CM		 869 bio_index++;
870 bvec++;
871 }
79787eaa 872 return ret;
ce3ed71a
CM		 873}
874
4a69a410
CM		 875static int __btree_submit_bio_start(struct inode *inode, int rw,
876 struct bio *bio, int mirror_num,
eaf25d93
CM		 877 unsigned long bio_flags,
878 u64 bio_offset)
22c59948 879{
8b712842
CM		 880 /*
881 * when we're called for a write, we're already in the async
5443be45 882 * submission context. Just jump into btrfs_map_bio
8b712842 883 */
79787eaa 884 return btree_csum_one_bio(bio);
4a69a410 885}
22c59948 886
4a69a410 887static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 888 int mirror_num, unsigned long bio_flags,
889 u64 bio_offset)
4a69a410 890{
61891923
SB		 891 int ret;
892
8b712842 893 /*
4a69a410
CM		 894 * when we're called for a write, we're already in the async
895 * submission context. Just jump into btrfs_map_bio
8b712842 896 */
61891923
SB		 897 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
898 if (ret)
899 bio_endio(bio, ret);
900 return ret;
0b86a832
CM		 901}
902
de0022b9
JB		 903static int check_async_write(struct inode *inode, unsigned long bio_flags)
904{
905 if (bio_flags & EXTENT_BIO_TREE_LOG)
906 return 0;
907#ifdef CONFIG_X86
908 if (cpu_has_xmm4_2)
909 return 0;
910#endif
911 return 1;
912}
913
44b8bd7e 914static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 915 int mirror_num, unsigned long bio_flags,
916 u64 bio_offset)
44b8bd7e 917{
de0022b9 918 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 919 int ret;
920
7b6d91da 921 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 922 /*
923 * called for a read, do the setup so that checksum validation
924 * can happen in the async kernel threads
925 */
f3f266ab
CM		 926 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
927 bio, 1);
1d4284bd 928 if (ret)
61891923
SB		 929 goto out_w_error;
930 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
931 mirror_num, 0);
de0022b9
JB		 932 } else if (!async) {
933 ret = btree_csum_one_bio(bio);
934 if (ret)
61891923
SB		 935 goto out_w_error;
936 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
937 mirror_num, 0);
938 } else {
939 /*
940 * kthread helpers are used to submit writes so that
941 * checksumming can happen in parallel across all CPUs
942 */
943 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
944 inode, rw, bio, mirror_num, 0,
945 bio_offset,
946 __btree_submit_bio_start,
947 __btree_submit_bio_done);
44b8bd7e 948 }
d313d7a3 949
61891923
SB		 950 if (ret) {
951out_w_error:
952 bio_endio(bio, ret);
953 }
954 return ret;
44b8bd7e
CM		 955}
956
3dd1462e 957#ifdef CONFIG_MIGRATION
784b4e29 958static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 959 struct page *newpage, struct page *page,
960 enum migrate_mode mode)
784b4e29
CM		 961{
962 /*
963 * we can't safely write a btree page from here,
964 * we haven't done the locking hook
965 */
966 if (PageDirty(page))
967 return -EAGAIN;
968 /*
969 * Buffers may be managed in a filesystem specific way.
970 * We must have no buffers or drop them.
971 */
972 if (page_has_private(page) &&
973 !try_to_release_page(page, GFP_KERNEL))
974 return -EAGAIN;
a6bc32b8 975 return migrate_page(mapping, newpage, page, mode);
784b4e29 976}
3dd1462e 977#endif
784b4e29 978
0da5468f
CM		 979
980static int btree_writepages(struct address_space *mapping,
981 struct writeback_control *wbc)
982{
d1310b2e 983 struct extent_io_tree *tree;
e2d84521
MX		 984 struct btrfs_fs_info *fs_info;
985 int ret;
986
d1310b2e 987 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 988 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 989
990 if (wbc->for_kupdate)
991 return 0;
992
e2d84521 993 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 994 /* this is a bit racy, but that's ok */
e2d84521
MX		 995 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
996 BTRFS_DIRTY_METADATA_THRESH);
997 if (ret < 0)
793955bc 998 return 0;
793955bc 999 }
0b32f4bb 1000 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 1001}
1002
b2950863 1003static int btree_readpage(struct file *file, struct page *page)
5f39d397 1004{
d1310b2e
CM		 1005 struct extent_io_tree *tree;
1006 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 1007 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 1008}
22b0ebda 1009
70dec807 1010static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 1011{
98509cfc 1012 if (PageWriteback(page) || PageDirty(page))
d397712b 1013 return 0;
0c4e538b 1014
f7a52a40 1015 return try_release_extent_buffer(page);
d98237b3
CM		 1016}
1017
d47992f8
LC		 1018static void btree_invalidatepage(struct page *page, unsigned int offset,
1019 unsigned int length)
d98237b3 1020{
d1310b2e
CM		 1021 struct extent_io_tree *tree;
1022 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1023 extent_invalidatepage(tree, page, offset);
1024 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1025 if (PagePrivate(page)) {
d397712b
CM		 1026 printk(KERN_WARNING "btrfs warning page private not zero "
1027 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1028 ClearPagePrivate(page);
1029 set_page_private(page, 0);
1030 page_cache_release(page);
1031 }
d98237b3
CM		 1032}
1033
0b32f4bb
JB		 1034static int btree_set_page_dirty(struct page *page)
1035{
bb146eb2 1036#ifdef DEBUG
0b32f4bb
JB		 1037 struct extent_buffer *eb;
1038
1039 BUG_ON(!PagePrivate(page));
1040 eb = (struct extent_buffer *)page->private;
1041 BUG_ON(!eb);
1042 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1043 BUG_ON(!atomic_read(&eb->refs));
1044 btrfs_assert_tree_locked(eb);
bb146eb2 1045#endif
0b32f4bb
JB		 1046 return __set_page_dirty_nobuffers(page);
1047}
1048
7f09410b 1049static const struct address_space_operations btree_aops = {
d98237b3 1050 .readpage = btree_readpage,
0da5468f 1051 .writepages = btree_writepages,
5f39d397
CM		 1052 .releasepage = btree_releasepage,
1053 .invalidatepage = btree_invalidatepage,
5a92bc88 1054#ifdef CONFIG_MIGRATION
784b4e29 1055 .migratepage = btree_migratepage,
5a92bc88 1056#endif
0b32f4bb 1057 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1058};
1059
ca7a79ad
CM		 1060int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1061 u64 parent_transid)
090d1875 1062{
5f39d397
CM		 1063 struct extent_buffer *buf = NULL;
1064 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1065 int ret = 0;
090d1875 1066
db94535d 1067 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1068 if (!buf)
090d1875 1069 return 0;
d1310b2e 1070 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1071 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1072 free_extent_buffer(buf);
de428b63 1073 return ret;
090d1875
CM		 1074}
1075
ab0fff03
AJ		 1076int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1077 int mirror_num, struct extent_buffer **eb)
1078{
1079 struct extent_buffer *buf = NULL;
1080 struct inode *btree_inode = root->fs_info->btree_inode;
1081 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1082 int ret;
1083
1084 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1085 if (!buf)
1086 return 0;
1087
1088 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1089
1090 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1091 btree_get_extent, mirror_num);
1092 if (ret) {
1093 free_extent_buffer(buf);
1094 return ret;
1095 }
1096
1097 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1098 free_extent_buffer(buf);
1099 return -EIO;
0b32f4bb 1100 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1101 *eb = buf;
1102 } else {
1103 free_extent_buffer(buf);
1104 }
1105 return 0;
1106}
1107
0999df54
CM		 1108struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1109 u64 bytenr, u32 blocksize)
1110{
1111 struct inode *btree_inode = root->fs_info->btree_inode;
1112 struct extent_buffer *eb;
1113 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
f09d1f60 1114 bytenr, blocksize);
0999df54
CM		 1115 return eb;
1116}
1117
1118struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1119 u64 bytenr, u32 blocksize)
1120{
1121 struct inode *btree_inode = root->fs_info->btree_inode;
1122 struct extent_buffer *eb;
1123
1124 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
727011e0 1125 bytenr, blocksize);
0999df54
CM		 1126 return eb;
1127}
1128
1129
e02119d5
CM		 1130int btrfs_write_tree_block(struct extent_buffer *buf)
1131{
727011e0 1132 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
8aa38c31 1133 buf->start + buf->len - 1);
e02119d5
CM		 1134}
1135
1136int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1137{
727011e0 1138 return filemap_fdatawait_range(buf->pages[0]->mapping,
8aa38c31 1139 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1140}
1141
0999df54 1142struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1143 u32 blocksize, u64 parent_transid)
0999df54
CM		 1144{
1145 struct extent_buffer *buf = NULL;
0999df54
CM		 1146 int ret;
1147
0999df54
CM		 1148 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1149 if (!buf)
1150 return NULL;
0999df54 1151
ca7a79ad 1152 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
0f0fe8f7
FDBM		 1153 if (ret) {
1154 free_extent_buffer(buf);
1155 return NULL;
1156 }
5f39d397 1157 return buf;
ce9adaa5 1158
eb60ceac
CM		 1159}
1160
d5c13f92
JM		 1161void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1162 struct extent_buffer *buf)
ed2ff2cb 1163{
e2d84521
MX		 1164 struct btrfs_fs_info *fs_info = root->fs_info;
1165
55c69072 1166 if (btrfs_header_generation(buf) ==
e2d84521 1167 fs_info->running_transaction->transid) {
b9447ef8 1168 btrfs_assert_tree_locked(buf);
b4ce94de 1169
b9473439 1170 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1171 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1172 -buf->len,
1173 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1174 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1175 btrfs_set_lock_blocking(buf);
1176 clear_extent_buffer_dirty(buf);
1177 }
925baedd 1178 }
5f39d397
CM		 1179}
1180
143bede5
JM		 1181static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1182 u32 stripesize, struct btrfs_root *root,
1183 struct btrfs_fs_info *fs_info,
1184 u64 objectid)
d97e63b6 1185{
cfaa7295 1186 root->node = NULL;
a28ec197 1187 root->commit_root = NULL;
db94535d
CM		 1188 root->sectorsize = sectorsize;
1189 root->nodesize = nodesize;
1190 root->leafsize = leafsize;
87ee04eb 1191 root->stripesize = stripesize;
123abc88 1192 root->ref_cows = 0;
0b86a832 1193 root->track_dirty = 0;
c71bf099 1194 root->in_radix = 0;
d68fc57b
YZ		 1195 root->orphan_item_inserted = 0;
1196 root->orphan_cleanup_state = 0;
0b86a832 1197
0f7d52f4
CM		 1198 root->objectid = objectid;
1199 root->last_trans = 0;
13a8a7c8 1200 root->highest_objectid = 0;
eb73c1b7 1201 root->nr_delalloc_inodes = 0;
199c2a9c 1202 root->nr_ordered_extents = 0;
58176a96 1203 root->name = NULL;
6bef4d31 1204 root->inode_tree = RB_ROOT;
16cdcec7 1205 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1206 root->block_rsv = NULL;
d68fc57b 1207 root->orphan_block_rsv = NULL;
0b86a832
CM		 1208
1209 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1210 INIT_LIST_HEAD(&root->root_list);
eb73c1b7
MX		 1211 INIT_LIST_HEAD(&root->delalloc_inodes);
1212 INIT_LIST_HEAD(&root->delalloc_root);
199c2a9c
MX		 1213 INIT_LIST_HEAD(&root->ordered_extents);
1214 INIT_LIST_HEAD(&root->ordered_root);
2ab28f32
JB		 1215 INIT_LIST_HEAD(&root->logged_list[0]);
1216 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1217 spin_lock_init(&root->orphan_lock);
5d4f98a2 1218 spin_lock_init(&root->inode_lock);
eb73c1b7 1219 spin_lock_init(&root->delalloc_lock);
199c2a9c 1220 spin_lock_init(&root->ordered_extent_lock);
f0486c68 1221 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1222 spin_lock_init(&root->log_extents_lock[0]);
1223 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1224 mutex_init(&root->objectid_mutex);
e02119d5 1225 mutex_init(&root->log_mutex);
7237f183
YZ		 1226 init_waitqueue_head(&root->log_writer_wait);
1227 init_waitqueue_head(&root->log_commit_wait[0]);
1228 init_waitqueue_head(&root->log_commit_wait[1]);
1229 atomic_set(&root->log_commit[0], 0);
1230 atomic_set(&root->log_commit[1], 0);
1231 atomic_set(&root->log_writers, 0);
2ecb7923 1232 atomic_set(&root->log_batch, 0);
8a35d95f 1233 atomic_set(&root->orphan_inodes, 0);
b0feb9d9 1234 atomic_set(&root->refs, 1);
7237f183 1235 root->log_transid = 0;
257c62e1 1236 root->last_log_commit = 0;
d0c803c4 1237 extent_io_tree_init(&root->dirty_log_pages,
f993c883 1238 fs_info->btree_inode->i_mapping);
017e5369 1239
3768f368
CM		 1240 memset(&root->root_key, 0, sizeof(root->root_key));
1241 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1242 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1243 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 1244 root->defrag_trans_start = fs_info->generation;
58176a96 1245 init_completion(&root->kobj_unregister);
6702ed49 1246 root->defrag_running = 0;
4d775673 1247 root->root_key.objectid = objectid;
0ee5dc67 1248 root->anon_dev = 0;
8ea05e3a 1249
5f3ab90a 1250 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1251}
1252
f84a8bd6 1253static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1254{
1255 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1256 if (root)
1257 root->fs_info = fs_info;
1258 return root;
1259}
1260
20897f5c
AJ		 1261struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1262 struct btrfs_fs_info *fs_info,
1263 u64 objectid)
1264{
1265 struct extent_buffer *leaf;
1266 struct btrfs_root *tree_root = fs_info->tree_root;
1267 struct btrfs_root *root;
1268 struct btrfs_key key;
1269 int ret = 0;
1270 u64 bytenr;
6463fe58 1271 uuid_le uuid;
20897f5c
AJ		 1272
1273 root = btrfs_alloc_root(fs_info);
1274 if (!root)
1275 return ERR_PTR(-ENOMEM);
1276
1277 __setup_root(tree_root->nodesize, tree_root->leafsize,
1278 tree_root->sectorsize, tree_root->stripesize,
1279 root, fs_info, objectid);
1280 root->root_key.objectid = objectid;
1281 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1282 root->root_key.offset = 0;
1283
1284 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1285 0, objectid, NULL, 0, 0, 0);
1286 if (IS_ERR(leaf)) {
1287 ret = PTR_ERR(leaf);
1dd05682 1288 leaf = NULL;
20897f5c
AJ		 1289 goto fail;
1290 }
1291
1292 bytenr = leaf->start;
1293 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1294 btrfs_set_header_bytenr(leaf, leaf->start);
1295 btrfs_set_header_generation(leaf, trans->transid);
1296 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1297 btrfs_set_header_owner(leaf, objectid);
1298 root->node = leaf;
1299
1300 write_extent_buffer(leaf, fs_info->fsid,
1301 (unsigned long)btrfs_header_fsid(leaf),
1302 BTRFS_FSID_SIZE);
1303 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1304 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1305 BTRFS_UUID_SIZE);
1306 btrfs_mark_buffer_dirty(leaf);
1307
1308 root->commit_root = btrfs_root_node(root);
1309 root->track_dirty = 1;
1310
1311
1312 root->root_item.flags = 0;
1313 root->root_item.byte_limit = 0;
1314 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1315 btrfs_set_root_generation(&root->root_item, trans->transid);
1316 btrfs_set_root_level(&root->root_item, 0);
1317 btrfs_set_root_refs(&root->root_item, 1);
1318 btrfs_set_root_used(&root->root_item, leaf->len);
1319 btrfs_set_root_last_snapshot(&root->root_item, 0);
1320 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1321 uuid_le_gen(&uuid);
1322 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1323 root->root_item.drop_level = 0;
1324
1325 key.objectid = objectid;
1326 key.type = BTRFS_ROOT_ITEM_KEY;
1327 key.offset = 0;
1328 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1329 if (ret)
1330 goto fail;
1331
1332 btrfs_tree_unlock(leaf);
1333
1dd05682
TI		 1334 return root;
1335
20897f5c 1336fail:
1dd05682
TI		 1337 if (leaf) {
1338 btrfs_tree_unlock(leaf);
1339 free_extent_buffer(leaf);
1340 }
1341 kfree(root);
20897f5c 1342
1dd05682 1343 return ERR_PTR(ret);
20897f5c
AJ		 1344}
1345
7237f183
YZ		 1346static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1347 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1348{
1349 struct btrfs_root *root;
1350 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1351 struct extent_buffer *leaf;
e02119d5 1352
6f07e42e 1353 root = btrfs_alloc_root(fs_info);
e02119d5 1354 if (!root)
7237f183 1355 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1356
1357 __setup_root(tree_root->nodesize, tree_root->leafsize,
1358 tree_root->sectorsize, tree_root->stripesize,
1359 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1360
1361 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1362 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1363 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1364 /*
1365 * log trees do not get reference counted because they go away
1366 * before a real commit is actually done. They do store pointers
1367 * to file data extents, and those reference counts still get
1368 * updated (along with back refs to the log tree).
1369 */
e02119d5
CM		 1370 root->ref_cows = 0;
1371
5d4f98a2 1372 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1373 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1374 0, 0, 0);
7237f183
YZ		 1375 if (IS_ERR(leaf)) {
1376 kfree(root);
1377 return ERR_CAST(leaf);
1378 }
e02119d5 1379
5d4f98a2
YZ		 1380 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1381 btrfs_set_header_bytenr(leaf, leaf->start);
1382 btrfs_set_header_generation(leaf, trans->transid);
1383 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1384 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1385 root->node = leaf;
e02119d5
CM		 1386
1387 write_extent_buffer(root->node, root->fs_info->fsid,
1388 (unsigned long)btrfs_header_fsid(root->node),
1389 BTRFS_FSID_SIZE);
1390 btrfs_mark_buffer_dirty(root->node);
1391 btrfs_tree_unlock(root->node);
7237f183
YZ		 1392 return root;
1393}
1394
1395int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1396 struct btrfs_fs_info *fs_info)
1397{
1398 struct btrfs_root *log_root;
1399
1400 log_root = alloc_log_tree(trans, fs_info);
1401 if (IS_ERR(log_root))
1402 return PTR_ERR(log_root);
1403 WARN_ON(fs_info->log_root_tree);
1404 fs_info->log_root_tree = log_root;
1405 return 0;
1406}
1407
1408int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1409 struct btrfs_root *root)
1410{
1411 struct btrfs_root *log_root;
1412 struct btrfs_inode_item *inode_item;
1413
1414 log_root = alloc_log_tree(trans, root->fs_info);
1415 if (IS_ERR(log_root))
1416 return PTR_ERR(log_root);
1417
1418 log_root->last_trans = trans->transid;
1419 log_root->root_key.offset = root->root_key.objectid;
1420
1421 inode_item = &log_root->root_item.inode;
3cae210f
QW		 1422 btrfs_set_stack_inode_generation(inode_item, 1);
1423 btrfs_set_stack_inode_size(inode_item, 3);
1424 btrfs_set_stack_inode_nlink(inode_item, 1);
1425 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1426 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
7237f183 1427
5d4f98a2 1428 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1429
1430 WARN_ON(root->log_root);
1431 root->log_root = log_root;
1432 root->log_transid = 0;
257c62e1 1433 root->last_log_commit = 0;
e02119d5
CM		 1434 return 0;
1435}
1436
35a3621b
SB		 1437static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1438 struct btrfs_key *key)
e02119d5
CM		 1439{
1440 struct btrfs_root *root;
1441 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1442 struct btrfs_path *path;
84234f3a 1443 u64 generation;
db94535d 1444 u32 blocksize;
cb517eab 1445 int ret;
0f7d52f4 1446
cb517eab
MX		 1447 path = btrfs_alloc_path();
1448 if (!path)
0f7d52f4 1449 return ERR_PTR(-ENOMEM);
cb517eab
MX		 1450
1451 root = btrfs_alloc_root(fs_info);
1452 if (!root) {
1453 ret = -ENOMEM;
1454 goto alloc_fail;
0f7d52f4
CM		 1455 }
1456
db94535d 1457 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb 1458 tree_root->sectorsize, tree_root->stripesize,
cb517eab 1459 root, fs_info, key->objectid);
0f7d52f4 1460
cb517eab
MX		 1461 ret = btrfs_find_root(tree_root, key, path,
1462 &root->root_item, &root->root_key);
0f7d52f4 1463 if (ret) {
13a8a7c8
YZ		 1464 if (ret > 0)
1465 ret = -ENOENT;
cb517eab 1466 goto find_fail;
0f7d52f4 1467 }
13a8a7c8 1468
84234f3a 1469 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1470 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1471 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1472 blocksize, generation);
cb517eab
MX		 1473 if (!root->node) {
1474 ret = -ENOMEM;
1475 goto find_fail;
1476 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1477 ret = -EIO;
1478 goto read_fail;
416bc658 1479 }
5d4f98a2 1480 root->commit_root = btrfs_root_node(root);
13a8a7c8 1481out:
cb517eab
MX		 1482 btrfs_free_path(path);
1483 return root;
1484
1485read_fail:
1486 free_extent_buffer(root->node);
1487find_fail:
1488 kfree(root);
1489alloc_fail:
1490 root = ERR_PTR(ret);
1491 goto out;
1492}
1493
1494struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1495 struct btrfs_key *location)
1496{
1497 struct btrfs_root *root;
1498
1499 root = btrfs_read_tree_root(tree_root, location);
1500 if (IS_ERR(root))
1501 return root;
1502
1503 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1504 root->ref_cows = 1;
08fe4db1
LZ		 1505 btrfs_check_and_init_root_item(&root->root_item);
1506 }
13a8a7c8 1507
5eda7b5e
CM		 1508 return root;
1509}
1510
cb517eab
MX		 1511int btrfs_init_fs_root(struct btrfs_root *root)
1512{
1513 int ret;
1514
1515 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1516 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1517 GFP_NOFS);
1518 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1519 ret = -ENOMEM;
1520 goto fail;
1521 }
1522
1523 btrfs_init_free_ino_ctl(root);
1524 mutex_init(&root->fs_commit_mutex);
1525 spin_lock_init(&root->cache_lock);
1526 init_waitqueue_head(&root->cache_wait);
1527
1528 ret = get_anon_bdev(&root->anon_dev);
1529 if (ret)
1530 goto fail;
1531 return 0;
1532fail:
1533 kfree(root->free_ino_ctl);
1534 kfree(root->free_ino_pinned);
1535 return ret;
1536}
1537
171170c1
ST		 1538static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1539 u64 root_id)
cb517eab
MX		 1540{
1541 struct btrfs_root *root;
1542
1543 spin_lock(&fs_info->fs_roots_radix_lock);
1544 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1545 (unsigned long)root_id);
1546 spin_unlock(&fs_info->fs_roots_radix_lock);
1547 return root;
1548}
1549
1550int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1551 struct btrfs_root *root)
1552{
1553 int ret;
1554
1555 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1556 if (ret)
1557 return ret;
1558
1559 spin_lock(&fs_info->fs_roots_radix_lock);
1560 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1561 (unsigned long)root->root_key.objectid,
1562 root);
1563 if (ret == 0)
1564 root->in_radix = 1;
1565 spin_unlock(&fs_info->fs_roots_radix_lock);
1566 radix_tree_preload_end();
1567
1568 return ret;
1569}
1570
edbd8d4e
CM		 1571struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1572 struct btrfs_key *location)
5eda7b5e
CM		 1573{
1574 struct btrfs_root *root;
1575 int ret;
1576
edbd8d4e
CM		 1577 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1578 return fs_info->tree_root;
1579 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1580 return fs_info->extent_root;
8f18cf13
CM		 1581 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1582 return fs_info->chunk_root;
1583 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1584 return fs_info->dev_root;
0403e47e
YZ		 1585 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1586 return fs_info->csum_root;
bcef60f2
AJ		 1587 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1588 return fs_info->quota_root ? fs_info->quota_root :
1589 ERR_PTR(-ENOENT);
f7a81ea4
SB		 1590 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1591 return fs_info->uuid_root ? fs_info->uuid_root :
1592 ERR_PTR(-ENOENT);
4df27c4d 1593again:
cb517eab 1594 root = btrfs_lookup_fs_root(fs_info, location->objectid);
5eda7b5e
CM		 1595 if (root)
1596 return root;
1597
cb517eab 1598 root = btrfs_read_fs_root(fs_info->tree_root, location);
5eda7b5e
CM		 1599 if (IS_ERR(root))
1600 return root;
3394e160 1601
cb517eab
MX		 1602 if (btrfs_root_refs(&root->root_item) == 0) {
1603 ret = -ENOENT;
581bb050 1604 goto fail;
35a30d7c 1605 }
581bb050 1606
cb517eab 1607 ret = btrfs_init_fs_root(root);
ac08aedf
CM		 1608 if (ret)
1609 goto fail;
3394e160 1610
d68fc57b
YZ		 1611 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1612 if (ret < 0)
1613 goto fail;
1614 if (ret == 0)
1615 root->orphan_item_inserted = 1;
1616
cb517eab 1617 ret = btrfs_insert_fs_root(fs_info, root);
0f7d52f4 1618 if (ret) {
4df27c4d
YZ		 1619 if (ret == -EEXIST) {
1620 free_fs_root(root);
1621 goto again;
1622 }
1623 goto fail;
0f7d52f4 1624 }
edbd8d4e 1625 return root;
4df27c4d
YZ		 1626fail:
1627 free_fs_root(root);
1628 return ERR_PTR(ret);
edbd8d4e
CM		 1629}
1630
04160088
CM		 1631static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1632{
1633 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1634 int ret = 0;
04160088
CM		 1635 struct btrfs_device *device;
1636 struct backing_dev_info *bdi;
b7967db7 1637
1f78160c
XG		 1638 rcu_read_lock();
1639 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1640 if (!device->bdev)
1641 continue;
04160088
CM		 1642 bdi = blk_get_backing_dev_info(device->bdev);
1643 if (bdi && bdi_congested(bdi, bdi_bits)) {
1644 ret = 1;
1645 break;
1646 }
1647 }
1f78160c 1648 rcu_read_unlock();
04160088
CM		 1649 return ret;
1650}
1651
ad081f14
JA		 1652/*
1653 * If this fails, caller must call bdi_destroy() to get rid of the
1654 * bdi again.
1655 */
04160088
CM		 1656static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1657{
ad081f14
JA		 1658 int err;
1659
1660 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1661 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1662 if (err)
1663 return err;
1664
4575c9cc 1665 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1666 bdi->congested_fn = btrfs_congested_fn;
1667 bdi->congested_data = info;
1668 return 0;
1669}
1670
8b712842
CM		 1671/*
1672 * called by the kthread helper functions to finally call the bio end_io
1673 * functions. This is where read checksum verification actually happens
1674 */
1675static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1676{
ce9adaa5 1677 struct bio *bio;
8b712842
CM		 1678 struct end_io_wq *end_io_wq;
1679 struct btrfs_fs_info *fs_info;
ce9adaa5 1680 int error;
ce9adaa5 1681
8b712842
CM		 1682 end_io_wq = container_of(work, struct end_io_wq, work);
1683 bio = end_io_wq->bio;
1684 fs_info = end_io_wq->info;
ce9adaa5 1685
8b712842
CM		 1686 error = end_io_wq->error;
1687 bio->bi_private = end_io_wq->private;
1688 bio->bi_end_io = end_io_wq->end_io;
1689 kfree(end_io_wq);
8b712842 1690 bio_endio(bio, error);
44b8bd7e
CM		 1691}
1692
a74a4b97
CM		 1693static int cleaner_kthread(void *arg)
1694{
1695 struct btrfs_root *root = arg;
d0278245 1696 int again;
a74a4b97
CM		 1697
1698 do {
d0278245 1699 again = 0;
a74a4b97 1700
d0278245 1701 /* Make the cleaner go to sleep early. */
babbf170 1702 if (btrfs_need_cleaner_sleep(root))
d0278245
MX		 1703 goto sleep;
1704
1705 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1706 goto sleep;
1707
dc7f370c
MX		 1708 /*
1709 * Avoid the problem that we change the status of the fs
1710 * during the above check and trylock.
1711 */
babbf170 1712 if (btrfs_need_cleaner_sleep(root)) {
dc7f370c
MX		 1713 mutex_unlock(&root->fs_info->cleaner_mutex);
1714 goto sleep;
76dda93c 1715 }
a74a4b97 1716
d0278245
MX		 1717 btrfs_run_delayed_iputs(root);
1718 again = btrfs_clean_one_deleted_snapshot(root);
1719 mutex_unlock(&root->fs_info->cleaner_mutex);
1720
1721 /*
05323cd1
MX		 1722 * The defragger has dealt with the R/O remount and umount,
1723 * needn't do anything special here.
d0278245
MX		 1724 */
1725 btrfs_run_defrag_inodes(root->fs_info);
1726sleep:
9d1a2a3a 1727 if (!try_to_freeze() && !again) {
a74a4b97 1728 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1729 if (!kthread_should_stop())
1730 schedule();
a74a4b97
CM		 1731 __set_current_state(TASK_RUNNING);
1732 }
1733 } while (!kthread_should_stop());
1734 return 0;
1735}
1736
1737static int transaction_kthread(void *arg)
1738{
1739 struct btrfs_root *root = arg;
1740 struct btrfs_trans_handle *trans;
1741 struct btrfs_transaction *cur;
8929ecfa 1742 u64 transid;
a74a4b97
CM		 1743 unsigned long now;
1744 unsigned long delay;
914b2007 1745 bool cannot_commit;
a74a4b97
CM		 1746
1747 do {
914b2007 1748 cannot_commit = false;
8b87dc17 1749 delay = HZ * root->fs_info->commit_interval;
a74a4b97
CM		 1750 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1751
a4abeea4 1752 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1753 cur = root->fs_info->running_transaction;
1754 if (!cur) {
a4abeea4 1755 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1756 goto sleep;
1757 }
31153d81 1758
a74a4b97 1759 now = get_seconds();
4a9d8bde 1760 if (cur->state < TRANS_STATE_BLOCKED &&
8b87dc17
DS		 1761 (now < cur->start_time ||
1762 now - cur->start_time < root->fs_info->commit_interval)) {
a4abeea4 1763 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1764 delay = HZ * 5;
1765 goto sleep;
1766 }
8929ecfa 1767 transid = cur->transid;
a4abeea4 1768 spin_unlock(&root->fs_info->trans_lock);
56bec294 1769
79787eaa 1770 /* If the file system is aborted, this will always fail. */
354aa0fb 1771 trans = btrfs_attach_transaction(root);
914b2007 1772 if (IS_ERR(trans)) {
354aa0fb
MX		 1773 if (PTR_ERR(trans) != -ENOENT)
1774 cannot_commit = true;
79787eaa 1775 goto sleep;
914b2007 1776 }
8929ecfa 1777 if (transid == trans->transid) {
79787eaa 1778 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1779 } else {
1780 btrfs_end_transaction(trans, root);
1781 }
a74a4b97
CM		 1782sleep:
1783 wake_up_process(root->fs_info->cleaner_kthread);
1784 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1785
a0acae0e 1786 if (!try_to_freeze()) {
a74a4b97 1787 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1788 if (!kthread_should_stop() &&
914b2007
JK		 1789 (!btrfs_transaction_blocked(root->fs_info) ||
1790 cannot_commit))
8929ecfa 1791 schedule_timeout(delay);
a74a4b97
CM		 1792 __set_current_state(TASK_RUNNING);
1793 }
1794 } while (!kthread_should_stop());
1795 return 0;
1796}
1797
af31f5e5
CM		 1798/*
1799 * this will find the highest generation in the array of
1800 * root backups. The index of the highest array is returned,
1801 * or -1 if we can't find anything.
1802 *
1803 * We check to make sure the array is valid by comparing the
1804 * generation of the latest root in the array with the generation
1805 * in the super block. If they don't match we pitch it.
1806 */
1807static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1808{
1809 u64 cur;
1810 int newest_index = -1;
1811 struct btrfs_root_backup *root_backup;
1812 int i;
1813
1814 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1815 root_backup = info->super_copy->super_roots + i;
1816 cur = btrfs_backup_tree_root_gen(root_backup);
1817 if (cur == newest_gen)
1818 newest_index = i;
1819 }
1820
1821 /* check to see if we actually wrapped around */
1822 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1823 root_backup = info->super_copy->super_roots;
1824 cur = btrfs_backup_tree_root_gen(root_backup);
1825 if (cur == newest_gen)
1826 newest_index = 0;
1827 }
1828 return newest_index;
1829}
1830
1831
1832/*
1833 * find the oldest backup so we know where to store new entries
1834 * in the backup array. This will set the backup_root_index
1835 * field in the fs_info struct
1836 */
1837static void find_oldest_super_backup(struct btrfs_fs_info *info,
1838 u64 newest_gen)
1839{
1840 int newest_index = -1;
1841
1842 newest_index = find_newest_super_backup(info, newest_gen);
1843 /* if there was garbage in there, just move along */
1844 if (newest_index == -1) {
1845 info->backup_root_index = 0;
1846 } else {
1847 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1848 }
1849}
1850
1851/*
1852 * copy all the root pointers into the super backup array.
1853 * this will bump the backup pointer by one when it is
1854 * done
1855 */
1856static void backup_super_roots(struct btrfs_fs_info *info)
1857{
1858 int next_backup;
1859 struct btrfs_root_backup *root_backup;
1860 int last_backup;
1861
1862 next_backup = info->backup_root_index;
1863 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1864 BTRFS_NUM_BACKUP_ROOTS;
1865
1866 /*
1867 * just overwrite the last backup if we're at the same generation
1868 * this happens only at umount
1869 */
1870 root_backup = info->super_for_commit->super_roots + last_backup;
1871 if (btrfs_backup_tree_root_gen(root_backup) ==
1872 btrfs_header_generation(info->tree_root->node))
1873 next_backup = last_backup;
1874
1875 root_backup = info->super_for_commit->super_roots + next_backup;
1876
1877 /*
1878 * make sure all of our padding and empty slots get zero filled
1879 * regardless of which ones we use today
1880 */
1881 memset(root_backup, 0, sizeof(*root_backup));
1882
1883 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1884
1885 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1886 btrfs_set_backup_tree_root_gen(root_backup,
1887 btrfs_header_generation(info->tree_root->node));
1888
1889 btrfs_set_backup_tree_root_level(root_backup,
1890 btrfs_header_level(info->tree_root->node));
1891
1892 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1893 btrfs_set_backup_chunk_root_gen(root_backup,
1894 btrfs_header_generation(info->chunk_root->node));
1895 btrfs_set_backup_chunk_root_level(root_backup,
1896 btrfs_header_level(info->chunk_root->node));
1897
1898 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1899 btrfs_set_backup_extent_root_gen(root_backup,
1900 btrfs_header_generation(info->extent_root->node));
1901 btrfs_set_backup_extent_root_level(root_backup,
1902 btrfs_header_level(info->extent_root->node));
1903
7c7e82a7
CM		 1904 /*
1905 * we might commit during log recovery, which happens before we set
1906 * the fs_root. Make sure it is valid before we fill it in.
1907 */
1908 if (info->fs_root && info->fs_root->node) {
1909 btrfs_set_backup_fs_root(root_backup,
1910 info->fs_root->node->start);
1911 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1912 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1913 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1914 btrfs_header_level(info->fs_root->node));
7c7e82a7 1915 }
af31f5e5
CM		 1916
1917 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1918 btrfs_set_backup_dev_root_gen(root_backup,
1919 btrfs_header_generation(info->dev_root->node));
1920 btrfs_set_backup_dev_root_level(root_backup,
1921 btrfs_header_level(info->dev_root->node));
1922
1923 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1924 btrfs_set_backup_csum_root_gen(root_backup,
1925 btrfs_header_generation(info->csum_root->node));
1926 btrfs_set_backup_csum_root_level(root_backup,
1927 btrfs_header_level(info->csum_root->node));
1928
1929 btrfs_set_backup_total_bytes(root_backup,
1930 btrfs_super_total_bytes(info->super_copy));
1931 btrfs_set_backup_bytes_used(root_backup,
1932 btrfs_super_bytes_used(info->super_copy));
1933 btrfs_set_backup_num_devices(root_backup,
1934 btrfs_super_num_devices(info->super_copy));
1935
1936 /*
1937 * if we don't copy this out to the super_copy, it won't get remembered
1938 * for the next commit
1939 */
1940 memcpy(&info->super_copy->super_roots,
1941 &info->super_for_commit->super_roots,
1942 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1943}
1944
1945/*
1946 * this copies info out of the root backup array and back into
1947 * the in-memory super block. It is meant to help iterate through
1948 * the array, so you send it the number of backups you've already
1949 * tried and the last backup index you used.
1950 *
1951 * this returns -1 when it has tried all the backups
1952 */
1953static noinline int next_root_backup(struct btrfs_fs_info *info,
1954 struct btrfs_super_block *super,
1955 int *num_backups_tried, int *backup_index)
1956{
1957 struct btrfs_root_backup *root_backup;
1958 int newest = *backup_index;
1959
1960 if (*num_backups_tried == 0) {
1961 u64 gen = btrfs_super_generation(super);
1962
1963 newest = find_newest_super_backup(info, gen);
1964 if (newest == -1)
1965 return -1;
1966
1967 *backup_index = newest;
1968 *num_backups_tried = 1;
1969 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1970 /* we've tried all the backups, all done */
1971 return -1;
1972 } else {
1973 /* jump to the next oldest backup */
1974 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1975 BTRFS_NUM_BACKUP_ROOTS;
1976 *backup_index = newest;
1977 *num_backups_tried += 1;
1978 }
1979 root_backup = super->super_roots + newest;
1980
1981 btrfs_set_super_generation(super,
1982 btrfs_backup_tree_root_gen(root_backup));
1983 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1984 btrfs_set_super_root_level(super,
1985 btrfs_backup_tree_root_level(root_backup));
1986 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1987
1988 /*
1989 * fixme: the total bytes and num_devices need to match or we should
1990 * need a fsck
1991 */
1992 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1993 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1994 return 0;
1995}
1996
7abadb64
LB		 1997/* helper to cleanup workers */
1998static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1999{
2000 btrfs_stop_workers(&fs_info->generic_worker);
2001 btrfs_stop_workers(&fs_info->fixup_workers);
2002 btrfs_stop_workers(&fs_info->delalloc_workers);
2003 btrfs_stop_workers(&fs_info->workers);
2004 btrfs_stop_workers(&fs_info->endio_workers);
2005 btrfs_stop_workers(&fs_info->endio_meta_workers);
2006 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2007 btrfs_stop_workers(&fs_info->rmw_workers);
2008 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2009 btrfs_stop_workers(&fs_info->endio_write_workers);
2010 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2011 btrfs_stop_workers(&fs_info->submit_workers);
2012 btrfs_stop_workers(&fs_info->delayed_workers);
2013 btrfs_stop_workers(&fs_info->caching_workers);
2014 btrfs_stop_workers(&fs_info->readahead_workers);
2015 btrfs_stop_workers(&fs_info->flush_workers);
2f232036 2016 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
7abadb64
LB		 2017}
2018
af31f5e5
CM		 2019/* helper to cleanup tree roots */
2020static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2021{
2022 free_extent_buffer(info->tree_root->node);
2023 free_extent_buffer(info->tree_root->commit_root);
af31f5e5
CM		 2024 info->tree_root->node = NULL;
2025 info->tree_root->commit_root = NULL;
655b09fe
JB		 2026
2027 if (info->dev_root) {
2028 free_extent_buffer(info->dev_root->node);
2029 free_extent_buffer(info->dev_root->commit_root);
2030 info->dev_root->node = NULL;
2031 info->dev_root->commit_root = NULL;
2032 }
2033 if (info->extent_root) {
2034 free_extent_buffer(info->extent_root->node);
2035 free_extent_buffer(info->extent_root->commit_root);
2036 info->extent_root->node = NULL;
2037 info->extent_root->commit_root = NULL;
2038 }
2039 if (info->csum_root) {
2040 free_extent_buffer(info->csum_root->node);
2041 free_extent_buffer(info->csum_root->commit_root);
2042 info->csum_root->node = NULL;
2043 info->csum_root->commit_root = NULL;
2044 }
bcef60f2 2045 if (info->quota_root) {
655b09fe
JB		 2046 free_extent_buffer(info->quota_root->node);
2047 free_extent_buffer(info->quota_root->commit_root);
bcef60f2
AJ		 2048 info->quota_root->node = NULL;
2049 info->quota_root->commit_root = NULL;
2050 }
f7a81ea4
SB		 2051 if (info->uuid_root) {
2052 free_extent_buffer(info->uuid_root->node);
2053 free_extent_buffer(info->uuid_root->commit_root);
2054 info->uuid_root->node = NULL;
2055 info->uuid_root->commit_root = NULL;
2056 }
af31f5e5
CM		 2057 if (chunk_root) {
2058 free_extent_buffer(info->chunk_root->node);
2059 free_extent_buffer(info->chunk_root->commit_root);
2060 info->chunk_root->node = NULL;
2061 info->chunk_root->commit_root = NULL;
2062 }
2063}
2064
171f6537
JB		 2065static void del_fs_roots(struct btrfs_fs_info *fs_info)
2066{
2067 int ret;
2068 struct btrfs_root *gang[8];
2069 int i;
2070
2071 while (!list_empty(&fs_info->dead_roots)) {
2072 gang[0] = list_entry(fs_info->dead_roots.next,
2073 struct btrfs_root, root_list);
2074 list_del(&gang[0]->root_list);
2075
2076 if (gang[0]->in_radix) {
cb517eab 2077 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
171f6537
JB		 2078 } else {
2079 free_extent_buffer(gang[0]->node);
2080 free_extent_buffer(gang[0]->commit_root);
b0feb9d9 2081 btrfs_put_fs_root(gang[0]);
171f6537
JB		 2082 }
2083 }
2084
2085 while (1) {
2086 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2087 (void **)gang, 0,
2088 ARRAY_SIZE(gang));
2089 if (!ret)
2090 break;
2091 for (i = 0; i < ret; i++)
cb517eab 2092 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
171f6537
JB		 2093 }
2094}
af31f5e5 2095
ad2b2c80
AV		 2096int open_ctree(struct super_block *sb,
2097 struct btrfs_fs_devices *fs_devices,
2098 char *options)
2e635a27 2099{
db94535d
CM		 2100 u32 sectorsize;
2101 u32 nodesize;
2102 u32 leafsize;
2103 u32 blocksize;
87ee04eb 2104 u32 stripesize;
84234f3a 2105 u64 generation;
f2b636e8 2106 u64 features;
3de4586c 2107 struct btrfs_key location;
a061fc8d 2108 struct buffer_head *bh;
4d34b278 2109 struct btrfs_super_block *disk_super;
815745cf 2110 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2111 struct btrfs_root *tree_root;
4d34b278
ID		 2112 struct btrfs_root *extent_root;
2113 struct btrfs_root *csum_root;
2114 struct btrfs_root *chunk_root;
2115 struct btrfs_root *dev_root;
bcef60f2 2116 struct btrfs_root *quota_root;
f7a81ea4 2117 struct btrfs_root *uuid_root;
e02119d5 2118 struct btrfs_root *log_tree_root;
eb60ceac 2119 int ret;
e58ca020 2120 int err = -EINVAL;
af31f5e5
CM		 2121 int num_backups_tried = 0;
2122 int backup_index = 0;
f7a81ea4 2123 bool create_uuid_tree = false;
4543df7e 2124
f84a8bd6 2125 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e 2126 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
cb517eab 2127 if (!tree_root || !chunk_root) {
39279cc3
CM		 2128 err = -ENOMEM;
2129 goto fail;
2130 }
76dda93c
YZ		 2131
2132 ret = init_srcu_struct(&fs_info->subvol_srcu);
2133 if (ret) {
2134 err = ret;
2135 goto fail;
2136 }
2137
2138 ret = setup_bdi(fs_info, &fs_info->bdi);
2139 if (ret) {
2140 err = ret;
2141 goto fail_srcu;
2142 }
2143
e2d84521
MX		 2144 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2145 if (ret) {
2146 err = ret;
2147 goto fail_bdi;
2148 }
2149 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2150 (1 + ilog2(nr_cpu_ids));
2151
963d678b
MX		 2152 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2153 if (ret) {
2154 err = ret;
2155 goto fail_dirty_metadata_bytes;
2156 }
2157
76dda93c
YZ		 2158 fs_info->btree_inode = new_inode(sb);
2159 if (!fs_info->btree_inode) {
2160 err = -ENOMEM;
963d678b 2161 goto fail_delalloc_bytes;
76dda93c
YZ		 2162 }
2163
a6591715 2164 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2165
76dda93c 2166 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 2167 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2168 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2169 INIT_LIST_HEAD(&fs_info->delayed_iputs);
eb73c1b7 2170 INIT_LIST_HEAD(&fs_info->delalloc_roots);
11833d66 2171 INIT_LIST_HEAD(&fs_info->caching_block_groups);
eb73c1b7 2172 spin_lock_init(&fs_info->delalloc_root_lock);
a4abeea4 2173 spin_lock_init(&fs_info->trans_lock);
76dda93c 2174 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2175 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2176 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2177 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2178 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2179 spin_lock_init(&fs_info->super_lock);
f29021b2 2180 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2181 mutex_init(&fs_info->reloc_mutex);
de98ced9 2182 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2183
58176a96 2184 init_completion(&fs_info->kobj_unregister);
0b86a832 2185 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2186 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2187 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2188 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2189 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2190 BTRFS_BLOCK_RSV_GLOBAL);
2191 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2192 BTRFS_BLOCK_RSV_DELALLOC);
2193 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2194 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2195 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2196 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2197 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2198 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2199 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2200 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2201 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2202 atomic_set(&fs_info->defrag_running, 0);
fc36ed7e 2203 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2204 fs_info->sb = sb;
6f568d35 2205 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2206 fs_info->metadata_ratio = 0;
4cb5300b 2207 fs_info->defrag_inodes = RB_ROOT;
2bf64758 2208 fs_info->free_chunk_space = 0;
f29021b2 2209 fs_info->tree_mod_log = RB_ROOT;
8b87dc17 2210 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
c8b97818 2211
90519d66
AJ		 2212 /* readahead state */
2213 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2214 spin_lock_init(&fs_info->reada_lock);
c8b97818 2215
b34b086c
CM		 2216 fs_info->thread_pool_size = min_t(unsigned long,
2217 num_online_cpus() + 2, 8);
0afbaf8c 2218
199c2a9c
MX		 2219 INIT_LIST_HEAD(&fs_info->ordered_roots);
2220 spin_lock_init(&fs_info->ordered_root_lock);
16cdcec7
MX		 2221 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2222 GFP_NOFS);
2223 if (!fs_info->delayed_root) {
2224 err = -ENOMEM;
2225 goto fail_iput;
2226 }
2227 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2228
a2de733c
AJ		 2229 mutex_init(&fs_info->scrub_lock);
2230 atomic_set(&fs_info->scrubs_running, 0);
2231 atomic_set(&fs_info->scrub_pause_req, 0);
2232 atomic_set(&fs_info->scrubs_paused, 0);
2233 atomic_set(&fs_info->scrub_cancel_req, 0);
2234 init_waitqueue_head(&fs_info->scrub_pause_wait);
2235 init_rwsem(&fs_info->scrub_super_lock);
2236 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2237#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2238 fs_info->check_integrity_print_mask = 0;
2239#endif
a2de733c 2240
c9e9f97b
ID		 2241 spin_lock_init(&fs_info->balance_lock);
2242 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2243 atomic_set(&fs_info->balance_running, 0);
2244 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2245 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2246 fs_info->balance_ctl = NULL;
837d5b6e 2247 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2248
a061fc8d
CM		 2249 sb->s_blocksize = 4096;
2250 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2251 sb->s_bdi = &fs_info->bdi;
a061fc8d 2252
76dda93c 2253 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2254 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2255 /*
2256 * we set the i_size on the btree inode to the max possible int.
2257 * the real end of the address space is determined by all of
2258 * the devices in the system
2259 */
2260 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2261 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2262 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2263
5d4f98a2 2264 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2265 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2266 fs_info->btree_inode->i_mapping);
0b32f4bb 2267 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2268 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2269
2270 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2271
76dda93c
YZ		 2272 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2273 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2274 sizeof(struct btrfs_key));
72ac3c0d
JB		 2275 set_bit(BTRFS_INODE_DUMMY,
2276 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
c65ddb52 2277 insert_inode_hash(fs_info->btree_inode);
76dda93c 2278
0f9dd46c 2279 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2280 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2281 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2282
11833d66 2283 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2284 fs_info->btree_inode->i_mapping);
11833d66 2285 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2286 fs_info->btree_inode->i_mapping);
11833d66 2287 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2288 fs_info->do_barriers = 1;
e18e4809 2289
39279cc3 2290
5a3f23d5 2291 mutex_init(&fs_info->ordered_operations_mutex);
9ffba8cd 2292 mutex_init(&fs_info->ordered_extent_flush_mutex);
e02119d5 2293 mutex_init(&fs_info->tree_log_mutex);
925baedd 2294 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2295 mutex_init(&fs_info->transaction_kthread_mutex);
2296 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2297 mutex_init(&fs_info->volume_mutex);
276e680d 2298 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2299 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2300 init_rwsem(&fs_info->subvol_sem);
803b2f54 2301 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
e922e087
SB		 2302 fs_info->dev_replace.lock_owner = 0;
2303 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2304 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2305 mutex_init(&fs_info->dev_replace.lock_management_lock);
2306 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2307
416ac51d 2308 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2309 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2310 fs_info->qgroup_tree = RB_ROOT;
2311 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2312 fs_info->qgroup_seq = 1;
2313 fs_info->quota_enabled = 0;
2314 fs_info->pending_quota_state = 0;
1e8f9158 2315 fs_info->qgroup_ulist = NULL;
2f232036 2316 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2317
fa9c0d79
CM		 2318 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2319 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2320
e6dcd2dc 2321 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2322 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2323 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2324 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2325
53b381b3
DW		 2326 ret = btrfs_alloc_stripe_hash_table(fs_info);
2327 if (ret) {
83c8266a 2328 err = ret;
53b381b3
DW		 2329 goto fail_alloc;
2330 }
2331
0b86a832 2332 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2333 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2334
3c4bb26b 2335 invalidate_bdev(fs_devices->latest_bdev);
1104a885
DS		 2336
2337 /*
2338 * Read super block and check the signature bytes only
2339 */
a512bbf8 2340 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2341 if (!bh) {
2342 err = -EINVAL;
16cdcec7 2343 goto fail_alloc;
20b45077 2344 }
39279cc3 2345
1104a885
DS		 2346 /*
2347 * We want to check superblock checksum, the type is stored inside.
2348 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2349 */
2350 if (btrfs_check_super_csum(bh->b_data)) {
2351 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2352 err = -EINVAL;
2353 goto fail_alloc;
2354 }
2355
2356 /*
2357 * super_copy is zeroed at allocation time and we never touch the
2358 * following bytes up to INFO_SIZE, the checksum is calculated from
2359 * the whole block of INFO_SIZE
2360 */
6c41761f
DS		 2361 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2362 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2363 sizeof(*fs_info->super_for_commit));
a061fc8d 2364 brelse(bh);
5f39d397 2365
6c41761f 2366 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2367
1104a885
DS		 2368 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2369 if (ret) {
2370 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2371 err = -EINVAL;
2372 goto fail_alloc;
2373 }
2374
6c41761f 2375 disk_super = fs_info->super_copy;
0f7d52f4 2376 if (!btrfs_super_root(disk_super))
16cdcec7 2377 goto fail_alloc;
0f7d52f4 2378
acce952b 2379 /* check FS state, whether FS is broken. */
87533c47
MX		 2380 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2381 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2382
af31f5e5
CM		 2383 /*
2384 * run through our array of backup supers and setup
2385 * our ring pointer to the oldest one
2386 */
2387 generation = btrfs_super_generation(disk_super);
2388 find_oldest_super_backup(fs_info, generation);
2389
75e7cb7f
LB		 2390 /*
2391 * In the long term, we'll store the compression type in the super
2392 * block, and it'll be used for per file compression control.
2393 */
2394 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2395
2b82032c
YZ		 2396 ret = btrfs_parse_options(tree_root, options);
2397 if (ret) {
2398 err = ret;
16cdcec7 2399 goto fail_alloc;
2b82032c 2400 }
dfe25020 2401
f2b636e8
JB		 2402 features = btrfs_super_incompat_flags(disk_super) &
2403 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2404 if (features) {
2405 printk(KERN_ERR "BTRFS: couldn't mount because of "
2406 "unsupported optional features (%Lx).\n",
21380931 2407 (unsigned long long)features);
f2b636e8 2408 err = -EINVAL;
16cdcec7 2409 goto fail_alloc;
f2b636e8
JB		 2410 }
2411
727011e0
CM		 2412 if (btrfs_super_leafsize(disk_super) !=
2413 btrfs_super_nodesize(disk_super)) {
2414 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2415 "blocksizes don't match. node %d leaf %d\n",
2416 btrfs_super_nodesize(disk_super),
2417 btrfs_super_leafsize(disk_super));
2418 err = -EINVAL;
2419 goto fail_alloc;
2420 }
2421 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2422 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2423 "blocksize (%d) was too large\n",
2424 btrfs_super_leafsize(disk_super));
2425 err = -EINVAL;
2426 goto fail_alloc;
2427 }
2428
5d4f98a2 2429 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2430 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2431 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2432 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2433
3173a18f
JB		 2434 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2435 printk(KERN_ERR "btrfs: has skinny extents\n");
2436
727011e0
CM		 2437 /*
2438 * flag our filesystem as having big metadata blocks if
2439 * they are bigger than the page size
2440 */
2441 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2442 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2443 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2444 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2445 }
2446
bc3f116f
CM		 2447 nodesize = btrfs_super_nodesize(disk_super);
2448 leafsize = btrfs_super_leafsize(disk_super);
2449 sectorsize = btrfs_super_sectorsize(disk_super);
2450 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2451 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2452 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2453
2454 /*
2455 * mixed block groups end up with duplicate but slightly offset
2456 * extent buffers for the same range. It leads to corruptions
2457 */
2458 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2459 (sectorsize != leafsize)) {
2460 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2461 "are not allowed for mixed block groups on %s\n",
2462 sb->s_id);
2463 goto fail_alloc;
2464 }
2465
ceda0864
MX		 2466 /*
2467 * Needn't use the lock because there is no other task which will
2468 * update the flag.
2469 */
a6fa6fae 2470 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2471
f2b636e8
JB		 2472 features = btrfs_super_compat_ro_flags(disk_super) &
2473 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2474 if (!(sb->s_flags & MS_RDONLY) && features) {
2475 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2476 "unsupported option features (%Lx).\n",
21380931 2477 (unsigned long long)features);
f2b636e8 2478 err = -EINVAL;
16cdcec7 2479 goto fail_alloc;
f2b636e8 2480 }
61d92c32
CM		 2481
2482 btrfs_init_workers(&fs_info->generic_worker,
2483 "genwork", 1, NULL);
2484
5443be45 2485 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 2486 fs_info->thread_pool_size,
2487 &fs_info->generic_worker);
c8b97818 2488
771ed689 2489 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 2490 fs_info->thread_pool_size,
2491 &fs_info->generic_worker);
771ed689 2492
8ccf6f19
MX		 2493 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2494 fs_info->thread_pool_size,
2495 &fs_info->generic_worker);
2496
5443be45 2497 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 2498 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 2499 fs_info->thread_pool_size),
2500 &fs_info->generic_worker);
61b49440 2501
bab39bf9
JB		 2502 btrfs_init_workers(&fs_info->caching_workers, "cache",
2503 2, &fs_info->generic_worker);
2504
61b49440
CM		 2505 /* a higher idle thresh on the submit workers makes it much more
2506 * likely that bios will be send down in a sane order to the
2507 * devices
2508 */
2509 fs_info->submit_workers.idle_thresh = 64;
53863232 2510
771ed689 2511 fs_info->workers.idle_thresh = 16;
4a69a410 2512 fs_info->workers.ordered = 1;
61b49440 2513
771ed689
CM		 2514 fs_info->delalloc_workers.idle_thresh = 2;
2515 fs_info->delalloc_workers.ordered = 1;
2516
61d92c32
CM		 2517 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2518 &fs_info->generic_worker);
5443be45 2519 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2520 fs_info->thread_pool_size,
2521 &fs_info->generic_worker);
d20f7043 2522 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2523 fs_info->thread_pool_size,
2524 &fs_info->generic_worker);
cad321ad 2525 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2526 "endio-meta-write", fs_info->thread_pool_size,
2527 &fs_info->generic_worker);
53b381b3
DW		 2528 btrfs_init_workers(&fs_info->endio_raid56_workers,
2529 "endio-raid56", fs_info->thread_pool_size,
2530 &fs_info->generic_worker);
2531 btrfs_init_workers(&fs_info->rmw_workers,
2532 "rmw", fs_info->thread_pool_size,
2533 &fs_info->generic_worker);
5443be45 2534 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2535 fs_info->thread_pool_size,
2536 &fs_info->generic_worker);
0cb59c99
JB		 2537 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2538 1, &fs_info->generic_worker);
16cdcec7
MX		 2539 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2540 fs_info->thread_pool_size,
2541 &fs_info->generic_worker);
90519d66
AJ		 2542 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2543 fs_info->thread_pool_size,
2544 &fs_info->generic_worker);
2f232036
JS		 2545 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2546 &fs_info->generic_worker);
61b49440
CM		 2547
2548 /*
2549 * endios are largely parallel and should have a very
2550 * low idle thresh
2551 */
2552 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2553 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2554 fs_info->endio_raid56_workers.idle_thresh = 4;
2555 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2556
9042846b
CM		 2557 fs_info->endio_write_workers.idle_thresh = 2;
2558 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2559 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2560
0dc3b84a
JB		 2561 /*
2562 * btrfs_start_workers can really only fail because of ENOMEM so just
2563 * return -ENOMEM if any of these fail.
2564 */
2565 ret = btrfs_start_workers(&fs_info->workers);
2566 ret |= btrfs_start_workers(&fs_info->generic_worker);
2567 ret |= btrfs_start_workers(&fs_info->submit_workers);
2568 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2569 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2570 ret |= btrfs_start_workers(&fs_info->endio_workers);
2571 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2572 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2573 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2574 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2575 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2576 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2577 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2578 ret |= btrfs_start_workers(&fs_info->caching_workers);
2579 ret |= btrfs_start_workers(&fs_info->readahead_workers);
8ccf6f19 2580 ret |= btrfs_start_workers(&fs_info->flush_workers);
2f232036 2581 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
0dc3b84a 2582 if (ret) {
fed425c7 2583 err = -ENOMEM;
0dc3b84a
JB		 2584 goto fail_sb_buffer;
2585 }
4543df7e 2586
4575c9cc 2587 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2588 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2589 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2590
db94535d
CM		 2591 tree_root->nodesize = nodesize;
2592 tree_root->leafsize = leafsize;
2593 tree_root->sectorsize = sectorsize;
87ee04eb 2594 tree_root->stripesize = stripesize;
a061fc8d
CM		 2595
2596 sb->s_blocksize = sectorsize;
2597 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2598
3cae210f 2599 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
d397712b 2600 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2601 goto fail_sb_buffer;
2602 }
19c00ddc 2603
8d082fb7
LB		 2604 if (sectorsize != PAGE_SIZE) {
2605 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2606 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2607 goto fail_sb_buffer;
2608 }
2609
925baedd 2610 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2611 ret = btrfs_read_sys_array(tree_root);
925baedd 2612 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2613 if (ret) {
d397712b
CM		 2614 printk(KERN_WARNING "btrfs: failed to read the system "
2615 "array on %s\n", sb->s_id);
5d4f98a2 2616 goto fail_sb_buffer;
84eed90f 2617 }
0b86a832
CM		 2618
2619 blocksize = btrfs_level_size(tree_root,
2620 btrfs_super_chunk_root_level(disk_super));
84234f3a 2621 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2622
2623 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2624 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2625
2626 chunk_root->node = read_tree_block(chunk_root,
2627 btrfs_super_chunk_root(disk_super),
84234f3a 2628 blocksize, generation);
416bc658
JB		 2629 if (!chunk_root->node ||
2630 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
83121942
DW		 2631 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2632 sb->s_id);
af31f5e5 2633 goto fail_tree_roots;
83121942 2634 }
5d4f98a2
YZ		 2635 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2636 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2637
e17cade2 2638 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 2639 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2640 BTRFS_UUID_SIZE);
e17cade2 2641
0b86a832 2642 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2643 if (ret) {
d397712b
CM		 2644 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2645 sb->s_id);
af31f5e5 2646 goto fail_tree_roots;
2b82032c 2647 }
0b86a832 2648
8dabb742
SB		 2649 /*
2650 * keep the device that is marked to be the target device for the
2651 * dev_replace procedure
2652 */
2653 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2654
a6b0d5c8
CM		 2655 if (!fs_devices->latest_bdev) {
2656 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2657 sb->s_id);
2658 goto fail_tree_roots;
2659 }
2660
af31f5e5 2661retry_root_backup:
db94535d
CM		 2662 blocksize = btrfs_level_size(tree_root,
2663 btrfs_super_root_level(disk_super));
84234f3a 2664 generation = btrfs_super_generation(disk_super);
0b86a832 2665
e20d96d6 2666 tree_root->node = read_tree_block(tree_root,
db94535d 2667 btrfs_super_root(disk_super),
84234f3a 2668 blocksize, generation);
af31f5e5
CM		 2669 if (!tree_root->node ||
2670 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
83121942
DW		 2671 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2672 sb->s_id);
af31f5e5
CM		 2673
2674 goto recovery_tree_root;
83121942 2675 }
af31f5e5 2676
5d4f98a2
YZ		 2677 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2678 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d 2679
cb517eab
MX		 2680 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2681 location.type = BTRFS_ROOT_ITEM_KEY;
2682 location.offset = 0;
2683
2684 extent_root = btrfs_read_tree_root(tree_root, &location);
2685 if (IS_ERR(extent_root)) {
2686 ret = PTR_ERR(extent_root);
af31f5e5 2687 goto recovery_tree_root;
cb517eab 2688 }
0b86a832 2689 extent_root->track_dirty = 1;
cb517eab 2690 fs_info->extent_root = extent_root;
0b86a832 2691
cb517eab
MX		 2692 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2693 dev_root = btrfs_read_tree_root(tree_root, &location);
2694 if (IS_ERR(dev_root)) {
2695 ret = PTR_ERR(dev_root);
af31f5e5 2696 goto recovery_tree_root;
cb517eab 2697 }
5d4f98a2 2698 dev_root->track_dirty = 1;
cb517eab
MX		 2699 fs_info->dev_root = dev_root;
2700 btrfs_init_devices_late(fs_info);
3768f368 2701
cb517eab
MX		 2702 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2703 csum_root = btrfs_read_tree_root(tree_root, &location);
2704 if (IS_ERR(csum_root)) {
2705 ret = PTR_ERR(csum_root);
af31f5e5 2706 goto recovery_tree_root;
cb517eab 2707 }
d20f7043 2708 csum_root->track_dirty = 1;
cb517eab 2709 fs_info->csum_root = csum_root;
d20f7043 2710
cb517eab
MX		 2711 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2712 quota_root = btrfs_read_tree_root(tree_root, &location);
2713 if (!IS_ERR(quota_root)) {
bcef60f2
AJ		 2714 quota_root->track_dirty = 1;
2715 fs_info->quota_enabled = 1;
2716 fs_info->pending_quota_state = 1;
cb517eab 2717 fs_info->quota_root = quota_root;
bcef60f2
AJ		 2718 }
2719
f7a81ea4
SB		 2720 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2721 uuid_root = btrfs_read_tree_root(tree_root, &location);
2722 if (IS_ERR(uuid_root)) {
2723 ret = PTR_ERR(uuid_root);
2724 if (ret != -ENOENT)
2725 goto recovery_tree_root;
2726 create_uuid_tree = true;
2727 } else {
2728 uuid_root->track_dirty = 1;
2729 fs_info->uuid_root = uuid_root;
2730 }
2731
8929ecfa
YZ		 2732 fs_info->generation = generation;
2733 fs_info->last_trans_committed = generation;
8929ecfa 2734
68310a5e
ID		 2735 ret = btrfs_recover_balance(fs_info);
2736 if (ret) {
2737 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2738 goto fail_block_groups;
2739 }
2740
733f4fbb
SB		 2741 ret = btrfs_init_dev_stats(fs_info);
2742 if (ret) {
2743 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2744 ret);
2745 goto fail_block_groups;
2746 }
2747
8dabb742
SB		 2748 ret = btrfs_init_dev_replace(fs_info);
2749 if (ret) {
2750 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2751 goto fail_block_groups;
2752 }
2753
2754 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2755
c59021f8 2756 ret = btrfs_init_space_info(fs_info);
2757 if (ret) {
2758 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2759 goto fail_block_groups;
2760 }
2761
1b1d1f66
JB		 2762 ret = btrfs_read_block_groups(extent_root);
2763 if (ret) {
2764 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2765 goto fail_block_groups;
2766 }
5af3e8cc
SB		 2767 fs_info->num_tolerated_disk_barrier_failures =
2768 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2769 if (fs_info->fs_devices->missing_devices >
2770 fs_info->num_tolerated_disk_barrier_failures &&
2771 !(sb->s_flags & MS_RDONLY)) {
2772 printk(KERN_WARNING
2773 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2774 goto fail_block_groups;
2775 }
9078a3e1 2776
a74a4b97
CM		 2777 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2778 "btrfs-cleaner");
57506d50 2779 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2780 goto fail_block_groups;
a74a4b97
CM		 2781
2782 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2783 tree_root,
2784 "btrfs-transaction");
57506d50 2785 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2786 goto fail_cleaner;
a74a4b97 2787
c289811c
CM		 2788 if (!btrfs_test_opt(tree_root, SSD) &&
2789 !btrfs_test_opt(tree_root, NOSSD) &&
2790 !fs_info->fs_devices->rotating) {
2791 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2792 "mode\n");
2793 btrfs_set_opt(fs_info->mount_opt, SSD);
2794 }
2795
21adbd5c
SB		 2796#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2797 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2798 ret = btrfsic_mount(tree_root, fs_devices,
2799 btrfs_test_opt(tree_root,
2800 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2801 1 : 0,
2802 fs_info->check_integrity_print_mask);
2803 if (ret)
2804 printk(KERN_WARNING "btrfs: failed to initialize"
2805 " integrity check module %s\n", sb->s_id);
2806 }
2807#endif
bcef60f2
AJ		 2808 ret = btrfs_read_qgroup_config(fs_info);
2809 if (ret)
2810 goto fail_trans_kthread;
21adbd5c 2811
acce952b 2812 /* do not make disk changes in broken FS */
68ce9682 2813 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2814 u64 bytenr = btrfs_super_log_root(disk_super);
2815
7c2ca468 2816 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2817 printk(KERN_WARNING "Btrfs log replay required "
2818 "on RO media\n");
7c2ca468 2819 err = -EIO;
bcef60f2 2820 goto fail_qgroup;
7c2ca468 2821 }
e02119d5
CM		 2822 blocksize =
2823 btrfs_level_size(tree_root,
2824 btrfs_super_log_root_level(disk_super));
d18a2c44 2825
6f07e42e 2826 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2827 if (!log_tree_root) {
2828 err = -ENOMEM;
bcef60f2 2829 goto fail_qgroup;
676e4c86 2830 }
e02119d5
CM		 2831
2832 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2833 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2834
2835 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2836 blocksize,
2837 generation + 1);
416bc658
JB		 2838 if (!log_tree_root->node ||
2839 !extent_buffer_uptodate(log_tree_root->node)) {
2840 printk(KERN_ERR "btrfs: failed to read log tree\n");
2841 free_extent_buffer(log_tree_root->node);
2842 kfree(log_tree_root);
2843 goto fail_trans_kthread;
2844 }
79787eaa 2845 /* returns with log_tree_root freed on success */
e02119d5 2846 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2847 if (ret) {
2848 btrfs_error(tree_root->fs_info, ret,
2849 "Failed to recover log tree");
2850 free_extent_buffer(log_tree_root->node);
2851 kfree(log_tree_root);
2852 goto fail_trans_kthread;
2853 }
e556ce2c
YZ		 2854
2855 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2856 ret = btrfs_commit_super(tree_root);
2857 if (ret)
2858 goto fail_trans_kthread;
e556ce2c 2859 }
e02119d5 2860 }
1a40e23b 2861
76dda93c 2862 ret = btrfs_find_orphan_roots(tree_root);
79787eaa
JM		 2863 if (ret)
2864 goto fail_trans_kthread;
76dda93c 2865
7c2ca468 2866 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2867 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2
ID		 2868 if (ret)
2869 goto fail_trans_kthread;
d68fc57b 2870
5d4f98a2 2871 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2872 if (ret < 0) {
2873 printk(KERN_WARNING
2874 "btrfs: failed to recover relocation\n");
2875 err = -EINVAL;
bcef60f2 2876 goto fail_qgroup;
d7ce5843 2877 }
7c2ca468 2878 }
1a40e23b 2879
3de4586c
CM		 2880 location.objectid = BTRFS_FS_TREE_OBJECTID;
2881 location.type = BTRFS_ROOT_ITEM_KEY;
cb517eab 2882 location.offset = 0;
3de4586c 2883
3de4586c 2884 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3140c9a3
DC		 2885 if (IS_ERR(fs_info->fs_root)) {
2886 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2887 goto fail_qgroup;
3140c9a3 2888 }
c289811c 2889
2b6ba629
ID		 2890 if (sb->s_flags & MS_RDONLY)
2891 return 0;
59641015 2892
2b6ba629
ID		 2893 down_read(&fs_info->cleanup_work_sem);
2894 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2895 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2896 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2897 close_ctree(tree_root);
2898 return ret;
2899 }
2900 up_read(&fs_info->cleanup_work_sem);
59641015 2901
2b6ba629
ID		 2902 ret = btrfs_resume_balance_async(fs_info);
2903 if (ret) {
2904 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2905 close_ctree(tree_root);
2906 return ret;
e3acc2a6
JB		 2907 }
2908
8dabb742
SB		 2909 ret = btrfs_resume_dev_replace_async(fs_info);
2910 if (ret) {
2911 pr_warn("btrfs: failed to resume dev_replace\n");
2912 close_ctree(tree_root);
2913 return ret;
2914 }
2915
b382a324
JS		 2916 btrfs_qgroup_rescan_resume(fs_info);
2917
f7a81ea4
SB		 2918 if (create_uuid_tree) {
2919 pr_info("btrfs: creating UUID tree\n");
2920 ret = btrfs_create_uuid_tree(fs_info);
2921 if (ret) {
2922 pr_warn("btrfs: failed to create the UUID tree %d\n",
2923 ret);
2924 close_ctree(tree_root);
2925 return ret;
2926 }
2927 }
2928
ad2b2c80 2929 return 0;
39279cc3 2930
bcef60f2
AJ		 2931fail_qgroup:
2932 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2933fail_trans_kthread:
2934 kthread_stop(fs_info->transaction_kthread);
54067ae9 2935 btrfs_cleanup_transaction(fs_info->tree_root);
7b5ff90e 2936 del_fs_roots(fs_info);
3f157a2f 2937fail_cleaner:
a74a4b97 2938 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2939
2940 /*
2941 * make sure we're done with the btree inode before we stop our
2942 * kthreads
2943 */
2944 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2945
1b1d1f66 2946fail_block_groups:
54067ae9 2947 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2948 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2949
2950fail_tree_roots:
2951 free_root_pointers(fs_info, 1);
2b8195bb 2952 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2953
39279cc3 2954fail_sb_buffer:
7abadb64 2955 btrfs_stop_all_workers(fs_info);
16cdcec7 2956fail_alloc:
4543df7e 2957fail_iput:
586e46e2
ID		 2958 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2959
4543df7e 2960 iput(fs_info->btree_inode);
963d678b
MX		 2961fail_delalloc_bytes:
2962 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 2963fail_dirty_metadata_bytes:
2964 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 2965fail_bdi:
7e662854 2966 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2967fail_srcu:
2968 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2969fail:
53b381b3 2970 btrfs_free_stripe_hash_table(fs_info);
586e46e2 2971 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 2972 return err;
af31f5e5
CM		 2973
2974recovery_tree_root:
af31f5e5
CM		 2975 if (!btrfs_test_opt(tree_root, RECOVERY))
2976 goto fail_tree_roots;
2977
2978 free_root_pointers(fs_info, 0);
2979
2980 /* don't use the log in recovery mode, it won't be valid */
2981 btrfs_set_super_log_root(disk_super, 0);
2982
2983 /* we can't trust the free space cache either */
2984 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2985
2986 ret = next_root_backup(fs_info, fs_info->super_copy,
2987 &num_backups_tried, &backup_index);
2988 if (ret == -1)
2989 goto fail_block_groups;
2990 goto retry_root_backup;
eb60ceac
CM		 2991}
2992
f2984462
CM		 2993static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2994{
f2984462
CM		 2995 if (uptodate) {
2996 set_buffer_uptodate(bh);
2997 } else {
442a4f63
SB		 2998 struct btrfs_device *device = (struct btrfs_device *)
2999 bh->b_private;
3000
606686ee
JB		 3001 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3002 "I/O error on %s\n",
3003 rcu_str_deref(device->name));
1259ab75
CM		 3004 /* note, we dont' set_buffer_write_io_error because we have
3005 * our own ways of dealing with the IO errors
3006 */
f2984462 3007 clear_buffer_uptodate(bh);
442a4f63 3008 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 3009 }
3010 unlock_buffer(bh);
3011 put_bh(bh);
3012}
3013
a512bbf8
YZ		 3014struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3015{
3016 struct buffer_head *bh;
3017 struct buffer_head *latest = NULL;
3018 struct btrfs_super_block *super;
3019 int i;
3020 u64 transid = 0;
3021 u64 bytenr;
3022
3023 /* we would like to check all the supers, but that would make
3024 * a btrfs mount succeed after a mkfs from a different FS.
3025 * So, we need to add a special mount option to scan for
3026 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3027 */
3028 for (i = 0; i < 1; i++) {
3029 bytenr = btrfs_sb_offset(i);
8068a47e
AJ		 3030 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3031 i_size_read(bdev->bd_inode))
a512bbf8 3032 break;
8068a47e
AJ		 3033 bh = __bread(bdev, bytenr / 4096,
3034 BTRFS_SUPER_INFO_SIZE);
a512bbf8
YZ		 3035 if (!bh)
3036 continue;
3037
3038 super = (struct btrfs_super_block *)bh->b_data;
3039 if (btrfs_super_bytenr(super) != bytenr ||
3cae210f 3040 btrfs_super_magic(super) != BTRFS_MAGIC) {
a512bbf8
YZ		 3041 brelse(bh);
3042 continue;
3043 }
3044
3045 if (!latest || btrfs_super_generation(super) > transid) {
3046 brelse(latest);
3047 latest = bh;
3048 transid = btrfs_super_generation(super);
3049 } else {
3050 brelse(bh);
3051 }
3052 }
3053 return latest;
3054}
3055
4eedeb75
HH		 3056/*
3057 * this should be called twice, once with wait == 0 and
3058 * once with wait == 1. When wait == 0 is done, all the buffer heads
3059 * we write are pinned.
3060 *
3061 * They are released when wait == 1 is done.
3062 * max_mirrors must be the same for both runs, and it indicates how
3063 * many supers on this one device should be written.
3064 *
3065 * max_mirrors == 0 means to write them all.
3066 */
a512bbf8
YZ		 3067static int write_dev_supers(struct btrfs_device *device,
3068 struct btrfs_super_block *sb,
3069 int do_barriers, int wait, int max_mirrors)
3070{
3071 struct buffer_head *bh;
3072 int i;
3073 int ret;
3074 int errors = 0;
3075 u32 crc;
3076 u64 bytenr;
a512bbf8
YZ		 3077
3078 if (max_mirrors == 0)
3079 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3080
a512bbf8
YZ		 3081 for (i = 0; i < max_mirrors; i++) {
3082 bytenr = btrfs_sb_offset(i);
3083 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3084 break;
3085
3086 if (wait) {
3087 bh = __find_get_block(device->bdev, bytenr / 4096,
3088 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3089 if (!bh) {
3090 errors++;
3091 continue;
3092 }
a512bbf8 3093 wait_on_buffer(bh);
4eedeb75
HH		 3094 if (!buffer_uptodate(bh))
3095 errors++;
3096
3097 /* drop our reference */
3098 brelse(bh);
3099
3100 /* drop the reference from the wait == 0 run */
3101 brelse(bh);
3102 continue;
a512bbf8
YZ		 3103 } else {
3104 btrfs_set_super_bytenr(sb, bytenr);
3105
3106 crc = ~(u32)0;
b0496686 3107 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 3108 BTRFS_CSUM_SIZE, crc,
3109 BTRFS_SUPER_INFO_SIZE -
3110 BTRFS_CSUM_SIZE);
3111 btrfs_csum_final(crc, sb->csum);
3112
4eedeb75
HH		 3113 /*
3114 * one reference for us, and we leave it for the
3115 * caller
3116 */
a512bbf8
YZ		 3117 bh = __getblk(device->bdev, bytenr / 4096,
3118 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3119 if (!bh) {
3120 printk(KERN_ERR "btrfs: couldn't get super "
3121 "buffer head for bytenr %Lu\n", bytenr);
3122 errors++;
3123 continue;
3124 }
3125
a512bbf8
YZ		 3126 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3127
4eedeb75 3128 /* one reference for submit_bh */
a512bbf8 3129 get_bh(bh);
4eedeb75
HH		 3130
3131 set_buffer_uptodate(bh);
a512bbf8
YZ		 3132 lock_buffer(bh);
3133 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3134 bh->b_private = device;
a512bbf8
YZ		 3135 }
3136
387125fc
CM		 3137 /*
3138 * we fua the first super. The others we allow
3139 * to go down lazy.
3140 */
21adbd5c 3141 ret = btrfsic_submit_bh(WRITE_FUA, bh);
4eedeb75 3142 if (ret)
a512bbf8 3143 errors++;
a512bbf8
YZ		 3144 }
3145 return errors < i ? 0 : -1;
3146}
3147
387125fc
CM		 3148/*
3149 * endio for the write_dev_flush, this will wake anyone waiting
3150 * for the barrier when it is done
3151 */
3152static void btrfs_end_empty_barrier(struct bio *bio, int err)
3153{
3154 if (err) {
3155 if (err == -EOPNOTSUPP)
3156 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3157 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3158 }
3159 if (bio->bi_private)
3160 complete(bio->bi_private);
3161 bio_put(bio);
3162}
3163
3164/*
3165 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3166 * sent down. With wait == 1, it waits for the previous flush.
3167 *
3168 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3169 * capable
3170 */
3171static int write_dev_flush(struct btrfs_device *device, int wait)
3172{
3173 struct bio *bio;
3174 int ret = 0;
3175
3176 if (device->nobarriers)
3177 return 0;
3178
3179 if (wait) {
3180 bio = device->flush_bio;
3181 if (!bio)
3182 return 0;
3183
3184 wait_for_completion(&device->flush_wait);
3185
3186 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
606686ee
JB		 3187 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3188 rcu_str_deref(device->name));
387125fc 3189 device->nobarriers = 1;
5af3e8cc 3190 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3191 ret = -EIO;
5af3e8cc
SB		 3192 btrfs_dev_stat_inc_and_print(device,
3193 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3194 }
3195
3196 /* drop the reference from the wait == 0 run */
3197 bio_put(bio);
3198 device->flush_bio = NULL;
3199
3200 return ret;
3201 }
3202
3203 /*
3204 * one reference for us, and we leave it for the
3205 * caller
3206 */
9c017abc 3207 device->flush_bio = NULL;
9be3395b 3208 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
387125fc
CM		 3209 if (!bio)
3210 return -ENOMEM;
3211
3212 bio->bi_end_io = btrfs_end_empty_barrier;
3213 bio->bi_bdev = device->bdev;
3214 init_completion(&device->flush_wait);
3215 bio->bi_private = &device->flush_wait;
3216 device->flush_bio = bio;
3217
3218 bio_get(bio);
21adbd5c 3219 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3220
3221 return 0;
3222}
3223
3224/*
3225 * send an empty flush down to each device in parallel,
3226 * then wait for them
3227 */
3228static int barrier_all_devices(struct btrfs_fs_info *info)
3229{
3230 struct list_head *head;
3231 struct btrfs_device *dev;
5af3e8cc
SB		 3232 int errors_send = 0;
3233 int errors_wait = 0;
387125fc
CM		 3234 int ret;
3235
3236 /* send down all the barriers */
3237 head = &info->fs_devices->devices;
3238 list_for_each_entry_rcu(dev, head, dev_list) {
3239 if (!dev->bdev) {
5af3e8cc 3240 errors_send++;
387125fc
CM		 3241 continue;
3242 }
3243 if (!dev->in_fs_metadata || !dev->writeable)
3244 continue;
3245
3246 ret = write_dev_flush(dev, 0);
3247 if (ret)
5af3e8cc 3248 errors_send++;
387125fc
CM		 3249 }
3250
3251 /* wait for all the barriers */
3252 list_for_each_entry_rcu(dev, head, dev_list) {
3253 if (!dev->bdev) {
5af3e8cc 3254 errors_wait++;
387125fc
CM		 3255 continue;
3256 }
3257 if (!dev->in_fs_metadata || !dev->writeable)
3258 continue;
3259
3260 ret = write_dev_flush(dev, 1);
3261 if (ret)
5af3e8cc 3262 errors_wait++;
387125fc 3263 }
5af3e8cc
SB		 3264 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3265 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3266 return -EIO;
3267 return 0;
3268}
3269
5af3e8cc
SB		 3270int btrfs_calc_num_tolerated_disk_barrier_failures(
3271 struct btrfs_fs_info *fs_info)
3272{
3273 struct btrfs_ioctl_space_info space;
3274 struct btrfs_space_info *sinfo;
3275 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3276 BTRFS_BLOCK_GROUP_SYSTEM,
3277 BTRFS_BLOCK_GROUP_METADATA,
3278 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3279 int num_types = 4;
3280 int i;
3281 int c;
3282 int num_tolerated_disk_barrier_failures =
3283 (int)fs_info->fs_devices->num_devices;
3284
3285 for (i = 0; i < num_types; i++) {
3286 struct btrfs_space_info *tmp;
3287
3288 sinfo = NULL;
3289 rcu_read_lock();
3290 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3291 if (tmp->flags == types[i]) {
3292 sinfo = tmp;
3293 break;
3294 }
3295 }
3296 rcu_read_unlock();
3297
3298 if (!sinfo)
3299 continue;
3300
3301 down_read(&sinfo->groups_sem);
3302 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3303 if (!list_empty(&sinfo->block_groups[c])) {
3304 u64 flags;
3305
3306 btrfs_get_block_group_info(
3307 &sinfo->block_groups[c], &space);
3308 if (space.total_bytes == 0 ||
3309 space.used_bytes == 0)
3310 continue;
3311 flags = space.flags;
3312 /*
3313 * return
3314 * 0: if dup, single or RAID0 is configured for
3315 * any of metadata, system or data, else
3316 * 1: if RAID5 is configured, or if RAID1 or
3317 * RAID10 is configured and only two mirrors
3318 * are used, else
3319 * 2: if RAID6 is configured, else
3320 * num_mirrors - 1: if RAID1 or RAID10 is
3321 * configured and more than
3322 * 2 mirrors are used.
3323 */
3324 if (num_tolerated_disk_barrier_failures > 0 &&
3325 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3326 BTRFS_BLOCK_GROUP_RAID0)) ||
3327 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3328 == 0)))
3329 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3330 else if (num_tolerated_disk_barrier_failures > 1) {
3331 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3332 BTRFS_BLOCK_GROUP_RAID5 |
3333 BTRFS_BLOCK_GROUP_RAID10)) {
3334 num_tolerated_disk_barrier_failures = 1;
3335 } else if (flags &
15b0a89d 3336 BTRFS_BLOCK_GROUP_RAID6) {
53b381b3
DW		 3337 num_tolerated_disk_barrier_failures = 2;
3338 }
3339 }
5af3e8cc
SB		 3340 }
3341 }
3342 up_read(&sinfo->groups_sem);
3343 }
3344
3345 return num_tolerated_disk_barrier_failures;
3346}
3347
48a3b636 3348static int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3349{
e5e9a520 3350 struct list_head *head;
f2984462 3351 struct btrfs_device *dev;
a061fc8d 3352 struct btrfs_super_block *sb;
f2984462 3353 struct btrfs_dev_item *dev_item;
f2984462
CM		 3354 int ret;
3355 int do_barriers;
a236aed1
CM		 3356 int max_errors;
3357 int total_errors = 0;
a061fc8d 3358 u64 flags;
f2984462 3359
6c41761f 3360 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
f2984462 3361 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3362 backup_super_roots(root->fs_info);
f2984462 3363
6c41761f 3364 sb = root->fs_info->super_for_commit;
a061fc8d 3365 dev_item = &sb->dev_item;
e5e9a520 3366
174ba509 3367 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3368 head = &root->fs_info->fs_devices->devices;
387125fc 3369
5af3e8cc
SB		 3370 if (do_barriers) {
3371 ret = barrier_all_devices(root->fs_info);
3372 if (ret) {
3373 mutex_unlock(
3374 &root->fs_info->fs_devices->device_list_mutex);
3375 btrfs_error(root->fs_info, ret,
3376 "errors while submitting device barriers.");
3377 return ret;
3378 }
3379 }
387125fc 3380
1f78160c 3381 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3382 if (!dev->bdev) {
3383 total_errors++;
3384 continue;
3385 }
2b82032c 3386 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3387 continue;
3388
2b82032c 3389 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3390 btrfs_set_stack_device_type(dev_item, dev->type);
3391 btrfs_set_stack_device_id(dev_item, dev->devid);
3392 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3393 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3394 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3395 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3396 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3397 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3398 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3399
a061fc8d
CM		 3400 flags = btrfs_super_flags(sb);
3401 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3402
a512bbf8 3403 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3404 if (ret)
3405 total_errors++;
f2984462 3406 }
a236aed1 3407 if (total_errors > max_errors) {
d397712b
CM		 3408 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3409 total_errors);
79787eaa
JM		 3410
3411 /* This shouldn't happen. FUA is masked off if unsupported */
a236aed1
CM		 3412 BUG();
3413 }
f2984462 3414
a512bbf8 3415 total_errors = 0;
1f78160c 3416 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3417 if (!dev->bdev)
3418 continue;
2b82032c 3419 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3420 continue;
3421
a512bbf8
YZ		 3422 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3423 if (ret)
3424 total_errors++;
f2984462 3425 }
174ba509 3426 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3427 if (total_errors > max_errors) {
79787eaa
JM		 3428 btrfs_error(root->fs_info, -EIO,
3429 "%d errors while writing supers", total_errors);
3430 return -EIO;
a236aed1 3431 }
f2984462
CM		 3432 return 0;
3433}
3434
a512bbf8
YZ		 3435int write_ctree_super(struct btrfs_trans_handle *trans,
3436 struct btrfs_root *root, int max_mirrors)
eb60ceac 3437{
e66f709b 3438 int ret;
5f39d397 3439
a512bbf8 3440 ret = write_all_supers(root, max_mirrors);
5f39d397 3441 return ret;
cfaa7295
CM		 3442}
3443
cb517eab
MX		 3444/* Drop a fs root from the radix tree and free it. */
3445void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3446 struct btrfs_root *root)
2619ba1f 3447{
4df27c4d 3448 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3449 radix_tree_delete(&fs_info->fs_roots_radix,
3450 (unsigned long)root->root_key.objectid);
4df27c4d 3451 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3452
3453 if (btrfs_root_refs(&root->root_item) == 0)
3454 synchronize_srcu(&fs_info->subvol_srcu);
3455
d7634482 3456 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3321719e
LB		 3457 btrfs_free_log(NULL, root);
3458 btrfs_free_log_root_tree(NULL, fs_info);
3459 }
3460
581bb050
LZ		 3461 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3462 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3463 free_fs_root(root);
4df27c4d
YZ		 3464}
3465
3466static void free_fs_root(struct btrfs_root *root)
3467{
82d5902d 3468 iput(root->cache_inode);
4df27c4d 3469 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
0ee5dc67
AV		 3470 if (root->anon_dev)
3471 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 3472 free_extent_buffer(root->node);
3473 free_extent_buffer(root->commit_root);
581bb050
LZ		 3474 kfree(root->free_ino_ctl);
3475 kfree(root->free_ino_pinned);
d397712b 3476 kfree(root->name);
b0feb9d9 3477 btrfs_put_fs_root(root);
2619ba1f
CM		 3478}
3479
cb517eab
MX		 3480void btrfs_free_fs_root(struct btrfs_root *root)
3481{
3482 free_fs_root(root);
2619ba1f
CM		 3483}
3484
c146afad 3485int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3486{
c146afad
YZ		 3487 u64 root_objectid = 0;
3488 struct btrfs_root *gang[8];
3489 int i;
3768f368 3490 int ret;
e089f05c 3491
c146afad
YZ		 3492 while (1) {
3493 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3494 (void **)gang, root_objectid,
3495 ARRAY_SIZE(gang));
3496 if (!ret)
3497 break;
5d4f98a2
YZ		 3498
3499 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 3500 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 3501 int err;
3502
c146afad 3503 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3504 err = btrfs_orphan_cleanup(gang[i]);
3505 if (err)
3506 return err;
c146afad
YZ		 3507 }
3508 root_objectid++;
3509 }
3510 return 0;
3511}
a2135011 3512
c146afad
YZ		 3513int btrfs_commit_super(struct btrfs_root *root)
3514{
3515 struct btrfs_trans_handle *trans;
3516 int ret;
a74a4b97 3517
c146afad 3518 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3519 btrfs_run_delayed_iputs(root);
c146afad 3520 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3521 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3522
3523 /* wait until ongoing cleanup work done */
3524 down_write(&root->fs_info->cleanup_work_sem);
3525 up_write(&root->fs_info->cleanup_work_sem);
3526
7a7eaa40 3527 trans = btrfs_join_transaction(root);
3612b495
TI		 3528 if (IS_ERR(trans))
3529 return PTR_ERR(trans);
54aa1f4d 3530 ret = btrfs_commit_transaction(trans, root);
79787eaa
JM		 3531 if (ret)
3532 return ret;
c146afad 3533 /* run commit again to drop the original snapshot */
7a7eaa40 3534 trans = btrfs_join_transaction(root);
3612b495
TI		 3535 if (IS_ERR(trans))
3536 return PTR_ERR(trans);
79787eaa
JM		 3537 ret = btrfs_commit_transaction(trans, root);
3538 if (ret)
3539 return ret;
79154b1b 3540 ret = btrfs_write_and_wait_transaction(NULL, root);
79787eaa
JM		 3541 if (ret) {
3542 btrfs_error(root->fs_info, ret,
3543 "Failed to sync btree inode to disk.");
3544 return ret;
3545 }
d6bfde87 3546
a512bbf8 3547 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 3548 return ret;
3549}
3550
3551int close_ctree(struct btrfs_root *root)
3552{
3553 struct btrfs_fs_info *fs_info = root->fs_info;
3554 int ret;
3555
3556 fs_info->closing = 1;
3557 smp_mb();
3558
803b2f54
SB		 3559 /* wait for the uuid_scan task to finish */
3560 down(&fs_info->uuid_tree_rescan_sem);
3561 /* avoid complains from lockdep et al., set sem back to initial state */
3562 up(&fs_info->uuid_tree_rescan_sem);
3563
837d5b6e 3564 /* pause restriper - we want to resume on mount */
aa1b8cd4 3565 btrfs_pause_balance(fs_info);
837d5b6e 3566
8dabb742
SB		 3567 btrfs_dev_replace_suspend_for_unmount(fs_info);
3568
aa1b8cd4 3569 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3570
3571 /* wait for any defraggers to finish */
3572 wait_event(fs_info->transaction_wait,
3573 (atomic_read(&fs_info->defrag_running) == 0));
3574
3575 /* clear out the rbtree of defraggable inodes */
26176e7c 3576 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3577
c146afad 3578 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3579 ret = btrfs_commit_super(root);
3580 if (ret)
3581 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3582 }
3583
87533c47 3584 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3585 btrfs_error_commit_super(root);
0f7d52f4 3586
300e4f8a
JB		 3587 btrfs_put_block_group_cache(fs_info);
3588
e3029d9f
AV		 3589 kthread_stop(fs_info->transaction_kthread);
3590 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3591
f25784b3
YZ		 3592 fs_info->closing = 2;
3593 smp_mb();
3594
bcef60f2
AJ		 3595 btrfs_free_qgroup_config(root->fs_info);
3596
963d678b
MX		 3597 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3598 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3599 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3600 }
bcc63abb 3601
e3029d9f 3602 btrfs_free_block_groups(fs_info);
d10c5f31 3603
13e6c37b 3604 btrfs_stop_all_workers(fs_info);
2932505a 3605
c146afad 3606 del_fs_roots(fs_info);
d10c5f31 3607
13e6c37b 3608 free_root_pointers(fs_info, 1);
9ad6b7bc 3609
13e6c37b 3610 iput(fs_info->btree_inode);
d6bfde87 3611
21adbd5c
SB		 3612#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3613 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3614 btrfsic_unmount(root, fs_info->fs_devices);
3615#endif
3616
dfe25020 3617 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3618 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3619
e2d84521 3620 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3621 percpu_counter_destroy(&fs_info->delalloc_bytes);
04160088 3622 bdi_destroy(&fs_info->bdi);
76dda93c 3623 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3624
53b381b3
DW		 3625 btrfs_free_stripe_hash_table(fs_info);
3626
eb60ceac
CM		 3627 return 0;
3628}
3629
b9fab919
CM		 3630int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3631 int atomic)
5f39d397 3632{
1259ab75 3633 int ret;
727011e0 3634 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3635
0b32f4bb 3636 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3637 if (!ret)
3638 return ret;
3639
3640 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3641 parent_transid, atomic);
3642 if (ret == -EAGAIN)
3643 return ret;
1259ab75 3644 return !ret;
5f39d397
CM		 3645}
3646
3647int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3648{
0b32f4bb 3649 return set_extent_buffer_uptodate(buf);
5f39d397 3650}
6702ed49 3651
5f39d397
CM		 3652void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3653{
727011e0 3654 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
5f39d397 3655 u64 transid = btrfs_header_generation(buf);
b9473439 3656 int was_dirty;
b4ce94de 3657
b9447ef8 3658 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3659 if (transid != root->fs_info->generation)
3660 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3661 "found %llu running %llu\n",
db94535d 3662 (unsigned long long)buf->start,
d397712b
CM		 3663 (unsigned long long)transid,
3664 (unsigned long long)root->fs_info->generation);
0b32f4bb 3665 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3666 if (!was_dirty)
3667 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3668 buf->len,
3669 root->fs_info->dirty_metadata_batch);
eb60ceac
CM		 3670}
3671
b53d3f5d
LB		 3672static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3673 int flush_delayed)
16cdcec7
MX		 3674{
3675 /*
3676 * looks as though older kernels can get into trouble with
3677 * this code, they end up stuck in balance_dirty_pages forever
3678 */
e2d84521 3679 int ret;
16cdcec7
MX		 3680
3681 if (current->flags & PF_MEMALLOC)
3682 return;
3683
b53d3f5d
LB		 3684 if (flush_delayed)
3685 btrfs_balance_delayed_items(root);
16cdcec7 3686
e2d84521
MX		 3687 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3688 BTRFS_DIRTY_METADATA_THRESH);
3689 if (ret > 0) {
d0e1d66b
NJ		 3690 balance_dirty_pages_ratelimited(
3691 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3692 }
3693 return;
3694}
3695
b53d3f5d 3696void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3697{
b53d3f5d
LB		 3698 __btrfs_btree_balance_dirty(root, 1);
3699}
585ad2c3 3700
b53d3f5d
LB		 3701void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3702{
3703 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3704}
6b80053d 3705
ca7a79ad 3706int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3707{
727011e0 3708 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3709 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3710}
0da5468f 3711
fcd1f065 3712static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3713 int read_only)
3714{
1104a885
DS		 3715 /*
3716 * Placeholder for checks
3717 */
fcd1f065 3718 return 0;
acce952b 3719}
3720
48a3b636 3721static void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3722{
acce952b 3723 mutex_lock(&root->fs_info->cleaner_mutex);
3724 btrfs_run_delayed_iputs(root);
3725 mutex_unlock(&root->fs_info->cleaner_mutex);
3726
3727 down_write(&root->fs_info->cleanup_work_sem);
3728 up_write(&root->fs_info->cleanup_work_sem);
3729
3730 /* cleanup FS via transaction */
3731 btrfs_cleanup_transaction(root);
acce952b 3732}
3733
569e0f35
JB		 3734static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3735 struct btrfs_root *root)
acce952b 3736{
3737 struct btrfs_inode *btrfs_inode;
3738 struct list_head splice;
3739
3740 INIT_LIST_HEAD(&splice);
3741
3742 mutex_lock(&root->fs_info->ordered_operations_mutex);
199c2a9c 3743 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3744
569e0f35 3745 list_splice_init(&t->ordered_operations, &splice);
acce952b 3746 while (!list_empty(&splice)) {
3747 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3748 ordered_operations);
3749
3750 list_del_init(&btrfs_inode->ordered_operations);
199c2a9c 3751 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3752
3753 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3754
199c2a9c 3755 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3756 }
3757
199c2a9c 3758 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3759 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3760}
3761
143bede5 3762static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3763{
acce952b 3764 struct btrfs_ordered_extent *ordered;
acce952b 3765
199c2a9c 3766 spin_lock(&root->ordered_extent_lock);
779880ef
JB		 3767 /*
3768 * This will just short circuit the ordered completion stuff which will
3769 * make sure the ordered extent gets properly cleaned up.
3770 */
199c2a9c 3771 list_for_each_entry(ordered, &root->ordered_extents,
779880ef
JB		 3772 root_extent_list)
3773 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
199c2a9c
MX		 3774 spin_unlock(&root->ordered_extent_lock);
3775}
3776
3777static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3778{
3779 struct btrfs_root *root;
3780 struct list_head splice;
3781
3782 INIT_LIST_HEAD(&splice);
3783
3784 spin_lock(&fs_info->ordered_root_lock);
3785 list_splice_init(&fs_info->ordered_roots, &splice);
3786 while (!list_empty(&splice)) {
3787 root = list_first_entry(&splice, struct btrfs_root,
3788 ordered_root);
3789 list_del_init(&root->ordered_root);
3790
3791 btrfs_destroy_ordered_extents(root);
3792
3793 cond_resched_lock(&fs_info->ordered_root_lock);
3794 }
3795 spin_unlock(&fs_info->ordered_root_lock);
acce952b 3796}
3797
35a3621b
SB		 3798static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3799 struct btrfs_root *root)
acce952b 3800{
3801 struct rb_node *node;
3802 struct btrfs_delayed_ref_root *delayed_refs;
3803 struct btrfs_delayed_ref_node *ref;
3804 int ret = 0;
3805
3806 delayed_refs = &trans->delayed_refs;
3807
3808 spin_lock(&delayed_refs->lock);
3809 if (delayed_refs->num_entries == 0) {
cfece4db 3810 spin_unlock(&delayed_refs->lock);
acce952b 3811 printk(KERN_INFO "delayed_refs has NO entry\n");
3812 return ret;
3813 }
3814
b939d1ab 3815 while ((node = rb_first(&delayed_refs->root)) != NULL) {
eb12db69 3816 struct btrfs_delayed_ref_head *head = NULL;
e78417d1 3817 bool pin_bytes = false;
acce952b 3818
eb12db69 3819 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
acce952b 3820 atomic_set(&ref->refs, 1);
3821 if (btrfs_delayed_ref_is_head(ref)) {
acce952b 3822
3823 head = btrfs_delayed_node_to_head(ref);
b939d1ab
JB		 3824 if (!mutex_trylock(&head->mutex)) {
3825 atomic_inc(&ref->refs);
3826 spin_unlock(&delayed_refs->lock);
3827
3828 /* Need to wait for the delayed ref to run */
3829 mutex_lock(&head->mutex);
3830 mutex_unlock(&head->mutex);
3831 btrfs_put_delayed_ref(ref);
3832
e18fca73 3833 spin_lock(&delayed_refs->lock);
b939d1ab
JB		 3834 continue;
3835 }
3836
54067ae9 3837 if (head->must_insert_reserved)
e78417d1 3838 pin_bytes = true;
78a6184a 3839 btrfs_free_delayed_extent_op(head->extent_op);
acce952b 3840 delayed_refs->num_heads--;
3841 if (list_empty(&head->cluster))
3842 delayed_refs->num_heads_ready--;
3843 list_del_init(&head->cluster);
acce952b 3844 }
eb12db69 3845
b939d1ab
JB		 3846 ref->in_tree = 0;
3847 rb_erase(&ref->rb_node, &delayed_refs->root);
3848 delayed_refs->num_entries--;
acce952b 3849 spin_unlock(&delayed_refs->lock);
e78417d1
JB		 3850 if (head) {
3851 if (pin_bytes)
3852 btrfs_pin_extent(root, ref->bytenr,
3853 ref->num_bytes, 1);
3854 mutex_unlock(&head->mutex);
3855 }
acce952b 3856 btrfs_put_delayed_ref(ref);
3857
3858 cond_resched();
3859 spin_lock(&delayed_refs->lock);
3860 }
3861
3862 spin_unlock(&delayed_refs->lock);
3863
3864 return ret;
3865}
3866
aec8030a 3867static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
acce952b 3868{
3869 struct btrfs_pending_snapshot *snapshot;
3870 struct list_head splice;
3871
3872 INIT_LIST_HEAD(&splice);
3873
3874 list_splice_init(&t->pending_snapshots, &splice);
3875
3876 while (!list_empty(&splice)) {
3877 snapshot = list_entry(splice.next,
3878 struct btrfs_pending_snapshot,
3879 list);
aec8030a 3880 snapshot->error = -ECANCELED;
acce952b 3881 list_del_init(&snapshot->list);
acce952b 3882 }
acce952b 3883}
3884
143bede5 3885static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3886{
3887 struct btrfs_inode *btrfs_inode;
3888 struct list_head splice;
3889
3890 INIT_LIST_HEAD(&splice);
3891
eb73c1b7
MX		 3892 spin_lock(&root->delalloc_lock);
3893 list_splice_init(&root->delalloc_inodes, &splice);
acce952b 3894
3895 while (!list_empty(&splice)) {
eb73c1b7
MX		 3896 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3897 delalloc_inodes);
acce952b 3898
3899 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3900 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3901 &btrfs_inode->runtime_flags);
eb73c1b7 3902 spin_unlock(&root->delalloc_lock);
acce952b 3903
3904 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3905
eb73c1b7 3906 spin_lock(&root->delalloc_lock);
acce952b 3907 }
3908
eb73c1b7
MX		 3909 spin_unlock(&root->delalloc_lock);
3910}
3911
3912static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3913{
3914 struct btrfs_root *root;
3915 struct list_head splice;
3916
3917 INIT_LIST_HEAD(&splice);
3918
3919 spin_lock(&fs_info->delalloc_root_lock);
3920 list_splice_init(&fs_info->delalloc_roots, &splice);
3921 while (!list_empty(&splice)) {
3922 root = list_first_entry(&splice, struct btrfs_root,
3923 delalloc_root);
3924 list_del_init(&root->delalloc_root);
3925 root = btrfs_grab_fs_root(root);
3926 BUG_ON(!root);
3927 spin_unlock(&fs_info->delalloc_root_lock);
3928
3929 btrfs_destroy_delalloc_inodes(root);
3930 btrfs_put_fs_root(root);
3931
3932 spin_lock(&fs_info->delalloc_root_lock);
3933 }
3934 spin_unlock(&fs_info->delalloc_root_lock);
acce952b 3935}
3936
3937static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3938 struct extent_io_tree *dirty_pages,
3939 int mark)
3940{
3941 int ret;
acce952b 3942 struct extent_buffer *eb;
3943 u64 start = 0;
3944 u64 end;
acce952b 3945
3946 while (1) {
3947 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3948 mark, NULL);
acce952b 3949 if (ret)
3950 break;
3951
3952 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3953 while (start <= end) {
fd8b2b61
JB		 3954 eb = btrfs_find_tree_block(root, start,
3955 root->leafsize);
69a85bd8 3956 start += root->leafsize;
fd8b2b61 3957 if (!eb)
acce952b 3958 continue;
fd8b2b61 3959 wait_on_extent_buffer_writeback(eb);
acce952b 3960
fd8b2b61
JB		 3961 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3962 &eb->bflags))
3963 clear_extent_buffer_dirty(eb);
3964 free_extent_buffer_stale(eb);
acce952b 3965 }
3966 }
3967
3968 return ret;
3969}
3970
3971static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3972 struct extent_io_tree *pinned_extents)
3973{
3974 struct extent_io_tree *unpin;
3975 u64 start;
3976 u64 end;
3977 int ret;
ed0eaa14 3978 bool loop = true;
acce952b 3979
3980 unpin = pinned_extents;
ed0eaa14 3981again:
acce952b 3982 while (1) {
3983 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 3984 EXTENT_DIRTY, NULL);
acce952b 3985 if (ret)
3986 break;
3987
3988 /* opt_discard */
5378e607
LD		 3989 if (btrfs_test_opt(root, DISCARD))
3990 ret = btrfs_error_discard_extent(root, start,
3991 end + 1 - start,
3992 NULL);
acce952b 3993
3994 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3995 btrfs_error_unpin_extent_range(root, start, end);
3996 cond_resched();
3997 }
3998
ed0eaa14
LB		 3999 if (loop) {
4000 if (unpin == &root->fs_info->freed_extents[0])
4001 unpin = &root->fs_info->freed_extents[1];
4002 else
4003 unpin = &root->fs_info->freed_extents[0];
4004 loop = false;
4005 goto again;
4006 }
4007
acce952b 4008 return 0;
4009}
4010
49b25e05
JM		 4011void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4012 struct btrfs_root *root)
4013{
4014 btrfs_destroy_delayed_refs(cur_trans, root);
4015 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
4016 cur_trans->dirty_pages.dirty_bytes);
4017
4a9d8bde 4018 cur_trans->state = TRANS_STATE_COMMIT_START;
d7096fc3 4019 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 4020
aec8030a
MX		 4021 btrfs_evict_pending_snapshots(cur_trans);
4022
4a9d8bde 4023 cur_trans->state = TRANS_STATE_UNBLOCKED;
d7096fc3 4024 wake_up(&root->fs_info->transaction_wait);
49b25e05 4025
67cde344
MX		 4026 btrfs_destroy_delayed_inodes(root);
4027 btrfs_assert_delayed_root_empty(root);
49b25e05 4028
49b25e05
JM		 4029 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4030 EXTENT_DIRTY);
6e841e32
LB		 4031 btrfs_destroy_pinned_extent(root,
4032 root->fs_info->pinned_extents);
49b25e05 4033
4a9d8bde
MX		 4034 cur_trans->state =TRANS_STATE_COMPLETED;
4035 wake_up(&cur_trans->commit_wait);
4036
49b25e05
JM		 4037 /*
4038 memset(cur_trans, 0, sizeof(*cur_trans));
4039 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4040 */
4041}
4042
48a3b636 4043static int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 4044{
4045 struct btrfs_transaction *t;
4046 LIST_HEAD(list);
4047
acce952b 4048 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4049
a4abeea4 4050 spin_lock(&root->fs_info->trans_lock);
acce952b 4051 list_splice_init(&root->fs_info->trans_list, &list);
ac673879 4052 root->fs_info->running_transaction = NULL;
a4abeea4
JB		 4053 spin_unlock(&root->fs_info->trans_lock);
4054
acce952b 4055 while (!list_empty(&list)) {
4056 t = list_entry(list.next, struct btrfs_transaction, list);
acce952b 4057
569e0f35 4058 btrfs_destroy_ordered_operations(t, root);
acce952b 4059
199c2a9c 4060 btrfs_destroy_all_ordered_extents(root->fs_info);
acce952b 4061
4062 btrfs_destroy_delayed_refs(t, root);
4063
4a9d8bde
MX		 4064 /*
4065 * FIXME: cleanup wait for commit
4066 * We needn't acquire the lock here, because we are during
4067 * the umount, there is no other task which will change it.
4068 */
4069 t->state = TRANS_STATE_COMMIT_START;
66657b31 4070 smp_mb();
acce952b 4071 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4072 wake_up(&root->fs_info->transaction_blocked_wait);
4073
aec8030a
MX		 4074 btrfs_evict_pending_snapshots(t);
4075
4a9d8bde 4076 t->state = TRANS_STATE_UNBLOCKED;
66657b31 4077 smp_mb();
acce952b 4078 if (waitqueue_active(&root->fs_info->transaction_wait))
4079 wake_up(&root->fs_info->transaction_wait);
acce952b 4080
67cde344
MX		 4081 btrfs_destroy_delayed_inodes(root);
4082 btrfs_assert_delayed_root_empty(root);
4083
eb73c1b7 4084 btrfs_destroy_all_delalloc_inodes(root->fs_info);
acce952b 4085
4086 btrfs_destroy_marked_extents(root, &t->dirty_pages,
4087 EXTENT_DIRTY);
4088
4089 btrfs_destroy_pinned_extent(root,
4090 root->fs_info->pinned_extents);
4091
4a9d8bde
MX		 4092 t->state = TRANS_STATE_COMPLETED;
4093 smp_mb();
4094 if (waitqueue_active(&t->commit_wait))
4095 wake_up(&t->commit_wait);
4096
13c5a93e 4097 atomic_set(&t->use_count, 0);
acce952b 4098 list_del_init(&t->list);
4099 memset(t, 0, sizeof(*t));
4100 kmem_cache_free(btrfs_transaction_cachep, t);
4101 }
4102
4103 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 4104
4105 return 0;
4106}
4107
d1310b2e 4108static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 4109 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 4110 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 4111 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 4112 /* note we're sharing with inode.c for the merge bio hook */
4113 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 4114};
Linux 6.x block layer and io_uring tree(s)