projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: add tests for btrfs_get_extent
[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);
143bede5 67static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
acce952b 68static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 struct extent_io_tree *dirty_pages,
70 int mark);
71static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 struct extent_io_tree *pinned_extents);
48a3b636
ES		 73static int btrfs_cleanup_transaction(struct btrfs_root *root);
74static void btrfs_error_commit_super(struct btrfs_root *root);
ce9adaa5 75
d352ac68
CM		 76/*
77 * end_io_wq structs are used to do processing in task context when an IO is
78 * complete. This is used during reads to verify checksums, and it is used
79 * by writes to insert metadata for new file extents after IO is complete.
80 */
ce9adaa5
CM		 81struct end_io_wq {
82 struct bio *bio;
83 bio_end_io_t *end_io;
84 void *private;
85 struct btrfs_fs_info *info;
86 int error;
22c59948 87 int metadata;
ce9adaa5 88 struct list_head list;
8b712842 89 struct btrfs_work work;
ce9adaa5 90};
0da5468f 91
d352ac68
CM		 92/*
93 * async submit bios are used to offload expensive checksumming
94 * onto the worker threads. They checksum file and metadata bios
95 * just before they are sent down the IO stack.
96 */
44b8bd7e
CM		 97struct async_submit_bio {
98 struct inode *inode;
99 struct bio *bio;
100 struct list_head list;
4a69a410
CM		 101 extent_submit_bio_hook_t *submit_bio_start;
102 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 103 int rw;
104 int mirror_num;
c8b97818 105 unsigned long bio_flags;
eaf25d93
CM		 106 /*
107 * bio_offset is optional, can be used if the pages in the bio
108 * can't tell us where in the file the bio should go
109 */
110 u64 bio_offset;
8b712842 111 struct btrfs_work work;
79787eaa 112 int error;
44b8bd7e
CM		 113};
114
85d4e461
CM		 115/*
116 * Lockdep class keys for extent_buffer->lock's in this root. For a given
117 * eb, the lockdep key is determined by the btrfs_root it belongs to and
118 * the level the eb occupies in the tree.
119 *
120 * Different roots are used for different purposes and may nest inside each
121 * other and they require separate keysets. As lockdep keys should be
122 * static, assign keysets according to the purpose of the root as indicated
123 * by btrfs_root->objectid. This ensures that all special purpose roots
124 * have separate keysets.
4008c04a 125 *
85d4e461
CM		 126 * Lock-nesting across peer nodes is always done with the immediate parent
127 * node locked thus preventing deadlock. As lockdep doesn't know this, use
128 * subclass to avoid triggering lockdep warning in such cases.
4008c04a 129 *
85d4e461
CM		 130 * The key is set by the readpage_end_io_hook after the buffer has passed
131 * csum validation but before the pages are unlocked. It is also set by
132 * btrfs_init_new_buffer on freshly allocated blocks.
4008c04a 133 *
85d4e461
CM		 134 * We also add a check to make sure the highest level of the tree is the
135 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
136 * needs update as well.
4008c04a
CM		 137 */
138#ifdef CONFIG_DEBUG_LOCK_ALLOC
139# if BTRFS_MAX_LEVEL != 8
140# error
141# endif
85d4e461
CM		 142
143static struct btrfs_lockdep_keyset {
144 u64 id; /* root objectid */
145 const char *name_stem; /* lock name stem */
146 char names[BTRFS_MAX_LEVEL + 1][20];
147 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
148} btrfs_lockdep_keysets[] = {
149 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
150 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
151 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
152 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
153 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
154 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
60b62978 155 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
85d4e461
CM		 156 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
157 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
158 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
13fd8da9 159 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
85d4e461 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",
c1c9ff7c
GU		 306 root->fs_info->sb->s_id, buf->start,
307 val, found, btrfs_header_level(buf));
607d432d
JB		 308 if (result != (char *)&inline_result)
309 kfree(result);
19c00ddc
CM		 310 return 1;
311 }
312 } else {
607d432d 313 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 314 }
607d432d
JB		 315 if (result != (char *)&inline_result)
316 kfree(result);
19c00ddc
CM		 317 return 0;
318}
319
d352ac68
CM		 320/*
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
325 */
1259ab75 326static int verify_parent_transid(struct extent_io_tree *io_tree,
b9fab919
CM		 327 struct extent_buffer *eb, u64 parent_transid,
328 int atomic)
1259ab75 329{
2ac55d41 330 struct extent_state *cached_state = NULL;
1259ab75
CM		 331 int ret;
332
333 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334 return 0;
335
b9fab919
CM		 336 if (atomic)
337 return -EAGAIN;
338
2ac55d41 339 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
d0082371 340 0, &cached_state);
0b32f4bb 341 if (extent_buffer_uptodate(eb) &&
1259ab75
CM		 342 btrfs_header_generation(eb) == parent_transid) {
343 ret = 0;
344 goto out;
345 }
7a36ddec 346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
193f284d 347 "found %llu\n",
c1c9ff7c 348 eb->start, parent_transid, btrfs_header_generation(eb));
1259ab75 349 ret = 1;
0b32f4bb 350 clear_extent_buffer_uptodate(eb);
33958dc6 351out:
2ac55d41
JB		 352 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
353 &cached_state, GFP_NOFS);
1259ab75 354 return ret;
1259ab75
CM		 355}
356
1104a885
DS		 357/*
358 * Return 0 if the superblock checksum type matches the checksum value of that
359 * algorithm. Pass the raw disk superblock data.
360 */
361static int btrfs_check_super_csum(char *raw_disk_sb)
362{
363 struct btrfs_super_block *disk_sb =
364 (struct btrfs_super_block *)raw_disk_sb;
365 u16 csum_type = btrfs_super_csum_type(disk_sb);
366 int ret = 0;
367
368 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
369 u32 crc = ~(u32)0;
370 const int csum_size = sizeof(crc);
371 char result[csum_size];
372
373 /*
374 * The super_block structure does not span the whole
375 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
376 * is filled with zeros and is included in the checkum.
377 */
378 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
379 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
380 btrfs_csum_final(crc, result);
381
382 if (memcmp(raw_disk_sb, result, csum_size))
383 ret = 1;
667e7d94
CM		 384
385 if (ret && btrfs_super_generation(disk_sb) < 10) {
386 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
387 ret = 0;
388 }
1104a885
DS		 389 }
390
391 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
392 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
393 csum_type);
394 ret = 1;
395 }
396
397 return ret;
398}
399
d352ac68
CM		 400/*
401 * helper to read a given tree block, doing retries as required when
402 * the checksums don't match and we have alternate mirrors to try.
403 */
f188591e
CM		 404static int btree_read_extent_buffer_pages(struct btrfs_root *root,
405 struct extent_buffer *eb,
ca7a79ad 406 u64 start, u64 parent_transid)
f188591e
CM		 407{
408 struct extent_io_tree *io_tree;
ea466794 409 int failed = 0;
f188591e
CM		 410 int ret;
411 int num_copies = 0;
412 int mirror_num = 0;
ea466794 413 int failed_mirror = 0;
f188591e 414
a826d6dc 415 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM		 416 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
417 while (1) {
bb82ab88
AJ		 418 ret = read_extent_buffer_pages(io_tree, eb, start,
419 WAIT_COMPLETE,
f188591e 420 btree_get_extent, mirror_num);
256dd1bb
SB		 421 if (!ret) {
422 if (!verify_parent_transid(io_tree, eb,
b9fab919 423 parent_transid, 0))
256dd1bb
SB		 424 break;
425 else
426 ret = -EIO;
427 }
d397712b 428
a826d6dc
JB		 429 /*
430 * This buffer's crc is fine, but its contents are corrupted, so
431 * there is no reason to read the other copies, they won't be
432 * any less wrong.
433 */
434 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
ea466794
JB		 435 break;
436
5d964051 437 num_copies = btrfs_num_copies(root->fs_info,
f188591e 438 eb->start, eb->len);
4235298e 439 if (num_copies == 1)
ea466794 440 break;
4235298e 441
5cf1ab56
JB		 442 if (!failed_mirror) {
443 failed = 1;
444 failed_mirror = eb->read_mirror;
445 }
446
f188591e 447 mirror_num++;
ea466794
JB		 448 if (mirror_num == failed_mirror)
449 mirror_num++;
450
4235298e 451 if (mirror_num > num_copies)
ea466794 452 break;
f188591e 453 }
ea466794 454
c0901581 455 if (failed && !ret && failed_mirror)
ea466794
JB		 456 repair_eb_io_failure(root, eb, failed_mirror);
457
458 return ret;
f188591e 459}
19c00ddc 460
d352ac68 461/*
d397712b
CM		 462 * checksum a dirty tree block before IO. This has extra checks to make sure
463 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 464 */
d397712b 465
b2950863 466static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 467{
d1310b2e 468 struct extent_io_tree *tree;
4eee4fa4 469 u64 start = page_offset(page);
19c00ddc 470 u64 found_start;
19c00ddc 471 struct extent_buffer *eb;
f188591e 472
d1310b2e 473 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc 474
4f2de97a
JB		 475 eb = (struct extent_buffer *)page->private;
476 if (page != eb->pages[0])
477 return 0;
19c00ddc
CM		 478 found_start = btrfs_header_bytenr(eb);
479 if (found_start != start) {
55c69072 480 WARN_ON(1);
4f2de97a 481 return 0;
55c69072 482 }
55c69072 483 if (!PageUptodate(page)) {
55c69072 484 WARN_ON(1);
4f2de97a 485 return 0;
19c00ddc 486 }
19c00ddc 487 csum_tree_block(root, eb, 0);
19c00ddc
CM		 488 return 0;
489}
490
2b82032c
YZ		 491static int check_tree_block_fsid(struct btrfs_root *root,
492 struct extent_buffer *eb)
493{
494 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
495 u8 fsid[BTRFS_UUID_SIZE];
496 int ret = 1;
497
0a4e5586 498 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
2b82032c
YZ		 499 while (fs_devices) {
500 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
501 ret = 0;
502 break;
503 }
504 fs_devices = fs_devices->seed;
505 }
506 return ret;
507}
508
a826d6dc
JB		 509#define CORRUPT(reason, eb, root, slot) \
510 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
511 "root=%llu, slot=%d\n", reason, \
c1c9ff7c 512 btrfs_header_bytenr(eb), root->objectid, slot)
a826d6dc
JB		 513
514static noinline int check_leaf(struct btrfs_root *root,
515 struct extent_buffer *leaf)
516{
517 struct btrfs_key key;
518 struct btrfs_key leaf_key;
519 u32 nritems = btrfs_header_nritems(leaf);
520 int slot;
521
522 if (nritems == 0)
523 return 0;
524
525 /* Check the 0 item */
526 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
527 BTRFS_LEAF_DATA_SIZE(root)) {
528 CORRUPT("invalid item offset size pair", leaf, root, 0);
529 return -EIO;
530 }
531
532 /*
533 * Check to make sure each items keys are in the correct order and their
534 * offsets make sense. We only have to loop through nritems-1 because
535 * we check the current slot against the next slot, which verifies the
536 * next slot's offset+size makes sense and that the current's slot
537 * offset is correct.
538 */
539 for (slot = 0; slot < nritems - 1; slot++) {
540 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
541 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
542
543 /* Make sure the keys are in the right order */
544 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
545 CORRUPT("bad key order", leaf, root, slot);
546 return -EIO;
547 }
548
549 /*
550 * Make sure the offset and ends are right, remember that the
551 * item data starts at the end of the leaf and grows towards the
552 * front.
553 */
554 if (btrfs_item_offset_nr(leaf, slot) !=
555 btrfs_item_end_nr(leaf, slot + 1)) {
556 CORRUPT("slot offset bad", leaf, root, slot);
557 return -EIO;
558 }
559
560 /*
561 * Check to make sure that we don't point outside of the leaf,
562 * just incase all the items are consistent to eachother, but
563 * all point outside of the leaf.
564 */
565 if (btrfs_item_end_nr(leaf, slot) >
566 BTRFS_LEAF_DATA_SIZE(root)) {
567 CORRUPT("slot end outside of leaf", leaf, root, slot);
568 return -EIO;
569 }
570 }
571
572 return 0;
573}
574
facc8a22
MX		 575static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
576 u64 phy_offset, struct page *page,
577 u64 start, u64 end, int mirror)
ce9adaa5
CM		 578{
579 struct extent_io_tree *tree;
580 u64 found_start;
581 int found_level;
ce9adaa5
CM		 582 struct extent_buffer *eb;
583 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 584 int ret = 0;
727011e0 585 int reads_done;
ce9adaa5 586
ce9adaa5
CM		 587 if (!page->private)
588 goto out;
d397712b 589
727011e0 590 tree = &BTRFS_I(page->mapping->host)->io_tree;
4f2de97a 591 eb = (struct extent_buffer *)page->private;
d397712b 592
0b32f4bb
JB		 593 /* the pending IO might have been the only thing that kept this buffer
594 * in memory. Make sure we have a ref for all this other checks
595 */
596 extent_buffer_get(eb);
597
598 reads_done = atomic_dec_and_test(&eb->io_pages);
727011e0
CM		 599 if (!reads_done)
600 goto err;
f188591e 601
5cf1ab56 602 eb->read_mirror = mirror;
ea466794
JB		 603 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
604 ret = -EIO;
605 goto err;
606 }
607
ce9adaa5 608 found_start = btrfs_header_bytenr(eb);
727011e0 609 if (found_start != eb->start) {
7a36ddec 610 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
193f284d 611 "%llu %llu\n",
c1c9ff7c 612 found_start, eb->start);
f188591e 613 ret = -EIO;
ce9adaa5
CM		 614 goto err;
615 }
2b82032c 616 if (check_tree_block_fsid(root, eb)) {
7a36ddec 617 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
c1c9ff7c 618 eb->start);
1259ab75
CM		 619 ret = -EIO;
620 goto err;
621 }
ce9adaa5 622 found_level = btrfs_header_level(eb);
1c24c3ce
JB		 623 if (found_level >= BTRFS_MAX_LEVEL) {
624 btrfs_info(root->fs_info, "bad tree block level %d\n",
625 (int)btrfs_header_level(eb));
626 ret = -EIO;
627 goto err;
628 }
ce9adaa5 629
85d4e461
CM		 630 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
631 eb, found_level);
4008c04a 632
ce9adaa5 633 ret = csum_tree_block(root, eb, 1);
a826d6dc 634 if (ret) {
f188591e 635 ret = -EIO;
a826d6dc
JB		 636 goto err;
637 }
638
639 /*
640 * If this is a leaf block and it is corrupt, set the corrupt bit so
641 * that we don't try and read the other copies of this block, just
642 * return -EIO.
643 */
644 if (found_level == 0 && check_leaf(root, eb)) {
645 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
646 ret = -EIO;
647 }
ce9adaa5 648
0b32f4bb
JB		 649 if (!ret)
650 set_extent_buffer_uptodate(eb);
ce9adaa5 651err:
79fb65a1
JB		 652 if (reads_done &&
653 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 654 btree_readahead_hook(root, eb, eb->start, ret);
4bb31e92 655
53b381b3
DW		 656 if (ret) {
657 /*
658 * our io error hook is going to dec the io pages
659 * again, we have to make sure it has something
660 * to decrement
661 */
662 atomic_inc(&eb->io_pages);
0b32f4bb 663 clear_extent_buffer_uptodate(eb);
53b381b3 664 }
0b32f4bb 665 free_extent_buffer(eb);
ce9adaa5 666out:
f188591e 667 return ret;
ce9adaa5
CM		 668}
669
ea466794 670static int btree_io_failed_hook(struct page *page, int failed_mirror)
4bb31e92 671{
4bb31e92
AJ		 672 struct extent_buffer *eb;
673 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
674
4f2de97a 675 eb = (struct extent_buffer *)page->private;
ea466794 676 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
5cf1ab56 677 eb->read_mirror = failed_mirror;
53b381b3 678 atomic_dec(&eb->io_pages);
ea466794 679 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 680 btree_readahead_hook(root, eb, eb->start, -EIO);
4bb31e92
AJ		 681 return -EIO; /* we fixed nothing */
682}
683
ce9adaa5 684static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 685{
686 struct end_io_wq *end_io_wq = bio->bi_private;
687 struct btrfs_fs_info *fs_info;
ce9adaa5 688
ce9adaa5 689 fs_info = end_io_wq->info;
ce9adaa5 690 end_io_wq->error = err;
8b712842
CM		 691 end_io_wq->work.func = end_workqueue_fn;
692 end_io_wq->work.flags = 0;
d20f7043 693
7b6d91da 694 if (bio->bi_rw & REQ_WRITE) {
53b381b3 695 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
cad321ad
CM		 696 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
697 &end_io_wq->work);
53b381b3 698 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
0cb59c99
JB		 699 btrfs_queue_worker(&fs_info->endio_freespace_worker,
700 &end_io_wq->work);
53b381b3
DW		 701 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
702 btrfs_queue_worker(&fs_info->endio_raid56_workers,
703 &end_io_wq->work);
cad321ad
CM		 704 else
705 btrfs_queue_worker(&fs_info->endio_write_workers,
706 &end_io_wq->work);
d20f7043 707 } else {
53b381b3
DW		 708 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
709 btrfs_queue_worker(&fs_info->endio_raid56_workers,
710 &end_io_wq->work);
711 else if (end_io_wq->metadata)
d20f7043
CM		 712 btrfs_queue_worker(&fs_info->endio_meta_workers,
713 &end_io_wq->work);
714 else
715 btrfs_queue_worker(&fs_info->endio_workers,
716 &end_io_wq->work);
717 }
ce9adaa5
CM		 718}
719
0cb59c99
JB		 720/*
721 * For the metadata arg you want
722 *
723 * 0 - if data
724 * 1 - if normal metadta
725 * 2 - if writing to the free space cache area
53b381b3 726 * 3 - raid parity work
0cb59c99 727 */
22c59948
CM		 728int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
729 int metadata)
0b86a832 730{
ce9adaa5 731 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 732 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
733 if (!end_io_wq)
734 return -ENOMEM;
735
736 end_io_wq->private = bio->bi_private;
737 end_io_wq->end_io = bio->bi_end_io;
22c59948 738 end_io_wq->info = info;
ce9adaa5
CM		 739 end_io_wq->error = 0;
740 end_io_wq->bio = bio;
22c59948 741 end_io_wq->metadata = metadata;
ce9adaa5
CM		 742
743 bio->bi_private = end_io_wq;
744 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 745 return 0;
746}
747
b64a2851 748unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 749{
4854ddd0
CM		 750 unsigned long limit = min_t(unsigned long,
751 info->workers.max_workers,
752 info->fs_devices->open_devices);
753 return 256 * limit;
754}
0986fe9e 755
4a69a410
CM		 756static void run_one_async_start(struct btrfs_work *work)
757{
4a69a410 758 struct async_submit_bio *async;
79787eaa 759 int ret;
4a69a410
CM		 760
761 async = container_of(work, struct async_submit_bio, work);
79787eaa
JM		 762 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
763 async->mirror_num, async->bio_flags,
764 async->bio_offset);
765 if (ret)
766 async->error = ret;
4a69a410
CM		 767}
768
769static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 770{
771 struct btrfs_fs_info *fs_info;
772 struct async_submit_bio *async;
4854ddd0 773 int limit;
8b712842
CM		 774
775 async = container_of(work, struct async_submit_bio, work);
776 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 777
b64a2851 778 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 779 limit = limit * 2 / 3;
780
66657b31 781 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
b64a2851 782 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 783 wake_up(&fs_info->async_submit_wait);
784
79787eaa
JM		 785 /* If an error occured we just want to clean up the bio and move on */
786 if (async->error) {
787 bio_endio(async->bio, async->error);
788 return;
789 }
790
4a69a410 791 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 792 async->mirror_num, async->bio_flags,
793 async->bio_offset);
4a69a410
CM		 794}
795
796static void run_one_async_free(struct btrfs_work *work)
797{
798 struct async_submit_bio *async;
799
800 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 801 kfree(async);
802}
803
44b8bd7e
CM		 804int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
805 int rw, struct bio *bio, int mirror_num,
c8b97818 806 unsigned long bio_flags,
eaf25d93 807 u64 bio_offset,
4a69a410
CM		 808 extent_submit_bio_hook_t *submit_bio_start,
809 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 810{
811 struct async_submit_bio *async;
812
813 async = kmalloc(sizeof(*async), GFP_NOFS);
814 if (!async)
815 return -ENOMEM;
816
817 async->inode = inode;
818 async->rw = rw;
819 async->bio = bio;
820 async->mirror_num = mirror_num;
4a69a410
CM		 821 async->submit_bio_start = submit_bio_start;
822 async->submit_bio_done = submit_bio_done;
823
824 async->work.func = run_one_async_start;
825 async->work.ordered_func = run_one_async_done;
826 async->work.ordered_free = run_one_async_free;
827
8b712842 828 async->work.flags = 0;
c8b97818 829 async->bio_flags = bio_flags;
eaf25d93 830 async->bio_offset = bio_offset;
8c8bee1d 831
79787eaa
JM		 832 async->error = 0;
833
cb03c743 834 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 835
7b6d91da 836 if (rw & REQ_SYNC)
d313d7a3
CM		 837 btrfs_set_work_high_prio(&async->work);
838
8b712842 839 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 840
d397712b 841 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 842 atomic_read(&fs_info->nr_async_submits)) {
843 wait_event(fs_info->async_submit_wait,
844 (atomic_read(&fs_info->nr_async_submits) == 0));
845 }
846
44b8bd7e
CM		 847 return 0;
848}
849
ce3ed71a
CM		 850static int btree_csum_one_bio(struct bio *bio)
851{
852 struct bio_vec *bvec = bio->bi_io_vec;
853 int bio_index = 0;
854 struct btrfs_root *root;
79787eaa 855 int ret = 0;
ce3ed71a
CM		 856
857 WARN_ON(bio->bi_vcnt <= 0);
d397712b 858 while (bio_index < bio->bi_vcnt) {
ce3ed71a 859 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 860 ret = csum_dirty_buffer(root, bvec->bv_page);
861 if (ret)
862 break;
ce3ed71a
CM		 863 bio_index++;
864 bvec++;
865 }
79787eaa 866 return ret;
ce3ed71a
CM		 867}
868
4a69a410
CM		 869static int __btree_submit_bio_start(struct inode *inode, int rw,
870 struct bio *bio, int mirror_num,
eaf25d93
CM		 871 unsigned long bio_flags,
872 u64 bio_offset)
22c59948 873{
8b712842
CM		 874 /*
875 * when we're called for a write, we're already in the async
5443be45 876 * submission context. Just jump into btrfs_map_bio
8b712842 877 */
79787eaa 878 return btree_csum_one_bio(bio);
4a69a410 879}
22c59948 880
4a69a410 881static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 882 int mirror_num, unsigned long bio_flags,
883 u64 bio_offset)
4a69a410 884{
61891923
SB		 885 int ret;
886
8b712842 887 /*
4a69a410
CM		 888 * when we're called for a write, we're already in the async
889 * submission context. Just jump into btrfs_map_bio
8b712842 890 */
61891923
SB		 891 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
892 if (ret)
893 bio_endio(bio, ret);
894 return ret;
0b86a832
CM		 895}
896
de0022b9
JB		 897static int check_async_write(struct inode *inode, unsigned long bio_flags)
898{
899 if (bio_flags & EXTENT_BIO_TREE_LOG)
900 return 0;
901#ifdef CONFIG_X86
902 if (cpu_has_xmm4_2)
903 return 0;
904#endif
905 return 1;
906}
907
44b8bd7e 908static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 909 int mirror_num, unsigned long bio_flags,
910 u64 bio_offset)
44b8bd7e 911{
de0022b9 912 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 913 int ret;
914
7b6d91da 915 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 916 /*
917 * called for a read, do the setup so that checksum validation
918 * can happen in the async kernel threads
919 */
f3f266ab
CM		 920 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
921 bio, 1);
1d4284bd 922 if (ret)
61891923
SB		 923 goto out_w_error;
924 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
925 mirror_num, 0);
de0022b9
JB		 926 } else if (!async) {
927 ret = btree_csum_one_bio(bio);
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);
932 } else {
933 /*
934 * kthread helpers are used to submit writes so that
935 * checksumming can happen in parallel across all CPUs
936 */
937 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
938 inode, rw, bio, mirror_num, 0,
939 bio_offset,
940 __btree_submit_bio_start,
941 __btree_submit_bio_done);
44b8bd7e 942 }
d313d7a3 943
61891923
SB		 944 if (ret) {
945out_w_error:
946 bio_endio(bio, ret);
947 }
948 return ret;
44b8bd7e
CM		 949}
950
3dd1462e 951#ifdef CONFIG_MIGRATION
784b4e29 952static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 953 struct page *newpage, struct page *page,
954 enum migrate_mode mode)
784b4e29
CM		 955{
956 /*
957 * we can't safely write a btree page from here,
958 * we haven't done the locking hook
959 */
960 if (PageDirty(page))
961 return -EAGAIN;
962 /*
963 * Buffers may be managed in a filesystem specific way.
964 * We must have no buffers or drop them.
965 */
966 if (page_has_private(page) &&
967 !try_to_release_page(page, GFP_KERNEL))
968 return -EAGAIN;
a6bc32b8 969 return migrate_page(mapping, newpage, page, mode);
784b4e29 970}
3dd1462e 971#endif
784b4e29 972
0da5468f
CM		 973
974static int btree_writepages(struct address_space *mapping,
975 struct writeback_control *wbc)
976{
d1310b2e 977 struct extent_io_tree *tree;
e2d84521
MX		 978 struct btrfs_fs_info *fs_info;
979 int ret;
980
d1310b2e 981 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 982 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 983
984 if (wbc->for_kupdate)
985 return 0;
986
e2d84521 987 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 988 /* this is a bit racy, but that's ok */
e2d84521
MX		 989 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
990 BTRFS_DIRTY_METADATA_THRESH);
991 if (ret < 0)
793955bc 992 return 0;
793955bc 993 }
0b32f4bb 994 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 995}
996
b2950863 997static int btree_readpage(struct file *file, struct page *page)
5f39d397 998{
d1310b2e
CM		 999 struct extent_io_tree *tree;
1000 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 1001 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 1002}
22b0ebda 1003
70dec807 1004static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 1005{
98509cfc 1006 if (PageWriteback(page) || PageDirty(page))
d397712b 1007 return 0;
0c4e538b 1008
f7a52a40 1009 return try_release_extent_buffer(page);
d98237b3
CM		 1010}
1011
d47992f8
LC		 1012static void btree_invalidatepage(struct page *page, unsigned int offset,
1013 unsigned int length)
d98237b3 1014{
d1310b2e
CM		 1015 struct extent_io_tree *tree;
1016 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1017 extent_invalidatepage(tree, page, offset);
1018 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1019 if (PagePrivate(page)) {
d397712b
CM		 1020 printk(KERN_WARNING "btrfs warning page private not zero "
1021 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1022 ClearPagePrivate(page);
1023 set_page_private(page, 0);
1024 page_cache_release(page);
1025 }
d98237b3
CM		 1026}
1027
0b32f4bb
JB		 1028static int btree_set_page_dirty(struct page *page)
1029{
bb146eb2 1030#ifdef DEBUG
0b32f4bb
JB		 1031 struct extent_buffer *eb;
1032
1033 BUG_ON(!PagePrivate(page));
1034 eb = (struct extent_buffer *)page->private;
1035 BUG_ON(!eb);
1036 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1037 BUG_ON(!atomic_read(&eb->refs));
1038 btrfs_assert_tree_locked(eb);
bb146eb2 1039#endif
0b32f4bb
JB		 1040 return __set_page_dirty_nobuffers(page);
1041}
1042
7f09410b 1043static const struct address_space_operations btree_aops = {
d98237b3 1044 .readpage = btree_readpage,
0da5468f 1045 .writepages = btree_writepages,
5f39d397
CM		 1046 .releasepage = btree_releasepage,
1047 .invalidatepage = btree_invalidatepage,
5a92bc88 1048#ifdef CONFIG_MIGRATION
784b4e29 1049 .migratepage = btree_migratepage,
5a92bc88 1050#endif
0b32f4bb 1051 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1052};
1053
ca7a79ad
CM		 1054int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1055 u64 parent_transid)
090d1875 1056{
5f39d397
CM		 1057 struct extent_buffer *buf = NULL;
1058 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1059 int ret = 0;
090d1875 1060
db94535d 1061 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1062 if (!buf)
090d1875 1063 return 0;
d1310b2e 1064 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1065 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1066 free_extent_buffer(buf);
de428b63 1067 return ret;
090d1875
CM		 1068}
1069
ab0fff03
AJ		 1070int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1071 int mirror_num, struct extent_buffer **eb)
1072{
1073 struct extent_buffer *buf = NULL;
1074 struct inode *btree_inode = root->fs_info->btree_inode;
1075 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1076 int ret;
1077
1078 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1079 if (!buf)
1080 return 0;
1081
1082 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1083
1084 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1085 btree_get_extent, mirror_num);
1086 if (ret) {
1087 free_extent_buffer(buf);
1088 return ret;
1089 }
1090
1091 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1092 free_extent_buffer(buf);
1093 return -EIO;
0b32f4bb 1094 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1095 *eb = buf;
1096 } else {
1097 free_extent_buffer(buf);
1098 }
1099 return 0;
1100}
1101
0999df54
CM		 1102struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1103 u64 bytenr, u32 blocksize)
1104{
1105 struct inode *btree_inode = root->fs_info->btree_inode;
1106 struct extent_buffer *eb;
1107 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
f09d1f60 1108 bytenr, blocksize);
0999df54
CM		 1109 return eb;
1110}
1111
1112struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1113 u64 bytenr, u32 blocksize)
1114{
1115 struct inode *btree_inode = root->fs_info->btree_inode;
1116 struct extent_buffer *eb;
1117
1118 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
727011e0 1119 bytenr, blocksize);
0999df54
CM		 1120 return eb;
1121}
1122
1123
e02119d5
CM		 1124int btrfs_write_tree_block(struct extent_buffer *buf)
1125{
727011e0 1126 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
8aa38c31 1127 buf->start + buf->len - 1);
e02119d5
CM		 1128}
1129
1130int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1131{
727011e0 1132 return filemap_fdatawait_range(buf->pages[0]->mapping,
8aa38c31 1133 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1134}
1135
0999df54 1136struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1137 u32 blocksize, u64 parent_transid)
0999df54
CM		 1138{
1139 struct extent_buffer *buf = NULL;
0999df54
CM		 1140 int ret;
1141
0999df54
CM		 1142 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1143 if (!buf)
1144 return NULL;
0999df54 1145
ca7a79ad 1146 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
0f0fe8f7
FDBM		 1147 if (ret) {
1148 free_extent_buffer(buf);
1149 return NULL;
1150 }
5f39d397 1151 return buf;
ce9adaa5 1152
eb60ceac
CM		 1153}
1154
d5c13f92
JM		 1155void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1156 struct extent_buffer *buf)
ed2ff2cb 1157{
e2d84521
MX		 1158 struct btrfs_fs_info *fs_info = root->fs_info;
1159
55c69072 1160 if (btrfs_header_generation(buf) ==
e2d84521 1161 fs_info->running_transaction->transid) {
b9447ef8 1162 btrfs_assert_tree_locked(buf);
b4ce94de 1163
b9473439 1164 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1165 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1166 -buf->len,
1167 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1168 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1169 btrfs_set_lock_blocking(buf);
1170 clear_extent_buffer_dirty(buf);
1171 }
925baedd 1172 }
5f39d397
CM		 1173}
1174
143bede5
JM		 1175static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1176 u32 stripesize, struct btrfs_root *root,
1177 struct btrfs_fs_info *fs_info,
1178 u64 objectid)
d97e63b6 1179{
cfaa7295 1180 root->node = NULL;
a28ec197 1181 root->commit_root = NULL;
db94535d
CM		 1182 root->sectorsize = sectorsize;
1183 root->nodesize = nodesize;
1184 root->leafsize = leafsize;
87ee04eb 1185 root->stripesize = stripesize;
123abc88 1186 root->ref_cows = 0;
0b86a832 1187 root->track_dirty = 0;
c71bf099 1188 root->in_radix = 0;
d68fc57b
YZ		 1189 root->orphan_item_inserted = 0;
1190 root->orphan_cleanup_state = 0;
0b86a832 1191
0f7d52f4
CM		 1192 root->objectid = objectid;
1193 root->last_trans = 0;
13a8a7c8 1194 root->highest_objectid = 0;
eb73c1b7 1195 root->nr_delalloc_inodes = 0;
199c2a9c 1196 root->nr_ordered_extents = 0;
58176a96 1197 root->name = NULL;
6bef4d31 1198 root->inode_tree = RB_ROOT;
16cdcec7 1199 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1200 root->block_rsv = NULL;
d68fc57b 1201 root->orphan_block_rsv = NULL;
0b86a832
CM		 1202
1203 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1204 INIT_LIST_HEAD(&root->root_list);
eb73c1b7
MX		 1205 INIT_LIST_HEAD(&root->delalloc_inodes);
1206 INIT_LIST_HEAD(&root->delalloc_root);
199c2a9c
MX		 1207 INIT_LIST_HEAD(&root->ordered_extents);
1208 INIT_LIST_HEAD(&root->ordered_root);
2ab28f32
JB		 1209 INIT_LIST_HEAD(&root->logged_list[0]);
1210 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1211 spin_lock_init(&root->orphan_lock);
5d4f98a2 1212 spin_lock_init(&root->inode_lock);
eb73c1b7 1213 spin_lock_init(&root->delalloc_lock);
199c2a9c 1214 spin_lock_init(&root->ordered_extent_lock);
f0486c68 1215 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1216 spin_lock_init(&root->log_extents_lock[0]);
1217 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1218 mutex_init(&root->objectid_mutex);
e02119d5 1219 mutex_init(&root->log_mutex);
7237f183
YZ		 1220 init_waitqueue_head(&root->log_writer_wait);
1221 init_waitqueue_head(&root->log_commit_wait[0]);
1222 init_waitqueue_head(&root->log_commit_wait[1]);
1223 atomic_set(&root->log_commit[0], 0);
1224 atomic_set(&root->log_commit[1], 0);
1225 atomic_set(&root->log_writers, 0);
2ecb7923 1226 atomic_set(&root->log_batch, 0);
8a35d95f 1227 atomic_set(&root->orphan_inodes, 0);
b0feb9d9 1228 atomic_set(&root->refs, 1);
7237f183 1229 root->log_transid = 0;
257c62e1 1230 root->last_log_commit = 0;
06ea65a3
JB		 1231 if (fs_info)
1232 extent_io_tree_init(&root->dirty_log_pages,
1233 fs_info->btree_inode->i_mapping);
017e5369 1234
3768f368
CM		 1235 memset(&root->root_key, 0, sizeof(root->root_key));
1236 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1237 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1238 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
06ea65a3
JB		 1239 if (fs_info)
1240 root->defrag_trans_start = fs_info->generation;
1241 else
1242 root->defrag_trans_start = 0;
58176a96 1243 init_completion(&root->kobj_unregister);
6702ed49 1244 root->defrag_running = 0;
4d775673 1245 root->root_key.objectid = objectid;
0ee5dc67 1246 root->anon_dev = 0;
8ea05e3a 1247
5f3ab90a 1248 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1249}
1250
f84a8bd6 1251static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1252{
1253 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1254 if (root)
1255 root->fs_info = fs_info;
1256 return root;
1257}
1258
06ea65a3
JB		 1259#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1260/* Should only be used by the testing infrastructure */
1261struct btrfs_root *btrfs_alloc_dummy_root(void)
1262{
1263 struct btrfs_root *root;
1264
1265 root = btrfs_alloc_root(NULL);
1266 if (!root)
1267 return ERR_PTR(-ENOMEM);
1268 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1269 root->dummy_root = 1;
1270
1271 return root;
1272}
1273#endif
1274
20897f5c
AJ		 1275struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1276 struct btrfs_fs_info *fs_info,
1277 u64 objectid)
1278{
1279 struct extent_buffer *leaf;
1280 struct btrfs_root *tree_root = fs_info->tree_root;
1281 struct btrfs_root *root;
1282 struct btrfs_key key;
1283 int ret = 0;
1284 u64 bytenr;
6463fe58 1285 uuid_le uuid;
20897f5c
AJ		 1286
1287 root = btrfs_alloc_root(fs_info);
1288 if (!root)
1289 return ERR_PTR(-ENOMEM);
1290
1291 __setup_root(tree_root->nodesize, tree_root->leafsize,
1292 tree_root->sectorsize, tree_root->stripesize,
1293 root, fs_info, objectid);
1294 root->root_key.objectid = objectid;
1295 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1296 root->root_key.offset = 0;
1297
1298 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1299 0, objectid, NULL, 0, 0, 0);
1300 if (IS_ERR(leaf)) {
1301 ret = PTR_ERR(leaf);
1dd05682 1302 leaf = NULL;
20897f5c
AJ		 1303 goto fail;
1304 }
1305
1306 bytenr = leaf->start;
1307 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1308 btrfs_set_header_bytenr(leaf, leaf->start);
1309 btrfs_set_header_generation(leaf, trans->transid);
1310 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1311 btrfs_set_header_owner(leaf, objectid);
1312 root->node = leaf;
1313
0a4e5586 1314 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
20897f5c
AJ		 1315 BTRFS_FSID_SIZE);
1316 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
b308bc2f 1317 btrfs_header_chunk_tree_uuid(leaf),
20897f5c
AJ		 1318 BTRFS_UUID_SIZE);
1319 btrfs_mark_buffer_dirty(leaf);
1320
1321 root->commit_root = btrfs_root_node(root);
1322 root->track_dirty = 1;
1323
1324
1325 root->root_item.flags = 0;
1326 root->root_item.byte_limit = 0;
1327 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1328 btrfs_set_root_generation(&root->root_item, trans->transid);
1329 btrfs_set_root_level(&root->root_item, 0);
1330 btrfs_set_root_refs(&root->root_item, 1);
1331 btrfs_set_root_used(&root->root_item, leaf->len);
1332 btrfs_set_root_last_snapshot(&root->root_item, 0);
1333 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1334 uuid_le_gen(&uuid);
1335 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1336 root->root_item.drop_level = 0;
1337
1338 key.objectid = objectid;
1339 key.type = BTRFS_ROOT_ITEM_KEY;
1340 key.offset = 0;
1341 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1342 if (ret)
1343 goto fail;
1344
1345 btrfs_tree_unlock(leaf);
1346
1dd05682
TI		 1347 return root;
1348
20897f5c 1349fail:
1dd05682
TI		 1350 if (leaf) {
1351 btrfs_tree_unlock(leaf);
1352 free_extent_buffer(leaf);
1353 }
1354 kfree(root);
20897f5c 1355
1dd05682 1356 return ERR_PTR(ret);
20897f5c
AJ		 1357}
1358
7237f183
YZ		 1359static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1360 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1361{
1362 struct btrfs_root *root;
1363 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1364 struct extent_buffer *leaf;
e02119d5 1365
6f07e42e 1366 root = btrfs_alloc_root(fs_info);
e02119d5 1367 if (!root)
7237f183 1368 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1369
1370 __setup_root(tree_root->nodesize, tree_root->leafsize,
1371 tree_root->sectorsize, tree_root->stripesize,
1372 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1373
1374 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1375 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1376 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1377 /*
1378 * log trees do not get reference counted because they go away
1379 * before a real commit is actually done. They do store pointers
1380 * to file data extents, and those reference counts still get
1381 * updated (along with back refs to the log tree).
1382 */
e02119d5
CM		 1383 root->ref_cows = 0;
1384
5d4f98a2 1385 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1386 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1387 0, 0, 0);
7237f183
YZ		 1388 if (IS_ERR(leaf)) {
1389 kfree(root);
1390 return ERR_CAST(leaf);
1391 }
e02119d5 1392
5d4f98a2
YZ		 1393 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1394 btrfs_set_header_bytenr(leaf, leaf->start);
1395 btrfs_set_header_generation(leaf, trans->transid);
1396 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1397 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1398 root->node = leaf;
e02119d5
CM		 1399
1400 write_extent_buffer(root->node, root->fs_info->fsid,
0a4e5586 1401 btrfs_header_fsid(), BTRFS_FSID_SIZE);
e02119d5
CM		 1402 btrfs_mark_buffer_dirty(root->node);
1403 btrfs_tree_unlock(root->node);
7237f183
YZ		 1404 return root;
1405}
1406
1407int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1408 struct btrfs_fs_info *fs_info)
1409{
1410 struct btrfs_root *log_root;
1411
1412 log_root = alloc_log_tree(trans, fs_info);
1413 if (IS_ERR(log_root))
1414 return PTR_ERR(log_root);
1415 WARN_ON(fs_info->log_root_tree);
1416 fs_info->log_root_tree = log_root;
1417 return 0;
1418}
1419
1420int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1421 struct btrfs_root *root)
1422{
1423 struct btrfs_root *log_root;
1424 struct btrfs_inode_item *inode_item;
1425
1426 log_root = alloc_log_tree(trans, root->fs_info);
1427 if (IS_ERR(log_root))
1428 return PTR_ERR(log_root);
1429
1430 log_root->last_trans = trans->transid;
1431 log_root->root_key.offset = root->root_key.objectid;
1432
1433 inode_item = &log_root->root_item.inode;
3cae210f
QW		 1434 btrfs_set_stack_inode_generation(inode_item, 1);
1435 btrfs_set_stack_inode_size(inode_item, 3);
1436 btrfs_set_stack_inode_nlink(inode_item, 1);
1437 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1438 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
7237f183 1439
5d4f98a2 1440 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1441
1442 WARN_ON(root->log_root);
1443 root->log_root = log_root;
1444 root->log_transid = 0;
257c62e1 1445 root->last_log_commit = 0;
e02119d5
CM		 1446 return 0;
1447}
1448
35a3621b
SB		 1449static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1450 struct btrfs_key *key)
e02119d5
CM		 1451{
1452 struct btrfs_root *root;
1453 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1454 struct btrfs_path *path;
84234f3a 1455 u64 generation;
db94535d 1456 u32 blocksize;
cb517eab 1457 int ret;
0f7d52f4 1458
cb517eab
MX		 1459 path = btrfs_alloc_path();
1460 if (!path)
0f7d52f4 1461 return ERR_PTR(-ENOMEM);
cb517eab
MX		 1462
1463 root = btrfs_alloc_root(fs_info);
1464 if (!root) {
1465 ret = -ENOMEM;
1466 goto alloc_fail;
0f7d52f4
CM		 1467 }
1468
db94535d 1469 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb 1470 tree_root->sectorsize, tree_root->stripesize,
cb517eab 1471 root, fs_info, key->objectid);
0f7d52f4 1472
cb517eab
MX		 1473 ret = btrfs_find_root(tree_root, key, path,
1474 &root->root_item, &root->root_key);
0f7d52f4 1475 if (ret) {
13a8a7c8
YZ		 1476 if (ret > 0)
1477 ret = -ENOENT;
cb517eab 1478 goto find_fail;
0f7d52f4 1479 }
13a8a7c8 1480
84234f3a 1481 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1482 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1483 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1484 blocksize, generation);
cb517eab
MX		 1485 if (!root->node) {
1486 ret = -ENOMEM;
1487 goto find_fail;
1488 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1489 ret = -EIO;
1490 goto read_fail;
416bc658 1491 }
5d4f98a2 1492 root->commit_root = btrfs_root_node(root);
13a8a7c8 1493out:
cb517eab
MX		 1494 btrfs_free_path(path);
1495 return root;
1496
1497read_fail:
1498 free_extent_buffer(root->node);
1499find_fail:
1500 kfree(root);
1501alloc_fail:
1502 root = ERR_PTR(ret);
1503 goto out;
1504}
1505
1506struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1507 struct btrfs_key *location)
1508{
1509 struct btrfs_root *root;
1510
1511 root = btrfs_read_tree_root(tree_root, location);
1512 if (IS_ERR(root))
1513 return root;
1514
1515 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1516 root->ref_cows = 1;
08fe4db1
LZ		 1517 btrfs_check_and_init_root_item(&root->root_item);
1518 }
13a8a7c8 1519
5eda7b5e
CM		 1520 return root;
1521}
1522
cb517eab
MX		 1523int btrfs_init_fs_root(struct btrfs_root *root)
1524{
1525 int ret;
1526
1527 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1528 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1529 GFP_NOFS);
1530 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1531 ret = -ENOMEM;
1532 goto fail;
1533 }
1534
1535 btrfs_init_free_ino_ctl(root);
1536 mutex_init(&root->fs_commit_mutex);
1537 spin_lock_init(&root->cache_lock);
1538 init_waitqueue_head(&root->cache_wait);
1539
1540 ret = get_anon_bdev(&root->anon_dev);
1541 if (ret)
1542 goto fail;
1543 return 0;
1544fail:
1545 kfree(root->free_ino_ctl);
1546 kfree(root->free_ino_pinned);
1547 return ret;
1548}
1549
171170c1
ST		 1550static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1551 u64 root_id)
cb517eab
MX		 1552{
1553 struct btrfs_root *root;
1554
1555 spin_lock(&fs_info->fs_roots_radix_lock);
1556 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1557 (unsigned long)root_id);
1558 spin_unlock(&fs_info->fs_roots_radix_lock);
1559 return root;
1560}
1561
1562int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1563 struct btrfs_root *root)
1564{
1565 int ret;
1566
1567 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1568 if (ret)
1569 return ret;
1570
1571 spin_lock(&fs_info->fs_roots_radix_lock);
1572 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1573 (unsigned long)root->root_key.objectid,
1574 root);
1575 if (ret == 0)
1576 root->in_radix = 1;
1577 spin_unlock(&fs_info->fs_roots_radix_lock);
1578 radix_tree_preload_end();
1579
1580 return ret;
1581}
1582
c00869f1
MX		 1583struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1584 struct btrfs_key *location,
1585 bool check_ref)
5eda7b5e
CM		 1586{
1587 struct btrfs_root *root;
1588 int ret;
1589
edbd8d4e
CM		 1590 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1591 return fs_info->tree_root;
1592 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1593 return fs_info->extent_root;
8f18cf13
CM		 1594 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1595 return fs_info->chunk_root;
1596 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1597 return fs_info->dev_root;
0403e47e
YZ		 1598 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1599 return fs_info->csum_root;
bcef60f2
AJ		 1600 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1601 return fs_info->quota_root ? fs_info->quota_root :
1602 ERR_PTR(-ENOENT);
f7a81ea4
SB		 1603 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1604 return fs_info->uuid_root ? fs_info->uuid_root :
1605 ERR_PTR(-ENOENT);
4df27c4d 1606again:
cb517eab 1607 root = btrfs_lookup_fs_root(fs_info, location->objectid);
48475471 1608 if (root) {
c00869f1 1609 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
48475471 1610 return ERR_PTR(-ENOENT);
5eda7b5e 1611 return root;
48475471 1612 }
5eda7b5e 1613
cb517eab 1614 root = btrfs_read_fs_root(fs_info->tree_root, location);
5eda7b5e
CM		 1615 if (IS_ERR(root))
1616 return root;
3394e160 1617
c00869f1 1618 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
cb517eab 1619 ret = -ENOENT;
581bb050 1620 goto fail;
35a30d7c 1621 }
581bb050 1622
cb517eab 1623 ret = btrfs_init_fs_root(root);
ac08aedf
CM		 1624 if (ret)
1625 goto fail;
3394e160 1626
d68fc57b
YZ		 1627 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1628 if (ret < 0)
1629 goto fail;
1630 if (ret == 0)
1631 root->orphan_item_inserted = 1;
1632
cb517eab 1633 ret = btrfs_insert_fs_root(fs_info, root);
0f7d52f4 1634 if (ret) {
4df27c4d
YZ		 1635 if (ret == -EEXIST) {
1636 free_fs_root(root);
1637 goto again;
1638 }
1639 goto fail;
0f7d52f4 1640 }
edbd8d4e 1641 return root;
4df27c4d
YZ		 1642fail:
1643 free_fs_root(root);
1644 return ERR_PTR(ret);
edbd8d4e
CM		 1645}
1646
04160088
CM		 1647static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1648{
1649 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1650 int ret = 0;
04160088
CM		 1651 struct btrfs_device *device;
1652 struct backing_dev_info *bdi;
b7967db7 1653
1f78160c
XG		 1654 rcu_read_lock();
1655 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1656 if (!device->bdev)
1657 continue;
04160088
CM		 1658 bdi = blk_get_backing_dev_info(device->bdev);
1659 if (bdi && bdi_congested(bdi, bdi_bits)) {
1660 ret = 1;
1661 break;
1662 }
1663 }
1f78160c 1664 rcu_read_unlock();
04160088
CM		 1665 return ret;
1666}
1667
ad081f14
JA		 1668/*
1669 * If this fails, caller must call bdi_destroy() to get rid of the
1670 * bdi again.
1671 */
04160088
CM		 1672static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1673{
ad081f14
JA		 1674 int err;
1675
1676 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1677 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1678 if (err)
1679 return err;
1680
4575c9cc 1681 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1682 bdi->congested_fn = btrfs_congested_fn;
1683 bdi->congested_data = info;
1684 return 0;
1685}
1686
8b712842
CM		 1687/*
1688 * called by the kthread helper functions to finally call the bio end_io
1689 * functions. This is where read checksum verification actually happens
1690 */
1691static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1692{
ce9adaa5 1693 struct bio *bio;
8b712842
CM		 1694 struct end_io_wq *end_io_wq;
1695 struct btrfs_fs_info *fs_info;
ce9adaa5 1696 int error;
ce9adaa5 1697
8b712842
CM		 1698 end_io_wq = container_of(work, struct end_io_wq, work);
1699 bio = end_io_wq->bio;
1700 fs_info = end_io_wq->info;
ce9adaa5 1701
8b712842
CM		 1702 error = end_io_wq->error;
1703 bio->bi_private = end_io_wq->private;
1704 bio->bi_end_io = end_io_wq->end_io;
1705 kfree(end_io_wq);
8b712842 1706 bio_endio(bio, error);
44b8bd7e
CM		 1707}
1708
a74a4b97
CM		 1709static int cleaner_kthread(void *arg)
1710{
1711 struct btrfs_root *root = arg;
d0278245 1712 int again;
a74a4b97
CM		 1713
1714 do {
d0278245 1715 again = 0;
a74a4b97 1716
d0278245 1717 /* Make the cleaner go to sleep early. */
babbf170 1718 if (btrfs_need_cleaner_sleep(root))
d0278245
MX		 1719 goto sleep;
1720
1721 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1722 goto sleep;
1723
dc7f370c
MX		 1724 /*
1725 * Avoid the problem that we change the status of the fs
1726 * during the above check and trylock.
1727 */
babbf170 1728 if (btrfs_need_cleaner_sleep(root)) {
dc7f370c
MX		 1729 mutex_unlock(&root->fs_info->cleaner_mutex);
1730 goto sleep;
76dda93c 1731 }
a74a4b97 1732
d0278245
MX		 1733 btrfs_run_delayed_iputs(root);
1734 again = btrfs_clean_one_deleted_snapshot(root);
1735 mutex_unlock(&root->fs_info->cleaner_mutex);
1736
1737 /*
05323cd1
MX		 1738 * The defragger has dealt with the R/O remount and umount,
1739 * needn't do anything special here.
d0278245
MX		 1740 */
1741 btrfs_run_defrag_inodes(root->fs_info);
1742sleep:
9d1a2a3a 1743 if (!try_to_freeze() && !again) {
a74a4b97 1744 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1745 if (!kthread_should_stop())
1746 schedule();
a74a4b97
CM		 1747 __set_current_state(TASK_RUNNING);
1748 }
1749 } while (!kthread_should_stop());
1750 return 0;
1751}
1752
1753static int transaction_kthread(void *arg)
1754{
1755 struct btrfs_root *root = arg;
1756 struct btrfs_trans_handle *trans;
1757 struct btrfs_transaction *cur;
8929ecfa 1758 u64 transid;
a74a4b97
CM		 1759 unsigned long now;
1760 unsigned long delay;
914b2007 1761 bool cannot_commit;
a74a4b97
CM		 1762
1763 do {
914b2007 1764 cannot_commit = false;
8b87dc17 1765 delay = HZ * root->fs_info->commit_interval;
a74a4b97
CM		 1766 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1767
a4abeea4 1768 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1769 cur = root->fs_info->running_transaction;
1770 if (!cur) {
a4abeea4 1771 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1772 goto sleep;
1773 }
31153d81 1774
a74a4b97 1775 now = get_seconds();
4a9d8bde 1776 if (cur->state < TRANS_STATE_BLOCKED &&
8b87dc17
DS		 1777 (now < cur->start_time ||
1778 now - cur->start_time < root->fs_info->commit_interval)) {
a4abeea4 1779 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1780 delay = HZ * 5;
1781 goto sleep;
1782 }
8929ecfa 1783 transid = cur->transid;
a4abeea4 1784 spin_unlock(&root->fs_info->trans_lock);
56bec294 1785
79787eaa 1786 /* If the file system is aborted, this will always fail. */
354aa0fb 1787 trans = btrfs_attach_transaction(root);
914b2007 1788 if (IS_ERR(trans)) {
354aa0fb
MX		 1789 if (PTR_ERR(trans) != -ENOENT)
1790 cannot_commit = true;
79787eaa 1791 goto sleep;
914b2007 1792 }
8929ecfa 1793 if (transid == trans->transid) {
79787eaa 1794 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1795 } else {
1796 btrfs_end_transaction(trans, root);
1797 }
a74a4b97
CM		 1798sleep:
1799 wake_up_process(root->fs_info->cleaner_kthread);
1800 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1801
4e121c06
JB		 1802 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1803 &root->fs_info->fs_state)))
1804 btrfs_cleanup_transaction(root);
a0acae0e 1805 if (!try_to_freeze()) {
a74a4b97 1806 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1807 if (!kthread_should_stop() &&
914b2007
JK		 1808 (!btrfs_transaction_blocked(root->fs_info) ||
1809 cannot_commit))
8929ecfa 1810 schedule_timeout(delay);
a74a4b97
CM		 1811 __set_current_state(TASK_RUNNING);
1812 }
1813 } while (!kthread_should_stop());
1814 return 0;
1815}
1816
af31f5e5
CM		 1817/*
1818 * this will find the highest generation in the array of
1819 * root backups. The index of the highest array is returned,
1820 * or -1 if we can't find anything.
1821 *
1822 * We check to make sure the array is valid by comparing the
1823 * generation of the latest root in the array with the generation
1824 * in the super block. If they don't match we pitch it.
1825 */
1826static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1827{
1828 u64 cur;
1829 int newest_index = -1;
1830 struct btrfs_root_backup *root_backup;
1831 int i;
1832
1833 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1834 root_backup = info->super_copy->super_roots + i;
1835 cur = btrfs_backup_tree_root_gen(root_backup);
1836 if (cur == newest_gen)
1837 newest_index = i;
1838 }
1839
1840 /* check to see if we actually wrapped around */
1841 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1842 root_backup = info->super_copy->super_roots;
1843 cur = btrfs_backup_tree_root_gen(root_backup);
1844 if (cur == newest_gen)
1845 newest_index = 0;
1846 }
1847 return newest_index;
1848}
1849
1850
1851/*
1852 * find the oldest backup so we know where to store new entries
1853 * in the backup array. This will set the backup_root_index
1854 * field in the fs_info struct
1855 */
1856static void find_oldest_super_backup(struct btrfs_fs_info *info,
1857 u64 newest_gen)
1858{
1859 int newest_index = -1;
1860
1861 newest_index = find_newest_super_backup(info, newest_gen);
1862 /* if there was garbage in there, just move along */
1863 if (newest_index == -1) {
1864 info->backup_root_index = 0;
1865 } else {
1866 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1867 }
1868}
1869
1870/*
1871 * copy all the root pointers into the super backup array.
1872 * this will bump the backup pointer by one when it is
1873 * done
1874 */
1875static void backup_super_roots(struct btrfs_fs_info *info)
1876{
1877 int next_backup;
1878 struct btrfs_root_backup *root_backup;
1879 int last_backup;
1880
1881 next_backup = info->backup_root_index;
1882 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1883 BTRFS_NUM_BACKUP_ROOTS;
1884
1885 /*
1886 * just overwrite the last backup if we're at the same generation
1887 * this happens only at umount
1888 */
1889 root_backup = info->super_for_commit->super_roots + last_backup;
1890 if (btrfs_backup_tree_root_gen(root_backup) ==
1891 btrfs_header_generation(info->tree_root->node))
1892 next_backup = last_backup;
1893
1894 root_backup = info->super_for_commit->super_roots + next_backup;
1895
1896 /*
1897 * make sure all of our padding and empty slots get zero filled
1898 * regardless of which ones we use today
1899 */
1900 memset(root_backup, 0, sizeof(*root_backup));
1901
1902 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1903
1904 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1905 btrfs_set_backup_tree_root_gen(root_backup,
1906 btrfs_header_generation(info->tree_root->node));
1907
1908 btrfs_set_backup_tree_root_level(root_backup,
1909 btrfs_header_level(info->tree_root->node));
1910
1911 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1912 btrfs_set_backup_chunk_root_gen(root_backup,
1913 btrfs_header_generation(info->chunk_root->node));
1914 btrfs_set_backup_chunk_root_level(root_backup,
1915 btrfs_header_level(info->chunk_root->node));
1916
1917 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1918 btrfs_set_backup_extent_root_gen(root_backup,
1919 btrfs_header_generation(info->extent_root->node));
1920 btrfs_set_backup_extent_root_level(root_backup,
1921 btrfs_header_level(info->extent_root->node));
1922
7c7e82a7
CM		 1923 /*
1924 * we might commit during log recovery, which happens before we set
1925 * the fs_root. Make sure it is valid before we fill it in.
1926 */
1927 if (info->fs_root && info->fs_root->node) {
1928 btrfs_set_backup_fs_root(root_backup,
1929 info->fs_root->node->start);
1930 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1931 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1932 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1933 btrfs_header_level(info->fs_root->node));
7c7e82a7 1934 }
af31f5e5
CM		 1935
1936 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1937 btrfs_set_backup_dev_root_gen(root_backup,
1938 btrfs_header_generation(info->dev_root->node));
1939 btrfs_set_backup_dev_root_level(root_backup,
1940 btrfs_header_level(info->dev_root->node));
1941
1942 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1943 btrfs_set_backup_csum_root_gen(root_backup,
1944 btrfs_header_generation(info->csum_root->node));
1945 btrfs_set_backup_csum_root_level(root_backup,
1946 btrfs_header_level(info->csum_root->node));
1947
1948 btrfs_set_backup_total_bytes(root_backup,
1949 btrfs_super_total_bytes(info->super_copy));
1950 btrfs_set_backup_bytes_used(root_backup,
1951 btrfs_super_bytes_used(info->super_copy));
1952 btrfs_set_backup_num_devices(root_backup,
1953 btrfs_super_num_devices(info->super_copy));
1954
1955 /*
1956 * if we don't copy this out to the super_copy, it won't get remembered
1957 * for the next commit
1958 */
1959 memcpy(&info->super_copy->super_roots,
1960 &info->super_for_commit->super_roots,
1961 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1962}
1963
1964/*
1965 * this copies info out of the root backup array and back into
1966 * the in-memory super block. It is meant to help iterate through
1967 * the array, so you send it the number of backups you've already
1968 * tried and the last backup index you used.
1969 *
1970 * this returns -1 when it has tried all the backups
1971 */
1972static noinline int next_root_backup(struct btrfs_fs_info *info,
1973 struct btrfs_super_block *super,
1974 int *num_backups_tried, int *backup_index)
1975{
1976 struct btrfs_root_backup *root_backup;
1977 int newest = *backup_index;
1978
1979 if (*num_backups_tried == 0) {
1980 u64 gen = btrfs_super_generation(super);
1981
1982 newest = find_newest_super_backup(info, gen);
1983 if (newest == -1)
1984 return -1;
1985
1986 *backup_index = newest;
1987 *num_backups_tried = 1;
1988 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1989 /* we've tried all the backups, all done */
1990 return -1;
1991 } else {
1992 /* jump to the next oldest backup */
1993 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1994 BTRFS_NUM_BACKUP_ROOTS;
1995 *backup_index = newest;
1996 *num_backups_tried += 1;
1997 }
1998 root_backup = super->super_roots + newest;
1999
2000 btrfs_set_super_generation(super,
2001 btrfs_backup_tree_root_gen(root_backup));
2002 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2003 btrfs_set_super_root_level(super,
2004 btrfs_backup_tree_root_level(root_backup));
2005 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2006
2007 /*
2008 * fixme: the total bytes and num_devices need to match or we should
2009 * need a fsck
2010 */
2011 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2012 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2013 return 0;
2014}
2015
7abadb64
LB		 2016/* helper to cleanup workers */
2017static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2018{
2019 btrfs_stop_workers(&fs_info->generic_worker);
2020 btrfs_stop_workers(&fs_info->fixup_workers);
2021 btrfs_stop_workers(&fs_info->delalloc_workers);
2022 btrfs_stop_workers(&fs_info->workers);
2023 btrfs_stop_workers(&fs_info->endio_workers);
2024 btrfs_stop_workers(&fs_info->endio_meta_workers);
2025 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2026 btrfs_stop_workers(&fs_info->rmw_workers);
2027 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2028 btrfs_stop_workers(&fs_info->endio_write_workers);
2029 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2030 btrfs_stop_workers(&fs_info->submit_workers);
2031 btrfs_stop_workers(&fs_info->delayed_workers);
2032 btrfs_stop_workers(&fs_info->caching_workers);
2033 btrfs_stop_workers(&fs_info->readahead_workers);
2034 btrfs_stop_workers(&fs_info->flush_workers);
2f232036 2035 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
7abadb64
LB		 2036}
2037
af31f5e5
CM		 2038/* helper to cleanup tree roots */
2039static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2040{
2041 free_extent_buffer(info->tree_root->node);
2042 free_extent_buffer(info->tree_root->commit_root);
af31f5e5
CM		 2043 info->tree_root->node = NULL;
2044 info->tree_root->commit_root = NULL;
655b09fe
JB		 2045
2046 if (info->dev_root) {
2047 free_extent_buffer(info->dev_root->node);
2048 free_extent_buffer(info->dev_root->commit_root);
2049 info->dev_root->node = NULL;
2050 info->dev_root->commit_root = NULL;
2051 }
2052 if (info->extent_root) {
2053 free_extent_buffer(info->extent_root->node);
2054 free_extent_buffer(info->extent_root->commit_root);
2055 info->extent_root->node = NULL;
2056 info->extent_root->commit_root = NULL;
2057 }
2058 if (info->csum_root) {
2059 free_extent_buffer(info->csum_root->node);
2060 free_extent_buffer(info->csum_root->commit_root);
2061 info->csum_root->node = NULL;
2062 info->csum_root->commit_root = NULL;
2063 }
bcef60f2 2064 if (info->quota_root) {
655b09fe
JB		 2065 free_extent_buffer(info->quota_root->node);
2066 free_extent_buffer(info->quota_root->commit_root);
bcef60f2
AJ		 2067 info->quota_root->node = NULL;
2068 info->quota_root->commit_root = NULL;
2069 }
f7a81ea4
SB		 2070 if (info->uuid_root) {
2071 free_extent_buffer(info->uuid_root->node);
2072 free_extent_buffer(info->uuid_root->commit_root);
2073 info->uuid_root->node = NULL;
2074 info->uuid_root->commit_root = NULL;
2075 }
af31f5e5
CM		 2076 if (chunk_root) {
2077 free_extent_buffer(info->chunk_root->node);
2078 free_extent_buffer(info->chunk_root->commit_root);
2079 info->chunk_root->node = NULL;
2080 info->chunk_root->commit_root = NULL;
2081 }
2082}
2083
171f6537
JB		 2084static void del_fs_roots(struct btrfs_fs_info *fs_info)
2085{
2086 int ret;
2087 struct btrfs_root *gang[8];
2088 int i;
2089
2090 while (!list_empty(&fs_info->dead_roots)) {
2091 gang[0] = list_entry(fs_info->dead_roots.next,
2092 struct btrfs_root, root_list);
2093 list_del(&gang[0]->root_list);
2094
2095 if (gang[0]->in_radix) {
cb517eab 2096 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
171f6537
JB		 2097 } else {
2098 free_extent_buffer(gang[0]->node);
2099 free_extent_buffer(gang[0]->commit_root);
b0feb9d9 2100 btrfs_put_fs_root(gang[0]);
171f6537
JB		 2101 }
2102 }
2103
2104 while (1) {
2105 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2106 (void **)gang, 0,
2107 ARRAY_SIZE(gang));
2108 if (!ret)
2109 break;
2110 for (i = 0; i < ret; i++)
cb517eab 2111 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
171f6537
JB		 2112 }
2113}
af31f5e5 2114
ad2b2c80
AV		 2115int open_ctree(struct super_block *sb,
2116 struct btrfs_fs_devices *fs_devices,
2117 char *options)
2e635a27 2118{
db94535d
CM		 2119 u32 sectorsize;
2120 u32 nodesize;
2121 u32 leafsize;
2122 u32 blocksize;
87ee04eb 2123 u32 stripesize;
84234f3a 2124 u64 generation;
f2b636e8 2125 u64 features;
3de4586c 2126 struct btrfs_key location;
a061fc8d 2127 struct buffer_head *bh;
4d34b278 2128 struct btrfs_super_block *disk_super;
815745cf 2129 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2130 struct btrfs_root *tree_root;
4d34b278
ID		 2131 struct btrfs_root *extent_root;
2132 struct btrfs_root *csum_root;
2133 struct btrfs_root *chunk_root;
2134 struct btrfs_root *dev_root;
bcef60f2 2135 struct btrfs_root *quota_root;
f7a81ea4 2136 struct btrfs_root *uuid_root;
e02119d5 2137 struct btrfs_root *log_tree_root;
eb60ceac 2138 int ret;
e58ca020 2139 int err = -EINVAL;
af31f5e5
CM		 2140 int num_backups_tried = 0;
2141 int backup_index = 0;
70f80175
SB		 2142 bool create_uuid_tree;
2143 bool check_uuid_tree;
4543df7e 2144
f84a8bd6 2145 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e 2146 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
cb517eab 2147 if (!tree_root || !chunk_root) {
39279cc3
CM		 2148 err = -ENOMEM;
2149 goto fail;
2150 }
76dda93c
YZ		 2151
2152 ret = init_srcu_struct(&fs_info->subvol_srcu);
2153 if (ret) {
2154 err = ret;
2155 goto fail;
2156 }
2157
2158 ret = setup_bdi(fs_info, &fs_info->bdi);
2159 if (ret) {
2160 err = ret;
2161 goto fail_srcu;
2162 }
2163
e2d84521
MX		 2164 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2165 if (ret) {
2166 err = ret;
2167 goto fail_bdi;
2168 }
2169 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2170 (1 + ilog2(nr_cpu_ids));
2171
963d678b
MX		 2172 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2173 if (ret) {
2174 err = ret;
2175 goto fail_dirty_metadata_bytes;
2176 }
2177
76dda93c
YZ		 2178 fs_info->btree_inode = new_inode(sb);
2179 if (!fs_info->btree_inode) {
2180 err = -ENOMEM;
963d678b 2181 goto fail_delalloc_bytes;
76dda93c
YZ		 2182 }
2183
a6591715 2184 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2185
76dda93c 2186 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 2187 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2188 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2189 INIT_LIST_HEAD(&fs_info->delayed_iputs);
eb73c1b7 2190 INIT_LIST_HEAD(&fs_info->delalloc_roots);
11833d66 2191 INIT_LIST_HEAD(&fs_info->caching_block_groups);
eb73c1b7 2192 spin_lock_init(&fs_info->delalloc_root_lock);
a4abeea4 2193 spin_lock_init(&fs_info->trans_lock);
76dda93c 2194 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2195 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2196 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2197 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2198 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2199 spin_lock_init(&fs_info->super_lock);
f29021b2 2200 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2201 mutex_init(&fs_info->reloc_mutex);
de98ced9 2202 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2203
58176a96 2204 init_completion(&fs_info->kobj_unregister);
0b86a832 2205 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2206 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2207 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2208 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2209 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2210 BTRFS_BLOCK_RSV_GLOBAL);
2211 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2212 BTRFS_BLOCK_RSV_DELALLOC);
2213 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2214 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2215 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2216 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2217 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2218 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2219 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2220 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2221 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2222 atomic_set(&fs_info->defrag_running, 0);
fc36ed7e 2223 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2224 fs_info->sb = sb;
6f568d35 2225 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2226 fs_info->metadata_ratio = 0;
4cb5300b 2227 fs_info->defrag_inodes = RB_ROOT;
2bf64758 2228 fs_info->free_chunk_space = 0;
f29021b2 2229 fs_info->tree_mod_log = RB_ROOT;
8b87dc17 2230 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
c8b97818 2231
90519d66
AJ		 2232 /* readahead state */
2233 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2234 spin_lock_init(&fs_info->reada_lock);
c8b97818 2235
b34b086c
CM		 2236 fs_info->thread_pool_size = min_t(unsigned long,
2237 num_online_cpus() + 2, 8);
0afbaf8c 2238
199c2a9c
MX		 2239 INIT_LIST_HEAD(&fs_info->ordered_roots);
2240 spin_lock_init(&fs_info->ordered_root_lock);
16cdcec7
MX		 2241 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2242 GFP_NOFS);
2243 if (!fs_info->delayed_root) {
2244 err = -ENOMEM;
2245 goto fail_iput;
2246 }
2247 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2248
a2de733c
AJ		 2249 mutex_init(&fs_info->scrub_lock);
2250 atomic_set(&fs_info->scrubs_running, 0);
2251 atomic_set(&fs_info->scrub_pause_req, 0);
2252 atomic_set(&fs_info->scrubs_paused, 0);
2253 atomic_set(&fs_info->scrub_cancel_req, 0);
2254 init_waitqueue_head(&fs_info->scrub_pause_wait);
2255 init_rwsem(&fs_info->scrub_super_lock);
2256 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2257#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2258 fs_info->check_integrity_print_mask = 0;
2259#endif
a2de733c 2260
c9e9f97b
ID		 2261 spin_lock_init(&fs_info->balance_lock);
2262 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2263 atomic_set(&fs_info->balance_running, 0);
2264 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2265 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2266 fs_info->balance_ctl = NULL;
837d5b6e 2267 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2268
a061fc8d
CM		 2269 sb->s_blocksize = 4096;
2270 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2271 sb->s_bdi = &fs_info->bdi;
a061fc8d 2272
76dda93c 2273 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2274 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2275 /*
2276 * we set the i_size on the btree inode to the max possible int.
2277 * the real end of the address space is determined by all of
2278 * the devices in the system
2279 */
2280 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2281 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2282 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2283
5d4f98a2 2284 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2285 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2286 fs_info->btree_inode->i_mapping);
0b32f4bb 2287 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2288 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2289
2290 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2291
76dda93c
YZ		 2292 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2293 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2294 sizeof(struct btrfs_key));
72ac3c0d
JB		 2295 set_bit(BTRFS_INODE_DUMMY,
2296 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
778ba82b 2297 btrfs_insert_inode_hash(fs_info->btree_inode);
76dda93c 2298
0f9dd46c 2299 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2300 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2301 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2302
11833d66 2303 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2304 fs_info->btree_inode->i_mapping);
11833d66 2305 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2306 fs_info->btree_inode->i_mapping);
11833d66 2307 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2308 fs_info->do_barriers = 1;
e18e4809 2309
39279cc3 2310
5a3f23d5 2311 mutex_init(&fs_info->ordered_operations_mutex);
9ffba8cd 2312 mutex_init(&fs_info->ordered_extent_flush_mutex);
e02119d5 2313 mutex_init(&fs_info->tree_log_mutex);
925baedd 2314 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2315 mutex_init(&fs_info->transaction_kthread_mutex);
2316 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2317 mutex_init(&fs_info->volume_mutex);
276e680d 2318 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2319 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2320 init_rwsem(&fs_info->subvol_sem);
803b2f54 2321 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
e922e087
SB		 2322 fs_info->dev_replace.lock_owner = 0;
2323 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2324 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2325 mutex_init(&fs_info->dev_replace.lock_management_lock);
2326 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2327
416ac51d 2328 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2329 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2330 fs_info->qgroup_tree = RB_ROOT;
2331 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2332 fs_info->qgroup_seq = 1;
2333 fs_info->quota_enabled = 0;
2334 fs_info->pending_quota_state = 0;
1e8f9158 2335 fs_info->qgroup_ulist = NULL;
2f232036 2336 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2337
fa9c0d79
CM		 2338 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2339 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2340
e6dcd2dc 2341 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2342 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2343 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2344 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2345
53b381b3
DW		 2346 ret = btrfs_alloc_stripe_hash_table(fs_info);
2347 if (ret) {
83c8266a 2348 err = ret;
53b381b3
DW		 2349 goto fail_alloc;
2350 }
2351
0b86a832 2352 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2353 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2354
3c4bb26b 2355 invalidate_bdev(fs_devices->latest_bdev);
1104a885
DS		 2356
2357 /*
2358 * Read super block and check the signature bytes only
2359 */
a512bbf8 2360 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2361 if (!bh) {
2362 err = -EINVAL;
16cdcec7 2363 goto fail_alloc;
20b45077 2364 }
39279cc3 2365
1104a885
DS		 2366 /*
2367 * We want to check superblock checksum, the type is stored inside.
2368 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2369 */
2370 if (btrfs_check_super_csum(bh->b_data)) {
2371 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2372 err = -EINVAL;
2373 goto fail_alloc;
2374 }
2375
2376 /*
2377 * super_copy is zeroed at allocation time and we never touch the
2378 * following bytes up to INFO_SIZE, the checksum is calculated from
2379 * the whole block of INFO_SIZE
2380 */
6c41761f
DS		 2381 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2382 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2383 sizeof(*fs_info->super_for_commit));
a061fc8d 2384 brelse(bh);
5f39d397 2385
6c41761f 2386 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2387
1104a885
DS		 2388 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2389 if (ret) {
2390 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2391 err = -EINVAL;
2392 goto fail_alloc;
2393 }
2394
6c41761f 2395 disk_super = fs_info->super_copy;
0f7d52f4 2396 if (!btrfs_super_root(disk_super))
16cdcec7 2397 goto fail_alloc;
0f7d52f4 2398
acce952b 2399 /* check FS state, whether FS is broken. */
87533c47
MX		 2400 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2401 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2402
af31f5e5
CM		 2403 /*
2404 * run through our array of backup supers and setup
2405 * our ring pointer to the oldest one
2406 */
2407 generation = btrfs_super_generation(disk_super);
2408 find_oldest_super_backup(fs_info, generation);
2409
75e7cb7f
LB		 2410 /*
2411 * In the long term, we'll store the compression type in the super
2412 * block, and it'll be used for per file compression control.
2413 */
2414 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2415
2b82032c
YZ		 2416 ret = btrfs_parse_options(tree_root, options);
2417 if (ret) {
2418 err = ret;
16cdcec7 2419 goto fail_alloc;
2b82032c 2420 }
dfe25020 2421
f2b636e8
JB		 2422 features = btrfs_super_incompat_flags(disk_super) &
2423 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2424 if (features) {
2425 printk(KERN_ERR "BTRFS: couldn't mount because of "
2426 "unsupported optional features (%Lx).\n",
c1c9ff7c 2427 features);
f2b636e8 2428 err = -EINVAL;
16cdcec7 2429 goto fail_alloc;
f2b636e8
JB		 2430 }
2431
727011e0
CM		 2432 if (btrfs_super_leafsize(disk_super) !=
2433 btrfs_super_nodesize(disk_super)) {
2434 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2435 "blocksizes don't match. node %d leaf %d\n",
2436 btrfs_super_nodesize(disk_super),
2437 btrfs_super_leafsize(disk_super));
2438 err = -EINVAL;
2439 goto fail_alloc;
2440 }
2441 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2442 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2443 "blocksize (%d) was too large\n",
2444 btrfs_super_leafsize(disk_super));
2445 err = -EINVAL;
2446 goto fail_alloc;
2447 }
2448
5d4f98a2 2449 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2450 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2451 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2452 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2453
3173a18f
JB		 2454 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2455 printk(KERN_ERR "btrfs: has skinny extents\n");
2456
727011e0
CM		 2457 /*
2458 * flag our filesystem as having big metadata blocks if
2459 * they are bigger than the page size
2460 */
2461 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2462 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2463 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2464 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2465 }
2466
bc3f116f
CM		 2467 nodesize = btrfs_super_nodesize(disk_super);
2468 leafsize = btrfs_super_leafsize(disk_super);
2469 sectorsize = btrfs_super_sectorsize(disk_super);
2470 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2471 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2472 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2473
2474 /*
2475 * mixed block groups end up with duplicate but slightly offset
2476 * extent buffers for the same range. It leads to corruptions
2477 */
2478 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2479 (sectorsize != leafsize)) {
2480 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2481 "are not allowed for mixed block groups on %s\n",
2482 sb->s_id);
2483 goto fail_alloc;
2484 }
2485
ceda0864
MX		 2486 /*
2487 * Needn't use the lock because there is no other task which will
2488 * update the flag.
2489 */
a6fa6fae 2490 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2491
f2b636e8
JB		 2492 features = btrfs_super_compat_ro_flags(disk_super) &
2493 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2494 if (!(sb->s_flags & MS_RDONLY) && features) {
2495 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2496 "unsupported option features (%Lx).\n",
c1c9ff7c 2497 features);
f2b636e8 2498 err = -EINVAL;
16cdcec7 2499 goto fail_alloc;
f2b636e8 2500 }
61d92c32
CM		 2501
2502 btrfs_init_workers(&fs_info->generic_worker,
2503 "genwork", 1, NULL);
2504
5443be45 2505 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 2506 fs_info->thread_pool_size,
2507 &fs_info->generic_worker);
c8b97818 2508
771ed689 2509 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
45d5fd14 2510 fs_info->thread_pool_size, NULL);
771ed689 2511
8ccf6f19 2512 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
45d5fd14 2513 fs_info->thread_pool_size, NULL);
8ccf6f19 2514
5443be45 2515 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 2516 min_t(u64, fs_devices->num_devices,
45d5fd14 2517 fs_info->thread_pool_size), NULL);
61b49440 2518
bab39bf9 2519 btrfs_init_workers(&fs_info->caching_workers, "cache",
45d5fd14 2520 fs_info->thread_pool_size, NULL);
bab39bf9 2521
61b49440
CM		 2522 /* a higher idle thresh on the submit workers makes it much more
2523 * likely that bios will be send down in a sane order to the
2524 * devices
2525 */
2526 fs_info->submit_workers.idle_thresh = 64;
53863232 2527
771ed689 2528 fs_info->workers.idle_thresh = 16;
4a69a410 2529 fs_info->workers.ordered = 1;
61b49440 2530
771ed689
CM		 2531 fs_info->delalloc_workers.idle_thresh = 2;
2532 fs_info->delalloc_workers.ordered = 1;
2533
61d92c32
CM		 2534 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2535 &fs_info->generic_worker);
5443be45 2536 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2537 fs_info->thread_pool_size,
2538 &fs_info->generic_worker);
d20f7043 2539 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2540 fs_info->thread_pool_size,
2541 &fs_info->generic_worker);
cad321ad 2542 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2543 "endio-meta-write", fs_info->thread_pool_size,
2544 &fs_info->generic_worker);
53b381b3
DW		 2545 btrfs_init_workers(&fs_info->endio_raid56_workers,
2546 "endio-raid56", fs_info->thread_pool_size,
2547 &fs_info->generic_worker);
2548 btrfs_init_workers(&fs_info->rmw_workers,
2549 "rmw", fs_info->thread_pool_size,
2550 &fs_info->generic_worker);
5443be45 2551 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2552 fs_info->thread_pool_size,
2553 &fs_info->generic_worker);
0cb59c99
JB		 2554 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2555 1, &fs_info->generic_worker);
16cdcec7
MX		 2556 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2557 fs_info->thread_pool_size,
2558 &fs_info->generic_worker);
90519d66
AJ		 2559 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2560 fs_info->thread_pool_size,
2561 &fs_info->generic_worker);
2f232036
JS		 2562 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2563 &fs_info->generic_worker);
61b49440
CM		 2564
2565 /*
2566 * endios are largely parallel and should have a very
2567 * low idle thresh
2568 */
2569 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2570 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2571 fs_info->endio_raid56_workers.idle_thresh = 4;
2572 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2573
9042846b
CM		 2574 fs_info->endio_write_workers.idle_thresh = 2;
2575 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2576 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2577
0dc3b84a
JB		 2578 /*
2579 * btrfs_start_workers can really only fail because of ENOMEM so just
2580 * return -ENOMEM if any of these fail.
2581 */
2582 ret = btrfs_start_workers(&fs_info->workers);
2583 ret |= btrfs_start_workers(&fs_info->generic_worker);
2584 ret |= btrfs_start_workers(&fs_info->submit_workers);
2585 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2586 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2587 ret |= btrfs_start_workers(&fs_info->endio_workers);
2588 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2589 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2590 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2591 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2592 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2593 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2594 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2595 ret |= btrfs_start_workers(&fs_info->caching_workers);
2596 ret |= btrfs_start_workers(&fs_info->readahead_workers);
8ccf6f19 2597 ret |= btrfs_start_workers(&fs_info->flush_workers);
2f232036 2598 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
0dc3b84a 2599 if (ret) {
fed425c7 2600 err = -ENOMEM;
0dc3b84a
JB		 2601 goto fail_sb_buffer;
2602 }
4543df7e 2603
4575c9cc 2604 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2605 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2606 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2607
db94535d
CM		 2608 tree_root->nodesize = nodesize;
2609 tree_root->leafsize = leafsize;
2610 tree_root->sectorsize = sectorsize;
87ee04eb 2611 tree_root->stripesize = stripesize;
a061fc8d
CM		 2612
2613 sb->s_blocksize = sectorsize;
2614 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2615
3cae210f 2616 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
d397712b 2617 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2618 goto fail_sb_buffer;
2619 }
19c00ddc 2620
8d082fb7
LB		 2621 if (sectorsize != PAGE_SIZE) {
2622 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2623 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2624 goto fail_sb_buffer;
2625 }
2626
925baedd 2627 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2628 ret = btrfs_read_sys_array(tree_root);
925baedd 2629 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2630 if (ret) {
d397712b
CM		 2631 printk(KERN_WARNING "btrfs: failed to read the system "
2632 "array on %s\n", sb->s_id);
5d4f98a2 2633 goto fail_sb_buffer;
84eed90f 2634 }
0b86a832
CM		 2635
2636 blocksize = btrfs_level_size(tree_root,
2637 btrfs_super_chunk_root_level(disk_super));
84234f3a 2638 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2639
2640 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2641 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2642
2643 chunk_root->node = read_tree_block(chunk_root,
2644 btrfs_super_chunk_root(disk_super),
84234f3a 2645 blocksize, generation);
416bc658
JB		 2646 if (!chunk_root->node ||
2647 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
83121942
DW		 2648 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2649 sb->s_id);
af31f5e5 2650 goto fail_tree_roots;
83121942 2651 }
5d4f98a2
YZ		 2652 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2653 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2654
e17cade2 2655 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
b308bc2f 2656 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
e17cade2 2657
0b86a832 2658 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2659 if (ret) {
d397712b
CM		 2660 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2661 sb->s_id);
af31f5e5 2662 goto fail_tree_roots;
2b82032c 2663 }
0b86a832 2664
8dabb742
SB		 2665 /*
2666 * keep the device that is marked to be the target device for the
2667 * dev_replace procedure
2668 */
2669 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2670
a6b0d5c8
CM		 2671 if (!fs_devices->latest_bdev) {
2672 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2673 sb->s_id);
2674 goto fail_tree_roots;
2675 }
2676
af31f5e5 2677retry_root_backup:
db94535d
CM		 2678 blocksize = btrfs_level_size(tree_root,
2679 btrfs_super_root_level(disk_super));
84234f3a 2680 generation = btrfs_super_generation(disk_super);
0b86a832 2681
e20d96d6 2682 tree_root->node = read_tree_block(tree_root,
db94535d 2683 btrfs_super_root(disk_super),
84234f3a 2684 blocksize, generation);
af31f5e5
CM		 2685 if (!tree_root->node ||
2686 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
83121942
DW		 2687 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2688 sb->s_id);
af31f5e5
CM		 2689
2690 goto recovery_tree_root;
83121942 2691 }
af31f5e5 2692
5d4f98a2
YZ		 2693 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2694 tree_root->commit_root = btrfs_root_node(tree_root);
69e9c6c6 2695 btrfs_set_root_refs(&tree_root->root_item, 1);
db94535d 2696
cb517eab
MX		 2697 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2698 location.type = BTRFS_ROOT_ITEM_KEY;
2699 location.offset = 0;
2700
2701 extent_root = btrfs_read_tree_root(tree_root, &location);
2702 if (IS_ERR(extent_root)) {
2703 ret = PTR_ERR(extent_root);
af31f5e5 2704 goto recovery_tree_root;
cb517eab 2705 }
0b86a832 2706 extent_root->track_dirty = 1;
cb517eab 2707 fs_info->extent_root = extent_root;
0b86a832 2708
cb517eab
MX		 2709 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2710 dev_root = btrfs_read_tree_root(tree_root, &location);
2711 if (IS_ERR(dev_root)) {
2712 ret = PTR_ERR(dev_root);
af31f5e5 2713 goto recovery_tree_root;
cb517eab 2714 }
5d4f98a2 2715 dev_root->track_dirty = 1;
cb517eab
MX		 2716 fs_info->dev_root = dev_root;
2717 btrfs_init_devices_late(fs_info);
3768f368 2718
cb517eab
MX		 2719 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2720 csum_root = btrfs_read_tree_root(tree_root, &location);
2721 if (IS_ERR(csum_root)) {
2722 ret = PTR_ERR(csum_root);
af31f5e5 2723 goto recovery_tree_root;
cb517eab 2724 }
d20f7043 2725 csum_root->track_dirty = 1;
cb517eab 2726 fs_info->csum_root = csum_root;
d20f7043 2727
cb517eab
MX		 2728 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2729 quota_root = btrfs_read_tree_root(tree_root, &location);
2730 if (!IS_ERR(quota_root)) {
bcef60f2
AJ		 2731 quota_root->track_dirty = 1;
2732 fs_info->quota_enabled = 1;
2733 fs_info->pending_quota_state = 1;
cb517eab 2734 fs_info->quota_root = quota_root;
bcef60f2
AJ		 2735 }
2736
f7a81ea4
SB		 2737 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2738 uuid_root = btrfs_read_tree_root(tree_root, &location);
2739 if (IS_ERR(uuid_root)) {
2740 ret = PTR_ERR(uuid_root);
2741 if (ret != -ENOENT)
2742 goto recovery_tree_root;
2743 create_uuid_tree = true;
70f80175 2744 check_uuid_tree = false;
f7a81ea4
SB		 2745 } else {
2746 uuid_root->track_dirty = 1;
2747 fs_info->uuid_root = uuid_root;
70f80175
SB		 2748 create_uuid_tree = false;
2749 check_uuid_tree =
2750 generation != btrfs_super_uuid_tree_generation(disk_super);
f7a81ea4
SB		 2751 }
2752
8929ecfa
YZ		 2753 fs_info->generation = generation;
2754 fs_info->last_trans_committed = generation;
8929ecfa 2755
68310a5e
ID		 2756 ret = btrfs_recover_balance(fs_info);
2757 if (ret) {
2758 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2759 goto fail_block_groups;
2760 }
2761
733f4fbb
SB		 2762 ret = btrfs_init_dev_stats(fs_info);
2763 if (ret) {
2764 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2765 ret);
2766 goto fail_block_groups;
2767 }
2768
8dabb742
SB		 2769 ret = btrfs_init_dev_replace(fs_info);
2770 if (ret) {
2771 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2772 goto fail_block_groups;
2773 }
2774
2775 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2776
c59021f8 2777 ret = btrfs_init_space_info(fs_info);
2778 if (ret) {
2779 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2780 goto fail_block_groups;
2781 }
2782
1b1d1f66
JB		 2783 ret = btrfs_read_block_groups(extent_root);
2784 if (ret) {
2785 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2786 goto fail_block_groups;
2787 }
5af3e8cc
SB		 2788 fs_info->num_tolerated_disk_barrier_failures =
2789 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2790 if (fs_info->fs_devices->missing_devices >
2791 fs_info->num_tolerated_disk_barrier_failures &&
2792 !(sb->s_flags & MS_RDONLY)) {
2793 printk(KERN_WARNING
2794 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2795 goto fail_block_groups;
2796 }
9078a3e1 2797
a74a4b97
CM		 2798 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2799 "btrfs-cleaner");
57506d50 2800 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2801 goto fail_block_groups;
a74a4b97
CM		 2802
2803 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2804 tree_root,
2805 "btrfs-transaction");
57506d50 2806 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2807 goto fail_cleaner;
a74a4b97 2808
c289811c
CM		 2809 if (!btrfs_test_opt(tree_root, SSD) &&
2810 !btrfs_test_opt(tree_root, NOSSD) &&
2811 !fs_info->fs_devices->rotating) {
2812 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2813 "mode\n");
2814 btrfs_set_opt(fs_info->mount_opt, SSD);
2815 }
2816
21adbd5c
SB		 2817#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2818 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2819 ret = btrfsic_mount(tree_root, fs_devices,
2820 btrfs_test_opt(tree_root,
2821 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2822 1 : 0,
2823 fs_info->check_integrity_print_mask);
2824 if (ret)
2825 printk(KERN_WARNING "btrfs: failed to initialize"
2826 " integrity check module %s\n", sb->s_id);
2827 }
2828#endif
bcef60f2
AJ		 2829 ret = btrfs_read_qgroup_config(fs_info);
2830 if (ret)
2831 goto fail_trans_kthread;
21adbd5c 2832
acce952b 2833 /* do not make disk changes in broken FS */
68ce9682 2834 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2835 u64 bytenr = btrfs_super_log_root(disk_super);
2836
7c2ca468 2837 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2838 printk(KERN_WARNING "Btrfs log replay required "
2839 "on RO media\n");
7c2ca468 2840 err = -EIO;
bcef60f2 2841 goto fail_qgroup;
7c2ca468 2842 }
e02119d5
CM		 2843 blocksize =
2844 btrfs_level_size(tree_root,
2845 btrfs_super_log_root_level(disk_super));
d18a2c44 2846
6f07e42e 2847 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2848 if (!log_tree_root) {
2849 err = -ENOMEM;
bcef60f2 2850 goto fail_qgroup;
676e4c86 2851 }
e02119d5
CM		 2852
2853 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2854 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2855
2856 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2857 blocksize,
2858 generation + 1);
416bc658
JB		 2859 if (!log_tree_root->node ||
2860 !extent_buffer_uptodate(log_tree_root->node)) {
2861 printk(KERN_ERR "btrfs: failed to read log tree\n");
2862 free_extent_buffer(log_tree_root->node);
2863 kfree(log_tree_root);
2864 goto fail_trans_kthread;
2865 }
79787eaa 2866 /* returns with log_tree_root freed on success */
e02119d5 2867 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2868 if (ret) {
2869 btrfs_error(tree_root->fs_info, ret,
2870 "Failed to recover log tree");
2871 free_extent_buffer(log_tree_root->node);
2872 kfree(log_tree_root);
2873 goto fail_trans_kthread;
2874 }
e556ce2c
YZ		 2875
2876 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2877 ret = btrfs_commit_super(tree_root);
2878 if (ret)
2879 goto fail_trans_kthread;
e556ce2c 2880 }
e02119d5 2881 }
1a40e23b 2882
76dda93c 2883 ret = btrfs_find_orphan_roots(tree_root);
79787eaa
JM		 2884 if (ret)
2885 goto fail_trans_kthread;
76dda93c 2886
7c2ca468 2887 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2888 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2
ID		 2889 if (ret)
2890 goto fail_trans_kthread;
d68fc57b 2891
5d4f98a2 2892 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2893 if (ret < 0) {
2894 printk(KERN_WARNING
2895 "btrfs: failed to recover relocation\n");
2896 err = -EINVAL;
bcef60f2 2897 goto fail_qgroup;
d7ce5843 2898 }
7c2ca468 2899 }
1a40e23b 2900
3de4586c
CM		 2901 location.objectid = BTRFS_FS_TREE_OBJECTID;
2902 location.type = BTRFS_ROOT_ITEM_KEY;
cb517eab 2903 location.offset = 0;
3de4586c 2904
3de4586c 2905 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3140c9a3
DC		 2906 if (IS_ERR(fs_info->fs_root)) {
2907 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2908 goto fail_qgroup;
3140c9a3 2909 }
c289811c 2910
2b6ba629
ID		 2911 if (sb->s_flags & MS_RDONLY)
2912 return 0;
59641015 2913
2b6ba629
ID		 2914 down_read(&fs_info->cleanup_work_sem);
2915 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2916 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2917 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2918 close_ctree(tree_root);
2919 return ret;
2920 }
2921 up_read(&fs_info->cleanup_work_sem);
59641015 2922
2b6ba629
ID		 2923 ret = btrfs_resume_balance_async(fs_info);
2924 if (ret) {
2925 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2926 close_ctree(tree_root);
2927 return ret;
e3acc2a6
JB		 2928 }
2929
8dabb742
SB		 2930 ret = btrfs_resume_dev_replace_async(fs_info);
2931 if (ret) {
2932 pr_warn("btrfs: failed to resume dev_replace\n");
2933 close_ctree(tree_root);
2934 return ret;
2935 }
2936
b382a324
JS		 2937 btrfs_qgroup_rescan_resume(fs_info);
2938
f7a81ea4
SB		 2939 if (create_uuid_tree) {
2940 pr_info("btrfs: creating UUID tree\n");
2941 ret = btrfs_create_uuid_tree(fs_info);
2942 if (ret) {
2943 pr_warn("btrfs: failed to create the UUID tree %d\n",
2944 ret);
2945 close_ctree(tree_root);
2946 return ret;
2947 }
f420ee1e
SB		 2948 } else if (check_uuid_tree ||
2949 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
70f80175
SB		 2950 pr_info("btrfs: checking UUID tree\n");
2951 ret = btrfs_check_uuid_tree(fs_info);
2952 if (ret) {
2953 pr_warn("btrfs: failed to check the UUID tree %d\n",
2954 ret);
2955 close_ctree(tree_root);
2956 return ret;
2957 }
2958 } else {
2959 fs_info->update_uuid_tree_gen = 1;
f7a81ea4
SB		 2960 }
2961
ad2b2c80 2962 return 0;
39279cc3 2963
bcef60f2
AJ		 2964fail_qgroup:
2965 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2966fail_trans_kthread:
2967 kthread_stop(fs_info->transaction_kthread);
54067ae9 2968 btrfs_cleanup_transaction(fs_info->tree_root);
7b5ff90e 2969 del_fs_roots(fs_info);
3f157a2f 2970fail_cleaner:
a74a4b97 2971 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2972
2973 /*
2974 * make sure we're done with the btree inode before we stop our
2975 * kthreads
2976 */
2977 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2978
1b1d1f66 2979fail_block_groups:
54067ae9 2980 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2981 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2982
2983fail_tree_roots:
2984 free_root_pointers(fs_info, 1);
2b8195bb 2985 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2986
39279cc3 2987fail_sb_buffer:
7abadb64 2988 btrfs_stop_all_workers(fs_info);
16cdcec7 2989fail_alloc:
4543df7e 2990fail_iput:
586e46e2
ID		 2991 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2992
4543df7e 2993 iput(fs_info->btree_inode);
963d678b
MX		 2994fail_delalloc_bytes:
2995 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 2996fail_dirty_metadata_bytes:
2997 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 2998fail_bdi:
7e662854 2999 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 3000fail_srcu:
3001 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 3002fail:
53b381b3 3003 btrfs_free_stripe_hash_table(fs_info);
586e46e2 3004 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 3005 return err;
af31f5e5
CM		 3006
3007recovery_tree_root:
af31f5e5
CM		 3008 if (!btrfs_test_opt(tree_root, RECOVERY))
3009 goto fail_tree_roots;
3010
3011 free_root_pointers(fs_info, 0);
3012
3013 /* don't use the log in recovery mode, it won't be valid */
3014 btrfs_set_super_log_root(disk_super, 0);
3015
3016 /* we can't trust the free space cache either */
3017 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3018
3019 ret = next_root_backup(fs_info, fs_info->super_copy,
3020 &num_backups_tried, &backup_index);
3021 if (ret == -1)
3022 goto fail_block_groups;
3023 goto retry_root_backup;
eb60ceac
CM		 3024}
3025
f2984462
CM		 3026static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3027{
f2984462
CM		 3028 if (uptodate) {
3029 set_buffer_uptodate(bh);
3030 } else {
442a4f63
SB		 3031 struct btrfs_device *device = (struct btrfs_device *)
3032 bh->b_private;
3033
606686ee
JB		 3034 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3035 "I/O error on %s\n",
3036 rcu_str_deref(device->name));
1259ab75
CM		 3037 /* note, we dont' set_buffer_write_io_error because we have
3038 * our own ways of dealing with the IO errors
3039 */
f2984462 3040 clear_buffer_uptodate(bh);
442a4f63 3041 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 3042 }
3043 unlock_buffer(bh);
3044 put_bh(bh);
3045}
3046
a512bbf8
YZ		 3047struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3048{
3049 struct buffer_head *bh;
3050 struct buffer_head *latest = NULL;
3051 struct btrfs_super_block *super;
3052 int i;
3053 u64 transid = 0;
3054 u64 bytenr;
3055
3056 /* we would like to check all the supers, but that would make
3057 * a btrfs mount succeed after a mkfs from a different FS.
3058 * So, we need to add a special mount option to scan for
3059 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3060 */
3061 for (i = 0; i < 1; i++) {
3062 bytenr = btrfs_sb_offset(i);
8068a47e
AJ		 3063 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3064 i_size_read(bdev->bd_inode))
a512bbf8 3065 break;
8068a47e
AJ		 3066 bh = __bread(bdev, bytenr / 4096,
3067 BTRFS_SUPER_INFO_SIZE);
a512bbf8
YZ		 3068 if (!bh)
3069 continue;
3070
3071 super = (struct btrfs_super_block *)bh->b_data;
3072 if (btrfs_super_bytenr(super) != bytenr ||
3cae210f 3073 btrfs_super_magic(super) != BTRFS_MAGIC) {
a512bbf8
YZ		 3074 brelse(bh);
3075 continue;
3076 }
3077
3078 if (!latest || btrfs_super_generation(super) > transid) {
3079 brelse(latest);
3080 latest = bh;
3081 transid = btrfs_super_generation(super);
3082 } else {
3083 brelse(bh);
3084 }
3085 }
3086 return latest;
3087}
3088
4eedeb75
HH		 3089/*
3090 * this should be called twice, once with wait == 0 and
3091 * once with wait == 1. When wait == 0 is done, all the buffer heads
3092 * we write are pinned.
3093 *
3094 * They are released when wait == 1 is done.
3095 * max_mirrors must be the same for both runs, and it indicates how
3096 * many supers on this one device should be written.
3097 *
3098 * max_mirrors == 0 means to write them all.
3099 */
a512bbf8
YZ		 3100static int write_dev_supers(struct btrfs_device *device,
3101 struct btrfs_super_block *sb,
3102 int do_barriers, int wait, int max_mirrors)
3103{
3104 struct buffer_head *bh;
3105 int i;
3106 int ret;
3107 int errors = 0;
3108 u32 crc;
3109 u64 bytenr;
a512bbf8
YZ		 3110
3111 if (max_mirrors == 0)
3112 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3113
a512bbf8
YZ		 3114 for (i = 0; i < max_mirrors; i++) {
3115 bytenr = btrfs_sb_offset(i);
3116 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3117 break;
3118
3119 if (wait) {
3120 bh = __find_get_block(device->bdev, bytenr / 4096,
3121 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3122 if (!bh) {
3123 errors++;
3124 continue;
3125 }
a512bbf8 3126 wait_on_buffer(bh);
4eedeb75
HH		 3127 if (!buffer_uptodate(bh))
3128 errors++;
3129
3130 /* drop our reference */
3131 brelse(bh);
3132
3133 /* drop the reference from the wait == 0 run */
3134 brelse(bh);
3135 continue;
a512bbf8
YZ		 3136 } else {
3137 btrfs_set_super_bytenr(sb, bytenr);
3138
3139 crc = ~(u32)0;
b0496686 3140 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 3141 BTRFS_CSUM_SIZE, crc,
3142 BTRFS_SUPER_INFO_SIZE -
3143 BTRFS_CSUM_SIZE);
3144 btrfs_csum_final(crc, sb->csum);
3145
4eedeb75
HH		 3146 /*
3147 * one reference for us, and we leave it for the
3148 * caller
3149 */
a512bbf8
YZ		 3150 bh = __getblk(device->bdev, bytenr / 4096,
3151 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3152 if (!bh) {
3153 printk(KERN_ERR "btrfs: couldn't get super "
3154 "buffer head for bytenr %Lu\n", bytenr);
3155 errors++;
3156 continue;
3157 }
3158
a512bbf8
YZ		 3159 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3160
4eedeb75 3161 /* one reference for submit_bh */
a512bbf8 3162 get_bh(bh);
4eedeb75
HH		 3163
3164 set_buffer_uptodate(bh);
a512bbf8
YZ		 3165 lock_buffer(bh);
3166 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3167 bh->b_private = device;
a512bbf8
YZ		 3168 }
3169
387125fc
CM		 3170 /*
3171 * we fua the first super. The others we allow
3172 * to go down lazy.
3173 */
21adbd5c 3174 ret = btrfsic_submit_bh(WRITE_FUA, bh);
4eedeb75 3175 if (ret)
a512bbf8 3176 errors++;
a512bbf8
YZ		 3177 }
3178 return errors < i ? 0 : -1;
3179}
3180
387125fc
CM		 3181/*
3182 * endio for the write_dev_flush, this will wake anyone waiting
3183 * for the barrier when it is done
3184 */
3185static void btrfs_end_empty_barrier(struct bio *bio, int err)
3186{
3187 if (err) {
3188 if (err == -EOPNOTSUPP)
3189 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3190 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3191 }
3192 if (bio->bi_private)
3193 complete(bio->bi_private);
3194 bio_put(bio);
3195}
3196
3197/*
3198 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3199 * sent down. With wait == 1, it waits for the previous flush.
3200 *
3201 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3202 * capable
3203 */
3204static int write_dev_flush(struct btrfs_device *device, int wait)
3205{
3206 struct bio *bio;
3207 int ret = 0;
3208
3209 if (device->nobarriers)
3210 return 0;
3211
3212 if (wait) {
3213 bio = device->flush_bio;
3214 if (!bio)
3215 return 0;
3216
3217 wait_for_completion(&device->flush_wait);
3218
3219 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
606686ee
JB		 3220 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3221 rcu_str_deref(device->name));
387125fc 3222 device->nobarriers = 1;
5af3e8cc 3223 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3224 ret = -EIO;
5af3e8cc
SB		 3225 btrfs_dev_stat_inc_and_print(device,
3226 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3227 }
3228
3229 /* drop the reference from the wait == 0 run */
3230 bio_put(bio);
3231 device->flush_bio = NULL;
3232
3233 return ret;
3234 }
3235
3236 /*
3237 * one reference for us, and we leave it for the
3238 * caller
3239 */
9c017abc 3240 device->flush_bio = NULL;
9be3395b 3241 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
387125fc
CM		 3242 if (!bio)
3243 return -ENOMEM;
3244
3245 bio->bi_end_io = btrfs_end_empty_barrier;
3246 bio->bi_bdev = device->bdev;
3247 init_completion(&device->flush_wait);
3248 bio->bi_private = &device->flush_wait;
3249 device->flush_bio = bio;
3250
3251 bio_get(bio);
21adbd5c 3252 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3253
3254 return 0;
3255}
3256
3257/*
3258 * send an empty flush down to each device in parallel,
3259 * then wait for them
3260 */
3261static int barrier_all_devices(struct btrfs_fs_info *info)
3262{
3263 struct list_head *head;
3264 struct btrfs_device *dev;
5af3e8cc
SB		 3265 int errors_send = 0;
3266 int errors_wait = 0;
387125fc
CM		 3267 int ret;
3268
3269 /* send down all the barriers */
3270 head = &info->fs_devices->devices;
3271 list_for_each_entry_rcu(dev, head, dev_list) {
3272 if (!dev->bdev) {
5af3e8cc 3273 errors_send++;
387125fc
CM		 3274 continue;
3275 }
3276 if (!dev->in_fs_metadata || !dev->writeable)
3277 continue;
3278
3279 ret = write_dev_flush(dev, 0);
3280 if (ret)
5af3e8cc 3281 errors_send++;
387125fc
CM		 3282 }
3283
3284 /* wait for all the barriers */
3285 list_for_each_entry_rcu(dev, head, dev_list) {
3286 if (!dev->bdev) {
5af3e8cc 3287 errors_wait++;
387125fc
CM		 3288 continue;
3289 }
3290 if (!dev->in_fs_metadata || !dev->writeable)
3291 continue;
3292
3293 ret = write_dev_flush(dev, 1);
3294 if (ret)
5af3e8cc 3295 errors_wait++;
387125fc 3296 }
5af3e8cc
SB		 3297 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3298 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3299 return -EIO;
3300 return 0;
3301}
3302
5af3e8cc
SB		 3303int btrfs_calc_num_tolerated_disk_barrier_failures(
3304 struct btrfs_fs_info *fs_info)
3305{
3306 struct btrfs_ioctl_space_info space;
3307 struct btrfs_space_info *sinfo;
3308 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3309 BTRFS_BLOCK_GROUP_SYSTEM,
3310 BTRFS_BLOCK_GROUP_METADATA,
3311 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3312 int num_types = 4;
3313 int i;
3314 int c;
3315 int num_tolerated_disk_barrier_failures =
3316 (int)fs_info->fs_devices->num_devices;
3317
3318 for (i = 0; i < num_types; i++) {
3319 struct btrfs_space_info *tmp;
3320
3321 sinfo = NULL;
3322 rcu_read_lock();
3323 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3324 if (tmp->flags == types[i]) {
3325 sinfo = tmp;
3326 break;
3327 }
3328 }
3329 rcu_read_unlock();
3330
3331 if (!sinfo)
3332 continue;
3333
3334 down_read(&sinfo->groups_sem);
3335 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3336 if (!list_empty(&sinfo->block_groups[c])) {
3337 u64 flags;
3338
3339 btrfs_get_block_group_info(
3340 &sinfo->block_groups[c], &space);
3341 if (space.total_bytes == 0 ||
3342 space.used_bytes == 0)
3343 continue;
3344 flags = space.flags;
3345 /*
3346 * return
3347 * 0: if dup, single or RAID0 is configured for
3348 * any of metadata, system or data, else
3349 * 1: if RAID5 is configured, or if RAID1 or
3350 * RAID10 is configured and only two mirrors
3351 * are used, else
3352 * 2: if RAID6 is configured, else
3353 * num_mirrors - 1: if RAID1 or RAID10 is
3354 * configured and more than
3355 * 2 mirrors are used.
3356 */
3357 if (num_tolerated_disk_barrier_failures > 0 &&
3358 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3359 BTRFS_BLOCK_GROUP_RAID0)) ||
3360 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3361 == 0)))
3362 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3363 else if (num_tolerated_disk_barrier_failures > 1) {
3364 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3365 BTRFS_BLOCK_GROUP_RAID5 |
3366 BTRFS_BLOCK_GROUP_RAID10)) {
3367 num_tolerated_disk_barrier_failures = 1;
3368 } else if (flags &
15b0a89d 3369 BTRFS_BLOCK_GROUP_RAID6) {
53b381b3
DW		 3370 num_tolerated_disk_barrier_failures = 2;
3371 }
3372 }
5af3e8cc
SB		 3373 }
3374 }
3375 up_read(&sinfo->groups_sem);
3376 }
3377
3378 return num_tolerated_disk_barrier_failures;
3379}
3380
48a3b636 3381static int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3382{
e5e9a520 3383 struct list_head *head;
f2984462 3384 struct btrfs_device *dev;
a061fc8d 3385 struct btrfs_super_block *sb;
f2984462 3386 struct btrfs_dev_item *dev_item;
f2984462
CM		 3387 int ret;
3388 int do_barriers;
a236aed1
CM		 3389 int max_errors;
3390 int total_errors = 0;
a061fc8d 3391 u64 flags;
f2984462
CM		 3392
3393 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3394 backup_super_roots(root->fs_info);
f2984462 3395
6c41761f 3396 sb = root->fs_info->super_for_commit;
a061fc8d 3397 dev_item = &sb->dev_item;
e5e9a520 3398
174ba509 3399 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3400 head = &root->fs_info->fs_devices->devices;
d7306801 3401 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
387125fc 3402
5af3e8cc
SB		 3403 if (do_barriers) {
3404 ret = barrier_all_devices(root->fs_info);
3405 if (ret) {
3406 mutex_unlock(
3407 &root->fs_info->fs_devices->device_list_mutex);
3408 btrfs_error(root->fs_info, ret,
3409 "errors while submitting device barriers.");
3410 return ret;
3411 }
3412 }
387125fc 3413
1f78160c 3414 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3415 if (!dev->bdev) {
3416 total_errors++;
3417 continue;
3418 }
2b82032c 3419 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3420 continue;
3421
2b82032c 3422 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3423 btrfs_set_stack_device_type(dev_item, dev->type);
3424 btrfs_set_stack_device_id(dev_item, dev->devid);
3425 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3426 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3427 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3428 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3429 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3430 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3431 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3432
a061fc8d
CM		 3433 flags = btrfs_super_flags(sb);
3434 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3435
a512bbf8 3436 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3437 if (ret)
3438 total_errors++;
f2984462 3439 }
a236aed1 3440 if (total_errors > max_errors) {
d397712b
CM		 3441 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3442 total_errors);
a724b436 3443 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
79787eaa 3444
9d565ba4
SB		 3445 /* FUA is masked off if unsupported and can't be the reason */
3446 btrfs_error(root->fs_info, -EIO,
3447 "%d errors while writing supers", total_errors);
3448 return -EIO;
a236aed1 3449 }
f2984462 3450
a512bbf8 3451 total_errors = 0;
1f78160c 3452 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3453 if (!dev->bdev)
3454 continue;
2b82032c 3455 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3456 continue;
3457
a512bbf8
YZ		 3458 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3459 if (ret)
3460 total_errors++;
f2984462 3461 }
174ba509 3462 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3463 if (total_errors > max_errors) {
79787eaa
JM		 3464 btrfs_error(root->fs_info, -EIO,
3465 "%d errors while writing supers", total_errors);
3466 return -EIO;
a236aed1 3467 }
f2984462
CM		 3468 return 0;
3469}
3470
a512bbf8
YZ		 3471int write_ctree_super(struct btrfs_trans_handle *trans,
3472 struct btrfs_root *root, int max_mirrors)
eb60ceac 3473{
e66f709b 3474 int ret;
5f39d397 3475
a512bbf8 3476 ret = write_all_supers(root, max_mirrors);
5f39d397 3477 return ret;
cfaa7295
CM		 3478}
3479
cb517eab
MX		 3480/* Drop a fs root from the radix tree and free it. */
3481void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3482 struct btrfs_root *root)
2619ba1f 3483{
4df27c4d 3484 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3485 radix_tree_delete(&fs_info->fs_roots_radix,
3486 (unsigned long)root->root_key.objectid);
4df27c4d 3487 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3488
3489 if (btrfs_root_refs(&root->root_item) == 0)
3490 synchronize_srcu(&fs_info->subvol_srcu);
3491
d7634482 3492 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3321719e
LB		 3493 btrfs_free_log(NULL, root);
3494 btrfs_free_log_root_tree(NULL, fs_info);
3495 }
3496
581bb050
LZ		 3497 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3498 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3499 free_fs_root(root);
4df27c4d
YZ		 3500}
3501
3502static void free_fs_root(struct btrfs_root *root)
3503{
82d5902d 3504 iput(root->cache_inode);
4df27c4d 3505 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1cb048f5
FDBM		 3506 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3507 root->orphan_block_rsv = NULL;
0ee5dc67
AV		 3508 if (root->anon_dev)
3509 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 3510 free_extent_buffer(root->node);
3511 free_extent_buffer(root->commit_root);
581bb050
LZ		 3512 kfree(root->free_ino_ctl);
3513 kfree(root->free_ino_pinned);
d397712b 3514 kfree(root->name);
b0feb9d9 3515 btrfs_put_fs_root(root);
2619ba1f
CM		 3516}
3517
cb517eab
MX		 3518void btrfs_free_fs_root(struct btrfs_root *root)
3519{
3520 free_fs_root(root);
2619ba1f
CM		 3521}
3522
c146afad 3523int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3524{
c146afad
YZ		 3525 u64 root_objectid = 0;
3526 struct btrfs_root *gang[8];
3527 int i;
3768f368 3528 int ret;
e089f05c 3529
c146afad
YZ		 3530 while (1) {
3531 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3532 (void **)gang, root_objectid,
3533 ARRAY_SIZE(gang));
3534 if (!ret)
3535 break;
5d4f98a2
YZ		 3536
3537 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 3538 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 3539 int err;
3540
c146afad 3541 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3542 err = btrfs_orphan_cleanup(gang[i]);
3543 if (err)
3544 return err;
c146afad
YZ		 3545 }
3546 root_objectid++;
3547 }
3548 return 0;
3549}
a2135011 3550
c146afad
YZ		 3551int btrfs_commit_super(struct btrfs_root *root)
3552{
3553 struct btrfs_trans_handle *trans;
3554 int ret;
a74a4b97 3555
c146afad 3556 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3557 btrfs_run_delayed_iputs(root);
c146afad 3558 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3559 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3560
3561 /* wait until ongoing cleanup work done */
3562 down_write(&root->fs_info->cleanup_work_sem);
3563 up_write(&root->fs_info->cleanup_work_sem);
3564
7a7eaa40 3565 trans = btrfs_join_transaction(root);
3612b495
TI		 3566 if (IS_ERR(trans))
3567 return PTR_ERR(trans);
54aa1f4d 3568 ret = btrfs_commit_transaction(trans, root);
79787eaa
JM		 3569 if (ret)
3570 return ret;
c146afad 3571 /* run commit again to drop the original snapshot */
7a7eaa40 3572 trans = btrfs_join_transaction(root);
3612b495
TI		 3573 if (IS_ERR(trans))
3574 return PTR_ERR(trans);
79787eaa
JM		 3575 ret = btrfs_commit_transaction(trans, root);
3576 if (ret)
3577 return ret;
79154b1b 3578 ret = btrfs_write_and_wait_transaction(NULL, root);
79787eaa
JM		 3579 if (ret) {
3580 btrfs_error(root->fs_info, ret,
3581 "Failed to sync btree inode to disk.");
3582 return ret;
3583 }
d6bfde87 3584
a512bbf8 3585 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 3586 return ret;
3587}
3588
3589int close_ctree(struct btrfs_root *root)
3590{
3591 struct btrfs_fs_info *fs_info = root->fs_info;
3592 int ret;
3593
3594 fs_info->closing = 1;
3595 smp_mb();
3596
803b2f54
SB		 3597 /* wait for the uuid_scan task to finish */
3598 down(&fs_info->uuid_tree_rescan_sem);
3599 /* avoid complains from lockdep et al., set sem back to initial state */
3600 up(&fs_info->uuid_tree_rescan_sem);
3601
837d5b6e 3602 /* pause restriper - we want to resume on mount */
aa1b8cd4 3603 btrfs_pause_balance(fs_info);
837d5b6e 3604
8dabb742
SB		 3605 btrfs_dev_replace_suspend_for_unmount(fs_info);
3606
aa1b8cd4 3607 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3608
3609 /* wait for any defraggers to finish */
3610 wait_event(fs_info->transaction_wait,
3611 (atomic_read(&fs_info->defrag_running) == 0));
3612
3613 /* clear out the rbtree of defraggable inodes */
26176e7c 3614 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3615
c146afad 3616 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3617 ret = btrfs_commit_super(root);
3618 if (ret)
3619 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3620 }
3621
87533c47 3622 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3623 btrfs_error_commit_super(root);
0f7d52f4 3624
300e4f8a
JB		 3625 btrfs_put_block_group_cache(fs_info);
3626
e3029d9f
AV		 3627 kthread_stop(fs_info->transaction_kthread);
3628 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3629
f25784b3
YZ		 3630 fs_info->closing = 2;
3631 smp_mb();
3632
bcef60f2
AJ		 3633 btrfs_free_qgroup_config(root->fs_info);
3634
963d678b
MX		 3635 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3636 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3637 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3638 }
bcc63abb 3639
13e6c37b 3640 btrfs_stop_all_workers(fs_info);
2932505a 3641
c146afad 3642 del_fs_roots(fs_info);
d10c5f31 3643
2b1360da
JB		 3644 btrfs_free_block_groups(fs_info);
3645
13e6c37b 3646 free_root_pointers(fs_info, 1);
9ad6b7bc 3647
13e6c37b 3648 iput(fs_info->btree_inode);
d6bfde87 3649
21adbd5c
SB		 3650#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3651 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3652 btrfsic_unmount(root, fs_info->fs_devices);
3653#endif
3654
dfe25020 3655 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3656 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3657
e2d84521 3658 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3659 percpu_counter_destroy(&fs_info->delalloc_bytes);
04160088 3660 bdi_destroy(&fs_info->bdi);
76dda93c 3661 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3662
53b381b3
DW		 3663 btrfs_free_stripe_hash_table(fs_info);
3664
1cb048f5
FDBM		 3665 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3666 root->orphan_block_rsv = NULL;
3667
eb60ceac
CM		 3668 return 0;
3669}
3670
b9fab919
CM		 3671int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3672 int atomic)
5f39d397 3673{
1259ab75 3674 int ret;
727011e0 3675 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3676
0b32f4bb 3677 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3678 if (!ret)
3679 return ret;
3680
3681 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3682 parent_transid, atomic);
3683 if (ret == -EAGAIN)
3684 return ret;
1259ab75 3685 return !ret;
5f39d397
CM		 3686}
3687
3688int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3689{
0b32f4bb 3690 return set_extent_buffer_uptodate(buf);
5f39d397 3691}
6702ed49 3692
5f39d397
CM		 3693void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3694{
06ea65a3 3695 struct btrfs_root *root;
5f39d397 3696 u64 transid = btrfs_header_generation(buf);
b9473439 3697 int was_dirty;
b4ce94de 3698
06ea65a3
JB		 3699#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3700 /*
3701 * This is a fast path so only do this check if we have sanity tests
3702 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3703 * outside of the sanity tests.
3704 */
3705 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3706 return;
3707#endif
3708 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
b9447ef8 3709 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3710 if (transid != root->fs_info->generation)
3711 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3712 "found %llu running %llu\n",
c1c9ff7c 3713 buf->start, transid, root->fs_info->generation);
0b32f4bb 3714 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3715 if (!was_dirty)
3716 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3717 buf->len,
3718 root->fs_info->dirty_metadata_batch);
eb60ceac
CM		 3719}
3720
b53d3f5d
LB		 3721static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3722 int flush_delayed)
16cdcec7
MX		 3723{
3724 /*
3725 * looks as though older kernels can get into trouble with
3726 * this code, they end up stuck in balance_dirty_pages forever
3727 */
e2d84521 3728 int ret;
16cdcec7
MX		 3729
3730 if (current->flags & PF_MEMALLOC)
3731 return;
3732
b53d3f5d
LB		 3733 if (flush_delayed)
3734 btrfs_balance_delayed_items(root);
16cdcec7 3735
e2d84521
MX		 3736 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3737 BTRFS_DIRTY_METADATA_THRESH);
3738 if (ret > 0) {
d0e1d66b
NJ		 3739 balance_dirty_pages_ratelimited(
3740 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3741 }
3742 return;
3743}
3744
b53d3f5d 3745void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3746{
b53d3f5d
LB		 3747 __btrfs_btree_balance_dirty(root, 1);
3748}
585ad2c3 3749
b53d3f5d
LB		 3750void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3751{
3752 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3753}
6b80053d 3754
ca7a79ad 3755int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3756{
727011e0 3757 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3758 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3759}
0da5468f 3760
fcd1f065 3761static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3762 int read_only)
3763{
1104a885
DS		 3764 /*
3765 * Placeholder for checks
3766 */
fcd1f065 3767 return 0;
acce952b 3768}
3769
48a3b636 3770static void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3771{
acce952b 3772 mutex_lock(&root->fs_info->cleaner_mutex);
3773 btrfs_run_delayed_iputs(root);
3774 mutex_unlock(&root->fs_info->cleaner_mutex);
3775
3776 down_write(&root->fs_info->cleanup_work_sem);
3777 up_write(&root->fs_info->cleanup_work_sem);
3778
3779 /* cleanup FS via transaction */
3780 btrfs_cleanup_transaction(root);
acce952b 3781}
3782
569e0f35
JB		 3783static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3784 struct btrfs_root *root)
acce952b 3785{
3786 struct btrfs_inode *btrfs_inode;
3787 struct list_head splice;
3788
3789 INIT_LIST_HEAD(&splice);
3790
3791 mutex_lock(&root->fs_info->ordered_operations_mutex);
199c2a9c 3792 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3793
569e0f35 3794 list_splice_init(&t->ordered_operations, &splice);
acce952b 3795 while (!list_empty(&splice)) {
3796 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3797 ordered_operations);
3798
3799 list_del_init(&btrfs_inode->ordered_operations);
199c2a9c 3800 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3801
3802 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3803
199c2a9c 3804 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3805 }
3806
199c2a9c 3807 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3808 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3809}
3810
143bede5 3811static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3812{
acce952b 3813 struct btrfs_ordered_extent *ordered;
acce952b 3814
199c2a9c 3815 spin_lock(&root->ordered_extent_lock);
779880ef
JB		 3816 /*
3817 * This will just short circuit the ordered completion stuff which will
3818 * make sure the ordered extent gets properly cleaned up.
3819 */
199c2a9c 3820 list_for_each_entry(ordered, &root->ordered_extents,
779880ef
JB		 3821 root_extent_list)
3822 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
199c2a9c
MX		 3823 spin_unlock(&root->ordered_extent_lock);
3824}
3825
3826static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3827{
3828 struct btrfs_root *root;
3829 struct list_head splice;
3830
3831 INIT_LIST_HEAD(&splice);
3832
3833 spin_lock(&fs_info->ordered_root_lock);
3834 list_splice_init(&fs_info->ordered_roots, &splice);
3835 while (!list_empty(&splice)) {
3836 root = list_first_entry(&splice, struct btrfs_root,
3837 ordered_root);
1de2cfde
JB		 3838 list_move_tail(&root->ordered_root,
3839 &fs_info->ordered_roots);
199c2a9c
MX		 3840
3841 btrfs_destroy_ordered_extents(root);
3842
3843 cond_resched_lock(&fs_info->ordered_root_lock);
3844 }
3845 spin_unlock(&fs_info->ordered_root_lock);
acce952b 3846}
3847
35a3621b
SB		 3848static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3849 struct btrfs_root *root)
acce952b 3850{
3851 struct rb_node *node;
3852 struct btrfs_delayed_ref_root *delayed_refs;
3853 struct btrfs_delayed_ref_node *ref;
3854 int ret = 0;
3855
3856 delayed_refs = &trans->delayed_refs;
3857
3858 spin_lock(&delayed_refs->lock);
3859 if (delayed_refs->num_entries == 0) {
cfece4db 3860 spin_unlock(&delayed_refs->lock);
acce952b 3861 printk(KERN_INFO "delayed_refs has NO entry\n");
3862 return ret;
3863 }
3864
b939d1ab 3865 while ((node = rb_first(&delayed_refs->root)) != NULL) {
eb12db69 3866 struct btrfs_delayed_ref_head *head = NULL;
e78417d1 3867 bool pin_bytes = false;
acce952b 3868
eb12db69 3869 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
acce952b 3870 atomic_set(&ref->refs, 1);
3871 if (btrfs_delayed_ref_is_head(ref)) {
acce952b 3872
3873 head = btrfs_delayed_node_to_head(ref);
b939d1ab
JB		 3874 if (!mutex_trylock(&head->mutex)) {
3875 atomic_inc(&ref->refs);
3876 spin_unlock(&delayed_refs->lock);
3877
3878 /* Need to wait for the delayed ref to run */
3879 mutex_lock(&head->mutex);
3880 mutex_unlock(&head->mutex);
3881 btrfs_put_delayed_ref(ref);
3882
e18fca73 3883 spin_lock(&delayed_refs->lock);
b939d1ab
JB		 3884 continue;
3885 }
3886
54067ae9 3887 if (head->must_insert_reserved)
e78417d1 3888 pin_bytes = true;
78a6184a 3889 btrfs_free_delayed_extent_op(head->extent_op);
acce952b 3890 delayed_refs->num_heads--;
3891 if (list_empty(&head->cluster))
3892 delayed_refs->num_heads_ready--;
3893 list_del_init(&head->cluster);
acce952b 3894 }
eb12db69 3895
b939d1ab
JB		 3896 ref->in_tree = 0;
3897 rb_erase(&ref->rb_node, &delayed_refs->root);
3898 delayed_refs->num_entries--;
acce952b 3899 spin_unlock(&delayed_refs->lock);
e78417d1
JB		 3900 if (head) {
3901 if (pin_bytes)
3902 btrfs_pin_extent(root, ref->bytenr,
3903 ref->num_bytes, 1);
3904 mutex_unlock(&head->mutex);
3905 }
acce952b 3906 btrfs_put_delayed_ref(ref);
3907
3908 cond_resched();
3909 spin_lock(&delayed_refs->lock);
3910 }
3911
3912 spin_unlock(&delayed_refs->lock);
3913
3914 return ret;
3915}
3916
143bede5 3917static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3918{
3919 struct btrfs_inode *btrfs_inode;
3920 struct list_head splice;
3921
3922 INIT_LIST_HEAD(&splice);
3923
eb73c1b7
MX		 3924 spin_lock(&root->delalloc_lock);
3925 list_splice_init(&root->delalloc_inodes, &splice);
acce952b 3926
3927 while (!list_empty(&splice)) {
eb73c1b7
MX		 3928 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3929 delalloc_inodes);
acce952b 3930
3931 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3932 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3933 &btrfs_inode->runtime_flags);
eb73c1b7 3934 spin_unlock(&root->delalloc_lock);
acce952b 3935
3936 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3937
eb73c1b7 3938 spin_lock(&root->delalloc_lock);
acce952b 3939 }
3940
eb73c1b7
MX		 3941 spin_unlock(&root->delalloc_lock);
3942}
3943
3944static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3945{
3946 struct btrfs_root *root;
3947 struct list_head splice;
3948
3949 INIT_LIST_HEAD(&splice);
3950
3951 spin_lock(&fs_info->delalloc_root_lock);
3952 list_splice_init(&fs_info->delalloc_roots, &splice);
3953 while (!list_empty(&splice)) {
3954 root = list_first_entry(&splice, struct btrfs_root,
3955 delalloc_root);
3956 list_del_init(&root->delalloc_root);
3957 root = btrfs_grab_fs_root(root);
3958 BUG_ON(!root);
3959 spin_unlock(&fs_info->delalloc_root_lock);
3960
3961 btrfs_destroy_delalloc_inodes(root);
3962 btrfs_put_fs_root(root);
3963
3964 spin_lock(&fs_info->delalloc_root_lock);
3965 }
3966 spin_unlock(&fs_info->delalloc_root_lock);
acce952b 3967}
3968
3969static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3970 struct extent_io_tree *dirty_pages,
3971 int mark)
3972{
3973 int ret;
acce952b 3974 struct extent_buffer *eb;
3975 u64 start = 0;
3976 u64 end;
acce952b 3977
3978 while (1) {
3979 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3980 mark, NULL);
acce952b 3981 if (ret)
3982 break;
3983
3984 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3985 while (start <= end) {
fd8b2b61
JB		 3986 eb = btrfs_find_tree_block(root, start,
3987 root->leafsize);
69a85bd8 3988 start += root->leafsize;
fd8b2b61 3989 if (!eb)
acce952b 3990 continue;
fd8b2b61 3991 wait_on_extent_buffer_writeback(eb);
acce952b 3992
fd8b2b61
JB		 3993 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3994 &eb->bflags))
3995 clear_extent_buffer_dirty(eb);
3996 free_extent_buffer_stale(eb);
acce952b 3997 }
3998 }
3999
4000 return ret;
4001}
4002
4003static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4004 struct extent_io_tree *pinned_extents)
4005{
4006 struct extent_io_tree *unpin;
4007 u64 start;
4008 u64 end;
4009 int ret;
ed0eaa14 4010 bool loop = true;
acce952b 4011
4012 unpin = pinned_extents;
ed0eaa14 4013again:
acce952b 4014 while (1) {
4015 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 4016 EXTENT_DIRTY, NULL);
acce952b 4017 if (ret)
4018 break;
4019
4020 /* opt_discard */
5378e607
LD		 4021 if (btrfs_test_opt(root, DISCARD))
4022 ret = btrfs_error_discard_extent(root, start,
4023 end + 1 - start,
4024 NULL);
acce952b 4025
4026 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4027 btrfs_error_unpin_extent_range(root, start, end);
4028 cond_resched();
4029 }
4030
ed0eaa14
LB		 4031 if (loop) {
4032 if (unpin == &root->fs_info->freed_extents[0])
4033 unpin = &root->fs_info->freed_extents[1];
4034 else
4035 unpin = &root->fs_info->freed_extents[0];
4036 loop = false;
4037 goto again;
4038 }
4039
acce952b 4040 return 0;
4041}
4042
49b25e05
JM		 4043void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4044 struct btrfs_root *root)
4045{
724e2315
JB		 4046 btrfs_destroy_ordered_operations(cur_trans, root);
4047
49b25e05 4048 btrfs_destroy_delayed_refs(cur_trans, root);
49b25e05 4049
4a9d8bde 4050 cur_trans->state = TRANS_STATE_COMMIT_START;
d7096fc3 4051 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 4052
4a9d8bde 4053 cur_trans->state = TRANS_STATE_UNBLOCKED;
d7096fc3 4054 wake_up(&root->fs_info->transaction_wait);
49b25e05 4055
67cde344
MX		 4056 btrfs_destroy_delayed_inodes(root);
4057 btrfs_assert_delayed_root_empty(root);
49b25e05 4058
49b25e05
JM		 4059 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4060 EXTENT_DIRTY);
6e841e32
LB		 4061 btrfs_destroy_pinned_extent(root,
4062 root->fs_info->pinned_extents);
49b25e05 4063
4a9d8bde
MX		 4064 cur_trans->state =TRANS_STATE_COMPLETED;
4065 wake_up(&cur_trans->commit_wait);
4066
49b25e05
JM		 4067 /*
4068 memset(cur_trans, 0, sizeof(*cur_trans));
4069 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4070 */
4071}
4072
48a3b636 4073static int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 4074{
4075 struct btrfs_transaction *t;
acce952b 4076
acce952b 4077 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4078
a4abeea4 4079 spin_lock(&root->fs_info->trans_lock);
724e2315
JB		 4080 while (!list_empty(&root->fs_info->trans_list)) {
4081 t = list_first_entry(&root->fs_info->trans_list,
4082 struct btrfs_transaction, list);
4083 if (t->state >= TRANS_STATE_COMMIT_START) {
4084 atomic_inc(&t->use_count);
4085 spin_unlock(&root->fs_info->trans_lock);
4086 btrfs_wait_for_commit(root, t->transid);
4087 btrfs_put_transaction(t);
4088 spin_lock(&root->fs_info->trans_lock);
4089 continue;
4090 }
4091 if (t == root->fs_info->running_transaction) {
4092 t->state = TRANS_STATE_COMMIT_DOING;
4093 spin_unlock(&root->fs_info->trans_lock);
4094 /*
4095 * We wait for 0 num_writers since we don't hold a trans
4096 * handle open currently for this transaction.
4097 */
4098 wait_event(t->writer_wait,
4099 atomic_read(&t->num_writers) == 0);
4100 } else {
4101 spin_unlock(&root->fs_info->trans_lock);
4102 }
4103 btrfs_cleanup_one_transaction(t, root);
4a9d8bde 4104
724e2315
JB		 4105 spin_lock(&root->fs_info->trans_lock);
4106 if (t == root->fs_info->running_transaction)
4107 root->fs_info->running_transaction = NULL;
acce952b 4108 list_del_init(&t->list);
724e2315 4109 spin_unlock(&root->fs_info->trans_lock);
acce952b 4110
724e2315
JB		 4111 btrfs_put_transaction(t);
4112 trace_btrfs_transaction_commit(root);
4113 spin_lock(&root->fs_info->trans_lock);
4114 }
4115 spin_unlock(&root->fs_info->trans_lock);
4116 btrfs_destroy_all_ordered_extents(root->fs_info);
4117 btrfs_destroy_delayed_inodes(root);
4118 btrfs_assert_delayed_root_empty(root);
4119 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4120 btrfs_destroy_all_delalloc_inodes(root->fs_info);
acce952b 4121 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 4122
4123 return 0;
4124}
4125
d1310b2e 4126static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 4127 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 4128 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 4129 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 4130 /* note we're sharing with inode.c for the merge bio hook */
4131 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 4132};
Linux 6.x block layer and io_uring tree(s)