projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: Make btrfs_header_fsid() return unsigned long
[linux-2.6-block.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
163e783e 29#include <linux/crc32c.h>
5a0e3ad6 30#include <linux/slab.h>
784b4e29 31#include <linux/migrate.h>
7a36ddec 32#include <linux/ratelimit.h>
6463fe58 33#include <linux/uuid.h>
803b2f54 34#include <linux/semaphore.h>
7e75bf3f 35#include <asm/unaligned.h>
4b4e25f2 36#include "compat.h"
eb60ceac
CM		 37#include "ctree.h"
38#include "disk-io.h"
e089f05c 39#include "transaction.h"
0f7d52f4 40#include "btrfs_inode.h"
0b86a832 41#include "volumes.h"
db94535d 42#include "print-tree.h"
8b712842 43#include "async-thread.h"
925baedd 44#include "locking.h"
e02119d5 45#include "tree-log.h"
fa9c0d79 46#include "free-space-cache.h"
581bb050 47#include "inode-map.h"
21adbd5c 48#include "check-integrity.h"
606686ee 49#include "rcu-string.h"
8dabb742 50#include "dev-replace.h"
53b381b3 51#include "raid56.h"
eb60ceac 52
de0022b9
JB		 53#ifdef CONFIG_X86
54#include <asm/cpufeature.h>
55#endif
56
d1310b2e 57static struct extent_io_ops btree_extent_io_ops;
8b712842 58static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 59static void free_fs_root(struct btrfs_root *root);
fcd1f065 60static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 61 int read_only);
569e0f35
JB		 62static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
63 struct btrfs_root *root);
143bede5 64static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
acce952b 65static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
66 struct btrfs_root *root);
aec8030a 67static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
143bede5 68static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
acce952b 69static int btrfs_destroy_marked_extents(struct btrfs_root *root,
70 struct extent_io_tree *dirty_pages,
71 int mark);
72static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
73 struct extent_io_tree *pinned_extents);
48a3b636
ES		 74static int btrfs_cleanup_transaction(struct btrfs_root *root);
75static void btrfs_error_commit_super(struct btrfs_root *root);
ce9adaa5 76
d352ac68
CM		 77/*
78 * end_io_wq structs are used to do processing in task context when an IO is
79 * complete. This is used during reads to verify checksums, and it is used
80 * by writes to insert metadata for new file extents after IO is complete.
81 */
ce9adaa5
CM		 82struct end_io_wq {
83 struct bio *bio;
84 bio_end_io_t *end_io;
85 void *private;
86 struct btrfs_fs_info *info;
87 int error;
22c59948 88 int metadata;
ce9adaa5 89 struct list_head list;
8b712842 90 struct btrfs_work work;
ce9adaa5 91};
0da5468f 92
d352ac68
CM		 93/*
94 * async submit bios are used to offload expensive checksumming
95 * onto the worker threads. They checksum file and metadata bios
96 * just before they are sent down the IO stack.
97 */
44b8bd7e
CM		 98struct async_submit_bio {
99 struct inode *inode;
100 struct bio *bio;
101 struct list_head list;
4a69a410
CM		 102 extent_submit_bio_hook_t *submit_bio_start;
103 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 104 int rw;
105 int mirror_num;
c8b97818 106 unsigned long bio_flags;
eaf25d93
CM		 107 /*
108 * bio_offset is optional, can be used if the pages in the bio
109 * can't tell us where in the file the bio should go
110 */
111 u64 bio_offset;
8b712842 112 struct btrfs_work work;
79787eaa 113 int error;
44b8bd7e
CM		 114};
115
85d4e461
CM		 116/*
117 * Lockdep class keys for extent_buffer->lock's in this root. For a given
118 * eb, the lockdep key is determined by the btrfs_root it belongs to and
119 * the level the eb occupies in the tree.
120 *
121 * Different roots are used for different purposes and may nest inside each
122 * other and they require separate keysets. As lockdep keys should be
123 * static, assign keysets according to the purpose of the root as indicated
124 * by btrfs_root->objectid. This ensures that all special purpose roots
125 * have separate keysets.
4008c04a 126 *
85d4e461
CM		 127 * Lock-nesting across peer nodes is always done with the immediate parent
128 * node locked thus preventing deadlock. As lockdep doesn't know this, use
129 * subclass to avoid triggering lockdep warning in such cases.
4008c04a 130 *
85d4e461
CM		 131 * The key is set by the readpage_end_io_hook after the buffer has passed
132 * csum validation but before the pages are unlocked. It is also set by
133 * btrfs_init_new_buffer on freshly allocated blocks.
4008c04a 134 *
85d4e461
CM		 135 * We also add a check to make sure the highest level of the tree is the
136 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
137 * needs update as well.
4008c04a
CM		 138 */
139#ifdef CONFIG_DEBUG_LOCK_ALLOC
140# if BTRFS_MAX_LEVEL != 8
141# error
142# endif
85d4e461
CM		 143
144static struct btrfs_lockdep_keyset {
145 u64 id; /* root objectid */
146 const char *name_stem; /* lock name stem */
147 char names[BTRFS_MAX_LEVEL + 1][20];
148 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
149} btrfs_lockdep_keysets[] = {
150 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
151 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
152 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
153 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
154 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
155 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
60b62978 156 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
85d4e461
CM		 157 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
158 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
159 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
160 { .id = 0, .name_stem = "tree" },
4008c04a 161};
85d4e461
CM		 162
163void __init btrfs_init_lockdep(void)
164{
165 int i, j;
166
167 /* initialize lockdep class names */
168 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
169 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
170
171 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
172 snprintf(ks->names[j], sizeof(ks->names[j]),
173 "btrfs-%s-%02d", ks->name_stem, j);
174 }
175}
176
177void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
178 int level)
179{
180 struct btrfs_lockdep_keyset *ks;
181
182 BUG_ON(level >= ARRAY_SIZE(ks->keys));
183
184 /* find the matching keyset, id 0 is the default entry */
185 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
186 if (ks->id == objectid)
187 break;
188
189 lockdep_set_class_and_name(&eb->lock,
190 &ks->keys[level], ks->names[level]);
191}
192
4008c04a
CM		 193#endif
194
d352ac68
CM		 195/*
196 * extents on the btree inode are pretty simple, there's one extent
197 * that covers the entire device
198 */
b2950863 199static struct extent_map *btree_get_extent(struct inode *inode,
306e16ce 200 struct page *page, size_t pg_offset, u64 start, u64 len,
b2950863 201 int create)
7eccb903 202{
5f39d397
CM		 203 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
204 struct extent_map *em;
205 int ret;
206
890871be 207 read_lock(&em_tree->lock);
d1310b2e 208 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 209 if (em) {
210 em->bdev =
211 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 212 read_unlock(&em_tree->lock);
5f39d397 213 goto out;
a061fc8d 214 }
890871be 215 read_unlock(&em_tree->lock);
7b13b7b1 216
172ddd60 217 em = alloc_extent_map();
5f39d397
CM		 218 if (!em) {
219 em = ERR_PTR(-ENOMEM);
220 goto out;
221 }
222 em->start = 0;
0afbaf8c 223 em->len = (u64)-1;
c8b97818 224 em->block_len = (u64)-1;
5f39d397 225 em->block_start = 0;
a061fc8d 226 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 227
890871be 228 write_lock(&em_tree->lock);
09a2a8f9 229 ret = add_extent_mapping(em_tree, em, 0);
5f39d397
CM		 230 if (ret == -EEXIST) {
231 free_extent_map(em);
7b13b7b1 232 em = lookup_extent_mapping(em_tree, start, len);
b4f359ab 233 if (!em)
0433f20d 234 em = ERR_PTR(-EIO);
5f39d397 235 } else if (ret) {
7b13b7b1 236 free_extent_map(em);
0433f20d 237 em = ERR_PTR(ret);
5f39d397 238 }
890871be 239 write_unlock(&em_tree->lock);
7b13b7b1 240
5f39d397
CM		 241out:
242 return em;
7eccb903
CM		 243}
244
b0496686 245u32 btrfs_csum_data(char *data, u32 seed, size_t len)
19c00ddc 246{
163e783e 247 return crc32c(seed, data, len);
19c00ddc
CM		 248}
249
250void btrfs_csum_final(u32 crc, char *result)
251{
7e75bf3f 252 put_unaligned_le32(~crc, result);
19c00ddc
CM		 253}
254
d352ac68
CM		 255/*
256 * compute the csum for a btree block, and either verify it or write it
257 * into the csum field of the block.
258 */
19c00ddc
CM		 259static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
260 int verify)
261{
6c41761f 262 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
607d432d 263 char *result = NULL;
19c00ddc
CM		 264 unsigned long len;
265 unsigned long cur_len;
266 unsigned long offset = BTRFS_CSUM_SIZE;
19c00ddc
CM		 267 char *kaddr;
268 unsigned long map_start;
269 unsigned long map_len;
270 int err;
271 u32 crc = ~(u32)0;
607d432d 272 unsigned long inline_result;
19c00ddc
CM		 273
274 len = buf->len - offset;
d397712b 275 while (len > 0) {
19c00ddc 276 err = map_private_extent_buffer(buf, offset, 32,
a6591715 277 &kaddr, &map_start, &map_len);
d397712b 278 if (err)
19c00ddc 279 return 1;
19c00ddc 280 cur_len = min(len, map_len - (offset - map_start));
b0496686 281 crc = btrfs_csum_data(kaddr + offset - map_start,
19c00ddc
CM		 282 crc, cur_len);
283 len -= cur_len;
284 offset += cur_len;
19c00ddc 285 }
607d432d
JB		 286 if (csum_size > sizeof(inline_result)) {
287 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
288 if (!result)
289 return 1;
290 } else {
291 result = (char *)&inline_result;
292 }
293
19c00ddc
CM		 294 btrfs_csum_final(crc, result);
295
296 if (verify) {
607d432d 297 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 298 u32 val;
299 u32 found = 0;
607d432d 300 memcpy(&found, result, csum_size);
e4204ded 301
607d432d 302 read_extent_buffer(buf, &val, 0, csum_size);
7a36ddec 303 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
193f284d
CM		 304 "failed on %llu wanted %X found %X "
305 "level %d\n",
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
fba6aa75 498 read_extent_buffer(eb, fsid, btrfs_header_fsid(eb), 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;
d0c803c4 1231 extent_io_tree_init(&root->dirty_log_pages,
f993c883 1232 fs_info->btree_inode->i_mapping);
017e5369 1233
3768f368
CM		 1234 memset(&root->root_key, 0, sizeof(root->root_key));
1235 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1236 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1237 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 1238 root->defrag_trans_start = fs_info->generation;
58176a96 1239 init_completion(&root->kobj_unregister);
6702ed49 1240 root->defrag_running = 0;
4d775673 1241 root->root_key.objectid = objectid;
0ee5dc67 1242 root->anon_dev = 0;
8ea05e3a 1243
5f3ab90a 1244 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1245}
1246
f84a8bd6 1247static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1248{
1249 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1250 if (root)
1251 root->fs_info = fs_info;
1252 return root;
1253}
1254
20897f5c
AJ		 1255struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1256 struct btrfs_fs_info *fs_info,
1257 u64 objectid)
1258{
1259 struct extent_buffer *leaf;
1260 struct btrfs_root *tree_root = fs_info->tree_root;
1261 struct btrfs_root *root;
1262 struct btrfs_key key;
1263 int ret = 0;
1264 u64 bytenr;
6463fe58 1265 uuid_le uuid;
20897f5c
AJ		 1266
1267 root = btrfs_alloc_root(fs_info);
1268 if (!root)
1269 return ERR_PTR(-ENOMEM);
1270
1271 __setup_root(tree_root->nodesize, tree_root->leafsize,
1272 tree_root->sectorsize, tree_root->stripesize,
1273 root, fs_info, objectid);
1274 root->root_key.objectid = objectid;
1275 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1276 root->root_key.offset = 0;
1277
1278 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1279 0, objectid, NULL, 0, 0, 0);
1280 if (IS_ERR(leaf)) {
1281 ret = PTR_ERR(leaf);
1dd05682 1282 leaf = NULL;
20897f5c
AJ		 1283 goto fail;
1284 }
1285
1286 bytenr = leaf->start;
1287 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1288 btrfs_set_header_bytenr(leaf, leaf->start);
1289 btrfs_set_header_generation(leaf, trans->transid);
1290 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1291 btrfs_set_header_owner(leaf, objectid);
1292 root->node = leaf;
1293
fba6aa75 1294 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(leaf),
20897f5c
AJ		 1295 BTRFS_FSID_SIZE);
1296 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1297 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1298 BTRFS_UUID_SIZE);
1299 btrfs_mark_buffer_dirty(leaf);
1300
1301 root->commit_root = btrfs_root_node(root);
1302 root->track_dirty = 1;
1303
1304
1305 root->root_item.flags = 0;
1306 root->root_item.byte_limit = 0;
1307 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1308 btrfs_set_root_generation(&root->root_item, trans->transid);
1309 btrfs_set_root_level(&root->root_item, 0);
1310 btrfs_set_root_refs(&root->root_item, 1);
1311 btrfs_set_root_used(&root->root_item, leaf->len);
1312 btrfs_set_root_last_snapshot(&root->root_item, 0);
1313 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1314 uuid_le_gen(&uuid);
1315 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1316 root->root_item.drop_level = 0;
1317
1318 key.objectid = objectid;
1319 key.type = BTRFS_ROOT_ITEM_KEY;
1320 key.offset = 0;
1321 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1322 if (ret)
1323 goto fail;
1324
1325 btrfs_tree_unlock(leaf);
1326
1dd05682
TI		 1327 return root;
1328
20897f5c 1329fail:
1dd05682
TI		 1330 if (leaf) {
1331 btrfs_tree_unlock(leaf);
1332 free_extent_buffer(leaf);
1333 }
1334 kfree(root);
20897f5c 1335
1dd05682 1336 return ERR_PTR(ret);
20897f5c
AJ		 1337}
1338
7237f183
YZ		 1339static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1340 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1341{
1342 struct btrfs_root *root;
1343 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1344 struct extent_buffer *leaf;
e02119d5 1345
6f07e42e 1346 root = btrfs_alloc_root(fs_info);
e02119d5 1347 if (!root)
7237f183 1348 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1349
1350 __setup_root(tree_root->nodesize, tree_root->leafsize,
1351 tree_root->sectorsize, tree_root->stripesize,
1352 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1353
1354 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1355 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1356 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1357 /*
1358 * log trees do not get reference counted because they go away
1359 * before a real commit is actually done. They do store pointers
1360 * to file data extents, and those reference counts still get
1361 * updated (along with back refs to the log tree).
1362 */
e02119d5
CM		 1363 root->ref_cows = 0;
1364
5d4f98a2 1365 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1366 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1367 0, 0, 0);
7237f183
YZ		 1368 if (IS_ERR(leaf)) {
1369 kfree(root);
1370 return ERR_CAST(leaf);
1371 }
e02119d5 1372
5d4f98a2
YZ		 1373 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1374 btrfs_set_header_bytenr(leaf, leaf->start);
1375 btrfs_set_header_generation(leaf, trans->transid);
1376 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1377 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1378 root->node = leaf;
e02119d5
CM		 1379
1380 write_extent_buffer(root->node, root->fs_info->fsid,
fba6aa75 1381 btrfs_header_fsid(root->node), BTRFS_FSID_SIZE);
e02119d5
CM		 1382 btrfs_mark_buffer_dirty(root->node);
1383 btrfs_tree_unlock(root->node);
7237f183
YZ		 1384 return root;
1385}
1386
1387int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1388 struct btrfs_fs_info *fs_info)
1389{
1390 struct btrfs_root *log_root;
1391
1392 log_root = alloc_log_tree(trans, fs_info);
1393 if (IS_ERR(log_root))
1394 return PTR_ERR(log_root);
1395 WARN_ON(fs_info->log_root_tree);
1396 fs_info->log_root_tree = log_root;
1397 return 0;
1398}
1399
1400int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1401 struct btrfs_root *root)
1402{
1403 struct btrfs_root *log_root;
1404 struct btrfs_inode_item *inode_item;
1405
1406 log_root = alloc_log_tree(trans, root->fs_info);
1407 if (IS_ERR(log_root))
1408 return PTR_ERR(log_root);
1409
1410 log_root->last_trans = trans->transid;
1411 log_root->root_key.offset = root->root_key.objectid;
1412
1413 inode_item = &log_root->root_item.inode;
3cae210f
QW		 1414 btrfs_set_stack_inode_generation(inode_item, 1);
1415 btrfs_set_stack_inode_size(inode_item, 3);
1416 btrfs_set_stack_inode_nlink(inode_item, 1);
1417 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1418 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
7237f183 1419
5d4f98a2 1420 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1421
1422 WARN_ON(root->log_root);
1423 root->log_root = log_root;
1424 root->log_transid = 0;
257c62e1 1425 root->last_log_commit = 0;
e02119d5
CM		 1426 return 0;
1427}
1428
35a3621b
SB		 1429static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1430 struct btrfs_key *key)
e02119d5
CM		 1431{
1432 struct btrfs_root *root;
1433 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1434 struct btrfs_path *path;
84234f3a 1435 u64 generation;
db94535d 1436 u32 blocksize;
cb517eab 1437 int ret;
0f7d52f4 1438
cb517eab
MX		 1439 path = btrfs_alloc_path();
1440 if (!path)
0f7d52f4 1441 return ERR_PTR(-ENOMEM);
cb517eab
MX		 1442
1443 root = btrfs_alloc_root(fs_info);
1444 if (!root) {
1445 ret = -ENOMEM;
1446 goto alloc_fail;
0f7d52f4
CM		 1447 }
1448
db94535d 1449 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb 1450 tree_root->sectorsize, tree_root->stripesize,
cb517eab 1451 root, fs_info, key->objectid);
0f7d52f4 1452
cb517eab
MX		 1453 ret = btrfs_find_root(tree_root, key, path,
1454 &root->root_item, &root->root_key);
0f7d52f4 1455 if (ret) {
13a8a7c8
YZ		 1456 if (ret > 0)
1457 ret = -ENOENT;
cb517eab 1458 goto find_fail;
0f7d52f4 1459 }
13a8a7c8 1460
84234f3a 1461 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1462 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1463 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1464 blocksize, generation);
cb517eab
MX		 1465 if (!root->node) {
1466 ret = -ENOMEM;
1467 goto find_fail;
1468 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1469 ret = -EIO;
1470 goto read_fail;
416bc658 1471 }
5d4f98a2 1472 root->commit_root = btrfs_root_node(root);
13a8a7c8 1473out:
cb517eab
MX		 1474 btrfs_free_path(path);
1475 return root;
1476
1477read_fail:
1478 free_extent_buffer(root->node);
1479find_fail:
1480 kfree(root);
1481alloc_fail:
1482 root = ERR_PTR(ret);
1483 goto out;
1484}
1485
1486struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1487 struct btrfs_key *location)
1488{
1489 struct btrfs_root *root;
1490
1491 root = btrfs_read_tree_root(tree_root, location);
1492 if (IS_ERR(root))
1493 return root;
1494
1495 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1496 root->ref_cows = 1;
08fe4db1
LZ		 1497 btrfs_check_and_init_root_item(&root->root_item);
1498 }
13a8a7c8 1499
5eda7b5e
CM		 1500 return root;
1501}
1502
cb517eab
MX		 1503int btrfs_init_fs_root(struct btrfs_root *root)
1504{
1505 int ret;
1506
1507 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1508 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1509 GFP_NOFS);
1510 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1511 ret = -ENOMEM;
1512 goto fail;
1513 }
1514
1515 btrfs_init_free_ino_ctl(root);
1516 mutex_init(&root->fs_commit_mutex);
1517 spin_lock_init(&root->cache_lock);
1518 init_waitqueue_head(&root->cache_wait);
1519
1520 ret = get_anon_bdev(&root->anon_dev);
1521 if (ret)
1522 goto fail;
1523 return 0;
1524fail:
1525 kfree(root->free_ino_ctl);
1526 kfree(root->free_ino_pinned);
1527 return ret;
1528}
1529
171170c1
ST		 1530static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1531 u64 root_id)
cb517eab
MX		 1532{
1533 struct btrfs_root *root;
1534
1535 spin_lock(&fs_info->fs_roots_radix_lock);
1536 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1537 (unsigned long)root_id);
1538 spin_unlock(&fs_info->fs_roots_radix_lock);
1539 return root;
1540}
1541
1542int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1543 struct btrfs_root *root)
1544{
1545 int ret;
1546
1547 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1548 if (ret)
1549 return ret;
1550
1551 spin_lock(&fs_info->fs_roots_radix_lock);
1552 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1553 (unsigned long)root->root_key.objectid,
1554 root);
1555 if (ret == 0)
1556 root->in_radix = 1;
1557 spin_unlock(&fs_info->fs_roots_radix_lock);
1558 radix_tree_preload_end();
1559
1560 return ret;
1561}
1562
edbd8d4e
CM		 1563struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1564 struct btrfs_key *location)
5eda7b5e
CM		 1565{
1566 struct btrfs_root *root;
1567 int ret;
1568
edbd8d4e
CM		 1569 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1570 return fs_info->tree_root;
1571 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1572 return fs_info->extent_root;
8f18cf13
CM		 1573 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1574 return fs_info->chunk_root;
1575 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1576 return fs_info->dev_root;
0403e47e
YZ		 1577 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1578 return fs_info->csum_root;
bcef60f2
AJ		 1579 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1580 return fs_info->quota_root ? fs_info->quota_root :
1581 ERR_PTR(-ENOENT);
f7a81ea4
SB		 1582 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1583 return fs_info->uuid_root ? fs_info->uuid_root :
1584 ERR_PTR(-ENOENT);
4df27c4d 1585again:
cb517eab 1586 root = btrfs_lookup_fs_root(fs_info, location->objectid);
5eda7b5e
CM		 1587 if (root)
1588 return root;
1589
cb517eab 1590 root = btrfs_read_fs_root(fs_info->tree_root, location);
5eda7b5e
CM		 1591 if (IS_ERR(root))
1592 return root;
3394e160 1593
cb517eab
MX		 1594 if (btrfs_root_refs(&root->root_item) == 0) {
1595 ret = -ENOENT;
581bb050 1596 goto fail;
35a30d7c 1597 }
581bb050 1598
cb517eab 1599 ret = btrfs_init_fs_root(root);
ac08aedf
CM		 1600 if (ret)
1601 goto fail;
3394e160 1602
d68fc57b
YZ		 1603 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1604 if (ret < 0)
1605 goto fail;
1606 if (ret == 0)
1607 root->orphan_item_inserted = 1;
1608
cb517eab 1609 ret = btrfs_insert_fs_root(fs_info, root);
0f7d52f4 1610 if (ret) {
4df27c4d
YZ		 1611 if (ret == -EEXIST) {
1612 free_fs_root(root);
1613 goto again;
1614 }
1615 goto fail;
0f7d52f4 1616 }
edbd8d4e 1617 return root;
4df27c4d
YZ		 1618fail:
1619 free_fs_root(root);
1620 return ERR_PTR(ret);
edbd8d4e
CM		 1621}
1622
04160088
CM		 1623static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1624{
1625 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1626 int ret = 0;
04160088
CM		 1627 struct btrfs_device *device;
1628 struct backing_dev_info *bdi;
b7967db7 1629
1f78160c
XG		 1630 rcu_read_lock();
1631 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1632 if (!device->bdev)
1633 continue;
04160088
CM		 1634 bdi = blk_get_backing_dev_info(device->bdev);
1635 if (bdi && bdi_congested(bdi, bdi_bits)) {
1636 ret = 1;
1637 break;
1638 }
1639 }
1f78160c 1640 rcu_read_unlock();
04160088
CM		 1641 return ret;
1642}
1643
ad081f14
JA		 1644/*
1645 * If this fails, caller must call bdi_destroy() to get rid of the
1646 * bdi again.
1647 */
04160088
CM		 1648static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1649{
ad081f14
JA		 1650 int err;
1651
1652 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1653 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1654 if (err)
1655 return err;
1656
4575c9cc 1657 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1658 bdi->congested_fn = btrfs_congested_fn;
1659 bdi->congested_data = info;
1660 return 0;
1661}
1662
8b712842
CM		 1663/*
1664 * called by the kthread helper functions to finally call the bio end_io
1665 * functions. This is where read checksum verification actually happens
1666 */
1667static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1668{
ce9adaa5 1669 struct bio *bio;
8b712842
CM		 1670 struct end_io_wq *end_io_wq;
1671 struct btrfs_fs_info *fs_info;
ce9adaa5 1672 int error;
ce9adaa5 1673
8b712842
CM		 1674 end_io_wq = container_of(work, struct end_io_wq, work);
1675 bio = end_io_wq->bio;
1676 fs_info = end_io_wq->info;
ce9adaa5 1677
8b712842
CM		 1678 error = end_io_wq->error;
1679 bio->bi_private = end_io_wq->private;
1680 bio->bi_end_io = end_io_wq->end_io;
1681 kfree(end_io_wq);
8b712842 1682 bio_endio(bio, error);
44b8bd7e
CM		 1683}
1684
a74a4b97
CM		 1685static int cleaner_kthread(void *arg)
1686{
1687 struct btrfs_root *root = arg;
d0278245 1688 int again;
a74a4b97
CM		 1689
1690 do {
d0278245 1691 again = 0;
a74a4b97 1692
d0278245 1693 /* Make the cleaner go to sleep early. */
babbf170 1694 if (btrfs_need_cleaner_sleep(root))
d0278245
MX		 1695 goto sleep;
1696
1697 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1698 goto sleep;
1699
dc7f370c
MX		 1700 /*
1701 * Avoid the problem that we change the status of the fs
1702 * during the above check and trylock.
1703 */
babbf170 1704 if (btrfs_need_cleaner_sleep(root)) {
dc7f370c
MX		 1705 mutex_unlock(&root->fs_info->cleaner_mutex);
1706 goto sleep;
76dda93c 1707 }
a74a4b97 1708
d0278245
MX		 1709 btrfs_run_delayed_iputs(root);
1710 again = btrfs_clean_one_deleted_snapshot(root);
1711 mutex_unlock(&root->fs_info->cleaner_mutex);
1712
1713 /*
05323cd1
MX		 1714 * The defragger has dealt with the R/O remount and umount,
1715 * needn't do anything special here.
d0278245
MX		 1716 */
1717 btrfs_run_defrag_inodes(root->fs_info);
1718sleep:
9d1a2a3a 1719 if (!try_to_freeze() && !again) {
a74a4b97 1720 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1721 if (!kthread_should_stop())
1722 schedule();
a74a4b97
CM		 1723 __set_current_state(TASK_RUNNING);
1724 }
1725 } while (!kthread_should_stop());
1726 return 0;
1727}
1728
1729static int transaction_kthread(void *arg)
1730{
1731 struct btrfs_root *root = arg;
1732 struct btrfs_trans_handle *trans;
1733 struct btrfs_transaction *cur;
8929ecfa 1734 u64 transid;
a74a4b97
CM		 1735 unsigned long now;
1736 unsigned long delay;
914b2007 1737 bool cannot_commit;
a74a4b97
CM		 1738
1739 do {
914b2007 1740 cannot_commit = false;
8b87dc17 1741 delay = HZ * root->fs_info->commit_interval;
a74a4b97
CM		 1742 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1743
a4abeea4 1744 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1745 cur = root->fs_info->running_transaction;
1746 if (!cur) {
a4abeea4 1747 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1748 goto sleep;
1749 }
31153d81 1750
a74a4b97 1751 now = get_seconds();
4a9d8bde 1752 if (cur->state < TRANS_STATE_BLOCKED &&
8b87dc17
DS		 1753 (now < cur->start_time ||
1754 now - cur->start_time < root->fs_info->commit_interval)) {
a4abeea4 1755 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1756 delay = HZ * 5;
1757 goto sleep;
1758 }
8929ecfa 1759 transid = cur->transid;
a4abeea4 1760 spin_unlock(&root->fs_info->trans_lock);
56bec294 1761
79787eaa 1762 /* If the file system is aborted, this will always fail. */
354aa0fb 1763 trans = btrfs_attach_transaction(root);
914b2007 1764 if (IS_ERR(trans)) {
354aa0fb
MX		 1765 if (PTR_ERR(trans) != -ENOENT)
1766 cannot_commit = true;
79787eaa 1767 goto sleep;
914b2007 1768 }
8929ecfa 1769 if (transid == trans->transid) {
79787eaa 1770 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1771 } else {
1772 btrfs_end_transaction(trans, root);
1773 }
a74a4b97
CM		 1774sleep:
1775 wake_up_process(root->fs_info->cleaner_kthread);
1776 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1777
a0acae0e 1778 if (!try_to_freeze()) {
a74a4b97 1779 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1780 if (!kthread_should_stop() &&
914b2007
JK		 1781 (!btrfs_transaction_blocked(root->fs_info) ||
1782 cannot_commit))
8929ecfa 1783 schedule_timeout(delay);
a74a4b97
CM		 1784 __set_current_state(TASK_RUNNING);
1785 }
1786 } while (!kthread_should_stop());
1787 return 0;
1788}
1789
af31f5e5
CM		 1790/*
1791 * this will find the highest generation in the array of
1792 * root backups. The index of the highest array is returned,
1793 * or -1 if we can't find anything.
1794 *
1795 * We check to make sure the array is valid by comparing the
1796 * generation of the latest root in the array with the generation
1797 * in the super block. If they don't match we pitch it.
1798 */
1799static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1800{
1801 u64 cur;
1802 int newest_index = -1;
1803 struct btrfs_root_backup *root_backup;
1804 int i;
1805
1806 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1807 root_backup = info->super_copy->super_roots + i;
1808 cur = btrfs_backup_tree_root_gen(root_backup);
1809 if (cur == newest_gen)
1810 newest_index = i;
1811 }
1812
1813 /* check to see if we actually wrapped around */
1814 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1815 root_backup = info->super_copy->super_roots;
1816 cur = btrfs_backup_tree_root_gen(root_backup);
1817 if (cur == newest_gen)
1818 newest_index = 0;
1819 }
1820 return newest_index;
1821}
1822
1823
1824/*
1825 * find the oldest backup so we know where to store new entries
1826 * in the backup array. This will set the backup_root_index
1827 * field in the fs_info struct
1828 */
1829static void find_oldest_super_backup(struct btrfs_fs_info *info,
1830 u64 newest_gen)
1831{
1832 int newest_index = -1;
1833
1834 newest_index = find_newest_super_backup(info, newest_gen);
1835 /* if there was garbage in there, just move along */
1836 if (newest_index == -1) {
1837 info->backup_root_index = 0;
1838 } else {
1839 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1840 }
1841}
1842
1843/*
1844 * copy all the root pointers into the super backup array.
1845 * this will bump the backup pointer by one when it is
1846 * done
1847 */
1848static void backup_super_roots(struct btrfs_fs_info *info)
1849{
1850 int next_backup;
1851 struct btrfs_root_backup *root_backup;
1852 int last_backup;
1853
1854 next_backup = info->backup_root_index;
1855 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1856 BTRFS_NUM_BACKUP_ROOTS;
1857
1858 /*
1859 * just overwrite the last backup if we're at the same generation
1860 * this happens only at umount
1861 */
1862 root_backup = info->super_for_commit->super_roots + last_backup;
1863 if (btrfs_backup_tree_root_gen(root_backup) ==
1864 btrfs_header_generation(info->tree_root->node))
1865 next_backup = last_backup;
1866
1867 root_backup = info->super_for_commit->super_roots + next_backup;
1868
1869 /*
1870 * make sure all of our padding and empty slots get zero filled
1871 * regardless of which ones we use today
1872 */
1873 memset(root_backup, 0, sizeof(*root_backup));
1874
1875 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1876
1877 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1878 btrfs_set_backup_tree_root_gen(root_backup,
1879 btrfs_header_generation(info->tree_root->node));
1880
1881 btrfs_set_backup_tree_root_level(root_backup,
1882 btrfs_header_level(info->tree_root->node));
1883
1884 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1885 btrfs_set_backup_chunk_root_gen(root_backup,
1886 btrfs_header_generation(info->chunk_root->node));
1887 btrfs_set_backup_chunk_root_level(root_backup,
1888 btrfs_header_level(info->chunk_root->node));
1889
1890 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1891 btrfs_set_backup_extent_root_gen(root_backup,
1892 btrfs_header_generation(info->extent_root->node));
1893 btrfs_set_backup_extent_root_level(root_backup,
1894 btrfs_header_level(info->extent_root->node));
1895
7c7e82a7
CM		 1896 /*
1897 * we might commit during log recovery, which happens before we set
1898 * the fs_root. Make sure it is valid before we fill it in.
1899 */
1900 if (info->fs_root && info->fs_root->node) {
1901 btrfs_set_backup_fs_root(root_backup,
1902 info->fs_root->node->start);
1903 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1904 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1905 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1906 btrfs_header_level(info->fs_root->node));
7c7e82a7 1907 }
af31f5e5
CM		 1908
1909 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1910 btrfs_set_backup_dev_root_gen(root_backup,
1911 btrfs_header_generation(info->dev_root->node));
1912 btrfs_set_backup_dev_root_level(root_backup,
1913 btrfs_header_level(info->dev_root->node));
1914
1915 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1916 btrfs_set_backup_csum_root_gen(root_backup,
1917 btrfs_header_generation(info->csum_root->node));
1918 btrfs_set_backup_csum_root_level(root_backup,
1919 btrfs_header_level(info->csum_root->node));
1920
1921 btrfs_set_backup_total_bytes(root_backup,
1922 btrfs_super_total_bytes(info->super_copy));
1923 btrfs_set_backup_bytes_used(root_backup,
1924 btrfs_super_bytes_used(info->super_copy));
1925 btrfs_set_backup_num_devices(root_backup,
1926 btrfs_super_num_devices(info->super_copy));
1927
1928 /*
1929 * if we don't copy this out to the super_copy, it won't get remembered
1930 * for the next commit
1931 */
1932 memcpy(&info->super_copy->super_roots,
1933 &info->super_for_commit->super_roots,
1934 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1935}
1936
1937/*
1938 * this copies info out of the root backup array and back into
1939 * the in-memory super block. It is meant to help iterate through
1940 * the array, so you send it the number of backups you've already
1941 * tried and the last backup index you used.
1942 *
1943 * this returns -1 when it has tried all the backups
1944 */
1945static noinline int next_root_backup(struct btrfs_fs_info *info,
1946 struct btrfs_super_block *super,
1947 int *num_backups_tried, int *backup_index)
1948{
1949 struct btrfs_root_backup *root_backup;
1950 int newest = *backup_index;
1951
1952 if (*num_backups_tried == 0) {
1953 u64 gen = btrfs_super_generation(super);
1954
1955 newest = find_newest_super_backup(info, gen);
1956 if (newest == -1)
1957 return -1;
1958
1959 *backup_index = newest;
1960 *num_backups_tried = 1;
1961 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1962 /* we've tried all the backups, all done */
1963 return -1;
1964 } else {
1965 /* jump to the next oldest backup */
1966 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1967 BTRFS_NUM_BACKUP_ROOTS;
1968 *backup_index = newest;
1969 *num_backups_tried += 1;
1970 }
1971 root_backup = super->super_roots + newest;
1972
1973 btrfs_set_super_generation(super,
1974 btrfs_backup_tree_root_gen(root_backup));
1975 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1976 btrfs_set_super_root_level(super,
1977 btrfs_backup_tree_root_level(root_backup));
1978 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1979
1980 /*
1981 * fixme: the total bytes and num_devices need to match or we should
1982 * need a fsck
1983 */
1984 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1985 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1986 return 0;
1987}
1988
7abadb64
LB		 1989/* helper to cleanup workers */
1990static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1991{
1992 btrfs_stop_workers(&fs_info->generic_worker);
1993 btrfs_stop_workers(&fs_info->fixup_workers);
1994 btrfs_stop_workers(&fs_info->delalloc_workers);
1995 btrfs_stop_workers(&fs_info->workers);
1996 btrfs_stop_workers(&fs_info->endio_workers);
1997 btrfs_stop_workers(&fs_info->endio_meta_workers);
1998 btrfs_stop_workers(&fs_info->endio_raid56_workers);
1999 btrfs_stop_workers(&fs_info->rmw_workers);
2000 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2001 btrfs_stop_workers(&fs_info->endio_write_workers);
2002 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2003 btrfs_stop_workers(&fs_info->submit_workers);
2004 btrfs_stop_workers(&fs_info->delayed_workers);
2005 btrfs_stop_workers(&fs_info->caching_workers);
2006 btrfs_stop_workers(&fs_info->readahead_workers);
2007 btrfs_stop_workers(&fs_info->flush_workers);
2f232036 2008 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
7abadb64
LB		 2009}
2010
af31f5e5
CM		 2011/* helper to cleanup tree roots */
2012static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2013{
2014 free_extent_buffer(info->tree_root->node);
2015 free_extent_buffer(info->tree_root->commit_root);
af31f5e5
CM		 2016 info->tree_root->node = NULL;
2017 info->tree_root->commit_root = NULL;
655b09fe
JB		 2018
2019 if (info->dev_root) {
2020 free_extent_buffer(info->dev_root->node);
2021 free_extent_buffer(info->dev_root->commit_root);
2022 info->dev_root->node = NULL;
2023 info->dev_root->commit_root = NULL;
2024 }
2025 if (info->extent_root) {
2026 free_extent_buffer(info->extent_root->node);
2027 free_extent_buffer(info->extent_root->commit_root);
2028 info->extent_root->node = NULL;
2029 info->extent_root->commit_root = NULL;
2030 }
2031 if (info->csum_root) {
2032 free_extent_buffer(info->csum_root->node);
2033 free_extent_buffer(info->csum_root->commit_root);
2034 info->csum_root->node = NULL;
2035 info->csum_root->commit_root = NULL;
2036 }
bcef60f2 2037 if (info->quota_root) {
655b09fe
JB		 2038 free_extent_buffer(info->quota_root->node);
2039 free_extent_buffer(info->quota_root->commit_root);
bcef60f2
AJ		 2040 info->quota_root->node = NULL;
2041 info->quota_root->commit_root = NULL;
2042 }
f7a81ea4
SB		 2043 if (info->uuid_root) {
2044 free_extent_buffer(info->uuid_root->node);
2045 free_extent_buffer(info->uuid_root->commit_root);
2046 info->uuid_root->node = NULL;
2047 info->uuid_root->commit_root = NULL;
2048 }
af31f5e5
CM		 2049 if (chunk_root) {
2050 free_extent_buffer(info->chunk_root->node);
2051 free_extent_buffer(info->chunk_root->commit_root);
2052 info->chunk_root->node = NULL;
2053 info->chunk_root->commit_root = NULL;
2054 }
2055}
2056
171f6537
JB		 2057static void del_fs_roots(struct btrfs_fs_info *fs_info)
2058{
2059 int ret;
2060 struct btrfs_root *gang[8];
2061 int i;
2062
2063 while (!list_empty(&fs_info->dead_roots)) {
2064 gang[0] = list_entry(fs_info->dead_roots.next,
2065 struct btrfs_root, root_list);
2066 list_del(&gang[0]->root_list);
2067
2068 if (gang[0]->in_radix) {
cb517eab 2069 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
171f6537
JB		 2070 } else {
2071 free_extent_buffer(gang[0]->node);
2072 free_extent_buffer(gang[0]->commit_root);
b0feb9d9 2073 btrfs_put_fs_root(gang[0]);
171f6537
JB		 2074 }
2075 }
2076
2077 while (1) {
2078 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2079 (void **)gang, 0,
2080 ARRAY_SIZE(gang));
2081 if (!ret)
2082 break;
2083 for (i = 0; i < ret; i++)
cb517eab 2084 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
171f6537
JB		 2085 }
2086}
af31f5e5 2087
ad2b2c80
AV		 2088int open_ctree(struct super_block *sb,
2089 struct btrfs_fs_devices *fs_devices,
2090 char *options)
2e635a27 2091{
db94535d
CM		 2092 u32 sectorsize;
2093 u32 nodesize;
2094 u32 leafsize;
2095 u32 blocksize;
87ee04eb 2096 u32 stripesize;
84234f3a 2097 u64 generation;
f2b636e8 2098 u64 features;
3de4586c 2099 struct btrfs_key location;
a061fc8d 2100 struct buffer_head *bh;
4d34b278 2101 struct btrfs_super_block *disk_super;
815745cf 2102 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2103 struct btrfs_root *tree_root;
4d34b278
ID		 2104 struct btrfs_root *extent_root;
2105 struct btrfs_root *csum_root;
2106 struct btrfs_root *chunk_root;
2107 struct btrfs_root *dev_root;
bcef60f2 2108 struct btrfs_root *quota_root;
f7a81ea4 2109 struct btrfs_root *uuid_root;
e02119d5 2110 struct btrfs_root *log_tree_root;
eb60ceac 2111 int ret;
e58ca020 2112 int err = -EINVAL;
af31f5e5
CM		 2113 int num_backups_tried = 0;
2114 int backup_index = 0;
70f80175
SB		 2115 bool create_uuid_tree;
2116 bool check_uuid_tree;
4543df7e 2117
f84a8bd6 2118 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e 2119 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
cb517eab 2120 if (!tree_root || !chunk_root) {
39279cc3
CM		 2121 err = -ENOMEM;
2122 goto fail;
2123 }
76dda93c
YZ		 2124
2125 ret = init_srcu_struct(&fs_info->subvol_srcu);
2126 if (ret) {
2127 err = ret;
2128 goto fail;
2129 }
2130
2131 ret = setup_bdi(fs_info, &fs_info->bdi);
2132 if (ret) {
2133 err = ret;
2134 goto fail_srcu;
2135 }
2136
e2d84521
MX		 2137 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2138 if (ret) {
2139 err = ret;
2140 goto fail_bdi;
2141 }
2142 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2143 (1 + ilog2(nr_cpu_ids));
2144
963d678b
MX		 2145 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2146 if (ret) {
2147 err = ret;
2148 goto fail_dirty_metadata_bytes;
2149 }
2150
76dda93c
YZ		 2151 fs_info->btree_inode = new_inode(sb);
2152 if (!fs_info->btree_inode) {
2153 err = -ENOMEM;
963d678b 2154 goto fail_delalloc_bytes;
76dda93c
YZ		 2155 }
2156
a6591715 2157 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2158
76dda93c 2159 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 2160 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2161 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2162 INIT_LIST_HEAD(&fs_info->delayed_iputs);
eb73c1b7 2163 INIT_LIST_HEAD(&fs_info->delalloc_roots);
11833d66 2164 INIT_LIST_HEAD(&fs_info->caching_block_groups);
eb73c1b7 2165 spin_lock_init(&fs_info->delalloc_root_lock);
a4abeea4 2166 spin_lock_init(&fs_info->trans_lock);
76dda93c 2167 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2168 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2169 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2170 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2171 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2172 spin_lock_init(&fs_info->super_lock);
f29021b2 2173 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2174 mutex_init(&fs_info->reloc_mutex);
de98ced9 2175 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2176
58176a96 2177 init_completion(&fs_info->kobj_unregister);
0b86a832 2178 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2179 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2180 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2181 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2182 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2183 BTRFS_BLOCK_RSV_GLOBAL);
2184 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2185 BTRFS_BLOCK_RSV_DELALLOC);
2186 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2187 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2188 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2189 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2190 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2191 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2192 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2193 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2194 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2195 atomic_set(&fs_info->defrag_running, 0);
fc36ed7e 2196 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2197 fs_info->sb = sb;
6f568d35 2198 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2199 fs_info->metadata_ratio = 0;
4cb5300b 2200 fs_info->defrag_inodes = RB_ROOT;
2bf64758 2201 fs_info->free_chunk_space = 0;
f29021b2 2202 fs_info->tree_mod_log = RB_ROOT;
8b87dc17 2203 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
c8b97818 2204
90519d66
AJ		 2205 /* readahead state */
2206 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2207 spin_lock_init(&fs_info->reada_lock);
c8b97818 2208
b34b086c
CM		 2209 fs_info->thread_pool_size = min_t(unsigned long,
2210 num_online_cpus() + 2, 8);
0afbaf8c 2211
199c2a9c
MX		 2212 INIT_LIST_HEAD(&fs_info->ordered_roots);
2213 spin_lock_init(&fs_info->ordered_root_lock);
16cdcec7
MX		 2214 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2215 GFP_NOFS);
2216 if (!fs_info->delayed_root) {
2217 err = -ENOMEM;
2218 goto fail_iput;
2219 }
2220 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2221
a2de733c
AJ		 2222 mutex_init(&fs_info->scrub_lock);
2223 atomic_set(&fs_info->scrubs_running, 0);
2224 atomic_set(&fs_info->scrub_pause_req, 0);
2225 atomic_set(&fs_info->scrubs_paused, 0);
2226 atomic_set(&fs_info->scrub_cancel_req, 0);
2227 init_waitqueue_head(&fs_info->scrub_pause_wait);
2228 init_rwsem(&fs_info->scrub_super_lock);
2229 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2230#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2231 fs_info->check_integrity_print_mask = 0;
2232#endif
a2de733c 2233
c9e9f97b
ID		 2234 spin_lock_init(&fs_info->balance_lock);
2235 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2236 atomic_set(&fs_info->balance_running, 0);
2237 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2238 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2239 fs_info->balance_ctl = NULL;
837d5b6e 2240 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2241
a061fc8d
CM		 2242 sb->s_blocksize = 4096;
2243 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2244 sb->s_bdi = &fs_info->bdi;
a061fc8d 2245
76dda93c 2246 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2247 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2248 /*
2249 * we set the i_size on the btree inode to the max possible int.
2250 * the real end of the address space is determined by all of
2251 * the devices in the system
2252 */
2253 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2254 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2255 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2256
5d4f98a2 2257 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2258 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2259 fs_info->btree_inode->i_mapping);
0b32f4bb 2260 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2261 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2262
2263 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2264
76dda93c
YZ		 2265 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2266 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2267 sizeof(struct btrfs_key));
72ac3c0d
JB		 2268 set_bit(BTRFS_INODE_DUMMY,
2269 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
c65ddb52 2270 insert_inode_hash(fs_info->btree_inode);
76dda93c 2271
0f9dd46c 2272 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2273 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2274 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2275
11833d66 2276 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2277 fs_info->btree_inode->i_mapping);
11833d66 2278 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2279 fs_info->btree_inode->i_mapping);
11833d66 2280 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2281 fs_info->do_barriers = 1;
e18e4809 2282
39279cc3 2283
5a3f23d5 2284 mutex_init(&fs_info->ordered_operations_mutex);
9ffba8cd 2285 mutex_init(&fs_info->ordered_extent_flush_mutex);
e02119d5 2286 mutex_init(&fs_info->tree_log_mutex);
925baedd 2287 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2288 mutex_init(&fs_info->transaction_kthread_mutex);
2289 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2290 mutex_init(&fs_info->volume_mutex);
276e680d 2291 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2292 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2293 init_rwsem(&fs_info->subvol_sem);
803b2f54 2294 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
e922e087
SB		 2295 fs_info->dev_replace.lock_owner = 0;
2296 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2297 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2298 mutex_init(&fs_info->dev_replace.lock_management_lock);
2299 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2300
416ac51d 2301 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2302 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2303 fs_info->qgroup_tree = RB_ROOT;
2304 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2305 fs_info->qgroup_seq = 1;
2306 fs_info->quota_enabled = 0;
2307 fs_info->pending_quota_state = 0;
1e8f9158 2308 fs_info->qgroup_ulist = NULL;
2f232036 2309 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2310
fa9c0d79
CM		 2311 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2312 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2313
e6dcd2dc 2314 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2315 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2316 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2317 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2318
53b381b3
DW		 2319 ret = btrfs_alloc_stripe_hash_table(fs_info);
2320 if (ret) {
83c8266a 2321 err = ret;
53b381b3
DW		 2322 goto fail_alloc;
2323 }
2324
0b86a832 2325 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2326 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2327
3c4bb26b 2328 invalidate_bdev(fs_devices->latest_bdev);
1104a885
DS		 2329
2330 /*
2331 * Read super block and check the signature bytes only
2332 */
a512bbf8 2333 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2334 if (!bh) {
2335 err = -EINVAL;
16cdcec7 2336 goto fail_alloc;
20b45077 2337 }
39279cc3 2338
1104a885
DS		 2339 /*
2340 * We want to check superblock checksum, the type is stored inside.
2341 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2342 */
2343 if (btrfs_check_super_csum(bh->b_data)) {
2344 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2345 err = -EINVAL;
2346 goto fail_alloc;
2347 }
2348
2349 /*
2350 * super_copy is zeroed at allocation time and we never touch the
2351 * following bytes up to INFO_SIZE, the checksum is calculated from
2352 * the whole block of INFO_SIZE
2353 */
6c41761f
DS		 2354 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2355 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2356 sizeof(*fs_info->super_for_commit));
a061fc8d 2357 brelse(bh);
5f39d397 2358
6c41761f 2359 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2360
1104a885
DS		 2361 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2362 if (ret) {
2363 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2364 err = -EINVAL;
2365 goto fail_alloc;
2366 }
2367
6c41761f 2368 disk_super = fs_info->super_copy;
0f7d52f4 2369 if (!btrfs_super_root(disk_super))
16cdcec7 2370 goto fail_alloc;
0f7d52f4 2371
acce952b 2372 /* check FS state, whether FS is broken. */
87533c47
MX		 2373 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2374 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2375
af31f5e5
CM		 2376 /*
2377 * run through our array of backup supers and setup
2378 * our ring pointer to the oldest one
2379 */
2380 generation = btrfs_super_generation(disk_super);
2381 find_oldest_super_backup(fs_info, generation);
2382
75e7cb7f
LB		 2383 /*
2384 * In the long term, we'll store the compression type in the super
2385 * block, and it'll be used for per file compression control.
2386 */
2387 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2388
2b82032c
YZ		 2389 ret = btrfs_parse_options(tree_root, options);
2390 if (ret) {
2391 err = ret;
16cdcec7 2392 goto fail_alloc;
2b82032c 2393 }
dfe25020 2394
f2b636e8
JB		 2395 features = btrfs_super_incompat_flags(disk_super) &
2396 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2397 if (features) {
2398 printk(KERN_ERR "BTRFS: couldn't mount because of "
2399 "unsupported optional features (%Lx).\n",
c1c9ff7c 2400 features);
f2b636e8 2401 err = -EINVAL;
16cdcec7 2402 goto fail_alloc;
f2b636e8
JB		 2403 }
2404
727011e0
CM		 2405 if (btrfs_super_leafsize(disk_super) !=
2406 btrfs_super_nodesize(disk_super)) {
2407 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2408 "blocksizes don't match. node %d leaf %d\n",
2409 btrfs_super_nodesize(disk_super),
2410 btrfs_super_leafsize(disk_super));
2411 err = -EINVAL;
2412 goto fail_alloc;
2413 }
2414 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2415 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2416 "blocksize (%d) was too large\n",
2417 btrfs_super_leafsize(disk_super));
2418 err = -EINVAL;
2419 goto fail_alloc;
2420 }
2421
5d4f98a2 2422 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2423 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2424 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2425 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2426
3173a18f
JB		 2427 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2428 printk(KERN_ERR "btrfs: has skinny extents\n");
2429
727011e0
CM		 2430 /*
2431 * flag our filesystem as having big metadata blocks if
2432 * they are bigger than the page size
2433 */
2434 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2435 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2436 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2437 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2438 }
2439
bc3f116f
CM		 2440 nodesize = btrfs_super_nodesize(disk_super);
2441 leafsize = btrfs_super_leafsize(disk_super);
2442 sectorsize = btrfs_super_sectorsize(disk_super);
2443 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2444 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2445 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2446
2447 /*
2448 * mixed block groups end up with duplicate but slightly offset
2449 * extent buffers for the same range. It leads to corruptions
2450 */
2451 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2452 (sectorsize != leafsize)) {
2453 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2454 "are not allowed for mixed block groups on %s\n",
2455 sb->s_id);
2456 goto fail_alloc;
2457 }
2458
ceda0864
MX		 2459 /*
2460 * Needn't use the lock because there is no other task which will
2461 * update the flag.
2462 */
a6fa6fae 2463 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2464
f2b636e8
JB		 2465 features = btrfs_super_compat_ro_flags(disk_super) &
2466 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2467 if (!(sb->s_flags & MS_RDONLY) && features) {
2468 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2469 "unsupported option features (%Lx).\n",
c1c9ff7c 2470 features);
f2b636e8 2471 err = -EINVAL;
16cdcec7 2472 goto fail_alloc;
f2b636e8 2473 }
61d92c32
CM		 2474
2475 btrfs_init_workers(&fs_info->generic_worker,
2476 "genwork", 1, NULL);
2477
5443be45 2478 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 2479 fs_info->thread_pool_size,
2480 &fs_info->generic_worker);
c8b97818 2481
771ed689 2482 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 2483 fs_info->thread_pool_size,
2484 &fs_info->generic_worker);
771ed689 2485
8ccf6f19
MX		 2486 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2487 fs_info->thread_pool_size,
2488 &fs_info->generic_worker);
2489
5443be45 2490 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 2491 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 2492 fs_info->thread_pool_size),
2493 &fs_info->generic_worker);
61b49440 2494
bab39bf9
JB		 2495 btrfs_init_workers(&fs_info->caching_workers, "cache",
2496 2, &fs_info->generic_worker);
2497
61b49440
CM		 2498 /* a higher idle thresh on the submit workers makes it much more
2499 * likely that bios will be send down in a sane order to the
2500 * devices
2501 */
2502 fs_info->submit_workers.idle_thresh = 64;
53863232 2503
771ed689 2504 fs_info->workers.idle_thresh = 16;
4a69a410 2505 fs_info->workers.ordered = 1;
61b49440 2506
771ed689
CM		 2507 fs_info->delalloc_workers.idle_thresh = 2;
2508 fs_info->delalloc_workers.ordered = 1;
2509
61d92c32
CM		 2510 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2511 &fs_info->generic_worker);
5443be45 2512 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2513 fs_info->thread_pool_size,
2514 &fs_info->generic_worker);
d20f7043 2515 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2516 fs_info->thread_pool_size,
2517 &fs_info->generic_worker);
cad321ad 2518 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2519 "endio-meta-write", fs_info->thread_pool_size,
2520 &fs_info->generic_worker);
53b381b3
DW		 2521 btrfs_init_workers(&fs_info->endio_raid56_workers,
2522 "endio-raid56", fs_info->thread_pool_size,
2523 &fs_info->generic_worker);
2524 btrfs_init_workers(&fs_info->rmw_workers,
2525 "rmw", fs_info->thread_pool_size,
2526 &fs_info->generic_worker);
5443be45 2527 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2528 fs_info->thread_pool_size,
2529 &fs_info->generic_worker);
0cb59c99
JB		 2530 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2531 1, &fs_info->generic_worker);
16cdcec7
MX		 2532 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2533 fs_info->thread_pool_size,
2534 &fs_info->generic_worker);
90519d66
AJ		 2535 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2536 fs_info->thread_pool_size,
2537 &fs_info->generic_worker);
2f232036
JS		 2538 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2539 &fs_info->generic_worker);
61b49440
CM		 2540
2541 /*
2542 * endios are largely parallel and should have a very
2543 * low idle thresh
2544 */
2545 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2546 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2547 fs_info->endio_raid56_workers.idle_thresh = 4;
2548 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2549
9042846b
CM		 2550 fs_info->endio_write_workers.idle_thresh = 2;
2551 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2552 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2553
0dc3b84a
JB		 2554 /*
2555 * btrfs_start_workers can really only fail because of ENOMEM so just
2556 * return -ENOMEM if any of these fail.
2557 */
2558 ret = btrfs_start_workers(&fs_info->workers);
2559 ret |= btrfs_start_workers(&fs_info->generic_worker);
2560 ret |= btrfs_start_workers(&fs_info->submit_workers);
2561 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2562 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2563 ret |= btrfs_start_workers(&fs_info->endio_workers);
2564 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2565 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2566 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2567 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2568 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2569 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2570 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2571 ret |= btrfs_start_workers(&fs_info->caching_workers);
2572 ret |= btrfs_start_workers(&fs_info->readahead_workers);
8ccf6f19 2573 ret |= btrfs_start_workers(&fs_info->flush_workers);
2f232036 2574 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
0dc3b84a 2575 if (ret) {
fed425c7 2576 err = -ENOMEM;
0dc3b84a
JB		 2577 goto fail_sb_buffer;
2578 }
4543df7e 2579
4575c9cc 2580 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2581 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2582 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2583
db94535d
CM		 2584 tree_root->nodesize = nodesize;
2585 tree_root->leafsize = leafsize;
2586 tree_root->sectorsize = sectorsize;
87ee04eb 2587 tree_root->stripesize = stripesize;
a061fc8d
CM		 2588
2589 sb->s_blocksize = sectorsize;
2590 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2591
3cae210f 2592 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
d397712b 2593 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2594 goto fail_sb_buffer;
2595 }
19c00ddc 2596
8d082fb7
LB		 2597 if (sectorsize != PAGE_SIZE) {
2598 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2599 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2600 goto fail_sb_buffer;
2601 }
2602
925baedd 2603 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2604 ret = btrfs_read_sys_array(tree_root);
925baedd 2605 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2606 if (ret) {
d397712b
CM		 2607 printk(KERN_WARNING "btrfs: failed to read the system "
2608 "array on %s\n", sb->s_id);
5d4f98a2 2609 goto fail_sb_buffer;
84eed90f 2610 }
0b86a832
CM		 2611
2612 blocksize = btrfs_level_size(tree_root,
2613 btrfs_super_chunk_root_level(disk_super));
84234f3a 2614 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2615
2616 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2617 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2618
2619 chunk_root->node = read_tree_block(chunk_root,
2620 btrfs_super_chunk_root(disk_super),
84234f3a 2621 blocksize, generation);
416bc658
JB		 2622 if (!chunk_root->node ||
2623 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
83121942
DW		 2624 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2625 sb->s_id);
af31f5e5 2626 goto fail_tree_roots;
83121942 2627 }
5d4f98a2
YZ		 2628 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2629 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2630
e17cade2 2631 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 2632 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2633 BTRFS_UUID_SIZE);
e17cade2 2634
0b86a832 2635 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2636 if (ret) {
d397712b
CM		 2637 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2638 sb->s_id);
af31f5e5 2639 goto fail_tree_roots;
2b82032c 2640 }
0b86a832 2641
8dabb742
SB		 2642 /*
2643 * keep the device that is marked to be the target device for the
2644 * dev_replace procedure
2645 */
2646 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2647
a6b0d5c8
CM		 2648 if (!fs_devices->latest_bdev) {
2649 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2650 sb->s_id);
2651 goto fail_tree_roots;
2652 }
2653
af31f5e5 2654retry_root_backup:
db94535d
CM		 2655 blocksize = btrfs_level_size(tree_root,
2656 btrfs_super_root_level(disk_super));
84234f3a 2657 generation = btrfs_super_generation(disk_super);
0b86a832 2658
e20d96d6 2659 tree_root->node = read_tree_block(tree_root,
db94535d 2660 btrfs_super_root(disk_super),
84234f3a 2661 blocksize, generation);
af31f5e5
CM		 2662 if (!tree_root->node ||
2663 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
83121942
DW		 2664 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2665 sb->s_id);
af31f5e5
CM		 2666
2667 goto recovery_tree_root;
83121942 2668 }
af31f5e5 2669
5d4f98a2
YZ		 2670 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2671 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d 2672
cb517eab
MX		 2673 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2674 location.type = BTRFS_ROOT_ITEM_KEY;
2675 location.offset = 0;
2676
2677 extent_root = btrfs_read_tree_root(tree_root, &location);
2678 if (IS_ERR(extent_root)) {
2679 ret = PTR_ERR(extent_root);
af31f5e5 2680 goto recovery_tree_root;
cb517eab 2681 }
0b86a832 2682 extent_root->track_dirty = 1;
cb517eab 2683 fs_info->extent_root = extent_root;
0b86a832 2684
cb517eab
MX		 2685 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2686 dev_root = btrfs_read_tree_root(tree_root, &location);
2687 if (IS_ERR(dev_root)) {
2688 ret = PTR_ERR(dev_root);
af31f5e5 2689 goto recovery_tree_root;
cb517eab 2690 }
5d4f98a2 2691 dev_root->track_dirty = 1;
cb517eab
MX		 2692 fs_info->dev_root = dev_root;
2693 btrfs_init_devices_late(fs_info);
3768f368 2694
cb517eab
MX		 2695 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2696 csum_root = btrfs_read_tree_root(tree_root, &location);
2697 if (IS_ERR(csum_root)) {
2698 ret = PTR_ERR(csum_root);
af31f5e5 2699 goto recovery_tree_root;
cb517eab 2700 }
d20f7043 2701 csum_root->track_dirty = 1;
cb517eab 2702 fs_info->csum_root = csum_root;
d20f7043 2703
cb517eab
MX		 2704 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2705 quota_root = btrfs_read_tree_root(tree_root, &location);
2706 if (!IS_ERR(quota_root)) {
bcef60f2
AJ		 2707 quota_root->track_dirty = 1;
2708 fs_info->quota_enabled = 1;
2709 fs_info->pending_quota_state = 1;
cb517eab 2710 fs_info->quota_root = quota_root;
bcef60f2
AJ		 2711 }
2712
f7a81ea4
SB		 2713 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2714 uuid_root = btrfs_read_tree_root(tree_root, &location);
2715 if (IS_ERR(uuid_root)) {
2716 ret = PTR_ERR(uuid_root);
2717 if (ret != -ENOENT)
2718 goto recovery_tree_root;
2719 create_uuid_tree = true;
70f80175 2720 check_uuid_tree = false;
f7a81ea4
SB		 2721 } else {
2722 uuid_root->track_dirty = 1;
2723 fs_info->uuid_root = uuid_root;
70f80175
SB		 2724 create_uuid_tree = false;
2725 check_uuid_tree =
2726 generation != btrfs_super_uuid_tree_generation(disk_super);
f7a81ea4
SB		 2727 }
2728
8929ecfa
YZ		 2729 fs_info->generation = generation;
2730 fs_info->last_trans_committed = generation;
8929ecfa 2731
68310a5e
ID		 2732 ret = btrfs_recover_balance(fs_info);
2733 if (ret) {
2734 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2735 goto fail_block_groups;
2736 }
2737
733f4fbb
SB		 2738 ret = btrfs_init_dev_stats(fs_info);
2739 if (ret) {
2740 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2741 ret);
2742 goto fail_block_groups;
2743 }
2744
8dabb742
SB		 2745 ret = btrfs_init_dev_replace(fs_info);
2746 if (ret) {
2747 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2748 goto fail_block_groups;
2749 }
2750
2751 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2752
c59021f8 2753 ret = btrfs_init_space_info(fs_info);
2754 if (ret) {
2755 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2756 goto fail_block_groups;
2757 }
2758
1b1d1f66
JB		 2759 ret = btrfs_read_block_groups(extent_root);
2760 if (ret) {
2761 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2762 goto fail_block_groups;
2763 }
5af3e8cc
SB		 2764 fs_info->num_tolerated_disk_barrier_failures =
2765 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2766 if (fs_info->fs_devices->missing_devices >
2767 fs_info->num_tolerated_disk_barrier_failures &&
2768 !(sb->s_flags & MS_RDONLY)) {
2769 printk(KERN_WARNING
2770 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2771 goto fail_block_groups;
2772 }
9078a3e1 2773
a74a4b97
CM		 2774 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2775 "btrfs-cleaner");
57506d50 2776 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2777 goto fail_block_groups;
a74a4b97
CM		 2778
2779 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2780 tree_root,
2781 "btrfs-transaction");
57506d50 2782 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2783 goto fail_cleaner;
a74a4b97 2784
c289811c
CM		 2785 if (!btrfs_test_opt(tree_root, SSD) &&
2786 !btrfs_test_opt(tree_root, NOSSD) &&
2787 !fs_info->fs_devices->rotating) {
2788 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2789 "mode\n");
2790 btrfs_set_opt(fs_info->mount_opt, SSD);
2791 }
2792
21adbd5c
SB		 2793#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2794 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2795 ret = btrfsic_mount(tree_root, fs_devices,
2796 btrfs_test_opt(tree_root,
2797 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2798 1 : 0,
2799 fs_info->check_integrity_print_mask);
2800 if (ret)
2801 printk(KERN_WARNING "btrfs: failed to initialize"
2802 " integrity check module %s\n", sb->s_id);
2803 }
2804#endif
bcef60f2
AJ		 2805 ret = btrfs_read_qgroup_config(fs_info);
2806 if (ret)
2807 goto fail_trans_kthread;
21adbd5c 2808
acce952b 2809 /* do not make disk changes in broken FS */
68ce9682 2810 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2811 u64 bytenr = btrfs_super_log_root(disk_super);
2812
7c2ca468 2813 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2814 printk(KERN_WARNING "Btrfs log replay required "
2815 "on RO media\n");
7c2ca468 2816 err = -EIO;
bcef60f2 2817 goto fail_qgroup;
7c2ca468 2818 }
e02119d5
CM		 2819 blocksize =
2820 btrfs_level_size(tree_root,
2821 btrfs_super_log_root_level(disk_super));
d18a2c44 2822
6f07e42e 2823 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2824 if (!log_tree_root) {
2825 err = -ENOMEM;
bcef60f2 2826 goto fail_qgroup;
676e4c86 2827 }
e02119d5
CM		 2828
2829 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2830 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2831
2832 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2833 blocksize,
2834 generation + 1);
416bc658
JB		 2835 if (!log_tree_root->node ||
2836 !extent_buffer_uptodate(log_tree_root->node)) {
2837 printk(KERN_ERR "btrfs: failed to read log tree\n");
2838 free_extent_buffer(log_tree_root->node);
2839 kfree(log_tree_root);
2840 goto fail_trans_kthread;
2841 }
79787eaa 2842 /* returns with log_tree_root freed on success */
e02119d5 2843 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2844 if (ret) {
2845 btrfs_error(tree_root->fs_info, ret,
2846 "Failed to recover log tree");
2847 free_extent_buffer(log_tree_root->node);
2848 kfree(log_tree_root);
2849 goto fail_trans_kthread;
2850 }
e556ce2c
YZ		 2851
2852 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2853 ret = btrfs_commit_super(tree_root);
2854 if (ret)
2855 goto fail_trans_kthread;
e556ce2c 2856 }
e02119d5 2857 }
1a40e23b 2858
76dda93c 2859 ret = btrfs_find_orphan_roots(tree_root);
79787eaa
JM		 2860 if (ret)
2861 goto fail_trans_kthread;
76dda93c 2862
7c2ca468 2863 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2864 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2
ID		 2865 if (ret)
2866 goto fail_trans_kthread;
d68fc57b 2867
5d4f98a2 2868 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2869 if (ret < 0) {
2870 printk(KERN_WARNING
2871 "btrfs: failed to recover relocation\n");
2872 err = -EINVAL;
bcef60f2 2873 goto fail_qgroup;
d7ce5843 2874 }
7c2ca468 2875 }
1a40e23b 2876
3de4586c
CM		 2877 location.objectid = BTRFS_FS_TREE_OBJECTID;
2878 location.type = BTRFS_ROOT_ITEM_KEY;
cb517eab 2879 location.offset = 0;
3de4586c 2880
3de4586c 2881 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3140c9a3
DC		 2882 if (IS_ERR(fs_info->fs_root)) {
2883 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2884 goto fail_qgroup;
3140c9a3 2885 }
c289811c 2886
2b6ba629
ID		 2887 if (sb->s_flags & MS_RDONLY)
2888 return 0;
59641015 2889
2b6ba629
ID		 2890 down_read(&fs_info->cleanup_work_sem);
2891 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2892 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2893 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2894 close_ctree(tree_root);
2895 return ret;
2896 }
2897 up_read(&fs_info->cleanup_work_sem);
59641015 2898
2b6ba629
ID		 2899 ret = btrfs_resume_balance_async(fs_info);
2900 if (ret) {
2901 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2902 close_ctree(tree_root);
2903 return ret;
e3acc2a6
JB		 2904 }
2905
8dabb742
SB		 2906 ret = btrfs_resume_dev_replace_async(fs_info);
2907 if (ret) {
2908 pr_warn("btrfs: failed to resume dev_replace\n");
2909 close_ctree(tree_root);
2910 return ret;
2911 }
2912
b382a324
JS		 2913 btrfs_qgroup_rescan_resume(fs_info);
2914
f7a81ea4
SB		 2915 if (create_uuid_tree) {
2916 pr_info("btrfs: creating UUID tree\n");
2917 ret = btrfs_create_uuid_tree(fs_info);
2918 if (ret) {
2919 pr_warn("btrfs: failed to create the UUID tree %d\n",
2920 ret);
2921 close_ctree(tree_root);
2922 return ret;
2923 }
f420ee1e
SB		 2924 } else if (check_uuid_tree ||
2925 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
70f80175
SB		 2926 pr_info("btrfs: checking UUID tree\n");
2927 ret = btrfs_check_uuid_tree(fs_info);
2928 if (ret) {
2929 pr_warn("btrfs: failed to check the UUID tree %d\n",
2930 ret);
2931 close_ctree(tree_root);
2932 return ret;
2933 }
2934 } else {
2935 fs_info->update_uuid_tree_gen = 1;
f7a81ea4
SB		 2936 }
2937
ad2b2c80 2938 return 0;
39279cc3 2939
bcef60f2
AJ		 2940fail_qgroup:
2941 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2942fail_trans_kthread:
2943 kthread_stop(fs_info->transaction_kthread);
54067ae9 2944 btrfs_cleanup_transaction(fs_info->tree_root);
7b5ff90e 2945 del_fs_roots(fs_info);
3f157a2f 2946fail_cleaner:
a74a4b97 2947 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2948
2949 /*
2950 * make sure we're done with the btree inode before we stop our
2951 * kthreads
2952 */
2953 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2954
1b1d1f66 2955fail_block_groups:
54067ae9 2956 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2957 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2958
2959fail_tree_roots:
2960 free_root_pointers(fs_info, 1);
2b8195bb 2961 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2962
39279cc3 2963fail_sb_buffer:
7abadb64 2964 btrfs_stop_all_workers(fs_info);
16cdcec7 2965fail_alloc:
4543df7e 2966fail_iput:
586e46e2
ID		 2967 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2968
4543df7e 2969 iput(fs_info->btree_inode);
963d678b
MX		 2970fail_delalloc_bytes:
2971 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 2972fail_dirty_metadata_bytes:
2973 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 2974fail_bdi:
7e662854 2975 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2976fail_srcu:
2977 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2978fail:
53b381b3 2979 btrfs_free_stripe_hash_table(fs_info);
586e46e2 2980 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 2981 return err;
af31f5e5
CM		 2982
2983recovery_tree_root:
af31f5e5
CM		 2984 if (!btrfs_test_opt(tree_root, RECOVERY))
2985 goto fail_tree_roots;
2986
2987 free_root_pointers(fs_info, 0);
2988
2989 /* don't use the log in recovery mode, it won't be valid */
2990 btrfs_set_super_log_root(disk_super, 0);
2991
2992 /* we can't trust the free space cache either */
2993 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2994
2995 ret = next_root_backup(fs_info, fs_info->super_copy,
2996 &num_backups_tried, &backup_index);
2997 if (ret == -1)
2998 goto fail_block_groups;
2999 goto retry_root_backup;
eb60ceac
CM		 3000}
3001
f2984462
CM		 3002static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3003{
f2984462
CM		 3004 if (uptodate) {
3005 set_buffer_uptodate(bh);
3006 } else {
442a4f63
SB		 3007 struct btrfs_device *device = (struct btrfs_device *)
3008 bh->b_private;
3009
606686ee
JB		 3010 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3011 "I/O error on %s\n",
3012 rcu_str_deref(device->name));
1259ab75
CM		 3013 /* note, we dont' set_buffer_write_io_error because we have
3014 * our own ways of dealing with the IO errors
3015 */
f2984462 3016 clear_buffer_uptodate(bh);
442a4f63 3017 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 3018 }
3019 unlock_buffer(bh);
3020 put_bh(bh);
3021}
3022
a512bbf8
YZ		 3023struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3024{
3025 struct buffer_head *bh;
3026 struct buffer_head *latest = NULL;
3027 struct btrfs_super_block *super;
3028 int i;
3029 u64 transid = 0;
3030 u64 bytenr;
3031
3032 /* we would like to check all the supers, but that would make
3033 * a btrfs mount succeed after a mkfs from a different FS.
3034 * So, we need to add a special mount option to scan for
3035 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3036 */
3037 for (i = 0; i < 1; i++) {
3038 bytenr = btrfs_sb_offset(i);
8068a47e
AJ		 3039 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3040 i_size_read(bdev->bd_inode))
a512bbf8 3041 break;
8068a47e
AJ		 3042 bh = __bread(bdev, bytenr / 4096,
3043 BTRFS_SUPER_INFO_SIZE);
a512bbf8
YZ		 3044 if (!bh)
3045 continue;
3046
3047 super = (struct btrfs_super_block *)bh->b_data;
3048 if (btrfs_super_bytenr(super) != bytenr ||
3cae210f 3049 btrfs_super_magic(super) != BTRFS_MAGIC) {
a512bbf8
YZ		 3050 brelse(bh);
3051 continue;
3052 }
3053
3054 if (!latest || btrfs_super_generation(super) > transid) {
3055 brelse(latest);
3056 latest = bh;
3057 transid = btrfs_super_generation(super);
3058 } else {
3059 brelse(bh);
3060 }
3061 }
3062 return latest;
3063}
3064
4eedeb75
HH		 3065/*
3066 * this should be called twice, once with wait == 0 and
3067 * once with wait == 1. When wait == 0 is done, all the buffer heads
3068 * we write are pinned.
3069 *
3070 * They are released when wait == 1 is done.
3071 * max_mirrors must be the same for both runs, and it indicates how
3072 * many supers on this one device should be written.
3073 *
3074 * max_mirrors == 0 means to write them all.
3075 */
a512bbf8
YZ		 3076static int write_dev_supers(struct btrfs_device *device,
3077 struct btrfs_super_block *sb,
3078 int do_barriers, int wait, int max_mirrors)
3079{
3080 struct buffer_head *bh;
3081 int i;
3082 int ret;
3083 int errors = 0;
3084 u32 crc;
3085 u64 bytenr;
a512bbf8
YZ		 3086
3087 if (max_mirrors == 0)
3088 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3089
a512bbf8
YZ		 3090 for (i = 0; i < max_mirrors; i++) {
3091 bytenr = btrfs_sb_offset(i);
3092 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3093 break;
3094
3095 if (wait) {
3096 bh = __find_get_block(device->bdev, bytenr / 4096,
3097 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3098 if (!bh) {
3099 errors++;
3100 continue;
3101 }
a512bbf8 3102 wait_on_buffer(bh);
4eedeb75
HH		 3103 if (!buffer_uptodate(bh))
3104 errors++;
3105
3106 /* drop our reference */
3107 brelse(bh);
3108
3109 /* drop the reference from the wait == 0 run */
3110 brelse(bh);
3111 continue;
a512bbf8
YZ		 3112 } else {
3113 btrfs_set_super_bytenr(sb, bytenr);
3114
3115 crc = ~(u32)0;
b0496686 3116 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 3117 BTRFS_CSUM_SIZE, crc,
3118 BTRFS_SUPER_INFO_SIZE -
3119 BTRFS_CSUM_SIZE);
3120 btrfs_csum_final(crc, sb->csum);
3121
4eedeb75
HH		 3122 /*
3123 * one reference for us, and we leave it for the
3124 * caller
3125 */
a512bbf8
YZ		 3126 bh = __getblk(device->bdev, bytenr / 4096,
3127 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3128 if (!bh) {
3129 printk(KERN_ERR "btrfs: couldn't get super "
3130 "buffer head for bytenr %Lu\n", bytenr);
3131 errors++;
3132 continue;
3133 }
3134
a512bbf8
YZ		 3135 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3136
4eedeb75 3137 /* one reference for submit_bh */
a512bbf8 3138 get_bh(bh);
4eedeb75
HH		 3139
3140 set_buffer_uptodate(bh);
a512bbf8
YZ		 3141 lock_buffer(bh);
3142 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3143 bh->b_private = device;
a512bbf8
YZ		 3144 }
3145
387125fc
CM		 3146 /*
3147 * we fua the first super. The others we allow
3148 * to go down lazy.
3149 */
21adbd5c 3150 ret = btrfsic_submit_bh(WRITE_FUA, bh);
4eedeb75 3151 if (ret)
a512bbf8 3152 errors++;
a512bbf8
YZ		 3153 }
3154 return errors < i ? 0 : -1;
3155}
3156
387125fc
CM		 3157/*
3158 * endio for the write_dev_flush, this will wake anyone waiting
3159 * for the barrier when it is done
3160 */
3161static void btrfs_end_empty_barrier(struct bio *bio, int err)
3162{
3163 if (err) {
3164 if (err == -EOPNOTSUPP)
3165 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3166 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3167 }
3168 if (bio->bi_private)
3169 complete(bio->bi_private);
3170 bio_put(bio);
3171}
3172
3173/*
3174 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3175 * sent down. With wait == 1, it waits for the previous flush.
3176 *
3177 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3178 * capable
3179 */
3180static int write_dev_flush(struct btrfs_device *device, int wait)
3181{
3182 struct bio *bio;
3183 int ret = 0;
3184
3185 if (device->nobarriers)
3186 return 0;
3187
3188 if (wait) {
3189 bio = device->flush_bio;
3190 if (!bio)
3191 return 0;
3192
3193 wait_for_completion(&device->flush_wait);
3194
3195 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
606686ee
JB		 3196 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3197 rcu_str_deref(device->name));
387125fc 3198 device->nobarriers = 1;
5af3e8cc 3199 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3200 ret = -EIO;
5af3e8cc
SB		 3201 btrfs_dev_stat_inc_and_print(device,
3202 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3203 }
3204
3205 /* drop the reference from the wait == 0 run */
3206 bio_put(bio);
3207 device->flush_bio = NULL;
3208
3209 return ret;
3210 }
3211
3212 /*
3213 * one reference for us, and we leave it for the
3214 * caller
3215 */
9c017abc 3216 device->flush_bio = NULL;
9be3395b 3217 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
387125fc
CM		 3218 if (!bio)
3219 return -ENOMEM;
3220
3221 bio->bi_end_io = btrfs_end_empty_barrier;
3222 bio->bi_bdev = device->bdev;
3223 init_completion(&device->flush_wait);
3224 bio->bi_private = &device->flush_wait;
3225 device->flush_bio = bio;
3226
3227 bio_get(bio);
21adbd5c 3228 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3229
3230 return 0;
3231}
3232
3233/*
3234 * send an empty flush down to each device in parallel,
3235 * then wait for them
3236 */
3237static int barrier_all_devices(struct btrfs_fs_info *info)
3238{
3239 struct list_head *head;
3240 struct btrfs_device *dev;
5af3e8cc
SB		 3241 int errors_send = 0;
3242 int errors_wait = 0;
387125fc
CM		 3243 int ret;
3244
3245 /* send down all the barriers */
3246 head = &info->fs_devices->devices;
3247 list_for_each_entry_rcu(dev, head, dev_list) {
3248 if (!dev->bdev) {
5af3e8cc 3249 errors_send++;
387125fc
CM		 3250 continue;
3251 }
3252 if (!dev->in_fs_metadata || !dev->writeable)
3253 continue;
3254
3255 ret = write_dev_flush(dev, 0);
3256 if (ret)
5af3e8cc 3257 errors_send++;
387125fc
CM		 3258 }
3259
3260 /* wait for all the barriers */
3261 list_for_each_entry_rcu(dev, head, dev_list) {
3262 if (!dev->bdev) {
5af3e8cc 3263 errors_wait++;
387125fc
CM		 3264 continue;
3265 }
3266 if (!dev->in_fs_metadata || !dev->writeable)
3267 continue;
3268
3269 ret = write_dev_flush(dev, 1);
3270 if (ret)
5af3e8cc 3271 errors_wait++;
387125fc 3272 }
5af3e8cc
SB		 3273 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3274 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3275 return -EIO;
3276 return 0;
3277}
3278
5af3e8cc
SB		 3279int btrfs_calc_num_tolerated_disk_barrier_failures(
3280 struct btrfs_fs_info *fs_info)
3281{
3282 struct btrfs_ioctl_space_info space;
3283 struct btrfs_space_info *sinfo;
3284 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3285 BTRFS_BLOCK_GROUP_SYSTEM,
3286 BTRFS_BLOCK_GROUP_METADATA,
3287 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3288 int num_types = 4;
3289 int i;
3290 int c;
3291 int num_tolerated_disk_barrier_failures =
3292 (int)fs_info->fs_devices->num_devices;
3293
3294 for (i = 0; i < num_types; i++) {
3295 struct btrfs_space_info *tmp;
3296
3297 sinfo = NULL;
3298 rcu_read_lock();
3299 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3300 if (tmp->flags == types[i]) {
3301 sinfo = tmp;
3302 break;
3303 }
3304 }
3305 rcu_read_unlock();
3306
3307 if (!sinfo)
3308 continue;
3309
3310 down_read(&sinfo->groups_sem);
3311 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3312 if (!list_empty(&sinfo->block_groups[c])) {
3313 u64 flags;
3314
3315 btrfs_get_block_group_info(
3316 &sinfo->block_groups[c], &space);
3317 if (space.total_bytes == 0 ||
3318 space.used_bytes == 0)
3319 continue;
3320 flags = space.flags;
3321 /*
3322 * return
3323 * 0: if dup, single or RAID0 is configured for
3324 * any of metadata, system or data, else
3325 * 1: if RAID5 is configured, or if RAID1 or
3326 * RAID10 is configured and only two mirrors
3327 * are used, else
3328 * 2: if RAID6 is configured, else
3329 * num_mirrors - 1: if RAID1 or RAID10 is
3330 * configured and more than
3331 * 2 mirrors are used.
3332 */
3333 if (num_tolerated_disk_barrier_failures > 0 &&
3334 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3335 BTRFS_BLOCK_GROUP_RAID0)) ||
3336 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3337 == 0)))
3338 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3339 else if (num_tolerated_disk_barrier_failures > 1) {
3340 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3341 BTRFS_BLOCK_GROUP_RAID5 |
3342 BTRFS_BLOCK_GROUP_RAID10)) {
3343 num_tolerated_disk_barrier_failures = 1;
3344 } else if (flags &
15b0a89d 3345 BTRFS_BLOCK_GROUP_RAID6) {
53b381b3
DW		 3346 num_tolerated_disk_barrier_failures = 2;
3347 }
3348 }
5af3e8cc
SB		 3349 }
3350 }
3351 up_read(&sinfo->groups_sem);
3352 }
3353
3354 return num_tolerated_disk_barrier_failures;
3355}
3356
48a3b636 3357static int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3358{
e5e9a520 3359 struct list_head *head;
f2984462 3360 struct btrfs_device *dev;
a061fc8d 3361 struct btrfs_super_block *sb;
f2984462 3362 struct btrfs_dev_item *dev_item;
f2984462
CM		 3363 int ret;
3364 int do_barriers;
a236aed1
CM		 3365 int max_errors;
3366 int total_errors = 0;
a061fc8d 3367 u64 flags;
f2984462 3368
6c41761f 3369 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
f2984462 3370 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3371 backup_super_roots(root->fs_info);
f2984462 3372
6c41761f 3373 sb = root->fs_info->super_for_commit;
a061fc8d 3374 dev_item = &sb->dev_item;
e5e9a520 3375
174ba509 3376 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3377 head = &root->fs_info->fs_devices->devices;
387125fc 3378
5af3e8cc
SB		 3379 if (do_barriers) {
3380 ret = barrier_all_devices(root->fs_info);
3381 if (ret) {
3382 mutex_unlock(
3383 &root->fs_info->fs_devices->device_list_mutex);
3384 btrfs_error(root->fs_info, ret,
3385 "errors while submitting device barriers.");
3386 return ret;
3387 }
3388 }
387125fc 3389
1f78160c 3390 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3391 if (!dev->bdev) {
3392 total_errors++;
3393 continue;
3394 }
2b82032c 3395 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3396 continue;
3397
2b82032c 3398 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3399 btrfs_set_stack_device_type(dev_item, dev->type);
3400 btrfs_set_stack_device_id(dev_item, dev->devid);
3401 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3402 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3403 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3404 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3405 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3406 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3407 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3408
a061fc8d
CM		 3409 flags = btrfs_super_flags(sb);
3410 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3411
a512bbf8 3412 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3413 if (ret)
3414 total_errors++;
f2984462 3415 }
a236aed1 3416 if (total_errors > max_errors) {
d397712b
CM		 3417 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3418 total_errors);
79787eaa
JM		 3419
3420 /* This shouldn't happen. FUA is masked off if unsupported */
a236aed1
CM		 3421 BUG();
3422 }
f2984462 3423
a512bbf8 3424 total_errors = 0;
1f78160c 3425 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3426 if (!dev->bdev)
3427 continue;
2b82032c 3428 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3429 continue;
3430
a512bbf8
YZ		 3431 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3432 if (ret)
3433 total_errors++;
f2984462 3434 }
174ba509 3435 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3436 if (total_errors > max_errors) {
79787eaa
JM		 3437 btrfs_error(root->fs_info, -EIO,
3438 "%d errors while writing supers", total_errors);
3439 return -EIO;
a236aed1 3440 }
f2984462
CM		 3441 return 0;
3442}
3443
a512bbf8
YZ		 3444int write_ctree_super(struct btrfs_trans_handle *trans,
3445 struct btrfs_root *root, int max_mirrors)
eb60ceac 3446{
e66f709b 3447 int ret;
5f39d397 3448
a512bbf8 3449 ret = write_all_supers(root, max_mirrors);
5f39d397 3450 return ret;
cfaa7295
CM		 3451}
3452
cb517eab
MX		 3453/* Drop a fs root from the radix tree and free it. */
3454void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3455 struct btrfs_root *root)
2619ba1f 3456{
4df27c4d 3457 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3458 radix_tree_delete(&fs_info->fs_roots_radix,
3459 (unsigned long)root->root_key.objectid);
4df27c4d 3460 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3461
3462 if (btrfs_root_refs(&root->root_item) == 0)
3463 synchronize_srcu(&fs_info->subvol_srcu);
3464
d7634482 3465 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3321719e
LB		 3466 btrfs_free_log(NULL, root);
3467 btrfs_free_log_root_tree(NULL, fs_info);
3468 }
3469
581bb050
LZ		 3470 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3471 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3472 free_fs_root(root);
4df27c4d
YZ		 3473}
3474
3475static void free_fs_root(struct btrfs_root *root)
3476{
82d5902d 3477 iput(root->cache_inode);
4df27c4d 3478 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1cb048f5
FDBM		 3479 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3480 root->orphan_block_rsv = NULL;
0ee5dc67
AV		 3481 if (root->anon_dev)
3482 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 3483 free_extent_buffer(root->node);
3484 free_extent_buffer(root->commit_root);
581bb050
LZ		 3485 kfree(root->free_ino_ctl);
3486 kfree(root->free_ino_pinned);
d397712b 3487 kfree(root->name);
b0feb9d9 3488 btrfs_put_fs_root(root);
2619ba1f
CM		 3489}
3490
cb517eab
MX		 3491void btrfs_free_fs_root(struct btrfs_root *root)
3492{
3493 free_fs_root(root);
2619ba1f
CM		 3494}
3495
c146afad 3496int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3497{
c146afad
YZ		 3498 u64 root_objectid = 0;
3499 struct btrfs_root *gang[8];
3500 int i;
3768f368 3501 int ret;
e089f05c 3502
c146afad
YZ		 3503 while (1) {
3504 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3505 (void **)gang, root_objectid,
3506 ARRAY_SIZE(gang));
3507 if (!ret)
3508 break;
5d4f98a2
YZ		 3509
3510 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 3511 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 3512 int err;
3513
c146afad 3514 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3515 err = btrfs_orphan_cleanup(gang[i]);
3516 if (err)
3517 return err;
c146afad
YZ		 3518 }
3519 root_objectid++;
3520 }
3521 return 0;
3522}
a2135011 3523
c146afad
YZ		 3524int btrfs_commit_super(struct btrfs_root *root)
3525{
3526 struct btrfs_trans_handle *trans;
3527 int ret;
a74a4b97 3528
c146afad 3529 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3530 btrfs_run_delayed_iputs(root);
c146afad 3531 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3532 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3533
3534 /* wait until ongoing cleanup work done */
3535 down_write(&root->fs_info->cleanup_work_sem);
3536 up_write(&root->fs_info->cleanup_work_sem);
3537
7a7eaa40 3538 trans = btrfs_join_transaction(root);
3612b495
TI		 3539 if (IS_ERR(trans))
3540 return PTR_ERR(trans);
54aa1f4d 3541 ret = btrfs_commit_transaction(trans, root);
79787eaa
JM		 3542 if (ret)
3543 return ret;
c146afad 3544 /* run commit again to drop the original snapshot */
7a7eaa40 3545 trans = btrfs_join_transaction(root);
3612b495
TI		 3546 if (IS_ERR(trans))
3547 return PTR_ERR(trans);
79787eaa
JM		 3548 ret = btrfs_commit_transaction(trans, root);
3549 if (ret)
3550 return ret;
79154b1b 3551 ret = btrfs_write_and_wait_transaction(NULL, root);
79787eaa
JM		 3552 if (ret) {
3553 btrfs_error(root->fs_info, ret,
3554 "Failed to sync btree inode to disk.");
3555 return ret;
3556 }
d6bfde87 3557
a512bbf8 3558 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 3559 return ret;
3560}
3561
3562int close_ctree(struct btrfs_root *root)
3563{
3564 struct btrfs_fs_info *fs_info = root->fs_info;
3565 int ret;
3566
3567 fs_info->closing = 1;
3568 smp_mb();
3569
803b2f54
SB		 3570 /* wait for the uuid_scan task to finish */
3571 down(&fs_info->uuid_tree_rescan_sem);
3572 /* avoid complains from lockdep et al., set sem back to initial state */
3573 up(&fs_info->uuid_tree_rescan_sem);
3574
837d5b6e 3575 /* pause restriper - we want to resume on mount */
aa1b8cd4 3576 btrfs_pause_balance(fs_info);
837d5b6e 3577
8dabb742
SB		 3578 btrfs_dev_replace_suspend_for_unmount(fs_info);
3579
aa1b8cd4 3580 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3581
3582 /* wait for any defraggers to finish */
3583 wait_event(fs_info->transaction_wait,
3584 (atomic_read(&fs_info->defrag_running) == 0));
3585
3586 /* clear out the rbtree of defraggable inodes */
26176e7c 3587 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3588
c146afad 3589 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3590 ret = btrfs_commit_super(root);
3591 if (ret)
3592 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3593 }
3594
87533c47 3595 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3596 btrfs_error_commit_super(root);
0f7d52f4 3597
300e4f8a
JB		 3598 btrfs_put_block_group_cache(fs_info);
3599
e3029d9f
AV		 3600 kthread_stop(fs_info->transaction_kthread);
3601 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3602
f25784b3
YZ		 3603 fs_info->closing = 2;
3604 smp_mb();
3605
bcef60f2
AJ		 3606 btrfs_free_qgroup_config(root->fs_info);
3607
963d678b
MX		 3608 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3609 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3610 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3611 }
bcc63abb 3612
e3029d9f 3613 btrfs_free_block_groups(fs_info);
d10c5f31 3614
13e6c37b 3615 btrfs_stop_all_workers(fs_info);
2932505a 3616
c146afad 3617 del_fs_roots(fs_info);
d10c5f31 3618
13e6c37b 3619 free_root_pointers(fs_info, 1);
9ad6b7bc 3620
13e6c37b 3621 iput(fs_info->btree_inode);
d6bfde87 3622
21adbd5c
SB		 3623#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3624 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3625 btrfsic_unmount(root, fs_info->fs_devices);
3626#endif
3627
dfe25020 3628 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3629 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3630
e2d84521 3631 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3632 percpu_counter_destroy(&fs_info->delalloc_bytes);
04160088 3633 bdi_destroy(&fs_info->bdi);
76dda93c 3634 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3635
53b381b3
DW		 3636 btrfs_free_stripe_hash_table(fs_info);
3637
1cb048f5
FDBM		 3638 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3639 root->orphan_block_rsv = NULL;
3640
eb60ceac
CM		 3641 return 0;
3642}
3643
b9fab919
CM		 3644int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3645 int atomic)
5f39d397 3646{
1259ab75 3647 int ret;
727011e0 3648 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3649
0b32f4bb 3650 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3651 if (!ret)
3652 return ret;
3653
3654 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3655 parent_transid, atomic);
3656 if (ret == -EAGAIN)
3657 return ret;
1259ab75 3658 return !ret;
5f39d397
CM		 3659}
3660
3661int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3662{
0b32f4bb 3663 return set_extent_buffer_uptodate(buf);
5f39d397 3664}
6702ed49 3665
5f39d397
CM		 3666void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3667{
727011e0 3668 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
5f39d397 3669 u64 transid = btrfs_header_generation(buf);
b9473439 3670 int was_dirty;
b4ce94de 3671
b9447ef8 3672 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3673 if (transid != root->fs_info->generation)
3674 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3675 "found %llu running %llu\n",
c1c9ff7c 3676 buf->start, transid, root->fs_info->generation);
0b32f4bb 3677 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3678 if (!was_dirty)
3679 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3680 buf->len,
3681 root->fs_info->dirty_metadata_batch);
eb60ceac
CM		 3682}
3683
b53d3f5d
LB		 3684static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3685 int flush_delayed)
16cdcec7
MX		 3686{
3687 /*
3688 * looks as though older kernels can get into trouble with
3689 * this code, they end up stuck in balance_dirty_pages forever
3690 */
e2d84521 3691 int ret;
16cdcec7
MX		 3692
3693 if (current->flags & PF_MEMALLOC)
3694 return;
3695
b53d3f5d
LB		 3696 if (flush_delayed)
3697 btrfs_balance_delayed_items(root);
16cdcec7 3698
e2d84521
MX		 3699 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3700 BTRFS_DIRTY_METADATA_THRESH);
3701 if (ret > 0) {
d0e1d66b
NJ		 3702 balance_dirty_pages_ratelimited(
3703 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3704 }
3705 return;
3706}
3707
b53d3f5d 3708void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3709{
b53d3f5d
LB		 3710 __btrfs_btree_balance_dirty(root, 1);
3711}
585ad2c3 3712
b53d3f5d
LB		 3713void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3714{
3715 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3716}
6b80053d 3717
ca7a79ad 3718int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3719{
727011e0 3720 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3721 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3722}
0da5468f 3723
fcd1f065 3724static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3725 int read_only)
3726{
1104a885
DS		 3727 /*
3728 * Placeholder for checks
3729 */
fcd1f065 3730 return 0;
acce952b 3731}
3732
48a3b636 3733static void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3734{
acce952b 3735 mutex_lock(&root->fs_info->cleaner_mutex);
3736 btrfs_run_delayed_iputs(root);
3737 mutex_unlock(&root->fs_info->cleaner_mutex);
3738
3739 down_write(&root->fs_info->cleanup_work_sem);
3740 up_write(&root->fs_info->cleanup_work_sem);
3741
3742 /* cleanup FS via transaction */
3743 btrfs_cleanup_transaction(root);
acce952b 3744}
3745
569e0f35
JB		 3746static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3747 struct btrfs_root *root)
acce952b 3748{
3749 struct btrfs_inode *btrfs_inode;
3750 struct list_head splice;
3751
3752 INIT_LIST_HEAD(&splice);
3753
3754 mutex_lock(&root->fs_info->ordered_operations_mutex);
199c2a9c 3755 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3756
569e0f35 3757 list_splice_init(&t->ordered_operations, &splice);
acce952b 3758 while (!list_empty(&splice)) {
3759 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3760 ordered_operations);
3761
3762 list_del_init(&btrfs_inode->ordered_operations);
199c2a9c 3763 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3764
3765 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3766
199c2a9c 3767 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3768 }
3769
199c2a9c 3770 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3771 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3772}
3773
143bede5 3774static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3775{
acce952b 3776 struct btrfs_ordered_extent *ordered;
acce952b 3777
199c2a9c 3778 spin_lock(&root->ordered_extent_lock);
779880ef
JB		 3779 /*
3780 * This will just short circuit the ordered completion stuff which will
3781 * make sure the ordered extent gets properly cleaned up.
3782 */
199c2a9c 3783 list_for_each_entry(ordered, &root->ordered_extents,
779880ef
JB		 3784 root_extent_list)
3785 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
199c2a9c
MX		 3786 spin_unlock(&root->ordered_extent_lock);
3787}
3788
3789static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3790{
3791 struct btrfs_root *root;
3792 struct list_head splice;
3793
3794 INIT_LIST_HEAD(&splice);
3795
3796 spin_lock(&fs_info->ordered_root_lock);
3797 list_splice_init(&fs_info->ordered_roots, &splice);
3798 while (!list_empty(&splice)) {
3799 root = list_first_entry(&splice, struct btrfs_root,
3800 ordered_root);
3801 list_del_init(&root->ordered_root);
3802
3803 btrfs_destroy_ordered_extents(root);
3804
3805 cond_resched_lock(&fs_info->ordered_root_lock);
3806 }
3807 spin_unlock(&fs_info->ordered_root_lock);
acce952b 3808}
3809
35a3621b
SB		 3810static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3811 struct btrfs_root *root)
acce952b 3812{
3813 struct rb_node *node;
3814 struct btrfs_delayed_ref_root *delayed_refs;
3815 struct btrfs_delayed_ref_node *ref;
3816 int ret = 0;
3817
3818 delayed_refs = &trans->delayed_refs;
3819
3820 spin_lock(&delayed_refs->lock);
3821 if (delayed_refs->num_entries == 0) {
cfece4db 3822 spin_unlock(&delayed_refs->lock);
acce952b 3823 printk(KERN_INFO "delayed_refs has NO entry\n");
3824 return ret;
3825 }
3826
b939d1ab 3827 while ((node = rb_first(&delayed_refs->root)) != NULL) {
eb12db69 3828 struct btrfs_delayed_ref_head *head = NULL;
e78417d1 3829 bool pin_bytes = false;
acce952b 3830
eb12db69 3831 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
acce952b 3832 atomic_set(&ref->refs, 1);
3833 if (btrfs_delayed_ref_is_head(ref)) {
acce952b 3834
3835 head = btrfs_delayed_node_to_head(ref);
b939d1ab
JB		 3836 if (!mutex_trylock(&head->mutex)) {
3837 atomic_inc(&ref->refs);
3838 spin_unlock(&delayed_refs->lock);
3839
3840 /* Need to wait for the delayed ref to run */
3841 mutex_lock(&head->mutex);
3842 mutex_unlock(&head->mutex);
3843 btrfs_put_delayed_ref(ref);
3844
e18fca73 3845 spin_lock(&delayed_refs->lock);
b939d1ab
JB		 3846 continue;
3847 }
3848
54067ae9 3849 if (head->must_insert_reserved)
e78417d1 3850 pin_bytes = true;
78a6184a 3851 btrfs_free_delayed_extent_op(head->extent_op);
acce952b 3852 delayed_refs->num_heads--;
3853 if (list_empty(&head->cluster))
3854 delayed_refs->num_heads_ready--;
3855 list_del_init(&head->cluster);
acce952b 3856 }
eb12db69 3857
b939d1ab
JB		 3858 ref->in_tree = 0;
3859 rb_erase(&ref->rb_node, &delayed_refs->root);
3860 delayed_refs->num_entries--;
acce952b 3861 spin_unlock(&delayed_refs->lock);
e78417d1
JB		 3862 if (head) {
3863 if (pin_bytes)
3864 btrfs_pin_extent(root, ref->bytenr,
3865 ref->num_bytes, 1);
3866 mutex_unlock(&head->mutex);
3867 }
acce952b 3868 btrfs_put_delayed_ref(ref);
3869
3870 cond_resched();
3871 spin_lock(&delayed_refs->lock);
3872 }
3873
3874 spin_unlock(&delayed_refs->lock);
3875
3876 return ret;
3877}
3878
aec8030a 3879static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
acce952b 3880{
3881 struct btrfs_pending_snapshot *snapshot;
3882 struct list_head splice;
3883
3884 INIT_LIST_HEAD(&splice);
3885
3886 list_splice_init(&t->pending_snapshots, &splice);
3887
3888 while (!list_empty(&splice)) {
3889 snapshot = list_entry(splice.next,
3890 struct btrfs_pending_snapshot,
3891 list);
aec8030a 3892 snapshot->error = -ECANCELED;
acce952b 3893 list_del_init(&snapshot->list);
acce952b 3894 }
acce952b 3895}
3896
143bede5 3897static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3898{
3899 struct btrfs_inode *btrfs_inode;
3900 struct list_head splice;
3901
3902 INIT_LIST_HEAD(&splice);
3903
eb73c1b7
MX		 3904 spin_lock(&root->delalloc_lock);
3905 list_splice_init(&root->delalloc_inodes, &splice);
acce952b 3906
3907 while (!list_empty(&splice)) {
eb73c1b7
MX		 3908 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3909 delalloc_inodes);
acce952b 3910
3911 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3912 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3913 &btrfs_inode->runtime_flags);
eb73c1b7 3914 spin_unlock(&root->delalloc_lock);
acce952b 3915
3916 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3917
eb73c1b7 3918 spin_lock(&root->delalloc_lock);
acce952b 3919 }
3920
eb73c1b7
MX		 3921 spin_unlock(&root->delalloc_lock);
3922}
3923
3924static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3925{
3926 struct btrfs_root *root;
3927 struct list_head splice;
3928
3929 INIT_LIST_HEAD(&splice);
3930
3931 spin_lock(&fs_info->delalloc_root_lock);
3932 list_splice_init(&fs_info->delalloc_roots, &splice);
3933 while (!list_empty(&splice)) {
3934 root = list_first_entry(&splice, struct btrfs_root,
3935 delalloc_root);
3936 list_del_init(&root->delalloc_root);
3937 root = btrfs_grab_fs_root(root);
3938 BUG_ON(!root);
3939 spin_unlock(&fs_info->delalloc_root_lock);
3940
3941 btrfs_destroy_delalloc_inodes(root);
3942 btrfs_put_fs_root(root);
3943
3944 spin_lock(&fs_info->delalloc_root_lock);
3945 }
3946 spin_unlock(&fs_info->delalloc_root_lock);
acce952b 3947}
3948
3949static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3950 struct extent_io_tree *dirty_pages,
3951 int mark)
3952{
3953 int ret;
acce952b 3954 struct extent_buffer *eb;
3955 u64 start = 0;
3956 u64 end;
acce952b 3957
3958 while (1) {
3959 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3960 mark, NULL);
acce952b 3961 if (ret)
3962 break;
3963
3964 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3965 while (start <= end) {
fd8b2b61
JB		 3966 eb = btrfs_find_tree_block(root, start,
3967 root->leafsize);
69a85bd8 3968 start += root->leafsize;
fd8b2b61 3969 if (!eb)
acce952b 3970 continue;
fd8b2b61 3971 wait_on_extent_buffer_writeback(eb);
acce952b 3972
fd8b2b61
JB		 3973 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3974 &eb->bflags))
3975 clear_extent_buffer_dirty(eb);
3976 free_extent_buffer_stale(eb);
acce952b 3977 }
3978 }
3979
3980 return ret;
3981}
3982
3983static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3984 struct extent_io_tree *pinned_extents)
3985{
3986 struct extent_io_tree *unpin;
3987 u64 start;
3988 u64 end;
3989 int ret;
ed0eaa14 3990 bool loop = true;
acce952b 3991
3992 unpin = pinned_extents;
ed0eaa14 3993again:
acce952b 3994 while (1) {
3995 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 3996 EXTENT_DIRTY, NULL);
acce952b 3997 if (ret)
3998 break;
3999
4000 /* opt_discard */
5378e607
LD		 4001 if (btrfs_test_opt(root, DISCARD))
4002 ret = btrfs_error_discard_extent(root, start,
4003 end + 1 - start,
4004 NULL);
acce952b 4005
4006 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4007 btrfs_error_unpin_extent_range(root, start, end);
4008 cond_resched();
4009 }
4010
ed0eaa14
LB		 4011 if (loop) {
4012 if (unpin == &root->fs_info->freed_extents[0])
4013 unpin = &root->fs_info->freed_extents[1];
4014 else
4015 unpin = &root->fs_info->freed_extents[0];
4016 loop = false;
4017 goto again;
4018 }
4019
acce952b 4020 return 0;
4021}
4022
49b25e05
JM		 4023void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4024 struct btrfs_root *root)
4025{
4026 btrfs_destroy_delayed_refs(cur_trans, root);
4027 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
4028 cur_trans->dirty_pages.dirty_bytes);
4029
4a9d8bde 4030 cur_trans->state = TRANS_STATE_COMMIT_START;
d7096fc3 4031 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 4032
aec8030a
MX		 4033 btrfs_evict_pending_snapshots(cur_trans);
4034
4a9d8bde 4035 cur_trans->state = TRANS_STATE_UNBLOCKED;
d7096fc3 4036 wake_up(&root->fs_info->transaction_wait);
49b25e05 4037
67cde344
MX		 4038 btrfs_destroy_delayed_inodes(root);
4039 btrfs_assert_delayed_root_empty(root);
49b25e05 4040
49b25e05
JM		 4041 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4042 EXTENT_DIRTY);
6e841e32
LB		 4043 btrfs_destroy_pinned_extent(root,
4044 root->fs_info->pinned_extents);
49b25e05 4045
4a9d8bde
MX		 4046 cur_trans->state =TRANS_STATE_COMPLETED;
4047 wake_up(&cur_trans->commit_wait);
4048
49b25e05
JM		 4049 /*
4050 memset(cur_trans, 0, sizeof(*cur_trans));
4051 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4052 */
4053}
4054
48a3b636 4055static int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 4056{
4057 struct btrfs_transaction *t;
4058 LIST_HEAD(list);
4059
acce952b 4060 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4061
a4abeea4 4062 spin_lock(&root->fs_info->trans_lock);
acce952b 4063 list_splice_init(&root->fs_info->trans_list, &list);
ac673879 4064 root->fs_info->running_transaction = NULL;
a4abeea4
JB		 4065 spin_unlock(&root->fs_info->trans_lock);
4066
acce952b 4067 while (!list_empty(&list)) {
4068 t = list_entry(list.next, struct btrfs_transaction, list);
acce952b 4069
569e0f35 4070 btrfs_destroy_ordered_operations(t, root);
acce952b 4071
199c2a9c 4072 btrfs_destroy_all_ordered_extents(root->fs_info);
acce952b 4073
4074 btrfs_destroy_delayed_refs(t, root);
4075
4a9d8bde
MX		 4076 /*
4077 * FIXME: cleanup wait for commit
4078 * We needn't acquire the lock here, because we are during
4079 * the umount, there is no other task which will change it.
4080 */
4081 t->state = TRANS_STATE_COMMIT_START;
66657b31 4082 smp_mb();
acce952b 4083 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4084 wake_up(&root->fs_info->transaction_blocked_wait);
4085
aec8030a
MX		 4086 btrfs_evict_pending_snapshots(t);
4087
4a9d8bde 4088 t->state = TRANS_STATE_UNBLOCKED;
66657b31 4089 smp_mb();
acce952b 4090 if (waitqueue_active(&root->fs_info->transaction_wait))
4091 wake_up(&root->fs_info->transaction_wait);
acce952b 4092
67cde344
MX		 4093 btrfs_destroy_delayed_inodes(root);
4094 btrfs_assert_delayed_root_empty(root);
4095
eb73c1b7 4096 btrfs_destroy_all_delalloc_inodes(root->fs_info);
acce952b 4097
4098 btrfs_destroy_marked_extents(root, &t->dirty_pages,
4099 EXTENT_DIRTY);
4100
4101 btrfs_destroy_pinned_extent(root,
4102 root->fs_info->pinned_extents);
4103
4a9d8bde
MX		 4104 t->state = TRANS_STATE_COMPLETED;
4105 smp_mb();
4106 if (waitqueue_active(&t->commit_wait))
4107 wake_up(&t->commit_wait);
4108
13c5a93e 4109 atomic_set(&t->use_count, 0);
acce952b 4110 list_del_init(&t->list);
4111 memset(t, 0, sizeof(*t));
4112 kmem_cache_free(btrfs_transaction_cachep, t);
4113 }
4114
4115 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 4116
4117 return 0;
4118}
4119
d1310b2e 4120static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 4121 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 4122 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 4123 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 4124 /* note we're sharing with inode.c for the merge bio hook */
4125 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 4126};
Linux 6.x block layer and io_uring tree(s)