projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
block: submit_bio_wait() conversions
[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>
eb60ceac
CM		 36#include "ctree.h"
37#include "disk-io.h"
e089f05c 38#include "transaction.h"
0f7d52f4 39#include "btrfs_inode.h"
0b86a832 40#include "volumes.h"
db94535d 41#include "print-tree.h"
8b712842 42#include "async-thread.h"
925baedd 43#include "locking.h"
e02119d5 44#include "tree-log.h"
fa9c0d79 45#include "free-space-cache.h"
581bb050 46#include "inode-map.h"
21adbd5c 47#include "check-integrity.h"
606686ee 48#include "rcu-string.h"
8dabb742 49#include "dev-replace.h"
53b381b3 50#include "raid56.h"
eb60ceac 51
de0022b9
JB		 52#ifdef CONFIG_X86
53#include <asm/cpufeature.h>
54#endif
55
d1310b2e 56static struct extent_io_ops btree_extent_io_ops;
8b712842 57static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 58static void free_fs_root(struct btrfs_root *root);
fcd1f065 59static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 60 int read_only);
569e0f35
JB		 61static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
62 struct btrfs_root *root);
143bede5 63static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
acce952b 64static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
65 struct btrfs_root *root);
143bede5 66static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
acce952b 67static int btrfs_destroy_marked_extents(struct btrfs_root *root,
68 struct extent_io_tree *dirty_pages,
69 int mark);
70static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
71 struct extent_io_tree *pinned_extents);
48a3b636
ES		 72static int btrfs_cleanup_transaction(struct btrfs_root *root);
73static void btrfs_error_commit_super(struct btrfs_root *root);
ce9adaa5 74
d352ac68
CM		 75/*
76 * end_io_wq structs are used to do processing in task context when an IO is
77 * complete. This is used during reads to verify checksums, and it is used
78 * by writes to insert metadata for new file extents after IO is complete.
79 */
ce9adaa5
CM		 80struct end_io_wq {
81 struct bio *bio;
82 bio_end_io_t *end_io;
83 void *private;
84 struct btrfs_fs_info *info;
85 int error;
22c59948 86 int metadata;
ce9adaa5 87 struct list_head list;
8b712842 88 struct btrfs_work work;
ce9adaa5 89};
0da5468f 90
d352ac68
CM		 91/*
92 * async submit bios are used to offload expensive checksumming
93 * onto the worker threads. They checksum file and metadata bios
94 * just before they are sent down the IO stack.
95 */
44b8bd7e
CM		 96struct async_submit_bio {
97 struct inode *inode;
98 struct bio *bio;
99 struct list_head list;
4a69a410
CM		 100 extent_submit_bio_hook_t *submit_bio_start;
101 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 102 int rw;
103 int mirror_num;
c8b97818 104 unsigned long bio_flags;
eaf25d93
CM		 105 /*
106 * bio_offset is optional, can be used if the pages in the bio
107 * can't tell us where in the file the bio should go
108 */
109 u64 bio_offset;
8b712842 110 struct btrfs_work work;
79787eaa 111 int error;
44b8bd7e
CM		 112};
113
85d4e461
CM		 114/*
115 * Lockdep class keys for extent_buffer->lock's in this root. For a given
116 * eb, the lockdep key is determined by the btrfs_root it belongs to and
117 * the level the eb occupies in the tree.
118 *
119 * Different roots are used for different purposes and may nest inside each
120 * other and they require separate keysets. As lockdep keys should be
121 * static, assign keysets according to the purpose of the root as indicated
122 * by btrfs_root->objectid. This ensures that all special purpose roots
123 * have separate keysets.
4008c04a 124 *
85d4e461
CM		 125 * Lock-nesting across peer nodes is always done with the immediate parent
126 * node locked thus preventing deadlock. As lockdep doesn't know this, use
127 * subclass to avoid triggering lockdep warning in such cases.
4008c04a 128 *
85d4e461
CM		 129 * The key is set by the readpage_end_io_hook after the buffer has passed
130 * csum validation but before the pages are unlocked. It is also set by
131 * btrfs_init_new_buffer on freshly allocated blocks.
4008c04a 132 *
85d4e461
CM		 133 * We also add a check to make sure the highest level of the tree is the
134 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
135 * needs update as well.
4008c04a
CM		 136 */
137#ifdef CONFIG_DEBUG_LOCK_ALLOC
138# if BTRFS_MAX_LEVEL != 8
139# error
140# endif
85d4e461
CM		 141
142static struct btrfs_lockdep_keyset {
143 u64 id; /* root objectid */
144 const char *name_stem; /* lock name stem */
145 char names[BTRFS_MAX_LEVEL + 1][20];
146 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
147} btrfs_lockdep_keysets[] = {
148 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
149 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
150 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
151 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
152 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
153 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
60b62978 154 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
85d4e461
CM		 155 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
156 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
157 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
13fd8da9 158 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
85d4e461 159 { .id = 0, .name_stem = "tree" },
4008c04a 160};
85d4e461
CM		 161
162void __init btrfs_init_lockdep(void)
163{
164 int i, j;
165
166 /* initialize lockdep class names */
167 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
168 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
169
170 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
171 snprintf(ks->names[j], sizeof(ks->names[j]),
172 "btrfs-%s-%02d", ks->name_stem, j);
173 }
174}
175
176void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
177 int level)
178{
179 struct btrfs_lockdep_keyset *ks;
180
181 BUG_ON(level >= ARRAY_SIZE(ks->keys));
182
183 /* find the matching keyset, id 0 is the default entry */
184 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
185 if (ks->id == objectid)
186 break;
187
188 lockdep_set_class_and_name(&eb->lock,
189 &ks->keys[level], ks->names[level]);
190}
191
4008c04a
CM		 192#endif
193
d352ac68
CM		 194/*
195 * extents on the btree inode are pretty simple, there's one extent
196 * that covers the entire device
197 */
b2950863 198static struct extent_map *btree_get_extent(struct inode *inode,
306e16ce 199 struct page *page, size_t pg_offset, u64 start, u64 len,
b2950863 200 int create)
7eccb903 201{
5f39d397
CM		 202 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
203 struct extent_map *em;
204 int ret;
205
890871be 206 read_lock(&em_tree->lock);
d1310b2e 207 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 208 if (em) {
209 em->bdev =
210 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 211 read_unlock(&em_tree->lock);
5f39d397 212 goto out;
a061fc8d 213 }
890871be 214 read_unlock(&em_tree->lock);
7b13b7b1 215
172ddd60 216 em = alloc_extent_map();
5f39d397
CM		 217 if (!em) {
218 em = ERR_PTR(-ENOMEM);
219 goto out;
220 }
221 em->start = 0;
0afbaf8c 222 em->len = (u64)-1;
c8b97818 223 em->block_len = (u64)-1;
5f39d397 224 em->block_start = 0;
a061fc8d 225 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 226
890871be 227 write_lock(&em_tree->lock);
09a2a8f9 228 ret = add_extent_mapping(em_tree, em, 0);
5f39d397
CM		 229 if (ret == -EEXIST) {
230 free_extent_map(em);
7b13b7b1 231 em = lookup_extent_mapping(em_tree, start, len);
b4f359ab 232 if (!em)
0433f20d 233 em = ERR_PTR(-EIO);
5f39d397 234 } else if (ret) {
7b13b7b1 235 free_extent_map(em);
0433f20d 236 em = ERR_PTR(ret);
5f39d397 237 }
890871be 238 write_unlock(&em_tree->lock);
7b13b7b1 239
5f39d397
CM		 240out:
241 return em;
7eccb903
CM		 242}
243
b0496686 244u32 btrfs_csum_data(char *data, u32 seed, size_t len)
19c00ddc 245{
163e783e 246 return crc32c(seed, data, len);
19c00ddc
CM		 247}
248
249void btrfs_csum_final(u32 crc, char *result)
250{
7e75bf3f 251 put_unaligned_le32(~crc, result);
19c00ddc
CM		 252}
253
d352ac68
CM		 254/*
255 * compute the csum for a btree block, and either verify it or write it
256 * into the csum field of the block.
257 */
19c00ddc
CM		 258static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
259 int verify)
260{
6c41761f 261 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
607d432d 262 char *result = NULL;
19c00ddc
CM		 263 unsigned long len;
264 unsigned long cur_len;
265 unsigned long offset = BTRFS_CSUM_SIZE;
19c00ddc
CM		 266 char *kaddr;
267 unsigned long map_start;
268 unsigned long map_len;
269 int err;
270 u32 crc = ~(u32)0;
607d432d 271 unsigned long inline_result;
19c00ddc
CM		 272
273 len = buf->len - offset;
d397712b 274 while (len > 0) {
19c00ddc 275 err = map_private_extent_buffer(buf, offset, 32,
a6591715 276 &kaddr, &map_start, &map_len);
d397712b 277 if (err)
19c00ddc 278 return 1;
19c00ddc 279 cur_len = min(len, map_len - (offset - map_start));
b0496686 280 crc = btrfs_csum_data(kaddr + offset - map_start,
19c00ddc
CM		 281 crc, cur_len);
282 len -= cur_len;
283 offset += cur_len;
19c00ddc 284 }
607d432d
JB		 285 if (csum_size > sizeof(inline_result)) {
286 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
287 if (!result)
288 return 1;
289 } else {
290 result = (char *)&inline_result;
291 }
292
19c00ddc
CM		 293 btrfs_csum_final(crc, result);
294
295 if (verify) {
607d432d 296 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 297 u32 val;
298 u32 found = 0;
607d432d 299 memcpy(&found, result, csum_size);
e4204ded 300
607d432d 301 read_extent_buffer(buf, &val, 0, csum_size);
7a36ddec 302 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
193f284d
CM		 303 "failed on %llu wanted %X found %X "
304 "level %d\n",
c1c9ff7c
GU		 305 root->fs_info->sb->s_id, buf->start,
306 val, found, btrfs_header_level(buf));
607d432d
JB		 307 if (result != (char *)&inline_result)
308 kfree(result);
19c00ddc
CM		 309 return 1;
310 }
311 } else {
607d432d 312 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 313 }
607d432d
JB		 314 if (result != (char *)&inline_result)
315 kfree(result);
19c00ddc
CM		 316 return 0;
317}
318
d352ac68
CM		 319/*
320 * we can't consider a given block up to date unless the transid of the
321 * block matches the transid in the parent node's pointer. This is how we
322 * detect blocks that either didn't get written at all or got written
323 * in the wrong place.
324 */
1259ab75 325static int verify_parent_transid(struct extent_io_tree *io_tree,
b9fab919
CM		 326 struct extent_buffer *eb, u64 parent_transid,
327 int atomic)
1259ab75 328{
2ac55d41 329 struct extent_state *cached_state = NULL;
1259ab75
CM		 330 int ret;
331
332 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
333 return 0;
334
b9fab919
CM		 335 if (atomic)
336 return -EAGAIN;
337
2ac55d41 338 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
d0082371 339 0, &cached_state);
0b32f4bb 340 if (extent_buffer_uptodate(eb) &&
1259ab75
CM		 341 btrfs_header_generation(eb) == parent_transid) {
342 ret = 0;
343 goto out;
344 }
7a36ddec 345 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
193f284d 346 "found %llu\n",
c1c9ff7c 347 eb->start, parent_transid, btrfs_header_generation(eb));
1259ab75 348 ret = 1;
0b32f4bb 349 clear_extent_buffer_uptodate(eb);
33958dc6 350out:
2ac55d41
JB		 351 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
352 &cached_state, GFP_NOFS);
1259ab75 353 return ret;
1259ab75
CM		 354}
355
1104a885
DS		 356/*
357 * Return 0 if the superblock checksum type matches the checksum value of that
358 * algorithm. Pass the raw disk superblock data.
359 */
360static int btrfs_check_super_csum(char *raw_disk_sb)
361{
362 struct btrfs_super_block *disk_sb =
363 (struct btrfs_super_block *)raw_disk_sb;
364 u16 csum_type = btrfs_super_csum_type(disk_sb);
365 int ret = 0;
366
367 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
368 u32 crc = ~(u32)0;
369 const int csum_size = sizeof(crc);
370 char result[csum_size];
371
372 /*
373 * The super_block structure does not span the whole
374 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
375 * is filled with zeros and is included in the checkum.
376 */
377 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
378 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
379 btrfs_csum_final(crc, result);
380
381 if (memcmp(raw_disk_sb, result, csum_size))
382 ret = 1;
667e7d94
CM		 383
384 if (ret && btrfs_super_generation(disk_sb) < 10) {
385 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
386 ret = 0;
387 }
1104a885
DS		 388 }
389
390 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
391 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
392 csum_type);
393 ret = 1;
394 }
395
396 return ret;
397}
398
d352ac68
CM		 399/*
400 * helper to read a given tree block, doing retries as required when
401 * the checksums don't match and we have alternate mirrors to try.
402 */
f188591e
CM		 403static int btree_read_extent_buffer_pages(struct btrfs_root *root,
404 struct extent_buffer *eb,
ca7a79ad 405 u64 start, u64 parent_transid)
f188591e
CM		 406{
407 struct extent_io_tree *io_tree;
ea466794 408 int failed = 0;
f188591e
CM		 409 int ret;
410 int num_copies = 0;
411 int mirror_num = 0;
ea466794 412 int failed_mirror = 0;
f188591e 413
a826d6dc 414 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM		 415 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
416 while (1) {
bb82ab88
AJ		 417 ret = read_extent_buffer_pages(io_tree, eb, start,
418 WAIT_COMPLETE,
f188591e 419 btree_get_extent, mirror_num);
256dd1bb
SB		 420 if (!ret) {
421 if (!verify_parent_transid(io_tree, eb,
b9fab919 422 parent_transid, 0))
256dd1bb
SB		 423 break;
424 else
425 ret = -EIO;
426 }
d397712b 427
a826d6dc
JB		 428 /*
429 * This buffer's crc is fine, but its contents are corrupted, so
430 * there is no reason to read the other copies, they won't be
431 * any less wrong.
432 */
433 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
ea466794
JB		 434 break;
435
5d964051 436 num_copies = btrfs_num_copies(root->fs_info,
f188591e 437 eb->start, eb->len);
4235298e 438 if (num_copies == 1)
ea466794 439 break;
4235298e 440
5cf1ab56
JB		 441 if (!failed_mirror) {
442 failed = 1;
443 failed_mirror = eb->read_mirror;
444 }
445
f188591e 446 mirror_num++;
ea466794
JB		 447 if (mirror_num == failed_mirror)
448 mirror_num++;
449
4235298e 450 if (mirror_num > num_copies)
ea466794 451 break;
f188591e 452 }
ea466794 453
c0901581 454 if (failed && !ret && failed_mirror)
ea466794
JB		 455 repair_eb_io_failure(root, eb, failed_mirror);
456
457 return ret;
f188591e 458}
19c00ddc 459
d352ac68 460/*
d397712b
CM		 461 * checksum a dirty tree block before IO. This has extra checks to make sure
462 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 463 */
d397712b 464
b2950863 465static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 466{
d1310b2e 467 struct extent_io_tree *tree;
4eee4fa4 468 u64 start = page_offset(page);
19c00ddc 469 u64 found_start;
19c00ddc 470 struct extent_buffer *eb;
f188591e 471
d1310b2e 472 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc 473
4f2de97a
JB		 474 eb = (struct extent_buffer *)page->private;
475 if (page != eb->pages[0])
476 return 0;
19c00ddc 477 found_start = btrfs_header_bytenr(eb);
fae7f21c 478 if (WARN_ON(found_start != start || !PageUptodate(page)))
4f2de97a 479 return 0;
19c00ddc 480 csum_tree_block(root, eb, 0);
19c00ddc
CM		 481 return 0;
482}
483
2b82032c
YZ		 484static int check_tree_block_fsid(struct btrfs_root *root,
485 struct extent_buffer *eb)
486{
487 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
488 u8 fsid[BTRFS_UUID_SIZE];
489 int ret = 1;
490
0a4e5586 491 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
2b82032c
YZ		 492 while (fs_devices) {
493 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
494 ret = 0;
495 break;
496 }
497 fs_devices = fs_devices->seed;
498 }
499 return ret;
500}
501
a826d6dc
JB		 502#define CORRUPT(reason, eb, root, slot) \
503 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
504 "root=%llu, slot=%d\n", reason, \
c1c9ff7c 505 btrfs_header_bytenr(eb), root->objectid, slot)
a826d6dc
JB		 506
507static noinline int check_leaf(struct btrfs_root *root,
508 struct extent_buffer *leaf)
509{
510 struct btrfs_key key;
511 struct btrfs_key leaf_key;
512 u32 nritems = btrfs_header_nritems(leaf);
513 int slot;
514
515 if (nritems == 0)
516 return 0;
517
518 /* Check the 0 item */
519 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
520 BTRFS_LEAF_DATA_SIZE(root)) {
521 CORRUPT("invalid item offset size pair", leaf, root, 0);
522 return -EIO;
523 }
524
525 /*
526 * Check to make sure each items keys are in the correct order and their
527 * offsets make sense. We only have to loop through nritems-1 because
528 * we check the current slot against the next slot, which verifies the
529 * next slot's offset+size makes sense and that the current's slot
530 * offset is correct.
531 */
532 for (slot = 0; slot < nritems - 1; slot++) {
533 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
534 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
535
536 /* Make sure the keys are in the right order */
537 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
538 CORRUPT("bad key order", leaf, root, slot);
539 return -EIO;
540 }
541
542 /*
543 * Make sure the offset and ends are right, remember that the
544 * item data starts at the end of the leaf and grows towards the
545 * front.
546 */
547 if (btrfs_item_offset_nr(leaf, slot) !=
548 btrfs_item_end_nr(leaf, slot + 1)) {
549 CORRUPT("slot offset bad", leaf, root, slot);
550 return -EIO;
551 }
552
553 /*
554 * Check to make sure that we don't point outside of the leaf,
555 * just incase all the items are consistent to eachother, but
556 * all point outside of the leaf.
557 */
558 if (btrfs_item_end_nr(leaf, slot) >
559 BTRFS_LEAF_DATA_SIZE(root)) {
560 CORRUPT("slot end outside of leaf", leaf, root, slot);
561 return -EIO;
562 }
563 }
564
565 return 0;
566}
567
facc8a22
MX		 568static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
569 u64 phy_offset, struct page *page,
570 u64 start, u64 end, int mirror)
ce9adaa5
CM		 571{
572 struct extent_io_tree *tree;
573 u64 found_start;
574 int found_level;
ce9adaa5
CM		 575 struct extent_buffer *eb;
576 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 577 int ret = 0;
727011e0 578 int reads_done;
ce9adaa5 579
ce9adaa5
CM		 580 if (!page->private)
581 goto out;
d397712b 582
727011e0 583 tree = &BTRFS_I(page->mapping->host)->io_tree;
4f2de97a 584 eb = (struct extent_buffer *)page->private;
d397712b 585
0b32f4bb
JB		 586 /* the pending IO might have been the only thing that kept this buffer
587 * in memory. Make sure we have a ref for all this other checks
588 */
589 extent_buffer_get(eb);
590
591 reads_done = atomic_dec_and_test(&eb->io_pages);
727011e0
CM		 592 if (!reads_done)
593 goto err;
f188591e 594
5cf1ab56 595 eb->read_mirror = mirror;
ea466794
JB		 596 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
597 ret = -EIO;
598 goto err;
599 }
600
ce9adaa5 601 found_start = btrfs_header_bytenr(eb);
727011e0 602 if (found_start != eb->start) {
7a36ddec 603 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
193f284d 604 "%llu %llu\n",
c1c9ff7c 605 found_start, eb->start);
f188591e 606 ret = -EIO;
ce9adaa5
CM		 607 goto err;
608 }
2b82032c 609 if (check_tree_block_fsid(root, eb)) {
7a36ddec 610 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
c1c9ff7c 611 eb->start);
1259ab75
CM		 612 ret = -EIO;
613 goto err;
614 }
ce9adaa5 615 found_level = btrfs_header_level(eb);
1c24c3ce
JB		 616 if (found_level >= BTRFS_MAX_LEVEL) {
617 btrfs_info(root->fs_info, "bad tree block level %d\n",
618 (int)btrfs_header_level(eb));
619 ret = -EIO;
620 goto err;
621 }
ce9adaa5 622
85d4e461
CM		 623 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
624 eb, found_level);
4008c04a 625
ce9adaa5 626 ret = csum_tree_block(root, eb, 1);
a826d6dc 627 if (ret) {
f188591e 628 ret = -EIO;
a826d6dc
JB		 629 goto err;
630 }
631
632 /*
633 * If this is a leaf block and it is corrupt, set the corrupt bit so
634 * that we don't try and read the other copies of this block, just
635 * return -EIO.
636 */
637 if (found_level == 0 && check_leaf(root, eb)) {
638 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
639 ret = -EIO;
640 }
ce9adaa5 641
0b32f4bb
JB		 642 if (!ret)
643 set_extent_buffer_uptodate(eb);
ce9adaa5 644err:
79fb65a1
JB		 645 if (reads_done &&
646 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 647 btree_readahead_hook(root, eb, eb->start, ret);
4bb31e92 648
53b381b3
DW		 649 if (ret) {
650 /*
651 * our io error hook is going to dec the io pages
652 * again, we have to make sure it has something
653 * to decrement
654 */
655 atomic_inc(&eb->io_pages);
0b32f4bb 656 clear_extent_buffer_uptodate(eb);
53b381b3 657 }
0b32f4bb 658 free_extent_buffer(eb);
ce9adaa5 659out:
f188591e 660 return ret;
ce9adaa5
CM		 661}
662
ea466794 663static int btree_io_failed_hook(struct page *page, int failed_mirror)
4bb31e92 664{
4bb31e92
AJ		 665 struct extent_buffer *eb;
666 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
667
4f2de97a 668 eb = (struct extent_buffer *)page->private;
ea466794 669 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
5cf1ab56 670 eb->read_mirror = failed_mirror;
53b381b3 671 atomic_dec(&eb->io_pages);
ea466794 672 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 673 btree_readahead_hook(root, eb, eb->start, -EIO);
4bb31e92
AJ		 674 return -EIO; /* we fixed nothing */
675}
676
ce9adaa5 677static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 678{
679 struct end_io_wq *end_io_wq = bio->bi_private;
680 struct btrfs_fs_info *fs_info;
ce9adaa5 681
ce9adaa5 682 fs_info = end_io_wq->info;
ce9adaa5 683 end_io_wq->error = err;
8b712842
CM		 684 end_io_wq->work.func = end_workqueue_fn;
685 end_io_wq->work.flags = 0;
d20f7043 686
7b6d91da 687 if (bio->bi_rw & REQ_WRITE) {
53b381b3 688 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
cad321ad
CM		 689 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
690 &end_io_wq->work);
53b381b3 691 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
0cb59c99
JB		 692 btrfs_queue_worker(&fs_info->endio_freespace_worker,
693 &end_io_wq->work);
53b381b3
DW		 694 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
695 btrfs_queue_worker(&fs_info->endio_raid56_workers,
696 &end_io_wq->work);
cad321ad
CM		 697 else
698 btrfs_queue_worker(&fs_info->endio_write_workers,
699 &end_io_wq->work);
d20f7043 700 } else {
53b381b3
DW		 701 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
702 btrfs_queue_worker(&fs_info->endio_raid56_workers,
703 &end_io_wq->work);
704 else if (end_io_wq->metadata)
d20f7043
CM		 705 btrfs_queue_worker(&fs_info->endio_meta_workers,
706 &end_io_wq->work);
707 else
708 btrfs_queue_worker(&fs_info->endio_workers,
709 &end_io_wq->work);
710 }
ce9adaa5
CM		 711}
712
0cb59c99
JB		 713/*
714 * For the metadata arg you want
715 *
716 * 0 - if data
717 * 1 - if normal metadta
718 * 2 - if writing to the free space cache area
53b381b3 719 * 3 - raid parity work
0cb59c99 720 */
22c59948
CM		 721int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
722 int metadata)
0b86a832 723{
ce9adaa5 724 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 725 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
726 if (!end_io_wq)
727 return -ENOMEM;
728
729 end_io_wq->private = bio->bi_private;
730 end_io_wq->end_io = bio->bi_end_io;
22c59948 731 end_io_wq->info = info;
ce9adaa5
CM		 732 end_io_wq->error = 0;
733 end_io_wq->bio = bio;
22c59948 734 end_io_wq->metadata = metadata;
ce9adaa5
CM		 735
736 bio->bi_private = end_io_wq;
737 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 738 return 0;
739}
740
b64a2851 741unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 742{
4854ddd0
CM		 743 unsigned long limit = min_t(unsigned long,
744 info->workers.max_workers,
745 info->fs_devices->open_devices);
746 return 256 * limit;
747}
0986fe9e 748
4a69a410
CM		 749static void run_one_async_start(struct btrfs_work *work)
750{
4a69a410 751 struct async_submit_bio *async;
79787eaa 752 int ret;
4a69a410
CM		 753
754 async = container_of(work, struct async_submit_bio, work);
79787eaa
JM		 755 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
756 async->mirror_num, async->bio_flags,
757 async->bio_offset);
758 if (ret)
759 async->error = ret;
4a69a410
CM		 760}
761
762static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 763{
764 struct btrfs_fs_info *fs_info;
765 struct async_submit_bio *async;
4854ddd0 766 int limit;
8b712842
CM		 767
768 async = container_of(work, struct async_submit_bio, work);
769 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 770
b64a2851 771 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 772 limit = limit * 2 / 3;
773
66657b31 774 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
b64a2851 775 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 776 wake_up(&fs_info->async_submit_wait);
777
79787eaa
JM		 778 /* If an error occured we just want to clean up the bio and move on */
779 if (async->error) {
780 bio_endio(async->bio, async->error);
781 return;
782 }
783
4a69a410 784 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 785 async->mirror_num, async->bio_flags,
786 async->bio_offset);
4a69a410
CM		 787}
788
789static void run_one_async_free(struct btrfs_work *work)
790{
791 struct async_submit_bio *async;
792
793 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 794 kfree(async);
795}
796
44b8bd7e
CM		 797int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
798 int rw, struct bio *bio, int mirror_num,
c8b97818 799 unsigned long bio_flags,
eaf25d93 800 u64 bio_offset,
4a69a410
CM		 801 extent_submit_bio_hook_t *submit_bio_start,
802 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 803{
804 struct async_submit_bio *async;
805
806 async = kmalloc(sizeof(*async), GFP_NOFS);
807 if (!async)
808 return -ENOMEM;
809
810 async->inode = inode;
811 async->rw = rw;
812 async->bio = bio;
813 async->mirror_num = mirror_num;
4a69a410
CM		 814 async->submit_bio_start = submit_bio_start;
815 async->submit_bio_done = submit_bio_done;
816
817 async->work.func = run_one_async_start;
818 async->work.ordered_func = run_one_async_done;
819 async->work.ordered_free = run_one_async_free;
820
8b712842 821 async->work.flags = 0;
c8b97818 822 async->bio_flags = bio_flags;
eaf25d93 823 async->bio_offset = bio_offset;
8c8bee1d 824
79787eaa
JM		 825 async->error = 0;
826
cb03c743 827 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 828
7b6d91da 829 if (rw & REQ_SYNC)
d313d7a3
CM		 830 btrfs_set_work_high_prio(&async->work);
831
8b712842 832 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 833
d397712b 834 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 835 atomic_read(&fs_info->nr_async_submits)) {
836 wait_event(fs_info->async_submit_wait,
837 (atomic_read(&fs_info->nr_async_submits) == 0));
838 }
839
44b8bd7e
CM		 840 return 0;
841}
842
ce3ed71a
CM		 843static int btree_csum_one_bio(struct bio *bio)
844{
845 struct bio_vec *bvec = bio->bi_io_vec;
846 int bio_index = 0;
847 struct btrfs_root *root;
79787eaa 848 int ret = 0;
ce3ed71a
CM		 849
850 WARN_ON(bio->bi_vcnt <= 0);
d397712b 851 while (bio_index < bio->bi_vcnt) {
ce3ed71a 852 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 853 ret = csum_dirty_buffer(root, bvec->bv_page);
854 if (ret)
855 break;
ce3ed71a
CM		 856 bio_index++;
857 bvec++;
858 }
79787eaa 859 return ret;
ce3ed71a
CM		 860}
861
4a69a410
CM		 862static int __btree_submit_bio_start(struct inode *inode, int rw,
863 struct bio *bio, int mirror_num,
eaf25d93
CM		 864 unsigned long bio_flags,
865 u64 bio_offset)
22c59948 866{
8b712842
CM		 867 /*
868 * when we're called for a write, we're already in the async
5443be45 869 * submission context. Just jump into btrfs_map_bio
8b712842 870 */
79787eaa 871 return btree_csum_one_bio(bio);
4a69a410 872}
22c59948 873
4a69a410 874static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 875 int mirror_num, unsigned long bio_flags,
876 u64 bio_offset)
4a69a410 877{
61891923
SB		 878 int ret;
879
8b712842 880 /*
4a69a410
CM		 881 * when we're called for a write, we're already in the async
882 * submission context. Just jump into btrfs_map_bio
8b712842 883 */
61891923
SB		 884 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
885 if (ret)
886 bio_endio(bio, ret);
887 return ret;
0b86a832
CM		 888}
889
de0022b9
JB		 890static int check_async_write(struct inode *inode, unsigned long bio_flags)
891{
892 if (bio_flags & EXTENT_BIO_TREE_LOG)
893 return 0;
894#ifdef CONFIG_X86
895 if (cpu_has_xmm4_2)
896 return 0;
897#endif
898 return 1;
899}
900
44b8bd7e 901static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 902 int mirror_num, unsigned long bio_flags,
903 u64 bio_offset)
44b8bd7e 904{
de0022b9 905 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 906 int ret;
907
7b6d91da 908 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 909 /*
910 * called for a read, do the setup so that checksum validation
911 * can happen in the async kernel threads
912 */
f3f266ab
CM		 913 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
914 bio, 1);
1d4284bd 915 if (ret)
61891923
SB		 916 goto out_w_error;
917 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
918 mirror_num, 0);
de0022b9
JB		 919 } else if (!async) {
920 ret = btree_csum_one_bio(bio);
921 if (ret)
61891923
SB		 922 goto out_w_error;
923 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
924 mirror_num, 0);
925 } else {
926 /*
927 * kthread helpers are used to submit writes so that
928 * checksumming can happen in parallel across all CPUs
929 */
930 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
931 inode, rw, bio, mirror_num, 0,
932 bio_offset,
933 __btree_submit_bio_start,
934 __btree_submit_bio_done);
44b8bd7e 935 }
d313d7a3 936
61891923
SB		 937 if (ret) {
938out_w_error:
939 bio_endio(bio, ret);
940 }
941 return ret;
44b8bd7e
CM		 942}
943
3dd1462e 944#ifdef CONFIG_MIGRATION
784b4e29 945static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 946 struct page *newpage, struct page *page,
947 enum migrate_mode mode)
784b4e29
CM		 948{
949 /*
950 * we can't safely write a btree page from here,
951 * we haven't done the locking hook
952 */
953 if (PageDirty(page))
954 return -EAGAIN;
955 /*
956 * Buffers may be managed in a filesystem specific way.
957 * We must have no buffers or drop them.
958 */
959 if (page_has_private(page) &&
960 !try_to_release_page(page, GFP_KERNEL))
961 return -EAGAIN;
a6bc32b8 962 return migrate_page(mapping, newpage, page, mode);
784b4e29 963}
3dd1462e 964#endif
784b4e29 965
0da5468f
CM		 966
967static int btree_writepages(struct address_space *mapping,
968 struct writeback_control *wbc)
969{
d1310b2e 970 struct extent_io_tree *tree;
e2d84521
MX		 971 struct btrfs_fs_info *fs_info;
972 int ret;
973
d1310b2e 974 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 975 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 976
977 if (wbc->for_kupdate)
978 return 0;
979
e2d84521 980 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 981 /* this is a bit racy, but that's ok */
e2d84521
MX		 982 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
983 BTRFS_DIRTY_METADATA_THRESH);
984 if (ret < 0)
793955bc 985 return 0;
793955bc 986 }
0b32f4bb 987 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 988}
989
b2950863 990static int btree_readpage(struct file *file, struct page *page)
5f39d397 991{
d1310b2e
CM		 992 struct extent_io_tree *tree;
993 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 994 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 995}
22b0ebda 996
70dec807 997static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 998{
98509cfc 999 if (PageWriteback(page) || PageDirty(page))
d397712b 1000 return 0;
0c4e538b 1001
f7a52a40 1002 return try_release_extent_buffer(page);
d98237b3
CM		 1003}
1004
d47992f8
LC		 1005static void btree_invalidatepage(struct page *page, unsigned int offset,
1006 unsigned int length)
d98237b3 1007{
d1310b2e
CM		 1008 struct extent_io_tree *tree;
1009 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1010 extent_invalidatepage(tree, page, offset);
1011 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1012 if (PagePrivate(page)) {
d397712b
CM		 1013 printk(KERN_WARNING "btrfs warning page private not zero "
1014 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1015 ClearPagePrivate(page);
1016 set_page_private(page, 0);
1017 page_cache_release(page);
1018 }
d98237b3
CM		 1019}
1020
0b32f4bb
JB		 1021static int btree_set_page_dirty(struct page *page)
1022{
bb146eb2 1023#ifdef DEBUG
0b32f4bb
JB		 1024 struct extent_buffer *eb;
1025
1026 BUG_ON(!PagePrivate(page));
1027 eb = (struct extent_buffer *)page->private;
1028 BUG_ON(!eb);
1029 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1030 BUG_ON(!atomic_read(&eb->refs));
1031 btrfs_assert_tree_locked(eb);
bb146eb2 1032#endif
0b32f4bb
JB		 1033 return __set_page_dirty_nobuffers(page);
1034}
1035
7f09410b 1036static const struct address_space_operations btree_aops = {
d98237b3 1037 .readpage = btree_readpage,
0da5468f 1038 .writepages = btree_writepages,
5f39d397
CM		 1039 .releasepage = btree_releasepage,
1040 .invalidatepage = btree_invalidatepage,
5a92bc88 1041#ifdef CONFIG_MIGRATION
784b4e29 1042 .migratepage = btree_migratepage,
5a92bc88 1043#endif
0b32f4bb 1044 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1045};
1046
ca7a79ad
CM		 1047int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1048 u64 parent_transid)
090d1875 1049{
5f39d397
CM		 1050 struct extent_buffer *buf = NULL;
1051 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1052 int ret = 0;
090d1875 1053
db94535d 1054 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1055 if (!buf)
090d1875 1056 return 0;
d1310b2e 1057 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1058 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1059 free_extent_buffer(buf);
de428b63 1060 return ret;
090d1875
CM		 1061}
1062
ab0fff03
AJ		 1063int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1064 int mirror_num, struct extent_buffer **eb)
1065{
1066 struct extent_buffer *buf = NULL;
1067 struct inode *btree_inode = root->fs_info->btree_inode;
1068 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1069 int ret;
1070
1071 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1072 if (!buf)
1073 return 0;
1074
1075 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1076
1077 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1078 btree_get_extent, mirror_num);
1079 if (ret) {
1080 free_extent_buffer(buf);
1081 return ret;
1082 }
1083
1084 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1085 free_extent_buffer(buf);
1086 return -EIO;
0b32f4bb 1087 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1088 *eb = buf;
1089 } else {
1090 free_extent_buffer(buf);
1091 }
1092 return 0;
1093}
1094
0999df54
CM		 1095struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1096 u64 bytenr, u32 blocksize)
1097{
1098 struct inode *btree_inode = root->fs_info->btree_inode;
1099 struct extent_buffer *eb;
452c75c3 1100 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree, bytenr);
0999df54
CM		 1101 return eb;
1102}
1103
1104struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1105 u64 bytenr, u32 blocksize)
1106{
1107 struct inode *btree_inode = root->fs_info->btree_inode;
1108 struct extent_buffer *eb;
1109
1110 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
727011e0 1111 bytenr, blocksize);
0999df54
CM		 1112 return eb;
1113}
1114
1115
e02119d5
CM		 1116int btrfs_write_tree_block(struct extent_buffer *buf)
1117{
727011e0 1118 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
8aa38c31 1119 buf->start + buf->len - 1);
e02119d5
CM		 1120}
1121
1122int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1123{
727011e0 1124 return filemap_fdatawait_range(buf->pages[0]->mapping,
8aa38c31 1125 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1126}
1127
0999df54 1128struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1129 u32 blocksize, u64 parent_transid)
0999df54
CM		 1130{
1131 struct extent_buffer *buf = NULL;
0999df54
CM		 1132 int ret;
1133
0999df54
CM		 1134 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1135 if (!buf)
1136 return NULL;
0999df54 1137
ca7a79ad 1138 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
0f0fe8f7
FDBM		 1139 if (ret) {
1140 free_extent_buffer(buf);
1141 return NULL;
1142 }
5f39d397 1143 return buf;
ce9adaa5 1144
eb60ceac
CM		 1145}
1146
d5c13f92
JM		 1147void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1148 struct extent_buffer *buf)
ed2ff2cb 1149{
e2d84521
MX		 1150 struct btrfs_fs_info *fs_info = root->fs_info;
1151
55c69072 1152 if (btrfs_header_generation(buf) ==
e2d84521 1153 fs_info->running_transaction->transid) {
b9447ef8 1154 btrfs_assert_tree_locked(buf);
b4ce94de 1155
b9473439 1156 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1157 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1158 -buf->len,
1159 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1160 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1161 btrfs_set_lock_blocking(buf);
1162 clear_extent_buffer_dirty(buf);
1163 }
925baedd 1164 }
5f39d397
CM		 1165}
1166
143bede5
JM		 1167static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1168 u32 stripesize, struct btrfs_root *root,
1169 struct btrfs_fs_info *fs_info,
1170 u64 objectid)
d97e63b6 1171{
cfaa7295 1172 root->node = NULL;
a28ec197 1173 root->commit_root = NULL;
db94535d
CM		 1174 root->sectorsize = sectorsize;
1175 root->nodesize = nodesize;
1176 root->leafsize = leafsize;
87ee04eb 1177 root->stripesize = stripesize;
123abc88 1178 root->ref_cows = 0;
0b86a832 1179 root->track_dirty = 0;
c71bf099 1180 root->in_radix = 0;
d68fc57b
YZ		 1181 root->orphan_item_inserted = 0;
1182 root->orphan_cleanup_state = 0;
0b86a832 1183
0f7d52f4
CM		 1184 root->objectid = objectid;
1185 root->last_trans = 0;
13a8a7c8 1186 root->highest_objectid = 0;
eb73c1b7 1187 root->nr_delalloc_inodes = 0;
199c2a9c 1188 root->nr_ordered_extents = 0;
58176a96 1189 root->name = NULL;
6bef4d31 1190 root->inode_tree = RB_ROOT;
16cdcec7 1191 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1192 root->block_rsv = NULL;
d68fc57b 1193 root->orphan_block_rsv = NULL;
0b86a832
CM		 1194
1195 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1196 INIT_LIST_HEAD(&root->root_list);
eb73c1b7
MX		 1197 INIT_LIST_HEAD(&root->delalloc_inodes);
1198 INIT_LIST_HEAD(&root->delalloc_root);
199c2a9c
MX		 1199 INIT_LIST_HEAD(&root->ordered_extents);
1200 INIT_LIST_HEAD(&root->ordered_root);
2ab28f32
JB		 1201 INIT_LIST_HEAD(&root->logged_list[0]);
1202 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1203 spin_lock_init(&root->orphan_lock);
5d4f98a2 1204 spin_lock_init(&root->inode_lock);
eb73c1b7 1205 spin_lock_init(&root->delalloc_lock);
199c2a9c 1206 spin_lock_init(&root->ordered_extent_lock);
f0486c68 1207 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1208 spin_lock_init(&root->log_extents_lock[0]);
1209 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1210 mutex_init(&root->objectid_mutex);
e02119d5 1211 mutex_init(&root->log_mutex);
7237f183
YZ		 1212 init_waitqueue_head(&root->log_writer_wait);
1213 init_waitqueue_head(&root->log_commit_wait[0]);
1214 init_waitqueue_head(&root->log_commit_wait[1]);
1215 atomic_set(&root->log_commit[0], 0);
1216 atomic_set(&root->log_commit[1], 0);
1217 atomic_set(&root->log_writers, 0);
2ecb7923 1218 atomic_set(&root->log_batch, 0);
8a35d95f 1219 atomic_set(&root->orphan_inodes, 0);
b0feb9d9 1220 atomic_set(&root->refs, 1);
7237f183 1221 root->log_transid = 0;
257c62e1 1222 root->last_log_commit = 0;
06ea65a3
JB		 1223 if (fs_info)
1224 extent_io_tree_init(&root->dirty_log_pages,
1225 fs_info->btree_inode->i_mapping);
017e5369 1226
3768f368
CM		 1227 memset(&root->root_key, 0, sizeof(root->root_key));
1228 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1229 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1230 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
06ea65a3
JB		 1231 if (fs_info)
1232 root->defrag_trans_start = fs_info->generation;
1233 else
1234 root->defrag_trans_start = 0;
58176a96 1235 init_completion(&root->kobj_unregister);
6702ed49 1236 root->defrag_running = 0;
4d775673 1237 root->root_key.objectid = objectid;
0ee5dc67 1238 root->anon_dev = 0;
8ea05e3a 1239
5f3ab90a 1240 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1241}
1242
f84a8bd6 1243static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1244{
1245 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1246 if (root)
1247 root->fs_info = fs_info;
1248 return root;
1249}
1250
06ea65a3
JB		 1251#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1252/* Should only be used by the testing infrastructure */
1253struct btrfs_root *btrfs_alloc_dummy_root(void)
1254{
1255 struct btrfs_root *root;
1256
1257 root = btrfs_alloc_root(NULL);
1258 if (!root)
1259 return ERR_PTR(-ENOMEM);
1260 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1261 root->dummy_root = 1;
1262
1263 return root;
1264}
1265#endif
1266
20897f5c
AJ		 1267struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1268 struct btrfs_fs_info *fs_info,
1269 u64 objectid)
1270{
1271 struct extent_buffer *leaf;
1272 struct btrfs_root *tree_root = fs_info->tree_root;
1273 struct btrfs_root *root;
1274 struct btrfs_key key;
1275 int ret = 0;
1276 u64 bytenr;
6463fe58 1277 uuid_le uuid;
20897f5c
AJ		 1278
1279 root = btrfs_alloc_root(fs_info);
1280 if (!root)
1281 return ERR_PTR(-ENOMEM);
1282
1283 __setup_root(tree_root->nodesize, tree_root->leafsize,
1284 tree_root->sectorsize, tree_root->stripesize,
1285 root, fs_info, objectid);
1286 root->root_key.objectid = objectid;
1287 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1288 root->root_key.offset = 0;
1289
1290 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1291 0, objectid, NULL, 0, 0, 0);
1292 if (IS_ERR(leaf)) {
1293 ret = PTR_ERR(leaf);
1dd05682 1294 leaf = NULL;
20897f5c
AJ		 1295 goto fail;
1296 }
1297
1298 bytenr = leaf->start;
1299 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1300 btrfs_set_header_bytenr(leaf, leaf->start);
1301 btrfs_set_header_generation(leaf, trans->transid);
1302 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1303 btrfs_set_header_owner(leaf, objectid);
1304 root->node = leaf;
1305
0a4e5586 1306 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
20897f5c
AJ		 1307 BTRFS_FSID_SIZE);
1308 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
b308bc2f 1309 btrfs_header_chunk_tree_uuid(leaf),
20897f5c
AJ		 1310 BTRFS_UUID_SIZE);
1311 btrfs_mark_buffer_dirty(leaf);
1312
1313 root->commit_root = btrfs_root_node(root);
1314 root->track_dirty = 1;
1315
1316
1317 root->root_item.flags = 0;
1318 root->root_item.byte_limit = 0;
1319 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1320 btrfs_set_root_generation(&root->root_item, trans->transid);
1321 btrfs_set_root_level(&root->root_item, 0);
1322 btrfs_set_root_refs(&root->root_item, 1);
1323 btrfs_set_root_used(&root->root_item, leaf->len);
1324 btrfs_set_root_last_snapshot(&root->root_item, 0);
1325 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1326 uuid_le_gen(&uuid);
1327 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1328 root->root_item.drop_level = 0;
1329
1330 key.objectid = objectid;
1331 key.type = BTRFS_ROOT_ITEM_KEY;
1332 key.offset = 0;
1333 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1334 if (ret)
1335 goto fail;
1336
1337 btrfs_tree_unlock(leaf);
1338
1dd05682
TI		 1339 return root;
1340
20897f5c 1341fail:
1dd05682
TI		 1342 if (leaf) {
1343 btrfs_tree_unlock(leaf);
1344 free_extent_buffer(leaf);
1345 }
1346 kfree(root);
20897f5c 1347
1dd05682 1348 return ERR_PTR(ret);
20897f5c
AJ		 1349}
1350
7237f183
YZ		 1351static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1352 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1353{
1354 struct btrfs_root *root;
1355 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1356 struct extent_buffer *leaf;
e02119d5 1357
6f07e42e 1358 root = btrfs_alloc_root(fs_info);
e02119d5 1359 if (!root)
7237f183 1360 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1361
1362 __setup_root(tree_root->nodesize, tree_root->leafsize,
1363 tree_root->sectorsize, tree_root->stripesize,
1364 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1365
1366 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1367 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1368 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1369 /*
1370 * log trees do not get reference counted because they go away
1371 * before a real commit is actually done. They do store pointers
1372 * to file data extents, and those reference counts still get
1373 * updated (along with back refs to the log tree).
1374 */
e02119d5
CM		 1375 root->ref_cows = 0;
1376
5d4f98a2 1377 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1378 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1379 0, 0, 0);
7237f183
YZ		 1380 if (IS_ERR(leaf)) {
1381 kfree(root);
1382 return ERR_CAST(leaf);
1383 }
e02119d5 1384
5d4f98a2
YZ		 1385 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1386 btrfs_set_header_bytenr(leaf, leaf->start);
1387 btrfs_set_header_generation(leaf, trans->transid);
1388 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1389 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1390 root->node = leaf;
e02119d5
CM		 1391
1392 write_extent_buffer(root->node, root->fs_info->fsid,
0a4e5586 1393 btrfs_header_fsid(), BTRFS_FSID_SIZE);
e02119d5
CM		 1394 btrfs_mark_buffer_dirty(root->node);
1395 btrfs_tree_unlock(root->node);
7237f183
YZ		 1396 return root;
1397}
1398
1399int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1400 struct btrfs_fs_info *fs_info)
1401{
1402 struct btrfs_root *log_root;
1403
1404 log_root = alloc_log_tree(trans, fs_info);
1405 if (IS_ERR(log_root))
1406 return PTR_ERR(log_root);
1407 WARN_ON(fs_info->log_root_tree);
1408 fs_info->log_root_tree = log_root;
1409 return 0;
1410}
1411
1412int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1413 struct btrfs_root *root)
1414{
1415 struct btrfs_root *log_root;
1416 struct btrfs_inode_item *inode_item;
1417
1418 log_root = alloc_log_tree(trans, root->fs_info);
1419 if (IS_ERR(log_root))
1420 return PTR_ERR(log_root);
1421
1422 log_root->last_trans = trans->transid;
1423 log_root->root_key.offset = root->root_key.objectid;
1424
1425 inode_item = &log_root->root_item.inode;
3cae210f
QW		 1426 btrfs_set_stack_inode_generation(inode_item, 1);
1427 btrfs_set_stack_inode_size(inode_item, 3);
1428 btrfs_set_stack_inode_nlink(inode_item, 1);
1429 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1430 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
7237f183 1431
5d4f98a2 1432 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1433
1434 WARN_ON(root->log_root);
1435 root->log_root = log_root;
1436 root->log_transid = 0;
257c62e1 1437 root->last_log_commit = 0;
e02119d5
CM		 1438 return 0;
1439}
1440
35a3621b
SB		 1441static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1442 struct btrfs_key *key)
e02119d5
CM		 1443{
1444 struct btrfs_root *root;
1445 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1446 struct btrfs_path *path;
84234f3a 1447 u64 generation;
db94535d 1448 u32 blocksize;
cb517eab 1449 int ret;
0f7d52f4 1450
cb517eab
MX		 1451 path = btrfs_alloc_path();
1452 if (!path)
0f7d52f4 1453 return ERR_PTR(-ENOMEM);
cb517eab
MX		 1454
1455 root = btrfs_alloc_root(fs_info);
1456 if (!root) {
1457 ret = -ENOMEM;
1458 goto alloc_fail;
0f7d52f4
CM		 1459 }
1460
db94535d 1461 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb 1462 tree_root->sectorsize, tree_root->stripesize,
cb517eab 1463 root, fs_info, key->objectid);
0f7d52f4 1464
cb517eab
MX		 1465 ret = btrfs_find_root(tree_root, key, path,
1466 &root->root_item, &root->root_key);
0f7d52f4 1467 if (ret) {
13a8a7c8
YZ		 1468 if (ret > 0)
1469 ret = -ENOENT;
cb517eab 1470 goto find_fail;
0f7d52f4 1471 }
13a8a7c8 1472
84234f3a 1473 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1474 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1475 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1476 blocksize, generation);
cb517eab
MX		 1477 if (!root->node) {
1478 ret = -ENOMEM;
1479 goto find_fail;
1480 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1481 ret = -EIO;
1482 goto read_fail;
416bc658 1483 }
5d4f98a2 1484 root->commit_root = btrfs_root_node(root);
13a8a7c8 1485out:
cb517eab
MX		 1486 btrfs_free_path(path);
1487 return root;
1488
1489read_fail:
1490 free_extent_buffer(root->node);
1491find_fail:
1492 kfree(root);
1493alloc_fail:
1494 root = ERR_PTR(ret);
1495 goto out;
1496}
1497
1498struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1499 struct btrfs_key *location)
1500{
1501 struct btrfs_root *root;
1502
1503 root = btrfs_read_tree_root(tree_root, location);
1504 if (IS_ERR(root))
1505 return root;
1506
1507 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1508 root->ref_cows = 1;
08fe4db1
LZ		 1509 btrfs_check_and_init_root_item(&root->root_item);
1510 }
13a8a7c8 1511
5eda7b5e
CM		 1512 return root;
1513}
1514
cb517eab
MX		 1515int btrfs_init_fs_root(struct btrfs_root *root)
1516{
1517 int ret;
1518
1519 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1520 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1521 GFP_NOFS);
1522 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1523 ret = -ENOMEM;
1524 goto fail;
1525 }
1526
1527 btrfs_init_free_ino_ctl(root);
1528 mutex_init(&root->fs_commit_mutex);
1529 spin_lock_init(&root->cache_lock);
1530 init_waitqueue_head(&root->cache_wait);
1531
1532 ret = get_anon_bdev(&root->anon_dev);
1533 if (ret)
1534 goto fail;
1535 return 0;
1536fail:
1537 kfree(root->free_ino_ctl);
1538 kfree(root->free_ino_pinned);
1539 return ret;
1540}
1541
171170c1
ST		 1542static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1543 u64 root_id)
cb517eab
MX		 1544{
1545 struct btrfs_root *root;
1546
1547 spin_lock(&fs_info->fs_roots_radix_lock);
1548 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1549 (unsigned long)root_id);
1550 spin_unlock(&fs_info->fs_roots_radix_lock);
1551 return root;
1552}
1553
1554int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1555 struct btrfs_root *root)
1556{
1557 int ret;
1558
1559 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1560 if (ret)
1561 return ret;
1562
1563 spin_lock(&fs_info->fs_roots_radix_lock);
1564 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1565 (unsigned long)root->root_key.objectid,
1566 root);
1567 if (ret == 0)
1568 root->in_radix = 1;
1569 spin_unlock(&fs_info->fs_roots_radix_lock);
1570 radix_tree_preload_end();
1571
1572 return ret;
1573}
1574
c00869f1
MX		 1575struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1576 struct btrfs_key *location,
1577 bool check_ref)
5eda7b5e
CM		 1578{
1579 struct btrfs_root *root;
1580 int ret;
1581
edbd8d4e
CM		 1582 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1583 return fs_info->tree_root;
1584 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1585 return fs_info->extent_root;
8f18cf13
CM		 1586 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1587 return fs_info->chunk_root;
1588 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1589 return fs_info->dev_root;
0403e47e
YZ		 1590 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1591 return fs_info->csum_root;
bcef60f2
AJ		 1592 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1593 return fs_info->quota_root ? fs_info->quota_root :
1594 ERR_PTR(-ENOENT);
f7a81ea4
SB		 1595 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1596 return fs_info->uuid_root ? fs_info->uuid_root :
1597 ERR_PTR(-ENOENT);
4df27c4d 1598again:
cb517eab 1599 root = btrfs_lookup_fs_root(fs_info, location->objectid);
48475471 1600 if (root) {
c00869f1 1601 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
48475471 1602 return ERR_PTR(-ENOENT);
5eda7b5e 1603 return root;
48475471 1604 }
5eda7b5e 1605
cb517eab 1606 root = btrfs_read_fs_root(fs_info->tree_root, location);
5eda7b5e
CM		 1607 if (IS_ERR(root))
1608 return root;
3394e160 1609
c00869f1 1610 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
cb517eab 1611 ret = -ENOENT;
581bb050 1612 goto fail;
35a30d7c 1613 }
581bb050 1614
cb517eab 1615 ret = btrfs_init_fs_root(root);
ac08aedf
CM		 1616 if (ret)
1617 goto fail;
3394e160 1618
d68fc57b
YZ		 1619 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1620 if (ret < 0)
1621 goto fail;
1622 if (ret == 0)
1623 root->orphan_item_inserted = 1;
1624
cb517eab 1625 ret = btrfs_insert_fs_root(fs_info, root);
0f7d52f4 1626 if (ret) {
4df27c4d
YZ		 1627 if (ret == -EEXIST) {
1628 free_fs_root(root);
1629 goto again;
1630 }
1631 goto fail;
0f7d52f4 1632 }
edbd8d4e 1633 return root;
4df27c4d
YZ		 1634fail:
1635 free_fs_root(root);
1636 return ERR_PTR(ret);
edbd8d4e
CM		 1637}
1638
04160088
CM		 1639static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1640{
1641 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1642 int ret = 0;
04160088
CM		 1643 struct btrfs_device *device;
1644 struct backing_dev_info *bdi;
b7967db7 1645
1f78160c
XG		 1646 rcu_read_lock();
1647 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1648 if (!device->bdev)
1649 continue;
04160088
CM		 1650 bdi = blk_get_backing_dev_info(device->bdev);
1651 if (bdi && bdi_congested(bdi, bdi_bits)) {
1652 ret = 1;
1653 break;
1654 }
1655 }
1f78160c 1656 rcu_read_unlock();
04160088
CM		 1657 return ret;
1658}
1659
ad081f14
JA		 1660/*
1661 * If this fails, caller must call bdi_destroy() to get rid of the
1662 * bdi again.
1663 */
04160088
CM		 1664static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1665{
ad081f14
JA		 1666 int err;
1667
1668 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1669 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1670 if (err)
1671 return err;
1672
4575c9cc 1673 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1674 bdi->congested_fn = btrfs_congested_fn;
1675 bdi->congested_data = info;
1676 return 0;
1677}
1678
8b712842
CM		 1679/*
1680 * called by the kthread helper functions to finally call the bio end_io
1681 * functions. This is where read checksum verification actually happens
1682 */
1683static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1684{
ce9adaa5 1685 struct bio *bio;
8b712842
CM		 1686 struct end_io_wq *end_io_wq;
1687 struct btrfs_fs_info *fs_info;
ce9adaa5 1688 int error;
ce9adaa5 1689
8b712842
CM		 1690 end_io_wq = container_of(work, struct end_io_wq, work);
1691 bio = end_io_wq->bio;
1692 fs_info = end_io_wq->info;
ce9adaa5 1693
8b712842
CM		 1694 error = end_io_wq->error;
1695 bio->bi_private = end_io_wq->private;
1696 bio->bi_end_io = end_io_wq->end_io;
1697 kfree(end_io_wq);
8b712842 1698 bio_endio(bio, error);
44b8bd7e
CM		 1699}
1700
a74a4b97
CM		 1701static int cleaner_kthread(void *arg)
1702{
1703 struct btrfs_root *root = arg;
d0278245 1704 int again;
a74a4b97
CM		 1705
1706 do {
d0278245 1707 again = 0;
a74a4b97 1708
d0278245 1709 /* Make the cleaner go to sleep early. */
babbf170 1710 if (btrfs_need_cleaner_sleep(root))
d0278245
MX		 1711 goto sleep;
1712
1713 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1714 goto sleep;
1715
dc7f370c
MX		 1716 /*
1717 * Avoid the problem that we change the status of the fs
1718 * during the above check and trylock.
1719 */
babbf170 1720 if (btrfs_need_cleaner_sleep(root)) {
dc7f370c
MX		 1721 mutex_unlock(&root->fs_info->cleaner_mutex);
1722 goto sleep;
76dda93c 1723 }
a74a4b97 1724
d0278245
MX		 1725 btrfs_run_delayed_iputs(root);
1726 again = btrfs_clean_one_deleted_snapshot(root);
1727 mutex_unlock(&root->fs_info->cleaner_mutex);
1728
1729 /*
05323cd1
MX		 1730 * The defragger has dealt with the R/O remount and umount,
1731 * needn't do anything special here.
d0278245
MX		 1732 */
1733 btrfs_run_defrag_inodes(root->fs_info);
1734sleep:
9d1a2a3a 1735 if (!try_to_freeze() && !again) {
a74a4b97 1736 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1737 if (!kthread_should_stop())
1738 schedule();
a74a4b97
CM		 1739 __set_current_state(TASK_RUNNING);
1740 }
1741 } while (!kthread_should_stop());
1742 return 0;
1743}
1744
1745static int transaction_kthread(void *arg)
1746{
1747 struct btrfs_root *root = arg;
1748 struct btrfs_trans_handle *trans;
1749 struct btrfs_transaction *cur;
8929ecfa 1750 u64 transid;
a74a4b97
CM		 1751 unsigned long now;
1752 unsigned long delay;
914b2007 1753 bool cannot_commit;
a74a4b97
CM		 1754
1755 do {
914b2007 1756 cannot_commit = false;
8b87dc17 1757 delay = HZ * root->fs_info->commit_interval;
a74a4b97
CM		 1758 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1759
a4abeea4 1760 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1761 cur = root->fs_info->running_transaction;
1762 if (!cur) {
a4abeea4 1763 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1764 goto sleep;
1765 }
31153d81 1766
a74a4b97 1767 now = get_seconds();
4a9d8bde 1768 if (cur->state < TRANS_STATE_BLOCKED &&
8b87dc17
DS		 1769 (now < cur->start_time ||
1770 now - cur->start_time < root->fs_info->commit_interval)) {
a4abeea4 1771 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1772 delay = HZ * 5;
1773 goto sleep;
1774 }
8929ecfa 1775 transid = cur->transid;
a4abeea4 1776 spin_unlock(&root->fs_info->trans_lock);
56bec294 1777
79787eaa 1778 /* If the file system is aborted, this will always fail. */
354aa0fb 1779 trans = btrfs_attach_transaction(root);
914b2007 1780 if (IS_ERR(trans)) {
354aa0fb
MX		 1781 if (PTR_ERR(trans) != -ENOENT)
1782 cannot_commit = true;
79787eaa 1783 goto sleep;
914b2007 1784 }
8929ecfa 1785 if (transid == trans->transid) {
79787eaa 1786 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1787 } else {
1788 btrfs_end_transaction(trans, root);
1789 }
a74a4b97
CM		 1790sleep:
1791 wake_up_process(root->fs_info->cleaner_kthread);
1792 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1793
4e121c06
JB		 1794 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1795 &root->fs_info->fs_state)))
1796 btrfs_cleanup_transaction(root);
a0acae0e 1797 if (!try_to_freeze()) {
a74a4b97 1798 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1799 if (!kthread_should_stop() &&
914b2007
JK		 1800 (!btrfs_transaction_blocked(root->fs_info) ||
1801 cannot_commit))
8929ecfa 1802 schedule_timeout(delay);
a74a4b97
CM		 1803 __set_current_state(TASK_RUNNING);
1804 }
1805 } while (!kthread_should_stop());
1806 return 0;
1807}
1808
af31f5e5
CM		 1809/*
1810 * this will find the highest generation in the array of
1811 * root backups. The index of the highest array is returned,
1812 * or -1 if we can't find anything.
1813 *
1814 * We check to make sure the array is valid by comparing the
1815 * generation of the latest root in the array with the generation
1816 * in the super block. If they don't match we pitch it.
1817 */
1818static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1819{
1820 u64 cur;
1821 int newest_index = -1;
1822 struct btrfs_root_backup *root_backup;
1823 int i;
1824
1825 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1826 root_backup = info->super_copy->super_roots + i;
1827 cur = btrfs_backup_tree_root_gen(root_backup);
1828 if (cur == newest_gen)
1829 newest_index = i;
1830 }
1831
1832 /* check to see if we actually wrapped around */
1833 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1834 root_backup = info->super_copy->super_roots;
1835 cur = btrfs_backup_tree_root_gen(root_backup);
1836 if (cur == newest_gen)
1837 newest_index = 0;
1838 }
1839 return newest_index;
1840}
1841
1842
1843/*
1844 * find the oldest backup so we know where to store new entries
1845 * in the backup array. This will set the backup_root_index
1846 * field in the fs_info struct
1847 */
1848static void find_oldest_super_backup(struct btrfs_fs_info *info,
1849 u64 newest_gen)
1850{
1851 int newest_index = -1;
1852
1853 newest_index = find_newest_super_backup(info, newest_gen);
1854 /* if there was garbage in there, just move along */
1855 if (newest_index == -1) {
1856 info->backup_root_index = 0;
1857 } else {
1858 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1859 }
1860}
1861
1862/*
1863 * copy all the root pointers into the super backup array.
1864 * this will bump the backup pointer by one when it is
1865 * done
1866 */
1867static void backup_super_roots(struct btrfs_fs_info *info)
1868{
1869 int next_backup;
1870 struct btrfs_root_backup *root_backup;
1871 int last_backup;
1872
1873 next_backup = info->backup_root_index;
1874 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1875 BTRFS_NUM_BACKUP_ROOTS;
1876
1877 /*
1878 * just overwrite the last backup if we're at the same generation
1879 * this happens only at umount
1880 */
1881 root_backup = info->super_for_commit->super_roots + last_backup;
1882 if (btrfs_backup_tree_root_gen(root_backup) ==
1883 btrfs_header_generation(info->tree_root->node))
1884 next_backup = last_backup;
1885
1886 root_backup = info->super_for_commit->super_roots + next_backup;
1887
1888 /*
1889 * make sure all of our padding and empty slots get zero filled
1890 * regardless of which ones we use today
1891 */
1892 memset(root_backup, 0, sizeof(*root_backup));
1893
1894 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1895
1896 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1897 btrfs_set_backup_tree_root_gen(root_backup,
1898 btrfs_header_generation(info->tree_root->node));
1899
1900 btrfs_set_backup_tree_root_level(root_backup,
1901 btrfs_header_level(info->tree_root->node));
1902
1903 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1904 btrfs_set_backup_chunk_root_gen(root_backup,
1905 btrfs_header_generation(info->chunk_root->node));
1906 btrfs_set_backup_chunk_root_level(root_backup,
1907 btrfs_header_level(info->chunk_root->node));
1908
1909 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1910 btrfs_set_backup_extent_root_gen(root_backup,
1911 btrfs_header_generation(info->extent_root->node));
1912 btrfs_set_backup_extent_root_level(root_backup,
1913 btrfs_header_level(info->extent_root->node));
1914
7c7e82a7
CM		 1915 /*
1916 * we might commit during log recovery, which happens before we set
1917 * the fs_root. Make sure it is valid before we fill it in.
1918 */
1919 if (info->fs_root && info->fs_root->node) {
1920 btrfs_set_backup_fs_root(root_backup,
1921 info->fs_root->node->start);
1922 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1923 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1924 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1925 btrfs_header_level(info->fs_root->node));
7c7e82a7 1926 }
af31f5e5
CM		 1927
1928 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1929 btrfs_set_backup_dev_root_gen(root_backup,
1930 btrfs_header_generation(info->dev_root->node));
1931 btrfs_set_backup_dev_root_level(root_backup,
1932 btrfs_header_level(info->dev_root->node));
1933
1934 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1935 btrfs_set_backup_csum_root_gen(root_backup,
1936 btrfs_header_generation(info->csum_root->node));
1937 btrfs_set_backup_csum_root_level(root_backup,
1938 btrfs_header_level(info->csum_root->node));
1939
1940 btrfs_set_backup_total_bytes(root_backup,
1941 btrfs_super_total_bytes(info->super_copy));
1942 btrfs_set_backup_bytes_used(root_backup,
1943 btrfs_super_bytes_used(info->super_copy));
1944 btrfs_set_backup_num_devices(root_backup,
1945 btrfs_super_num_devices(info->super_copy));
1946
1947 /*
1948 * if we don't copy this out to the super_copy, it won't get remembered
1949 * for the next commit
1950 */
1951 memcpy(&info->super_copy->super_roots,
1952 &info->super_for_commit->super_roots,
1953 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1954}
1955
1956/*
1957 * this copies info out of the root backup array and back into
1958 * the in-memory super block. It is meant to help iterate through
1959 * the array, so you send it the number of backups you've already
1960 * tried and the last backup index you used.
1961 *
1962 * this returns -1 when it has tried all the backups
1963 */
1964static noinline int next_root_backup(struct btrfs_fs_info *info,
1965 struct btrfs_super_block *super,
1966 int *num_backups_tried, int *backup_index)
1967{
1968 struct btrfs_root_backup *root_backup;
1969 int newest = *backup_index;
1970
1971 if (*num_backups_tried == 0) {
1972 u64 gen = btrfs_super_generation(super);
1973
1974 newest = find_newest_super_backup(info, gen);
1975 if (newest == -1)
1976 return -1;
1977
1978 *backup_index = newest;
1979 *num_backups_tried = 1;
1980 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1981 /* we've tried all the backups, all done */
1982 return -1;
1983 } else {
1984 /* jump to the next oldest backup */
1985 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1986 BTRFS_NUM_BACKUP_ROOTS;
1987 *backup_index = newest;
1988 *num_backups_tried += 1;
1989 }
1990 root_backup = super->super_roots + newest;
1991
1992 btrfs_set_super_generation(super,
1993 btrfs_backup_tree_root_gen(root_backup));
1994 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1995 btrfs_set_super_root_level(super,
1996 btrfs_backup_tree_root_level(root_backup));
1997 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1998
1999 /*
2000 * fixme: the total bytes and num_devices need to match or we should
2001 * need a fsck
2002 */
2003 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2004 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2005 return 0;
2006}
2007
7abadb64
LB		 2008/* helper to cleanup workers */
2009static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2010{
2011 btrfs_stop_workers(&fs_info->generic_worker);
2012 btrfs_stop_workers(&fs_info->fixup_workers);
2013 btrfs_stop_workers(&fs_info->delalloc_workers);
2014 btrfs_stop_workers(&fs_info->workers);
2015 btrfs_stop_workers(&fs_info->endio_workers);
2016 btrfs_stop_workers(&fs_info->endio_meta_workers);
2017 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2018 btrfs_stop_workers(&fs_info->rmw_workers);
2019 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2020 btrfs_stop_workers(&fs_info->endio_write_workers);
2021 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2022 btrfs_stop_workers(&fs_info->submit_workers);
2023 btrfs_stop_workers(&fs_info->delayed_workers);
2024 btrfs_stop_workers(&fs_info->caching_workers);
2025 btrfs_stop_workers(&fs_info->readahead_workers);
2026 btrfs_stop_workers(&fs_info->flush_workers);
2f232036 2027 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
7abadb64
LB		 2028}
2029
2e9f5954
R		 2030static void free_root_extent_buffers(struct btrfs_root *root)
2031{
2032 if (root) {
2033 free_extent_buffer(root->node);
2034 free_extent_buffer(root->commit_root);
2035 root->node = NULL;
2036 root->commit_root = NULL;
2037 }
2038}
2039
af31f5e5
CM		 2040/* helper to cleanup tree roots */
2041static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2042{
2e9f5954 2043 free_root_extent_buffers(info->tree_root);
655b09fe 2044
2e9f5954
R		 2045 free_root_extent_buffers(info->dev_root);
2046 free_root_extent_buffers(info->extent_root);
2047 free_root_extent_buffers(info->csum_root);
2048 free_root_extent_buffers(info->quota_root);
2049 free_root_extent_buffers(info->uuid_root);
2050 if (chunk_root)
2051 free_root_extent_buffers(info->chunk_root);
af31f5e5
CM		 2052}
2053
171f6537
JB		 2054static void del_fs_roots(struct btrfs_fs_info *fs_info)
2055{
2056 int ret;
2057 struct btrfs_root *gang[8];
2058 int i;
2059
2060 while (!list_empty(&fs_info->dead_roots)) {
2061 gang[0] = list_entry(fs_info->dead_roots.next,
2062 struct btrfs_root, root_list);
2063 list_del(&gang[0]->root_list);
2064
2065 if (gang[0]->in_radix) {
cb517eab 2066 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
171f6537
JB		 2067 } else {
2068 free_extent_buffer(gang[0]->node);
2069 free_extent_buffer(gang[0]->commit_root);
b0feb9d9 2070 btrfs_put_fs_root(gang[0]);
171f6537
JB		 2071 }
2072 }
2073
2074 while (1) {
2075 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2076 (void **)gang, 0,
2077 ARRAY_SIZE(gang));
2078 if (!ret)
2079 break;
2080 for (i = 0; i < ret; i++)
cb517eab 2081 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
171f6537
JB		 2082 }
2083}
af31f5e5 2084
ad2b2c80
AV		 2085int open_ctree(struct super_block *sb,
2086 struct btrfs_fs_devices *fs_devices,
2087 char *options)
2e635a27 2088{
db94535d
CM		 2089 u32 sectorsize;
2090 u32 nodesize;
2091 u32 leafsize;
2092 u32 blocksize;
87ee04eb 2093 u32 stripesize;
84234f3a 2094 u64 generation;
f2b636e8 2095 u64 features;
3de4586c 2096 struct btrfs_key location;
a061fc8d 2097 struct buffer_head *bh;
4d34b278 2098 struct btrfs_super_block *disk_super;
815745cf 2099 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2100 struct btrfs_root *tree_root;
4d34b278
ID		 2101 struct btrfs_root *extent_root;
2102 struct btrfs_root *csum_root;
2103 struct btrfs_root *chunk_root;
2104 struct btrfs_root *dev_root;
bcef60f2 2105 struct btrfs_root *quota_root;
f7a81ea4 2106 struct btrfs_root *uuid_root;
e02119d5 2107 struct btrfs_root *log_tree_root;
eb60ceac 2108 int ret;
e58ca020 2109 int err = -EINVAL;
af31f5e5
CM		 2110 int num_backups_tried = 0;
2111 int backup_index = 0;
70f80175
SB		 2112 bool create_uuid_tree;
2113 bool check_uuid_tree;
4543df7e 2114
f84a8bd6 2115 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e 2116 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
cb517eab 2117 if (!tree_root || !chunk_root) {
39279cc3
CM		 2118 err = -ENOMEM;
2119 goto fail;
2120 }
76dda93c
YZ		 2121
2122 ret = init_srcu_struct(&fs_info->subvol_srcu);
2123 if (ret) {
2124 err = ret;
2125 goto fail;
2126 }
2127
2128 ret = setup_bdi(fs_info, &fs_info->bdi);
2129 if (ret) {
2130 err = ret;
2131 goto fail_srcu;
2132 }
2133
e2d84521
MX		 2134 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2135 if (ret) {
2136 err = ret;
2137 goto fail_bdi;
2138 }
2139 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2140 (1 + ilog2(nr_cpu_ids));
2141
963d678b
MX		 2142 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2143 if (ret) {
2144 err = ret;
2145 goto fail_dirty_metadata_bytes;
2146 }
2147
76dda93c
YZ		 2148 fs_info->btree_inode = new_inode(sb);
2149 if (!fs_info->btree_inode) {
2150 err = -ENOMEM;
963d678b 2151 goto fail_delalloc_bytes;
76dda93c
YZ		 2152 }
2153
a6591715 2154 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2155
76dda93c 2156 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 2157 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2158 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2159 INIT_LIST_HEAD(&fs_info->delayed_iputs);
eb73c1b7 2160 INIT_LIST_HEAD(&fs_info->delalloc_roots);
11833d66 2161 INIT_LIST_HEAD(&fs_info->caching_block_groups);
eb73c1b7 2162 spin_lock_init(&fs_info->delalloc_root_lock);
a4abeea4 2163 spin_lock_init(&fs_info->trans_lock);
76dda93c 2164 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2165 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2166 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2167 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2168 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2169 spin_lock_init(&fs_info->super_lock);
f29021b2 2170 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2171 mutex_init(&fs_info->reloc_mutex);
de98ced9 2172 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2173
58176a96 2174 init_completion(&fs_info->kobj_unregister);
0b86a832 2175 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2176 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2177 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2178 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2179 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2180 BTRFS_BLOCK_RSV_GLOBAL);
2181 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2182 BTRFS_BLOCK_RSV_DELALLOC);
2183 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2184 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2185 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2186 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2187 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2188 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2189 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2190 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2191 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2192 atomic_set(&fs_info->defrag_running, 0);
fc36ed7e 2193 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2194 fs_info->sb = sb;
6f568d35 2195 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2196 fs_info->metadata_ratio = 0;
4cb5300b 2197 fs_info->defrag_inodes = RB_ROOT;
2bf64758 2198 fs_info->free_chunk_space = 0;
f29021b2 2199 fs_info->tree_mod_log = RB_ROOT;
8b87dc17 2200 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
c8b97818 2201
90519d66
AJ		 2202 /* readahead state */
2203 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2204 spin_lock_init(&fs_info->reada_lock);
c8b97818 2205
b34b086c
CM		 2206 fs_info->thread_pool_size = min_t(unsigned long,
2207 num_online_cpus() + 2, 8);
0afbaf8c 2208
199c2a9c
MX		 2209 INIT_LIST_HEAD(&fs_info->ordered_roots);
2210 spin_lock_init(&fs_info->ordered_root_lock);
16cdcec7
MX		 2211 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2212 GFP_NOFS);
2213 if (!fs_info->delayed_root) {
2214 err = -ENOMEM;
2215 goto fail_iput;
2216 }
2217 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2218
a2de733c
AJ		 2219 mutex_init(&fs_info->scrub_lock);
2220 atomic_set(&fs_info->scrubs_running, 0);
2221 atomic_set(&fs_info->scrub_pause_req, 0);
2222 atomic_set(&fs_info->scrubs_paused, 0);
2223 atomic_set(&fs_info->scrub_cancel_req, 0);
2224 init_waitqueue_head(&fs_info->scrub_pause_wait);
a2de733c 2225 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2226#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2227 fs_info->check_integrity_print_mask = 0;
2228#endif
a2de733c 2229
c9e9f97b
ID		 2230 spin_lock_init(&fs_info->balance_lock);
2231 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2232 atomic_set(&fs_info->balance_running, 0);
2233 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2234 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2235 fs_info->balance_ctl = NULL;
837d5b6e 2236 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2237
a061fc8d
CM		 2238 sb->s_blocksize = 4096;
2239 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2240 sb->s_bdi = &fs_info->bdi;
a061fc8d 2241
76dda93c 2242 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2243 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2244 /*
2245 * we set the i_size on the btree inode to the max possible int.
2246 * the real end of the address space is determined by all of
2247 * the devices in the system
2248 */
2249 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2250 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2251 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2252
5d4f98a2 2253 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2254 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2255 fs_info->btree_inode->i_mapping);
0b32f4bb 2256 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2257 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2258
2259 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2260
76dda93c
YZ		 2261 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2262 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2263 sizeof(struct btrfs_key));
72ac3c0d
JB		 2264 set_bit(BTRFS_INODE_DUMMY,
2265 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
778ba82b 2266 btrfs_insert_inode_hash(fs_info->btree_inode);
76dda93c 2267
0f9dd46c 2268 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2269 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2270 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2271
11833d66 2272 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2273 fs_info->btree_inode->i_mapping);
11833d66 2274 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2275 fs_info->btree_inode->i_mapping);
11833d66 2276 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2277 fs_info->do_barriers = 1;
e18e4809 2278
39279cc3 2279
5a3f23d5 2280 mutex_init(&fs_info->ordered_operations_mutex);
9ffba8cd 2281 mutex_init(&fs_info->ordered_extent_flush_mutex);
e02119d5 2282 mutex_init(&fs_info->tree_log_mutex);
925baedd 2283 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2284 mutex_init(&fs_info->transaction_kthread_mutex);
2285 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2286 mutex_init(&fs_info->volume_mutex);
276e680d 2287 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2288 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2289 init_rwsem(&fs_info->subvol_sem);
803b2f54 2290 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
e922e087
SB		 2291 fs_info->dev_replace.lock_owner = 0;
2292 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2293 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2294 mutex_init(&fs_info->dev_replace.lock_management_lock);
2295 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2296
416ac51d 2297 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2298 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2299 fs_info->qgroup_tree = RB_ROOT;
2300 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2301 fs_info->qgroup_seq = 1;
2302 fs_info->quota_enabled = 0;
2303 fs_info->pending_quota_state = 0;
1e8f9158 2304 fs_info->qgroup_ulist = NULL;
2f232036 2305 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2306
fa9c0d79
CM		 2307 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2308 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2309
e6dcd2dc 2310 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2311 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2312 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2313 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2314
53b381b3
DW		 2315 ret = btrfs_alloc_stripe_hash_table(fs_info);
2316 if (ret) {
83c8266a 2317 err = ret;
53b381b3
DW		 2318 goto fail_alloc;
2319 }
2320
0b86a832 2321 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2322 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2323
3c4bb26b 2324 invalidate_bdev(fs_devices->latest_bdev);
1104a885
DS		 2325
2326 /*
2327 * Read super block and check the signature bytes only
2328 */
a512bbf8 2329 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2330 if (!bh) {
2331 err = -EINVAL;
16cdcec7 2332 goto fail_alloc;
20b45077 2333 }
39279cc3 2334
1104a885
DS		 2335 /*
2336 * We want to check superblock checksum, the type is stored inside.
2337 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2338 */
2339 if (btrfs_check_super_csum(bh->b_data)) {
2340 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2341 err = -EINVAL;
2342 goto fail_alloc;
2343 }
2344
2345 /*
2346 * super_copy is zeroed at allocation time and we never touch the
2347 * following bytes up to INFO_SIZE, the checksum is calculated from
2348 * the whole block of INFO_SIZE
2349 */
6c41761f
DS		 2350 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2351 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2352 sizeof(*fs_info->super_for_commit));
a061fc8d 2353 brelse(bh);
5f39d397 2354
6c41761f 2355 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2356
1104a885
DS		 2357 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2358 if (ret) {
2359 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2360 err = -EINVAL;
2361 goto fail_alloc;
2362 }
2363
6c41761f 2364 disk_super = fs_info->super_copy;
0f7d52f4 2365 if (!btrfs_super_root(disk_super))
16cdcec7 2366 goto fail_alloc;
0f7d52f4 2367
acce952b 2368 /* check FS state, whether FS is broken. */
87533c47
MX		 2369 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2370 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2371
af31f5e5
CM		 2372 /*
2373 * run through our array of backup supers and setup
2374 * our ring pointer to the oldest one
2375 */
2376 generation = btrfs_super_generation(disk_super);
2377 find_oldest_super_backup(fs_info, generation);
2378
75e7cb7f
LB		 2379 /*
2380 * In the long term, we'll store the compression type in the super
2381 * block, and it'll be used for per file compression control.
2382 */
2383 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2384
2b82032c
YZ		 2385 ret = btrfs_parse_options(tree_root, options);
2386 if (ret) {
2387 err = ret;
16cdcec7 2388 goto fail_alloc;
2b82032c 2389 }
dfe25020 2390
f2b636e8
JB		 2391 features = btrfs_super_incompat_flags(disk_super) &
2392 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2393 if (features) {
2394 printk(KERN_ERR "BTRFS: couldn't mount because of "
2395 "unsupported optional features (%Lx).\n",
c1c9ff7c 2396 features);
f2b636e8 2397 err = -EINVAL;
16cdcec7 2398 goto fail_alloc;
f2b636e8
JB		 2399 }
2400
727011e0
CM		 2401 if (btrfs_super_leafsize(disk_super) !=
2402 btrfs_super_nodesize(disk_super)) {
2403 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2404 "blocksizes don't match. node %d leaf %d\n",
2405 btrfs_super_nodesize(disk_super),
2406 btrfs_super_leafsize(disk_super));
2407 err = -EINVAL;
2408 goto fail_alloc;
2409 }
2410 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2411 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2412 "blocksize (%d) was too large\n",
2413 btrfs_super_leafsize(disk_super));
2414 err = -EINVAL;
2415 goto fail_alloc;
2416 }
2417
5d4f98a2 2418 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2419 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2420 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2421 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2422
3173a18f
JB		 2423 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2424 printk(KERN_ERR "btrfs: has skinny extents\n");
2425
727011e0
CM		 2426 /*
2427 * flag our filesystem as having big metadata blocks if
2428 * they are bigger than the page size
2429 */
2430 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2431 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2432 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2433 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2434 }
2435
bc3f116f
CM		 2436 nodesize = btrfs_super_nodesize(disk_super);
2437 leafsize = btrfs_super_leafsize(disk_super);
2438 sectorsize = btrfs_super_sectorsize(disk_super);
2439 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2440 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2441 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2442
2443 /*
2444 * mixed block groups end up with duplicate but slightly offset
2445 * extent buffers for the same range. It leads to corruptions
2446 */
2447 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2448 (sectorsize != leafsize)) {
2449 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2450 "are not allowed for mixed block groups on %s\n",
2451 sb->s_id);
2452 goto fail_alloc;
2453 }
2454
ceda0864
MX		 2455 /*
2456 * Needn't use the lock because there is no other task which will
2457 * update the flag.
2458 */
a6fa6fae 2459 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2460
f2b636e8
JB		 2461 features = btrfs_super_compat_ro_flags(disk_super) &
2462 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2463 if (!(sb->s_flags & MS_RDONLY) && features) {
2464 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2465 "unsupported option features (%Lx).\n",
c1c9ff7c 2466 features);
f2b636e8 2467 err = -EINVAL;
16cdcec7 2468 goto fail_alloc;
f2b636e8 2469 }
61d92c32
CM		 2470
2471 btrfs_init_workers(&fs_info->generic_worker,
2472 "genwork", 1, NULL);
2473
5443be45 2474 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 2475 fs_info->thread_pool_size,
2476 &fs_info->generic_worker);
c8b97818 2477
771ed689 2478 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
45d5fd14 2479 fs_info->thread_pool_size, NULL);
771ed689 2480
8ccf6f19 2481 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
45d5fd14 2482 fs_info->thread_pool_size, NULL);
8ccf6f19 2483
5443be45 2484 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 2485 min_t(u64, fs_devices->num_devices,
45d5fd14 2486 fs_info->thread_pool_size), NULL);
61b49440 2487
bab39bf9 2488 btrfs_init_workers(&fs_info->caching_workers, "cache",
45d5fd14 2489 fs_info->thread_pool_size, NULL);
bab39bf9 2490
61b49440
CM		 2491 /* a higher idle thresh on the submit workers makes it much more
2492 * likely that bios will be send down in a sane order to the
2493 * devices
2494 */
2495 fs_info->submit_workers.idle_thresh = 64;
53863232 2496
771ed689 2497 fs_info->workers.idle_thresh = 16;
4a69a410 2498 fs_info->workers.ordered = 1;
61b49440 2499
771ed689
CM		 2500 fs_info->delalloc_workers.idle_thresh = 2;
2501 fs_info->delalloc_workers.ordered = 1;
2502
61d92c32
CM		 2503 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2504 &fs_info->generic_worker);
5443be45 2505 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2506 fs_info->thread_pool_size,
2507 &fs_info->generic_worker);
d20f7043 2508 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2509 fs_info->thread_pool_size,
2510 &fs_info->generic_worker);
cad321ad 2511 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2512 "endio-meta-write", fs_info->thread_pool_size,
2513 &fs_info->generic_worker);
53b381b3
DW		 2514 btrfs_init_workers(&fs_info->endio_raid56_workers,
2515 "endio-raid56", fs_info->thread_pool_size,
2516 &fs_info->generic_worker);
2517 btrfs_init_workers(&fs_info->rmw_workers,
2518 "rmw", fs_info->thread_pool_size,
2519 &fs_info->generic_worker);
5443be45 2520 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2521 fs_info->thread_pool_size,
2522 &fs_info->generic_worker);
0cb59c99
JB		 2523 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2524 1, &fs_info->generic_worker);
16cdcec7
MX		 2525 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2526 fs_info->thread_pool_size,
2527 &fs_info->generic_worker);
90519d66
AJ		 2528 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2529 fs_info->thread_pool_size,
2530 &fs_info->generic_worker);
2f232036
JS		 2531 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2532 &fs_info->generic_worker);
61b49440
CM		 2533
2534 /*
2535 * endios are largely parallel and should have a very
2536 * low idle thresh
2537 */
2538 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2539 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2540 fs_info->endio_raid56_workers.idle_thresh = 4;
2541 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2542
9042846b
CM		 2543 fs_info->endio_write_workers.idle_thresh = 2;
2544 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2545 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2546
0dc3b84a
JB		 2547 /*
2548 * btrfs_start_workers can really only fail because of ENOMEM so just
2549 * return -ENOMEM if any of these fail.
2550 */
2551 ret = btrfs_start_workers(&fs_info->workers);
2552 ret |= btrfs_start_workers(&fs_info->generic_worker);
2553 ret |= btrfs_start_workers(&fs_info->submit_workers);
2554 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2555 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2556 ret |= btrfs_start_workers(&fs_info->endio_workers);
2557 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2558 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2559 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2560 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2561 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2562 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2563 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2564 ret |= btrfs_start_workers(&fs_info->caching_workers);
2565 ret |= btrfs_start_workers(&fs_info->readahead_workers);
8ccf6f19 2566 ret |= btrfs_start_workers(&fs_info->flush_workers);
2f232036 2567 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
0dc3b84a 2568 if (ret) {
fed425c7 2569 err = -ENOMEM;
0dc3b84a
JB		 2570 goto fail_sb_buffer;
2571 }
4543df7e 2572
4575c9cc 2573 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2574 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2575 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2576
db94535d
CM		 2577 tree_root->nodesize = nodesize;
2578 tree_root->leafsize = leafsize;
2579 tree_root->sectorsize = sectorsize;
87ee04eb 2580 tree_root->stripesize = stripesize;
a061fc8d
CM		 2581
2582 sb->s_blocksize = sectorsize;
2583 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2584
3cae210f 2585 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
d397712b 2586 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2587 goto fail_sb_buffer;
2588 }
19c00ddc 2589
8d082fb7
LB		 2590 if (sectorsize != PAGE_SIZE) {
2591 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2592 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2593 goto fail_sb_buffer;
2594 }
2595
925baedd 2596 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2597 ret = btrfs_read_sys_array(tree_root);
925baedd 2598 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2599 if (ret) {
d397712b
CM		 2600 printk(KERN_WARNING "btrfs: failed to read the system "
2601 "array on %s\n", sb->s_id);
5d4f98a2 2602 goto fail_sb_buffer;
84eed90f 2603 }
0b86a832
CM		 2604
2605 blocksize = btrfs_level_size(tree_root,
2606 btrfs_super_chunk_root_level(disk_super));
84234f3a 2607 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2608
2609 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2610 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2611
2612 chunk_root->node = read_tree_block(chunk_root,
2613 btrfs_super_chunk_root(disk_super),
84234f3a 2614 blocksize, generation);
416bc658
JB		 2615 if (!chunk_root->node ||
2616 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
83121942
DW		 2617 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2618 sb->s_id);
af31f5e5 2619 goto fail_tree_roots;
83121942 2620 }
5d4f98a2
YZ		 2621 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2622 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2623
e17cade2 2624 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
b308bc2f 2625 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
e17cade2 2626
0b86a832 2627 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2628 if (ret) {
d397712b
CM		 2629 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2630 sb->s_id);
af31f5e5 2631 goto fail_tree_roots;
2b82032c 2632 }
0b86a832 2633
8dabb742
SB		 2634 /*
2635 * keep the device that is marked to be the target device for the
2636 * dev_replace procedure
2637 */
2638 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2639
a6b0d5c8
CM		 2640 if (!fs_devices->latest_bdev) {
2641 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2642 sb->s_id);
2643 goto fail_tree_roots;
2644 }
2645
af31f5e5 2646retry_root_backup:
db94535d
CM		 2647 blocksize = btrfs_level_size(tree_root,
2648 btrfs_super_root_level(disk_super));
84234f3a 2649 generation = btrfs_super_generation(disk_super);
0b86a832 2650
e20d96d6 2651 tree_root->node = read_tree_block(tree_root,
db94535d 2652 btrfs_super_root(disk_super),
84234f3a 2653 blocksize, generation);
af31f5e5
CM		 2654 if (!tree_root->node ||
2655 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
83121942
DW		 2656 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2657 sb->s_id);
af31f5e5
CM		 2658
2659 goto recovery_tree_root;
83121942 2660 }
af31f5e5 2661
5d4f98a2
YZ		 2662 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2663 tree_root->commit_root = btrfs_root_node(tree_root);
69e9c6c6 2664 btrfs_set_root_refs(&tree_root->root_item, 1);
db94535d 2665
cb517eab
MX		 2666 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2667 location.type = BTRFS_ROOT_ITEM_KEY;
2668 location.offset = 0;
2669
2670 extent_root = btrfs_read_tree_root(tree_root, &location);
2671 if (IS_ERR(extent_root)) {
2672 ret = PTR_ERR(extent_root);
af31f5e5 2673 goto recovery_tree_root;
cb517eab 2674 }
0b86a832 2675 extent_root->track_dirty = 1;
cb517eab 2676 fs_info->extent_root = extent_root;
0b86a832 2677
cb517eab
MX		 2678 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2679 dev_root = btrfs_read_tree_root(tree_root, &location);
2680 if (IS_ERR(dev_root)) {
2681 ret = PTR_ERR(dev_root);
af31f5e5 2682 goto recovery_tree_root;
cb517eab 2683 }
5d4f98a2 2684 dev_root->track_dirty = 1;
cb517eab
MX		 2685 fs_info->dev_root = dev_root;
2686 btrfs_init_devices_late(fs_info);
3768f368 2687
cb517eab
MX		 2688 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2689 csum_root = btrfs_read_tree_root(tree_root, &location);
2690 if (IS_ERR(csum_root)) {
2691 ret = PTR_ERR(csum_root);
af31f5e5 2692 goto recovery_tree_root;
cb517eab 2693 }
d20f7043 2694 csum_root->track_dirty = 1;
cb517eab 2695 fs_info->csum_root = csum_root;
d20f7043 2696
cb517eab
MX		 2697 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2698 quota_root = btrfs_read_tree_root(tree_root, &location);
2699 if (!IS_ERR(quota_root)) {
bcef60f2
AJ		 2700 quota_root->track_dirty = 1;
2701 fs_info->quota_enabled = 1;
2702 fs_info->pending_quota_state = 1;
cb517eab 2703 fs_info->quota_root = quota_root;
bcef60f2
AJ		 2704 }
2705
f7a81ea4
SB		 2706 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2707 uuid_root = btrfs_read_tree_root(tree_root, &location);
2708 if (IS_ERR(uuid_root)) {
2709 ret = PTR_ERR(uuid_root);
2710 if (ret != -ENOENT)
2711 goto recovery_tree_root;
2712 create_uuid_tree = true;
70f80175 2713 check_uuid_tree = false;
f7a81ea4
SB		 2714 } else {
2715 uuid_root->track_dirty = 1;
2716 fs_info->uuid_root = uuid_root;
70f80175
SB		 2717 create_uuid_tree = false;
2718 check_uuid_tree =
2719 generation != btrfs_super_uuid_tree_generation(disk_super);
f7a81ea4
SB		 2720 }
2721
8929ecfa
YZ		 2722 fs_info->generation = generation;
2723 fs_info->last_trans_committed = generation;
8929ecfa 2724
68310a5e
ID		 2725 ret = btrfs_recover_balance(fs_info);
2726 if (ret) {
2727 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2728 goto fail_block_groups;
2729 }
2730
733f4fbb
SB		 2731 ret = btrfs_init_dev_stats(fs_info);
2732 if (ret) {
2733 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2734 ret);
2735 goto fail_block_groups;
2736 }
2737
8dabb742
SB		 2738 ret = btrfs_init_dev_replace(fs_info);
2739 if (ret) {
2740 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2741 goto fail_block_groups;
2742 }
2743
2744 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2745
c59021f8 2746 ret = btrfs_init_space_info(fs_info);
2747 if (ret) {
2748 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2749 goto fail_block_groups;
2750 }
2751
1b1d1f66
JB		 2752 ret = btrfs_read_block_groups(extent_root);
2753 if (ret) {
2754 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2755 goto fail_block_groups;
2756 }
5af3e8cc
SB		 2757 fs_info->num_tolerated_disk_barrier_failures =
2758 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2759 if (fs_info->fs_devices->missing_devices >
2760 fs_info->num_tolerated_disk_barrier_failures &&
2761 !(sb->s_flags & MS_RDONLY)) {
2762 printk(KERN_WARNING
2763 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2764 goto fail_block_groups;
2765 }
9078a3e1 2766
a74a4b97
CM		 2767 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2768 "btrfs-cleaner");
57506d50 2769 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2770 goto fail_block_groups;
a74a4b97
CM		 2771
2772 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2773 tree_root,
2774 "btrfs-transaction");
57506d50 2775 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2776 goto fail_cleaner;
a74a4b97 2777
c289811c
CM		 2778 if (!btrfs_test_opt(tree_root, SSD) &&
2779 !btrfs_test_opt(tree_root, NOSSD) &&
2780 !fs_info->fs_devices->rotating) {
2781 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2782 "mode\n");
2783 btrfs_set_opt(fs_info->mount_opt, SSD);
2784 }
2785
21adbd5c
SB		 2786#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2787 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2788 ret = btrfsic_mount(tree_root, fs_devices,
2789 btrfs_test_opt(tree_root,
2790 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2791 1 : 0,
2792 fs_info->check_integrity_print_mask);
2793 if (ret)
2794 printk(KERN_WARNING "btrfs: failed to initialize"
2795 " integrity check module %s\n", sb->s_id);
2796 }
2797#endif
bcef60f2
AJ		 2798 ret = btrfs_read_qgroup_config(fs_info);
2799 if (ret)
2800 goto fail_trans_kthread;
21adbd5c 2801
acce952b 2802 /* do not make disk changes in broken FS */
68ce9682 2803 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2804 u64 bytenr = btrfs_super_log_root(disk_super);
2805
7c2ca468 2806 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2807 printk(KERN_WARNING "Btrfs log replay required "
2808 "on RO media\n");
7c2ca468 2809 err = -EIO;
bcef60f2 2810 goto fail_qgroup;
7c2ca468 2811 }
e02119d5
CM		 2812 blocksize =
2813 btrfs_level_size(tree_root,
2814 btrfs_super_log_root_level(disk_super));
d18a2c44 2815
6f07e42e 2816 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2817 if (!log_tree_root) {
2818 err = -ENOMEM;
bcef60f2 2819 goto fail_qgroup;
676e4c86 2820 }
e02119d5
CM		 2821
2822 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2823 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2824
2825 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2826 blocksize,
2827 generation + 1);
416bc658
JB		 2828 if (!log_tree_root->node ||
2829 !extent_buffer_uptodate(log_tree_root->node)) {
2830 printk(KERN_ERR "btrfs: failed to read log tree\n");
2831 free_extent_buffer(log_tree_root->node);
2832 kfree(log_tree_root);
2833 goto fail_trans_kthread;
2834 }
79787eaa 2835 /* returns with log_tree_root freed on success */
e02119d5 2836 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2837 if (ret) {
2838 btrfs_error(tree_root->fs_info, ret,
2839 "Failed to recover log tree");
2840 free_extent_buffer(log_tree_root->node);
2841 kfree(log_tree_root);
2842 goto fail_trans_kthread;
2843 }
e556ce2c
YZ		 2844
2845 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2846 ret = btrfs_commit_super(tree_root);
2847 if (ret)
2848 goto fail_trans_kthread;
e556ce2c 2849 }
e02119d5 2850 }
1a40e23b 2851
76dda93c 2852 ret = btrfs_find_orphan_roots(tree_root);
79787eaa
JM		 2853 if (ret)
2854 goto fail_trans_kthread;
76dda93c 2855
7c2ca468 2856 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2857 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2
ID		 2858 if (ret)
2859 goto fail_trans_kthread;
d68fc57b 2860
5d4f98a2 2861 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2862 if (ret < 0) {
2863 printk(KERN_WARNING
2864 "btrfs: failed to recover relocation\n");
2865 err = -EINVAL;
bcef60f2 2866 goto fail_qgroup;
d7ce5843 2867 }
7c2ca468 2868 }
1a40e23b 2869
3de4586c
CM		 2870 location.objectid = BTRFS_FS_TREE_OBJECTID;
2871 location.type = BTRFS_ROOT_ITEM_KEY;
cb517eab 2872 location.offset = 0;
3de4586c 2873
3de4586c 2874 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3140c9a3
DC		 2875 if (IS_ERR(fs_info->fs_root)) {
2876 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2877 goto fail_qgroup;
3140c9a3 2878 }
c289811c 2879
2b6ba629
ID		 2880 if (sb->s_flags & MS_RDONLY)
2881 return 0;
59641015 2882
2b6ba629
ID		 2883 down_read(&fs_info->cleanup_work_sem);
2884 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2885 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2886 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2887 close_ctree(tree_root);
2888 return ret;
2889 }
2890 up_read(&fs_info->cleanup_work_sem);
59641015 2891
2b6ba629
ID		 2892 ret = btrfs_resume_balance_async(fs_info);
2893 if (ret) {
2894 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2895 close_ctree(tree_root);
2896 return ret;
e3acc2a6
JB		 2897 }
2898
8dabb742
SB		 2899 ret = btrfs_resume_dev_replace_async(fs_info);
2900 if (ret) {
2901 pr_warn("btrfs: failed to resume dev_replace\n");
2902 close_ctree(tree_root);
2903 return ret;
2904 }
2905
b382a324
JS		 2906 btrfs_qgroup_rescan_resume(fs_info);
2907
f7a81ea4
SB		 2908 if (create_uuid_tree) {
2909 pr_info("btrfs: creating UUID tree\n");
2910 ret = btrfs_create_uuid_tree(fs_info);
2911 if (ret) {
2912 pr_warn("btrfs: failed to create the UUID tree %d\n",
2913 ret);
2914 close_ctree(tree_root);
2915 return ret;
2916 }
f420ee1e
SB		 2917 } else if (check_uuid_tree ||
2918 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
70f80175
SB		 2919 pr_info("btrfs: checking UUID tree\n");
2920 ret = btrfs_check_uuid_tree(fs_info);
2921 if (ret) {
2922 pr_warn("btrfs: failed to check the UUID tree %d\n",
2923 ret);
2924 close_ctree(tree_root);
2925 return ret;
2926 }
2927 } else {
2928 fs_info->update_uuid_tree_gen = 1;
f7a81ea4
SB		 2929 }
2930
ad2b2c80 2931 return 0;
39279cc3 2932
bcef60f2
AJ		 2933fail_qgroup:
2934 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2935fail_trans_kthread:
2936 kthread_stop(fs_info->transaction_kthread);
54067ae9 2937 btrfs_cleanup_transaction(fs_info->tree_root);
7b5ff90e 2938 del_fs_roots(fs_info);
3f157a2f 2939fail_cleaner:
a74a4b97 2940 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2941
2942 /*
2943 * make sure we're done with the btree inode before we stop our
2944 * kthreads
2945 */
2946 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2947
1b1d1f66 2948fail_block_groups:
54067ae9 2949 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2950 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2951
2952fail_tree_roots:
2953 free_root_pointers(fs_info, 1);
2b8195bb 2954 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2955
39279cc3 2956fail_sb_buffer:
7abadb64 2957 btrfs_stop_all_workers(fs_info);
16cdcec7 2958fail_alloc:
4543df7e 2959fail_iput:
586e46e2
ID		 2960 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2961
4543df7e 2962 iput(fs_info->btree_inode);
963d678b
MX		 2963fail_delalloc_bytes:
2964 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 2965fail_dirty_metadata_bytes:
2966 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 2967fail_bdi:
7e662854 2968 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2969fail_srcu:
2970 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2971fail:
53b381b3 2972 btrfs_free_stripe_hash_table(fs_info);
586e46e2 2973 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 2974 return err;
af31f5e5
CM		 2975
2976recovery_tree_root:
af31f5e5
CM		 2977 if (!btrfs_test_opt(tree_root, RECOVERY))
2978 goto fail_tree_roots;
2979
2980 free_root_pointers(fs_info, 0);
2981
2982 /* don't use the log in recovery mode, it won't be valid */
2983 btrfs_set_super_log_root(disk_super, 0);
2984
2985 /* we can't trust the free space cache either */
2986 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2987
2988 ret = next_root_backup(fs_info, fs_info->super_copy,
2989 &num_backups_tried, &backup_index);
2990 if (ret == -1)
2991 goto fail_block_groups;
2992 goto retry_root_backup;
eb60ceac
CM		 2993}
2994
f2984462
CM		 2995static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2996{
f2984462
CM		 2997 if (uptodate) {
2998 set_buffer_uptodate(bh);
2999 } else {
442a4f63
SB		 3000 struct btrfs_device *device = (struct btrfs_device *)
3001 bh->b_private;
3002
606686ee
JB		 3003 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3004 "I/O error on %s\n",
3005 rcu_str_deref(device->name));
1259ab75
CM		 3006 /* note, we dont' set_buffer_write_io_error because we have
3007 * our own ways of dealing with the IO errors
3008 */
f2984462 3009 clear_buffer_uptodate(bh);
442a4f63 3010 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 3011 }
3012 unlock_buffer(bh);
3013 put_bh(bh);
3014}
3015
a512bbf8
YZ		 3016struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3017{
3018 struct buffer_head *bh;
3019 struct buffer_head *latest = NULL;
3020 struct btrfs_super_block *super;
3021 int i;
3022 u64 transid = 0;
3023 u64 bytenr;
3024
3025 /* we would like to check all the supers, but that would make
3026 * a btrfs mount succeed after a mkfs from a different FS.
3027 * So, we need to add a special mount option to scan for
3028 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3029 */
3030 for (i = 0; i < 1; i++) {
3031 bytenr = btrfs_sb_offset(i);
8068a47e
AJ		 3032 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3033 i_size_read(bdev->bd_inode))
a512bbf8 3034 break;
8068a47e
AJ		 3035 bh = __bread(bdev, bytenr / 4096,
3036 BTRFS_SUPER_INFO_SIZE);
a512bbf8
YZ		 3037 if (!bh)
3038 continue;
3039
3040 super = (struct btrfs_super_block *)bh->b_data;
3041 if (btrfs_super_bytenr(super) != bytenr ||
3cae210f 3042 btrfs_super_magic(super) != BTRFS_MAGIC) {
a512bbf8
YZ		 3043 brelse(bh);
3044 continue;
3045 }
3046
3047 if (!latest || btrfs_super_generation(super) > transid) {
3048 brelse(latest);
3049 latest = bh;
3050 transid = btrfs_super_generation(super);
3051 } else {
3052 brelse(bh);
3053 }
3054 }
3055 return latest;
3056}
3057
4eedeb75
HH		 3058/*
3059 * this should be called twice, once with wait == 0 and
3060 * once with wait == 1. When wait == 0 is done, all the buffer heads
3061 * we write are pinned.
3062 *
3063 * They are released when wait == 1 is done.
3064 * max_mirrors must be the same for both runs, and it indicates how
3065 * many supers on this one device should be written.
3066 *
3067 * max_mirrors == 0 means to write them all.
3068 */
a512bbf8
YZ		 3069static int write_dev_supers(struct btrfs_device *device,
3070 struct btrfs_super_block *sb,
3071 int do_barriers, int wait, int max_mirrors)
3072{
3073 struct buffer_head *bh;
3074 int i;
3075 int ret;
3076 int errors = 0;
3077 u32 crc;
3078 u64 bytenr;
a512bbf8
YZ		 3079
3080 if (max_mirrors == 0)
3081 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3082
a512bbf8
YZ		 3083 for (i = 0; i < max_mirrors; i++) {
3084 bytenr = btrfs_sb_offset(i);
3085 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3086 break;
3087
3088 if (wait) {
3089 bh = __find_get_block(device->bdev, bytenr / 4096,
3090 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3091 if (!bh) {
3092 errors++;
3093 continue;
3094 }
a512bbf8 3095 wait_on_buffer(bh);
4eedeb75
HH		 3096 if (!buffer_uptodate(bh))
3097 errors++;
3098
3099 /* drop our reference */
3100 brelse(bh);
3101
3102 /* drop the reference from the wait == 0 run */
3103 brelse(bh);
3104 continue;
a512bbf8
YZ		 3105 } else {
3106 btrfs_set_super_bytenr(sb, bytenr);
3107
3108 crc = ~(u32)0;
b0496686 3109 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 3110 BTRFS_CSUM_SIZE, crc,
3111 BTRFS_SUPER_INFO_SIZE -
3112 BTRFS_CSUM_SIZE);
3113 btrfs_csum_final(crc, sb->csum);
3114
4eedeb75
HH		 3115 /*
3116 * one reference for us, and we leave it for the
3117 * caller
3118 */
a512bbf8
YZ		 3119 bh = __getblk(device->bdev, bytenr / 4096,
3120 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3121 if (!bh) {
3122 printk(KERN_ERR "btrfs: couldn't get super "
3123 "buffer head for bytenr %Lu\n", bytenr);
3124 errors++;
3125 continue;
3126 }
3127
a512bbf8
YZ		 3128 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3129
4eedeb75 3130 /* one reference for submit_bh */
a512bbf8 3131 get_bh(bh);
4eedeb75
HH		 3132
3133 set_buffer_uptodate(bh);
a512bbf8
YZ		 3134 lock_buffer(bh);
3135 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3136 bh->b_private = device;
a512bbf8
YZ		 3137 }
3138
387125fc
CM		 3139 /*
3140 * we fua the first super. The others we allow
3141 * to go down lazy.
3142 */
21adbd5c 3143 ret = btrfsic_submit_bh(WRITE_FUA, bh);
4eedeb75 3144 if (ret)
a512bbf8 3145 errors++;
a512bbf8
YZ		 3146 }
3147 return errors < i ? 0 : -1;
3148}
3149
387125fc
CM		 3150/*
3151 * endio for the write_dev_flush, this will wake anyone waiting
3152 * for the barrier when it is done
3153 */
3154static void btrfs_end_empty_barrier(struct bio *bio, int err)
3155{
3156 if (err) {
3157 if (err == -EOPNOTSUPP)
3158 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3159 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3160 }
3161 if (bio->bi_private)
3162 complete(bio->bi_private);
3163 bio_put(bio);
3164}
3165
3166/*
3167 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3168 * sent down. With wait == 1, it waits for the previous flush.
3169 *
3170 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3171 * capable
3172 */
3173static int write_dev_flush(struct btrfs_device *device, int wait)
3174{
3175 struct bio *bio;
3176 int ret = 0;
3177
3178 if (device->nobarriers)
3179 return 0;
3180
3181 if (wait) {
3182 bio = device->flush_bio;
3183 if (!bio)
3184 return 0;
3185
3186 wait_for_completion(&device->flush_wait);
3187
3188 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
606686ee
JB		 3189 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3190 rcu_str_deref(device->name));
387125fc 3191 device->nobarriers = 1;
5af3e8cc 3192 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3193 ret = -EIO;
5af3e8cc
SB		 3194 btrfs_dev_stat_inc_and_print(device,
3195 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3196 }
3197
3198 /* drop the reference from the wait == 0 run */
3199 bio_put(bio);
3200 device->flush_bio = NULL;
3201
3202 return ret;
3203 }
3204
3205 /*
3206 * one reference for us, and we leave it for the
3207 * caller
3208 */
9c017abc 3209 device->flush_bio = NULL;
9be3395b 3210 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
387125fc
CM		 3211 if (!bio)
3212 return -ENOMEM;
3213
3214 bio->bi_end_io = btrfs_end_empty_barrier;
3215 bio->bi_bdev = device->bdev;
3216 init_completion(&device->flush_wait);
3217 bio->bi_private = &device->flush_wait;
3218 device->flush_bio = bio;
3219
3220 bio_get(bio);
21adbd5c 3221 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3222
3223 return 0;
3224}
3225
3226/*
3227 * send an empty flush down to each device in parallel,
3228 * then wait for them
3229 */
3230static int barrier_all_devices(struct btrfs_fs_info *info)
3231{
3232 struct list_head *head;
3233 struct btrfs_device *dev;
5af3e8cc
SB		 3234 int errors_send = 0;
3235 int errors_wait = 0;
387125fc
CM		 3236 int ret;
3237
3238 /* send down all the barriers */
3239 head = &info->fs_devices->devices;
3240 list_for_each_entry_rcu(dev, head, dev_list) {
3241 if (!dev->bdev) {
5af3e8cc 3242 errors_send++;
387125fc
CM		 3243 continue;
3244 }
3245 if (!dev->in_fs_metadata || !dev->writeable)
3246 continue;
3247
3248 ret = write_dev_flush(dev, 0);
3249 if (ret)
5af3e8cc 3250 errors_send++;
387125fc
CM		 3251 }
3252
3253 /* wait for all the barriers */
3254 list_for_each_entry_rcu(dev, head, dev_list) {
3255 if (!dev->bdev) {
5af3e8cc 3256 errors_wait++;
387125fc
CM		 3257 continue;
3258 }
3259 if (!dev->in_fs_metadata || !dev->writeable)
3260 continue;
3261
3262 ret = write_dev_flush(dev, 1);
3263 if (ret)
5af3e8cc 3264 errors_wait++;
387125fc 3265 }
5af3e8cc
SB		 3266 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3267 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3268 return -EIO;
3269 return 0;
3270}
3271
5af3e8cc
SB		 3272int btrfs_calc_num_tolerated_disk_barrier_failures(
3273 struct btrfs_fs_info *fs_info)
3274{
3275 struct btrfs_ioctl_space_info space;
3276 struct btrfs_space_info *sinfo;
3277 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3278 BTRFS_BLOCK_GROUP_SYSTEM,
3279 BTRFS_BLOCK_GROUP_METADATA,
3280 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3281 int num_types = 4;
3282 int i;
3283 int c;
3284 int num_tolerated_disk_barrier_failures =
3285 (int)fs_info->fs_devices->num_devices;
3286
3287 for (i = 0; i < num_types; i++) {
3288 struct btrfs_space_info *tmp;
3289
3290 sinfo = NULL;
3291 rcu_read_lock();
3292 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3293 if (tmp->flags == types[i]) {
3294 sinfo = tmp;
3295 break;
3296 }
3297 }
3298 rcu_read_unlock();
3299
3300 if (!sinfo)
3301 continue;
3302
3303 down_read(&sinfo->groups_sem);
3304 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3305 if (!list_empty(&sinfo->block_groups[c])) {
3306 u64 flags;
3307
3308 btrfs_get_block_group_info(
3309 &sinfo->block_groups[c], &space);
3310 if (space.total_bytes == 0 ||
3311 space.used_bytes == 0)
3312 continue;
3313 flags = space.flags;
3314 /*
3315 * return
3316 * 0: if dup, single or RAID0 is configured for
3317 * any of metadata, system or data, else
3318 * 1: if RAID5 is configured, or if RAID1 or
3319 * RAID10 is configured and only two mirrors
3320 * are used, else
3321 * 2: if RAID6 is configured, else
3322 * num_mirrors - 1: if RAID1 or RAID10 is
3323 * configured and more than
3324 * 2 mirrors are used.
3325 */
3326 if (num_tolerated_disk_barrier_failures > 0 &&
3327 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3328 BTRFS_BLOCK_GROUP_RAID0)) ||
3329 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3330 == 0)))
3331 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3332 else if (num_tolerated_disk_barrier_failures > 1) {
3333 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3334 BTRFS_BLOCK_GROUP_RAID5 |
3335 BTRFS_BLOCK_GROUP_RAID10)) {
3336 num_tolerated_disk_barrier_failures = 1;
3337 } else if (flags &
15b0a89d 3338 BTRFS_BLOCK_GROUP_RAID6) {
53b381b3
DW		 3339 num_tolerated_disk_barrier_failures = 2;
3340 }
3341 }
5af3e8cc
SB		 3342 }
3343 }
3344 up_read(&sinfo->groups_sem);
3345 }
3346
3347 return num_tolerated_disk_barrier_failures;
3348}
3349
48a3b636 3350static int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3351{
e5e9a520 3352 struct list_head *head;
f2984462 3353 struct btrfs_device *dev;
a061fc8d 3354 struct btrfs_super_block *sb;
f2984462 3355 struct btrfs_dev_item *dev_item;
f2984462
CM		 3356 int ret;
3357 int do_barriers;
a236aed1
CM		 3358 int max_errors;
3359 int total_errors = 0;
a061fc8d 3360 u64 flags;
f2984462
CM		 3361
3362 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3363 backup_super_roots(root->fs_info);
f2984462 3364
6c41761f 3365 sb = root->fs_info->super_for_commit;
a061fc8d 3366 dev_item = &sb->dev_item;
e5e9a520 3367
174ba509 3368 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3369 head = &root->fs_info->fs_devices->devices;
d7306801 3370 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
387125fc 3371
5af3e8cc
SB		 3372 if (do_barriers) {
3373 ret = barrier_all_devices(root->fs_info);
3374 if (ret) {
3375 mutex_unlock(
3376 &root->fs_info->fs_devices->device_list_mutex);
3377 btrfs_error(root->fs_info, ret,
3378 "errors while submitting device barriers.");
3379 return ret;
3380 }
3381 }
387125fc 3382
1f78160c 3383 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3384 if (!dev->bdev) {
3385 total_errors++;
3386 continue;
3387 }
2b82032c 3388 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3389 continue;
3390
2b82032c 3391 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3392 btrfs_set_stack_device_type(dev_item, dev->type);
3393 btrfs_set_stack_device_id(dev_item, dev->devid);
3394 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3395 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3396 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3397 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3398 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3399 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3400 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3401
a061fc8d
CM		 3402 flags = btrfs_super_flags(sb);
3403 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3404
a512bbf8 3405 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3406 if (ret)
3407 total_errors++;
f2984462 3408 }
a236aed1 3409 if (total_errors > max_errors) {
d397712b
CM		 3410 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3411 total_errors);
a724b436 3412 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
79787eaa 3413
9d565ba4
SB		 3414 /* FUA is masked off if unsupported and can't be the reason */
3415 btrfs_error(root->fs_info, -EIO,
3416 "%d errors while writing supers", total_errors);
3417 return -EIO;
a236aed1 3418 }
f2984462 3419
a512bbf8 3420 total_errors = 0;
1f78160c 3421 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3422 if (!dev->bdev)
3423 continue;
2b82032c 3424 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3425 continue;
3426
a512bbf8
YZ		 3427 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3428 if (ret)
3429 total_errors++;
f2984462 3430 }
174ba509 3431 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3432 if (total_errors > max_errors) {
79787eaa
JM		 3433 btrfs_error(root->fs_info, -EIO,
3434 "%d errors while writing supers", total_errors);
3435 return -EIO;
a236aed1 3436 }
f2984462
CM		 3437 return 0;
3438}
3439
a512bbf8
YZ		 3440int write_ctree_super(struct btrfs_trans_handle *trans,
3441 struct btrfs_root *root, int max_mirrors)
eb60ceac 3442{
f570e757 3443 return write_all_supers(root, max_mirrors);
cfaa7295
CM		 3444}
3445
cb517eab
MX		 3446/* Drop a fs root from the radix tree and free it. */
3447void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3448 struct btrfs_root *root)
2619ba1f 3449{
4df27c4d 3450 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3451 radix_tree_delete(&fs_info->fs_roots_radix,
3452 (unsigned long)root->root_key.objectid);
4df27c4d 3453 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3454
3455 if (btrfs_root_refs(&root->root_item) == 0)
3456 synchronize_srcu(&fs_info->subvol_srcu);
3457
d7634482 3458 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3321719e
LB		 3459 btrfs_free_log(NULL, root);
3460 btrfs_free_log_root_tree(NULL, fs_info);
3461 }
3462
581bb050
LZ		 3463 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3464 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3465 free_fs_root(root);
4df27c4d
YZ		 3466}
3467
3468static void free_fs_root(struct btrfs_root *root)
3469{
82d5902d 3470 iput(root->cache_inode);
4df27c4d 3471 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1cb048f5
FDBM		 3472 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3473 root->orphan_block_rsv = NULL;
0ee5dc67
AV		 3474 if (root->anon_dev)
3475 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 3476 free_extent_buffer(root->node);
3477 free_extent_buffer(root->commit_root);
581bb050
LZ		 3478 kfree(root->free_ino_ctl);
3479 kfree(root->free_ino_pinned);
d397712b 3480 kfree(root->name);
b0feb9d9 3481 btrfs_put_fs_root(root);
2619ba1f
CM		 3482}
3483
cb517eab
MX		 3484void btrfs_free_fs_root(struct btrfs_root *root)
3485{
3486 free_fs_root(root);
2619ba1f
CM		 3487}
3488
c146afad 3489int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3490{
c146afad
YZ		 3491 u64 root_objectid = 0;
3492 struct btrfs_root *gang[8];
3493 int i;
3768f368 3494 int ret;
e089f05c 3495
c146afad
YZ		 3496 while (1) {
3497 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3498 (void **)gang, root_objectid,
3499 ARRAY_SIZE(gang));
3500 if (!ret)
3501 break;
5d4f98a2
YZ		 3502
3503 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 3504 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 3505 int err;
3506
c146afad 3507 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3508 err = btrfs_orphan_cleanup(gang[i]);
3509 if (err)
3510 return err;
c146afad
YZ		 3511 }
3512 root_objectid++;
3513 }
3514 return 0;
3515}
a2135011 3516
c146afad
YZ		 3517int btrfs_commit_super(struct btrfs_root *root)
3518{
3519 struct btrfs_trans_handle *trans;
a74a4b97 3520
c146afad 3521 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3522 btrfs_run_delayed_iputs(root);
c146afad 3523 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3524 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3525
3526 /* wait until ongoing cleanup work done */
3527 down_write(&root->fs_info->cleanup_work_sem);
3528 up_write(&root->fs_info->cleanup_work_sem);
3529
7a7eaa40 3530 trans = btrfs_join_transaction(root);
3612b495
TI		 3531 if (IS_ERR(trans))
3532 return PTR_ERR(trans);
d52c1bcc 3533 return btrfs_commit_transaction(trans, root);
c146afad
YZ		 3534}
3535
3536int close_ctree(struct btrfs_root *root)
3537{
3538 struct btrfs_fs_info *fs_info = root->fs_info;
3539 int ret;
3540
3541 fs_info->closing = 1;
3542 smp_mb();
3543
803b2f54
SB		 3544 /* wait for the uuid_scan task to finish */
3545 down(&fs_info->uuid_tree_rescan_sem);
3546 /* avoid complains from lockdep et al., set sem back to initial state */
3547 up(&fs_info->uuid_tree_rescan_sem);
3548
837d5b6e 3549 /* pause restriper - we want to resume on mount */
aa1b8cd4 3550 btrfs_pause_balance(fs_info);
837d5b6e 3551
8dabb742
SB		 3552 btrfs_dev_replace_suspend_for_unmount(fs_info);
3553
aa1b8cd4 3554 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3555
3556 /* wait for any defraggers to finish */
3557 wait_event(fs_info->transaction_wait,
3558 (atomic_read(&fs_info->defrag_running) == 0));
3559
3560 /* clear out the rbtree of defraggable inodes */
26176e7c 3561 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3562
c146afad 3563 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3564 ret = btrfs_commit_super(root);
3565 if (ret)
3566 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3567 }
3568
87533c47 3569 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3570 btrfs_error_commit_super(root);
0f7d52f4 3571
300e4f8a
JB		 3572 btrfs_put_block_group_cache(fs_info);
3573
e3029d9f
AV		 3574 kthread_stop(fs_info->transaction_kthread);
3575 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3576
f25784b3
YZ		 3577 fs_info->closing = 2;
3578 smp_mb();
3579
bcef60f2
AJ		 3580 btrfs_free_qgroup_config(root->fs_info);
3581
963d678b
MX		 3582 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3583 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3584 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3585 }
bcc63abb 3586
c146afad 3587 del_fs_roots(fs_info);
d10c5f31 3588
2b1360da
JB		 3589 btrfs_free_block_groups(fs_info);
3590
96192499
JB		 3591 btrfs_stop_all_workers(fs_info);
3592
13e6c37b 3593 free_root_pointers(fs_info, 1);
9ad6b7bc 3594
13e6c37b 3595 iput(fs_info->btree_inode);
d6bfde87 3596
21adbd5c
SB		 3597#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3598 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3599 btrfsic_unmount(root, fs_info->fs_devices);
3600#endif
3601
dfe25020 3602 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3603 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3604
e2d84521 3605 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3606 percpu_counter_destroy(&fs_info->delalloc_bytes);
04160088 3607 bdi_destroy(&fs_info->bdi);
76dda93c 3608 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3609
53b381b3
DW		 3610 btrfs_free_stripe_hash_table(fs_info);
3611
1cb048f5
FDBM		 3612 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3613 root->orphan_block_rsv = NULL;
3614
eb60ceac
CM		 3615 return 0;
3616}
3617
b9fab919
CM		 3618int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3619 int atomic)
5f39d397 3620{
1259ab75 3621 int ret;
727011e0 3622 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3623
0b32f4bb 3624 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3625 if (!ret)
3626 return ret;
3627
3628 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3629 parent_transid, atomic);
3630 if (ret == -EAGAIN)
3631 return ret;
1259ab75 3632 return !ret;
5f39d397
CM		 3633}
3634
3635int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3636{
0b32f4bb 3637 return set_extent_buffer_uptodate(buf);
5f39d397 3638}
6702ed49 3639
5f39d397
CM		 3640void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3641{
06ea65a3 3642 struct btrfs_root *root;
5f39d397 3643 u64 transid = btrfs_header_generation(buf);
b9473439 3644 int was_dirty;
b4ce94de 3645
06ea65a3
JB		 3646#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3647 /*
3648 * This is a fast path so only do this check if we have sanity tests
3649 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3650 * outside of the sanity tests.
3651 */
3652 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3653 return;
3654#endif
3655 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
b9447ef8 3656 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3657 if (transid != root->fs_info->generation)
3658 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3659 "found %llu running %llu\n",
c1c9ff7c 3660 buf->start, transid, root->fs_info->generation);
0b32f4bb 3661 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3662 if (!was_dirty)
3663 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3664 buf->len,
3665 root->fs_info->dirty_metadata_batch);
eb60ceac
CM		 3666}
3667
b53d3f5d
LB		 3668static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3669 int flush_delayed)
16cdcec7
MX		 3670{
3671 /*
3672 * looks as though older kernels can get into trouble with
3673 * this code, they end up stuck in balance_dirty_pages forever
3674 */
e2d84521 3675 int ret;
16cdcec7
MX		 3676
3677 if (current->flags & PF_MEMALLOC)
3678 return;
3679
b53d3f5d
LB		 3680 if (flush_delayed)
3681 btrfs_balance_delayed_items(root);
16cdcec7 3682
e2d84521
MX		 3683 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3684 BTRFS_DIRTY_METADATA_THRESH);
3685 if (ret > 0) {
d0e1d66b
NJ		 3686 balance_dirty_pages_ratelimited(
3687 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3688 }
3689 return;
3690}
3691
b53d3f5d 3692void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3693{
b53d3f5d
LB		 3694 __btrfs_btree_balance_dirty(root, 1);
3695}
585ad2c3 3696
b53d3f5d
LB		 3697void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3698{
3699 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3700}
6b80053d 3701
ca7a79ad 3702int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3703{
727011e0 3704 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3705 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3706}
0da5468f 3707
fcd1f065 3708static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3709 int read_only)
3710{
1104a885
DS		 3711 /*
3712 * Placeholder for checks
3713 */
fcd1f065 3714 return 0;
acce952b 3715}
3716
48a3b636 3717static void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3718{
acce952b 3719 mutex_lock(&root->fs_info->cleaner_mutex);
3720 btrfs_run_delayed_iputs(root);
3721 mutex_unlock(&root->fs_info->cleaner_mutex);
3722
3723 down_write(&root->fs_info->cleanup_work_sem);
3724 up_write(&root->fs_info->cleanup_work_sem);
3725
3726 /* cleanup FS via transaction */
3727 btrfs_cleanup_transaction(root);
acce952b 3728}
3729
569e0f35
JB		 3730static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3731 struct btrfs_root *root)
acce952b 3732{
3733 struct btrfs_inode *btrfs_inode;
3734 struct list_head splice;
3735
3736 INIT_LIST_HEAD(&splice);
3737
3738 mutex_lock(&root->fs_info->ordered_operations_mutex);
199c2a9c 3739 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3740
569e0f35 3741 list_splice_init(&t->ordered_operations, &splice);
acce952b 3742 while (!list_empty(&splice)) {
3743 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3744 ordered_operations);
3745
3746 list_del_init(&btrfs_inode->ordered_operations);
199c2a9c 3747 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3748
3749 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3750
199c2a9c 3751 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3752 }
3753
199c2a9c 3754 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3755 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3756}
3757
143bede5 3758static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3759{
acce952b 3760 struct btrfs_ordered_extent *ordered;
acce952b 3761
199c2a9c 3762 spin_lock(&root->ordered_extent_lock);
779880ef
JB		 3763 /*
3764 * This will just short circuit the ordered completion stuff which will
3765 * make sure the ordered extent gets properly cleaned up.
3766 */
199c2a9c 3767 list_for_each_entry(ordered, &root->ordered_extents,
779880ef
JB		 3768 root_extent_list)
3769 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
199c2a9c
MX		 3770 spin_unlock(&root->ordered_extent_lock);
3771}
3772
3773static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3774{
3775 struct btrfs_root *root;
3776 struct list_head splice;
3777
3778 INIT_LIST_HEAD(&splice);
3779
3780 spin_lock(&fs_info->ordered_root_lock);
3781 list_splice_init(&fs_info->ordered_roots, &splice);
3782 while (!list_empty(&splice)) {
3783 root = list_first_entry(&splice, struct btrfs_root,
3784 ordered_root);
1de2cfde
JB		 3785 list_move_tail(&root->ordered_root,
3786 &fs_info->ordered_roots);
199c2a9c
MX		 3787
3788 btrfs_destroy_ordered_extents(root);
3789
3790 cond_resched_lock(&fs_info->ordered_root_lock);
3791 }
3792 spin_unlock(&fs_info->ordered_root_lock);
acce952b 3793}
3794
35a3621b
SB		 3795static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3796 struct btrfs_root *root)
acce952b 3797{
3798 struct rb_node *node;
3799 struct btrfs_delayed_ref_root *delayed_refs;
3800 struct btrfs_delayed_ref_node *ref;
3801 int ret = 0;
3802
3803 delayed_refs = &trans->delayed_refs;
3804
3805 spin_lock(&delayed_refs->lock);
3806 if (delayed_refs->num_entries == 0) {
cfece4db 3807 spin_unlock(&delayed_refs->lock);
acce952b 3808 printk(KERN_INFO "delayed_refs has NO entry\n");
3809 return ret;
3810 }
3811
b939d1ab 3812 while ((node = rb_first(&delayed_refs->root)) != NULL) {
eb12db69 3813 struct btrfs_delayed_ref_head *head = NULL;
e78417d1 3814 bool pin_bytes = false;
acce952b 3815
eb12db69 3816 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
acce952b 3817 atomic_set(&ref->refs, 1);
3818 if (btrfs_delayed_ref_is_head(ref)) {
acce952b 3819
3820 head = btrfs_delayed_node_to_head(ref);
b939d1ab
JB		 3821 if (!mutex_trylock(&head->mutex)) {
3822 atomic_inc(&ref->refs);
3823 spin_unlock(&delayed_refs->lock);
3824
3825 /* Need to wait for the delayed ref to run */
3826 mutex_lock(&head->mutex);
3827 mutex_unlock(&head->mutex);
3828 btrfs_put_delayed_ref(ref);
3829
e18fca73 3830 spin_lock(&delayed_refs->lock);
b939d1ab
JB		 3831 continue;
3832 }
3833
54067ae9 3834 if (head->must_insert_reserved)
e78417d1 3835 pin_bytes = true;
78a6184a 3836 btrfs_free_delayed_extent_op(head->extent_op);
acce952b 3837 delayed_refs->num_heads--;
3838 if (list_empty(&head->cluster))
3839 delayed_refs->num_heads_ready--;
3840 list_del_init(&head->cluster);
acce952b 3841 }
eb12db69 3842
b939d1ab
JB		 3843 ref->in_tree = 0;
3844 rb_erase(&ref->rb_node, &delayed_refs->root);
3845 delayed_refs->num_entries--;
acce952b 3846 spin_unlock(&delayed_refs->lock);
e78417d1
JB		 3847 if (head) {
3848 if (pin_bytes)
3849 btrfs_pin_extent(root, ref->bytenr,
3850 ref->num_bytes, 1);
3851 mutex_unlock(&head->mutex);
3852 }
acce952b 3853 btrfs_put_delayed_ref(ref);
3854
3855 cond_resched();
3856 spin_lock(&delayed_refs->lock);
3857 }
3858
3859 spin_unlock(&delayed_refs->lock);
3860
3861 return ret;
3862}
3863
143bede5 3864static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3865{
3866 struct btrfs_inode *btrfs_inode;
3867 struct list_head splice;
3868
3869 INIT_LIST_HEAD(&splice);
3870
eb73c1b7
MX		 3871 spin_lock(&root->delalloc_lock);
3872 list_splice_init(&root->delalloc_inodes, &splice);
acce952b 3873
3874 while (!list_empty(&splice)) {
eb73c1b7
MX		 3875 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3876 delalloc_inodes);
acce952b 3877
3878 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3879 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3880 &btrfs_inode->runtime_flags);
eb73c1b7 3881 spin_unlock(&root->delalloc_lock);
acce952b 3882
3883 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3884
eb73c1b7 3885 spin_lock(&root->delalloc_lock);
acce952b 3886 }
3887
eb73c1b7
MX		 3888 spin_unlock(&root->delalloc_lock);
3889}
3890
3891static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3892{
3893 struct btrfs_root *root;
3894 struct list_head splice;
3895
3896 INIT_LIST_HEAD(&splice);
3897
3898 spin_lock(&fs_info->delalloc_root_lock);
3899 list_splice_init(&fs_info->delalloc_roots, &splice);
3900 while (!list_empty(&splice)) {
3901 root = list_first_entry(&splice, struct btrfs_root,
3902 delalloc_root);
3903 list_del_init(&root->delalloc_root);
3904 root = btrfs_grab_fs_root(root);
3905 BUG_ON(!root);
3906 spin_unlock(&fs_info->delalloc_root_lock);
3907
3908 btrfs_destroy_delalloc_inodes(root);
3909 btrfs_put_fs_root(root);
3910
3911 spin_lock(&fs_info->delalloc_root_lock);
3912 }
3913 spin_unlock(&fs_info->delalloc_root_lock);
acce952b 3914}
3915
3916static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3917 struct extent_io_tree *dirty_pages,
3918 int mark)
3919{
3920 int ret;
acce952b 3921 struct extent_buffer *eb;
3922 u64 start = 0;
3923 u64 end;
acce952b 3924
3925 while (1) {
3926 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3927 mark, NULL);
acce952b 3928 if (ret)
3929 break;
3930
3931 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3932 while (start <= end) {
fd8b2b61
JB		 3933 eb = btrfs_find_tree_block(root, start,
3934 root->leafsize);
69a85bd8 3935 start += root->leafsize;
fd8b2b61 3936 if (!eb)
acce952b 3937 continue;
fd8b2b61 3938 wait_on_extent_buffer_writeback(eb);
acce952b 3939
fd8b2b61
JB		 3940 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3941 &eb->bflags))
3942 clear_extent_buffer_dirty(eb);
3943 free_extent_buffer_stale(eb);
acce952b 3944 }
3945 }
3946
3947 return ret;
3948}
3949
3950static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3951 struct extent_io_tree *pinned_extents)
3952{
3953 struct extent_io_tree *unpin;
3954 u64 start;
3955 u64 end;
3956 int ret;
ed0eaa14 3957 bool loop = true;
acce952b 3958
3959 unpin = pinned_extents;
ed0eaa14 3960again:
acce952b 3961 while (1) {
3962 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 3963 EXTENT_DIRTY, NULL);
acce952b 3964 if (ret)
3965 break;
3966
3967 /* opt_discard */
5378e607
LD		 3968 if (btrfs_test_opt(root, DISCARD))
3969 ret = btrfs_error_discard_extent(root, start,
3970 end + 1 - start,
3971 NULL);
acce952b 3972
3973 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3974 btrfs_error_unpin_extent_range(root, start, end);
3975 cond_resched();
3976 }
3977
ed0eaa14
LB		 3978 if (loop) {
3979 if (unpin == &root->fs_info->freed_extents[0])
3980 unpin = &root->fs_info->freed_extents[1];
3981 else
3982 unpin = &root->fs_info->freed_extents[0];
3983 loop = false;
3984 goto again;
3985 }
3986
acce952b 3987 return 0;
3988}
3989
49b25e05
JM		 3990void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3991 struct btrfs_root *root)
3992{
724e2315
JB		 3993 btrfs_destroy_ordered_operations(cur_trans, root);
3994
49b25e05 3995 btrfs_destroy_delayed_refs(cur_trans, root);
49b25e05 3996
4a9d8bde 3997 cur_trans->state = TRANS_STATE_COMMIT_START;
d7096fc3 3998 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 3999
4a9d8bde 4000 cur_trans->state = TRANS_STATE_UNBLOCKED;
d7096fc3 4001 wake_up(&root->fs_info->transaction_wait);
49b25e05 4002
67cde344
MX		 4003 btrfs_destroy_delayed_inodes(root);
4004 btrfs_assert_delayed_root_empty(root);
49b25e05 4005
49b25e05
JM		 4006 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4007 EXTENT_DIRTY);
6e841e32
LB		 4008 btrfs_destroy_pinned_extent(root,
4009 root->fs_info->pinned_extents);
49b25e05 4010
4a9d8bde
MX		 4011 cur_trans->state =TRANS_STATE_COMPLETED;
4012 wake_up(&cur_trans->commit_wait);
4013
49b25e05
JM		 4014 /*
4015 memset(cur_trans, 0, sizeof(*cur_trans));
4016 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4017 */
4018}
4019
48a3b636 4020static int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 4021{
4022 struct btrfs_transaction *t;
acce952b 4023
acce952b 4024 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4025
a4abeea4 4026 spin_lock(&root->fs_info->trans_lock);
724e2315
JB		 4027 while (!list_empty(&root->fs_info->trans_list)) {
4028 t = list_first_entry(&root->fs_info->trans_list,
4029 struct btrfs_transaction, list);
4030 if (t->state >= TRANS_STATE_COMMIT_START) {
4031 atomic_inc(&t->use_count);
4032 spin_unlock(&root->fs_info->trans_lock);
4033 btrfs_wait_for_commit(root, t->transid);
4034 btrfs_put_transaction(t);
4035 spin_lock(&root->fs_info->trans_lock);
4036 continue;
4037 }
4038 if (t == root->fs_info->running_transaction) {
4039 t->state = TRANS_STATE_COMMIT_DOING;
4040 spin_unlock(&root->fs_info->trans_lock);
4041 /*
4042 * We wait for 0 num_writers since we don't hold a trans
4043 * handle open currently for this transaction.
4044 */
4045 wait_event(t->writer_wait,
4046 atomic_read(&t->num_writers) == 0);
4047 } else {
4048 spin_unlock(&root->fs_info->trans_lock);
4049 }
4050 btrfs_cleanup_one_transaction(t, root);
4a9d8bde 4051
724e2315
JB		 4052 spin_lock(&root->fs_info->trans_lock);
4053 if (t == root->fs_info->running_transaction)
4054 root->fs_info->running_transaction = NULL;
acce952b 4055 list_del_init(&t->list);
724e2315 4056 spin_unlock(&root->fs_info->trans_lock);
acce952b 4057
724e2315
JB		 4058 btrfs_put_transaction(t);
4059 trace_btrfs_transaction_commit(root);
4060 spin_lock(&root->fs_info->trans_lock);
4061 }
4062 spin_unlock(&root->fs_info->trans_lock);
4063 btrfs_destroy_all_ordered_extents(root->fs_info);
4064 btrfs_destroy_delayed_inodes(root);
4065 btrfs_assert_delayed_root_empty(root);
4066 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4067 btrfs_destroy_all_delalloc_inodes(root->fs_info);
acce952b 4068 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 4069
4070 return 0;
4071}
4072
d1310b2e 4073static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 4074 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 4075 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 4076 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 4077 /* note we're sharing with inode.c for the merge bio hook */
4078 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 4079};
Linux 6.x block layer and io_uring tree(s)