projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: do not release metadata for space cache inodes
[linux-2.6-block.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM		 1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
163e783e 29#include <linux/crc32c.h>
5a0e3ad6 30#include <linux/slab.h>
784b4e29 31#include <linux/migrate.h>
7a36ddec 32#include <linux/ratelimit.h>
6463fe58 33#include <linux/uuid.h>
803b2f54 34#include <linux/semaphore.h>
7e75bf3f 35#include <asm/unaligned.h>
4b4e25f2 36#include "compat.h"
eb60ceac
CM		 37#include "ctree.h"
38#include "disk-io.h"
e089f05c 39#include "transaction.h"
0f7d52f4 40#include "btrfs_inode.h"
0b86a832 41#include "volumes.h"
db94535d 42#include "print-tree.h"
8b712842 43#include "async-thread.h"
925baedd 44#include "locking.h"
e02119d5 45#include "tree-log.h"
fa9c0d79 46#include "free-space-cache.h"
581bb050 47#include "inode-map.h"
21adbd5c 48#include "check-integrity.h"
606686ee 49#include "rcu-string.h"
8dabb742 50#include "dev-replace.h"
53b381b3 51#include "raid56.h"
eb60ceac 52
de0022b9
JB		 53#ifdef CONFIG_X86
54#include <asm/cpufeature.h>
55#endif
56
d1310b2e 57static struct extent_io_ops btree_extent_io_ops;
8b712842 58static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 59static void free_fs_root(struct btrfs_root *root);
fcd1f065 60static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 61 int read_only);
569e0f35
JB		 62static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
63 struct btrfs_root *root);
143bede5 64static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
acce952b 65static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
66 struct btrfs_root *root);
aec8030a 67static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
143bede5 68static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
acce952b 69static int btrfs_destroy_marked_extents(struct btrfs_root *root,
70 struct extent_io_tree *dirty_pages,
71 int mark);
72static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
73 struct extent_io_tree *pinned_extents);
48a3b636
ES		 74static int btrfs_cleanup_transaction(struct btrfs_root *root);
75static void btrfs_error_commit_super(struct btrfs_root *root);
ce9adaa5 76
d352ac68
CM		 77/*
78 * end_io_wq structs are used to do processing in task context when an IO is
79 * complete. This is used during reads to verify checksums, and it is used
80 * by writes to insert metadata for new file extents after IO is complete.
81 */
ce9adaa5
CM		 82struct end_io_wq {
83 struct bio *bio;
84 bio_end_io_t *end_io;
85 void *private;
86 struct btrfs_fs_info *info;
87 int error;
22c59948 88 int metadata;
ce9adaa5 89 struct list_head list;
8b712842 90 struct btrfs_work work;
ce9adaa5 91};
0da5468f 92
d352ac68
CM		 93/*
94 * async submit bios are used to offload expensive checksumming
95 * onto the worker threads. They checksum file and metadata bios
96 * just before they are sent down the IO stack.
97 */
44b8bd7e
CM		 98struct async_submit_bio {
99 struct inode *inode;
100 struct bio *bio;
101 struct list_head list;
4a69a410
CM		 102 extent_submit_bio_hook_t *submit_bio_start;
103 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 104 int rw;
105 int mirror_num;
c8b97818 106 unsigned long bio_flags;
eaf25d93
CM		 107 /*
108 * bio_offset is optional, can be used if the pages in the bio
109 * can't tell us where in the file the bio should go
110 */
111 u64 bio_offset;
8b712842 112 struct btrfs_work work;
79787eaa 113 int error;
44b8bd7e
CM		 114};
115
85d4e461
CM		 116/*
117 * Lockdep class keys for extent_buffer->lock's in this root. For a given
118 * eb, the lockdep key is determined by the btrfs_root it belongs to and
119 * the level the eb occupies in the tree.
120 *
121 * Different roots are used for different purposes and may nest inside each
122 * other and they require separate keysets. As lockdep keys should be
123 * static, assign keysets according to the purpose of the root as indicated
124 * by btrfs_root->objectid. This ensures that all special purpose roots
125 * have separate keysets.
4008c04a 126 *
85d4e461
CM		 127 * Lock-nesting across peer nodes is always done with the immediate parent
128 * node locked thus preventing deadlock. As lockdep doesn't know this, use
129 * subclass to avoid triggering lockdep warning in such cases.
4008c04a 130 *
85d4e461
CM		 131 * The key is set by the readpage_end_io_hook after the buffer has passed
132 * csum validation but before the pages are unlocked. It is also set by
133 * btrfs_init_new_buffer on freshly allocated blocks.
4008c04a 134 *
85d4e461
CM		 135 * We also add a check to make sure the highest level of the tree is the
136 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
137 * needs update as well.
4008c04a
CM		 138 */
139#ifdef CONFIG_DEBUG_LOCK_ALLOC
140# if BTRFS_MAX_LEVEL != 8
141# error
142# endif
85d4e461
CM		 143
144static struct btrfs_lockdep_keyset {
145 u64 id; /* root objectid */
146 const char *name_stem; /* lock name stem */
147 char names[BTRFS_MAX_LEVEL + 1][20];
148 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
149} btrfs_lockdep_keysets[] = {
150 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
151 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
152 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
153 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
154 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
155 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
60b62978 156 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
85d4e461
CM		 157 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
158 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
159 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
13fd8da9 160 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
85d4e461 161 { .id = 0, .name_stem = "tree" },
4008c04a 162};
85d4e461
CM		 163
164void __init btrfs_init_lockdep(void)
165{
166 int i, j;
167
168 /* initialize lockdep class names */
169 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
170 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
171
172 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
173 snprintf(ks->names[j], sizeof(ks->names[j]),
174 "btrfs-%s-%02d", ks->name_stem, j);
175 }
176}
177
178void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
179 int level)
180{
181 struct btrfs_lockdep_keyset *ks;
182
183 BUG_ON(level >= ARRAY_SIZE(ks->keys));
184
185 /* find the matching keyset, id 0 is the default entry */
186 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
187 if (ks->id == objectid)
188 break;
189
190 lockdep_set_class_and_name(&eb->lock,
191 &ks->keys[level], ks->names[level]);
192}
193
4008c04a
CM		 194#endif
195
d352ac68
CM		 196/*
197 * extents on the btree inode are pretty simple, there's one extent
198 * that covers the entire device
199 */
b2950863 200static struct extent_map *btree_get_extent(struct inode *inode,
306e16ce 201 struct page *page, size_t pg_offset, u64 start, u64 len,
b2950863 202 int create)
7eccb903 203{
5f39d397
CM		 204 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
205 struct extent_map *em;
206 int ret;
207
890871be 208 read_lock(&em_tree->lock);
d1310b2e 209 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 210 if (em) {
211 em->bdev =
212 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 213 read_unlock(&em_tree->lock);
5f39d397 214 goto out;
a061fc8d 215 }
890871be 216 read_unlock(&em_tree->lock);
7b13b7b1 217
172ddd60 218 em = alloc_extent_map();
5f39d397
CM		 219 if (!em) {
220 em = ERR_PTR(-ENOMEM);
221 goto out;
222 }
223 em->start = 0;
0afbaf8c 224 em->len = (u64)-1;
c8b97818 225 em->block_len = (u64)-1;
5f39d397 226 em->block_start = 0;
a061fc8d 227 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 228
890871be 229 write_lock(&em_tree->lock);
09a2a8f9 230 ret = add_extent_mapping(em_tree, em, 0);
5f39d397
CM		 231 if (ret == -EEXIST) {
232 free_extent_map(em);
7b13b7b1 233 em = lookup_extent_mapping(em_tree, start, len);
b4f359ab 234 if (!em)
0433f20d 235 em = ERR_PTR(-EIO);
5f39d397 236 } else if (ret) {
7b13b7b1 237 free_extent_map(em);
0433f20d 238 em = ERR_PTR(ret);
5f39d397 239 }
890871be 240 write_unlock(&em_tree->lock);
7b13b7b1 241
5f39d397
CM		 242out:
243 return em;
7eccb903
CM		 244}
245
b0496686 246u32 btrfs_csum_data(char *data, u32 seed, size_t len)
19c00ddc 247{
163e783e 248 return crc32c(seed, data, len);
19c00ddc
CM		 249}
250
251void btrfs_csum_final(u32 crc, char *result)
252{
7e75bf3f 253 put_unaligned_le32(~crc, result);
19c00ddc
CM		 254}
255
d352ac68
CM		 256/*
257 * compute the csum for a btree block, and either verify it or write it
258 * into the csum field of the block.
259 */
19c00ddc
CM		 260static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
261 int verify)
262{
6c41761f 263 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
607d432d 264 char *result = NULL;
19c00ddc
CM		 265 unsigned long len;
266 unsigned long cur_len;
267 unsigned long offset = BTRFS_CSUM_SIZE;
19c00ddc
CM		 268 char *kaddr;
269 unsigned long map_start;
270 unsigned long map_len;
271 int err;
272 u32 crc = ~(u32)0;
607d432d 273 unsigned long inline_result;
19c00ddc
CM		 274
275 len = buf->len - offset;
d397712b 276 while (len > 0) {
19c00ddc 277 err = map_private_extent_buffer(buf, offset, 32,
a6591715 278 &kaddr, &map_start, &map_len);
d397712b 279 if (err)
19c00ddc 280 return 1;
19c00ddc 281 cur_len = min(len, map_len - (offset - map_start));
b0496686 282 crc = btrfs_csum_data(kaddr + offset - map_start,
19c00ddc
CM		 283 crc, cur_len);
284 len -= cur_len;
285 offset += cur_len;
19c00ddc 286 }
607d432d
JB		 287 if (csum_size > sizeof(inline_result)) {
288 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
289 if (!result)
290 return 1;
291 } else {
292 result = (char *)&inline_result;
293 }
294
19c00ddc
CM		 295 btrfs_csum_final(crc, result);
296
297 if (verify) {
607d432d 298 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 299 u32 val;
300 u32 found = 0;
607d432d 301 memcpy(&found, result, csum_size);
e4204ded 302
607d432d 303 read_extent_buffer(buf, &val, 0, csum_size);
7a36ddec 304 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
193f284d
CM		 305 "failed on %llu wanted %X found %X "
306 "level %d\n",
c1c9ff7c
GU		 307 root->fs_info->sb->s_id, buf->start,
308 val, found, btrfs_header_level(buf));
607d432d
JB		 309 if (result != (char *)&inline_result)
310 kfree(result);
19c00ddc
CM		 311 return 1;
312 }
313 } else {
607d432d 314 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 315 }
607d432d
JB		 316 if (result != (char *)&inline_result)
317 kfree(result);
19c00ddc
CM		 318 return 0;
319}
320
d352ac68
CM		 321/*
322 * we can't consider a given block up to date unless the transid of the
323 * block matches the transid in the parent node's pointer. This is how we
324 * detect blocks that either didn't get written at all or got written
325 * in the wrong place.
326 */
1259ab75 327static int verify_parent_transid(struct extent_io_tree *io_tree,
b9fab919
CM		 328 struct extent_buffer *eb, u64 parent_transid,
329 int atomic)
1259ab75 330{
2ac55d41 331 struct extent_state *cached_state = NULL;
1259ab75
CM		 332 int ret;
333
334 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
335 return 0;
336
b9fab919
CM		 337 if (atomic)
338 return -EAGAIN;
339
2ac55d41 340 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
d0082371 341 0, &cached_state);
0b32f4bb 342 if (extent_buffer_uptodate(eb) &&
1259ab75
CM		 343 btrfs_header_generation(eb) == parent_transid) {
344 ret = 0;
345 goto out;
346 }
7a36ddec 347 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
193f284d 348 "found %llu\n",
c1c9ff7c 349 eb->start, parent_transid, btrfs_header_generation(eb));
1259ab75 350 ret = 1;
0b32f4bb 351 clear_extent_buffer_uptodate(eb);
33958dc6 352out:
2ac55d41
JB		 353 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
354 &cached_state, GFP_NOFS);
1259ab75 355 return ret;
1259ab75
CM		 356}
357
1104a885
DS		 358/*
359 * Return 0 if the superblock checksum type matches the checksum value of that
360 * algorithm. Pass the raw disk superblock data.
361 */
362static int btrfs_check_super_csum(char *raw_disk_sb)
363{
364 struct btrfs_super_block *disk_sb =
365 (struct btrfs_super_block *)raw_disk_sb;
366 u16 csum_type = btrfs_super_csum_type(disk_sb);
367 int ret = 0;
368
369 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
370 u32 crc = ~(u32)0;
371 const int csum_size = sizeof(crc);
372 char result[csum_size];
373
374 /*
375 * The super_block structure does not span the whole
376 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
377 * is filled with zeros and is included in the checkum.
378 */
379 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
380 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
381 btrfs_csum_final(crc, result);
382
383 if (memcmp(raw_disk_sb, result, csum_size))
384 ret = 1;
667e7d94
CM		 385
386 if (ret && btrfs_super_generation(disk_sb) < 10) {
387 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
388 ret = 0;
389 }
1104a885
DS		 390 }
391
392 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
393 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
394 csum_type);
395 ret = 1;
396 }
397
398 return ret;
399}
400
d352ac68
CM		 401/*
402 * helper to read a given tree block, doing retries as required when
403 * the checksums don't match and we have alternate mirrors to try.
404 */
f188591e
CM		 405static int btree_read_extent_buffer_pages(struct btrfs_root *root,
406 struct extent_buffer *eb,
ca7a79ad 407 u64 start, u64 parent_transid)
f188591e
CM		 408{
409 struct extent_io_tree *io_tree;
ea466794 410 int failed = 0;
f188591e
CM		 411 int ret;
412 int num_copies = 0;
413 int mirror_num = 0;
ea466794 414 int failed_mirror = 0;
f188591e 415
a826d6dc 416 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM		 417 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
418 while (1) {
bb82ab88
AJ		 419 ret = read_extent_buffer_pages(io_tree, eb, start,
420 WAIT_COMPLETE,
f188591e 421 btree_get_extent, mirror_num);
256dd1bb
SB		 422 if (!ret) {
423 if (!verify_parent_transid(io_tree, eb,
b9fab919 424 parent_transid, 0))
256dd1bb
SB		 425 break;
426 else
427 ret = -EIO;
428 }
d397712b 429
a826d6dc
JB		 430 /*
431 * This buffer's crc is fine, but its contents are corrupted, so
432 * there is no reason to read the other copies, they won't be
433 * any less wrong.
434 */
435 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
ea466794
JB		 436 break;
437
5d964051 438 num_copies = btrfs_num_copies(root->fs_info,
f188591e 439 eb->start, eb->len);
4235298e 440 if (num_copies == 1)
ea466794 441 break;
4235298e 442
5cf1ab56
JB		 443 if (!failed_mirror) {
444 failed = 1;
445 failed_mirror = eb->read_mirror;
446 }
447
f188591e 448 mirror_num++;
ea466794
JB		 449 if (mirror_num == failed_mirror)
450 mirror_num++;
451
4235298e 452 if (mirror_num > num_copies)
ea466794 453 break;
f188591e 454 }
ea466794 455
c0901581 456 if (failed && !ret && failed_mirror)
ea466794
JB		 457 repair_eb_io_failure(root, eb, failed_mirror);
458
459 return ret;
f188591e 460}
19c00ddc 461
d352ac68 462/*
d397712b
CM		 463 * checksum a dirty tree block before IO. This has extra checks to make sure
464 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 465 */
d397712b 466
b2950863 467static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 468{
d1310b2e 469 struct extent_io_tree *tree;
4eee4fa4 470 u64 start = page_offset(page);
19c00ddc 471 u64 found_start;
19c00ddc 472 struct extent_buffer *eb;
f188591e 473
d1310b2e 474 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc 475
4f2de97a
JB		 476 eb = (struct extent_buffer *)page->private;
477 if (page != eb->pages[0])
478 return 0;
19c00ddc
CM		 479 found_start = btrfs_header_bytenr(eb);
480 if (found_start != start) {
55c69072 481 WARN_ON(1);
4f2de97a 482 return 0;
55c69072 483 }
55c69072 484 if (!PageUptodate(page)) {
55c69072 485 WARN_ON(1);
4f2de97a 486 return 0;
19c00ddc 487 }
19c00ddc 488 csum_tree_block(root, eb, 0);
19c00ddc
CM		 489 return 0;
490}
491
2b82032c
YZ		 492static int check_tree_block_fsid(struct btrfs_root *root,
493 struct extent_buffer *eb)
494{
495 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
496 u8 fsid[BTRFS_UUID_SIZE];
497 int ret = 1;
498
fba6aa75 499 read_extent_buffer(eb, fsid, btrfs_header_fsid(eb), BTRFS_FSID_SIZE);
2b82032c
YZ		 500 while (fs_devices) {
501 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
502 ret = 0;
503 break;
504 }
505 fs_devices = fs_devices->seed;
506 }
507 return ret;
508}
509
a826d6dc
JB		 510#define CORRUPT(reason, eb, root, slot) \
511 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
512 "root=%llu, slot=%d\n", reason, \
c1c9ff7c 513 btrfs_header_bytenr(eb), root->objectid, slot)
a826d6dc
JB		 514
515static noinline int check_leaf(struct btrfs_root *root,
516 struct extent_buffer *leaf)
517{
518 struct btrfs_key key;
519 struct btrfs_key leaf_key;
520 u32 nritems = btrfs_header_nritems(leaf);
521 int slot;
522
523 if (nritems == 0)
524 return 0;
525
526 /* Check the 0 item */
527 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
528 BTRFS_LEAF_DATA_SIZE(root)) {
529 CORRUPT("invalid item offset size pair", leaf, root, 0);
530 return -EIO;
531 }
532
533 /*
534 * Check to make sure each items keys are in the correct order and their
535 * offsets make sense. We only have to loop through nritems-1 because
536 * we check the current slot against the next slot, which verifies the
537 * next slot's offset+size makes sense and that the current's slot
538 * offset is correct.
539 */
540 for (slot = 0; slot < nritems - 1; slot++) {
541 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
542 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
543
544 /* Make sure the keys are in the right order */
545 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
546 CORRUPT("bad key order", leaf, root, slot);
547 return -EIO;
548 }
549
550 /*
551 * Make sure the offset and ends are right, remember that the
552 * item data starts at the end of the leaf and grows towards the
553 * front.
554 */
555 if (btrfs_item_offset_nr(leaf, slot) !=
556 btrfs_item_end_nr(leaf, slot + 1)) {
557 CORRUPT("slot offset bad", leaf, root, slot);
558 return -EIO;
559 }
560
561 /*
562 * Check to make sure that we don't point outside of the leaf,
563 * just incase all the items are consistent to eachother, but
564 * all point outside of the leaf.
565 */
566 if (btrfs_item_end_nr(leaf, slot) >
567 BTRFS_LEAF_DATA_SIZE(root)) {
568 CORRUPT("slot end outside of leaf", leaf, root, slot);
569 return -EIO;
570 }
571 }
572
573 return 0;
574}
575
facc8a22
MX		 576static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
577 u64 phy_offset, struct page *page,
578 u64 start, u64 end, int mirror)
ce9adaa5
CM		 579{
580 struct extent_io_tree *tree;
581 u64 found_start;
582 int found_level;
ce9adaa5
CM		 583 struct extent_buffer *eb;
584 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 585 int ret = 0;
727011e0 586 int reads_done;
ce9adaa5 587
ce9adaa5
CM		 588 if (!page->private)
589 goto out;
d397712b 590
727011e0 591 tree = &BTRFS_I(page->mapping->host)->io_tree;
4f2de97a 592 eb = (struct extent_buffer *)page->private;
d397712b 593
0b32f4bb
JB		 594 /* the pending IO might have been the only thing that kept this buffer
595 * in memory. Make sure we have a ref for all this other checks
596 */
597 extent_buffer_get(eb);
598
599 reads_done = atomic_dec_and_test(&eb->io_pages);
727011e0
CM		 600 if (!reads_done)
601 goto err;
f188591e 602
5cf1ab56 603 eb->read_mirror = mirror;
ea466794
JB		 604 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
605 ret = -EIO;
606 goto err;
607 }
608
ce9adaa5 609 found_start = btrfs_header_bytenr(eb);
727011e0 610 if (found_start != eb->start) {
7a36ddec 611 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
193f284d 612 "%llu %llu\n",
c1c9ff7c 613 found_start, eb->start);
f188591e 614 ret = -EIO;
ce9adaa5
CM		 615 goto err;
616 }
2b82032c 617 if (check_tree_block_fsid(root, eb)) {
7a36ddec 618 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
c1c9ff7c 619 eb->start);
1259ab75
CM		 620 ret = -EIO;
621 goto err;
622 }
ce9adaa5 623 found_level = btrfs_header_level(eb);
1c24c3ce
JB		 624 if (found_level >= BTRFS_MAX_LEVEL) {
625 btrfs_info(root->fs_info, "bad tree block level %d\n",
626 (int)btrfs_header_level(eb));
627 ret = -EIO;
628 goto err;
629 }
ce9adaa5 630
85d4e461
CM		 631 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
632 eb, found_level);
4008c04a 633
ce9adaa5 634 ret = csum_tree_block(root, eb, 1);
a826d6dc 635 if (ret) {
f188591e 636 ret = -EIO;
a826d6dc
JB		 637 goto err;
638 }
639
640 /*
641 * If this is a leaf block and it is corrupt, set the corrupt bit so
642 * that we don't try and read the other copies of this block, just
643 * return -EIO.
644 */
645 if (found_level == 0 && check_leaf(root, eb)) {
646 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
647 ret = -EIO;
648 }
ce9adaa5 649
0b32f4bb
JB		 650 if (!ret)
651 set_extent_buffer_uptodate(eb);
ce9adaa5 652err:
79fb65a1
JB		 653 if (reads_done &&
654 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 655 btree_readahead_hook(root, eb, eb->start, ret);
4bb31e92 656
53b381b3
DW		 657 if (ret) {
658 /*
659 * our io error hook is going to dec the io pages
660 * again, we have to make sure it has something
661 * to decrement
662 */
663 atomic_inc(&eb->io_pages);
0b32f4bb 664 clear_extent_buffer_uptodate(eb);
53b381b3 665 }
0b32f4bb 666 free_extent_buffer(eb);
ce9adaa5 667out:
f188591e 668 return ret;
ce9adaa5
CM		 669}
670
ea466794 671static int btree_io_failed_hook(struct page *page, int failed_mirror)
4bb31e92 672{
4bb31e92
AJ		 673 struct extent_buffer *eb;
674 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
675
4f2de97a 676 eb = (struct extent_buffer *)page->private;
ea466794 677 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
5cf1ab56 678 eb->read_mirror = failed_mirror;
53b381b3 679 atomic_dec(&eb->io_pages);
ea466794 680 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
4bb31e92 681 btree_readahead_hook(root, eb, eb->start, -EIO);
4bb31e92
AJ		 682 return -EIO; /* we fixed nothing */
683}
684
ce9adaa5 685static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 686{
687 struct end_io_wq *end_io_wq = bio->bi_private;
688 struct btrfs_fs_info *fs_info;
ce9adaa5 689
ce9adaa5 690 fs_info = end_io_wq->info;
ce9adaa5 691 end_io_wq->error = err;
8b712842
CM		 692 end_io_wq->work.func = end_workqueue_fn;
693 end_io_wq->work.flags = 0;
d20f7043 694
7b6d91da 695 if (bio->bi_rw & REQ_WRITE) {
53b381b3 696 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
cad321ad
CM		 697 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
698 &end_io_wq->work);
53b381b3 699 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
0cb59c99
JB		 700 btrfs_queue_worker(&fs_info->endio_freespace_worker,
701 &end_io_wq->work);
53b381b3
DW		 702 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
703 btrfs_queue_worker(&fs_info->endio_raid56_workers,
704 &end_io_wq->work);
cad321ad
CM		 705 else
706 btrfs_queue_worker(&fs_info->endio_write_workers,
707 &end_io_wq->work);
d20f7043 708 } else {
53b381b3
DW		 709 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
710 btrfs_queue_worker(&fs_info->endio_raid56_workers,
711 &end_io_wq->work);
712 else if (end_io_wq->metadata)
d20f7043
CM		 713 btrfs_queue_worker(&fs_info->endio_meta_workers,
714 &end_io_wq->work);
715 else
716 btrfs_queue_worker(&fs_info->endio_workers,
717 &end_io_wq->work);
718 }
ce9adaa5
CM		 719}
720
0cb59c99
JB		 721/*
722 * For the metadata arg you want
723 *
724 * 0 - if data
725 * 1 - if normal metadta
726 * 2 - if writing to the free space cache area
53b381b3 727 * 3 - raid parity work
0cb59c99 728 */
22c59948
CM		 729int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
730 int metadata)
0b86a832 731{
ce9adaa5 732 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 733 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
734 if (!end_io_wq)
735 return -ENOMEM;
736
737 end_io_wq->private = bio->bi_private;
738 end_io_wq->end_io = bio->bi_end_io;
22c59948 739 end_io_wq->info = info;
ce9adaa5
CM		 740 end_io_wq->error = 0;
741 end_io_wq->bio = bio;
22c59948 742 end_io_wq->metadata = metadata;
ce9adaa5
CM		 743
744 bio->bi_private = end_io_wq;
745 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 746 return 0;
747}
748
b64a2851 749unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 750{
4854ddd0
CM		 751 unsigned long limit = min_t(unsigned long,
752 info->workers.max_workers,
753 info->fs_devices->open_devices);
754 return 256 * limit;
755}
0986fe9e 756
4a69a410
CM		 757static void run_one_async_start(struct btrfs_work *work)
758{
4a69a410 759 struct async_submit_bio *async;
79787eaa 760 int ret;
4a69a410
CM		 761
762 async = container_of(work, struct async_submit_bio, work);
79787eaa
JM		 763 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
764 async->mirror_num, async->bio_flags,
765 async->bio_offset);
766 if (ret)
767 async->error = ret;
4a69a410
CM		 768}
769
770static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 771{
772 struct btrfs_fs_info *fs_info;
773 struct async_submit_bio *async;
4854ddd0 774 int limit;
8b712842
CM		 775
776 async = container_of(work, struct async_submit_bio, work);
777 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 778
b64a2851 779 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 780 limit = limit * 2 / 3;
781
66657b31 782 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
b64a2851 783 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 784 wake_up(&fs_info->async_submit_wait);
785
79787eaa
JM		 786 /* If an error occured we just want to clean up the bio and move on */
787 if (async->error) {
788 bio_endio(async->bio, async->error);
789 return;
790 }
791
4a69a410 792 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 793 async->mirror_num, async->bio_flags,
794 async->bio_offset);
4a69a410
CM		 795}
796
797static void run_one_async_free(struct btrfs_work *work)
798{
799 struct async_submit_bio *async;
800
801 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 802 kfree(async);
803}
804
44b8bd7e
CM		 805int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
806 int rw, struct bio *bio, int mirror_num,
c8b97818 807 unsigned long bio_flags,
eaf25d93 808 u64 bio_offset,
4a69a410
CM		 809 extent_submit_bio_hook_t *submit_bio_start,
810 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 811{
812 struct async_submit_bio *async;
813
814 async = kmalloc(sizeof(*async), GFP_NOFS);
815 if (!async)
816 return -ENOMEM;
817
818 async->inode = inode;
819 async->rw = rw;
820 async->bio = bio;
821 async->mirror_num = mirror_num;
4a69a410
CM		 822 async->submit_bio_start = submit_bio_start;
823 async->submit_bio_done = submit_bio_done;
824
825 async->work.func = run_one_async_start;
826 async->work.ordered_func = run_one_async_done;
827 async->work.ordered_free = run_one_async_free;
828
8b712842 829 async->work.flags = 0;
c8b97818 830 async->bio_flags = bio_flags;
eaf25d93 831 async->bio_offset = bio_offset;
8c8bee1d 832
79787eaa
JM		 833 async->error = 0;
834
cb03c743 835 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 836
7b6d91da 837 if (rw & REQ_SYNC)
d313d7a3
CM		 838 btrfs_set_work_high_prio(&async->work);
839
8b712842 840 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 841
d397712b 842 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 843 atomic_read(&fs_info->nr_async_submits)) {
844 wait_event(fs_info->async_submit_wait,
845 (atomic_read(&fs_info->nr_async_submits) == 0));
846 }
847
44b8bd7e
CM		 848 return 0;
849}
850
ce3ed71a
CM		 851static int btree_csum_one_bio(struct bio *bio)
852{
853 struct bio_vec *bvec = bio->bi_io_vec;
854 int bio_index = 0;
855 struct btrfs_root *root;
79787eaa 856 int ret = 0;
ce3ed71a
CM		 857
858 WARN_ON(bio->bi_vcnt <= 0);
d397712b 859 while (bio_index < bio->bi_vcnt) {
ce3ed71a 860 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
79787eaa
JM		 861 ret = csum_dirty_buffer(root, bvec->bv_page);
862 if (ret)
863 break;
ce3ed71a
CM		 864 bio_index++;
865 bvec++;
866 }
79787eaa 867 return ret;
ce3ed71a
CM		 868}
869
4a69a410
CM		 870static int __btree_submit_bio_start(struct inode *inode, int rw,
871 struct bio *bio, int mirror_num,
eaf25d93
CM		 872 unsigned long bio_flags,
873 u64 bio_offset)
22c59948 874{
8b712842
CM		 875 /*
876 * when we're called for a write, we're already in the async
5443be45 877 * submission context. Just jump into btrfs_map_bio
8b712842 878 */
79787eaa 879 return btree_csum_one_bio(bio);
4a69a410 880}
22c59948 881
4a69a410 882static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 883 int mirror_num, unsigned long bio_flags,
884 u64 bio_offset)
4a69a410 885{
61891923
SB		 886 int ret;
887
8b712842 888 /*
4a69a410
CM		 889 * when we're called for a write, we're already in the async
890 * submission context. Just jump into btrfs_map_bio
8b712842 891 */
61891923
SB		 892 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
893 if (ret)
894 bio_endio(bio, ret);
895 return ret;
0b86a832
CM		 896}
897
de0022b9
JB		 898static int check_async_write(struct inode *inode, unsigned long bio_flags)
899{
900 if (bio_flags & EXTENT_BIO_TREE_LOG)
901 return 0;
902#ifdef CONFIG_X86
903 if (cpu_has_xmm4_2)
904 return 0;
905#endif
906 return 1;
907}
908
44b8bd7e 909static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 910 int mirror_num, unsigned long bio_flags,
911 u64 bio_offset)
44b8bd7e 912{
de0022b9 913 int async = check_async_write(inode, bio_flags);
cad321ad
CM		 914 int ret;
915
7b6d91da 916 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 917 /*
918 * called for a read, do the setup so that checksum validation
919 * can happen in the async kernel threads
920 */
f3f266ab
CM		 921 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
922 bio, 1);
1d4284bd 923 if (ret)
61891923
SB		 924 goto out_w_error;
925 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
926 mirror_num, 0);
de0022b9
JB		 927 } else if (!async) {
928 ret = btree_csum_one_bio(bio);
929 if (ret)
61891923
SB		 930 goto out_w_error;
931 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
932 mirror_num, 0);
933 } else {
934 /*
935 * kthread helpers are used to submit writes so that
936 * checksumming can happen in parallel across all CPUs
937 */
938 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
939 inode, rw, bio, mirror_num, 0,
940 bio_offset,
941 __btree_submit_bio_start,
942 __btree_submit_bio_done);
44b8bd7e 943 }
d313d7a3 944
61891923
SB		 945 if (ret) {
946out_w_error:
947 bio_endio(bio, ret);
948 }
949 return ret;
44b8bd7e
CM		 950}
951
3dd1462e 952#ifdef CONFIG_MIGRATION
784b4e29 953static int btree_migratepage(struct address_space *mapping,
a6bc32b8
MG		 954 struct page *newpage, struct page *page,
955 enum migrate_mode mode)
784b4e29
CM		 956{
957 /*
958 * we can't safely write a btree page from here,
959 * we haven't done the locking hook
960 */
961 if (PageDirty(page))
962 return -EAGAIN;
963 /*
964 * Buffers may be managed in a filesystem specific way.
965 * We must have no buffers or drop them.
966 */
967 if (page_has_private(page) &&
968 !try_to_release_page(page, GFP_KERNEL))
969 return -EAGAIN;
a6bc32b8 970 return migrate_page(mapping, newpage, page, mode);
784b4e29 971}
3dd1462e 972#endif
784b4e29 973
0da5468f
CM		 974
975static int btree_writepages(struct address_space *mapping,
976 struct writeback_control *wbc)
977{
d1310b2e 978 struct extent_io_tree *tree;
e2d84521
MX		 979 struct btrfs_fs_info *fs_info;
980 int ret;
981
d1310b2e 982 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 983 if (wbc->sync_mode == WB_SYNC_NONE) {
448d640b
CM		 984
985 if (wbc->for_kupdate)
986 return 0;
987
e2d84521 988 fs_info = BTRFS_I(mapping->host)->root->fs_info;
b9473439 989 /* this is a bit racy, but that's ok */
e2d84521
MX		 990 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
991 BTRFS_DIRTY_METADATA_THRESH);
992 if (ret < 0)
793955bc 993 return 0;
793955bc 994 }
0b32f4bb 995 return btree_write_cache_pages(mapping, wbc);
0da5468f
CM		 996}
997
b2950863 998static int btree_readpage(struct file *file, struct page *page)
5f39d397 999{
d1310b2e
CM		 1000 struct extent_io_tree *tree;
1001 tree = &BTRFS_I(page->mapping->host)->io_tree;
8ddc7d9c 1002 return extent_read_full_page(tree, page, btree_get_extent, 0);
5f39d397 1003}
22b0ebda 1004
70dec807 1005static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 1006{
98509cfc 1007 if (PageWriteback(page) || PageDirty(page))
d397712b 1008 return 0;
0c4e538b 1009
f7a52a40 1010 return try_release_extent_buffer(page);
d98237b3
CM		 1011}
1012
d47992f8
LC		 1013static void btree_invalidatepage(struct page *page, unsigned int offset,
1014 unsigned int length)
d98237b3 1015{
d1310b2e
CM		 1016 struct extent_io_tree *tree;
1017 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 1018 extent_invalidatepage(tree, page, offset);
1019 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 1020 if (PagePrivate(page)) {
d397712b
CM		 1021 printk(KERN_WARNING "btrfs warning page private not zero "
1022 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 1023 ClearPagePrivate(page);
1024 set_page_private(page, 0);
1025 page_cache_release(page);
1026 }
d98237b3
CM		 1027}
1028
0b32f4bb
JB		 1029static int btree_set_page_dirty(struct page *page)
1030{
bb146eb2 1031#ifdef DEBUG
0b32f4bb
JB		 1032 struct extent_buffer *eb;
1033
1034 BUG_ON(!PagePrivate(page));
1035 eb = (struct extent_buffer *)page->private;
1036 BUG_ON(!eb);
1037 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 BUG_ON(!atomic_read(&eb->refs));
1039 btrfs_assert_tree_locked(eb);
bb146eb2 1040#endif
0b32f4bb
JB		 1041 return __set_page_dirty_nobuffers(page);
1042}
1043
7f09410b 1044static const struct address_space_operations btree_aops = {
d98237b3 1045 .readpage = btree_readpage,
0da5468f 1046 .writepages = btree_writepages,
5f39d397
CM		 1047 .releasepage = btree_releasepage,
1048 .invalidatepage = btree_invalidatepage,
5a92bc88 1049#ifdef CONFIG_MIGRATION
784b4e29 1050 .migratepage = btree_migratepage,
5a92bc88 1051#endif
0b32f4bb 1052 .set_page_dirty = btree_set_page_dirty,
d98237b3
CM		 1053};
1054
ca7a79ad
CM		 1055int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1056 u64 parent_transid)
090d1875 1057{
5f39d397
CM		 1058 struct extent_buffer *buf = NULL;
1059 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 1060 int ret = 0;
090d1875 1061
db94535d 1062 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 1063 if (!buf)
090d1875 1064 return 0;
d1310b2e 1065 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
bb82ab88 1066 buf, 0, WAIT_NONE, btree_get_extent, 0);
5f39d397 1067 free_extent_buffer(buf);
de428b63 1068 return ret;
090d1875
CM		 1069}
1070
ab0fff03
AJ		 1071int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1072 int mirror_num, struct extent_buffer **eb)
1073{
1074 struct extent_buffer *buf = NULL;
1075 struct inode *btree_inode = root->fs_info->btree_inode;
1076 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1077 int ret;
1078
1079 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1080 if (!buf)
1081 return 0;
1082
1083 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1084
1085 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1086 btree_get_extent, mirror_num);
1087 if (ret) {
1088 free_extent_buffer(buf);
1089 return ret;
1090 }
1091
1092 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1093 free_extent_buffer(buf);
1094 return -EIO;
0b32f4bb 1095 } else if (extent_buffer_uptodate(buf)) {
ab0fff03
AJ		 1096 *eb = buf;
1097 } else {
1098 free_extent_buffer(buf);
1099 }
1100 return 0;
1101}
1102
0999df54
CM		 1103struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1104 u64 bytenr, u32 blocksize)
1105{
1106 struct inode *btree_inode = root->fs_info->btree_inode;
1107 struct extent_buffer *eb;
1108 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
f09d1f60 1109 bytenr, blocksize);
0999df54
CM		 1110 return eb;
1111}
1112
1113struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1114 u64 bytenr, u32 blocksize)
1115{
1116 struct inode *btree_inode = root->fs_info->btree_inode;
1117 struct extent_buffer *eb;
1118
1119 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
727011e0 1120 bytenr, blocksize);
0999df54
CM		 1121 return eb;
1122}
1123
1124
e02119d5
CM		 1125int btrfs_write_tree_block(struct extent_buffer *buf)
1126{
727011e0 1127 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
8aa38c31 1128 buf->start + buf->len - 1);
e02119d5
CM		 1129}
1130
1131int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1132{
727011e0 1133 return filemap_fdatawait_range(buf->pages[0]->mapping,
8aa38c31 1134 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 1135}
1136
0999df54 1137struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 1138 u32 blocksize, u64 parent_transid)
0999df54
CM		 1139{
1140 struct extent_buffer *buf = NULL;
0999df54
CM		 1141 int ret;
1142
0999df54
CM		 1143 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1144 if (!buf)
1145 return NULL;
0999df54 1146
ca7a79ad 1147 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
0f0fe8f7
FDBM		 1148 if (ret) {
1149 free_extent_buffer(buf);
1150 return NULL;
1151 }
5f39d397 1152 return buf;
ce9adaa5 1153
eb60ceac
CM		 1154}
1155
d5c13f92
JM		 1156void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1157 struct extent_buffer *buf)
ed2ff2cb 1158{
e2d84521
MX		 1159 struct btrfs_fs_info *fs_info = root->fs_info;
1160
55c69072 1161 if (btrfs_header_generation(buf) ==
e2d84521 1162 fs_info->running_transaction->transid) {
b9447ef8 1163 btrfs_assert_tree_locked(buf);
b4ce94de 1164
b9473439 1165 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
e2d84521
MX		 1166 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1167 -buf->len,
1168 fs_info->dirty_metadata_batch);
ed7b63eb
JB		 1169 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1170 btrfs_set_lock_blocking(buf);
1171 clear_extent_buffer_dirty(buf);
1172 }
925baedd 1173 }
5f39d397
CM		 1174}
1175
143bede5
JM		 1176static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1177 u32 stripesize, struct btrfs_root *root,
1178 struct btrfs_fs_info *fs_info,
1179 u64 objectid)
d97e63b6 1180{
cfaa7295 1181 root->node = NULL;
a28ec197 1182 root->commit_root = NULL;
db94535d
CM		 1183 root->sectorsize = sectorsize;
1184 root->nodesize = nodesize;
1185 root->leafsize = leafsize;
87ee04eb 1186 root->stripesize = stripesize;
123abc88 1187 root->ref_cows = 0;
0b86a832 1188 root->track_dirty = 0;
c71bf099 1189 root->in_radix = 0;
d68fc57b
YZ		 1190 root->orphan_item_inserted = 0;
1191 root->orphan_cleanup_state = 0;
0b86a832 1192
0f7d52f4
CM		 1193 root->objectid = objectid;
1194 root->last_trans = 0;
13a8a7c8 1195 root->highest_objectid = 0;
eb73c1b7 1196 root->nr_delalloc_inodes = 0;
199c2a9c 1197 root->nr_ordered_extents = 0;
58176a96 1198 root->name = NULL;
6bef4d31 1199 root->inode_tree = RB_ROOT;
16cdcec7 1200 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1201 root->block_rsv = NULL;
d68fc57b 1202 root->orphan_block_rsv = NULL;
0b86a832
CM		 1203
1204 INIT_LIST_HEAD(&root->dirty_list);
5d4f98a2 1205 INIT_LIST_HEAD(&root->root_list);
eb73c1b7
MX		 1206 INIT_LIST_HEAD(&root->delalloc_inodes);
1207 INIT_LIST_HEAD(&root->delalloc_root);
199c2a9c
MX		 1208 INIT_LIST_HEAD(&root->ordered_extents);
1209 INIT_LIST_HEAD(&root->ordered_root);
2ab28f32
JB		 1210 INIT_LIST_HEAD(&root->logged_list[0]);
1211 INIT_LIST_HEAD(&root->logged_list[1]);
d68fc57b 1212 spin_lock_init(&root->orphan_lock);
5d4f98a2 1213 spin_lock_init(&root->inode_lock);
eb73c1b7 1214 spin_lock_init(&root->delalloc_lock);
199c2a9c 1215 spin_lock_init(&root->ordered_extent_lock);
f0486c68 1216 spin_lock_init(&root->accounting_lock);
2ab28f32
JB		 1217 spin_lock_init(&root->log_extents_lock[0]);
1218 spin_lock_init(&root->log_extents_lock[1]);
a2135011 1219 mutex_init(&root->objectid_mutex);
e02119d5 1220 mutex_init(&root->log_mutex);
7237f183
YZ		 1221 init_waitqueue_head(&root->log_writer_wait);
1222 init_waitqueue_head(&root->log_commit_wait[0]);
1223 init_waitqueue_head(&root->log_commit_wait[1]);
1224 atomic_set(&root->log_commit[0], 0);
1225 atomic_set(&root->log_commit[1], 0);
1226 atomic_set(&root->log_writers, 0);
2ecb7923 1227 atomic_set(&root->log_batch, 0);
8a35d95f 1228 atomic_set(&root->orphan_inodes, 0);
b0feb9d9 1229 atomic_set(&root->refs, 1);
7237f183 1230 root->log_transid = 0;
257c62e1 1231 root->last_log_commit = 0;
06ea65a3
JB		 1232 if (fs_info)
1233 extent_io_tree_init(&root->dirty_log_pages,
1234 fs_info->btree_inode->i_mapping);
017e5369 1235
3768f368
CM		 1236 memset(&root->root_key, 0, sizeof(root->root_key));
1237 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1238 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1239 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
06ea65a3
JB		 1240 if (fs_info)
1241 root->defrag_trans_start = fs_info->generation;
1242 else
1243 root->defrag_trans_start = 0;
58176a96 1244 init_completion(&root->kobj_unregister);
6702ed49 1245 root->defrag_running = 0;
4d775673 1246 root->root_key.objectid = objectid;
0ee5dc67 1247 root->anon_dev = 0;
8ea05e3a 1248
5f3ab90a 1249 spin_lock_init(&root->root_item_lock);
3768f368
CM		 1250}
1251
f84a8bd6 1252static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
6f07e42e
AV		 1253{
1254 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1255 if (root)
1256 root->fs_info = fs_info;
1257 return root;
1258}
1259
06ea65a3
JB		 1260#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1261/* Should only be used by the testing infrastructure */
1262struct btrfs_root *btrfs_alloc_dummy_root(void)
1263{
1264 struct btrfs_root *root;
1265
1266 root = btrfs_alloc_root(NULL);
1267 if (!root)
1268 return ERR_PTR(-ENOMEM);
1269 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1270 root->dummy_root = 1;
1271
1272 return root;
1273}
1274#endif
1275
20897f5c
AJ		 1276struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1277 struct btrfs_fs_info *fs_info,
1278 u64 objectid)
1279{
1280 struct extent_buffer *leaf;
1281 struct btrfs_root *tree_root = fs_info->tree_root;
1282 struct btrfs_root *root;
1283 struct btrfs_key key;
1284 int ret = 0;
1285 u64 bytenr;
6463fe58 1286 uuid_le uuid;
20897f5c
AJ		 1287
1288 root = btrfs_alloc_root(fs_info);
1289 if (!root)
1290 return ERR_PTR(-ENOMEM);
1291
1292 __setup_root(tree_root->nodesize, tree_root->leafsize,
1293 tree_root->sectorsize, tree_root->stripesize,
1294 root, fs_info, objectid);
1295 root->root_key.objectid = objectid;
1296 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1297 root->root_key.offset = 0;
1298
1299 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1300 0, objectid, NULL, 0, 0, 0);
1301 if (IS_ERR(leaf)) {
1302 ret = PTR_ERR(leaf);
1dd05682 1303 leaf = NULL;
20897f5c
AJ		 1304 goto fail;
1305 }
1306
1307 bytenr = leaf->start;
1308 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1309 btrfs_set_header_bytenr(leaf, leaf->start);
1310 btrfs_set_header_generation(leaf, trans->transid);
1311 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1312 btrfs_set_header_owner(leaf, objectid);
1313 root->node = leaf;
1314
fba6aa75 1315 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(leaf),
20897f5c
AJ		 1316 BTRFS_FSID_SIZE);
1317 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
b308bc2f 1318 btrfs_header_chunk_tree_uuid(leaf),
20897f5c
AJ		 1319 BTRFS_UUID_SIZE);
1320 btrfs_mark_buffer_dirty(leaf);
1321
1322 root->commit_root = btrfs_root_node(root);
1323 root->track_dirty = 1;
1324
1325
1326 root->root_item.flags = 0;
1327 root->root_item.byte_limit = 0;
1328 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1329 btrfs_set_root_generation(&root->root_item, trans->transid);
1330 btrfs_set_root_level(&root->root_item, 0);
1331 btrfs_set_root_refs(&root->root_item, 1);
1332 btrfs_set_root_used(&root->root_item, leaf->len);
1333 btrfs_set_root_last_snapshot(&root->root_item, 0);
1334 btrfs_set_root_dirid(&root->root_item, 0);
6463fe58
SB		 1335 uuid_le_gen(&uuid);
1336 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
20897f5c
AJ		 1337 root->root_item.drop_level = 0;
1338
1339 key.objectid = objectid;
1340 key.type = BTRFS_ROOT_ITEM_KEY;
1341 key.offset = 0;
1342 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1343 if (ret)
1344 goto fail;
1345
1346 btrfs_tree_unlock(leaf);
1347
1dd05682
TI		 1348 return root;
1349
20897f5c 1350fail:
1dd05682
TI		 1351 if (leaf) {
1352 btrfs_tree_unlock(leaf);
1353 free_extent_buffer(leaf);
1354 }
1355 kfree(root);
20897f5c 1356
1dd05682 1357 return ERR_PTR(ret);
20897f5c
AJ		 1358}
1359
7237f183
YZ		 1360static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1361 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1362{
1363 struct btrfs_root *root;
1364 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1365 struct extent_buffer *leaf;
e02119d5 1366
6f07e42e 1367 root = btrfs_alloc_root(fs_info);
e02119d5 1368 if (!root)
7237f183 1369 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1370
1371 __setup_root(tree_root->nodesize, tree_root->leafsize,
1372 tree_root->sectorsize, tree_root->stripesize,
1373 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1374
1375 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1376 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1377 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1378 /*
1379 * log trees do not get reference counted because they go away
1380 * before a real commit is actually done. They do store pointers
1381 * to file data extents, and those reference counts still get
1382 * updated (along with back refs to the log tree).
1383 */
e02119d5
CM		 1384 root->ref_cows = 0;
1385
5d4f98a2 1386 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
66d7e7f0 1387 BTRFS_TREE_LOG_OBJECTID, NULL,
5581a51a 1388 0, 0, 0);
7237f183
YZ		 1389 if (IS_ERR(leaf)) {
1390 kfree(root);
1391 return ERR_CAST(leaf);
1392 }
e02119d5 1393
5d4f98a2
YZ		 1394 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1395 btrfs_set_header_bytenr(leaf, leaf->start);
1396 btrfs_set_header_generation(leaf, trans->transid);
1397 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1398 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1399 root->node = leaf;
e02119d5
CM		 1400
1401 write_extent_buffer(root->node, root->fs_info->fsid,
fba6aa75 1402 btrfs_header_fsid(root->node), BTRFS_FSID_SIZE);
e02119d5
CM		 1403 btrfs_mark_buffer_dirty(root->node);
1404 btrfs_tree_unlock(root->node);
7237f183
YZ		 1405 return root;
1406}
1407
1408int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1409 struct btrfs_fs_info *fs_info)
1410{
1411 struct btrfs_root *log_root;
1412
1413 log_root = alloc_log_tree(trans, fs_info);
1414 if (IS_ERR(log_root))
1415 return PTR_ERR(log_root);
1416 WARN_ON(fs_info->log_root_tree);
1417 fs_info->log_root_tree = log_root;
1418 return 0;
1419}
1420
1421int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1422 struct btrfs_root *root)
1423{
1424 struct btrfs_root *log_root;
1425 struct btrfs_inode_item *inode_item;
1426
1427 log_root = alloc_log_tree(trans, root->fs_info);
1428 if (IS_ERR(log_root))
1429 return PTR_ERR(log_root);
1430
1431 log_root->last_trans = trans->transid;
1432 log_root->root_key.offset = root->root_key.objectid;
1433
1434 inode_item = &log_root->root_item.inode;
3cae210f
QW		 1435 btrfs_set_stack_inode_generation(inode_item, 1);
1436 btrfs_set_stack_inode_size(inode_item, 3);
1437 btrfs_set_stack_inode_nlink(inode_item, 1);
1438 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1439 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
7237f183 1440
5d4f98a2 1441 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1442
1443 WARN_ON(root->log_root);
1444 root->log_root = log_root;
1445 root->log_transid = 0;
257c62e1 1446 root->last_log_commit = 0;
e02119d5
CM		 1447 return 0;
1448}
1449
35a3621b
SB		 1450static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1451 struct btrfs_key *key)
e02119d5
CM		 1452{
1453 struct btrfs_root *root;
1454 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1455 struct btrfs_path *path;
84234f3a 1456 u64 generation;
db94535d 1457 u32 blocksize;
cb517eab 1458 int ret;
0f7d52f4 1459
cb517eab
MX		 1460 path = btrfs_alloc_path();
1461 if (!path)
0f7d52f4 1462 return ERR_PTR(-ENOMEM);
cb517eab
MX		 1463
1464 root = btrfs_alloc_root(fs_info);
1465 if (!root) {
1466 ret = -ENOMEM;
1467 goto alloc_fail;
0f7d52f4
CM		 1468 }
1469
db94535d 1470 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb 1471 tree_root->sectorsize, tree_root->stripesize,
cb517eab 1472 root, fs_info, key->objectid);
0f7d52f4 1473
cb517eab
MX		 1474 ret = btrfs_find_root(tree_root, key, path,
1475 &root->root_item, &root->root_key);
0f7d52f4 1476 if (ret) {
13a8a7c8
YZ		 1477 if (ret > 0)
1478 ret = -ENOENT;
cb517eab 1479 goto find_fail;
0f7d52f4 1480 }
13a8a7c8 1481
84234f3a 1482 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1483 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1484 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1485 blocksize, generation);
cb517eab
MX		 1486 if (!root->node) {
1487 ret = -ENOMEM;
1488 goto find_fail;
1489 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1490 ret = -EIO;
1491 goto read_fail;
416bc658 1492 }
5d4f98a2 1493 root->commit_root = btrfs_root_node(root);
13a8a7c8 1494out:
cb517eab
MX		 1495 btrfs_free_path(path);
1496 return root;
1497
1498read_fail:
1499 free_extent_buffer(root->node);
1500find_fail:
1501 kfree(root);
1502alloc_fail:
1503 root = ERR_PTR(ret);
1504 goto out;
1505}
1506
1507struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1508 struct btrfs_key *location)
1509{
1510 struct btrfs_root *root;
1511
1512 root = btrfs_read_tree_root(tree_root, location);
1513 if (IS_ERR(root))
1514 return root;
1515
1516 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1517 root->ref_cows = 1;
08fe4db1
LZ		 1518 btrfs_check_and_init_root_item(&root->root_item);
1519 }
13a8a7c8 1520
5eda7b5e
CM		 1521 return root;
1522}
1523
cb517eab
MX		 1524int btrfs_init_fs_root(struct btrfs_root *root)
1525{
1526 int ret;
1527
1528 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1529 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1530 GFP_NOFS);
1531 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1532 ret = -ENOMEM;
1533 goto fail;
1534 }
1535
1536 btrfs_init_free_ino_ctl(root);
1537 mutex_init(&root->fs_commit_mutex);
1538 spin_lock_init(&root->cache_lock);
1539 init_waitqueue_head(&root->cache_wait);
1540
1541 ret = get_anon_bdev(&root->anon_dev);
1542 if (ret)
1543 goto fail;
1544 return 0;
1545fail:
1546 kfree(root->free_ino_ctl);
1547 kfree(root->free_ino_pinned);
1548 return ret;
1549}
1550
171170c1
ST		 1551static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1552 u64 root_id)
cb517eab
MX		 1553{
1554 struct btrfs_root *root;
1555
1556 spin_lock(&fs_info->fs_roots_radix_lock);
1557 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1558 (unsigned long)root_id);
1559 spin_unlock(&fs_info->fs_roots_radix_lock);
1560 return root;
1561}
1562
1563int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1564 struct btrfs_root *root)
1565{
1566 int ret;
1567
1568 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1569 if (ret)
1570 return ret;
1571
1572 spin_lock(&fs_info->fs_roots_radix_lock);
1573 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1574 (unsigned long)root->root_key.objectid,
1575 root);
1576 if (ret == 0)
1577 root->in_radix = 1;
1578 spin_unlock(&fs_info->fs_roots_radix_lock);
1579 radix_tree_preload_end();
1580
1581 return ret;
1582}
1583
c00869f1
MX		 1584struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1585 struct btrfs_key *location,
1586 bool check_ref)
5eda7b5e
CM		 1587{
1588 struct btrfs_root *root;
1589 int ret;
1590
edbd8d4e
CM		 1591 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1592 return fs_info->tree_root;
1593 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1594 return fs_info->extent_root;
8f18cf13
CM		 1595 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1596 return fs_info->chunk_root;
1597 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1598 return fs_info->dev_root;
0403e47e
YZ		 1599 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1600 return fs_info->csum_root;
bcef60f2
AJ		 1601 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1602 return fs_info->quota_root ? fs_info->quota_root :
1603 ERR_PTR(-ENOENT);
f7a81ea4
SB		 1604 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1605 return fs_info->uuid_root ? fs_info->uuid_root :
1606 ERR_PTR(-ENOENT);
4df27c4d 1607again:
cb517eab 1608 root = btrfs_lookup_fs_root(fs_info, location->objectid);
48475471 1609 if (root) {
c00869f1 1610 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
48475471 1611 return ERR_PTR(-ENOENT);
5eda7b5e 1612 return root;
48475471 1613 }
5eda7b5e 1614
cb517eab 1615 root = btrfs_read_fs_root(fs_info->tree_root, location);
5eda7b5e
CM		 1616 if (IS_ERR(root))
1617 return root;
3394e160 1618
c00869f1 1619 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
cb517eab 1620 ret = -ENOENT;
581bb050 1621 goto fail;
35a30d7c 1622 }
581bb050 1623
cb517eab 1624 ret = btrfs_init_fs_root(root);
ac08aedf
CM		 1625 if (ret)
1626 goto fail;
3394e160 1627
d68fc57b
YZ		 1628 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1629 if (ret < 0)
1630 goto fail;
1631 if (ret == 0)
1632 root->orphan_item_inserted = 1;
1633
cb517eab 1634 ret = btrfs_insert_fs_root(fs_info, root);
0f7d52f4 1635 if (ret) {
4df27c4d
YZ		 1636 if (ret == -EEXIST) {
1637 free_fs_root(root);
1638 goto again;
1639 }
1640 goto fail;
0f7d52f4 1641 }
edbd8d4e 1642 return root;
4df27c4d
YZ		 1643fail:
1644 free_fs_root(root);
1645 return ERR_PTR(ret);
edbd8d4e
CM		 1646}
1647
04160088
CM		 1648static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1649{
1650 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1651 int ret = 0;
04160088
CM		 1652 struct btrfs_device *device;
1653 struct backing_dev_info *bdi;
b7967db7 1654
1f78160c
XG		 1655 rcu_read_lock();
1656 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1657 if (!device->bdev)
1658 continue;
04160088
CM		 1659 bdi = blk_get_backing_dev_info(device->bdev);
1660 if (bdi && bdi_congested(bdi, bdi_bits)) {
1661 ret = 1;
1662 break;
1663 }
1664 }
1f78160c 1665 rcu_read_unlock();
04160088
CM		 1666 return ret;
1667}
1668
ad081f14
JA		 1669/*
1670 * If this fails, caller must call bdi_destroy() to get rid of the
1671 * bdi again.
1672 */
04160088
CM		 1673static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1674{
ad081f14
JA		 1675 int err;
1676
1677 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1678 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1679 if (err)
1680 return err;
1681
4575c9cc 1682 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1683 bdi->congested_fn = btrfs_congested_fn;
1684 bdi->congested_data = info;
1685 return 0;
1686}
1687
8b712842
CM		 1688/*
1689 * called by the kthread helper functions to finally call the bio end_io
1690 * functions. This is where read checksum verification actually happens
1691 */
1692static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1693{
ce9adaa5 1694 struct bio *bio;
8b712842
CM		 1695 struct end_io_wq *end_io_wq;
1696 struct btrfs_fs_info *fs_info;
ce9adaa5 1697 int error;
ce9adaa5 1698
8b712842
CM		 1699 end_io_wq = container_of(work, struct end_io_wq, work);
1700 bio = end_io_wq->bio;
1701 fs_info = end_io_wq->info;
ce9adaa5 1702
8b712842
CM		 1703 error = end_io_wq->error;
1704 bio->bi_private = end_io_wq->private;
1705 bio->bi_end_io = end_io_wq->end_io;
1706 kfree(end_io_wq);
8b712842 1707 bio_endio(bio, error);
44b8bd7e
CM		 1708}
1709
a74a4b97
CM		 1710static int cleaner_kthread(void *arg)
1711{
1712 struct btrfs_root *root = arg;
d0278245 1713 int again;
a74a4b97
CM		 1714
1715 do {
d0278245 1716 again = 0;
a74a4b97 1717
d0278245 1718 /* Make the cleaner go to sleep early. */
babbf170 1719 if (btrfs_need_cleaner_sleep(root))
d0278245
MX		 1720 goto sleep;
1721
1722 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1723 goto sleep;
1724
dc7f370c
MX		 1725 /*
1726 * Avoid the problem that we change the status of the fs
1727 * during the above check and trylock.
1728 */
babbf170 1729 if (btrfs_need_cleaner_sleep(root)) {
dc7f370c
MX		 1730 mutex_unlock(&root->fs_info->cleaner_mutex);
1731 goto sleep;
76dda93c 1732 }
a74a4b97 1733
d0278245
MX		 1734 btrfs_run_delayed_iputs(root);
1735 again = btrfs_clean_one_deleted_snapshot(root);
1736 mutex_unlock(&root->fs_info->cleaner_mutex);
1737
1738 /*
05323cd1
MX		 1739 * The defragger has dealt with the R/O remount and umount,
1740 * needn't do anything special here.
d0278245
MX		 1741 */
1742 btrfs_run_defrag_inodes(root->fs_info);
1743sleep:
9d1a2a3a 1744 if (!try_to_freeze() && !again) {
a74a4b97 1745 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1746 if (!kthread_should_stop())
1747 schedule();
a74a4b97
CM		 1748 __set_current_state(TASK_RUNNING);
1749 }
1750 } while (!kthread_should_stop());
1751 return 0;
1752}
1753
1754static int transaction_kthread(void *arg)
1755{
1756 struct btrfs_root *root = arg;
1757 struct btrfs_trans_handle *trans;
1758 struct btrfs_transaction *cur;
8929ecfa 1759 u64 transid;
a74a4b97
CM		 1760 unsigned long now;
1761 unsigned long delay;
914b2007 1762 bool cannot_commit;
a74a4b97
CM		 1763
1764 do {
914b2007 1765 cannot_commit = false;
8b87dc17 1766 delay = HZ * root->fs_info->commit_interval;
a74a4b97
CM		 1767 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1768
a4abeea4 1769 spin_lock(&root->fs_info->trans_lock);
a74a4b97
CM		 1770 cur = root->fs_info->running_transaction;
1771 if (!cur) {
a4abeea4 1772 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1773 goto sleep;
1774 }
31153d81 1775
a74a4b97 1776 now = get_seconds();
4a9d8bde 1777 if (cur->state < TRANS_STATE_BLOCKED &&
8b87dc17
DS		 1778 (now < cur->start_time ||
1779 now - cur->start_time < root->fs_info->commit_interval)) {
a4abeea4 1780 spin_unlock(&root->fs_info->trans_lock);
a74a4b97
CM		 1781 delay = HZ * 5;
1782 goto sleep;
1783 }
8929ecfa 1784 transid = cur->transid;
a4abeea4 1785 spin_unlock(&root->fs_info->trans_lock);
56bec294 1786
79787eaa 1787 /* If the file system is aborted, this will always fail. */
354aa0fb 1788 trans = btrfs_attach_transaction(root);
914b2007 1789 if (IS_ERR(trans)) {
354aa0fb
MX		 1790 if (PTR_ERR(trans) != -ENOENT)
1791 cannot_commit = true;
79787eaa 1792 goto sleep;
914b2007 1793 }
8929ecfa 1794 if (transid == trans->transid) {
79787eaa 1795 btrfs_commit_transaction(trans, root);
8929ecfa
YZ		 1796 } else {
1797 btrfs_end_transaction(trans, root);
1798 }
a74a4b97
CM		 1799sleep:
1800 wake_up_process(root->fs_info->cleaner_kthread);
1801 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1802
a0acae0e 1803 if (!try_to_freeze()) {
a74a4b97 1804 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa 1805 if (!kthread_should_stop() &&
914b2007
JK		 1806 (!btrfs_transaction_blocked(root->fs_info) ||
1807 cannot_commit))
8929ecfa 1808 schedule_timeout(delay);
a74a4b97
CM		 1809 __set_current_state(TASK_RUNNING);
1810 }
1811 } while (!kthread_should_stop());
1812 return 0;
1813}
1814
af31f5e5
CM		 1815/*
1816 * this will find the highest generation in the array of
1817 * root backups. The index of the highest array is returned,
1818 * or -1 if we can't find anything.
1819 *
1820 * We check to make sure the array is valid by comparing the
1821 * generation of the latest root in the array with the generation
1822 * in the super block. If they don't match we pitch it.
1823 */
1824static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1825{
1826 u64 cur;
1827 int newest_index = -1;
1828 struct btrfs_root_backup *root_backup;
1829 int i;
1830
1831 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1832 root_backup = info->super_copy->super_roots + i;
1833 cur = btrfs_backup_tree_root_gen(root_backup);
1834 if (cur == newest_gen)
1835 newest_index = i;
1836 }
1837
1838 /* check to see if we actually wrapped around */
1839 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1840 root_backup = info->super_copy->super_roots;
1841 cur = btrfs_backup_tree_root_gen(root_backup);
1842 if (cur == newest_gen)
1843 newest_index = 0;
1844 }
1845 return newest_index;
1846}
1847
1848
1849/*
1850 * find the oldest backup so we know where to store new entries
1851 * in the backup array. This will set the backup_root_index
1852 * field in the fs_info struct
1853 */
1854static void find_oldest_super_backup(struct btrfs_fs_info *info,
1855 u64 newest_gen)
1856{
1857 int newest_index = -1;
1858
1859 newest_index = find_newest_super_backup(info, newest_gen);
1860 /* if there was garbage in there, just move along */
1861 if (newest_index == -1) {
1862 info->backup_root_index = 0;
1863 } else {
1864 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1865 }
1866}
1867
1868/*
1869 * copy all the root pointers into the super backup array.
1870 * this will bump the backup pointer by one when it is
1871 * done
1872 */
1873static void backup_super_roots(struct btrfs_fs_info *info)
1874{
1875 int next_backup;
1876 struct btrfs_root_backup *root_backup;
1877 int last_backup;
1878
1879 next_backup = info->backup_root_index;
1880 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1881 BTRFS_NUM_BACKUP_ROOTS;
1882
1883 /*
1884 * just overwrite the last backup if we're at the same generation
1885 * this happens only at umount
1886 */
1887 root_backup = info->super_for_commit->super_roots + last_backup;
1888 if (btrfs_backup_tree_root_gen(root_backup) ==
1889 btrfs_header_generation(info->tree_root->node))
1890 next_backup = last_backup;
1891
1892 root_backup = info->super_for_commit->super_roots + next_backup;
1893
1894 /*
1895 * make sure all of our padding and empty slots get zero filled
1896 * regardless of which ones we use today
1897 */
1898 memset(root_backup, 0, sizeof(*root_backup));
1899
1900 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1901
1902 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1903 btrfs_set_backup_tree_root_gen(root_backup,
1904 btrfs_header_generation(info->tree_root->node));
1905
1906 btrfs_set_backup_tree_root_level(root_backup,
1907 btrfs_header_level(info->tree_root->node));
1908
1909 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1910 btrfs_set_backup_chunk_root_gen(root_backup,
1911 btrfs_header_generation(info->chunk_root->node));
1912 btrfs_set_backup_chunk_root_level(root_backup,
1913 btrfs_header_level(info->chunk_root->node));
1914
1915 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1916 btrfs_set_backup_extent_root_gen(root_backup,
1917 btrfs_header_generation(info->extent_root->node));
1918 btrfs_set_backup_extent_root_level(root_backup,
1919 btrfs_header_level(info->extent_root->node));
1920
7c7e82a7
CM		 1921 /*
1922 * we might commit during log recovery, which happens before we set
1923 * the fs_root. Make sure it is valid before we fill it in.
1924 */
1925 if (info->fs_root && info->fs_root->node) {
1926 btrfs_set_backup_fs_root(root_backup,
1927 info->fs_root->node->start);
1928 btrfs_set_backup_fs_root_gen(root_backup,
af31f5e5 1929 btrfs_header_generation(info->fs_root->node));
7c7e82a7 1930 btrfs_set_backup_fs_root_level(root_backup,
af31f5e5 1931 btrfs_header_level(info->fs_root->node));
7c7e82a7 1932 }
af31f5e5
CM		 1933
1934 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1935 btrfs_set_backup_dev_root_gen(root_backup,
1936 btrfs_header_generation(info->dev_root->node));
1937 btrfs_set_backup_dev_root_level(root_backup,
1938 btrfs_header_level(info->dev_root->node));
1939
1940 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1941 btrfs_set_backup_csum_root_gen(root_backup,
1942 btrfs_header_generation(info->csum_root->node));
1943 btrfs_set_backup_csum_root_level(root_backup,
1944 btrfs_header_level(info->csum_root->node));
1945
1946 btrfs_set_backup_total_bytes(root_backup,
1947 btrfs_super_total_bytes(info->super_copy));
1948 btrfs_set_backup_bytes_used(root_backup,
1949 btrfs_super_bytes_used(info->super_copy));
1950 btrfs_set_backup_num_devices(root_backup,
1951 btrfs_super_num_devices(info->super_copy));
1952
1953 /*
1954 * if we don't copy this out to the super_copy, it won't get remembered
1955 * for the next commit
1956 */
1957 memcpy(&info->super_copy->super_roots,
1958 &info->super_for_commit->super_roots,
1959 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1960}
1961
1962/*
1963 * this copies info out of the root backup array and back into
1964 * the in-memory super block. It is meant to help iterate through
1965 * the array, so you send it the number of backups you've already
1966 * tried and the last backup index you used.
1967 *
1968 * this returns -1 when it has tried all the backups
1969 */
1970static noinline int next_root_backup(struct btrfs_fs_info *info,
1971 struct btrfs_super_block *super,
1972 int *num_backups_tried, int *backup_index)
1973{
1974 struct btrfs_root_backup *root_backup;
1975 int newest = *backup_index;
1976
1977 if (*num_backups_tried == 0) {
1978 u64 gen = btrfs_super_generation(super);
1979
1980 newest = find_newest_super_backup(info, gen);
1981 if (newest == -1)
1982 return -1;
1983
1984 *backup_index = newest;
1985 *num_backups_tried = 1;
1986 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1987 /* we've tried all the backups, all done */
1988 return -1;
1989 } else {
1990 /* jump to the next oldest backup */
1991 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1992 BTRFS_NUM_BACKUP_ROOTS;
1993 *backup_index = newest;
1994 *num_backups_tried += 1;
1995 }
1996 root_backup = super->super_roots + newest;
1997
1998 btrfs_set_super_generation(super,
1999 btrfs_backup_tree_root_gen(root_backup));
2000 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2001 btrfs_set_super_root_level(super,
2002 btrfs_backup_tree_root_level(root_backup));
2003 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2004
2005 /*
2006 * fixme: the total bytes and num_devices need to match or we should
2007 * need a fsck
2008 */
2009 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2010 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2011 return 0;
2012}
2013
7abadb64
LB		 2014/* helper to cleanup workers */
2015static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2016{
2017 btrfs_stop_workers(&fs_info->generic_worker);
2018 btrfs_stop_workers(&fs_info->fixup_workers);
2019 btrfs_stop_workers(&fs_info->delalloc_workers);
2020 btrfs_stop_workers(&fs_info->workers);
2021 btrfs_stop_workers(&fs_info->endio_workers);
2022 btrfs_stop_workers(&fs_info->endio_meta_workers);
2023 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2024 btrfs_stop_workers(&fs_info->rmw_workers);
2025 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2026 btrfs_stop_workers(&fs_info->endio_write_workers);
2027 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2028 btrfs_stop_workers(&fs_info->submit_workers);
2029 btrfs_stop_workers(&fs_info->delayed_workers);
2030 btrfs_stop_workers(&fs_info->caching_workers);
2031 btrfs_stop_workers(&fs_info->readahead_workers);
2032 btrfs_stop_workers(&fs_info->flush_workers);
2f232036 2033 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
7abadb64
LB		 2034}
2035
af31f5e5
CM		 2036/* helper to cleanup tree roots */
2037static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2038{
2039 free_extent_buffer(info->tree_root->node);
2040 free_extent_buffer(info->tree_root->commit_root);
af31f5e5
CM		 2041 info->tree_root->node = NULL;
2042 info->tree_root->commit_root = NULL;
655b09fe
JB		 2043
2044 if (info->dev_root) {
2045 free_extent_buffer(info->dev_root->node);
2046 free_extent_buffer(info->dev_root->commit_root);
2047 info->dev_root->node = NULL;
2048 info->dev_root->commit_root = NULL;
2049 }
2050 if (info->extent_root) {
2051 free_extent_buffer(info->extent_root->node);
2052 free_extent_buffer(info->extent_root->commit_root);
2053 info->extent_root->node = NULL;
2054 info->extent_root->commit_root = NULL;
2055 }
2056 if (info->csum_root) {
2057 free_extent_buffer(info->csum_root->node);
2058 free_extent_buffer(info->csum_root->commit_root);
2059 info->csum_root->node = NULL;
2060 info->csum_root->commit_root = NULL;
2061 }
bcef60f2 2062 if (info->quota_root) {
655b09fe
JB		 2063 free_extent_buffer(info->quota_root->node);
2064 free_extent_buffer(info->quota_root->commit_root);
bcef60f2
AJ		 2065 info->quota_root->node = NULL;
2066 info->quota_root->commit_root = NULL;
2067 }
f7a81ea4
SB		 2068 if (info->uuid_root) {
2069 free_extent_buffer(info->uuid_root->node);
2070 free_extent_buffer(info->uuid_root->commit_root);
2071 info->uuid_root->node = NULL;
2072 info->uuid_root->commit_root = NULL;
2073 }
af31f5e5
CM		 2074 if (chunk_root) {
2075 free_extent_buffer(info->chunk_root->node);
2076 free_extent_buffer(info->chunk_root->commit_root);
2077 info->chunk_root->node = NULL;
2078 info->chunk_root->commit_root = NULL;
2079 }
2080}
2081
171f6537
JB		 2082static void del_fs_roots(struct btrfs_fs_info *fs_info)
2083{
2084 int ret;
2085 struct btrfs_root *gang[8];
2086 int i;
2087
2088 while (!list_empty(&fs_info->dead_roots)) {
2089 gang[0] = list_entry(fs_info->dead_roots.next,
2090 struct btrfs_root, root_list);
2091 list_del(&gang[0]->root_list);
2092
2093 if (gang[0]->in_radix) {
cb517eab 2094 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
171f6537
JB		 2095 } else {
2096 free_extent_buffer(gang[0]->node);
2097 free_extent_buffer(gang[0]->commit_root);
b0feb9d9 2098 btrfs_put_fs_root(gang[0]);
171f6537
JB		 2099 }
2100 }
2101
2102 while (1) {
2103 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2104 (void **)gang, 0,
2105 ARRAY_SIZE(gang));
2106 if (!ret)
2107 break;
2108 for (i = 0; i < ret; i++)
cb517eab 2109 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
171f6537
JB		 2110 }
2111}
af31f5e5 2112
ad2b2c80
AV		 2113int open_ctree(struct super_block *sb,
2114 struct btrfs_fs_devices *fs_devices,
2115 char *options)
2e635a27 2116{
db94535d
CM		 2117 u32 sectorsize;
2118 u32 nodesize;
2119 u32 leafsize;
2120 u32 blocksize;
87ee04eb 2121 u32 stripesize;
84234f3a 2122 u64 generation;
f2b636e8 2123 u64 features;
3de4586c 2124 struct btrfs_key location;
a061fc8d 2125 struct buffer_head *bh;
4d34b278 2126 struct btrfs_super_block *disk_super;
815745cf 2127 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
f84a8bd6 2128 struct btrfs_root *tree_root;
4d34b278
ID		 2129 struct btrfs_root *extent_root;
2130 struct btrfs_root *csum_root;
2131 struct btrfs_root *chunk_root;
2132 struct btrfs_root *dev_root;
bcef60f2 2133 struct btrfs_root *quota_root;
f7a81ea4 2134 struct btrfs_root *uuid_root;
e02119d5 2135 struct btrfs_root *log_tree_root;
eb60ceac 2136 int ret;
e58ca020 2137 int err = -EINVAL;
af31f5e5
CM		 2138 int num_backups_tried = 0;
2139 int backup_index = 0;
70f80175
SB		 2140 bool create_uuid_tree;
2141 bool check_uuid_tree;
4543df7e 2142
f84a8bd6 2143 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
6f07e42e 2144 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
cb517eab 2145 if (!tree_root || !chunk_root) {
39279cc3
CM		 2146 err = -ENOMEM;
2147 goto fail;
2148 }
76dda93c
YZ		 2149
2150 ret = init_srcu_struct(&fs_info->subvol_srcu);
2151 if (ret) {
2152 err = ret;
2153 goto fail;
2154 }
2155
2156 ret = setup_bdi(fs_info, &fs_info->bdi);
2157 if (ret) {
2158 err = ret;
2159 goto fail_srcu;
2160 }
2161
e2d84521
MX		 2162 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2163 if (ret) {
2164 err = ret;
2165 goto fail_bdi;
2166 }
2167 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2168 (1 + ilog2(nr_cpu_ids));
2169
963d678b
MX		 2170 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2171 if (ret) {
2172 err = ret;
2173 goto fail_dirty_metadata_bytes;
2174 }
2175
76dda93c
YZ		 2176 fs_info->btree_inode = new_inode(sb);
2177 if (!fs_info->btree_inode) {
2178 err = -ENOMEM;
963d678b 2179 goto fail_delalloc_bytes;
76dda93c
YZ		 2180 }
2181
a6591715 2182 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1561deda 2183
76dda93c 2184 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 2185 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 2186 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 2187 INIT_LIST_HEAD(&fs_info->delayed_iputs);
eb73c1b7 2188 INIT_LIST_HEAD(&fs_info->delalloc_roots);
11833d66 2189 INIT_LIST_HEAD(&fs_info->caching_block_groups);
eb73c1b7 2190 spin_lock_init(&fs_info->delalloc_root_lock);
a4abeea4 2191 spin_lock_init(&fs_info->trans_lock);
76dda93c 2192 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 2193 spin_lock_init(&fs_info->delayed_iput_lock);
4cb5300b 2194 spin_lock_init(&fs_info->defrag_inodes_lock);
2bf64758 2195 spin_lock_init(&fs_info->free_chunk_lock);
f29021b2 2196 spin_lock_init(&fs_info->tree_mod_seq_lock);
ceda0864 2197 spin_lock_init(&fs_info->super_lock);
f29021b2 2198 rwlock_init(&fs_info->tree_mod_log_lock);
7585717f 2199 mutex_init(&fs_info->reloc_mutex);
de98ced9 2200 seqlock_init(&fs_info->profiles_lock);
19c00ddc 2201
58176a96 2202 init_completion(&fs_info->kobj_unregister);
0b86a832 2203 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 2204 INIT_LIST_HEAD(&fs_info->space_info);
f29021b2 2205 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
0b86a832 2206 btrfs_mapping_init(&fs_info->mapping_tree);
66d8f3dd
MX		 2207 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2208 BTRFS_BLOCK_RSV_GLOBAL);
2209 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2210 BTRFS_BLOCK_RSV_DELALLOC);
2211 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2212 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2213 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2214 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2215 BTRFS_BLOCK_RSV_DELOPS);
cb03c743 2216 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 2217 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 2218 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 2219 atomic_set(&fs_info->nr_async_bios, 0);
4cb5300b 2220 atomic_set(&fs_info->defrag_running, 0);
fc36ed7e 2221 atomic64_set(&fs_info->tree_mod_seq, 0);
e20d96d6 2222 fs_info->sb = sb;
6f568d35 2223 fs_info->max_inline = 8192 * 1024;
9ed74f2d 2224 fs_info->metadata_ratio = 0;
4cb5300b 2225 fs_info->defrag_inodes = RB_ROOT;
2bf64758 2226 fs_info->free_chunk_space = 0;
f29021b2 2227 fs_info->tree_mod_log = RB_ROOT;
8b87dc17 2228 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
c8b97818 2229
90519d66
AJ		 2230 /* readahead state */
2231 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2232 spin_lock_init(&fs_info->reada_lock);
c8b97818 2233
b34b086c
CM		 2234 fs_info->thread_pool_size = min_t(unsigned long,
2235 num_online_cpus() + 2, 8);
0afbaf8c 2236
199c2a9c
MX		 2237 INIT_LIST_HEAD(&fs_info->ordered_roots);
2238 spin_lock_init(&fs_info->ordered_root_lock);
16cdcec7
MX		 2239 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2240 GFP_NOFS);
2241 if (!fs_info->delayed_root) {
2242 err = -ENOMEM;
2243 goto fail_iput;
2244 }
2245 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 2246
a2de733c
AJ		 2247 mutex_init(&fs_info->scrub_lock);
2248 atomic_set(&fs_info->scrubs_running, 0);
2249 atomic_set(&fs_info->scrub_pause_req, 0);
2250 atomic_set(&fs_info->scrubs_paused, 0);
2251 atomic_set(&fs_info->scrub_cancel_req, 0);
2252 init_waitqueue_head(&fs_info->scrub_pause_wait);
2253 init_rwsem(&fs_info->scrub_super_lock);
2254 fs_info->scrub_workers_refcnt = 0;
21adbd5c
SB		 2255#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2256 fs_info->check_integrity_print_mask = 0;
2257#endif
a2de733c 2258
c9e9f97b
ID		 2259 spin_lock_init(&fs_info->balance_lock);
2260 mutex_init(&fs_info->balance_mutex);
837d5b6e
ID		 2261 atomic_set(&fs_info->balance_running, 0);
2262 atomic_set(&fs_info->balance_pause_req, 0);
a7e99c69 2263 atomic_set(&fs_info->balance_cancel_req, 0);
c9e9f97b 2264 fs_info->balance_ctl = NULL;
837d5b6e 2265 init_waitqueue_head(&fs_info->balance_wait_q);
a2de733c 2266
a061fc8d
CM		 2267 sb->s_blocksize = 4096;
2268 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 2269 sb->s_bdi = &fs_info->bdi;
a061fc8d 2270
76dda93c 2271 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
bfe86848 2272 set_nlink(fs_info->btree_inode, 1);
0afbaf8c
CM		 2273 /*
2274 * we set the i_size on the btree inode to the max possible int.
2275 * the real end of the address space is determined by all of
2276 * the devices in the system
2277 */
2278 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 2279 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 2280 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2281
5d4f98a2 2282 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 2283 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 2284 fs_info->btree_inode->i_mapping);
0b32f4bb 2285 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
a8067e02 2286 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM		 2287
2288 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 2289
76dda93c
YZ		 2290 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2291 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2292 sizeof(struct btrfs_key));
72ac3c0d
JB		 2293 set_bit(BTRFS_INODE_DUMMY,
2294 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
c65ddb52 2295 insert_inode_hash(fs_info->btree_inode);
76dda93c 2296
0f9dd46c 2297 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 2298 fs_info->block_group_cache_tree = RB_ROOT;
a1897fdd 2299 fs_info->first_logical_byte = (u64)-1;
0f9dd46c 2300
11833d66 2301 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 2302 fs_info->btree_inode->i_mapping);
11833d66 2303 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 2304 fs_info->btree_inode->i_mapping);
11833d66 2305 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 2306 fs_info->do_barriers = 1;
e18e4809 2307
39279cc3 2308
5a3f23d5 2309 mutex_init(&fs_info->ordered_operations_mutex);
9ffba8cd 2310 mutex_init(&fs_info->ordered_extent_flush_mutex);
e02119d5 2311 mutex_init(&fs_info->tree_log_mutex);
925baedd 2312 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 2313 mutex_init(&fs_info->transaction_kthread_mutex);
2314 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 2315 mutex_init(&fs_info->volume_mutex);
276e680d 2316 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 2317 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 2318 init_rwsem(&fs_info->subvol_sem);
803b2f54 2319 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
e922e087
SB		 2320 fs_info->dev_replace.lock_owner = 0;
2321 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2322 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2323 mutex_init(&fs_info->dev_replace.lock_management_lock);
2324 mutex_init(&fs_info->dev_replace.lock);
fa9c0d79 2325
416ac51d 2326 spin_lock_init(&fs_info->qgroup_lock);
f2f6ed3d 2327 mutex_init(&fs_info->qgroup_ioctl_lock);
416ac51d
AJ		 2328 fs_info->qgroup_tree = RB_ROOT;
2329 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2330 fs_info->qgroup_seq = 1;
2331 fs_info->quota_enabled = 0;
2332 fs_info->pending_quota_state = 0;
1e8f9158 2333 fs_info->qgroup_ulist = NULL;
2f232036 2334 mutex_init(&fs_info->qgroup_rescan_lock);
416ac51d 2335
fa9c0d79
CM		 2336 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2337 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2338
e6dcd2dc 2339 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 2340 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 2341 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 2342 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 2343
53b381b3
DW		 2344 ret = btrfs_alloc_stripe_hash_table(fs_info);
2345 if (ret) {
83c8266a 2346 err = ret;
53b381b3
DW		 2347 goto fail_alloc;
2348 }
2349
0b86a832 2350 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 2351 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 2352
3c4bb26b 2353 invalidate_bdev(fs_devices->latest_bdev);
1104a885
DS		 2354
2355 /*
2356 * Read super block and check the signature bytes only
2357 */
a512bbf8 2358 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 2359 if (!bh) {
2360 err = -EINVAL;
16cdcec7 2361 goto fail_alloc;
20b45077 2362 }
39279cc3 2363
1104a885
DS		 2364 /*
2365 * We want to check superblock checksum, the type is stored inside.
2366 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2367 */
2368 if (btrfs_check_super_csum(bh->b_data)) {
2369 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2370 err = -EINVAL;
2371 goto fail_alloc;
2372 }
2373
2374 /*
2375 * super_copy is zeroed at allocation time and we never touch the
2376 * following bytes up to INFO_SIZE, the checksum is calculated from
2377 * the whole block of INFO_SIZE
2378 */
6c41761f
DS		 2379 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2380 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2381 sizeof(*fs_info->super_for_commit));
a061fc8d 2382 brelse(bh);
5f39d397 2383
6c41761f 2384 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
0b86a832 2385
1104a885
DS		 2386 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2387 if (ret) {
2388 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2389 err = -EINVAL;
2390 goto fail_alloc;
2391 }
2392
6c41761f 2393 disk_super = fs_info->super_copy;
0f7d52f4 2394 if (!btrfs_super_root(disk_super))
16cdcec7 2395 goto fail_alloc;
0f7d52f4 2396
acce952b 2397 /* check FS state, whether FS is broken. */
87533c47
MX		 2398 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2399 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
acce952b 2400
af31f5e5
CM		 2401 /*
2402 * run through our array of backup supers and setup
2403 * our ring pointer to the oldest one
2404 */
2405 generation = btrfs_super_generation(disk_super);
2406 find_oldest_super_backup(fs_info, generation);
2407
75e7cb7f
LB		 2408 /*
2409 * In the long term, we'll store the compression type in the super
2410 * block, and it'll be used for per file compression control.
2411 */
2412 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2413
2b82032c
YZ		 2414 ret = btrfs_parse_options(tree_root, options);
2415 if (ret) {
2416 err = ret;
16cdcec7 2417 goto fail_alloc;
2b82032c 2418 }
dfe25020 2419
f2b636e8
JB		 2420 features = btrfs_super_incompat_flags(disk_super) &
2421 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2422 if (features) {
2423 printk(KERN_ERR "BTRFS: couldn't mount because of "
2424 "unsupported optional features (%Lx).\n",
c1c9ff7c 2425 features);
f2b636e8 2426 err = -EINVAL;
16cdcec7 2427 goto fail_alloc;
f2b636e8
JB		 2428 }
2429
727011e0
CM		 2430 if (btrfs_super_leafsize(disk_super) !=
2431 btrfs_super_nodesize(disk_super)) {
2432 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2433 "blocksizes don't match. node %d leaf %d\n",
2434 btrfs_super_nodesize(disk_super),
2435 btrfs_super_leafsize(disk_super));
2436 err = -EINVAL;
2437 goto fail_alloc;
2438 }
2439 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2440 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2441 "blocksize (%d) was too large\n",
2442 btrfs_super_leafsize(disk_super));
2443 err = -EINVAL;
2444 goto fail_alloc;
2445 }
2446
5d4f98a2 2447 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae 2448 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
69e380d1 2449 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
a6fa6fae 2450 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
727011e0 2451
3173a18f
JB		 2452 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2453 printk(KERN_ERR "btrfs: has skinny extents\n");
2454
727011e0
CM		 2455 /*
2456 * flag our filesystem as having big metadata blocks if
2457 * they are bigger than the page size
2458 */
2459 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2460 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2461 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2462 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2463 }
2464
bc3f116f
CM		 2465 nodesize = btrfs_super_nodesize(disk_super);
2466 leafsize = btrfs_super_leafsize(disk_super);
2467 sectorsize = btrfs_super_sectorsize(disk_super);
2468 stripesize = btrfs_super_stripesize(disk_super);
e2d84521 2469 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
963d678b 2470 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
bc3f116f
CM		 2471
2472 /*
2473 * mixed block groups end up with duplicate but slightly offset
2474 * extent buffers for the same range. It leads to corruptions
2475 */
2476 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2477 (sectorsize != leafsize)) {
2478 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2479 "are not allowed for mixed block groups on %s\n",
2480 sb->s_id);
2481 goto fail_alloc;
2482 }
2483
ceda0864
MX		 2484 /*
2485 * Needn't use the lock because there is no other task which will
2486 * update the flag.
2487 */
a6fa6fae 2488 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 2489
f2b636e8
JB		 2490 features = btrfs_super_compat_ro_flags(disk_super) &
2491 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2492 if (!(sb->s_flags & MS_RDONLY) && features) {
2493 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2494 "unsupported option features (%Lx).\n",
c1c9ff7c 2495 features);
f2b636e8 2496 err = -EINVAL;
16cdcec7 2497 goto fail_alloc;
f2b636e8 2498 }
61d92c32
CM		 2499
2500 btrfs_init_workers(&fs_info->generic_worker,
2501 "genwork", 1, NULL);
2502
5443be45 2503 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 2504 fs_info->thread_pool_size,
2505 &fs_info->generic_worker);
c8b97818 2506
771ed689 2507 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
45d5fd14 2508 fs_info->thread_pool_size, NULL);
771ed689 2509
8ccf6f19 2510 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
45d5fd14 2511 fs_info->thread_pool_size, NULL);
8ccf6f19 2512
5443be45 2513 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 2514 min_t(u64, fs_devices->num_devices,
45d5fd14 2515 fs_info->thread_pool_size), NULL);
61b49440 2516
bab39bf9 2517 btrfs_init_workers(&fs_info->caching_workers, "cache",
45d5fd14 2518 fs_info->thread_pool_size, NULL);
bab39bf9 2519
61b49440
CM		 2520 /* a higher idle thresh on the submit workers makes it much more
2521 * likely that bios will be send down in a sane order to the
2522 * devices
2523 */
2524 fs_info->submit_workers.idle_thresh = 64;
53863232 2525
771ed689 2526 fs_info->workers.idle_thresh = 16;
4a69a410 2527 fs_info->workers.ordered = 1;
61b49440 2528
771ed689
CM		 2529 fs_info->delalloc_workers.idle_thresh = 2;
2530 fs_info->delalloc_workers.ordered = 1;
2531
61d92c32
CM		 2532 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2533 &fs_info->generic_worker);
5443be45 2534 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 2535 fs_info->thread_pool_size,
2536 &fs_info->generic_worker);
d20f7043 2537 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 2538 fs_info->thread_pool_size,
2539 &fs_info->generic_worker);
cad321ad 2540 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 2541 "endio-meta-write", fs_info->thread_pool_size,
2542 &fs_info->generic_worker);
53b381b3
DW		 2543 btrfs_init_workers(&fs_info->endio_raid56_workers,
2544 "endio-raid56", fs_info->thread_pool_size,
2545 &fs_info->generic_worker);
2546 btrfs_init_workers(&fs_info->rmw_workers,
2547 "rmw", fs_info->thread_pool_size,
2548 &fs_info->generic_worker);
5443be45 2549 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 2550 fs_info->thread_pool_size,
2551 &fs_info->generic_worker);
0cb59c99
JB		 2552 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2553 1, &fs_info->generic_worker);
16cdcec7
MX		 2554 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2555 fs_info->thread_pool_size,
2556 &fs_info->generic_worker);
90519d66
AJ		 2557 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2558 fs_info->thread_pool_size,
2559 &fs_info->generic_worker);
2f232036
JS		 2560 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2561 &fs_info->generic_worker);
61b49440
CM		 2562
2563 /*
2564 * endios are largely parallel and should have a very
2565 * low idle thresh
2566 */
2567 fs_info->endio_workers.idle_thresh = 4;
b51912c9 2568 fs_info->endio_meta_workers.idle_thresh = 4;
53b381b3
DW		 2569 fs_info->endio_raid56_workers.idle_thresh = 4;
2570 fs_info->rmw_workers.idle_thresh = 2;
b51912c9 2571
9042846b
CM		 2572 fs_info->endio_write_workers.idle_thresh = 2;
2573 fs_info->endio_meta_write_workers.idle_thresh = 2;
90519d66 2574 fs_info->readahead_workers.idle_thresh = 2;
9042846b 2575
0dc3b84a
JB		 2576 /*
2577 * btrfs_start_workers can really only fail because of ENOMEM so just
2578 * return -ENOMEM if any of these fail.
2579 */
2580 ret = btrfs_start_workers(&fs_info->workers);
2581 ret |= btrfs_start_workers(&fs_info->generic_worker);
2582 ret |= btrfs_start_workers(&fs_info->submit_workers);
2583 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2584 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2585 ret |= btrfs_start_workers(&fs_info->endio_workers);
2586 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
53b381b3
DW		 2587 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2588 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
0dc3b84a
JB		 2589 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2590 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2591 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2592 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2593 ret |= btrfs_start_workers(&fs_info->caching_workers);
2594 ret |= btrfs_start_workers(&fs_info->readahead_workers);
8ccf6f19 2595 ret |= btrfs_start_workers(&fs_info->flush_workers);
2f232036 2596 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
0dc3b84a 2597 if (ret) {
fed425c7 2598 err = -ENOMEM;
0dc3b84a
JB		 2599 goto fail_sb_buffer;
2600 }
4543df7e 2601
4575c9cc 2602 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 2603 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2604 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 2605
db94535d
CM		 2606 tree_root->nodesize = nodesize;
2607 tree_root->leafsize = leafsize;
2608 tree_root->sectorsize = sectorsize;
87ee04eb 2609 tree_root->stripesize = stripesize;
a061fc8d
CM		 2610
2611 sb->s_blocksize = sectorsize;
2612 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 2613
3cae210f 2614 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
d397712b 2615 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 2616 goto fail_sb_buffer;
2617 }
19c00ddc 2618
8d082fb7
LB		 2619 if (sectorsize != PAGE_SIZE) {
2620 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2621 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
941b2ddf
KM		 2622 goto fail_sb_buffer;
2623 }
2624
925baedd 2625 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 2626 ret = btrfs_read_sys_array(tree_root);
925baedd 2627 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 2628 if (ret) {
d397712b
CM		 2629 printk(KERN_WARNING "btrfs: failed to read the system "
2630 "array on %s\n", sb->s_id);
5d4f98a2 2631 goto fail_sb_buffer;
84eed90f 2632 }
0b86a832
CM		 2633
2634 blocksize = btrfs_level_size(tree_root,
2635 btrfs_super_chunk_root_level(disk_super));
84234f3a 2636 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 2637
2638 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2639 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2640
2641 chunk_root->node = read_tree_block(chunk_root,
2642 btrfs_super_chunk_root(disk_super),
84234f3a 2643 blocksize, generation);
416bc658
JB		 2644 if (!chunk_root->node ||
2645 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
83121942
DW		 2646 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2647 sb->s_id);
af31f5e5 2648 goto fail_tree_roots;
83121942 2649 }
5d4f98a2
YZ		 2650 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2651 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2652
e17cade2 2653 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
b308bc2f 2654 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
e17cade2 2655
0b86a832 2656 ret = btrfs_read_chunk_tree(chunk_root);
2b82032c 2657 if (ret) {
d397712b
CM		 2658 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2659 sb->s_id);
af31f5e5 2660 goto fail_tree_roots;
2b82032c 2661 }
0b86a832 2662
8dabb742
SB		 2663 /*
2664 * keep the device that is marked to be the target device for the
2665 * dev_replace procedure
2666 */
2667 btrfs_close_extra_devices(fs_info, fs_devices, 0);
dfe25020 2668
a6b0d5c8
CM		 2669 if (!fs_devices->latest_bdev) {
2670 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2671 sb->s_id);
2672 goto fail_tree_roots;
2673 }
2674
af31f5e5 2675retry_root_backup:
db94535d
CM		 2676 blocksize = btrfs_level_size(tree_root,
2677 btrfs_super_root_level(disk_super));
84234f3a 2678 generation = btrfs_super_generation(disk_super);
0b86a832 2679
e20d96d6 2680 tree_root->node = read_tree_block(tree_root,
db94535d 2681 btrfs_super_root(disk_super),
84234f3a 2682 blocksize, generation);
af31f5e5
CM		 2683 if (!tree_root->node ||
2684 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
83121942
DW		 2685 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2686 sb->s_id);
af31f5e5
CM		 2687
2688 goto recovery_tree_root;
83121942 2689 }
af31f5e5 2690
5d4f98a2
YZ		 2691 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2692 tree_root->commit_root = btrfs_root_node(tree_root);
69e9c6c6 2693 btrfs_set_root_refs(&tree_root->root_item, 1);
db94535d 2694
cb517eab
MX		 2695 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2696 location.type = BTRFS_ROOT_ITEM_KEY;
2697 location.offset = 0;
2698
2699 extent_root = btrfs_read_tree_root(tree_root, &location);
2700 if (IS_ERR(extent_root)) {
2701 ret = PTR_ERR(extent_root);
af31f5e5 2702 goto recovery_tree_root;
cb517eab 2703 }
0b86a832 2704 extent_root->track_dirty = 1;
cb517eab 2705 fs_info->extent_root = extent_root;
0b86a832 2706
cb517eab
MX		 2707 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2708 dev_root = btrfs_read_tree_root(tree_root, &location);
2709 if (IS_ERR(dev_root)) {
2710 ret = PTR_ERR(dev_root);
af31f5e5 2711 goto recovery_tree_root;
cb517eab 2712 }
5d4f98a2 2713 dev_root->track_dirty = 1;
cb517eab
MX		 2714 fs_info->dev_root = dev_root;
2715 btrfs_init_devices_late(fs_info);
3768f368 2716
cb517eab
MX		 2717 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2718 csum_root = btrfs_read_tree_root(tree_root, &location);
2719 if (IS_ERR(csum_root)) {
2720 ret = PTR_ERR(csum_root);
af31f5e5 2721 goto recovery_tree_root;
cb517eab 2722 }
d20f7043 2723 csum_root->track_dirty = 1;
cb517eab 2724 fs_info->csum_root = csum_root;
d20f7043 2725
cb517eab
MX		 2726 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2727 quota_root = btrfs_read_tree_root(tree_root, &location);
2728 if (!IS_ERR(quota_root)) {
bcef60f2
AJ		 2729 quota_root->track_dirty = 1;
2730 fs_info->quota_enabled = 1;
2731 fs_info->pending_quota_state = 1;
cb517eab 2732 fs_info->quota_root = quota_root;
bcef60f2
AJ		 2733 }
2734
f7a81ea4
SB		 2735 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2736 uuid_root = btrfs_read_tree_root(tree_root, &location);
2737 if (IS_ERR(uuid_root)) {
2738 ret = PTR_ERR(uuid_root);
2739 if (ret != -ENOENT)
2740 goto recovery_tree_root;
2741 create_uuid_tree = true;
70f80175 2742 check_uuid_tree = false;
f7a81ea4
SB		 2743 } else {
2744 uuid_root->track_dirty = 1;
2745 fs_info->uuid_root = uuid_root;
70f80175
SB		 2746 create_uuid_tree = false;
2747 check_uuid_tree =
2748 generation != btrfs_super_uuid_tree_generation(disk_super);
f7a81ea4
SB		 2749 }
2750
8929ecfa
YZ		 2751 fs_info->generation = generation;
2752 fs_info->last_trans_committed = generation;
8929ecfa 2753
68310a5e
ID		 2754 ret = btrfs_recover_balance(fs_info);
2755 if (ret) {
2756 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2757 goto fail_block_groups;
2758 }
2759
733f4fbb
SB		 2760 ret = btrfs_init_dev_stats(fs_info);
2761 if (ret) {
2762 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2763 ret);
2764 goto fail_block_groups;
2765 }
2766
8dabb742
SB		 2767 ret = btrfs_init_dev_replace(fs_info);
2768 if (ret) {
2769 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2770 goto fail_block_groups;
2771 }
2772
2773 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2774
c59021f8 2775 ret = btrfs_init_space_info(fs_info);
2776 if (ret) {
2777 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2778 goto fail_block_groups;
2779 }
2780
1b1d1f66
JB		 2781 ret = btrfs_read_block_groups(extent_root);
2782 if (ret) {
2783 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2784 goto fail_block_groups;
2785 }
5af3e8cc
SB		 2786 fs_info->num_tolerated_disk_barrier_failures =
2787 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
292fd7fc
SB		 2788 if (fs_info->fs_devices->missing_devices >
2789 fs_info->num_tolerated_disk_barrier_failures &&
2790 !(sb->s_flags & MS_RDONLY)) {
2791 printk(KERN_WARNING
2792 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2793 goto fail_block_groups;
2794 }
9078a3e1 2795
a74a4b97
CM		 2796 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2797 "btrfs-cleaner");
57506d50 2798 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2799 goto fail_block_groups;
a74a4b97
CM		 2800
2801 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2802 tree_root,
2803 "btrfs-transaction");
57506d50 2804 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2805 goto fail_cleaner;
a74a4b97 2806
c289811c
CM		 2807 if (!btrfs_test_opt(tree_root, SSD) &&
2808 !btrfs_test_opt(tree_root, NOSSD) &&
2809 !fs_info->fs_devices->rotating) {
2810 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2811 "mode\n");
2812 btrfs_set_opt(fs_info->mount_opt, SSD);
2813 }
2814
21adbd5c
SB		 2815#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2816 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2817 ret = btrfsic_mount(tree_root, fs_devices,
2818 btrfs_test_opt(tree_root,
2819 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2820 1 : 0,
2821 fs_info->check_integrity_print_mask);
2822 if (ret)
2823 printk(KERN_WARNING "btrfs: failed to initialize"
2824 " integrity check module %s\n", sb->s_id);
2825 }
2826#endif
bcef60f2
AJ		 2827 ret = btrfs_read_qgroup_config(fs_info);
2828 if (ret)
2829 goto fail_trans_kthread;
21adbd5c 2830
acce952b 2831 /* do not make disk changes in broken FS */
68ce9682 2832 if (btrfs_super_log_root(disk_super) != 0) {
e02119d5
CM		 2833 u64 bytenr = btrfs_super_log_root(disk_super);
2834
7c2ca468 2835 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2836 printk(KERN_WARNING "Btrfs log replay required "
2837 "on RO media\n");
7c2ca468 2838 err = -EIO;
bcef60f2 2839 goto fail_qgroup;
7c2ca468 2840 }
e02119d5
CM		 2841 blocksize =
2842 btrfs_level_size(tree_root,
2843 btrfs_super_log_root_level(disk_super));
d18a2c44 2844
6f07e42e 2845 log_tree_root = btrfs_alloc_root(fs_info);
676e4c86
DC		 2846 if (!log_tree_root) {
2847 err = -ENOMEM;
bcef60f2 2848 goto fail_qgroup;
676e4c86 2849 }
e02119d5
CM		 2850
2851 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2852 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2853
2854 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2855 blocksize,
2856 generation + 1);
416bc658
JB		 2857 if (!log_tree_root->node ||
2858 !extent_buffer_uptodate(log_tree_root->node)) {
2859 printk(KERN_ERR "btrfs: failed to read log tree\n");
2860 free_extent_buffer(log_tree_root->node);
2861 kfree(log_tree_root);
2862 goto fail_trans_kthread;
2863 }
79787eaa 2864 /* returns with log_tree_root freed on success */
e02119d5 2865 ret = btrfs_recover_log_trees(log_tree_root);
79787eaa
JM		 2866 if (ret) {
2867 btrfs_error(tree_root->fs_info, ret,
2868 "Failed to recover log tree");
2869 free_extent_buffer(log_tree_root->node);
2870 kfree(log_tree_root);
2871 goto fail_trans_kthread;
2872 }
e556ce2c
YZ		 2873
2874 if (sb->s_flags & MS_RDONLY) {
79787eaa
JM		 2875 ret = btrfs_commit_super(tree_root);
2876 if (ret)
2877 goto fail_trans_kthread;
e556ce2c 2878 }
e02119d5 2879 }
1a40e23b 2880
76dda93c 2881 ret = btrfs_find_orphan_roots(tree_root);
79787eaa
JM		 2882 if (ret)
2883 goto fail_trans_kthread;
76dda93c 2884
7c2ca468 2885 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b 2886 ret = btrfs_cleanup_fs_roots(fs_info);
44c44af2
ID		 2887 if (ret)
2888 goto fail_trans_kthread;
d68fc57b 2889
5d4f98a2 2890 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2891 if (ret < 0) {
2892 printk(KERN_WARNING
2893 "btrfs: failed to recover relocation\n");
2894 err = -EINVAL;
bcef60f2 2895 goto fail_qgroup;
d7ce5843 2896 }
7c2ca468 2897 }
1a40e23b 2898
3de4586c
CM		 2899 location.objectid = BTRFS_FS_TREE_OBJECTID;
2900 location.type = BTRFS_ROOT_ITEM_KEY;
cb517eab 2901 location.offset = 0;
3de4586c 2902
3de4586c 2903 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3140c9a3
DC		 2904 if (IS_ERR(fs_info->fs_root)) {
2905 err = PTR_ERR(fs_info->fs_root);
bcef60f2 2906 goto fail_qgroup;
3140c9a3 2907 }
c289811c 2908
2b6ba629
ID		 2909 if (sb->s_flags & MS_RDONLY)
2910 return 0;
59641015 2911
2b6ba629
ID		 2912 down_read(&fs_info->cleanup_work_sem);
2913 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2914 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
e3acc2a6 2915 up_read(&fs_info->cleanup_work_sem);
2b6ba629
ID		 2916 close_ctree(tree_root);
2917 return ret;
2918 }
2919 up_read(&fs_info->cleanup_work_sem);
59641015 2920
2b6ba629
ID		 2921 ret = btrfs_resume_balance_async(fs_info);
2922 if (ret) {
2923 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2924 close_ctree(tree_root);
2925 return ret;
e3acc2a6
JB		 2926 }
2927
8dabb742
SB		 2928 ret = btrfs_resume_dev_replace_async(fs_info);
2929 if (ret) {
2930 pr_warn("btrfs: failed to resume dev_replace\n");
2931 close_ctree(tree_root);
2932 return ret;
2933 }
2934
b382a324
JS		 2935 btrfs_qgroup_rescan_resume(fs_info);
2936
f7a81ea4
SB		 2937 if (create_uuid_tree) {
2938 pr_info("btrfs: creating UUID tree\n");
2939 ret = btrfs_create_uuid_tree(fs_info);
2940 if (ret) {
2941 pr_warn("btrfs: failed to create the UUID tree %d\n",
2942 ret);
2943 close_ctree(tree_root);
2944 return ret;
2945 }
f420ee1e
SB		 2946 } else if (check_uuid_tree ||
2947 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
70f80175
SB		 2948 pr_info("btrfs: checking UUID tree\n");
2949 ret = btrfs_check_uuid_tree(fs_info);
2950 if (ret) {
2951 pr_warn("btrfs: failed to check the UUID tree %d\n",
2952 ret);
2953 close_ctree(tree_root);
2954 return ret;
2955 }
2956 } else {
2957 fs_info->update_uuid_tree_gen = 1;
f7a81ea4
SB		 2958 }
2959
ad2b2c80 2960 return 0;
39279cc3 2961
bcef60f2
AJ		 2962fail_qgroup:
2963 btrfs_free_qgroup_config(fs_info);
7c2ca468
CM		 2964fail_trans_kthread:
2965 kthread_stop(fs_info->transaction_kthread);
54067ae9 2966 btrfs_cleanup_transaction(fs_info->tree_root);
7b5ff90e 2967 del_fs_roots(fs_info);
3f157a2f 2968fail_cleaner:
a74a4b97 2969 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2970
2971 /*
2972 * make sure we're done with the btree inode before we stop our
2973 * kthreads
2974 */
2975 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
7c2ca468 2976
1b1d1f66 2977fail_block_groups:
54067ae9 2978 btrfs_put_block_group_cache(fs_info);
1b1d1f66 2979 btrfs_free_block_groups(fs_info);
af31f5e5
CM		 2980
2981fail_tree_roots:
2982 free_root_pointers(fs_info, 1);
2b8195bb 2983 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
af31f5e5 2984
39279cc3 2985fail_sb_buffer:
7abadb64 2986 btrfs_stop_all_workers(fs_info);
16cdcec7 2987fail_alloc:
4543df7e 2988fail_iput:
586e46e2
ID		 2989 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2990
4543df7e 2991 iput(fs_info->btree_inode);
963d678b
MX		 2992fail_delalloc_bytes:
2993 percpu_counter_destroy(&fs_info->delalloc_bytes);
e2d84521
MX		 2994fail_dirty_metadata_bytes:
2995 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
ad081f14 2996fail_bdi:
7e662854 2997 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2998fail_srcu:
2999 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 3000fail:
53b381b3 3001 btrfs_free_stripe_hash_table(fs_info);
586e46e2 3002 btrfs_close_devices(fs_info->fs_devices);
ad2b2c80 3003 return err;
af31f5e5
CM		 3004
3005recovery_tree_root:
af31f5e5
CM		 3006 if (!btrfs_test_opt(tree_root, RECOVERY))
3007 goto fail_tree_roots;
3008
3009 free_root_pointers(fs_info, 0);
3010
3011 /* don't use the log in recovery mode, it won't be valid */
3012 btrfs_set_super_log_root(disk_super, 0);
3013
3014 /* we can't trust the free space cache either */
3015 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3016
3017 ret = next_root_backup(fs_info, fs_info->super_copy,
3018 &num_backups_tried, &backup_index);
3019 if (ret == -1)
3020 goto fail_block_groups;
3021 goto retry_root_backup;
eb60ceac
CM		 3022}
3023
f2984462
CM		 3024static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3025{
f2984462
CM		 3026 if (uptodate) {
3027 set_buffer_uptodate(bh);
3028 } else {
442a4f63
SB		 3029 struct btrfs_device *device = (struct btrfs_device *)
3030 bh->b_private;
3031
606686ee
JB		 3032 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3033 "I/O error on %s\n",
3034 rcu_str_deref(device->name));
1259ab75
CM		 3035 /* note, we dont' set_buffer_write_io_error because we have
3036 * our own ways of dealing with the IO errors
3037 */
f2984462 3038 clear_buffer_uptodate(bh);
442a4f63 3039 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
f2984462
CM		 3040 }
3041 unlock_buffer(bh);
3042 put_bh(bh);
3043}
3044
a512bbf8
YZ		 3045struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3046{
3047 struct buffer_head *bh;
3048 struct buffer_head *latest = NULL;
3049 struct btrfs_super_block *super;
3050 int i;
3051 u64 transid = 0;
3052 u64 bytenr;
3053
3054 /* we would like to check all the supers, but that would make
3055 * a btrfs mount succeed after a mkfs from a different FS.
3056 * So, we need to add a special mount option to scan for
3057 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3058 */
3059 for (i = 0; i < 1; i++) {
3060 bytenr = btrfs_sb_offset(i);
8068a47e
AJ		 3061 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3062 i_size_read(bdev->bd_inode))
a512bbf8 3063 break;
8068a47e
AJ		 3064 bh = __bread(bdev, bytenr / 4096,
3065 BTRFS_SUPER_INFO_SIZE);
a512bbf8
YZ		 3066 if (!bh)
3067 continue;
3068
3069 super = (struct btrfs_super_block *)bh->b_data;
3070 if (btrfs_super_bytenr(super) != bytenr ||
3cae210f 3071 btrfs_super_magic(super) != BTRFS_MAGIC) {
a512bbf8
YZ		 3072 brelse(bh);
3073 continue;
3074 }
3075
3076 if (!latest || btrfs_super_generation(super) > transid) {
3077 brelse(latest);
3078 latest = bh;
3079 transid = btrfs_super_generation(super);
3080 } else {
3081 brelse(bh);
3082 }
3083 }
3084 return latest;
3085}
3086
4eedeb75
HH		 3087/*
3088 * this should be called twice, once with wait == 0 and
3089 * once with wait == 1. When wait == 0 is done, all the buffer heads
3090 * we write are pinned.
3091 *
3092 * They are released when wait == 1 is done.
3093 * max_mirrors must be the same for both runs, and it indicates how
3094 * many supers on this one device should be written.
3095 *
3096 * max_mirrors == 0 means to write them all.
3097 */
a512bbf8
YZ		 3098static int write_dev_supers(struct btrfs_device *device,
3099 struct btrfs_super_block *sb,
3100 int do_barriers, int wait, int max_mirrors)
3101{
3102 struct buffer_head *bh;
3103 int i;
3104 int ret;
3105 int errors = 0;
3106 u32 crc;
3107 u64 bytenr;
a512bbf8
YZ		 3108
3109 if (max_mirrors == 0)
3110 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3111
a512bbf8
YZ		 3112 for (i = 0; i < max_mirrors; i++) {
3113 bytenr = btrfs_sb_offset(i);
3114 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3115 break;
3116
3117 if (wait) {
3118 bh = __find_get_block(device->bdev, bytenr / 4096,
3119 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3120 if (!bh) {
3121 errors++;
3122 continue;
3123 }
a512bbf8 3124 wait_on_buffer(bh);
4eedeb75
HH		 3125 if (!buffer_uptodate(bh))
3126 errors++;
3127
3128 /* drop our reference */
3129 brelse(bh);
3130
3131 /* drop the reference from the wait == 0 run */
3132 brelse(bh);
3133 continue;
a512bbf8
YZ		 3134 } else {
3135 btrfs_set_super_bytenr(sb, bytenr);
3136
3137 crc = ~(u32)0;
b0496686 3138 crc = btrfs_csum_data((char *)sb +
a512bbf8
YZ		 3139 BTRFS_CSUM_SIZE, crc,
3140 BTRFS_SUPER_INFO_SIZE -
3141 BTRFS_CSUM_SIZE);
3142 btrfs_csum_final(crc, sb->csum);
3143
4eedeb75
HH		 3144 /*
3145 * one reference for us, and we leave it for the
3146 * caller
3147 */
a512bbf8
YZ		 3148 bh = __getblk(device->bdev, bytenr / 4096,
3149 BTRFS_SUPER_INFO_SIZE);
634554dc
JB		 3150 if (!bh) {
3151 printk(KERN_ERR "btrfs: couldn't get super "
3152 "buffer head for bytenr %Lu\n", bytenr);
3153 errors++;
3154 continue;
3155 }
3156
a512bbf8
YZ		 3157 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3158
4eedeb75 3159 /* one reference for submit_bh */
a512bbf8 3160 get_bh(bh);
4eedeb75
HH		 3161
3162 set_buffer_uptodate(bh);
a512bbf8
YZ		 3163 lock_buffer(bh);
3164 bh->b_end_io = btrfs_end_buffer_write_sync;
442a4f63 3165 bh->b_private = device;
a512bbf8
YZ		 3166 }
3167
387125fc
CM		 3168 /*
3169 * we fua the first super. The others we allow
3170 * to go down lazy.
3171 */
21adbd5c 3172 ret = btrfsic_submit_bh(WRITE_FUA, bh);
4eedeb75 3173 if (ret)
a512bbf8 3174 errors++;
a512bbf8
YZ		 3175 }
3176 return errors < i ? 0 : -1;
3177}
3178
387125fc
CM		 3179/*
3180 * endio for the write_dev_flush, this will wake anyone waiting
3181 * for the barrier when it is done
3182 */
3183static void btrfs_end_empty_barrier(struct bio *bio, int err)
3184{
3185 if (err) {
3186 if (err == -EOPNOTSUPP)
3187 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3188 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3189 }
3190 if (bio->bi_private)
3191 complete(bio->bi_private);
3192 bio_put(bio);
3193}
3194
3195/*
3196 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3197 * sent down. With wait == 1, it waits for the previous flush.
3198 *
3199 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3200 * capable
3201 */
3202static int write_dev_flush(struct btrfs_device *device, int wait)
3203{
3204 struct bio *bio;
3205 int ret = 0;
3206
3207 if (device->nobarriers)
3208 return 0;
3209
3210 if (wait) {
3211 bio = device->flush_bio;
3212 if (!bio)
3213 return 0;
3214
3215 wait_for_completion(&device->flush_wait);
3216
3217 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
606686ee
JB		 3218 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3219 rcu_str_deref(device->name));
387125fc 3220 device->nobarriers = 1;
5af3e8cc 3221 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
387125fc 3222 ret = -EIO;
5af3e8cc
SB		 3223 btrfs_dev_stat_inc_and_print(device,
3224 BTRFS_DEV_STAT_FLUSH_ERRS);
387125fc
CM		 3225 }
3226
3227 /* drop the reference from the wait == 0 run */
3228 bio_put(bio);
3229 device->flush_bio = NULL;
3230
3231 return ret;
3232 }
3233
3234 /*
3235 * one reference for us, and we leave it for the
3236 * caller
3237 */
9c017abc 3238 device->flush_bio = NULL;
9be3395b 3239 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
387125fc
CM		 3240 if (!bio)
3241 return -ENOMEM;
3242
3243 bio->bi_end_io = btrfs_end_empty_barrier;
3244 bio->bi_bdev = device->bdev;
3245 init_completion(&device->flush_wait);
3246 bio->bi_private = &device->flush_wait;
3247 device->flush_bio = bio;
3248
3249 bio_get(bio);
21adbd5c 3250 btrfsic_submit_bio(WRITE_FLUSH, bio);
387125fc
CM		 3251
3252 return 0;
3253}
3254
3255/*
3256 * send an empty flush down to each device in parallel,
3257 * then wait for them
3258 */
3259static int barrier_all_devices(struct btrfs_fs_info *info)
3260{
3261 struct list_head *head;
3262 struct btrfs_device *dev;
5af3e8cc
SB		 3263 int errors_send = 0;
3264 int errors_wait = 0;
387125fc
CM		 3265 int ret;
3266
3267 /* send down all the barriers */
3268 head = &info->fs_devices->devices;
3269 list_for_each_entry_rcu(dev, head, dev_list) {
3270 if (!dev->bdev) {
5af3e8cc 3271 errors_send++;
387125fc
CM		 3272 continue;
3273 }
3274 if (!dev->in_fs_metadata || !dev->writeable)
3275 continue;
3276
3277 ret = write_dev_flush(dev, 0);
3278 if (ret)
5af3e8cc 3279 errors_send++;
387125fc
CM		 3280 }
3281
3282 /* wait for all the barriers */
3283 list_for_each_entry_rcu(dev, head, dev_list) {
3284 if (!dev->bdev) {
5af3e8cc 3285 errors_wait++;
387125fc
CM		 3286 continue;
3287 }
3288 if (!dev->in_fs_metadata || !dev->writeable)
3289 continue;
3290
3291 ret = write_dev_flush(dev, 1);
3292 if (ret)
5af3e8cc 3293 errors_wait++;
387125fc 3294 }
5af3e8cc
SB		 3295 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3296 errors_wait > info->num_tolerated_disk_barrier_failures)
387125fc
CM		 3297 return -EIO;
3298 return 0;
3299}
3300
5af3e8cc
SB		 3301int btrfs_calc_num_tolerated_disk_barrier_failures(
3302 struct btrfs_fs_info *fs_info)
3303{
3304 struct btrfs_ioctl_space_info space;
3305 struct btrfs_space_info *sinfo;
3306 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3307 BTRFS_BLOCK_GROUP_SYSTEM,
3308 BTRFS_BLOCK_GROUP_METADATA,
3309 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3310 int num_types = 4;
3311 int i;
3312 int c;
3313 int num_tolerated_disk_barrier_failures =
3314 (int)fs_info->fs_devices->num_devices;
3315
3316 for (i = 0; i < num_types; i++) {
3317 struct btrfs_space_info *tmp;
3318
3319 sinfo = NULL;
3320 rcu_read_lock();
3321 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3322 if (tmp->flags == types[i]) {
3323 sinfo = tmp;
3324 break;
3325 }
3326 }
3327 rcu_read_unlock();
3328
3329 if (!sinfo)
3330 continue;
3331
3332 down_read(&sinfo->groups_sem);
3333 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3334 if (!list_empty(&sinfo->block_groups[c])) {
3335 u64 flags;
3336
3337 btrfs_get_block_group_info(
3338 &sinfo->block_groups[c], &space);
3339 if (space.total_bytes == 0 ||
3340 space.used_bytes == 0)
3341 continue;
3342 flags = space.flags;
3343 /*
3344 * return
3345 * 0: if dup, single or RAID0 is configured for
3346 * any of metadata, system or data, else
3347 * 1: if RAID5 is configured, or if RAID1 or
3348 * RAID10 is configured and only two mirrors
3349 * are used, else
3350 * 2: if RAID6 is configured, else
3351 * num_mirrors - 1: if RAID1 or RAID10 is
3352 * configured and more than
3353 * 2 mirrors are used.
3354 */
3355 if (num_tolerated_disk_barrier_failures > 0 &&
3356 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3357 BTRFS_BLOCK_GROUP_RAID0)) ||
3358 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3359 == 0)))
3360 num_tolerated_disk_barrier_failures = 0;
53b381b3
DW		 3361 else if (num_tolerated_disk_barrier_failures > 1) {
3362 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3363 BTRFS_BLOCK_GROUP_RAID5 |
3364 BTRFS_BLOCK_GROUP_RAID10)) {
3365 num_tolerated_disk_barrier_failures = 1;
3366 } else if (flags &
15b0a89d 3367 BTRFS_BLOCK_GROUP_RAID6) {
53b381b3
DW		 3368 num_tolerated_disk_barrier_failures = 2;
3369 }
3370 }
5af3e8cc
SB		 3371 }
3372 }
3373 up_read(&sinfo->groups_sem);
3374 }
3375
3376 return num_tolerated_disk_barrier_failures;
3377}
3378
48a3b636 3379static int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 3380{
e5e9a520 3381 struct list_head *head;
f2984462 3382 struct btrfs_device *dev;
a061fc8d 3383 struct btrfs_super_block *sb;
f2984462 3384 struct btrfs_dev_item *dev_item;
f2984462
CM		 3385 int ret;
3386 int do_barriers;
a236aed1
CM		 3387 int max_errors;
3388 int total_errors = 0;
a061fc8d 3389 u64 flags;
f2984462
CM		 3390
3391 do_barriers = !btrfs_test_opt(root, NOBARRIER);
af31f5e5 3392 backup_super_roots(root->fs_info);
f2984462 3393
6c41761f 3394 sb = root->fs_info->super_for_commit;
a061fc8d 3395 dev_item = &sb->dev_item;
e5e9a520 3396
174ba509 3397 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
e5e9a520 3398 head = &root->fs_info->fs_devices->devices;
d7306801 3399 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
387125fc 3400
5af3e8cc
SB		 3401 if (do_barriers) {
3402 ret = barrier_all_devices(root->fs_info);
3403 if (ret) {
3404 mutex_unlock(
3405 &root->fs_info->fs_devices->device_list_mutex);
3406 btrfs_error(root->fs_info, ret,
3407 "errors while submitting device barriers.");
3408 return ret;
3409 }
3410 }
387125fc 3411
1f78160c 3412 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3413 if (!dev->bdev) {
3414 total_errors++;
3415 continue;
3416 }
2b82032c 3417 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3418 continue;
3419
2b82032c 3420 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 3421 btrfs_set_stack_device_type(dev_item, dev->type);
3422 btrfs_set_stack_device_id(dev_item, dev->devid);
3423 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3424 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3425 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3426 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3427 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3428 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 3429 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 3430
a061fc8d
CM		 3431 flags = btrfs_super_flags(sb);
3432 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3433
a512bbf8 3434 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 3435 if (ret)
3436 total_errors++;
f2984462 3437 }
a236aed1 3438 if (total_errors > max_errors) {
d397712b
CM		 3439 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3440 total_errors);
a724b436 3441 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
79787eaa 3442
9d565ba4
SB		 3443 /* FUA is masked off if unsupported and can't be the reason */
3444 btrfs_error(root->fs_info, -EIO,
3445 "%d errors while writing supers", total_errors);
3446 return -EIO;
a236aed1 3447 }
f2984462 3448
a512bbf8 3449 total_errors = 0;
1f78160c 3450 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM		 3451 if (!dev->bdev)
3452 continue;
2b82032c 3453 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 3454 continue;
3455
a512bbf8
YZ		 3456 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3457 if (ret)
3458 total_errors++;
f2984462 3459 }
174ba509 3460 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 3461 if (total_errors > max_errors) {
79787eaa
JM		 3462 btrfs_error(root->fs_info, -EIO,
3463 "%d errors while writing supers", total_errors);
3464 return -EIO;
a236aed1 3465 }
f2984462
CM		 3466 return 0;
3467}
3468
a512bbf8
YZ		 3469int write_ctree_super(struct btrfs_trans_handle *trans,
3470 struct btrfs_root *root, int max_mirrors)
eb60ceac 3471{
e66f709b 3472 int ret;
5f39d397 3473
a512bbf8 3474 ret = write_all_supers(root, max_mirrors);
5f39d397 3475 return ret;
cfaa7295
CM		 3476}
3477
cb517eab
MX		 3478/* Drop a fs root from the radix tree and free it. */
3479void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3480 struct btrfs_root *root)
2619ba1f 3481{
4df27c4d 3482 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 3483 radix_tree_delete(&fs_info->fs_roots_radix,
3484 (unsigned long)root->root_key.objectid);
4df27c4d 3485 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 3486
3487 if (btrfs_root_refs(&root->root_item) == 0)
3488 synchronize_srcu(&fs_info->subvol_srcu);
3489
d7634482 3490 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3321719e
LB		 3491 btrfs_free_log(NULL, root);
3492 btrfs_free_log_root_tree(NULL, fs_info);
3493 }
3494
581bb050
LZ		 3495 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3496 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d 3497 free_fs_root(root);
4df27c4d
YZ		 3498}
3499
3500static void free_fs_root(struct btrfs_root *root)
3501{
82d5902d 3502 iput(root->cache_inode);
4df27c4d 3503 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
1cb048f5
FDBM		 3504 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3505 root->orphan_block_rsv = NULL;
0ee5dc67
AV		 3506 if (root->anon_dev)
3507 free_anon_bdev(root->anon_dev);
4df27c4d
YZ		 3508 free_extent_buffer(root->node);
3509 free_extent_buffer(root->commit_root);
581bb050
LZ		 3510 kfree(root->free_ino_ctl);
3511 kfree(root->free_ino_pinned);
d397712b 3512 kfree(root->name);
b0feb9d9 3513 btrfs_put_fs_root(root);
2619ba1f
CM		 3514}
3515
cb517eab
MX		 3516void btrfs_free_fs_root(struct btrfs_root *root)
3517{
3518 free_fs_root(root);
2619ba1f
CM		 3519}
3520
c146afad 3521int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 3522{
c146afad
YZ		 3523 u64 root_objectid = 0;
3524 struct btrfs_root *gang[8];
3525 int i;
3768f368 3526 int ret;
e089f05c 3527
c146afad
YZ		 3528 while (1) {
3529 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3530 (void **)gang, root_objectid,
3531 ARRAY_SIZE(gang));
3532 if (!ret)
3533 break;
5d4f98a2
YZ		 3534
3535 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 3536 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 3537 int err;
3538
c146afad 3539 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 3540 err = btrfs_orphan_cleanup(gang[i]);
3541 if (err)
3542 return err;
c146afad
YZ		 3543 }
3544 root_objectid++;
3545 }
3546 return 0;
3547}
a2135011 3548
c146afad
YZ		 3549int btrfs_commit_super(struct btrfs_root *root)
3550{
3551 struct btrfs_trans_handle *trans;
3552 int ret;
a74a4b97 3553
c146afad 3554 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 3555 btrfs_run_delayed_iputs(root);
c146afad 3556 mutex_unlock(&root->fs_info->cleaner_mutex);
9d1a2a3a 3557 wake_up_process(root->fs_info->cleaner_kthread);
c71bf099
YZ		 3558
3559 /* wait until ongoing cleanup work done */
3560 down_write(&root->fs_info->cleanup_work_sem);
3561 up_write(&root->fs_info->cleanup_work_sem);
3562
7a7eaa40 3563 trans = btrfs_join_transaction(root);
3612b495
TI		 3564 if (IS_ERR(trans))
3565 return PTR_ERR(trans);
54aa1f4d 3566 ret = btrfs_commit_transaction(trans, root);
79787eaa
JM		 3567 if (ret)
3568 return ret;
c146afad 3569 /* run commit again to drop the original snapshot */
7a7eaa40 3570 trans = btrfs_join_transaction(root);
3612b495
TI		 3571 if (IS_ERR(trans))
3572 return PTR_ERR(trans);
79787eaa
JM		 3573 ret = btrfs_commit_transaction(trans, root);
3574 if (ret)
3575 return ret;
79154b1b 3576 ret = btrfs_write_and_wait_transaction(NULL, root);
79787eaa
JM		 3577 if (ret) {
3578 btrfs_error(root->fs_info, ret,
3579 "Failed to sync btree inode to disk.");
3580 return ret;
3581 }
d6bfde87 3582
a512bbf8 3583 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 3584 return ret;
3585}
3586
3587int close_ctree(struct btrfs_root *root)
3588{
3589 struct btrfs_fs_info *fs_info = root->fs_info;
3590 int ret;
3591
3592 fs_info->closing = 1;
3593 smp_mb();
3594
803b2f54
SB		 3595 /* wait for the uuid_scan task to finish */
3596 down(&fs_info->uuid_tree_rescan_sem);
3597 /* avoid complains from lockdep et al., set sem back to initial state */
3598 up(&fs_info->uuid_tree_rescan_sem);
3599
837d5b6e 3600 /* pause restriper - we want to resume on mount */
aa1b8cd4 3601 btrfs_pause_balance(fs_info);
837d5b6e 3602
8dabb742
SB		 3603 btrfs_dev_replace_suspend_for_unmount(fs_info);
3604
aa1b8cd4 3605 btrfs_scrub_cancel(fs_info);
4cb5300b
CM		 3606
3607 /* wait for any defraggers to finish */
3608 wait_event(fs_info->transaction_wait,
3609 (atomic_read(&fs_info->defrag_running) == 0));
3610
3611 /* clear out the rbtree of defraggable inodes */
26176e7c 3612 btrfs_cleanup_defrag_inodes(fs_info);
4cb5300b 3613
c146afad 3614 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 3615 ret = btrfs_commit_super(root);
3616 if (ret)
3617 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3618 }
3619
87533c47 3620 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
68ce9682 3621 btrfs_error_commit_super(root);
0f7d52f4 3622
300e4f8a
JB		 3623 btrfs_put_block_group_cache(fs_info);
3624
e3029d9f
AV		 3625 kthread_stop(fs_info->transaction_kthread);
3626 kthread_stop(fs_info->cleaner_kthread);
8929ecfa 3627
f25784b3
YZ		 3628 fs_info->closing = 2;
3629 smp_mb();
3630
bcef60f2
AJ		 3631 btrfs_free_qgroup_config(root->fs_info);
3632
963d678b
MX		 3633 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3634 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3635 percpu_counter_sum(&fs_info->delalloc_bytes));
b0c68f8b 3636 }
bcc63abb 3637
e3029d9f 3638 btrfs_free_block_groups(fs_info);
d10c5f31 3639
13e6c37b 3640 btrfs_stop_all_workers(fs_info);
2932505a 3641
c146afad 3642 del_fs_roots(fs_info);
d10c5f31 3643
13e6c37b 3644 free_root_pointers(fs_info, 1);
9ad6b7bc 3645
13e6c37b 3646 iput(fs_info->btree_inode);
d6bfde87 3647
21adbd5c
SB		 3648#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3649 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3650 btrfsic_unmount(root, fs_info->fs_devices);
3651#endif
3652
dfe25020 3653 btrfs_close_devices(fs_info->fs_devices);
0b86a832 3654 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 3655
e2d84521 3656 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
963d678b 3657 percpu_counter_destroy(&fs_info->delalloc_bytes);
04160088 3658 bdi_destroy(&fs_info->bdi);
76dda93c 3659 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 3660
53b381b3
DW		 3661 btrfs_free_stripe_hash_table(fs_info);
3662
1cb048f5
FDBM		 3663 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3664 root->orphan_block_rsv = NULL;
3665
eb60ceac
CM		 3666 return 0;
3667}
3668
b9fab919
CM		 3669int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3670 int atomic)
5f39d397 3671{
1259ab75 3672 int ret;
727011e0 3673 struct inode *btree_inode = buf->pages[0]->mapping->host;
1259ab75 3674
0b32f4bb 3675 ret = extent_buffer_uptodate(buf);
1259ab75
CM		 3676 if (!ret)
3677 return ret;
3678
3679 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
b9fab919
CM		 3680 parent_transid, atomic);
3681 if (ret == -EAGAIN)
3682 return ret;
1259ab75 3683 return !ret;
5f39d397
CM		 3684}
3685
3686int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 3687{
0b32f4bb 3688 return set_extent_buffer_uptodate(buf);
5f39d397 3689}
6702ed49 3690
5f39d397
CM		 3691void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3692{
06ea65a3 3693 struct btrfs_root *root;
5f39d397 3694 u64 transid = btrfs_header_generation(buf);
b9473439 3695 int was_dirty;
b4ce94de 3696
06ea65a3
JB		 3697#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3698 /*
3699 * This is a fast path so only do this check if we have sanity tests
3700 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3701 * outside of the sanity tests.
3702 */
3703 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3704 return;
3705#endif
3706 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
b9447ef8 3707 btrfs_assert_tree_locked(buf);
31b1a2bd
JL		 3708 if (transid != root->fs_info->generation)
3709 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
d397712b 3710 "found %llu running %llu\n",
c1c9ff7c 3711 buf->start, transid, root->fs_info->generation);
0b32f4bb 3712 was_dirty = set_extent_buffer_dirty(buf);
e2d84521
MX		 3713 if (!was_dirty)
3714 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3715 buf->len,
3716 root->fs_info->dirty_metadata_batch);
eb60ceac
CM		 3717}
3718
b53d3f5d
LB		 3719static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3720 int flush_delayed)
16cdcec7
MX		 3721{
3722 /*
3723 * looks as though older kernels can get into trouble with
3724 * this code, they end up stuck in balance_dirty_pages forever
3725 */
e2d84521 3726 int ret;
16cdcec7
MX		 3727
3728 if (current->flags & PF_MEMALLOC)
3729 return;
3730
b53d3f5d
LB		 3731 if (flush_delayed)
3732 btrfs_balance_delayed_items(root);
16cdcec7 3733
e2d84521
MX		 3734 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3735 BTRFS_DIRTY_METADATA_THRESH);
3736 if (ret > 0) {
d0e1d66b
NJ		 3737 balance_dirty_pages_ratelimited(
3738 root->fs_info->btree_inode->i_mapping);
16cdcec7
MX		 3739 }
3740 return;
3741}
3742
b53d3f5d 3743void btrfs_btree_balance_dirty(struct btrfs_root *root)
35b7e476 3744{
b53d3f5d
LB		 3745 __btrfs_btree_balance_dirty(root, 1);
3746}
585ad2c3 3747
b53d3f5d
LB		 3748void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3749{
3750 __btrfs_btree_balance_dirty(root, 0);
35b7e476 3751}
6b80053d 3752
ca7a79ad 3753int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 3754{
727011e0 3755 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
0b32f4bb 3756 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
6b80053d 3757}
0da5468f 3758
fcd1f065 3759static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
acce952b 3760 int read_only)
3761{
1104a885
DS		 3762 /*
3763 * Placeholder for checks
3764 */
fcd1f065 3765 return 0;
acce952b 3766}
3767
48a3b636 3768static void btrfs_error_commit_super(struct btrfs_root *root)
acce952b 3769{
acce952b 3770 mutex_lock(&root->fs_info->cleaner_mutex);
3771 btrfs_run_delayed_iputs(root);
3772 mutex_unlock(&root->fs_info->cleaner_mutex);
3773
3774 down_write(&root->fs_info->cleanup_work_sem);
3775 up_write(&root->fs_info->cleanup_work_sem);
3776
3777 /* cleanup FS via transaction */
3778 btrfs_cleanup_transaction(root);
acce952b 3779}
3780
569e0f35
JB		 3781static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3782 struct btrfs_root *root)
acce952b 3783{
3784 struct btrfs_inode *btrfs_inode;
3785 struct list_head splice;
3786
3787 INIT_LIST_HEAD(&splice);
3788
3789 mutex_lock(&root->fs_info->ordered_operations_mutex);
199c2a9c 3790 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3791
569e0f35 3792 list_splice_init(&t->ordered_operations, &splice);
acce952b 3793 while (!list_empty(&splice)) {
3794 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3795 ordered_operations);
3796
3797 list_del_init(&btrfs_inode->ordered_operations);
199c2a9c 3798 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3799
3800 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3801
199c2a9c 3802 spin_lock(&root->fs_info->ordered_root_lock);
acce952b 3803 }
3804
199c2a9c 3805 spin_unlock(&root->fs_info->ordered_root_lock);
acce952b 3806 mutex_unlock(&root->fs_info->ordered_operations_mutex);
acce952b 3807}
3808
143bede5 3809static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
acce952b 3810{
acce952b 3811 struct btrfs_ordered_extent *ordered;
acce952b 3812
199c2a9c 3813 spin_lock(&root->ordered_extent_lock);
779880ef
JB		 3814 /*
3815 * This will just short circuit the ordered completion stuff which will
3816 * make sure the ordered extent gets properly cleaned up.
3817 */
199c2a9c 3818 list_for_each_entry(ordered, &root->ordered_extents,
779880ef
JB		 3819 root_extent_list)
3820 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
199c2a9c
MX		 3821 spin_unlock(&root->ordered_extent_lock);
3822}
3823
3824static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3825{
3826 struct btrfs_root *root;
3827 struct list_head splice;
3828
3829 INIT_LIST_HEAD(&splice);
3830
3831 spin_lock(&fs_info->ordered_root_lock);
3832 list_splice_init(&fs_info->ordered_roots, &splice);
3833 while (!list_empty(&splice)) {
3834 root = list_first_entry(&splice, struct btrfs_root,
3835 ordered_root);
3836 list_del_init(&root->ordered_root);
3837
3838 btrfs_destroy_ordered_extents(root);
3839
3840 cond_resched_lock(&fs_info->ordered_root_lock);
3841 }
3842 spin_unlock(&fs_info->ordered_root_lock);
acce952b 3843}
3844
35a3621b
SB		 3845static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3846 struct btrfs_root *root)
acce952b 3847{
3848 struct rb_node *node;
3849 struct btrfs_delayed_ref_root *delayed_refs;
3850 struct btrfs_delayed_ref_node *ref;
3851 int ret = 0;
3852
3853 delayed_refs = &trans->delayed_refs;
3854
3855 spin_lock(&delayed_refs->lock);
3856 if (delayed_refs->num_entries == 0) {
cfece4db 3857 spin_unlock(&delayed_refs->lock);
acce952b 3858 printk(KERN_INFO "delayed_refs has NO entry\n");
3859 return ret;
3860 }
3861
b939d1ab 3862 while ((node = rb_first(&delayed_refs->root)) != NULL) {
eb12db69 3863 struct btrfs_delayed_ref_head *head = NULL;
e78417d1 3864 bool pin_bytes = false;
acce952b 3865
eb12db69 3866 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
acce952b 3867 atomic_set(&ref->refs, 1);
3868 if (btrfs_delayed_ref_is_head(ref)) {
acce952b 3869
3870 head = btrfs_delayed_node_to_head(ref);
b939d1ab
JB		 3871 if (!mutex_trylock(&head->mutex)) {
3872 atomic_inc(&ref->refs);
3873 spin_unlock(&delayed_refs->lock);
3874
3875 /* Need to wait for the delayed ref to run */
3876 mutex_lock(&head->mutex);
3877 mutex_unlock(&head->mutex);
3878 btrfs_put_delayed_ref(ref);
3879
e18fca73 3880 spin_lock(&delayed_refs->lock);
b939d1ab
JB		 3881 continue;
3882 }
3883
54067ae9 3884 if (head->must_insert_reserved)
e78417d1 3885 pin_bytes = true;
78a6184a 3886 btrfs_free_delayed_extent_op(head->extent_op);
acce952b 3887 delayed_refs->num_heads--;
3888 if (list_empty(&head->cluster))
3889 delayed_refs->num_heads_ready--;
3890 list_del_init(&head->cluster);
acce952b 3891 }
eb12db69 3892
b939d1ab
JB		 3893 ref->in_tree = 0;
3894 rb_erase(&ref->rb_node, &delayed_refs->root);
3895 delayed_refs->num_entries--;
acce952b 3896 spin_unlock(&delayed_refs->lock);
e78417d1
JB		 3897 if (head) {
3898 if (pin_bytes)
3899 btrfs_pin_extent(root, ref->bytenr,
3900 ref->num_bytes, 1);
3901 mutex_unlock(&head->mutex);
3902 }
acce952b 3903 btrfs_put_delayed_ref(ref);
3904
3905 cond_resched();
3906 spin_lock(&delayed_refs->lock);
3907 }
3908
3909 spin_unlock(&delayed_refs->lock);
3910
3911 return ret;
3912}
3913
aec8030a 3914static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
acce952b 3915{
3916 struct btrfs_pending_snapshot *snapshot;
3917 struct list_head splice;
3918
3919 INIT_LIST_HEAD(&splice);
3920
3921 list_splice_init(&t->pending_snapshots, &splice);
3922
3923 while (!list_empty(&splice)) {
3924 snapshot = list_entry(splice.next,
3925 struct btrfs_pending_snapshot,
3926 list);
aec8030a 3927 snapshot->error = -ECANCELED;
acce952b 3928 list_del_init(&snapshot->list);
acce952b 3929 }
acce952b 3930}
3931
143bede5 3932static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
acce952b 3933{
3934 struct btrfs_inode *btrfs_inode;
3935 struct list_head splice;
3936
3937 INIT_LIST_HEAD(&splice);
3938
eb73c1b7
MX		 3939 spin_lock(&root->delalloc_lock);
3940 list_splice_init(&root->delalloc_inodes, &splice);
acce952b 3941
3942 while (!list_empty(&splice)) {
eb73c1b7
MX		 3943 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3944 delalloc_inodes);
acce952b 3945
3946 list_del_init(&btrfs_inode->delalloc_inodes);
df0af1a5
MX		 3947 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3948 &btrfs_inode->runtime_flags);
eb73c1b7 3949 spin_unlock(&root->delalloc_lock);
acce952b 3950
3951 btrfs_invalidate_inodes(btrfs_inode->root);
b216cbfb 3952
eb73c1b7 3953 spin_lock(&root->delalloc_lock);
acce952b 3954 }
3955
eb73c1b7
MX		 3956 spin_unlock(&root->delalloc_lock);
3957}
3958
3959static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3960{
3961 struct btrfs_root *root;
3962 struct list_head splice;
3963
3964 INIT_LIST_HEAD(&splice);
3965
3966 spin_lock(&fs_info->delalloc_root_lock);
3967 list_splice_init(&fs_info->delalloc_roots, &splice);
3968 while (!list_empty(&splice)) {
3969 root = list_first_entry(&splice, struct btrfs_root,
3970 delalloc_root);
3971 list_del_init(&root->delalloc_root);
3972 root = btrfs_grab_fs_root(root);
3973 BUG_ON(!root);
3974 spin_unlock(&fs_info->delalloc_root_lock);
3975
3976 btrfs_destroy_delalloc_inodes(root);
3977 btrfs_put_fs_root(root);
3978
3979 spin_lock(&fs_info->delalloc_root_lock);
3980 }
3981 spin_unlock(&fs_info->delalloc_root_lock);
acce952b 3982}
3983
3984static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3985 struct extent_io_tree *dirty_pages,
3986 int mark)
3987{
3988 int ret;
acce952b 3989 struct extent_buffer *eb;
3990 u64 start = 0;
3991 u64 end;
acce952b 3992
3993 while (1) {
3994 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
e6138876 3995 mark, NULL);
acce952b 3996 if (ret)
3997 break;
3998
3999 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4000 while (start <= end) {
fd8b2b61
JB		 4001 eb = btrfs_find_tree_block(root, start,
4002 root->leafsize);
69a85bd8 4003 start += root->leafsize;
fd8b2b61 4004 if (!eb)
acce952b 4005 continue;
fd8b2b61 4006 wait_on_extent_buffer_writeback(eb);
acce952b 4007
fd8b2b61
JB		 4008 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4009 &eb->bflags))
4010 clear_extent_buffer_dirty(eb);
4011 free_extent_buffer_stale(eb);
acce952b 4012 }
4013 }
4014
4015 return ret;
4016}
4017
4018static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4019 struct extent_io_tree *pinned_extents)
4020{
4021 struct extent_io_tree *unpin;
4022 u64 start;
4023 u64 end;
4024 int ret;
ed0eaa14 4025 bool loop = true;
acce952b 4026
4027 unpin = pinned_extents;
ed0eaa14 4028again:
acce952b 4029 while (1) {
4030 ret = find_first_extent_bit(unpin, 0, &start, &end,
e6138876 4031 EXTENT_DIRTY, NULL);
acce952b 4032 if (ret)
4033 break;
4034
4035 /* opt_discard */
5378e607
LD		 4036 if (btrfs_test_opt(root, DISCARD))
4037 ret = btrfs_error_discard_extent(root, start,
4038 end + 1 - start,
4039 NULL);
acce952b 4040
4041 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4042 btrfs_error_unpin_extent_range(root, start, end);
4043 cond_resched();
4044 }
4045
ed0eaa14
LB		 4046 if (loop) {
4047 if (unpin == &root->fs_info->freed_extents[0])
4048 unpin = &root->fs_info->freed_extents[1];
4049 else
4050 unpin = &root->fs_info->freed_extents[0];
4051 loop = false;
4052 goto again;
4053 }
4054
acce952b 4055 return 0;
4056}
4057
49b25e05
JM		 4058void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4059 struct btrfs_root *root)
4060{
4061 btrfs_destroy_delayed_refs(cur_trans, root);
4062 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
4063 cur_trans->dirty_pages.dirty_bytes);
4064
4a9d8bde 4065 cur_trans->state = TRANS_STATE_COMMIT_START;
d7096fc3 4066 wake_up(&root->fs_info->transaction_blocked_wait);
49b25e05 4067
aec8030a
MX		 4068 btrfs_evict_pending_snapshots(cur_trans);
4069
4a9d8bde 4070 cur_trans->state = TRANS_STATE_UNBLOCKED;
d7096fc3 4071 wake_up(&root->fs_info->transaction_wait);
49b25e05 4072
67cde344
MX		 4073 btrfs_destroy_delayed_inodes(root);
4074 btrfs_assert_delayed_root_empty(root);
49b25e05 4075
49b25e05
JM		 4076 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4077 EXTENT_DIRTY);
6e841e32
LB		 4078 btrfs_destroy_pinned_extent(root,
4079 root->fs_info->pinned_extents);
49b25e05 4080
4a9d8bde
MX		 4081 cur_trans->state =TRANS_STATE_COMPLETED;
4082 wake_up(&cur_trans->commit_wait);
4083
49b25e05
JM		 4084 /*
4085 memset(cur_trans, 0, sizeof(*cur_trans));
4086 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4087 */
4088}
4089
48a3b636 4090static int btrfs_cleanup_transaction(struct btrfs_root *root)
acce952b 4091{
4092 struct btrfs_transaction *t;
4093 LIST_HEAD(list);
4094
acce952b 4095 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4096
a4abeea4 4097 spin_lock(&root->fs_info->trans_lock);
acce952b 4098 list_splice_init(&root->fs_info->trans_list, &list);
ac673879 4099 root->fs_info->running_transaction = NULL;
a4abeea4
JB		 4100 spin_unlock(&root->fs_info->trans_lock);
4101
acce952b 4102 while (!list_empty(&list)) {
4103 t = list_entry(list.next, struct btrfs_transaction, list);
acce952b 4104
569e0f35 4105 btrfs_destroy_ordered_operations(t, root);
acce952b 4106
199c2a9c 4107 btrfs_destroy_all_ordered_extents(root->fs_info);
acce952b 4108
4109 btrfs_destroy_delayed_refs(t, root);
4110
4a9d8bde
MX		 4111 /*
4112 * FIXME: cleanup wait for commit
4113 * We needn't acquire the lock here, because we are during
4114 * the umount, there is no other task which will change it.
4115 */
4116 t->state = TRANS_STATE_COMMIT_START;
66657b31 4117 smp_mb();
acce952b 4118 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4119 wake_up(&root->fs_info->transaction_blocked_wait);
4120
aec8030a
MX		 4121 btrfs_evict_pending_snapshots(t);
4122
4a9d8bde 4123 t->state = TRANS_STATE_UNBLOCKED;
66657b31 4124 smp_mb();
acce952b 4125 if (waitqueue_active(&root->fs_info->transaction_wait))
4126 wake_up(&root->fs_info->transaction_wait);
acce952b 4127
67cde344
MX		 4128 btrfs_destroy_delayed_inodes(root);
4129 btrfs_assert_delayed_root_empty(root);
4130
eb73c1b7 4131 btrfs_destroy_all_delalloc_inodes(root->fs_info);
acce952b 4132
4133 btrfs_destroy_marked_extents(root, &t->dirty_pages,
4134 EXTENT_DIRTY);
4135
4136 btrfs_destroy_pinned_extent(root,
4137 root->fs_info->pinned_extents);
4138
4a9d8bde
MX		 4139 t->state = TRANS_STATE_COMPLETED;
4140 smp_mb();
4141 if (waitqueue_active(&t->commit_wait))
4142 wake_up(&t->commit_wait);
4143
13c5a93e 4144 atomic_set(&t->use_count, 0);
acce952b 4145 list_del_init(&t->list);
4146 memset(t, 0, sizeof(*t));
4147 kmem_cache_free(btrfs_transaction_cachep, t);
4148 }
4149
4150 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
acce952b 4151
4152 return 0;
4153}
4154
d1310b2e 4155static struct extent_io_ops btree_extent_io_ops = {
ce9adaa5 4156 .readpage_end_io_hook = btree_readpage_end_io_hook,
4bb31e92 4157 .readpage_io_failed_hook = btree_io_failed_hook,
0b86a832 4158 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 4159 /* note we're sharing with inode.c for the merge bio hook */
4160 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 4161};
Linux 6.x block layer and io_uring tree(s)