projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Btrfs: Make free space cache code generic
[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>
7e75bf3f 32#include <asm/unaligned.h>
4b4e25f2 33#include "compat.h"
eb60ceac
CM		 34#include "ctree.h"
35#include "disk-io.h"
e089f05c 36#include "transaction.h"
0f7d52f4 37#include "btrfs_inode.h"
0b86a832 38#include "volumes.h"
db94535d 39#include "print-tree.h"
8b712842 40#include "async-thread.h"
925baedd 41#include "locking.h"
e02119d5 42#include "tree-log.h"
fa9c0d79 43#include "free-space-cache.h"
eb60ceac 44
d1310b2e 45static struct extent_io_ops btree_extent_io_ops;
8b712842 46static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 47static void free_fs_root(struct btrfs_root *root);
acce952b 48static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
49 int read_only);
50static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
51static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
52static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
53 struct btrfs_root *root);
54static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
55static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
56static int btrfs_destroy_marked_extents(struct btrfs_root *root,
57 struct extent_io_tree *dirty_pages,
58 int mark);
59static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
60 struct extent_io_tree *pinned_extents);
61static int btrfs_cleanup_transaction(struct btrfs_root *root);
ce9adaa5 62
d352ac68
CM		 63/*
64 * end_io_wq structs are used to do processing in task context when an IO is
65 * complete. This is used during reads to verify checksums, and it is used
66 * by writes to insert metadata for new file extents after IO is complete.
67 */
ce9adaa5
CM		 68struct end_io_wq {
69 struct bio *bio;
70 bio_end_io_t *end_io;
71 void *private;
72 struct btrfs_fs_info *info;
73 int error;
22c59948 74 int metadata;
ce9adaa5 75 struct list_head list;
8b712842 76 struct btrfs_work work;
ce9adaa5 77};
0da5468f 78
d352ac68
CM		 79/*
80 * async submit bios are used to offload expensive checksumming
81 * onto the worker threads. They checksum file and metadata bios
82 * just before they are sent down the IO stack.
83 */
44b8bd7e
CM		 84struct async_submit_bio {
85 struct inode *inode;
86 struct bio *bio;
87 struct list_head list;
4a69a410
CM		 88 extent_submit_bio_hook_t *submit_bio_start;
89 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM		 90 int rw;
91 int mirror_num;
c8b97818 92 unsigned long bio_flags;
eaf25d93
CM		 93 /*
94 * bio_offset is optional, can be used if the pages in the bio
95 * can't tell us where in the file the bio should go
96 */
97 u64 bio_offset;
8b712842 98 struct btrfs_work work;
44b8bd7e
CM		 99};
100
4008c04a
CM		 101/* These are used to set the lockdep class on the extent buffer locks.
102 * The class is set by the readpage_end_io_hook after the buffer has
103 * passed csum validation but before the pages are unlocked.
104 *
105 * The lockdep class is also set by btrfs_init_new_buffer on freshly
106 * allocated blocks.
107 *
108 * The class is based on the level in the tree block, which allows lockdep
109 * to know that lower nodes nest inside the locks of higher nodes.
110 *
111 * We also add a check to make sure the highest level of the tree is
112 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
113 * code needs update as well.
114 */
115#ifdef CONFIG_DEBUG_LOCK_ALLOC
116# if BTRFS_MAX_LEVEL != 8
117# error
118# endif
119static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
120static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
121 /* leaf */
122 "btrfs-extent-00",
123 "btrfs-extent-01",
124 "btrfs-extent-02",
125 "btrfs-extent-03",
126 "btrfs-extent-04",
127 "btrfs-extent-05",
128 "btrfs-extent-06",
129 "btrfs-extent-07",
130 /* highest possible level */
131 "btrfs-extent-08",
132};
133#endif
134
d352ac68
CM		 135/*
136 * extents on the btree inode are pretty simple, there's one extent
137 * that covers the entire device
138 */
b2950863
CH		 139static struct extent_map *btree_get_extent(struct inode *inode,
140 struct page *page, size_t page_offset, u64 start, u64 len,
141 int create)
7eccb903 142{
5f39d397
CM		 143 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
144 struct extent_map *em;
145 int ret;
146
890871be 147 read_lock(&em_tree->lock);
d1310b2e 148 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM		 149 if (em) {
150 em->bdev =
151 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 152 read_unlock(&em_tree->lock);
5f39d397 153 goto out;
a061fc8d 154 }
890871be 155 read_unlock(&em_tree->lock);
7b13b7b1 156
5f39d397
CM		 157 em = alloc_extent_map(GFP_NOFS);
158 if (!em) {
159 em = ERR_PTR(-ENOMEM);
160 goto out;
161 }
162 em->start = 0;
0afbaf8c 163 em->len = (u64)-1;
c8b97818 164 em->block_len = (u64)-1;
5f39d397 165 em->block_start = 0;
a061fc8d 166 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 167
890871be 168 write_lock(&em_tree->lock);
5f39d397
CM		 169 ret = add_extent_mapping(em_tree, em);
170 if (ret == -EEXIST) {
0afbaf8c
CM		 171 u64 failed_start = em->start;
172 u64 failed_len = em->len;
173
5f39d397 174 free_extent_map(em);
7b13b7b1 175 em = lookup_extent_mapping(em_tree, start, len);
0afbaf8c 176 if (em) {
7b13b7b1 177 ret = 0;
0afbaf8c
CM		 178 } else {
179 em = lookup_extent_mapping(em_tree, failed_start,
180 failed_len);
7b13b7b1 181 ret = -EIO;
0afbaf8c 182 }
5f39d397 183 } else if (ret) {
7b13b7b1
CM		 184 free_extent_map(em);
185 em = NULL;
5f39d397 186 }
890871be 187 write_unlock(&em_tree->lock);
7b13b7b1
CM		 188
189 if (ret)
190 em = ERR_PTR(ret);
5f39d397
CM		 191out:
192 return em;
7eccb903
CM		 193}
194
19c00ddc
CM		 195u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
196{
163e783e 197 return crc32c(seed, data, len);
19c00ddc
CM		 198}
199
200void btrfs_csum_final(u32 crc, char *result)
201{
7e75bf3f 202 put_unaligned_le32(~crc, result);
19c00ddc
CM		 203}
204
d352ac68
CM		 205/*
206 * compute the csum for a btree block, and either verify it or write it
207 * into the csum field of the block.
208 */
19c00ddc
CM		 209static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
210 int verify)
211{
607d432d
JB		 212 u16 csum_size =
213 btrfs_super_csum_size(&root->fs_info->super_copy);
214 char *result = NULL;
19c00ddc
CM		 215 unsigned long len;
216 unsigned long cur_len;
217 unsigned long offset = BTRFS_CSUM_SIZE;
218 char *map_token = NULL;
219 char *kaddr;
220 unsigned long map_start;
221 unsigned long map_len;
222 int err;
223 u32 crc = ~(u32)0;
607d432d 224 unsigned long inline_result;
19c00ddc
CM		 225
226 len = buf->len - offset;
d397712b 227 while (len > 0) {
19c00ddc
CM		 228 err = map_private_extent_buffer(buf, offset, 32,
229 &map_token, &kaddr,
230 &map_start, &map_len, KM_USER0);
d397712b 231 if (err)
19c00ddc 232 return 1;
19c00ddc
CM		 233 cur_len = min(len, map_len - (offset - map_start));
234 crc = btrfs_csum_data(root, kaddr + offset - map_start,
235 crc, cur_len);
236 len -= cur_len;
237 offset += cur_len;
238 unmap_extent_buffer(buf, map_token, KM_USER0);
239 }
607d432d
JB		 240 if (csum_size > sizeof(inline_result)) {
241 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
242 if (!result)
243 return 1;
244 } else {
245 result = (char *)&inline_result;
246 }
247
19c00ddc
CM		 248 btrfs_csum_final(crc, result);
249
250 if (verify) {
607d432d 251 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM		 252 u32 val;
253 u32 found = 0;
607d432d 254 memcpy(&found, result, csum_size);
e4204ded 255
607d432d 256 read_extent_buffer(buf, &val, 0, csum_size);
193f284d
CM		 257 if (printk_ratelimit()) {
258 printk(KERN_INFO "btrfs: %s checksum verify "
259 "failed on %llu wanted %X found %X "
260 "level %d\n",
261 root->fs_info->sb->s_id,
262 (unsigned long long)buf->start, val, found,
263 btrfs_header_level(buf));
264 }
607d432d
JB		 265 if (result != (char *)&inline_result)
266 kfree(result);
19c00ddc
CM		 267 return 1;
268 }
269 } else {
607d432d 270 write_extent_buffer(buf, result, 0, csum_size);
19c00ddc 271 }
607d432d
JB		 272 if (result != (char *)&inline_result)
273 kfree(result);
19c00ddc
CM		 274 return 0;
275}
276
d352ac68
CM		 277/*
278 * we can't consider a given block up to date unless the transid of the
279 * block matches the transid in the parent node's pointer. This is how we
280 * detect blocks that either didn't get written at all or got written
281 * in the wrong place.
282 */
1259ab75
CM		 283static int verify_parent_transid(struct extent_io_tree *io_tree,
284 struct extent_buffer *eb, u64 parent_transid)
285{
2ac55d41 286 struct extent_state *cached_state = NULL;
1259ab75
CM		 287 int ret;
288
289 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
290 return 0;
291
2ac55d41
JB		 292 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
293 0, &cached_state, GFP_NOFS);
294 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
1259ab75
CM		 295 btrfs_header_generation(eb) == parent_transid) {
296 ret = 0;
297 goto out;
298 }
193f284d
CM		 299 if (printk_ratelimit()) {
300 printk("parent transid verify failed on %llu wanted %llu "
301 "found %llu\n",
302 (unsigned long long)eb->start,
303 (unsigned long long)parent_transid,
304 (unsigned long long)btrfs_header_generation(eb));
305 }
1259ab75 306 ret = 1;
2ac55d41 307 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
33958dc6 308out:
2ac55d41
JB		 309 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
310 &cached_state, GFP_NOFS);
1259ab75 311 return ret;
1259ab75
CM		 312}
313
d352ac68
CM		 314/*
315 * helper to read a given tree block, doing retries as required when
316 * the checksums don't match and we have alternate mirrors to try.
317 */
f188591e
CM		 318static int btree_read_extent_buffer_pages(struct btrfs_root *root,
319 struct extent_buffer *eb,
ca7a79ad 320 u64 start, u64 parent_transid)
f188591e
CM		 321{
322 struct extent_io_tree *io_tree;
323 int ret;
324 int num_copies = 0;
325 int mirror_num = 0;
326
a826d6dc 327 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM		 328 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
329 while (1) {
330 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
331 btree_get_extent, mirror_num);
1259ab75
CM		 332 if (!ret &&
333 !verify_parent_transid(io_tree, eb, parent_transid))
f188591e 334 return ret;
d397712b 335
a826d6dc
JB		 336 /*
337 * This buffer's crc is fine, but its contents are corrupted, so
338 * there is no reason to read the other copies, they won't be
339 * any less wrong.
340 */
341 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
342 return ret;
343
f188591e
CM		 344 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
345 eb->start, eb->len);
4235298e 346 if (num_copies == 1)
f188591e 347 return ret;
4235298e 348
f188591e 349 mirror_num++;
4235298e 350 if (mirror_num > num_copies)
f188591e 351 return ret;
f188591e 352 }
f188591e
CM		 353 return -EIO;
354}
19c00ddc 355
d352ac68 356/*
d397712b
CM		 357 * checksum a dirty tree block before IO. This has extra checks to make sure
358 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 359 */
d397712b 360
b2950863 361static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 362{
d1310b2e 363 struct extent_io_tree *tree;
35ebb934 364 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
19c00ddc 365 u64 found_start;
19c00ddc
CM		 366 unsigned long len;
367 struct extent_buffer *eb;
f188591e
CM		 368 int ret;
369
d1310b2e 370 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc 371
eb14ab8e
CM		 372 if (page->private == EXTENT_PAGE_PRIVATE) {
373 WARN_ON(1);
19c00ddc 374 goto out;
eb14ab8e
CM		 375 }
376 if (!page->private) {
377 WARN_ON(1);
19c00ddc 378 goto out;
eb14ab8e 379 }
19c00ddc 380 len = page->private >> 2;
d397712b
CM		 381 WARN_ON(len == 0);
382
19c00ddc 383 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
91ca338d
TI		 384 if (eb == NULL) {
385 WARN_ON(1);
386 goto out;
387 }
ca7a79ad
CM		 388 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
389 btrfs_header_generation(eb));
f188591e 390 BUG_ON(ret);
784b4e29
CM		 391 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
392
19c00ddc
CM		 393 found_start = btrfs_header_bytenr(eb);
394 if (found_start != start) {
55c69072
CM		 395 WARN_ON(1);
396 goto err;
397 }
398 if (eb->first_page != page) {
55c69072
CM		 399 WARN_ON(1);
400 goto err;
401 }
402 if (!PageUptodate(page)) {
55c69072
CM		 403 WARN_ON(1);
404 goto err;
19c00ddc 405 }
19c00ddc 406 csum_tree_block(root, eb, 0);
55c69072 407err:
19c00ddc
CM		 408 free_extent_buffer(eb);
409out:
410 return 0;
411}
412
2b82032c
YZ		 413static int check_tree_block_fsid(struct btrfs_root *root,
414 struct extent_buffer *eb)
415{
416 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
417 u8 fsid[BTRFS_UUID_SIZE];
418 int ret = 1;
419
420 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
421 BTRFS_FSID_SIZE);
422 while (fs_devices) {
423 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
424 ret = 0;
425 break;
426 }
427 fs_devices = fs_devices->seed;
428 }
429 return ret;
430}
431
a826d6dc
JB		 432#define CORRUPT(reason, eb, root, slot) \
433 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
434 "root=%llu, slot=%d\n", reason, \
435 (unsigned long long)btrfs_header_bytenr(eb), \
436 (unsigned long long)root->objectid, slot)
437
438static noinline int check_leaf(struct btrfs_root *root,
439 struct extent_buffer *leaf)
440{
441 struct btrfs_key key;
442 struct btrfs_key leaf_key;
443 u32 nritems = btrfs_header_nritems(leaf);
444 int slot;
445
446 if (nritems == 0)
447 return 0;
448
449 /* Check the 0 item */
450 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
451 BTRFS_LEAF_DATA_SIZE(root)) {
452 CORRUPT("invalid item offset size pair", leaf, root, 0);
453 return -EIO;
454 }
455
456 /*
457 * Check to make sure each items keys are in the correct order and their
458 * offsets make sense. We only have to loop through nritems-1 because
459 * we check the current slot against the next slot, which verifies the
460 * next slot's offset+size makes sense and that the current's slot
461 * offset is correct.
462 */
463 for (slot = 0; slot < nritems - 1; slot++) {
464 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
465 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
466
467 /* Make sure the keys are in the right order */
468 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
469 CORRUPT("bad key order", leaf, root, slot);
470 return -EIO;
471 }
472
473 /*
474 * Make sure the offset and ends are right, remember that the
475 * item data starts at the end of the leaf and grows towards the
476 * front.
477 */
478 if (btrfs_item_offset_nr(leaf, slot) !=
479 btrfs_item_end_nr(leaf, slot + 1)) {
480 CORRUPT("slot offset bad", leaf, root, slot);
481 return -EIO;
482 }
483
484 /*
485 * Check to make sure that we don't point outside of the leaf,
486 * just incase all the items are consistent to eachother, but
487 * all point outside of the leaf.
488 */
489 if (btrfs_item_end_nr(leaf, slot) >
490 BTRFS_LEAF_DATA_SIZE(root)) {
491 CORRUPT("slot end outside of leaf", leaf, root, slot);
492 return -EIO;
493 }
494 }
495
496 return 0;
497}
498
4008c04a
CM		 499#ifdef CONFIG_DEBUG_LOCK_ALLOC
500void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
501{
502 lockdep_set_class_and_name(&eb->lock,
503 &btrfs_eb_class[level],
504 btrfs_eb_name[level]);
505}
506#endif
507
b2950863 508static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
ce9adaa5
CM		 509 struct extent_state *state)
510{
511 struct extent_io_tree *tree;
512 u64 found_start;
513 int found_level;
514 unsigned long len;
515 struct extent_buffer *eb;
516 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 517 int ret = 0;
ce9adaa5
CM		 518
519 tree = &BTRFS_I(page->mapping->host)->io_tree;
520 if (page->private == EXTENT_PAGE_PRIVATE)
521 goto out;
522 if (!page->private)
523 goto out;
d397712b 524
ce9adaa5 525 len = page->private >> 2;
d397712b
CM		 526 WARN_ON(len == 0);
527
ce9adaa5 528 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
91ca338d
TI		 529 if (eb == NULL) {
530 ret = -EIO;
531 goto out;
532 }
f188591e 533
ce9adaa5 534 found_start = btrfs_header_bytenr(eb);
23a07867 535 if (found_start != start) {
193f284d
CM		 536 if (printk_ratelimit()) {
537 printk(KERN_INFO "btrfs bad tree block start "
538 "%llu %llu\n",
539 (unsigned long long)found_start,
540 (unsigned long long)eb->start);
541 }
f188591e 542 ret = -EIO;
ce9adaa5
CM		 543 goto err;
544 }
545 if (eb->first_page != page) {
d397712b
CM		 546 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
547 eb->first_page->index, page->index);
ce9adaa5 548 WARN_ON(1);
f188591e 549 ret = -EIO;
ce9adaa5
CM		 550 goto err;
551 }
2b82032c 552 if (check_tree_block_fsid(root, eb)) {
193f284d
CM		 553 if (printk_ratelimit()) {
554 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
555 (unsigned long long)eb->start);
556 }
1259ab75
CM		 557 ret = -EIO;
558 goto err;
559 }
ce9adaa5
CM		 560 found_level = btrfs_header_level(eb);
561
4008c04a
CM		 562 btrfs_set_buffer_lockdep_class(eb, found_level);
563
ce9adaa5 564 ret = csum_tree_block(root, eb, 1);
a826d6dc 565 if (ret) {
f188591e 566 ret = -EIO;
a826d6dc
JB		 567 goto err;
568 }
569
570 /*
571 * If this is a leaf block and it is corrupt, set the corrupt bit so
572 * that we don't try and read the other copies of this block, just
573 * return -EIO.
574 */
575 if (found_level == 0 && check_leaf(root, eb)) {
576 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
577 ret = -EIO;
578 }
ce9adaa5
CM		 579
580 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
581 end = eb->start + end - 1;
ce9adaa5
CM		 582err:
583 free_extent_buffer(eb);
584out:
f188591e 585 return ret;
ce9adaa5
CM		 586}
587
ce9adaa5 588static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM		 589{
590 struct end_io_wq *end_io_wq = bio->bi_private;
591 struct btrfs_fs_info *fs_info;
ce9adaa5 592
ce9adaa5 593 fs_info = end_io_wq->info;
ce9adaa5 594 end_io_wq->error = err;
8b712842
CM		 595 end_io_wq->work.func = end_workqueue_fn;
596 end_io_wq->work.flags = 0;
d20f7043 597
7b6d91da 598 if (bio->bi_rw & REQ_WRITE) {
0cb59c99 599 if (end_io_wq->metadata == 1)
cad321ad
CM		 600 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
601 &end_io_wq->work);
0cb59c99
JB		 602 else if (end_io_wq->metadata == 2)
603 btrfs_queue_worker(&fs_info->endio_freespace_worker,
604 &end_io_wq->work);
cad321ad
CM		 605 else
606 btrfs_queue_worker(&fs_info->endio_write_workers,
607 &end_io_wq->work);
d20f7043
CM		 608 } else {
609 if (end_io_wq->metadata)
610 btrfs_queue_worker(&fs_info->endio_meta_workers,
611 &end_io_wq->work);
612 else
613 btrfs_queue_worker(&fs_info->endio_workers,
614 &end_io_wq->work);
615 }
ce9adaa5
CM		 616}
617
0cb59c99
JB		 618/*
619 * For the metadata arg you want
620 *
621 * 0 - if data
622 * 1 - if normal metadta
623 * 2 - if writing to the free space cache area
624 */
22c59948
CM		 625int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
626 int metadata)
0b86a832 627{
ce9adaa5 628 struct end_io_wq *end_io_wq;
ce9adaa5
CM		 629 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
630 if (!end_io_wq)
631 return -ENOMEM;
632
633 end_io_wq->private = bio->bi_private;
634 end_io_wq->end_io = bio->bi_end_io;
22c59948 635 end_io_wq->info = info;
ce9adaa5
CM		 636 end_io_wq->error = 0;
637 end_io_wq->bio = bio;
22c59948 638 end_io_wq->metadata = metadata;
ce9adaa5
CM		 639
640 bio->bi_private = end_io_wq;
641 bio->bi_end_io = end_workqueue_bio;
22c59948
CM		 642 return 0;
643}
644
b64a2851 645unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 646{
4854ddd0
CM		 647 unsigned long limit = min_t(unsigned long,
648 info->workers.max_workers,
649 info->fs_devices->open_devices);
650 return 256 * limit;
651}
0986fe9e 652
4854ddd0
CM		 653int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
654{
b64a2851
CM		 655 return atomic_read(&info->nr_async_bios) >
656 btrfs_async_submit_limit(info);
0986fe9e
CM		 657}
658
4a69a410
CM		 659static void run_one_async_start(struct btrfs_work *work)
660{
4a69a410
CM		 661 struct async_submit_bio *async;
662
663 async = container_of(work, struct async_submit_bio, work);
4a69a410 664 async->submit_bio_start(async->inode, async->rw, async->bio,
eaf25d93
CM		 665 async->mirror_num, async->bio_flags,
666 async->bio_offset);
4a69a410
CM		 667}
668
669static void run_one_async_done(struct btrfs_work *work)
8b712842
CM		 670{
671 struct btrfs_fs_info *fs_info;
672 struct async_submit_bio *async;
4854ddd0 673 int limit;
8b712842
CM		 674
675 async = container_of(work, struct async_submit_bio, work);
676 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 677
b64a2851 678 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM		 679 limit = limit * 2 / 3;
680
8b712842 681 atomic_dec(&fs_info->nr_async_submits);
0986fe9e 682
b64a2851
CM		 683 if (atomic_read(&fs_info->nr_async_submits) < limit &&
684 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM		 685 wake_up(&fs_info->async_submit_wait);
686
4a69a410 687 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM		 688 async->mirror_num, async->bio_flags,
689 async->bio_offset);
4a69a410
CM		 690}
691
692static void run_one_async_free(struct btrfs_work *work)
693{
694 struct async_submit_bio *async;
695
696 async = container_of(work, struct async_submit_bio, work);
8b712842
CM		 697 kfree(async);
698}
699
44b8bd7e
CM		 700int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
701 int rw, struct bio *bio, int mirror_num,
c8b97818 702 unsigned long bio_flags,
eaf25d93 703 u64 bio_offset,
4a69a410
CM		 704 extent_submit_bio_hook_t *submit_bio_start,
705 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM		 706{
707 struct async_submit_bio *async;
708
709 async = kmalloc(sizeof(*async), GFP_NOFS);
710 if (!async)
711 return -ENOMEM;
712
713 async->inode = inode;
714 async->rw = rw;
715 async->bio = bio;
716 async->mirror_num = mirror_num;
4a69a410
CM		 717 async->submit_bio_start = submit_bio_start;
718 async->submit_bio_done = submit_bio_done;
719
720 async->work.func = run_one_async_start;
721 async->work.ordered_func = run_one_async_done;
722 async->work.ordered_free = run_one_async_free;
723
8b712842 724 async->work.flags = 0;
c8b97818 725 async->bio_flags = bio_flags;
eaf25d93 726 async->bio_offset = bio_offset;
8c8bee1d 727
cb03c743 728 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 729
7b6d91da 730 if (rw & REQ_SYNC)
d313d7a3
CM		 731 btrfs_set_work_high_prio(&async->work);
732
8b712842 733 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 734
d397712b 735 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM		 736 atomic_read(&fs_info->nr_async_submits)) {
737 wait_event(fs_info->async_submit_wait,
738 (atomic_read(&fs_info->nr_async_submits) == 0));
739 }
740
44b8bd7e
CM		 741 return 0;
742}
743
ce3ed71a
CM		 744static int btree_csum_one_bio(struct bio *bio)
745{
746 struct bio_vec *bvec = bio->bi_io_vec;
747 int bio_index = 0;
748 struct btrfs_root *root;
749
750 WARN_ON(bio->bi_vcnt <= 0);
d397712b 751 while (bio_index < bio->bi_vcnt) {
ce3ed71a
CM		 752 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
753 csum_dirty_buffer(root, bvec->bv_page);
754 bio_index++;
755 bvec++;
756 }
757 return 0;
758}
759
4a69a410
CM		 760static int __btree_submit_bio_start(struct inode *inode, int rw,
761 struct bio *bio, int mirror_num,
eaf25d93
CM		 762 unsigned long bio_flags,
763 u64 bio_offset)
22c59948 764{
8b712842
CM		 765 /*
766 * when we're called for a write, we're already in the async
5443be45 767 * submission context. Just jump into btrfs_map_bio
8b712842 768 */
4a69a410
CM		 769 btree_csum_one_bio(bio);
770 return 0;
771}
22c59948 772
4a69a410 773static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 774 int mirror_num, unsigned long bio_flags,
775 u64 bio_offset)
4a69a410 776{
8b712842 777 /*
4a69a410
CM		 778 * when we're called for a write, we're already in the async
779 * submission context. Just jump into btrfs_map_bio
8b712842 780 */
8b712842 781 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
0b86a832
CM		 782}
783
44b8bd7e 784static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM		 785 int mirror_num, unsigned long bio_flags,
786 u64 bio_offset)
44b8bd7e 787{
cad321ad
CM		 788 int ret;
789
790 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
791 bio, 1);
792 BUG_ON(ret);
793
7b6d91da 794 if (!(rw & REQ_WRITE)) {
4a69a410
CM		 795 /*
796 * called for a read, do the setup so that checksum validation
797 * can happen in the async kernel threads
798 */
4a69a410 799 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
6f3577bd 800 mirror_num, 0);
44b8bd7e 801 }
d313d7a3 802
cad321ad
CM		 803 /*
804 * kthread helpers are used to submit writes so that checksumming
805 * can happen in parallel across all CPUs
806 */
44b8bd7e 807 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
c8b97818 808 inode, rw, bio, mirror_num, 0,
eaf25d93 809 bio_offset,
4a69a410
CM		 810 __btree_submit_bio_start,
811 __btree_submit_bio_done);
44b8bd7e
CM		 812}
813
3dd1462e 814#ifdef CONFIG_MIGRATION
784b4e29
CM		 815static int btree_migratepage(struct address_space *mapping,
816 struct page *newpage, struct page *page)
817{
818 /*
819 * we can't safely write a btree page from here,
820 * we haven't done the locking hook
821 */
822 if (PageDirty(page))
823 return -EAGAIN;
824 /*
825 * Buffers may be managed in a filesystem specific way.
826 * We must have no buffers or drop them.
827 */
828 if (page_has_private(page) &&
829 !try_to_release_page(page, GFP_KERNEL))
830 return -EAGAIN;
784b4e29
CM		 831 return migrate_page(mapping, newpage, page);
832}
3dd1462e 833#endif
784b4e29 834
0da5468f
CM		 835static int btree_writepage(struct page *page, struct writeback_control *wbc)
836{
d1310b2e 837 struct extent_io_tree *tree;
b9473439
CM		 838 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
839 struct extent_buffer *eb;
840 int was_dirty;
841
d1310b2e 842 tree = &BTRFS_I(page->mapping->host)->io_tree;
b9473439
CM		 843 if (!(current->flags & PF_MEMALLOC)) {
844 return extent_write_full_page(tree, page,
845 btree_get_extent, wbc);
846 }
5443be45 847
b9473439 848 redirty_page_for_writepage(wbc, page);
784b4e29 849 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
b9473439
CM		 850 WARN_ON(!eb);
851
852 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
853 if (!was_dirty) {
854 spin_lock(&root->fs_info->delalloc_lock);
855 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
856 spin_unlock(&root->fs_info->delalloc_lock);
5443be45 857 }
b9473439
CM		 858 free_extent_buffer(eb);
859
860 unlock_page(page);
861 return 0;
5f39d397 862}
0da5468f
CM		 863
864static int btree_writepages(struct address_space *mapping,
865 struct writeback_control *wbc)
866{
d1310b2e
CM		 867 struct extent_io_tree *tree;
868 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 869 if (wbc->sync_mode == WB_SYNC_NONE) {
b9473439 870 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
793955bc 871 u64 num_dirty;
24ab9cd8 872 unsigned long thresh = 32 * 1024 * 1024;
448d640b
CM		 873
874 if (wbc->for_kupdate)
875 return 0;
876
b9473439
CM		 877 /* this is a bit racy, but that's ok */
878 num_dirty = root->fs_info->dirty_metadata_bytes;
d397712b 879 if (num_dirty < thresh)
793955bc 880 return 0;
793955bc 881 }
0da5468f
CM		 882 return extent_writepages(tree, mapping, btree_get_extent, wbc);
883}
884
b2950863 885static int btree_readpage(struct file *file, struct page *page)
5f39d397 886{
d1310b2e
CM		 887 struct extent_io_tree *tree;
888 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 889 return extent_read_full_page(tree, page, btree_get_extent);
890}
22b0ebda 891
70dec807 892static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 893{
d1310b2e
CM		 894 struct extent_io_tree *tree;
895 struct extent_map_tree *map;
5f39d397 896 int ret;
d98237b3 897
98509cfc 898 if (PageWriteback(page) || PageDirty(page))
d397712b 899 return 0;
98509cfc 900
d1310b2e
CM		 901 tree = &BTRFS_I(page->mapping->host)->io_tree;
902 map = &BTRFS_I(page->mapping->host)->extent_tree;
6af118ce 903
7b13b7b1 904 ret = try_release_extent_state(map, tree, page, gfp_flags);
d397712b 905 if (!ret)
6af118ce 906 return 0;
6af118ce
CM		 907
908 ret = try_release_extent_buffer(tree, page);
5f39d397
CM		 909 if (ret == 1) {
910 ClearPagePrivate(page);
911 set_page_private(page, 0);
912 page_cache_release(page);
913 }
6af118ce 914
d98237b3
CM		 915 return ret;
916}
917
5f39d397 918static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 919{
d1310b2e
CM		 920 struct extent_io_tree *tree;
921 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM		 922 extent_invalidatepage(tree, page, offset);
923 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 924 if (PagePrivate(page)) {
d397712b
CM		 925 printk(KERN_WARNING "btrfs warning page private not zero "
926 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM		 927 ClearPagePrivate(page);
928 set_page_private(page, 0);
929 page_cache_release(page);
930 }
d98237b3
CM		 931}
932
7f09410b 933static const struct address_space_operations btree_aops = {
d98237b3
CM		 934 .readpage = btree_readpage,
935 .writepage = btree_writepage,
0da5468f 936 .writepages = btree_writepages,
5f39d397
CM		 937 .releasepage = btree_releasepage,
938 .invalidatepage = btree_invalidatepage,
d98237b3 939 .sync_page = block_sync_page,
5a92bc88 940#ifdef CONFIG_MIGRATION
784b4e29 941 .migratepage = btree_migratepage,
5a92bc88 942#endif
d98237b3
CM		 943};
944
ca7a79ad
CM		 945int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
946 u64 parent_transid)
090d1875 947{
5f39d397
CM		 948 struct extent_buffer *buf = NULL;
949 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 950 int ret = 0;
090d1875 951
db94535d 952 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 953 if (!buf)
090d1875 954 return 0;
d1310b2e 955 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
f188591e 956 buf, 0, 0, btree_get_extent, 0);
5f39d397 957 free_extent_buffer(buf);
de428b63 958 return ret;
090d1875
CM		 959}
960
0999df54
CM		 961struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
962 u64 bytenr, u32 blocksize)
963{
964 struct inode *btree_inode = root->fs_info->btree_inode;
965 struct extent_buffer *eb;
966 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
967 bytenr, blocksize, GFP_NOFS);
968 return eb;
969}
970
971struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
972 u64 bytenr, u32 blocksize)
973{
974 struct inode *btree_inode = root->fs_info->btree_inode;
975 struct extent_buffer *eb;
976
977 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
978 bytenr, blocksize, NULL, GFP_NOFS);
979 return eb;
980}
981
982
e02119d5
CM		 983int btrfs_write_tree_block(struct extent_buffer *buf)
984{
8aa38c31
CH		 985 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
986 buf->start + buf->len - 1);
e02119d5
CM		 987}
988
989int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
990{
8aa38c31
CH		 991 return filemap_fdatawait_range(buf->first_page->mapping,
992 buf->start, buf->start + buf->len - 1);
e02119d5
CM		 993}
994
0999df54 995struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 996 u32 blocksize, u64 parent_transid)
0999df54
CM		 997{
998 struct extent_buffer *buf = NULL;
0999df54
CM		 999 int ret;
1000
0999df54
CM		 1001 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1002 if (!buf)
1003 return NULL;
0999df54 1004
ca7a79ad 1005 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
ce9adaa5 1006
d397712b 1007 if (ret == 0)
b4ce94de 1008 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
5f39d397 1009 return buf;
ce9adaa5 1010
eb60ceac
CM		 1011}
1012
e089f05c 1013int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5f39d397 1014 struct extent_buffer *buf)
ed2ff2cb 1015{
5f39d397 1016 struct inode *btree_inode = root->fs_info->btree_inode;
55c69072 1017 if (btrfs_header_generation(buf) ==
925baedd 1018 root->fs_info->running_transaction->transid) {
b9447ef8 1019 btrfs_assert_tree_locked(buf);
b4ce94de 1020
b9473439
CM		 1021 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1022 spin_lock(&root->fs_info->delalloc_lock);
1023 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1024 root->fs_info->dirty_metadata_bytes -= buf->len;
1025 else
1026 WARN_ON(1);
1027 spin_unlock(&root->fs_info->delalloc_lock);
1028 }
b4ce94de 1029
b9473439
CM		 1030 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1031 btrfs_set_lock_blocking(buf);
d1310b2e 1032 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
55c69072 1033 buf);
925baedd 1034 }
5f39d397
CM		 1035 return 0;
1036}
1037
db94535d 1038static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
87ee04eb 1039 u32 stripesize, struct btrfs_root *root,
9f5fae2f 1040 struct btrfs_fs_info *fs_info,
e20d96d6 1041 u64 objectid)
d97e63b6 1042{
cfaa7295 1043 root->node = NULL;
a28ec197 1044 root->commit_root = NULL;
db94535d
CM		 1045 root->sectorsize = sectorsize;
1046 root->nodesize = nodesize;
1047 root->leafsize = leafsize;
87ee04eb 1048 root->stripesize = stripesize;
123abc88 1049 root->ref_cows = 0;
0b86a832 1050 root->track_dirty = 0;
c71bf099 1051 root->in_radix = 0;
d68fc57b
YZ		 1052 root->orphan_item_inserted = 0;
1053 root->orphan_cleanup_state = 0;
0b86a832 1054
9f5fae2f 1055 root->fs_info = fs_info;
0f7d52f4
CM		 1056 root->objectid = objectid;
1057 root->last_trans = 0;
13a8a7c8 1058 root->highest_objectid = 0;
58176a96 1059 root->name = NULL;
4313b399 1060 root->in_sysfs = 0;
6bef4d31 1061 root->inode_tree = RB_ROOT;
f0486c68 1062 root->block_rsv = NULL;
d68fc57b 1063 root->orphan_block_rsv = NULL;
0b86a832
CM		 1064
1065 INIT_LIST_HEAD(&root->dirty_list);
7b128766 1066 INIT_LIST_HEAD(&root->orphan_list);
5d4f98a2 1067 INIT_LIST_HEAD(&root->root_list);
925baedd 1068 spin_lock_init(&root->node_lock);
d68fc57b 1069 spin_lock_init(&root->orphan_lock);
5d4f98a2 1070 spin_lock_init(&root->inode_lock);
f0486c68 1071 spin_lock_init(&root->accounting_lock);
a2135011 1072 mutex_init(&root->objectid_mutex);
e02119d5 1073 mutex_init(&root->log_mutex);
7237f183
YZ		 1074 init_waitqueue_head(&root->log_writer_wait);
1075 init_waitqueue_head(&root->log_commit_wait[0]);
1076 init_waitqueue_head(&root->log_commit_wait[1]);
1077 atomic_set(&root->log_commit[0], 0);
1078 atomic_set(&root->log_commit[1], 0);
1079 atomic_set(&root->log_writers, 0);
1080 root->log_batch = 0;
1081 root->log_transid = 0;
257c62e1 1082 root->last_log_commit = 0;
d0c803c4
CM		 1083 extent_io_tree_init(&root->dirty_log_pages,
1084 fs_info->btree_inode->i_mapping, GFP_NOFS);
017e5369 1085
3768f368
CM		 1086 memset(&root->root_key, 0, sizeof(root->root_key));
1087 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1088 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1089 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 1090 root->defrag_trans_start = fs_info->generation;
58176a96 1091 init_completion(&root->kobj_unregister);
6702ed49 1092 root->defrag_running = 0;
4d775673 1093 root->root_key.objectid = objectid;
3394e160
CM		 1094 root->anon_super.s_root = NULL;
1095 root->anon_super.s_dev = 0;
1096 INIT_LIST_HEAD(&root->anon_super.s_list);
1097 INIT_LIST_HEAD(&root->anon_super.s_instances);
1098 init_rwsem(&root->anon_super.s_umount);
1099
3768f368
CM		 1100 return 0;
1101}
1102
db94535d 1103static int find_and_setup_root(struct btrfs_root *tree_root,
9f5fae2f
CM		 1104 struct btrfs_fs_info *fs_info,
1105 u64 objectid,
e20d96d6 1106 struct btrfs_root *root)
3768f368
CM		 1107{
1108 int ret;
db94535d 1109 u32 blocksize;
84234f3a 1110 u64 generation;
3768f368 1111
db94535d 1112 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1113 tree_root->sectorsize, tree_root->stripesize,
1114 root, fs_info, objectid);
3768f368
CM		 1115 ret = btrfs_find_last_root(tree_root, objectid,
1116 &root->root_item, &root->root_key);
4df27c4d
YZ		 1117 if (ret > 0)
1118 return -ENOENT;
3768f368
CM		 1119 BUG_ON(ret);
1120
84234f3a 1121 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1122 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1123 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1124 blocksize, generation);
68433b73
CM		 1125 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1126 free_extent_buffer(root->node);
1127 return -EIO;
1128 }
4df27c4d 1129 root->commit_root = btrfs_root_node(root);
d97e63b6
CM		 1130 return 0;
1131}
1132
7237f183
YZ		 1133static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1134 struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 1135{
1136 struct btrfs_root *root;
1137 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1138 struct extent_buffer *leaf;
e02119d5
CM		 1139
1140 root = kzalloc(sizeof(*root), GFP_NOFS);
1141 if (!root)
7237f183 1142 return ERR_PTR(-ENOMEM);
e02119d5
CM		 1143
1144 __setup_root(tree_root->nodesize, tree_root->leafsize,
1145 tree_root->sectorsize, tree_root->stripesize,
1146 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1147
1148 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1149 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1150 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ		 1151 /*
1152 * log trees do not get reference counted because they go away
1153 * before a real commit is actually done. They do store pointers
1154 * to file data extents, and those reference counts still get
1155 * updated (along with back refs to the log tree).
1156 */
e02119d5
CM		 1157 root->ref_cows = 0;
1158
5d4f98a2
YZ		 1159 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1160 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
7237f183
YZ		 1161 if (IS_ERR(leaf)) {
1162 kfree(root);
1163 return ERR_CAST(leaf);
1164 }
e02119d5 1165
5d4f98a2
YZ		 1166 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1167 btrfs_set_header_bytenr(leaf, leaf->start);
1168 btrfs_set_header_generation(leaf, trans->transid);
1169 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1170 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1171 root->node = leaf;
e02119d5
CM		 1172
1173 write_extent_buffer(root->node, root->fs_info->fsid,
1174 (unsigned long)btrfs_header_fsid(root->node),
1175 BTRFS_FSID_SIZE);
1176 btrfs_mark_buffer_dirty(root->node);
1177 btrfs_tree_unlock(root->node);
7237f183
YZ		 1178 return root;
1179}
1180
1181int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1182 struct btrfs_fs_info *fs_info)
1183{
1184 struct btrfs_root *log_root;
1185
1186 log_root = alloc_log_tree(trans, fs_info);
1187 if (IS_ERR(log_root))
1188 return PTR_ERR(log_root);
1189 WARN_ON(fs_info->log_root_tree);
1190 fs_info->log_root_tree = log_root;
1191 return 0;
1192}
1193
1194int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1195 struct btrfs_root *root)
1196{
1197 struct btrfs_root *log_root;
1198 struct btrfs_inode_item *inode_item;
1199
1200 log_root = alloc_log_tree(trans, root->fs_info);
1201 if (IS_ERR(log_root))
1202 return PTR_ERR(log_root);
1203
1204 log_root->last_trans = trans->transid;
1205 log_root->root_key.offset = root->root_key.objectid;
1206
1207 inode_item = &log_root->root_item.inode;
1208 inode_item->generation = cpu_to_le64(1);
1209 inode_item->size = cpu_to_le64(3);
1210 inode_item->nlink = cpu_to_le32(1);
1211 inode_item->nbytes = cpu_to_le64(root->leafsize);
1212 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1213
5d4f98a2 1214 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ		 1215
1216 WARN_ON(root->log_root);
1217 root->log_root = log_root;
1218 root->log_transid = 0;
257c62e1 1219 root->last_log_commit = 0;
e02119d5
CM		 1220 return 0;
1221}
1222
1223struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1224 struct btrfs_key *location)
1225{
1226 struct btrfs_root *root;
1227 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1228 struct btrfs_path *path;
5f39d397 1229 struct extent_buffer *l;
84234f3a 1230 u64 generation;
db94535d 1231 u32 blocksize;
0f7d52f4
CM		 1232 int ret = 0;
1233
5eda7b5e 1234 root = kzalloc(sizeof(*root), GFP_NOFS);
0cf6c620 1235 if (!root)
0f7d52f4 1236 return ERR_PTR(-ENOMEM);
0f7d52f4 1237 if (location->offset == (u64)-1) {
db94535d 1238 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM		 1239 location->objectid, root);
1240 if (ret) {
0f7d52f4
CM		 1241 kfree(root);
1242 return ERR_PTR(ret);
1243 }
13a8a7c8 1244 goto out;
0f7d52f4
CM		 1245 }
1246
db94535d 1247 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM		 1248 tree_root->sectorsize, tree_root->stripesize,
1249 root, fs_info, location->objectid);
0f7d52f4
CM		 1250
1251 path = btrfs_alloc_path();
db5b493a
TI		 1252 if (!path) {
1253 kfree(root);
1254 return ERR_PTR(-ENOMEM);
1255 }
0f7d52f4 1256 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
13a8a7c8
YZ		 1257 if (ret == 0) {
1258 l = path->nodes[0];
1259 read_extent_buffer(l, &root->root_item,
1260 btrfs_item_ptr_offset(l, path->slots[0]),
1261 sizeof(root->root_item));
1262 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4 1263 }
0f7d52f4
CM		 1264 btrfs_free_path(path);
1265 if (ret) {
5e540f77 1266 kfree(root);
13a8a7c8
YZ		 1267 if (ret > 0)
1268 ret = -ENOENT;
0f7d52f4
CM		 1269 return ERR_PTR(ret);
1270 }
13a8a7c8 1271
84234f3a 1272 generation = btrfs_root_generation(&root->root_item);
db94535d
CM		 1273 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1274 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1275 blocksize, generation);
5d4f98a2 1276 root->commit_root = btrfs_root_node(root);
0f7d52f4 1277 BUG_ON(!root->node);
13a8a7c8 1278out:
08fe4db1 1279 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1280 root->ref_cows = 1;
08fe4db1
LZ		 1281 btrfs_check_and_init_root_item(&root->root_item);
1282 }
13a8a7c8 1283
5eda7b5e
CM		 1284 return root;
1285}
1286
dc17ff8f
CM		 1287struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1288 u64 root_objectid)
1289{
1290 struct btrfs_root *root;
1291
1292 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1293 return fs_info->tree_root;
1294 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1295 return fs_info->extent_root;
1296
1297 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1298 (unsigned long)root_objectid);
1299 return root;
1300}
1301
edbd8d4e
CM		 1302struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1303 struct btrfs_key *location)
5eda7b5e
CM		 1304{
1305 struct btrfs_root *root;
1306 int ret;
1307
edbd8d4e
CM		 1308 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1309 return fs_info->tree_root;
1310 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1311 return fs_info->extent_root;
8f18cf13
CM		 1312 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1313 return fs_info->chunk_root;
1314 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1315 return fs_info->dev_root;
0403e47e
YZ		 1316 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1317 return fs_info->csum_root;
4df27c4d
YZ		 1318again:
1319 spin_lock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1320 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1321 (unsigned long)location->objectid);
4df27c4d 1322 spin_unlock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM		 1323 if (root)
1324 return root;
1325
e02119d5 1326 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM		 1327 if (IS_ERR(root))
1328 return root;
3394e160
CM		 1329
1330 set_anon_super(&root->anon_super, NULL);
1331
d68fc57b
YZ		 1332 if (btrfs_root_refs(&root->root_item) == 0) {
1333 ret = -ENOENT;
1334 goto fail;
1335 }
1336
1337 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1338 if (ret < 0)
1339 goto fail;
1340 if (ret == 0)
1341 root->orphan_item_inserted = 1;
1342
4df27c4d
YZ		 1343 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1344 if (ret)
1345 goto fail;
1346
1347 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 1348 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1349 (unsigned long)root->root_key.objectid,
0f7d52f4 1350 root);
d68fc57b 1351 if (ret == 0)
4df27c4d 1352 root->in_radix = 1;
d68fc57b 1353
4df27c4d
YZ		 1354 spin_unlock(&fs_info->fs_roots_radix_lock);
1355 radix_tree_preload_end();
0f7d52f4 1356 if (ret) {
4df27c4d
YZ		 1357 if (ret == -EEXIST) {
1358 free_fs_root(root);
1359 goto again;
1360 }
1361 goto fail;
0f7d52f4 1362 }
4df27c4d
YZ		 1363
1364 ret = btrfs_find_dead_roots(fs_info->tree_root,
1365 root->root_key.objectid);
1366 WARN_ON(ret);
edbd8d4e 1367 return root;
4df27c4d
YZ		 1368fail:
1369 free_fs_root(root);
1370 return ERR_PTR(ret);
edbd8d4e
CM		 1371}
1372
1373struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1374 struct btrfs_key *location,
1375 const char *name, int namelen)
1376{
4df27c4d
YZ		 1377 return btrfs_read_fs_root_no_name(fs_info, location);
1378#if 0
edbd8d4e
CM		 1379 struct btrfs_root *root;
1380 int ret;
1381
1382 root = btrfs_read_fs_root_no_name(fs_info, location);
1383 if (!root)
1384 return NULL;
58176a96 1385
4313b399
CM		 1386 if (root->in_sysfs)
1387 return root;
1388
58176a96
JB		 1389 ret = btrfs_set_root_name(root, name, namelen);
1390 if (ret) {
5f39d397 1391 free_extent_buffer(root->node);
58176a96
JB		 1392 kfree(root);
1393 return ERR_PTR(ret);
1394 }
4df27c4d 1395
58176a96
JB		 1396 ret = btrfs_sysfs_add_root(root);
1397 if (ret) {
5f39d397 1398 free_extent_buffer(root->node);
58176a96
JB		 1399 kfree(root->name);
1400 kfree(root);
1401 return ERR_PTR(ret);
1402 }
4313b399 1403 root->in_sysfs = 1;
0f7d52f4 1404 return root;
4df27c4d 1405#endif
0f7d52f4 1406}
04160088
CM		 1407
1408static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1409{
1410 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1411 int ret = 0;
04160088
CM		 1412 struct btrfs_device *device;
1413 struct backing_dev_info *bdi;
b7967db7 1414
c6e30871 1415 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM		 1416 if (!device->bdev)
1417 continue;
04160088
CM		 1418 bdi = blk_get_backing_dev_info(device->bdev);
1419 if (bdi && bdi_congested(bdi, bdi_bits)) {
1420 ret = 1;
1421 break;
1422 }
1423 }
1424 return ret;
1425}
1426
38b66988
CM		 1427/*
1428 * this unplugs every device on the box, and it is only used when page
1429 * is null
1430 */
1431static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1432{
38b66988
CM		 1433 struct btrfs_device *device;
1434 struct btrfs_fs_info *info;
1435
1436 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
c6e30871 1437 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
d20f7043
CM		 1438 if (!device->bdev)
1439 continue;
1440
38b66988 1441 bdi = blk_get_backing_dev_info(device->bdev);
d397712b 1442 if (bdi->unplug_io_fn)
38b66988 1443 bdi->unplug_io_fn(bdi, page);
38b66988
CM		 1444 }
1445}
1446
b2950863 1447static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
04160088 1448{
38b66988 1449 struct inode *inode;
f2d8d74d
CM		 1450 struct extent_map_tree *em_tree;
1451 struct extent_map *em;
bcbfce8a 1452 struct address_space *mapping;
38b66988
CM		 1453 u64 offset;
1454
bcbfce8a 1455 /* the generic O_DIRECT read code does this */
9f0ba5bd 1456 if (1 || !page) {
38b66988
CM		 1457 __unplug_io_fn(bdi, page);
1458 return;
1459 }
1460
bcbfce8a
CM		 1461 /*
1462 * page->mapping may change at any time. Get a consistent copy
1463 * and use that for everything below
1464 */
1465 smp_mb();
1466 mapping = page->mapping;
1467 if (!mapping)
1468 return;
1469
1470 inode = mapping->host;
240d5d48
CM		 1471
1472 /*
1473 * don't do the expensive searching for a small number of
1474 * devices
1475 */
1476 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1477 __unplug_io_fn(bdi, page);
1478 return;
1479 }
1480
38b66988 1481 offset = page_offset(page);
04160088 1482
f2d8d74d 1483 em_tree = &BTRFS_I(inode)->extent_tree;
890871be 1484 read_lock(&em_tree->lock);
f2d8d74d 1485 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
890871be 1486 read_unlock(&em_tree->lock);
89642229
CM		 1487 if (!em) {
1488 __unplug_io_fn(bdi, page);
f2d8d74d 1489 return;
89642229 1490 }
f2d8d74d 1491
89642229
CM		 1492 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1493 free_extent_map(em);
1494 __unplug_io_fn(bdi, page);
1495 return;
1496 }
f2d8d74d
CM		 1497 offset = offset - em->start;
1498 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1499 em->block_start + offset, page);
1500 free_extent_map(em);
04160088
CM		 1501}
1502
ad081f14
JA		 1503/*
1504 * If this fails, caller must call bdi_destroy() to get rid of the
1505 * bdi again.
1506 */
04160088
CM		 1507static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1508{
ad081f14
JA		 1509 int err;
1510
1511 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1512 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA		 1513 if (err)
1514 return err;
1515
4575c9cc 1516 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM		 1517 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1518 bdi->unplug_io_data = info;
1519 bdi->congested_fn = btrfs_congested_fn;
1520 bdi->congested_data = info;
1521 return 0;
1522}
1523
ce9adaa5
CM		 1524static int bio_ready_for_csum(struct bio *bio)
1525{
1526 u64 length = 0;
1527 u64 buf_len = 0;
1528 u64 start = 0;
1529 struct page *page;
1530 struct extent_io_tree *io_tree = NULL;
ce9adaa5
CM		 1531 struct bio_vec *bvec;
1532 int i;
1533 int ret;
1534
1535 bio_for_each_segment(bvec, bio, i) {
1536 page = bvec->bv_page;
1537 if (page->private == EXTENT_PAGE_PRIVATE) {
1538 length += bvec->bv_len;
1539 continue;
1540 }
1541 if (!page->private) {
1542 length += bvec->bv_len;
1543 continue;
1544 }
1545 length = bvec->bv_len;
1546 buf_len = page->private >> 2;
1547 start = page_offset(page) + bvec->bv_offset;
1548 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
ce9adaa5
CM		 1549 }
1550 /* are we fully contained in this bio? */
1551 if (buf_len <= length)
1552 return 1;
1553
1554 ret = extent_range_uptodate(io_tree, start + length,
1555 start + buf_len - 1);
ce9adaa5
CM		 1556 return ret;
1557}
1558
8b712842
CM		 1559/*
1560 * called by the kthread helper functions to finally call the bio end_io
1561 * functions. This is where read checksum verification actually happens
1562 */
1563static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1564{
ce9adaa5 1565 struct bio *bio;
8b712842
CM		 1566 struct end_io_wq *end_io_wq;
1567 struct btrfs_fs_info *fs_info;
ce9adaa5 1568 int error;
ce9adaa5 1569
8b712842
CM		 1570 end_io_wq = container_of(work, struct end_io_wq, work);
1571 bio = end_io_wq->bio;
1572 fs_info = end_io_wq->info;
ce9adaa5 1573
cad321ad 1574 /* metadata bio reads are special because the whole tree block must
8b712842
CM		 1575 * be checksummed at once. This makes sure the entire block is in
1576 * ram and up to date before trying to verify things. For
1577 * blocksize <= pagesize, it is basically a noop
1578 */
7b6d91da 1579 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
cad321ad 1580 !bio_ready_for_csum(bio)) {
d20f7043 1581 btrfs_queue_worker(&fs_info->endio_meta_workers,
8b712842
CM		 1582 &end_io_wq->work);
1583 return;
1584 }
1585 error = end_io_wq->error;
1586 bio->bi_private = end_io_wq->private;
1587 bio->bi_end_io = end_io_wq->end_io;
1588 kfree(end_io_wq);
8b712842 1589 bio_endio(bio, error);
44b8bd7e
CM		 1590}
1591
a74a4b97
CM		 1592static int cleaner_kthread(void *arg)
1593{
1594 struct btrfs_root *root = arg;
1595
1596 do {
a74a4b97 1597 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
76dda93c
YZ		 1598
1599 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1600 mutex_trylock(&root->fs_info->cleaner_mutex)) {
24bbcf04 1601 btrfs_run_delayed_iputs(root);
76dda93c
YZ		 1602 btrfs_clean_old_snapshots(root);
1603 mutex_unlock(&root->fs_info->cleaner_mutex);
1604 }
a74a4b97
CM		 1605
1606 if (freezing(current)) {
1607 refrigerator();
1608 } else {
a74a4b97 1609 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1610 if (!kthread_should_stop())
1611 schedule();
a74a4b97
CM		 1612 __set_current_state(TASK_RUNNING);
1613 }
1614 } while (!kthread_should_stop());
1615 return 0;
1616}
1617
1618static int transaction_kthread(void *arg)
1619{
1620 struct btrfs_root *root = arg;
1621 struct btrfs_trans_handle *trans;
1622 struct btrfs_transaction *cur;
8929ecfa 1623 u64 transid;
a74a4b97
CM		 1624 unsigned long now;
1625 unsigned long delay;
1626 int ret;
1627
1628 do {
a74a4b97
CM		 1629 delay = HZ * 30;
1630 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1631 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1632
8929ecfa 1633 spin_lock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1634 cur = root->fs_info->running_transaction;
1635 if (!cur) {
8929ecfa 1636 spin_unlock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1637 goto sleep;
1638 }
31153d81 1639
a74a4b97 1640 now = get_seconds();
8929ecfa
YZ		 1641 if (!cur->blocked &&
1642 (now < cur->start_time || now - cur->start_time < 30)) {
1643 spin_unlock(&root->fs_info->new_trans_lock);
a74a4b97
CM		 1644 delay = HZ * 5;
1645 goto sleep;
1646 }
8929ecfa
YZ		 1647 transid = cur->transid;
1648 spin_unlock(&root->fs_info->new_trans_lock);
56bec294 1649
8929ecfa 1650 trans = btrfs_join_transaction(root, 1);
3612b495 1651 BUG_ON(IS_ERR(trans));
8929ecfa
YZ		 1652 if (transid == trans->transid) {
1653 ret = btrfs_commit_transaction(trans, root);
1654 BUG_ON(ret);
1655 } else {
1656 btrfs_end_transaction(trans, root);
1657 }
a74a4b97
CM		 1658sleep:
1659 wake_up_process(root->fs_info->cleaner_kthread);
1660 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1661
1662 if (freezing(current)) {
1663 refrigerator();
1664 } else {
a74a4b97 1665 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ		 1666 if (!kthread_should_stop() &&
1667 !btrfs_transaction_blocked(root->fs_info))
1668 schedule_timeout(delay);
a74a4b97
CM		 1669 __set_current_state(TASK_RUNNING);
1670 }
1671 } while (!kthread_should_stop());
1672 return 0;
1673}
1674
8a4b83cc 1675struct btrfs_root *open_ctree(struct super_block *sb,
dfe25020
CM		 1676 struct btrfs_fs_devices *fs_devices,
1677 char *options)
2e635a27 1678{
db94535d
CM		 1679 u32 sectorsize;
1680 u32 nodesize;
1681 u32 leafsize;
1682 u32 blocksize;
87ee04eb 1683 u32 stripesize;
84234f3a 1684 u64 generation;
f2b636e8 1685 u64 features;
3de4586c 1686 struct btrfs_key location;
a061fc8d 1687 struct buffer_head *bh;
e02119d5 1688 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1689 GFP_NOFS);
d20f7043
CM		 1690 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1691 GFP_NOFS);
450ba0ea
JB		 1692 struct btrfs_root *tree_root = btrfs_sb(sb);
1693 struct btrfs_fs_info *fs_info = tree_root->fs_info;
e02119d5 1694 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1695 GFP_NOFS);
e02119d5 1696 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1697 GFP_NOFS);
e02119d5
CM		 1698 struct btrfs_root *log_tree_root;
1699
eb60ceac 1700 int ret;
e58ca020 1701 int err = -EINVAL;
4543df7e 1702
2c90e5d6 1703 struct btrfs_super_block *disk_super;
8790d502 1704
0463bb4e 1705 if (!extent_root || !tree_root || !fs_info ||
d20f7043 1706 !chunk_root || !dev_root || !csum_root) {
39279cc3
CM		 1707 err = -ENOMEM;
1708 goto fail;
1709 }
76dda93c
YZ		 1710
1711 ret = init_srcu_struct(&fs_info->subvol_srcu);
1712 if (ret) {
1713 err = ret;
1714 goto fail;
1715 }
1716
1717 ret = setup_bdi(fs_info, &fs_info->bdi);
1718 if (ret) {
1719 err = ret;
1720 goto fail_srcu;
1721 }
1722
1723 fs_info->btree_inode = new_inode(sb);
1724 if (!fs_info->btree_inode) {
1725 err = -ENOMEM;
1726 goto fail_bdi;
1727 }
1728
1561deda
MX		 1729 fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
1730
76dda93c 1731 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 1732 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 1733 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 1734 INIT_LIST_HEAD(&fs_info->delayed_iputs);
19c00ddc 1735 INIT_LIST_HEAD(&fs_info->hashers);
ea8c2819 1736 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
5a3f23d5 1737 INIT_LIST_HEAD(&fs_info->ordered_operations);
11833d66 1738 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1832a6d5 1739 spin_lock_init(&fs_info->delalloc_lock);
cee36a03 1740 spin_lock_init(&fs_info->new_trans_lock);
31153d81 1741 spin_lock_init(&fs_info->ref_cache_lock);
76dda93c 1742 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 1743 spin_lock_init(&fs_info->delayed_iput_lock);
19c00ddc 1744
58176a96 1745 init_completion(&fs_info->kobj_unregister);
9f5fae2f
CM		 1746 fs_info->tree_root = tree_root;
1747 fs_info->extent_root = extent_root;
d20f7043 1748 fs_info->csum_root = csum_root;
0b86a832
CM		 1749 fs_info->chunk_root = chunk_root;
1750 fs_info->dev_root = dev_root;
8a4b83cc 1751 fs_info->fs_devices = fs_devices;
0b86a832 1752 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 1753 INIT_LIST_HEAD(&fs_info->space_info);
0b86a832 1754 btrfs_mapping_init(&fs_info->mapping_tree);
f0486c68
YZ		 1755 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1756 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1757 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1758 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1759 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1760 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1761 mutex_init(&fs_info->durable_block_rsv_mutex);
cb03c743 1762 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 1763 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 1764 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 1765 atomic_set(&fs_info->nr_async_bios, 0);
e20d96d6 1766 fs_info->sb = sb;
6f568d35 1767 fs_info->max_inline = 8192 * 1024;
9ed74f2d 1768 fs_info->metadata_ratio = 0;
c8b97818 1769
b34b086c
CM		 1770 fs_info->thread_pool_size = min_t(unsigned long,
1771 num_online_cpus() + 2, 8);
0afbaf8c 1772
3eaa2885
CM		 1773 INIT_LIST_HEAD(&fs_info->ordered_extents);
1774 spin_lock_init(&fs_info->ordered_extent_lock);
1775
a061fc8d
CM		 1776 sb->s_blocksize = 4096;
1777 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 1778 sb->s_bdi = &fs_info->bdi;
a061fc8d 1779
76dda93c
YZ		 1780 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1781 fs_info->btree_inode->i_nlink = 1;
0afbaf8c
CM		 1782 /*
1783 * we set the i_size on the btree inode to the max possible int.
1784 * the real end of the address space is determined by all of
1785 * the devices in the system
1786 */
1787 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 1788 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM		 1789 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1790
5d4f98a2 1791 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 1792 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
5f39d397
CM		 1793 fs_info->btree_inode->i_mapping,
1794 GFP_NOFS);
d1310b2e
CM		 1795 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1796 GFP_NOFS);
1797
1798 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 1799
76dda93c
YZ		 1800 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1801 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1802 sizeof(struct btrfs_key));
1803 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
c65ddb52 1804 insert_inode_hash(fs_info->btree_inode);
76dda93c 1805
0f9dd46c 1806 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 1807 fs_info->block_group_cache_tree = RB_ROOT;
0f9dd46c 1808
11833d66 1809 extent_io_tree_init(&fs_info->freed_extents[0],
1a5bc167 1810 fs_info->btree_inode->i_mapping, GFP_NOFS);
11833d66
YZ		 1811 extent_io_tree_init(&fs_info->freed_extents[1],
1812 fs_info->btree_inode->i_mapping, GFP_NOFS);
1813 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 1814 fs_info->do_barriers = 1;
e18e4809 1815
39279cc3 1816
79154b1b 1817 mutex_init(&fs_info->trans_mutex);
5a3f23d5 1818 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 1819 mutex_init(&fs_info->tree_log_mutex);
925baedd 1820 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM		 1821 mutex_init(&fs_info->transaction_kthread_mutex);
1822 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 1823 mutex_init(&fs_info->volume_mutex);
276e680d 1824 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 1825 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 1826 init_rwsem(&fs_info->subvol_sem);
fa9c0d79
CM		 1827
1828 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1829 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1830
e6dcd2dc 1831 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 1832 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 1833 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 1834 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 1835
0b86a832 1836 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 1837 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 1838
a512bbf8 1839 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY		 1840 if (!bh) {
1841 err = -EINVAL;
39279cc3 1842 goto fail_iput;
20b45077 1843 }
39279cc3 1844
a061fc8d 1845 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
2d69a0f8
YZ		 1846 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1847 sizeof(fs_info->super_for_commit));
a061fc8d 1848 brelse(bh);
5f39d397 1849
a061fc8d 1850 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
0b86a832 1851
5f39d397 1852 disk_super = &fs_info->super_copy;
0f7d52f4 1853 if (!btrfs_super_root(disk_super))
c6e2bac1 1854 goto fail_iput;
0f7d52f4 1855
acce952b 1856 /* check FS state, whether FS is broken. */
1857 fs_info->fs_state |= btrfs_super_flags(disk_super);
1858
1859 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1860
75e7cb7f
LB		 1861 /*
1862 * In the long term, we'll store the compression type in the super
1863 * block, and it'll be used for per file compression control.
1864 */
1865 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1866
2b82032c
YZ		 1867 ret = btrfs_parse_options(tree_root, options);
1868 if (ret) {
1869 err = ret;
c6e2bac1 1870 goto fail_iput;
2b82032c 1871 }
dfe25020 1872
f2b636e8
JB		 1873 features = btrfs_super_incompat_flags(disk_super) &
1874 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1875 if (features) {
1876 printk(KERN_ERR "BTRFS: couldn't mount because of "
1877 "unsupported optional features (%Lx).\n",
21380931 1878 (unsigned long long)features);
f2b636e8 1879 err = -EINVAL;
c6e2bac1 1880 goto fail_iput;
f2b636e8
JB		 1881 }
1882
5d4f98a2 1883 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae
LZ		 1884 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1885 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1886 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1887 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 1888
f2b636e8
JB		 1889 features = btrfs_super_compat_ro_flags(disk_super) &
1890 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1891 if (!(sb->s_flags & MS_RDONLY) && features) {
1892 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1893 "unsupported option features (%Lx).\n",
21380931 1894 (unsigned long long)features);
f2b636e8 1895 err = -EINVAL;
c6e2bac1 1896 goto fail_iput;
f2b636e8 1897 }
61d92c32
CM		 1898
1899 btrfs_init_workers(&fs_info->generic_worker,
1900 "genwork", 1, NULL);
1901
5443be45 1902 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM		 1903 fs_info->thread_pool_size,
1904 &fs_info->generic_worker);
c8b97818 1905
771ed689 1906 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM		 1907 fs_info->thread_pool_size,
1908 &fs_info->generic_worker);
771ed689 1909
5443be45 1910 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 1911 min_t(u64, fs_devices->num_devices,
61d92c32
CM		 1912 fs_info->thread_pool_size),
1913 &fs_info->generic_worker);
61b49440
CM		 1914
1915 /* a higher idle thresh on the submit workers makes it much more
1916 * likely that bios will be send down in a sane order to the
1917 * devices
1918 */
1919 fs_info->submit_workers.idle_thresh = 64;
53863232 1920
771ed689 1921 fs_info->workers.idle_thresh = 16;
4a69a410 1922 fs_info->workers.ordered = 1;
61b49440 1923
771ed689
CM		 1924 fs_info->delalloc_workers.idle_thresh = 2;
1925 fs_info->delalloc_workers.ordered = 1;
1926
61d92c32
CM		 1927 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1928 &fs_info->generic_worker);
5443be45 1929 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM		 1930 fs_info->thread_pool_size,
1931 &fs_info->generic_worker);
d20f7043 1932 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM		 1933 fs_info->thread_pool_size,
1934 &fs_info->generic_worker);
cad321ad 1935 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM		 1936 "endio-meta-write", fs_info->thread_pool_size,
1937 &fs_info->generic_worker);
5443be45 1938 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM		 1939 fs_info->thread_pool_size,
1940 &fs_info->generic_worker);
0cb59c99
JB		 1941 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1942 1, &fs_info->generic_worker);
61b49440
CM		 1943
1944 /*
1945 * endios are largely parallel and should have a very
1946 * low idle thresh
1947 */
1948 fs_info->endio_workers.idle_thresh = 4;
b51912c9
CM		 1949 fs_info->endio_meta_workers.idle_thresh = 4;
1950
9042846b
CM		 1951 fs_info->endio_write_workers.idle_thresh = 2;
1952 fs_info->endio_meta_write_workers.idle_thresh = 2;
1953
4543df7e 1954 btrfs_start_workers(&fs_info->workers, 1);
61d92c32 1955 btrfs_start_workers(&fs_info->generic_worker, 1);
1cc127b5 1956 btrfs_start_workers(&fs_info->submit_workers, 1);
771ed689 1957 btrfs_start_workers(&fs_info->delalloc_workers, 1);
247e743c 1958 btrfs_start_workers(&fs_info->fixup_workers, 1);
9042846b
CM		 1959 btrfs_start_workers(&fs_info->endio_workers, 1);
1960 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1961 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1962 btrfs_start_workers(&fs_info->endio_write_workers, 1);
0cb59c99 1963 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
4543df7e 1964
4575c9cc 1965 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM		 1966 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1967 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 1968
db94535d
CM		 1969 nodesize = btrfs_super_nodesize(disk_super);
1970 leafsize = btrfs_super_leafsize(disk_super);
1971 sectorsize = btrfs_super_sectorsize(disk_super);
87ee04eb 1972 stripesize = btrfs_super_stripesize(disk_super);
db94535d
CM		 1973 tree_root->nodesize = nodesize;
1974 tree_root->leafsize = leafsize;
1975 tree_root->sectorsize = sectorsize;
87ee04eb 1976 tree_root->stripesize = stripesize;
a061fc8d
CM		 1977
1978 sb->s_blocksize = sectorsize;
1979 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 1980
39279cc3
CM		 1981 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1982 sizeof(disk_super->magic))) {
d397712b 1983 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM		 1984 goto fail_sb_buffer;
1985 }
19c00ddc 1986
925baedd 1987 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 1988 ret = btrfs_read_sys_array(tree_root);
925baedd 1989 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 1990 if (ret) {
d397712b
CM		 1991 printk(KERN_WARNING "btrfs: failed to read the system "
1992 "array on %s\n", sb->s_id);
5d4f98a2 1993 goto fail_sb_buffer;
84eed90f 1994 }
0b86a832
CM		 1995
1996 blocksize = btrfs_level_size(tree_root,
1997 btrfs_super_chunk_root_level(disk_super));
84234f3a 1998 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM		 1999
2000 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2001 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2002
2003 chunk_root->node = read_tree_block(chunk_root,
2004 btrfs_super_chunk_root(disk_super),
84234f3a 2005 blocksize, generation);
0b86a832 2006 BUG_ON(!chunk_root->node);
83121942
DW		 2007 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2008 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2009 sb->s_id);
2010 goto fail_chunk_root;
2011 }
5d4f98a2
YZ		 2012 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2013 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 2014
e17cade2 2015 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM		 2016 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2017 BTRFS_UUID_SIZE);
e17cade2 2018
925baedd 2019 mutex_lock(&fs_info->chunk_mutex);
0b86a832 2020 ret = btrfs_read_chunk_tree(chunk_root);
925baedd 2021 mutex_unlock(&fs_info->chunk_mutex);
2b82032c 2022 if (ret) {
d397712b
CM		 2023 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2024 sb->s_id);
2b82032c
YZ		 2025 goto fail_chunk_root;
2026 }
0b86a832 2027
dfe25020
CM		 2028 btrfs_close_extra_devices(fs_devices);
2029
db94535d
CM		 2030 blocksize = btrfs_level_size(tree_root,
2031 btrfs_super_root_level(disk_super));
84234f3a 2032 generation = btrfs_super_generation(disk_super);
0b86a832 2033
e20d96d6 2034 tree_root->node = read_tree_block(tree_root,
db94535d 2035 btrfs_super_root(disk_super),
84234f3a 2036 blocksize, generation);
39279cc3 2037 if (!tree_root->node)
2b82032c 2038 goto fail_chunk_root;
83121942
DW		 2039 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2040 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2041 sb->s_id);
2042 goto fail_tree_root;
2043 }
5d4f98a2
YZ		 2044 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2045 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d
CM		 2046
2047 ret = find_and_setup_root(tree_root, fs_info,
e20d96d6 2048 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
0b86a832 2049 if (ret)
39279cc3 2050 goto fail_tree_root;
0b86a832
CM		 2051 extent_root->track_dirty = 1;
2052
2053 ret = find_and_setup_root(tree_root, fs_info,
2054 BTRFS_DEV_TREE_OBJECTID, dev_root);
0b86a832
CM		 2055 if (ret)
2056 goto fail_extent_root;
5d4f98a2 2057 dev_root->track_dirty = 1;
3768f368 2058
d20f7043
CM		 2059 ret = find_and_setup_root(tree_root, fs_info,
2060 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2061 if (ret)
5d4f98a2 2062 goto fail_dev_root;
d20f7043
CM		 2063
2064 csum_root->track_dirty = 1;
2065
8929ecfa
YZ		 2066 fs_info->generation = generation;
2067 fs_info->last_trans_committed = generation;
2068 fs_info->data_alloc_profile = (u64)-1;
2069 fs_info->metadata_alloc_profile = (u64)-1;
2070 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2071
c59021f8 2072 ret = btrfs_init_space_info(fs_info);
2073 if (ret) {
2074 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2075 goto fail_block_groups;
2076 }
2077
1b1d1f66
JB		 2078 ret = btrfs_read_block_groups(extent_root);
2079 if (ret) {
2080 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2081 goto fail_block_groups;
2082 }
9078a3e1 2083
a74a4b97
CM		 2084 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2085 "btrfs-cleaner");
57506d50 2086 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 2087 goto fail_block_groups;
a74a4b97
CM		 2088
2089 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2090 tree_root,
2091 "btrfs-transaction");
57506d50 2092 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 2093 goto fail_cleaner;
a74a4b97 2094
c289811c
CM		 2095 if (!btrfs_test_opt(tree_root, SSD) &&
2096 !btrfs_test_opt(tree_root, NOSSD) &&
2097 !fs_info->fs_devices->rotating) {
2098 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2099 "mode\n");
2100 btrfs_set_opt(fs_info->mount_opt, SSD);
2101 }
2102
acce952b 2103 /* do not make disk changes in broken FS */
2104 if (btrfs_super_log_root(disk_super) != 0 &&
2105 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
e02119d5
CM		 2106 u64 bytenr = btrfs_super_log_root(disk_super);
2107
7c2ca468 2108 if (fs_devices->rw_devices == 0) {
d397712b
CM		 2109 printk(KERN_WARNING "Btrfs log replay required "
2110 "on RO media\n");
7c2ca468
CM		 2111 err = -EIO;
2112 goto fail_trans_kthread;
2113 }
e02119d5
CM		 2114 blocksize =
2115 btrfs_level_size(tree_root,
2116 btrfs_super_log_root_level(disk_super));
d18a2c44 2117
676e4c86
DC		 2118 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2119 if (!log_tree_root) {
2120 err = -ENOMEM;
2121 goto fail_trans_kthread;
2122 }
e02119d5
CM		 2123
2124 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2125 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2126
2127 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ		 2128 blocksize,
2129 generation + 1);
e02119d5
CM		 2130 ret = btrfs_recover_log_trees(log_tree_root);
2131 BUG_ON(ret);
e556ce2c
YZ		 2132
2133 if (sb->s_flags & MS_RDONLY) {
2134 ret = btrfs_commit_super(tree_root);
2135 BUG_ON(ret);
2136 }
e02119d5 2137 }
1a40e23b 2138
76dda93c
YZ		 2139 ret = btrfs_find_orphan_roots(tree_root);
2140 BUG_ON(ret);
2141
7c2ca468 2142 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b
YZ		 2143 ret = btrfs_cleanup_fs_roots(fs_info);
2144 BUG_ON(ret);
2145
5d4f98a2 2146 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX		 2147 if (ret < 0) {
2148 printk(KERN_WARNING
2149 "btrfs: failed to recover relocation\n");
2150 err = -EINVAL;
2151 goto fail_trans_kthread;
2152 }
7c2ca468 2153 }
1a40e23b 2154
3de4586c
CM		 2155 location.objectid = BTRFS_FS_TREE_OBJECTID;
2156 location.type = BTRFS_ROOT_ITEM_KEY;
2157 location.offset = (u64)-1;
2158
3de4586c
CM		 2159 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2160 if (!fs_info->fs_root)
7c2ca468 2161 goto fail_trans_kthread;
3140c9a3
DC		 2162 if (IS_ERR(fs_info->fs_root)) {
2163 err = PTR_ERR(fs_info->fs_root);
2164 goto fail_trans_kthread;
2165 }
c289811c 2166
e3acc2a6
JB		 2167 if (!(sb->s_flags & MS_RDONLY)) {
2168 down_read(&fs_info->cleanup_work_sem);
66b4ffd1
JB		 2169 err = btrfs_orphan_cleanup(fs_info->fs_root);
2170 if (!err)
2171 err = btrfs_orphan_cleanup(fs_info->tree_root);
e3acc2a6 2172 up_read(&fs_info->cleanup_work_sem);
66b4ffd1
JB		 2173 if (err) {
2174 close_ctree(tree_root);
2175 return ERR_PTR(err);
2176 }
e3acc2a6
JB		 2177 }
2178
0f7d52f4 2179 return tree_root;
39279cc3 2180
7c2ca468
CM		 2181fail_trans_kthread:
2182 kthread_stop(fs_info->transaction_kthread);
3f157a2f 2183fail_cleaner:
a74a4b97 2184 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM		 2185
2186 /*
2187 * make sure we're done with the btree inode before we stop our
2188 * kthreads
2189 */
2190 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2191 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2192
1b1d1f66
JB		 2193fail_block_groups:
2194 btrfs_free_block_groups(fs_info);
d20f7043 2195 free_extent_buffer(csum_root->node);
5d4f98a2
YZ		 2196 free_extent_buffer(csum_root->commit_root);
2197fail_dev_root:
2198 free_extent_buffer(dev_root->node);
2199 free_extent_buffer(dev_root->commit_root);
0b86a832
CM		 2200fail_extent_root:
2201 free_extent_buffer(extent_root->node);
5d4f98a2 2202 free_extent_buffer(extent_root->commit_root);
39279cc3 2203fail_tree_root:
5f39d397 2204 free_extent_buffer(tree_root->node);
5d4f98a2 2205 free_extent_buffer(tree_root->commit_root);
2b82032c
YZ		 2206fail_chunk_root:
2207 free_extent_buffer(chunk_root->node);
5d4f98a2 2208 free_extent_buffer(chunk_root->commit_root);
39279cc3 2209fail_sb_buffer:
61d92c32 2210 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2211 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2212 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2213 btrfs_stop_workers(&fs_info->workers);
2214 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2215 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2216 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2217 btrfs_stop_workers(&fs_info->endio_write_workers);
0cb59c99 2218 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1cc127b5 2219 btrfs_stop_workers(&fs_info->submit_workers);
4543df7e 2220fail_iput:
7c2ca468 2221 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4543df7e 2222 iput(fs_info->btree_inode);
7e662854 2223
dfe25020 2224 btrfs_close_devices(fs_info->fs_devices);
84eed90f 2225 btrfs_mapping_tree_free(&fs_info->mapping_tree);
ad081f14 2226fail_bdi:
7e662854 2227 bdi_destroy(&fs_info->bdi);
76dda93c
YZ		 2228fail_srcu:
2229 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2230fail:
39279cc3
CM		 2231 kfree(extent_root);
2232 kfree(tree_root);
2233 kfree(fs_info);
83afeac4
JM		 2234 kfree(chunk_root);
2235 kfree(dev_root);
d20f7043 2236 kfree(csum_root);
39279cc3 2237 return ERR_PTR(err);
eb60ceac
CM		 2238}
2239
f2984462
CM		 2240static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2241{
2242 char b[BDEVNAME_SIZE];
2243
2244 if (uptodate) {
2245 set_buffer_uptodate(bh);
2246 } else {
c3b9a62c 2247 if (printk_ratelimit()) {
f2984462
CM		 2248 printk(KERN_WARNING "lost page write due to "
2249 "I/O error on %s\n",
2250 bdevname(bh->b_bdev, b));
2251 }
1259ab75
CM		 2252 /* note, we dont' set_buffer_write_io_error because we have
2253 * our own ways of dealing with the IO errors
2254 */
f2984462
CM		 2255 clear_buffer_uptodate(bh);
2256 }
2257 unlock_buffer(bh);
2258 put_bh(bh);
2259}
2260
a512bbf8
YZ		 2261struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2262{
2263 struct buffer_head *bh;
2264 struct buffer_head *latest = NULL;
2265 struct btrfs_super_block *super;
2266 int i;
2267 u64 transid = 0;
2268 u64 bytenr;
2269
2270 /* we would like to check all the supers, but that would make
2271 * a btrfs mount succeed after a mkfs from a different FS.
2272 * So, we need to add a special mount option to scan for
2273 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2274 */
2275 for (i = 0; i < 1; i++) {
2276 bytenr = btrfs_sb_offset(i);
2277 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2278 break;
2279 bh = __bread(bdev, bytenr / 4096, 4096);
2280 if (!bh)
2281 continue;
2282
2283 super = (struct btrfs_super_block *)bh->b_data;
2284 if (btrfs_super_bytenr(super) != bytenr ||
2285 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2286 sizeof(super->magic))) {
2287 brelse(bh);
2288 continue;
2289 }
2290
2291 if (!latest || btrfs_super_generation(super) > transid) {
2292 brelse(latest);
2293 latest = bh;
2294 transid = btrfs_super_generation(super);
2295 } else {
2296 brelse(bh);
2297 }
2298 }
2299 return latest;
2300}
2301
4eedeb75
HH		 2302/*
2303 * this should be called twice, once with wait == 0 and
2304 * once with wait == 1. When wait == 0 is done, all the buffer heads
2305 * we write are pinned.
2306 *
2307 * They are released when wait == 1 is done.
2308 * max_mirrors must be the same for both runs, and it indicates how
2309 * many supers on this one device should be written.
2310 *
2311 * max_mirrors == 0 means to write them all.
2312 */
a512bbf8
YZ		 2313static int write_dev_supers(struct btrfs_device *device,
2314 struct btrfs_super_block *sb,
2315 int do_barriers, int wait, int max_mirrors)
2316{
2317 struct buffer_head *bh;
2318 int i;
2319 int ret;
2320 int errors = 0;
2321 u32 crc;
2322 u64 bytenr;
2323 int last_barrier = 0;
2324
2325 if (max_mirrors == 0)
2326 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2327
2328 /* make sure only the last submit_bh does a barrier */
2329 if (do_barriers) {
2330 for (i = 0; i < max_mirrors; i++) {
2331 bytenr = btrfs_sb_offset(i);
2332 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2333 device->total_bytes)
2334 break;
2335 last_barrier = i;
2336 }
2337 }
2338
2339 for (i = 0; i < max_mirrors; i++) {
2340 bytenr = btrfs_sb_offset(i);
2341 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2342 break;
2343
2344 if (wait) {
2345 bh = __find_get_block(device->bdev, bytenr / 4096,
2346 BTRFS_SUPER_INFO_SIZE);
2347 BUG_ON(!bh);
a512bbf8 2348 wait_on_buffer(bh);
4eedeb75
HH		 2349 if (!buffer_uptodate(bh))
2350 errors++;
2351
2352 /* drop our reference */
2353 brelse(bh);
2354
2355 /* drop the reference from the wait == 0 run */
2356 brelse(bh);
2357 continue;
a512bbf8
YZ		 2358 } else {
2359 btrfs_set_super_bytenr(sb, bytenr);
2360
2361 crc = ~(u32)0;
2362 crc = btrfs_csum_data(NULL, (char *)sb +
2363 BTRFS_CSUM_SIZE, crc,
2364 BTRFS_SUPER_INFO_SIZE -
2365 BTRFS_CSUM_SIZE);
2366 btrfs_csum_final(crc, sb->csum);
2367
4eedeb75
HH		 2368 /*
2369 * one reference for us, and we leave it for the
2370 * caller
2371 */
a512bbf8
YZ		 2372 bh = __getblk(device->bdev, bytenr / 4096,
2373 BTRFS_SUPER_INFO_SIZE);
2374 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2375
4eedeb75 2376 /* one reference for submit_bh */
a512bbf8 2377 get_bh(bh);
4eedeb75
HH		 2378
2379 set_buffer_uptodate(bh);
a512bbf8
YZ		 2380 lock_buffer(bh);
2381 bh->b_end_io = btrfs_end_buffer_write_sync;
2382 }
2383
c3b9a62c
CH		 2384 if (i == last_barrier && do_barriers)
2385 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2386 else
ffbd517d 2387 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8 2388
4eedeb75 2389 if (ret)
a512bbf8 2390 errors++;
a512bbf8
YZ		 2391 }
2392 return errors < i ? 0 : -1;
2393}
2394
2395int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 2396{
e5e9a520 2397 struct list_head *head;
f2984462 2398 struct btrfs_device *dev;
a061fc8d 2399 struct btrfs_super_block *sb;
f2984462 2400 struct btrfs_dev_item *dev_item;
f2984462
CM		 2401 int ret;
2402 int do_barriers;
a236aed1
CM		 2403 int max_errors;
2404 int total_errors = 0;
a061fc8d 2405 u64 flags;
f2984462 2406
a236aed1 2407 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
f2984462
CM		 2408 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2409
a061fc8d
CM		 2410 sb = &root->fs_info->super_for_commit;
2411 dev_item = &sb->dev_item;
e5e9a520
CM		 2412
2413 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2414 head = &root->fs_info->fs_devices->devices;
c6e30871 2415 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2416 if (!dev->bdev) {
2417 total_errors++;
2418 continue;
2419 }
2b82032c 2420 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2421 continue;
2422
2b82032c 2423 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM		 2424 btrfs_set_stack_device_type(dev_item, dev->type);
2425 btrfs_set_stack_device_id(dev_item, dev->devid);
2426 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2427 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2428 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2429 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2430 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2431 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 2432 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 2433
a061fc8d
CM		 2434 flags = btrfs_super_flags(sb);
2435 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2436
a512bbf8 2437 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM		 2438 if (ret)
2439 total_errors++;
f2984462 2440 }
a236aed1 2441 if (total_errors > max_errors) {
d397712b
CM		 2442 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2443 total_errors);
a236aed1
CM		 2444 BUG();
2445 }
f2984462 2446
a512bbf8 2447 total_errors = 0;
c6e30871 2448 list_for_each_entry(dev, head, dev_list) {
dfe25020
CM		 2449 if (!dev->bdev)
2450 continue;
2b82032c 2451 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM		 2452 continue;
2453
a512bbf8
YZ		 2454 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2455 if (ret)
2456 total_errors++;
f2984462 2457 }
e5e9a520 2458 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
a236aed1 2459 if (total_errors > max_errors) {
d397712b
CM		 2460 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2461 total_errors);
a236aed1
CM		 2462 BUG();
2463 }
f2984462
CM		 2464 return 0;
2465}
2466
a512bbf8
YZ		 2467int write_ctree_super(struct btrfs_trans_handle *trans,
2468 struct btrfs_root *root, int max_mirrors)
eb60ceac 2469{
e66f709b 2470 int ret;
5f39d397 2471
a512bbf8 2472 ret = write_all_supers(root, max_mirrors);
5f39d397 2473 return ret;
cfaa7295
CM		 2474}
2475
5eda7b5e 2476int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 2477{
4df27c4d 2478 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM		 2479 radix_tree_delete(&fs_info->fs_roots_radix,
2480 (unsigned long)root->root_key.objectid);
4df27c4d 2481 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ		 2482
2483 if (btrfs_root_refs(&root->root_item) == 0)
2484 synchronize_srcu(&fs_info->subvol_srcu);
2485
4df27c4d
YZ		 2486 free_fs_root(root);
2487 return 0;
2488}
2489
2490static void free_fs_root(struct btrfs_root *root)
2491{
2492 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3394e160
CM		 2493 if (root->anon_super.s_dev) {
2494 down_write(&root->anon_super.s_umount);
2495 kill_anon_super(&root->anon_super);
2496 }
4df27c4d
YZ		 2497 free_extent_buffer(root->node);
2498 free_extent_buffer(root->commit_root);
d397712b 2499 kfree(root->name);
2619ba1f 2500 kfree(root);
2619ba1f
CM		 2501}
2502
35b7e476 2503static int del_fs_roots(struct btrfs_fs_info *fs_info)
0f7d52f4
CM		 2504{
2505 int ret;
2506 struct btrfs_root *gang[8];
2507 int i;
2508
76dda93c
YZ		 2509 while (!list_empty(&fs_info->dead_roots)) {
2510 gang[0] = list_entry(fs_info->dead_roots.next,
2511 struct btrfs_root, root_list);
2512 list_del(&gang[0]->root_list);
2513
2514 if (gang[0]->in_radix) {
2515 btrfs_free_fs_root(fs_info, gang[0]);
2516 } else {
2517 free_extent_buffer(gang[0]->node);
2518 free_extent_buffer(gang[0]->commit_root);
2519 kfree(gang[0]);
2520 }
2521 }
2522
d397712b 2523 while (1) {
0f7d52f4
CM		 2524 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2525 (void **)gang, 0,
2526 ARRAY_SIZE(gang));
2527 if (!ret)
2528 break;
2619ba1f 2529 for (i = 0; i < ret; i++)
5eda7b5e 2530 btrfs_free_fs_root(fs_info, gang[i]);
0f7d52f4
CM		 2531 }
2532 return 0;
2533}
b4100d64 2534
c146afad 2535int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 2536{
c146afad
YZ		 2537 u64 root_objectid = 0;
2538 struct btrfs_root *gang[8];
2539 int i;
3768f368 2540 int ret;
e089f05c 2541
c146afad
YZ		 2542 while (1) {
2543 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2544 (void **)gang, root_objectid,
2545 ARRAY_SIZE(gang));
2546 if (!ret)
2547 break;
5d4f98a2
YZ		 2548
2549 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 2550 for (i = 0; i < ret; i++) {
66b4ffd1
JB		 2551 int err;
2552
c146afad 2553 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB		 2554 err = btrfs_orphan_cleanup(gang[i]);
2555 if (err)
2556 return err;
c146afad
YZ		 2557 }
2558 root_objectid++;
2559 }
2560 return 0;
2561}
a2135011 2562
c146afad
YZ		 2563int btrfs_commit_super(struct btrfs_root *root)
2564{
2565 struct btrfs_trans_handle *trans;
2566 int ret;
a74a4b97 2567
c146afad 2568 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 2569 btrfs_run_delayed_iputs(root);
a74a4b97 2570 btrfs_clean_old_snapshots(root);
c146afad 2571 mutex_unlock(&root->fs_info->cleaner_mutex);
c71bf099
YZ		 2572
2573 /* wait until ongoing cleanup work done */
2574 down_write(&root->fs_info->cleanup_work_sem);
2575 up_write(&root->fs_info->cleanup_work_sem);
2576
a22285a6 2577 trans = btrfs_join_transaction(root, 1);
3612b495
TI		 2578 if (IS_ERR(trans))
2579 return PTR_ERR(trans);
54aa1f4d 2580 ret = btrfs_commit_transaction(trans, root);
c146afad
YZ		 2581 BUG_ON(ret);
2582 /* run commit again to drop the original snapshot */
a22285a6 2583 trans = btrfs_join_transaction(root, 1);
3612b495
TI		 2584 if (IS_ERR(trans))
2585 return PTR_ERR(trans);
79154b1b
CM		 2586 btrfs_commit_transaction(trans, root);
2587 ret = btrfs_write_and_wait_transaction(NULL, root);
3768f368 2588 BUG_ON(ret);
d6bfde87 2589
a512bbf8 2590 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ		 2591 return ret;
2592}
2593
2594int close_ctree(struct btrfs_root *root)
2595{
2596 struct btrfs_fs_info *fs_info = root->fs_info;
2597 int ret;
2598
2599 fs_info->closing = 1;
2600 smp_mb();
2601
0af3d00b 2602 btrfs_put_block_group_cache(fs_info);
acce952b 2603
2604 /*
2605 * Here come 2 situations when btrfs is broken to flip readonly:
2606 *
2607 * 1. when btrfs flips readonly somewhere else before
2608 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2609 * and btrfs will skip to write sb directly to keep
2610 * ERROR state on disk.
2611 *
2612 * 2. when btrfs flips readonly just in btrfs_commit_super,
2613 * and in such case, btrfs cannnot write sb via btrfs_commit_super,
2614 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2615 * btrfs will cleanup all FS resources first and write sb then.
2616 */
c146afad 2617 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 2618 ret = btrfs_commit_super(root);
2619 if (ret)
2620 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2621 }
2622
2623 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2624 ret = btrfs_error_commit_super(root);
d397712b
CM		 2625 if (ret)
2626 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
c146afad 2627 }
0f7d52f4 2628
8929ecfa
YZ		 2629 kthread_stop(root->fs_info->transaction_kthread);
2630 kthread_stop(root->fs_info->cleaner_kthread);
2631
f25784b3
YZ		 2632 fs_info->closing = 2;
2633 smp_mb();
2634
b0c68f8b 2635 if (fs_info->delalloc_bytes) {
d397712b 2636 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
21380931 2637 (unsigned long long)fs_info->delalloc_bytes);
b0c68f8b 2638 }
31153d81 2639 if (fs_info->total_ref_cache_size) {
d397712b
CM		 2640 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2641 (unsigned long long)fs_info->total_ref_cache_size);
31153d81 2642 }
bcc63abb 2643
5d4f98a2
YZ		 2644 free_extent_buffer(fs_info->extent_root->node);
2645 free_extent_buffer(fs_info->extent_root->commit_root);
2646 free_extent_buffer(fs_info->tree_root->node);
2647 free_extent_buffer(fs_info->tree_root->commit_root);
2648 free_extent_buffer(root->fs_info->chunk_root->node);
2649 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2650 free_extent_buffer(root->fs_info->dev_root->node);
2651 free_extent_buffer(root->fs_info->dev_root->commit_root);
2652 free_extent_buffer(root->fs_info->csum_root->node);
2653 free_extent_buffer(root->fs_info->csum_root->commit_root);
d20f7043 2654
9078a3e1 2655 btrfs_free_block_groups(root->fs_info);
d10c5f31 2656
c146afad 2657 del_fs_roots(fs_info);
d10c5f31 2658
c146afad 2659 iput(fs_info->btree_inode);
9ad6b7bc 2660
61d92c32 2661 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2662 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2663 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM		 2664 btrfs_stop_workers(&fs_info->workers);
2665 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2666 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2667 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2668 btrfs_stop_workers(&fs_info->endio_write_workers);
0cb59c99 2669 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1cc127b5 2670 btrfs_stop_workers(&fs_info->submit_workers);
d6bfde87 2671
dfe25020 2672 btrfs_close_devices(fs_info->fs_devices);
0b86a832 2673 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 2674
04160088 2675 bdi_destroy(&fs_info->bdi);
76dda93c 2676 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 2677
0f7d52f4 2678 kfree(fs_info->extent_root);
0f7d52f4 2679 kfree(fs_info->tree_root);
0b86a832
CM		 2680 kfree(fs_info->chunk_root);
2681 kfree(fs_info->dev_root);
d20f7043 2682 kfree(fs_info->csum_root);
83a4d548
LZ		 2683 kfree(fs_info);
2684
eb60ceac
CM		 2685 return 0;
2686}
2687
1259ab75 2688int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
5f39d397 2689{
1259ab75 2690 int ret;
810191ff 2691 struct inode *btree_inode = buf->first_page->mapping->host;
1259ab75 2692
2ac55d41
JB		 2693 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2694 NULL);
1259ab75
CM		 2695 if (!ret)
2696 return ret;
2697
2698 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2699 parent_transid);
2700 return !ret;
5f39d397
CM		 2701}
2702
2703int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 2704{
810191ff 2705 struct inode *btree_inode = buf->first_page->mapping->host;
d1310b2e 2706 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
5f39d397
CM		 2707 buf);
2708}
6702ed49 2709
5f39d397
CM		 2710void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2711{
810191ff 2712 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
5f39d397
CM		 2713 u64 transid = btrfs_header_generation(buf);
2714 struct inode *btree_inode = root->fs_info->btree_inode;
b9473439 2715 int was_dirty;
b4ce94de 2716
b9447ef8 2717 btrfs_assert_tree_locked(buf);
ccd467d6 2718 if (transid != root->fs_info->generation) {
d397712b
CM		 2719 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2720 "found %llu running %llu\n",
db94535d 2721 (unsigned long long)buf->start,
d397712b
CM		 2722 (unsigned long long)transid,
2723 (unsigned long long)root->fs_info->generation);
ccd467d6
CM		 2724 WARN_ON(1);
2725 }
b9473439
CM		 2726 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2727 buf);
2728 if (!was_dirty) {
2729 spin_lock(&root->fs_info->delalloc_lock);
2730 root->fs_info->dirty_metadata_bytes += buf->len;
2731 spin_unlock(&root->fs_info->delalloc_lock);
2732 }
eb60ceac
CM		 2733}
2734
d3c2fdcf 2735void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
35b7e476 2736{
188de649
CM		 2737 /*
2738 * looks as though older kernels can get into trouble with
2739 * this code, they end up stuck in balance_dirty_pages forever
2740 */
d6bfde87 2741 u64 num_dirty;
771ed689 2742 unsigned long thresh = 32 * 1024 * 1024;
d6bfde87 2743
6933c02e 2744 if (current->flags & PF_MEMALLOC)
d6bfde87
CM		 2745 return;
2746
585ad2c3
CM		 2747 num_dirty = root->fs_info->dirty_metadata_bytes;
2748
d6bfde87
CM		 2749 if (num_dirty > thresh) {
2750 balance_dirty_pages_ratelimited_nr(
d7fc640e 2751 root->fs_info->btree_inode->i_mapping, 1);
d6bfde87 2752 }
188de649 2753 return;
35b7e476 2754}
6b80053d 2755
ca7a79ad 2756int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 2757{
810191ff 2758 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
ce9adaa5 2759 int ret;
ca7a79ad 2760 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
d397712b 2761 if (ret == 0)
b4ce94de 2762 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
ce9adaa5 2763 return ret;
6b80053d 2764}
0da5468f 2765
4bef0848
CM		 2766int btree_lock_page_hook(struct page *page)
2767{
2768 struct inode *inode = page->mapping->host;
b9473439 2769 struct btrfs_root *root = BTRFS_I(inode)->root;
4bef0848
CM		 2770 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2771 struct extent_buffer *eb;
2772 unsigned long len;
2773 u64 bytenr = page_offset(page);
2774
2775 if (page->private == EXTENT_PAGE_PRIVATE)
2776 goto out;
2777
2778 len = page->private >> 2;
2779 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2780 if (!eb)
2781 goto out;
2782
2783 btrfs_tree_lock(eb);
4bef0848 2784 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
b9473439
CM		 2785
2786 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2787 spin_lock(&root->fs_info->delalloc_lock);
2788 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2789 root->fs_info->dirty_metadata_bytes -= eb->len;
2790 else
2791 WARN_ON(1);
2792 spin_unlock(&root->fs_info->delalloc_lock);
2793 }
2794
4bef0848
CM		 2795 btrfs_tree_unlock(eb);
2796 free_extent_buffer(eb);
2797out:
2798 lock_page(page);
2799 return 0;
2800}
2801
acce952b 2802static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2803 int read_only)
2804{
2805 if (read_only)
2806 return;
2807
2808 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2809 printk(KERN_WARNING "warning: mount fs with errors, "
2810 "running btrfsck is recommended\n");
2811}
2812
2813int btrfs_error_commit_super(struct btrfs_root *root)
2814{
2815 int ret;
2816
2817 mutex_lock(&root->fs_info->cleaner_mutex);
2818 btrfs_run_delayed_iputs(root);
2819 mutex_unlock(&root->fs_info->cleaner_mutex);
2820
2821 down_write(&root->fs_info->cleanup_work_sem);
2822 up_write(&root->fs_info->cleanup_work_sem);
2823
2824 /* cleanup FS via transaction */
2825 btrfs_cleanup_transaction(root);
2826
2827 ret = write_ctree_super(NULL, root, 0);
2828
2829 return ret;
2830}
2831
2832static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2833{
2834 struct btrfs_inode *btrfs_inode;
2835 struct list_head splice;
2836
2837 INIT_LIST_HEAD(&splice);
2838
2839 mutex_lock(&root->fs_info->ordered_operations_mutex);
2840 spin_lock(&root->fs_info->ordered_extent_lock);
2841
2842 list_splice_init(&root->fs_info->ordered_operations, &splice);
2843 while (!list_empty(&splice)) {
2844 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2845 ordered_operations);
2846
2847 list_del_init(&btrfs_inode->ordered_operations);
2848
2849 btrfs_invalidate_inodes(btrfs_inode->root);
2850 }
2851
2852 spin_unlock(&root->fs_info->ordered_extent_lock);
2853 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2854
2855 return 0;
2856}
2857
2858static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2859{
2860 struct list_head splice;
2861 struct btrfs_ordered_extent *ordered;
2862 struct inode *inode;
2863
2864 INIT_LIST_HEAD(&splice);
2865
2866 spin_lock(&root->fs_info->ordered_extent_lock);
2867
2868 list_splice_init(&root->fs_info->ordered_extents, &splice);
2869 while (!list_empty(&splice)) {
2870 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2871 root_extent_list);
2872
2873 list_del_init(&ordered->root_extent_list);
2874 atomic_inc(&ordered->refs);
2875
2876 /* the inode may be getting freed (in sys_unlink path). */
2877 inode = igrab(ordered->inode);
2878
2879 spin_unlock(&root->fs_info->ordered_extent_lock);
2880 if (inode)
2881 iput(inode);
2882
2883 atomic_set(&ordered->refs, 1);
2884 btrfs_put_ordered_extent(ordered);
2885
2886 spin_lock(&root->fs_info->ordered_extent_lock);
2887 }
2888
2889 spin_unlock(&root->fs_info->ordered_extent_lock);
2890
2891 return 0;
2892}
2893
2894static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2895 struct btrfs_root *root)
2896{
2897 struct rb_node *node;
2898 struct btrfs_delayed_ref_root *delayed_refs;
2899 struct btrfs_delayed_ref_node *ref;
2900 int ret = 0;
2901
2902 delayed_refs = &trans->delayed_refs;
2903
2904 spin_lock(&delayed_refs->lock);
2905 if (delayed_refs->num_entries == 0) {
2906 printk(KERN_INFO "delayed_refs has NO entry\n");
2907 return ret;
2908 }
2909
2910 node = rb_first(&delayed_refs->root);
2911 while (node) {
2912 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2913 node = rb_next(node);
2914
2915 ref->in_tree = 0;
2916 rb_erase(&ref->rb_node, &delayed_refs->root);
2917 delayed_refs->num_entries--;
2918
2919 atomic_set(&ref->refs, 1);
2920 if (btrfs_delayed_ref_is_head(ref)) {
2921 struct btrfs_delayed_ref_head *head;
2922
2923 head = btrfs_delayed_node_to_head(ref);
2924 mutex_lock(&head->mutex);
2925 kfree(head->extent_op);
2926 delayed_refs->num_heads--;
2927 if (list_empty(&head->cluster))
2928 delayed_refs->num_heads_ready--;
2929 list_del_init(&head->cluster);
2930 mutex_unlock(&head->mutex);
2931 }
2932
2933 spin_unlock(&delayed_refs->lock);
2934 btrfs_put_delayed_ref(ref);
2935
2936 cond_resched();
2937 spin_lock(&delayed_refs->lock);
2938 }
2939
2940 spin_unlock(&delayed_refs->lock);
2941
2942 return ret;
2943}
2944
2945static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2946{
2947 struct btrfs_pending_snapshot *snapshot;
2948 struct list_head splice;
2949
2950 INIT_LIST_HEAD(&splice);
2951
2952 list_splice_init(&t->pending_snapshots, &splice);
2953
2954 while (!list_empty(&splice)) {
2955 snapshot = list_entry(splice.next,
2956 struct btrfs_pending_snapshot,
2957 list);
2958
2959 list_del_init(&snapshot->list);
2960
2961 kfree(snapshot);
2962 }
2963
2964 return 0;
2965}
2966
2967static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2968{
2969 struct btrfs_inode *btrfs_inode;
2970 struct list_head splice;
2971
2972 INIT_LIST_HEAD(&splice);
2973
2974 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2975
2976 spin_lock(&root->fs_info->delalloc_lock);
2977
2978 while (!list_empty(&splice)) {
2979 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2980 delalloc_inodes);
2981
2982 list_del_init(&btrfs_inode->delalloc_inodes);
2983
2984 btrfs_invalidate_inodes(btrfs_inode->root);
2985 }
2986
2987 spin_unlock(&root->fs_info->delalloc_lock);
2988
2989 return 0;
2990}
2991
2992static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2993 struct extent_io_tree *dirty_pages,
2994 int mark)
2995{
2996 int ret;
2997 struct page *page;
2998 struct inode *btree_inode = root->fs_info->btree_inode;
2999 struct extent_buffer *eb;
3000 u64 start = 0;
3001 u64 end;
3002 u64 offset;
3003 unsigned long index;
3004
3005 while (1) {
3006 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3007 mark);
3008 if (ret)
3009 break;
3010
3011 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3012 while (start <= end) {
3013 index = start >> PAGE_CACHE_SHIFT;
3014 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3015 page = find_get_page(btree_inode->i_mapping, index);
3016 if (!page)
3017 continue;
3018 offset = page_offset(page);
3019
3020 spin_lock(&dirty_pages->buffer_lock);
3021 eb = radix_tree_lookup(
3022 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3023 offset >> PAGE_CACHE_SHIFT);
3024 spin_unlock(&dirty_pages->buffer_lock);
3025 if (eb) {
3026 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3027 &eb->bflags);
3028 atomic_set(&eb->refs, 1);
3029 }
3030 if (PageWriteback(page))
3031 end_page_writeback(page);
3032
3033 lock_page(page);
3034 if (PageDirty(page)) {
3035 clear_page_dirty_for_io(page);
3036 spin_lock_irq(&page->mapping->tree_lock);
3037 radix_tree_tag_clear(&page->mapping->page_tree,
3038 page_index(page),
3039 PAGECACHE_TAG_DIRTY);
3040 spin_unlock_irq(&page->mapping->tree_lock);
3041 }
3042
3043 page->mapping->a_ops->invalidatepage(page, 0);
3044 unlock_page(page);
3045 }
3046 }
3047
3048 return ret;
3049}
3050
3051static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3052 struct extent_io_tree *pinned_extents)
3053{
3054 struct extent_io_tree *unpin;
3055 u64 start;
3056 u64 end;
3057 int ret;
3058
3059 unpin = pinned_extents;
3060 while (1) {
3061 ret = find_first_extent_bit(unpin, 0, &start, &end,
3062 EXTENT_DIRTY);
3063 if (ret)
3064 break;
3065
3066 /* opt_discard */
5378e607
LD		 3067 if (btrfs_test_opt(root, DISCARD))
3068 ret = btrfs_error_discard_extent(root, start,
3069 end + 1 - start,
3070 NULL);
acce952b 3071
3072 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3073 btrfs_error_unpin_extent_range(root, start, end);
3074 cond_resched();
3075 }
3076
3077 return 0;
3078}
3079
3080static int btrfs_cleanup_transaction(struct btrfs_root *root)
3081{
3082 struct btrfs_transaction *t;
3083 LIST_HEAD(list);
3084
3085 WARN_ON(1);
3086
3087 mutex_lock(&root->fs_info->trans_mutex);
3088 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3089
3090 list_splice_init(&root->fs_info->trans_list, &list);
3091 while (!list_empty(&list)) {
3092 t = list_entry(list.next, struct btrfs_transaction, list);
3093 if (!t)
3094 break;
3095
3096 btrfs_destroy_ordered_operations(root);
3097
3098 btrfs_destroy_ordered_extents(root);
3099
3100 btrfs_destroy_delayed_refs(t, root);
3101
3102 btrfs_block_rsv_release(root,
3103 &root->fs_info->trans_block_rsv,
3104 t->dirty_pages.dirty_bytes);
3105
3106 /* FIXME: cleanup wait for commit */
3107 t->in_commit = 1;
3108 t->blocked = 1;
3109 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3110 wake_up(&root->fs_info->transaction_blocked_wait);
3111
3112 t->blocked = 0;
3113 if (waitqueue_active(&root->fs_info->transaction_wait))
3114 wake_up(&root->fs_info->transaction_wait);
3115 mutex_unlock(&root->fs_info->trans_mutex);
3116
3117 mutex_lock(&root->fs_info->trans_mutex);
3118 t->commit_done = 1;
3119 if (waitqueue_active(&t->commit_wait))
3120 wake_up(&t->commit_wait);
3121 mutex_unlock(&root->fs_info->trans_mutex);
3122
3123 mutex_lock(&root->fs_info->trans_mutex);
3124
3125 btrfs_destroy_pending_snapshots(t);
3126
3127 btrfs_destroy_delalloc_inodes(root);
3128
3129 spin_lock(&root->fs_info->new_trans_lock);
3130 root->fs_info->running_transaction = NULL;
3131 spin_unlock(&root->fs_info->new_trans_lock);
3132
3133 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3134 EXTENT_DIRTY);
3135
3136 btrfs_destroy_pinned_extent(root,
3137 root->fs_info->pinned_extents);
3138
13c5a93e 3139 atomic_set(&t->use_count, 0);
acce952b 3140 list_del_init(&t->list);
3141 memset(t, 0, sizeof(*t));
3142 kmem_cache_free(btrfs_transaction_cachep, t);
3143 }
3144
3145 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3146 mutex_unlock(&root->fs_info->trans_mutex);
3147
3148 return 0;
3149}
3150
d1310b2e 3151static struct extent_io_ops btree_extent_io_ops = {
4bef0848 3152 .write_cache_pages_lock_hook = btree_lock_page_hook,
ce9adaa5 3153 .readpage_end_io_hook = btree_readpage_end_io_hook,
0b86a832 3154 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM		 3155 /* note we're sharing with inode.c for the merge bio hook */
3156 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 3157};
Linux 6.x block layer and io_uring tree(s)