Merge branch 'cleanups_and_fixes' into inode_numbers
[linux-2.6-block.git] / fs / btrfs / disk-io.c
CommitLineData
6cbd5570
CM
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
e20d96d6 19#include <linux/fs.h>
d98237b3 20#include <linux/blkdev.h>
87cbda5c 21#include <linux/scatterlist.h>
22b0ebda 22#include <linux/swap.h>
0f7d52f4 23#include <linux/radix-tree.h>
35b7e476 24#include <linux/writeback.h>
d397712b 25#include <linux/buffer_head.h>
ce9adaa5 26#include <linux/workqueue.h>
a74a4b97 27#include <linux/kthread.h>
4b4e25f2 28#include <linux/freezer.h>
163e783e 29#include <linux/crc32c.h>
5a0e3ad6 30#include <linux/slab.h>
784b4e29 31#include <linux/migrate.h>
7a36ddec 32#include <linux/ratelimit.h>
7e75bf3f 33#include <asm/unaligned.h>
4b4e25f2 34#include "compat.h"
eb60ceac
CM
35#include "ctree.h"
36#include "disk-io.h"
e089f05c 37#include "transaction.h"
0f7d52f4 38#include "btrfs_inode.h"
0b86a832 39#include "volumes.h"
db94535d 40#include "print-tree.h"
8b712842 41#include "async-thread.h"
925baedd 42#include "locking.h"
e02119d5 43#include "tree-log.h"
fa9c0d79 44#include "free-space-cache.h"
581bb050 45#include "inode-map.h"
eb60ceac 46
d1310b2e 47static struct extent_io_ops btree_extent_io_ops;
8b712842 48static void end_workqueue_fn(struct btrfs_work *work);
4df27c4d 49static void free_fs_root(struct btrfs_root *root);
acce952b 50static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
51 int read_only);
52static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
53static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
54static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
55 struct btrfs_root *root);
56static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
57static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58static int btrfs_destroy_marked_extents(struct btrfs_root *root,
59 struct extent_io_tree *dirty_pages,
60 int mark);
61static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
62 struct extent_io_tree *pinned_extents);
63static int btrfs_cleanup_transaction(struct btrfs_root *root);
ce9adaa5 64
d352ac68
CM
65/*
66 * end_io_wq structs are used to do processing in task context when an IO is
67 * complete. This is used during reads to verify checksums, and it is used
68 * by writes to insert metadata for new file extents after IO is complete.
69 */
ce9adaa5
CM
70struct end_io_wq {
71 struct bio *bio;
72 bio_end_io_t *end_io;
73 void *private;
74 struct btrfs_fs_info *info;
75 int error;
22c59948 76 int metadata;
ce9adaa5 77 struct list_head list;
8b712842 78 struct btrfs_work work;
ce9adaa5 79};
0da5468f 80
d352ac68
CM
81/*
82 * async submit bios are used to offload expensive checksumming
83 * onto the worker threads. They checksum file and metadata bios
84 * just before they are sent down the IO stack.
85 */
44b8bd7e
CM
86struct async_submit_bio {
87 struct inode *inode;
88 struct bio *bio;
89 struct list_head list;
4a69a410
CM
90 extent_submit_bio_hook_t *submit_bio_start;
91 extent_submit_bio_hook_t *submit_bio_done;
44b8bd7e
CM
92 int rw;
93 int mirror_num;
c8b97818 94 unsigned long bio_flags;
eaf25d93
CM
95 /*
96 * bio_offset is optional, can be used if the pages in the bio
97 * can't tell us where in the file the bio should go
98 */
99 u64 bio_offset;
8b712842 100 struct btrfs_work work;
44b8bd7e
CM
101};
102
4008c04a
CM
103/* These are used to set the lockdep class on the extent buffer locks.
104 * The class is set by the readpage_end_io_hook after the buffer has
105 * passed csum validation but before the pages are unlocked.
106 *
107 * The lockdep class is also set by btrfs_init_new_buffer on freshly
108 * allocated blocks.
109 *
110 * The class is based on the level in the tree block, which allows lockdep
111 * to know that lower nodes nest inside the locks of higher nodes.
112 *
113 * We also add a check to make sure the highest level of the tree is
114 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
115 * code needs update as well.
116 */
117#ifdef CONFIG_DEBUG_LOCK_ALLOC
118# if BTRFS_MAX_LEVEL != 8
119# error
120# endif
121static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
122static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
123 /* leaf */
124 "btrfs-extent-00",
125 "btrfs-extent-01",
126 "btrfs-extent-02",
127 "btrfs-extent-03",
128 "btrfs-extent-04",
129 "btrfs-extent-05",
130 "btrfs-extent-06",
131 "btrfs-extent-07",
132 /* highest possible level */
133 "btrfs-extent-08",
134};
135#endif
136
d352ac68
CM
137/*
138 * extents on the btree inode are pretty simple, there's one extent
139 * that covers the entire device
140 */
b2950863 141static struct extent_map *btree_get_extent(struct inode *inode,
306e16ce 142 struct page *page, size_t pg_offset, u64 start, u64 len,
b2950863 143 int create)
7eccb903 144{
5f39d397
CM
145 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
146 struct extent_map *em;
147 int ret;
148
890871be 149 read_lock(&em_tree->lock);
d1310b2e 150 em = lookup_extent_mapping(em_tree, start, len);
a061fc8d
CM
151 if (em) {
152 em->bdev =
153 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
890871be 154 read_unlock(&em_tree->lock);
5f39d397 155 goto out;
a061fc8d 156 }
890871be 157 read_unlock(&em_tree->lock);
7b13b7b1 158
172ddd60 159 em = alloc_extent_map();
5f39d397
CM
160 if (!em) {
161 em = ERR_PTR(-ENOMEM);
162 goto out;
163 }
164 em->start = 0;
0afbaf8c 165 em->len = (u64)-1;
c8b97818 166 em->block_len = (u64)-1;
5f39d397 167 em->block_start = 0;
a061fc8d 168 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
d1310b2e 169
890871be 170 write_lock(&em_tree->lock);
5f39d397
CM
171 ret = add_extent_mapping(em_tree, em);
172 if (ret == -EEXIST) {
0afbaf8c
CM
173 u64 failed_start = em->start;
174 u64 failed_len = em->len;
175
5f39d397 176 free_extent_map(em);
7b13b7b1 177 em = lookup_extent_mapping(em_tree, start, len);
0afbaf8c 178 if (em) {
7b13b7b1 179 ret = 0;
0afbaf8c
CM
180 } else {
181 em = lookup_extent_mapping(em_tree, failed_start,
182 failed_len);
7b13b7b1 183 ret = -EIO;
0afbaf8c 184 }
5f39d397 185 } else if (ret) {
7b13b7b1
CM
186 free_extent_map(em);
187 em = NULL;
5f39d397 188 }
890871be 189 write_unlock(&em_tree->lock);
7b13b7b1
CM
190
191 if (ret)
192 em = ERR_PTR(ret);
5f39d397
CM
193out:
194 return em;
7eccb903
CM
195}
196
19c00ddc
CM
197u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
198{
163e783e 199 return crc32c(seed, data, len);
19c00ddc
CM
200}
201
202void btrfs_csum_final(u32 crc, char *result)
203{
7e75bf3f 204 put_unaligned_le32(~crc, result);
19c00ddc
CM
205}
206
d352ac68
CM
207/*
208 * compute the csum for a btree block, and either verify it or write it
209 * into the csum field of the block.
210 */
19c00ddc
CM
211static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
212 int verify)
213{
607d432d
JB
214 u16 csum_size =
215 btrfs_super_csum_size(&root->fs_info->super_copy);
216 char *result = NULL;
19c00ddc
CM
217 unsigned long len;
218 unsigned long cur_len;
219 unsigned long offset = BTRFS_CSUM_SIZE;
220 char *map_token = NULL;
221 char *kaddr;
222 unsigned long map_start;
223 unsigned long map_len;
224 int err;
225 u32 crc = ~(u32)0;
607d432d 226 unsigned long inline_result;
19c00ddc
CM
227
228 len = buf->len - offset;
d397712b 229 while (len > 0) {
19c00ddc
CM
230 err = map_private_extent_buffer(buf, offset, 32,
231 &map_token, &kaddr,
232 &map_start, &map_len, KM_USER0);
d397712b 233 if (err)
19c00ddc 234 return 1;
19c00ddc
CM
235 cur_len = min(len, map_len - (offset - map_start));
236 crc = btrfs_csum_data(root, kaddr + offset - map_start,
237 crc, cur_len);
238 len -= cur_len;
239 offset += cur_len;
240 unmap_extent_buffer(buf, map_token, KM_USER0);
241 }
607d432d
JB
242 if (csum_size > sizeof(inline_result)) {
243 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
244 if (!result)
245 return 1;
246 } else {
247 result = (char *)&inline_result;
248 }
249
19c00ddc
CM
250 btrfs_csum_final(crc, result);
251
252 if (verify) {
607d432d 253 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
e4204ded
CM
254 u32 val;
255 u32 found = 0;
607d432d 256 memcpy(&found, result, csum_size);
e4204ded 257
607d432d 258 read_extent_buffer(buf, &val, 0, csum_size);
7a36ddec 259 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
193f284d
CM
260 "failed on %llu wanted %X found %X "
261 "level %d\n",
262 root->fs_info->sb->s_id,
263 (unsigned long long)buf->start, val, found,
264 btrfs_header_level(buf));
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 }
7a36ddec 299 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
193f284d
CM
300 "found %llu\n",
301 (unsigned long long)eb->start,
302 (unsigned long long)parent_transid,
303 (unsigned long long)btrfs_header_generation(eb));
1259ab75 304 ret = 1;
2ac55d41 305 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
33958dc6 306out:
2ac55d41
JB
307 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
308 &cached_state, GFP_NOFS);
1259ab75 309 return ret;
1259ab75
CM
310}
311
d352ac68
CM
312/*
313 * helper to read a given tree block, doing retries as required when
314 * the checksums don't match and we have alternate mirrors to try.
315 */
f188591e
CM
316static int btree_read_extent_buffer_pages(struct btrfs_root *root,
317 struct extent_buffer *eb,
ca7a79ad 318 u64 start, u64 parent_transid)
f188591e
CM
319{
320 struct extent_io_tree *io_tree;
321 int ret;
322 int num_copies = 0;
323 int mirror_num = 0;
324
a826d6dc 325 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
f188591e
CM
326 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
327 while (1) {
328 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
329 btree_get_extent, mirror_num);
1259ab75
CM
330 if (!ret &&
331 !verify_parent_transid(io_tree, eb, parent_transid))
f188591e 332 return ret;
d397712b 333
a826d6dc
JB
334 /*
335 * This buffer's crc is fine, but its contents are corrupted, so
336 * there is no reason to read the other copies, they won't be
337 * any less wrong.
338 */
339 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
340 return ret;
341
f188591e
CM
342 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
343 eb->start, eb->len);
4235298e 344 if (num_copies == 1)
f188591e 345 return ret;
4235298e 346
f188591e 347 mirror_num++;
4235298e 348 if (mirror_num > num_copies)
f188591e 349 return ret;
f188591e 350 }
f188591e
CM
351 return -EIO;
352}
19c00ddc 353
d352ac68 354/*
d397712b
CM
355 * checksum a dirty tree block before IO. This has extra checks to make sure
356 * we only fill in the checksum field in the first page of a multi-page block
d352ac68 357 */
d397712b 358
b2950863 359static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
19c00ddc 360{
d1310b2e 361 struct extent_io_tree *tree;
35ebb934 362 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
19c00ddc 363 u64 found_start;
19c00ddc
CM
364 unsigned long len;
365 struct extent_buffer *eb;
f188591e
CM
366 int ret;
367
d1310b2e 368 tree = &BTRFS_I(page->mapping->host)->io_tree;
19c00ddc 369
eb14ab8e
CM
370 if (page->private == EXTENT_PAGE_PRIVATE) {
371 WARN_ON(1);
19c00ddc 372 goto out;
eb14ab8e
CM
373 }
374 if (!page->private) {
375 WARN_ON(1);
19c00ddc 376 goto out;
eb14ab8e 377 }
19c00ddc 378 len = page->private >> 2;
d397712b
CM
379 WARN_ON(len == 0);
380
ba144192 381 eb = alloc_extent_buffer(tree, start, len, page);
91ca338d
TI
382 if (eb == NULL) {
383 WARN_ON(1);
384 goto out;
385 }
ca7a79ad
CM
386 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
387 btrfs_header_generation(eb));
f188591e 388 BUG_ON(ret);
784b4e29
CM
389 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
390
19c00ddc
CM
391 found_start = btrfs_header_bytenr(eb);
392 if (found_start != start) {
55c69072
CM
393 WARN_ON(1);
394 goto err;
395 }
396 if (eb->first_page != page) {
55c69072
CM
397 WARN_ON(1);
398 goto err;
399 }
400 if (!PageUptodate(page)) {
55c69072
CM
401 WARN_ON(1);
402 goto err;
19c00ddc 403 }
19c00ddc 404 csum_tree_block(root, eb, 0);
55c69072 405err:
19c00ddc
CM
406 free_extent_buffer(eb);
407out:
408 return 0;
409}
410
2b82032c
YZ
411static int check_tree_block_fsid(struct btrfs_root *root,
412 struct extent_buffer *eb)
413{
414 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
415 u8 fsid[BTRFS_UUID_SIZE];
416 int ret = 1;
417
418 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
419 BTRFS_FSID_SIZE);
420 while (fs_devices) {
421 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
422 ret = 0;
423 break;
424 }
425 fs_devices = fs_devices->seed;
426 }
427 return ret;
428}
429
a826d6dc
JB
430#define CORRUPT(reason, eb, root, slot) \
431 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
432 "root=%llu, slot=%d\n", reason, \
433 (unsigned long long)btrfs_header_bytenr(eb), \
434 (unsigned long long)root->objectid, slot)
435
436static noinline int check_leaf(struct btrfs_root *root,
437 struct extent_buffer *leaf)
438{
439 struct btrfs_key key;
440 struct btrfs_key leaf_key;
441 u32 nritems = btrfs_header_nritems(leaf);
442 int slot;
443
444 if (nritems == 0)
445 return 0;
446
447 /* Check the 0 item */
448 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
449 BTRFS_LEAF_DATA_SIZE(root)) {
450 CORRUPT("invalid item offset size pair", leaf, root, 0);
451 return -EIO;
452 }
453
454 /*
455 * Check to make sure each items keys are in the correct order and their
456 * offsets make sense. We only have to loop through nritems-1 because
457 * we check the current slot against the next slot, which verifies the
458 * next slot's offset+size makes sense and that the current's slot
459 * offset is correct.
460 */
461 for (slot = 0; slot < nritems - 1; slot++) {
462 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
463 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
464
465 /* Make sure the keys are in the right order */
466 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
467 CORRUPT("bad key order", leaf, root, slot);
468 return -EIO;
469 }
470
471 /*
472 * Make sure the offset and ends are right, remember that the
473 * item data starts at the end of the leaf and grows towards the
474 * front.
475 */
476 if (btrfs_item_offset_nr(leaf, slot) !=
477 btrfs_item_end_nr(leaf, slot + 1)) {
478 CORRUPT("slot offset bad", leaf, root, slot);
479 return -EIO;
480 }
481
482 /*
483 * Check to make sure that we don't point outside of the leaf,
484 * just incase all the items are consistent to eachother, but
485 * all point outside of the leaf.
486 */
487 if (btrfs_item_end_nr(leaf, slot) >
488 BTRFS_LEAF_DATA_SIZE(root)) {
489 CORRUPT("slot end outside of leaf", leaf, root, slot);
490 return -EIO;
491 }
492 }
493
494 return 0;
495}
496
4008c04a
CM
497#ifdef CONFIG_DEBUG_LOCK_ALLOC
498void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
499{
500 lockdep_set_class_and_name(&eb->lock,
501 &btrfs_eb_class[level],
502 btrfs_eb_name[level]);
503}
504#endif
505
b2950863 506static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
ce9adaa5
CM
507 struct extent_state *state)
508{
509 struct extent_io_tree *tree;
510 u64 found_start;
511 int found_level;
512 unsigned long len;
513 struct extent_buffer *eb;
514 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
f188591e 515 int ret = 0;
ce9adaa5
CM
516
517 tree = &BTRFS_I(page->mapping->host)->io_tree;
518 if (page->private == EXTENT_PAGE_PRIVATE)
519 goto out;
520 if (!page->private)
521 goto out;
d397712b 522
ce9adaa5 523 len = page->private >> 2;
d397712b
CM
524 WARN_ON(len == 0);
525
ba144192 526 eb = alloc_extent_buffer(tree, start, len, page);
91ca338d
TI
527 if (eb == NULL) {
528 ret = -EIO;
529 goto out;
530 }
f188591e 531
ce9adaa5 532 found_start = btrfs_header_bytenr(eb);
23a07867 533 if (found_start != start) {
7a36ddec 534 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
193f284d
CM
535 "%llu %llu\n",
536 (unsigned long long)found_start,
537 (unsigned long long)eb->start);
f188591e 538 ret = -EIO;
ce9adaa5
CM
539 goto err;
540 }
541 if (eb->first_page != page) {
d397712b
CM
542 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
543 eb->first_page->index, page->index);
ce9adaa5 544 WARN_ON(1);
f188591e 545 ret = -EIO;
ce9adaa5
CM
546 goto err;
547 }
2b82032c 548 if (check_tree_block_fsid(root, eb)) {
7a36ddec 549 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
193f284d 550 (unsigned long long)eb->start);
1259ab75
CM
551 ret = -EIO;
552 goto err;
553 }
ce9adaa5
CM
554 found_level = btrfs_header_level(eb);
555
4008c04a
CM
556 btrfs_set_buffer_lockdep_class(eb, found_level);
557
ce9adaa5 558 ret = csum_tree_block(root, eb, 1);
a826d6dc 559 if (ret) {
f188591e 560 ret = -EIO;
a826d6dc
JB
561 goto err;
562 }
563
564 /*
565 * If this is a leaf block and it is corrupt, set the corrupt bit so
566 * that we don't try and read the other copies of this block, just
567 * return -EIO.
568 */
569 if (found_level == 0 && check_leaf(root, eb)) {
570 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
571 ret = -EIO;
572 }
ce9adaa5
CM
573
574 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
575 end = eb->start + end - 1;
ce9adaa5
CM
576err:
577 free_extent_buffer(eb);
578out:
f188591e 579 return ret;
ce9adaa5
CM
580}
581
ce9adaa5 582static void end_workqueue_bio(struct bio *bio, int err)
ce9adaa5
CM
583{
584 struct end_io_wq *end_io_wq = bio->bi_private;
585 struct btrfs_fs_info *fs_info;
ce9adaa5 586
ce9adaa5 587 fs_info = end_io_wq->info;
ce9adaa5 588 end_io_wq->error = err;
8b712842
CM
589 end_io_wq->work.func = end_workqueue_fn;
590 end_io_wq->work.flags = 0;
d20f7043 591
7b6d91da 592 if (bio->bi_rw & REQ_WRITE) {
0cb59c99 593 if (end_io_wq->metadata == 1)
cad321ad
CM
594 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
595 &end_io_wq->work);
0cb59c99
JB
596 else if (end_io_wq->metadata == 2)
597 btrfs_queue_worker(&fs_info->endio_freespace_worker,
598 &end_io_wq->work);
cad321ad
CM
599 else
600 btrfs_queue_worker(&fs_info->endio_write_workers,
601 &end_io_wq->work);
d20f7043
CM
602 } else {
603 if (end_io_wq->metadata)
604 btrfs_queue_worker(&fs_info->endio_meta_workers,
605 &end_io_wq->work);
606 else
607 btrfs_queue_worker(&fs_info->endio_workers,
608 &end_io_wq->work);
609 }
ce9adaa5
CM
610}
611
0cb59c99
JB
612/*
613 * For the metadata arg you want
614 *
615 * 0 - if data
616 * 1 - if normal metadta
617 * 2 - if writing to the free space cache area
618 */
22c59948
CM
619int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
620 int metadata)
0b86a832 621{
ce9adaa5 622 struct end_io_wq *end_io_wq;
ce9adaa5
CM
623 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
624 if (!end_io_wq)
625 return -ENOMEM;
626
627 end_io_wq->private = bio->bi_private;
628 end_io_wq->end_io = bio->bi_end_io;
22c59948 629 end_io_wq->info = info;
ce9adaa5
CM
630 end_io_wq->error = 0;
631 end_io_wq->bio = bio;
22c59948 632 end_io_wq->metadata = metadata;
ce9adaa5
CM
633
634 bio->bi_private = end_io_wq;
635 bio->bi_end_io = end_workqueue_bio;
22c59948
CM
636 return 0;
637}
638
b64a2851 639unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
0986fe9e 640{
4854ddd0
CM
641 unsigned long limit = min_t(unsigned long,
642 info->workers.max_workers,
643 info->fs_devices->open_devices);
644 return 256 * limit;
645}
0986fe9e 646
4a69a410
CM
647static void run_one_async_start(struct btrfs_work *work)
648{
4a69a410
CM
649 struct async_submit_bio *async;
650
651 async = container_of(work, struct async_submit_bio, work);
4a69a410 652 async->submit_bio_start(async->inode, async->rw, async->bio,
eaf25d93
CM
653 async->mirror_num, async->bio_flags,
654 async->bio_offset);
4a69a410
CM
655}
656
657static void run_one_async_done(struct btrfs_work *work)
8b712842
CM
658{
659 struct btrfs_fs_info *fs_info;
660 struct async_submit_bio *async;
4854ddd0 661 int limit;
8b712842
CM
662
663 async = container_of(work, struct async_submit_bio, work);
664 fs_info = BTRFS_I(async->inode)->root->fs_info;
4854ddd0 665
b64a2851 666 limit = btrfs_async_submit_limit(fs_info);
4854ddd0
CM
667 limit = limit * 2 / 3;
668
8b712842 669 atomic_dec(&fs_info->nr_async_submits);
0986fe9e 670
b64a2851
CM
671 if (atomic_read(&fs_info->nr_async_submits) < limit &&
672 waitqueue_active(&fs_info->async_submit_wait))
4854ddd0
CM
673 wake_up(&fs_info->async_submit_wait);
674
4a69a410 675 async->submit_bio_done(async->inode, async->rw, async->bio,
eaf25d93
CM
676 async->mirror_num, async->bio_flags,
677 async->bio_offset);
4a69a410
CM
678}
679
680static void run_one_async_free(struct btrfs_work *work)
681{
682 struct async_submit_bio *async;
683
684 async = container_of(work, struct async_submit_bio, work);
8b712842
CM
685 kfree(async);
686}
687
44b8bd7e
CM
688int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
689 int rw, struct bio *bio, int mirror_num,
c8b97818 690 unsigned long bio_flags,
eaf25d93 691 u64 bio_offset,
4a69a410
CM
692 extent_submit_bio_hook_t *submit_bio_start,
693 extent_submit_bio_hook_t *submit_bio_done)
44b8bd7e
CM
694{
695 struct async_submit_bio *async;
696
697 async = kmalloc(sizeof(*async), GFP_NOFS);
698 if (!async)
699 return -ENOMEM;
700
701 async->inode = inode;
702 async->rw = rw;
703 async->bio = bio;
704 async->mirror_num = mirror_num;
4a69a410
CM
705 async->submit_bio_start = submit_bio_start;
706 async->submit_bio_done = submit_bio_done;
707
708 async->work.func = run_one_async_start;
709 async->work.ordered_func = run_one_async_done;
710 async->work.ordered_free = run_one_async_free;
711
8b712842 712 async->work.flags = 0;
c8b97818 713 async->bio_flags = bio_flags;
eaf25d93 714 async->bio_offset = bio_offset;
8c8bee1d 715
cb03c743 716 atomic_inc(&fs_info->nr_async_submits);
d313d7a3 717
7b6d91da 718 if (rw & REQ_SYNC)
d313d7a3
CM
719 btrfs_set_work_high_prio(&async->work);
720
8b712842 721 btrfs_queue_worker(&fs_info->workers, &async->work);
9473f16c 722
d397712b 723 while (atomic_read(&fs_info->async_submit_draining) &&
771ed689
CM
724 atomic_read(&fs_info->nr_async_submits)) {
725 wait_event(fs_info->async_submit_wait,
726 (atomic_read(&fs_info->nr_async_submits) == 0));
727 }
728
44b8bd7e
CM
729 return 0;
730}
731
ce3ed71a
CM
732static int btree_csum_one_bio(struct bio *bio)
733{
734 struct bio_vec *bvec = bio->bi_io_vec;
735 int bio_index = 0;
736 struct btrfs_root *root;
737
738 WARN_ON(bio->bi_vcnt <= 0);
d397712b 739 while (bio_index < bio->bi_vcnt) {
ce3ed71a
CM
740 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
741 csum_dirty_buffer(root, bvec->bv_page);
742 bio_index++;
743 bvec++;
744 }
745 return 0;
746}
747
4a69a410
CM
748static int __btree_submit_bio_start(struct inode *inode, int rw,
749 struct bio *bio, int mirror_num,
eaf25d93
CM
750 unsigned long bio_flags,
751 u64 bio_offset)
22c59948 752{
8b712842
CM
753 /*
754 * when we're called for a write, we're already in the async
5443be45 755 * submission context. Just jump into btrfs_map_bio
8b712842 756 */
4a69a410
CM
757 btree_csum_one_bio(bio);
758 return 0;
759}
22c59948 760
4a69a410 761static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM
762 int mirror_num, unsigned long bio_flags,
763 u64 bio_offset)
4a69a410 764{
8b712842 765 /*
4a69a410
CM
766 * when we're called for a write, we're already in the async
767 * submission context. Just jump into btrfs_map_bio
8b712842 768 */
8b712842 769 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
0b86a832
CM
770}
771
44b8bd7e 772static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
eaf25d93
CM
773 int mirror_num, unsigned long bio_flags,
774 u64 bio_offset)
44b8bd7e 775{
cad321ad
CM
776 int ret;
777
778 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
779 bio, 1);
780 BUG_ON(ret);
781
7b6d91da 782 if (!(rw & REQ_WRITE)) {
4a69a410
CM
783 /*
784 * called for a read, do the setup so that checksum validation
785 * can happen in the async kernel threads
786 */
4a69a410 787 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
6f3577bd 788 mirror_num, 0);
44b8bd7e 789 }
d313d7a3 790
cad321ad
CM
791 /*
792 * kthread helpers are used to submit writes so that checksumming
793 * can happen in parallel across all CPUs
794 */
44b8bd7e 795 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
c8b97818 796 inode, rw, bio, mirror_num, 0,
eaf25d93 797 bio_offset,
4a69a410
CM
798 __btree_submit_bio_start,
799 __btree_submit_bio_done);
44b8bd7e
CM
800}
801
3dd1462e 802#ifdef CONFIG_MIGRATION
784b4e29
CM
803static int btree_migratepage(struct address_space *mapping,
804 struct page *newpage, struct page *page)
805{
806 /*
807 * we can't safely write a btree page from here,
808 * we haven't done the locking hook
809 */
810 if (PageDirty(page))
811 return -EAGAIN;
812 /*
813 * Buffers may be managed in a filesystem specific way.
814 * We must have no buffers or drop them.
815 */
816 if (page_has_private(page) &&
817 !try_to_release_page(page, GFP_KERNEL))
818 return -EAGAIN;
784b4e29
CM
819 return migrate_page(mapping, newpage, page);
820}
3dd1462e 821#endif
784b4e29 822
0da5468f
CM
823static int btree_writepage(struct page *page, struct writeback_control *wbc)
824{
d1310b2e 825 struct extent_io_tree *tree;
b9473439
CM
826 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
827 struct extent_buffer *eb;
828 int was_dirty;
829
d1310b2e 830 tree = &BTRFS_I(page->mapping->host)->io_tree;
b9473439
CM
831 if (!(current->flags & PF_MEMALLOC)) {
832 return extent_write_full_page(tree, page,
833 btree_get_extent, wbc);
834 }
5443be45 835
b9473439 836 redirty_page_for_writepage(wbc, page);
784b4e29 837 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
b9473439
CM
838 WARN_ON(!eb);
839
840 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
841 if (!was_dirty) {
842 spin_lock(&root->fs_info->delalloc_lock);
843 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
844 spin_unlock(&root->fs_info->delalloc_lock);
5443be45 845 }
b9473439
CM
846 free_extent_buffer(eb);
847
848 unlock_page(page);
849 return 0;
5f39d397 850}
0da5468f
CM
851
852static int btree_writepages(struct address_space *mapping,
853 struct writeback_control *wbc)
854{
d1310b2e
CM
855 struct extent_io_tree *tree;
856 tree = &BTRFS_I(mapping->host)->io_tree;
d8d5f3e1 857 if (wbc->sync_mode == WB_SYNC_NONE) {
b9473439 858 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
793955bc 859 u64 num_dirty;
24ab9cd8 860 unsigned long thresh = 32 * 1024 * 1024;
448d640b
CM
861
862 if (wbc->for_kupdate)
863 return 0;
864
b9473439
CM
865 /* this is a bit racy, but that's ok */
866 num_dirty = root->fs_info->dirty_metadata_bytes;
d397712b 867 if (num_dirty < thresh)
793955bc 868 return 0;
793955bc 869 }
0da5468f
CM
870 return extent_writepages(tree, mapping, btree_get_extent, wbc);
871}
872
b2950863 873static int btree_readpage(struct file *file, struct page *page)
5f39d397 874{
d1310b2e
CM
875 struct extent_io_tree *tree;
876 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM
877 return extent_read_full_page(tree, page, btree_get_extent);
878}
22b0ebda 879
70dec807 880static int btree_releasepage(struct page *page, gfp_t gfp_flags)
5f39d397 881{
d1310b2e
CM
882 struct extent_io_tree *tree;
883 struct extent_map_tree *map;
5f39d397 884 int ret;
d98237b3 885
98509cfc 886 if (PageWriteback(page) || PageDirty(page))
d397712b 887 return 0;
98509cfc 888
d1310b2e
CM
889 tree = &BTRFS_I(page->mapping->host)->io_tree;
890 map = &BTRFS_I(page->mapping->host)->extent_tree;
6af118ce 891
7b13b7b1 892 ret = try_release_extent_state(map, tree, page, gfp_flags);
d397712b 893 if (!ret)
6af118ce 894 return 0;
6af118ce
CM
895
896 ret = try_release_extent_buffer(tree, page);
5f39d397
CM
897 if (ret == 1) {
898 ClearPagePrivate(page);
899 set_page_private(page, 0);
900 page_cache_release(page);
901 }
6af118ce 902
d98237b3
CM
903 return ret;
904}
905
5f39d397 906static void btree_invalidatepage(struct page *page, unsigned long offset)
d98237b3 907{
d1310b2e
CM
908 struct extent_io_tree *tree;
909 tree = &BTRFS_I(page->mapping->host)->io_tree;
5f39d397
CM
910 extent_invalidatepage(tree, page, offset);
911 btree_releasepage(page, GFP_NOFS);
9ad6b7bc 912 if (PagePrivate(page)) {
d397712b
CM
913 printk(KERN_WARNING "btrfs warning page private not zero "
914 "on page %llu\n", (unsigned long long)page_offset(page));
9ad6b7bc
CM
915 ClearPagePrivate(page);
916 set_page_private(page, 0);
917 page_cache_release(page);
918 }
d98237b3
CM
919}
920
7f09410b 921static const struct address_space_operations btree_aops = {
d98237b3
CM
922 .readpage = btree_readpage,
923 .writepage = btree_writepage,
0da5468f 924 .writepages = btree_writepages,
5f39d397
CM
925 .releasepage = btree_releasepage,
926 .invalidatepage = btree_invalidatepage,
5a92bc88 927#ifdef CONFIG_MIGRATION
784b4e29 928 .migratepage = btree_migratepage,
5a92bc88 929#endif
d98237b3
CM
930};
931
ca7a79ad
CM
932int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
933 u64 parent_transid)
090d1875 934{
5f39d397
CM
935 struct extent_buffer *buf = NULL;
936 struct inode *btree_inode = root->fs_info->btree_inode;
de428b63 937 int ret = 0;
090d1875 938
db94535d 939 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
5f39d397 940 if (!buf)
090d1875 941 return 0;
d1310b2e 942 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
f188591e 943 buf, 0, 0, btree_get_extent, 0);
5f39d397 944 free_extent_buffer(buf);
de428b63 945 return ret;
090d1875
CM
946}
947
0999df54
CM
948struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
949 u64 bytenr, u32 blocksize)
950{
951 struct inode *btree_inode = root->fs_info->btree_inode;
952 struct extent_buffer *eb;
953 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
f09d1f60 954 bytenr, blocksize);
0999df54
CM
955 return eb;
956}
957
958struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
959 u64 bytenr, u32 blocksize)
960{
961 struct inode *btree_inode = root->fs_info->btree_inode;
962 struct extent_buffer *eb;
963
964 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
ba144192 965 bytenr, blocksize, NULL);
0999df54
CM
966 return eb;
967}
968
969
e02119d5
CM
970int btrfs_write_tree_block(struct extent_buffer *buf)
971{
8aa38c31
CH
972 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
973 buf->start + buf->len - 1);
e02119d5
CM
974}
975
976int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
977{
8aa38c31
CH
978 return filemap_fdatawait_range(buf->first_page->mapping,
979 buf->start, buf->start + buf->len - 1);
e02119d5
CM
980}
981
0999df54 982struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
ca7a79ad 983 u32 blocksize, u64 parent_transid)
0999df54
CM
984{
985 struct extent_buffer *buf = NULL;
0999df54
CM
986 int ret;
987
0999df54
CM
988 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
989 if (!buf)
990 return NULL;
0999df54 991
ca7a79ad 992 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
ce9adaa5 993
d397712b 994 if (ret == 0)
b4ce94de 995 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
5f39d397 996 return buf;
ce9adaa5 997
eb60ceac
CM
998}
999
e089f05c 1000int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
5f39d397 1001 struct extent_buffer *buf)
ed2ff2cb 1002{
5f39d397 1003 struct inode *btree_inode = root->fs_info->btree_inode;
55c69072 1004 if (btrfs_header_generation(buf) ==
925baedd 1005 root->fs_info->running_transaction->transid) {
b9447ef8 1006 btrfs_assert_tree_locked(buf);
b4ce94de 1007
b9473439
CM
1008 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1009 spin_lock(&root->fs_info->delalloc_lock);
1010 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1011 root->fs_info->dirty_metadata_bytes -= buf->len;
1012 else
1013 WARN_ON(1);
1014 spin_unlock(&root->fs_info->delalloc_lock);
1015 }
b4ce94de 1016
b9473439
CM
1017 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1018 btrfs_set_lock_blocking(buf);
d1310b2e 1019 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
55c69072 1020 buf);
925baedd 1021 }
5f39d397
CM
1022 return 0;
1023}
1024
db94535d 1025static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
87ee04eb 1026 u32 stripesize, struct btrfs_root *root,
9f5fae2f 1027 struct btrfs_fs_info *fs_info,
e20d96d6 1028 u64 objectid)
d97e63b6 1029{
cfaa7295 1030 root->node = NULL;
a28ec197 1031 root->commit_root = NULL;
db94535d
CM
1032 root->sectorsize = sectorsize;
1033 root->nodesize = nodesize;
1034 root->leafsize = leafsize;
87ee04eb 1035 root->stripesize = stripesize;
123abc88 1036 root->ref_cows = 0;
0b86a832 1037 root->track_dirty = 0;
c71bf099 1038 root->in_radix = 0;
d68fc57b
YZ
1039 root->orphan_item_inserted = 0;
1040 root->orphan_cleanup_state = 0;
0b86a832 1041
9f5fae2f 1042 root->fs_info = fs_info;
0f7d52f4
CM
1043 root->objectid = objectid;
1044 root->last_trans = 0;
13a8a7c8 1045 root->highest_objectid = 0;
58176a96 1046 root->name = NULL;
4313b399 1047 root->in_sysfs = 0;
6bef4d31 1048 root->inode_tree = RB_ROOT;
16cdcec7 1049 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
f0486c68 1050 root->block_rsv = NULL;
d68fc57b 1051 root->orphan_block_rsv = NULL;
0b86a832
CM
1052
1053 INIT_LIST_HEAD(&root->dirty_list);
7b128766 1054 INIT_LIST_HEAD(&root->orphan_list);
5d4f98a2 1055 INIT_LIST_HEAD(&root->root_list);
d68fc57b 1056 spin_lock_init(&root->orphan_lock);
5d4f98a2 1057 spin_lock_init(&root->inode_lock);
f0486c68 1058 spin_lock_init(&root->accounting_lock);
a2135011 1059 mutex_init(&root->objectid_mutex);
e02119d5 1060 mutex_init(&root->log_mutex);
7237f183
YZ
1061 init_waitqueue_head(&root->log_writer_wait);
1062 init_waitqueue_head(&root->log_commit_wait[0]);
1063 init_waitqueue_head(&root->log_commit_wait[1]);
1064 atomic_set(&root->log_commit[0], 0);
1065 atomic_set(&root->log_commit[1], 0);
1066 atomic_set(&root->log_writers, 0);
1067 root->log_batch = 0;
1068 root->log_transid = 0;
257c62e1 1069 root->last_log_commit = 0;
d0c803c4 1070 extent_io_tree_init(&root->dirty_log_pages,
f993c883 1071 fs_info->btree_inode->i_mapping);
017e5369 1072
3768f368
CM
1073 memset(&root->root_key, 0, sizeof(root->root_key));
1074 memset(&root->root_item, 0, sizeof(root->root_item));
6702ed49 1075 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
58176a96 1076 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
3f157a2f 1077 root->defrag_trans_start = fs_info->generation;
58176a96 1078 init_completion(&root->kobj_unregister);
6702ed49 1079 root->defrag_running = 0;
4d775673 1080 root->root_key.objectid = objectid;
3394e160
CM
1081 root->anon_super.s_root = NULL;
1082 root->anon_super.s_dev = 0;
1083 INIT_LIST_HEAD(&root->anon_super.s_list);
1084 INIT_LIST_HEAD(&root->anon_super.s_instances);
1085 init_rwsem(&root->anon_super.s_umount);
1086
3768f368
CM
1087 return 0;
1088}
1089
db94535d 1090static int find_and_setup_root(struct btrfs_root *tree_root,
9f5fae2f
CM
1091 struct btrfs_fs_info *fs_info,
1092 u64 objectid,
e20d96d6 1093 struct btrfs_root *root)
3768f368
CM
1094{
1095 int ret;
db94535d 1096 u32 blocksize;
84234f3a 1097 u64 generation;
3768f368 1098
db94535d 1099 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM
1100 tree_root->sectorsize, tree_root->stripesize,
1101 root, fs_info, objectid);
3768f368
CM
1102 ret = btrfs_find_last_root(tree_root, objectid,
1103 &root->root_item, &root->root_key);
4df27c4d
YZ
1104 if (ret > 0)
1105 return -ENOENT;
3768f368
CM
1106 BUG_ON(ret);
1107
84234f3a 1108 generation = btrfs_root_generation(&root->root_item);
db94535d
CM
1109 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1110 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1111 blocksize, generation);
68433b73
CM
1112 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1113 free_extent_buffer(root->node);
1114 return -EIO;
1115 }
4df27c4d 1116 root->commit_root = btrfs_root_node(root);
d97e63b6
CM
1117 return 0;
1118}
1119
7237f183
YZ
1120static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1121 struct btrfs_fs_info *fs_info)
0f7d52f4
CM
1122{
1123 struct btrfs_root *root;
1124 struct btrfs_root *tree_root = fs_info->tree_root;
7237f183 1125 struct extent_buffer *leaf;
e02119d5
CM
1126
1127 root = kzalloc(sizeof(*root), GFP_NOFS);
1128 if (!root)
7237f183 1129 return ERR_PTR(-ENOMEM);
e02119d5
CM
1130
1131 __setup_root(tree_root->nodesize, tree_root->leafsize,
1132 tree_root->sectorsize, tree_root->stripesize,
1133 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1134
1135 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1136 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1137 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
7237f183
YZ
1138 /*
1139 * log trees do not get reference counted because they go away
1140 * before a real commit is actually done. They do store pointers
1141 * to file data extents, and those reference counts still get
1142 * updated (along with back refs to the log tree).
1143 */
e02119d5
CM
1144 root->ref_cows = 0;
1145
5d4f98a2
YZ
1146 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1147 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
7237f183
YZ
1148 if (IS_ERR(leaf)) {
1149 kfree(root);
1150 return ERR_CAST(leaf);
1151 }
e02119d5 1152
5d4f98a2
YZ
1153 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1154 btrfs_set_header_bytenr(leaf, leaf->start);
1155 btrfs_set_header_generation(leaf, trans->transid);
1156 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1157 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
7237f183 1158 root->node = leaf;
e02119d5
CM
1159
1160 write_extent_buffer(root->node, root->fs_info->fsid,
1161 (unsigned long)btrfs_header_fsid(root->node),
1162 BTRFS_FSID_SIZE);
1163 btrfs_mark_buffer_dirty(root->node);
1164 btrfs_tree_unlock(root->node);
7237f183
YZ
1165 return root;
1166}
1167
1168int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1169 struct btrfs_fs_info *fs_info)
1170{
1171 struct btrfs_root *log_root;
1172
1173 log_root = alloc_log_tree(trans, fs_info);
1174 if (IS_ERR(log_root))
1175 return PTR_ERR(log_root);
1176 WARN_ON(fs_info->log_root_tree);
1177 fs_info->log_root_tree = log_root;
1178 return 0;
1179}
1180
1181int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1182 struct btrfs_root *root)
1183{
1184 struct btrfs_root *log_root;
1185 struct btrfs_inode_item *inode_item;
1186
1187 log_root = alloc_log_tree(trans, root->fs_info);
1188 if (IS_ERR(log_root))
1189 return PTR_ERR(log_root);
1190
1191 log_root->last_trans = trans->transid;
1192 log_root->root_key.offset = root->root_key.objectid;
1193
1194 inode_item = &log_root->root_item.inode;
1195 inode_item->generation = cpu_to_le64(1);
1196 inode_item->size = cpu_to_le64(3);
1197 inode_item->nlink = cpu_to_le32(1);
1198 inode_item->nbytes = cpu_to_le64(root->leafsize);
1199 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1200
5d4f98a2 1201 btrfs_set_root_node(&log_root->root_item, log_root->node);
7237f183
YZ
1202
1203 WARN_ON(root->log_root);
1204 root->log_root = log_root;
1205 root->log_transid = 0;
257c62e1 1206 root->last_log_commit = 0;
e02119d5
CM
1207 return 0;
1208}
1209
1210struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1211 struct btrfs_key *location)
1212{
1213 struct btrfs_root *root;
1214 struct btrfs_fs_info *fs_info = tree_root->fs_info;
0f7d52f4 1215 struct btrfs_path *path;
5f39d397 1216 struct extent_buffer *l;
84234f3a 1217 u64 generation;
db94535d 1218 u32 blocksize;
0f7d52f4
CM
1219 int ret = 0;
1220
5eda7b5e 1221 root = kzalloc(sizeof(*root), GFP_NOFS);
0cf6c620 1222 if (!root)
0f7d52f4 1223 return ERR_PTR(-ENOMEM);
0f7d52f4 1224 if (location->offset == (u64)-1) {
db94535d 1225 ret = find_and_setup_root(tree_root, fs_info,
0f7d52f4
CM
1226 location->objectid, root);
1227 if (ret) {
0f7d52f4
CM
1228 kfree(root);
1229 return ERR_PTR(ret);
1230 }
13a8a7c8 1231 goto out;
0f7d52f4
CM
1232 }
1233
db94535d 1234 __setup_root(tree_root->nodesize, tree_root->leafsize,
87ee04eb
CM
1235 tree_root->sectorsize, tree_root->stripesize,
1236 root, fs_info, location->objectid);
0f7d52f4
CM
1237
1238 path = btrfs_alloc_path();
db5b493a
TI
1239 if (!path) {
1240 kfree(root);
1241 return ERR_PTR(-ENOMEM);
1242 }
0f7d52f4 1243 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
13a8a7c8
YZ
1244 if (ret == 0) {
1245 l = path->nodes[0];
1246 read_extent_buffer(l, &root->root_item,
1247 btrfs_item_ptr_offset(l, path->slots[0]),
1248 sizeof(root->root_item));
1249 memcpy(&root->root_key, location, sizeof(*location));
0f7d52f4 1250 }
0f7d52f4
CM
1251 btrfs_free_path(path);
1252 if (ret) {
5e540f77 1253 kfree(root);
13a8a7c8
YZ
1254 if (ret > 0)
1255 ret = -ENOENT;
0f7d52f4
CM
1256 return ERR_PTR(ret);
1257 }
13a8a7c8 1258
84234f3a 1259 generation = btrfs_root_generation(&root->root_item);
db94535d
CM
1260 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1261 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
84234f3a 1262 blocksize, generation);
5d4f98a2 1263 root->commit_root = btrfs_root_node(root);
0f7d52f4 1264 BUG_ON(!root->node);
13a8a7c8 1265out:
08fe4db1 1266 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
e02119d5 1267 root->ref_cows = 1;
08fe4db1
LZ
1268 btrfs_check_and_init_root_item(&root->root_item);
1269 }
13a8a7c8 1270
5eda7b5e
CM
1271 return root;
1272}
1273
edbd8d4e
CM
1274struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1275 struct btrfs_key *location)
5eda7b5e
CM
1276{
1277 struct btrfs_root *root;
1278 int ret;
1279
edbd8d4e
CM
1280 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1281 return fs_info->tree_root;
1282 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1283 return fs_info->extent_root;
8f18cf13
CM
1284 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1285 return fs_info->chunk_root;
1286 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1287 return fs_info->dev_root;
0403e47e
YZ
1288 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1289 return fs_info->csum_root;
4df27c4d
YZ
1290again:
1291 spin_lock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM
1292 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1293 (unsigned long)location->objectid);
4df27c4d 1294 spin_unlock(&fs_info->fs_roots_radix_lock);
5eda7b5e
CM
1295 if (root)
1296 return root;
1297
e02119d5 1298 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
5eda7b5e
CM
1299 if (IS_ERR(root))
1300 return root;
3394e160 1301
581bb050
LZ
1302 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1303 if (!root->free_ino_ctl)
1304 goto fail;
1305 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1306 GFP_NOFS);
1307 if (!root->free_ino_pinned)
1308 goto fail;
1309
1310 btrfs_init_free_ino_ctl(root);
1311 mutex_init(&root->fs_commit_mutex);
1312 spin_lock_init(&root->cache_lock);
1313 init_waitqueue_head(&root->cache_wait);
1314
3394e160
CM
1315 set_anon_super(&root->anon_super, NULL);
1316
d68fc57b
YZ
1317 if (btrfs_root_refs(&root->root_item) == 0) {
1318 ret = -ENOENT;
1319 goto fail;
1320 }
1321
1322 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1323 if (ret < 0)
1324 goto fail;
1325 if (ret == 0)
1326 root->orphan_item_inserted = 1;
1327
4df27c4d
YZ
1328 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1329 if (ret)
1330 goto fail;
1331
1332 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM
1333 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1334 (unsigned long)root->root_key.objectid,
0f7d52f4 1335 root);
d68fc57b 1336 if (ret == 0)
4df27c4d 1337 root->in_radix = 1;
d68fc57b 1338
4df27c4d
YZ
1339 spin_unlock(&fs_info->fs_roots_radix_lock);
1340 radix_tree_preload_end();
0f7d52f4 1341 if (ret) {
4df27c4d
YZ
1342 if (ret == -EEXIST) {
1343 free_fs_root(root);
1344 goto again;
1345 }
1346 goto fail;
0f7d52f4 1347 }
4df27c4d
YZ
1348
1349 ret = btrfs_find_dead_roots(fs_info->tree_root,
1350 root->root_key.objectid);
1351 WARN_ON(ret);
edbd8d4e 1352 return root;
4df27c4d
YZ
1353fail:
1354 free_fs_root(root);
1355 return ERR_PTR(ret);
edbd8d4e
CM
1356}
1357
04160088
CM
1358static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1359{
1360 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1361 int ret = 0;
04160088
CM
1362 struct btrfs_device *device;
1363 struct backing_dev_info *bdi;
b7967db7 1364
1f78160c
XG
1365 rcu_read_lock();
1366 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
dfe25020
CM
1367 if (!device->bdev)
1368 continue;
04160088
CM
1369 bdi = blk_get_backing_dev_info(device->bdev);
1370 if (bdi && bdi_congested(bdi, bdi_bits)) {
1371 ret = 1;
1372 break;
1373 }
1374 }
1f78160c 1375 rcu_read_unlock();
04160088
CM
1376 return ret;
1377}
1378
ad081f14
JA
1379/*
1380 * If this fails, caller must call bdi_destroy() to get rid of the
1381 * bdi again.
1382 */
04160088
CM
1383static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1384{
ad081f14
JA
1385 int err;
1386
1387 bdi->capabilities = BDI_CAP_MAP_COPY;
e6d086d8 1388 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
ad081f14
JA
1389 if (err)
1390 return err;
1391
4575c9cc 1392 bdi->ra_pages = default_backing_dev_info.ra_pages;
04160088
CM
1393 bdi->congested_fn = btrfs_congested_fn;
1394 bdi->congested_data = info;
1395 return 0;
1396}
1397
ce9adaa5
CM
1398static int bio_ready_for_csum(struct bio *bio)
1399{
1400 u64 length = 0;
1401 u64 buf_len = 0;
1402 u64 start = 0;
1403 struct page *page;
1404 struct extent_io_tree *io_tree = NULL;
ce9adaa5
CM
1405 struct bio_vec *bvec;
1406 int i;
1407 int ret;
1408
1409 bio_for_each_segment(bvec, bio, i) {
1410 page = bvec->bv_page;
1411 if (page->private == EXTENT_PAGE_PRIVATE) {
1412 length += bvec->bv_len;
1413 continue;
1414 }
1415 if (!page->private) {
1416 length += bvec->bv_len;
1417 continue;
1418 }
1419 length = bvec->bv_len;
1420 buf_len = page->private >> 2;
1421 start = page_offset(page) + bvec->bv_offset;
1422 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
ce9adaa5
CM
1423 }
1424 /* are we fully contained in this bio? */
1425 if (buf_len <= length)
1426 return 1;
1427
1428 ret = extent_range_uptodate(io_tree, start + length,
1429 start + buf_len - 1);
ce9adaa5
CM
1430 return ret;
1431}
1432
8b712842
CM
1433/*
1434 * called by the kthread helper functions to finally call the bio end_io
1435 * functions. This is where read checksum verification actually happens
1436 */
1437static void end_workqueue_fn(struct btrfs_work *work)
ce9adaa5 1438{
ce9adaa5 1439 struct bio *bio;
8b712842
CM
1440 struct end_io_wq *end_io_wq;
1441 struct btrfs_fs_info *fs_info;
ce9adaa5 1442 int error;
ce9adaa5 1443
8b712842
CM
1444 end_io_wq = container_of(work, struct end_io_wq, work);
1445 bio = end_io_wq->bio;
1446 fs_info = end_io_wq->info;
ce9adaa5 1447
cad321ad 1448 /* metadata bio reads are special because the whole tree block must
8b712842
CM
1449 * be checksummed at once. This makes sure the entire block is in
1450 * ram and up to date before trying to verify things. For
1451 * blocksize <= pagesize, it is basically a noop
1452 */
7b6d91da 1453 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
cad321ad 1454 !bio_ready_for_csum(bio)) {
d20f7043 1455 btrfs_queue_worker(&fs_info->endio_meta_workers,
8b712842
CM
1456 &end_io_wq->work);
1457 return;
1458 }
1459 error = end_io_wq->error;
1460 bio->bi_private = end_io_wq->private;
1461 bio->bi_end_io = end_io_wq->end_io;
1462 kfree(end_io_wq);
8b712842 1463 bio_endio(bio, error);
44b8bd7e
CM
1464}
1465
a74a4b97
CM
1466static int cleaner_kthread(void *arg)
1467{
1468 struct btrfs_root *root = arg;
1469
1470 do {
a74a4b97 1471 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
76dda93c
YZ
1472
1473 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1474 mutex_trylock(&root->fs_info->cleaner_mutex)) {
24bbcf04 1475 btrfs_run_delayed_iputs(root);
76dda93c
YZ
1476 btrfs_clean_old_snapshots(root);
1477 mutex_unlock(&root->fs_info->cleaner_mutex);
1478 }
a74a4b97
CM
1479
1480 if (freezing(current)) {
1481 refrigerator();
1482 } else {
a74a4b97 1483 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ
1484 if (!kthread_should_stop())
1485 schedule();
a74a4b97
CM
1486 __set_current_state(TASK_RUNNING);
1487 }
1488 } while (!kthread_should_stop());
1489 return 0;
1490}
1491
1492static int transaction_kthread(void *arg)
1493{
1494 struct btrfs_root *root = arg;
1495 struct btrfs_trans_handle *trans;
1496 struct btrfs_transaction *cur;
8929ecfa 1497 u64 transid;
a74a4b97
CM
1498 unsigned long now;
1499 unsigned long delay;
1500 int ret;
1501
1502 do {
a74a4b97
CM
1503 delay = HZ * 30;
1504 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1505 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1506
8929ecfa 1507 spin_lock(&root->fs_info->new_trans_lock);
a74a4b97
CM
1508 cur = root->fs_info->running_transaction;
1509 if (!cur) {
8929ecfa 1510 spin_unlock(&root->fs_info->new_trans_lock);
a74a4b97
CM
1511 goto sleep;
1512 }
31153d81 1513
a74a4b97 1514 now = get_seconds();
8929ecfa
YZ
1515 if (!cur->blocked &&
1516 (now < cur->start_time || now - cur->start_time < 30)) {
1517 spin_unlock(&root->fs_info->new_trans_lock);
a74a4b97
CM
1518 delay = HZ * 5;
1519 goto sleep;
1520 }
8929ecfa
YZ
1521 transid = cur->transid;
1522 spin_unlock(&root->fs_info->new_trans_lock);
56bec294 1523
8929ecfa 1524 trans = btrfs_join_transaction(root, 1);
3612b495 1525 BUG_ON(IS_ERR(trans));
8929ecfa
YZ
1526 if (transid == trans->transid) {
1527 ret = btrfs_commit_transaction(trans, root);
1528 BUG_ON(ret);
1529 } else {
1530 btrfs_end_transaction(trans, root);
1531 }
a74a4b97
CM
1532sleep:
1533 wake_up_process(root->fs_info->cleaner_kthread);
1534 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1535
1536 if (freezing(current)) {
1537 refrigerator();
1538 } else {
a74a4b97 1539 set_current_state(TASK_INTERRUPTIBLE);
8929ecfa
YZ
1540 if (!kthread_should_stop() &&
1541 !btrfs_transaction_blocked(root->fs_info))
1542 schedule_timeout(delay);
a74a4b97
CM
1543 __set_current_state(TASK_RUNNING);
1544 }
1545 } while (!kthread_should_stop());
1546 return 0;
1547}
1548
8a4b83cc 1549struct btrfs_root *open_ctree(struct super_block *sb,
dfe25020
CM
1550 struct btrfs_fs_devices *fs_devices,
1551 char *options)
2e635a27 1552{
db94535d
CM
1553 u32 sectorsize;
1554 u32 nodesize;
1555 u32 leafsize;
1556 u32 blocksize;
87ee04eb 1557 u32 stripesize;
84234f3a 1558 u64 generation;
f2b636e8 1559 u64 features;
3de4586c 1560 struct btrfs_key location;
a061fc8d 1561 struct buffer_head *bh;
e02119d5 1562 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
e20d96d6 1563 GFP_NOFS);
d20f7043
CM
1564 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1565 GFP_NOFS);
450ba0ea 1566 struct btrfs_root *tree_root = btrfs_sb(sb);
4891aca2 1567 struct btrfs_fs_info *fs_info = NULL;
e02119d5 1568 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1569 GFP_NOFS);
e02119d5 1570 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
0b86a832 1571 GFP_NOFS);
e02119d5
CM
1572 struct btrfs_root *log_tree_root;
1573
eb60ceac 1574 int ret;
e58ca020 1575 int err = -EINVAL;
4543df7e 1576
2c90e5d6 1577 struct btrfs_super_block *disk_super;
8790d502 1578
4891aca2 1579 if (!extent_root || !tree_root || !tree_root->fs_info ||
d20f7043 1580 !chunk_root || !dev_root || !csum_root) {
39279cc3
CM
1581 err = -ENOMEM;
1582 goto fail;
1583 }
4891aca2 1584 fs_info = tree_root->fs_info;
76dda93c
YZ
1585
1586 ret = init_srcu_struct(&fs_info->subvol_srcu);
1587 if (ret) {
1588 err = ret;
1589 goto fail;
1590 }
1591
1592 ret = setup_bdi(fs_info, &fs_info->bdi);
1593 if (ret) {
1594 err = ret;
1595 goto fail_srcu;
1596 }
1597
1598 fs_info->btree_inode = new_inode(sb);
1599 if (!fs_info->btree_inode) {
1600 err = -ENOMEM;
1601 goto fail_bdi;
1602 }
1603
1561deda
MX
1604 fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
1605
76dda93c 1606 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
8fd17795 1607 INIT_LIST_HEAD(&fs_info->trans_list);
facda1e7 1608 INIT_LIST_HEAD(&fs_info->dead_roots);
24bbcf04 1609 INIT_LIST_HEAD(&fs_info->delayed_iputs);
19c00ddc 1610 INIT_LIST_HEAD(&fs_info->hashers);
ea8c2819 1611 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
5a3f23d5 1612 INIT_LIST_HEAD(&fs_info->ordered_operations);
11833d66 1613 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1832a6d5 1614 spin_lock_init(&fs_info->delalloc_lock);
cee36a03 1615 spin_lock_init(&fs_info->new_trans_lock);
31153d81 1616 spin_lock_init(&fs_info->ref_cache_lock);
76dda93c 1617 spin_lock_init(&fs_info->fs_roots_radix_lock);
24bbcf04 1618 spin_lock_init(&fs_info->delayed_iput_lock);
19c00ddc 1619
58176a96 1620 init_completion(&fs_info->kobj_unregister);
9f5fae2f
CM
1621 fs_info->tree_root = tree_root;
1622 fs_info->extent_root = extent_root;
d20f7043 1623 fs_info->csum_root = csum_root;
0b86a832
CM
1624 fs_info->chunk_root = chunk_root;
1625 fs_info->dev_root = dev_root;
8a4b83cc 1626 fs_info->fs_devices = fs_devices;
0b86a832 1627 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
6324fbf3 1628 INIT_LIST_HEAD(&fs_info->space_info);
0b86a832 1629 btrfs_mapping_init(&fs_info->mapping_tree);
f0486c68
YZ
1630 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1631 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1632 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1633 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1634 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1635 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1636 mutex_init(&fs_info->durable_block_rsv_mutex);
cb03c743 1637 atomic_set(&fs_info->nr_async_submits, 0);
771ed689 1638 atomic_set(&fs_info->async_delalloc_pages, 0);
8c8bee1d 1639 atomic_set(&fs_info->async_submit_draining, 0);
0986fe9e 1640 atomic_set(&fs_info->nr_async_bios, 0);
e20d96d6 1641 fs_info->sb = sb;
6f568d35 1642 fs_info->max_inline = 8192 * 1024;
9ed74f2d 1643 fs_info->metadata_ratio = 0;
c8b97818 1644
b34b086c
CM
1645 fs_info->thread_pool_size = min_t(unsigned long,
1646 num_online_cpus() + 2, 8);
0afbaf8c 1647
3eaa2885
CM
1648 INIT_LIST_HEAD(&fs_info->ordered_extents);
1649 spin_lock_init(&fs_info->ordered_extent_lock);
16cdcec7
MX
1650 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1651 GFP_NOFS);
1652 if (!fs_info->delayed_root) {
1653 err = -ENOMEM;
1654 goto fail_iput;
1655 }
1656 btrfs_init_delayed_root(fs_info->delayed_root);
3eaa2885 1657
a2de733c
AJ
1658 mutex_init(&fs_info->scrub_lock);
1659 atomic_set(&fs_info->scrubs_running, 0);
1660 atomic_set(&fs_info->scrub_pause_req, 0);
1661 atomic_set(&fs_info->scrubs_paused, 0);
1662 atomic_set(&fs_info->scrub_cancel_req, 0);
1663 init_waitqueue_head(&fs_info->scrub_pause_wait);
1664 init_rwsem(&fs_info->scrub_super_lock);
1665 fs_info->scrub_workers_refcnt = 0;
1666 btrfs_init_workers(&fs_info->scrub_workers, "scrub",
1667 fs_info->thread_pool_size, &fs_info->generic_worker);
1668
a061fc8d
CM
1669 sb->s_blocksize = 4096;
1670 sb->s_blocksize_bits = blksize_bits(4096);
32a88aa1 1671 sb->s_bdi = &fs_info->bdi;
a061fc8d 1672
76dda93c
YZ
1673 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1674 fs_info->btree_inode->i_nlink = 1;
0afbaf8c
CM
1675 /*
1676 * we set the i_size on the btree inode to the max possible int.
1677 * the real end of the address space is determined by all of
1678 * the devices in the system
1679 */
1680 fs_info->btree_inode->i_size = OFFSET_MAX;
d98237b3 1681 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
04160088
CM
1682 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1683
5d4f98a2 1684 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
d1310b2e 1685 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
f993c883 1686 fs_info->btree_inode->i_mapping);
a8067e02 1687 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
d1310b2e
CM
1688
1689 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
0da5468f 1690
76dda93c
YZ
1691 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1692 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1693 sizeof(struct btrfs_key));
1694 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
c65ddb52 1695 insert_inode_hash(fs_info->btree_inode);
76dda93c 1696
0f9dd46c 1697 spin_lock_init(&fs_info->block_group_cache_lock);
6bef4d31 1698 fs_info->block_group_cache_tree = RB_ROOT;
0f9dd46c 1699
11833d66 1700 extent_io_tree_init(&fs_info->freed_extents[0],
f993c883 1701 fs_info->btree_inode->i_mapping);
11833d66 1702 extent_io_tree_init(&fs_info->freed_extents[1],
f993c883 1703 fs_info->btree_inode->i_mapping);
11833d66 1704 fs_info->pinned_extents = &fs_info->freed_extents[0];
e66f709b 1705 fs_info->do_barriers = 1;
e18e4809 1706
39279cc3 1707
79154b1b 1708 mutex_init(&fs_info->trans_mutex);
5a3f23d5 1709 mutex_init(&fs_info->ordered_operations_mutex);
e02119d5 1710 mutex_init(&fs_info->tree_log_mutex);
925baedd 1711 mutex_init(&fs_info->chunk_mutex);
a74a4b97
CM
1712 mutex_init(&fs_info->transaction_kthread_mutex);
1713 mutex_init(&fs_info->cleaner_mutex);
7d9eb12c 1714 mutex_init(&fs_info->volume_mutex);
276e680d 1715 init_rwsem(&fs_info->extent_commit_sem);
c71bf099 1716 init_rwsem(&fs_info->cleanup_work_sem);
76dda93c 1717 init_rwsem(&fs_info->subvol_sem);
fa9c0d79
CM
1718
1719 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1720 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1721
e6dcd2dc 1722 init_waitqueue_head(&fs_info->transaction_throttle);
f9295749 1723 init_waitqueue_head(&fs_info->transaction_wait);
bb9c12c9 1724 init_waitqueue_head(&fs_info->transaction_blocked_wait);
4854ddd0 1725 init_waitqueue_head(&fs_info->async_submit_wait);
3768f368 1726
0b86a832 1727 __setup_root(4096, 4096, 4096, 4096, tree_root,
2c90e5d6 1728 fs_info, BTRFS_ROOT_TREE_OBJECTID);
7eccb903 1729
a512bbf8 1730 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
20b45077
DY
1731 if (!bh) {
1732 err = -EINVAL;
16cdcec7 1733 goto fail_alloc;
20b45077 1734 }
39279cc3 1735
a061fc8d 1736 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
2d69a0f8
YZ
1737 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1738 sizeof(fs_info->super_for_commit));
a061fc8d 1739 brelse(bh);
5f39d397 1740
a061fc8d 1741 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
0b86a832 1742
5f39d397 1743 disk_super = &fs_info->super_copy;
0f7d52f4 1744 if (!btrfs_super_root(disk_super))
16cdcec7 1745 goto fail_alloc;
0f7d52f4 1746
acce952b 1747 /* check FS state, whether FS is broken. */
1748 fs_info->fs_state |= btrfs_super_flags(disk_super);
1749
1750 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1751
75e7cb7f
LB
1752 /*
1753 * In the long term, we'll store the compression type in the super
1754 * block, and it'll be used for per file compression control.
1755 */
1756 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1757
2b82032c
YZ
1758 ret = btrfs_parse_options(tree_root, options);
1759 if (ret) {
1760 err = ret;
16cdcec7 1761 goto fail_alloc;
2b82032c 1762 }
dfe25020 1763
f2b636e8
JB
1764 features = btrfs_super_incompat_flags(disk_super) &
1765 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1766 if (features) {
1767 printk(KERN_ERR "BTRFS: couldn't mount because of "
1768 "unsupported optional features (%Lx).\n",
21380931 1769 (unsigned long long)features);
f2b636e8 1770 err = -EINVAL;
16cdcec7 1771 goto fail_alloc;
f2b636e8
JB
1772 }
1773
5d4f98a2 1774 features = btrfs_super_incompat_flags(disk_super);
a6fa6fae
LZ
1775 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1776 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1777 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1778 btrfs_set_super_incompat_flags(disk_super, features);
5d4f98a2 1779
f2b636e8
JB
1780 features = btrfs_super_compat_ro_flags(disk_super) &
1781 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1782 if (!(sb->s_flags & MS_RDONLY) && features) {
1783 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1784 "unsupported option features (%Lx).\n",
21380931 1785 (unsigned long long)features);
f2b636e8 1786 err = -EINVAL;
16cdcec7 1787 goto fail_alloc;
f2b636e8 1788 }
61d92c32
CM
1789
1790 btrfs_init_workers(&fs_info->generic_worker,
1791 "genwork", 1, NULL);
1792
5443be45 1793 btrfs_init_workers(&fs_info->workers, "worker",
61d92c32
CM
1794 fs_info->thread_pool_size,
1795 &fs_info->generic_worker);
c8b97818 1796
771ed689 1797 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
61d92c32
CM
1798 fs_info->thread_pool_size,
1799 &fs_info->generic_worker);
771ed689 1800
5443be45 1801 btrfs_init_workers(&fs_info->submit_workers, "submit",
b720d209 1802 min_t(u64, fs_devices->num_devices,
61d92c32
CM
1803 fs_info->thread_pool_size),
1804 &fs_info->generic_worker);
61b49440
CM
1805
1806 /* a higher idle thresh on the submit workers makes it much more
1807 * likely that bios will be send down in a sane order to the
1808 * devices
1809 */
1810 fs_info->submit_workers.idle_thresh = 64;
53863232 1811
771ed689 1812 fs_info->workers.idle_thresh = 16;
4a69a410 1813 fs_info->workers.ordered = 1;
61b49440 1814
771ed689
CM
1815 fs_info->delalloc_workers.idle_thresh = 2;
1816 fs_info->delalloc_workers.ordered = 1;
1817
61d92c32
CM
1818 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1819 &fs_info->generic_worker);
5443be45 1820 btrfs_init_workers(&fs_info->endio_workers, "endio",
61d92c32
CM
1821 fs_info->thread_pool_size,
1822 &fs_info->generic_worker);
d20f7043 1823 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
61d92c32
CM
1824 fs_info->thread_pool_size,
1825 &fs_info->generic_worker);
cad321ad 1826 btrfs_init_workers(&fs_info->endio_meta_write_workers,
61d92c32
CM
1827 "endio-meta-write", fs_info->thread_pool_size,
1828 &fs_info->generic_worker);
5443be45 1829 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
61d92c32
CM
1830 fs_info->thread_pool_size,
1831 &fs_info->generic_worker);
0cb59c99
JB
1832 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1833 1, &fs_info->generic_worker);
16cdcec7
MX
1834 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
1835 fs_info->thread_pool_size,
1836 &fs_info->generic_worker);
61b49440
CM
1837
1838 /*
1839 * endios are largely parallel and should have a very
1840 * low idle thresh
1841 */
1842 fs_info->endio_workers.idle_thresh = 4;
b51912c9
CM
1843 fs_info->endio_meta_workers.idle_thresh = 4;
1844
9042846b
CM
1845 fs_info->endio_write_workers.idle_thresh = 2;
1846 fs_info->endio_meta_write_workers.idle_thresh = 2;
1847
4543df7e 1848 btrfs_start_workers(&fs_info->workers, 1);
61d92c32 1849 btrfs_start_workers(&fs_info->generic_worker, 1);
1cc127b5 1850 btrfs_start_workers(&fs_info->submit_workers, 1);
771ed689 1851 btrfs_start_workers(&fs_info->delalloc_workers, 1);
247e743c 1852 btrfs_start_workers(&fs_info->fixup_workers, 1);
9042846b
CM
1853 btrfs_start_workers(&fs_info->endio_workers, 1);
1854 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1855 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1856 btrfs_start_workers(&fs_info->endio_write_workers, 1);
0cb59c99 1857 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
16cdcec7 1858 btrfs_start_workers(&fs_info->delayed_workers, 1);
4543df7e 1859
4575c9cc 1860 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
c8b97818
CM
1861 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1862 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
4575c9cc 1863
db94535d
CM
1864 nodesize = btrfs_super_nodesize(disk_super);
1865 leafsize = btrfs_super_leafsize(disk_super);
1866 sectorsize = btrfs_super_sectorsize(disk_super);
87ee04eb 1867 stripesize = btrfs_super_stripesize(disk_super);
db94535d
CM
1868 tree_root->nodesize = nodesize;
1869 tree_root->leafsize = leafsize;
1870 tree_root->sectorsize = sectorsize;
87ee04eb 1871 tree_root->stripesize = stripesize;
a061fc8d
CM
1872
1873 sb->s_blocksize = sectorsize;
1874 sb->s_blocksize_bits = blksize_bits(sectorsize);
db94535d 1875
39279cc3
CM
1876 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1877 sizeof(disk_super->magic))) {
d397712b 1878 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
39279cc3
CM
1879 goto fail_sb_buffer;
1880 }
19c00ddc 1881
925baedd 1882 mutex_lock(&fs_info->chunk_mutex);
e4404d6e 1883 ret = btrfs_read_sys_array(tree_root);
925baedd 1884 mutex_unlock(&fs_info->chunk_mutex);
84eed90f 1885 if (ret) {
d397712b
CM
1886 printk(KERN_WARNING "btrfs: failed to read the system "
1887 "array on %s\n", sb->s_id);
5d4f98a2 1888 goto fail_sb_buffer;
84eed90f 1889 }
0b86a832
CM
1890
1891 blocksize = btrfs_level_size(tree_root,
1892 btrfs_super_chunk_root_level(disk_super));
84234f3a 1893 generation = btrfs_super_chunk_root_generation(disk_super);
0b86a832
CM
1894
1895 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1896 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1897
1898 chunk_root->node = read_tree_block(chunk_root,
1899 btrfs_super_chunk_root(disk_super),
84234f3a 1900 blocksize, generation);
0b86a832 1901 BUG_ON(!chunk_root->node);
83121942
DW
1902 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1903 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1904 sb->s_id);
1905 goto fail_chunk_root;
1906 }
5d4f98a2
YZ
1907 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1908 chunk_root->commit_root = btrfs_root_node(chunk_root);
0b86a832 1909
e17cade2 1910 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
d397712b
CM
1911 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1912 BTRFS_UUID_SIZE);
e17cade2 1913
925baedd 1914 mutex_lock(&fs_info->chunk_mutex);
0b86a832 1915 ret = btrfs_read_chunk_tree(chunk_root);
925baedd 1916 mutex_unlock(&fs_info->chunk_mutex);
2b82032c 1917 if (ret) {
d397712b
CM
1918 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1919 sb->s_id);
2b82032c
YZ
1920 goto fail_chunk_root;
1921 }
0b86a832 1922
dfe25020
CM
1923 btrfs_close_extra_devices(fs_devices);
1924
db94535d
CM
1925 blocksize = btrfs_level_size(tree_root,
1926 btrfs_super_root_level(disk_super));
84234f3a 1927 generation = btrfs_super_generation(disk_super);
0b86a832 1928
e20d96d6 1929 tree_root->node = read_tree_block(tree_root,
db94535d 1930 btrfs_super_root(disk_super),
84234f3a 1931 blocksize, generation);
39279cc3 1932 if (!tree_root->node)
2b82032c 1933 goto fail_chunk_root;
83121942
DW
1934 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1935 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1936 sb->s_id);
1937 goto fail_tree_root;
1938 }
5d4f98a2
YZ
1939 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1940 tree_root->commit_root = btrfs_root_node(tree_root);
db94535d
CM
1941
1942 ret = find_and_setup_root(tree_root, fs_info,
e20d96d6 1943 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
0b86a832 1944 if (ret)
39279cc3 1945 goto fail_tree_root;
0b86a832
CM
1946 extent_root->track_dirty = 1;
1947
1948 ret = find_and_setup_root(tree_root, fs_info,
1949 BTRFS_DEV_TREE_OBJECTID, dev_root);
0b86a832
CM
1950 if (ret)
1951 goto fail_extent_root;
5d4f98a2 1952 dev_root->track_dirty = 1;
3768f368 1953
d20f7043
CM
1954 ret = find_and_setup_root(tree_root, fs_info,
1955 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1956 if (ret)
5d4f98a2 1957 goto fail_dev_root;
d20f7043
CM
1958
1959 csum_root->track_dirty = 1;
1960
8929ecfa
YZ
1961 fs_info->generation = generation;
1962 fs_info->last_trans_committed = generation;
1963 fs_info->data_alloc_profile = (u64)-1;
1964 fs_info->metadata_alloc_profile = (u64)-1;
1965 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1966
c59021f8 1967 ret = btrfs_init_space_info(fs_info);
1968 if (ret) {
1969 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
1970 goto fail_block_groups;
1971 }
1972
1b1d1f66
JB
1973 ret = btrfs_read_block_groups(extent_root);
1974 if (ret) {
1975 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1976 goto fail_block_groups;
1977 }
9078a3e1 1978
a74a4b97
CM
1979 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1980 "btrfs-cleaner");
57506d50 1981 if (IS_ERR(fs_info->cleaner_kthread))
1b1d1f66 1982 goto fail_block_groups;
a74a4b97
CM
1983
1984 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1985 tree_root,
1986 "btrfs-transaction");
57506d50 1987 if (IS_ERR(fs_info->transaction_kthread))
3f157a2f 1988 goto fail_cleaner;
a74a4b97 1989
c289811c
CM
1990 if (!btrfs_test_opt(tree_root, SSD) &&
1991 !btrfs_test_opt(tree_root, NOSSD) &&
1992 !fs_info->fs_devices->rotating) {
1993 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1994 "mode\n");
1995 btrfs_set_opt(fs_info->mount_opt, SSD);
1996 }
1997
acce952b 1998 /* do not make disk changes in broken FS */
1999 if (btrfs_super_log_root(disk_super) != 0 &&
2000 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
e02119d5
CM
2001 u64 bytenr = btrfs_super_log_root(disk_super);
2002
7c2ca468 2003 if (fs_devices->rw_devices == 0) {
d397712b
CM
2004 printk(KERN_WARNING "Btrfs log replay required "
2005 "on RO media\n");
7c2ca468
CM
2006 err = -EIO;
2007 goto fail_trans_kthread;
2008 }
e02119d5
CM
2009 blocksize =
2010 btrfs_level_size(tree_root,
2011 btrfs_super_log_root_level(disk_super));
d18a2c44 2012
676e4c86
DC
2013 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2014 if (!log_tree_root) {
2015 err = -ENOMEM;
2016 goto fail_trans_kthread;
2017 }
e02119d5
CM
2018
2019 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2020 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2021
2022 log_tree_root->node = read_tree_block(tree_root, bytenr,
84234f3a
YZ
2023 blocksize,
2024 generation + 1);
e02119d5
CM
2025 ret = btrfs_recover_log_trees(log_tree_root);
2026 BUG_ON(ret);
e556ce2c
YZ
2027
2028 if (sb->s_flags & MS_RDONLY) {
2029 ret = btrfs_commit_super(tree_root);
2030 BUG_ON(ret);
2031 }
e02119d5 2032 }
1a40e23b 2033
76dda93c
YZ
2034 ret = btrfs_find_orphan_roots(tree_root);
2035 BUG_ON(ret);
2036
7c2ca468 2037 if (!(sb->s_flags & MS_RDONLY)) {
d68fc57b
YZ
2038 ret = btrfs_cleanup_fs_roots(fs_info);
2039 BUG_ON(ret);
2040
5d4f98a2 2041 ret = btrfs_recover_relocation(tree_root);
d7ce5843
MX
2042 if (ret < 0) {
2043 printk(KERN_WARNING
2044 "btrfs: failed to recover relocation\n");
2045 err = -EINVAL;
2046 goto fail_trans_kthread;
2047 }
7c2ca468 2048 }
1a40e23b 2049
3de4586c
CM
2050 location.objectid = BTRFS_FS_TREE_OBJECTID;
2051 location.type = BTRFS_ROOT_ITEM_KEY;
2052 location.offset = (u64)-1;
2053
3de4586c
CM
2054 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2055 if (!fs_info->fs_root)
7c2ca468 2056 goto fail_trans_kthread;
3140c9a3
DC
2057 if (IS_ERR(fs_info->fs_root)) {
2058 err = PTR_ERR(fs_info->fs_root);
2059 goto fail_trans_kthread;
2060 }
c289811c 2061
e3acc2a6
JB
2062 if (!(sb->s_flags & MS_RDONLY)) {
2063 down_read(&fs_info->cleanup_work_sem);
66b4ffd1
JB
2064 err = btrfs_orphan_cleanup(fs_info->fs_root);
2065 if (!err)
2066 err = btrfs_orphan_cleanup(fs_info->tree_root);
e3acc2a6 2067 up_read(&fs_info->cleanup_work_sem);
66b4ffd1
JB
2068 if (err) {
2069 close_ctree(tree_root);
2070 return ERR_PTR(err);
2071 }
e3acc2a6
JB
2072 }
2073
0f7d52f4 2074 return tree_root;
39279cc3 2075
7c2ca468
CM
2076fail_trans_kthread:
2077 kthread_stop(fs_info->transaction_kthread);
3f157a2f 2078fail_cleaner:
a74a4b97 2079 kthread_stop(fs_info->cleaner_kthread);
7c2ca468
CM
2080
2081 /*
2082 * make sure we're done with the btree inode before we stop our
2083 * kthreads
2084 */
2085 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2086 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2087
1b1d1f66
JB
2088fail_block_groups:
2089 btrfs_free_block_groups(fs_info);
d20f7043 2090 free_extent_buffer(csum_root->node);
5d4f98a2
YZ
2091 free_extent_buffer(csum_root->commit_root);
2092fail_dev_root:
2093 free_extent_buffer(dev_root->node);
2094 free_extent_buffer(dev_root->commit_root);
0b86a832
CM
2095fail_extent_root:
2096 free_extent_buffer(extent_root->node);
5d4f98a2 2097 free_extent_buffer(extent_root->commit_root);
39279cc3 2098fail_tree_root:
5f39d397 2099 free_extent_buffer(tree_root->node);
5d4f98a2 2100 free_extent_buffer(tree_root->commit_root);
2b82032c
YZ
2101fail_chunk_root:
2102 free_extent_buffer(chunk_root->node);
5d4f98a2 2103 free_extent_buffer(chunk_root->commit_root);
39279cc3 2104fail_sb_buffer:
61d92c32 2105 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2106 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2107 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM
2108 btrfs_stop_workers(&fs_info->workers);
2109 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2110 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2111 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2112 btrfs_stop_workers(&fs_info->endio_write_workers);
0cb59c99 2113 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1cc127b5 2114 btrfs_stop_workers(&fs_info->submit_workers);
16cdcec7
MX
2115 btrfs_stop_workers(&fs_info->delayed_workers);
2116fail_alloc:
2117 kfree(fs_info->delayed_root);
4543df7e 2118fail_iput:
7c2ca468 2119 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
4543df7e 2120 iput(fs_info->btree_inode);
7e662854 2121
dfe25020 2122 btrfs_close_devices(fs_info->fs_devices);
84eed90f 2123 btrfs_mapping_tree_free(&fs_info->mapping_tree);
ad081f14 2124fail_bdi:
7e662854 2125 bdi_destroy(&fs_info->bdi);
76dda93c
YZ
2126fail_srcu:
2127 cleanup_srcu_struct(&fs_info->subvol_srcu);
7e662854 2128fail:
39279cc3
CM
2129 kfree(extent_root);
2130 kfree(tree_root);
2131 kfree(fs_info);
83afeac4
JM
2132 kfree(chunk_root);
2133 kfree(dev_root);
d20f7043 2134 kfree(csum_root);
39279cc3 2135 return ERR_PTR(err);
eb60ceac
CM
2136}
2137
f2984462
CM
2138static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2139{
2140 char b[BDEVNAME_SIZE];
2141
2142 if (uptodate) {
2143 set_buffer_uptodate(bh);
2144 } else {
7a36ddec 2145 printk_ratelimited(KERN_WARNING "lost page write due to "
f2984462
CM
2146 "I/O error on %s\n",
2147 bdevname(bh->b_bdev, b));
1259ab75
CM
2148 /* note, we dont' set_buffer_write_io_error because we have
2149 * our own ways of dealing with the IO errors
2150 */
f2984462
CM
2151 clear_buffer_uptodate(bh);
2152 }
2153 unlock_buffer(bh);
2154 put_bh(bh);
2155}
2156
a512bbf8
YZ
2157struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2158{
2159 struct buffer_head *bh;
2160 struct buffer_head *latest = NULL;
2161 struct btrfs_super_block *super;
2162 int i;
2163 u64 transid = 0;
2164 u64 bytenr;
2165
2166 /* we would like to check all the supers, but that would make
2167 * a btrfs mount succeed after a mkfs from a different FS.
2168 * So, we need to add a special mount option to scan for
2169 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2170 */
2171 for (i = 0; i < 1; i++) {
2172 bytenr = btrfs_sb_offset(i);
2173 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2174 break;
2175 bh = __bread(bdev, bytenr / 4096, 4096);
2176 if (!bh)
2177 continue;
2178
2179 super = (struct btrfs_super_block *)bh->b_data;
2180 if (btrfs_super_bytenr(super) != bytenr ||
2181 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2182 sizeof(super->magic))) {
2183 brelse(bh);
2184 continue;
2185 }
2186
2187 if (!latest || btrfs_super_generation(super) > transid) {
2188 brelse(latest);
2189 latest = bh;
2190 transid = btrfs_super_generation(super);
2191 } else {
2192 brelse(bh);
2193 }
2194 }
2195 return latest;
2196}
2197
4eedeb75
HH
2198/*
2199 * this should be called twice, once with wait == 0 and
2200 * once with wait == 1. When wait == 0 is done, all the buffer heads
2201 * we write are pinned.
2202 *
2203 * They are released when wait == 1 is done.
2204 * max_mirrors must be the same for both runs, and it indicates how
2205 * many supers on this one device should be written.
2206 *
2207 * max_mirrors == 0 means to write them all.
2208 */
a512bbf8
YZ
2209static int write_dev_supers(struct btrfs_device *device,
2210 struct btrfs_super_block *sb,
2211 int do_barriers, int wait, int max_mirrors)
2212{
2213 struct buffer_head *bh;
2214 int i;
2215 int ret;
2216 int errors = 0;
2217 u32 crc;
2218 u64 bytenr;
2219 int last_barrier = 0;
2220
2221 if (max_mirrors == 0)
2222 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2223
2224 /* make sure only the last submit_bh does a barrier */
2225 if (do_barriers) {
2226 for (i = 0; i < max_mirrors; i++) {
2227 bytenr = btrfs_sb_offset(i);
2228 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2229 device->total_bytes)
2230 break;
2231 last_barrier = i;
2232 }
2233 }
2234
2235 for (i = 0; i < max_mirrors; i++) {
2236 bytenr = btrfs_sb_offset(i);
2237 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2238 break;
2239
2240 if (wait) {
2241 bh = __find_get_block(device->bdev, bytenr / 4096,
2242 BTRFS_SUPER_INFO_SIZE);
2243 BUG_ON(!bh);
a512bbf8 2244 wait_on_buffer(bh);
4eedeb75
HH
2245 if (!buffer_uptodate(bh))
2246 errors++;
2247
2248 /* drop our reference */
2249 brelse(bh);
2250
2251 /* drop the reference from the wait == 0 run */
2252 brelse(bh);
2253 continue;
a512bbf8
YZ
2254 } else {
2255 btrfs_set_super_bytenr(sb, bytenr);
2256
2257 crc = ~(u32)0;
2258 crc = btrfs_csum_data(NULL, (char *)sb +
2259 BTRFS_CSUM_SIZE, crc,
2260 BTRFS_SUPER_INFO_SIZE -
2261 BTRFS_CSUM_SIZE);
2262 btrfs_csum_final(crc, sb->csum);
2263
4eedeb75
HH
2264 /*
2265 * one reference for us, and we leave it for the
2266 * caller
2267 */
a512bbf8
YZ
2268 bh = __getblk(device->bdev, bytenr / 4096,
2269 BTRFS_SUPER_INFO_SIZE);
2270 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2271
4eedeb75 2272 /* one reference for submit_bh */
a512bbf8 2273 get_bh(bh);
4eedeb75
HH
2274
2275 set_buffer_uptodate(bh);
a512bbf8
YZ
2276 lock_buffer(bh);
2277 bh->b_end_io = btrfs_end_buffer_write_sync;
2278 }
2279
c3b9a62c
CH
2280 if (i == last_barrier && do_barriers)
2281 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2282 else
ffbd517d 2283 ret = submit_bh(WRITE_SYNC, bh);
a512bbf8 2284
4eedeb75 2285 if (ret)
a512bbf8 2286 errors++;
a512bbf8
YZ
2287 }
2288 return errors < i ? 0 : -1;
2289}
2290
2291int write_all_supers(struct btrfs_root *root, int max_mirrors)
f2984462 2292{
e5e9a520 2293 struct list_head *head;
f2984462 2294 struct btrfs_device *dev;
a061fc8d 2295 struct btrfs_super_block *sb;
f2984462 2296 struct btrfs_dev_item *dev_item;
f2984462
CM
2297 int ret;
2298 int do_barriers;
a236aed1
CM
2299 int max_errors;
2300 int total_errors = 0;
a061fc8d 2301 u64 flags;
f2984462 2302
a236aed1 2303 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
f2984462
CM
2304 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2305
a061fc8d
CM
2306 sb = &root->fs_info->super_for_commit;
2307 dev_item = &sb->dev_item;
e5e9a520 2308
1f78160c 2309 rcu_read_lock();
e5e9a520 2310 head = &root->fs_info->fs_devices->devices;
1f78160c 2311 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM
2312 if (!dev->bdev) {
2313 total_errors++;
2314 continue;
2315 }
2b82032c 2316 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM
2317 continue;
2318
2b82032c 2319 btrfs_set_stack_device_generation(dev_item, 0);
a061fc8d
CM
2320 btrfs_set_stack_device_type(dev_item, dev->type);
2321 btrfs_set_stack_device_id(dev_item, dev->devid);
2322 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2323 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2324 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2325 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2326 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2327 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2b82032c 2328 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
a512bbf8 2329
a061fc8d
CM
2330 flags = btrfs_super_flags(sb);
2331 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2332
a512bbf8 2333 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
a236aed1
CM
2334 if (ret)
2335 total_errors++;
f2984462 2336 }
a236aed1 2337 if (total_errors > max_errors) {
d397712b
CM
2338 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2339 total_errors);
a236aed1
CM
2340 BUG();
2341 }
f2984462 2342
a512bbf8 2343 total_errors = 0;
1f78160c 2344 list_for_each_entry_rcu(dev, head, dev_list) {
dfe25020
CM
2345 if (!dev->bdev)
2346 continue;
2b82032c 2347 if (!dev->in_fs_metadata || !dev->writeable)
dfe25020
CM
2348 continue;
2349
a512bbf8
YZ
2350 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2351 if (ret)
2352 total_errors++;
f2984462 2353 }
1f78160c 2354 rcu_read_unlock();
a236aed1 2355 if (total_errors > max_errors) {
d397712b
CM
2356 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2357 total_errors);
a236aed1
CM
2358 BUG();
2359 }
f2984462
CM
2360 return 0;
2361}
2362
a512bbf8
YZ
2363int write_ctree_super(struct btrfs_trans_handle *trans,
2364 struct btrfs_root *root, int max_mirrors)
eb60ceac 2365{
e66f709b 2366 int ret;
5f39d397 2367
a512bbf8 2368 ret = write_all_supers(root, max_mirrors);
5f39d397 2369 return ret;
cfaa7295
CM
2370}
2371
5eda7b5e 2372int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2619ba1f 2373{
4df27c4d 2374 spin_lock(&fs_info->fs_roots_radix_lock);
2619ba1f
CM
2375 radix_tree_delete(&fs_info->fs_roots_radix,
2376 (unsigned long)root->root_key.objectid);
4df27c4d 2377 spin_unlock(&fs_info->fs_roots_radix_lock);
76dda93c
YZ
2378
2379 if (btrfs_root_refs(&root->root_item) == 0)
2380 synchronize_srcu(&fs_info->subvol_srcu);
2381
581bb050
LZ
2382 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2383 __btrfs_remove_free_space_cache(root->free_ino_ctl);
4df27c4d
YZ
2384 free_fs_root(root);
2385 return 0;
2386}
2387
2388static void free_fs_root(struct btrfs_root *root)
2389{
82d5902d 2390 iput(root->cache_inode);
4df27c4d 2391 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3394e160
CM
2392 if (root->anon_super.s_dev) {
2393 down_write(&root->anon_super.s_umount);
2394 kill_anon_super(&root->anon_super);
2395 }
4df27c4d
YZ
2396 free_extent_buffer(root->node);
2397 free_extent_buffer(root->commit_root);
581bb050
LZ
2398 kfree(root->free_ino_ctl);
2399 kfree(root->free_ino_pinned);
d397712b 2400 kfree(root->name);
2619ba1f 2401 kfree(root);
2619ba1f
CM
2402}
2403
35b7e476 2404static int del_fs_roots(struct btrfs_fs_info *fs_info)
0f7d52f4
CM
2405{
2406 int ret;
2407 struct btrfs_root *gang[8];
2408 int i;
2409
76dda93c
YZ
2410 while (!list_empty(&fs_info->dead_roots)) {
2411 gang[0] = list_entry(fs_info->dead_roots.next,
2412 struct btrfs_root, root_list);
2413 list_del(&gang[0]->root_list);
2414
2415 if (gang[0]->in_radix) {
2416 btrfs_free_fs_root(fs_info, gang[0]);
2417 } else {
2418 free_extent_buffer(gang[0]->node);
2419 free_extent_buffer(gang[0]->commit_root);
2420 kfree(gang[0]);
2421 }
2422 }
2423
d397712b 2424 while (1) {
0f7d52f4
CM
2425 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2426 (void **)gang, 0,
2427 ARRAY_SIZE(gang));
2428 if (!ret)
2429 break;
2619ba1f 2430 for (i = 0; i < ret; i++)
5eda7b5e 2431 btrfs_free_fs_root(fs_info, gang[i]);
0f7d52f4
CM
2432 }
2433 return 0;
2434}
b4100d64 2435
c146afad 2436int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
cfaa7295 2437{
c146afad
YZ
2438 u64 root_objectid = 0;
2439 struct btrfs_root *gang[8];
2440 int i;
3768f368 2441 int ret;
e089f05c 2442
c146afad
YZ
2443 while (1) {
2444 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2445 (void **)gang, root_objectid,
2446 ARRAY_SIZE(gang));
2447 if (!ret)
2448 break;
5d4f98a2
YZ
2449
2450 root_objectid = gang[ret - 1]->root_key.objectid + 1;
c146afad 2451 for (i = 0; i < ret; i++) {
66b4ffd1
JB
2452 int err;
2453
c146afad 2454 root_objectid = gang[i]->root_key.objectid;
66b4ffd1
JB
2455 err = btrfs_orphan_cleanup(gang[i]);
2456 if (err)
2457 return err;
c146afad
YZ
2458 }
2459 root_objectid++;
2460 }
2461 return 0;
2462}
a2135011 2463
c146afad
YZ
2464int btrfs_commit_super(struct btrfs_root *root)
2465{
2466 struct btrfs_trans_handle *trans;
2467 int ret;
a74a4b97 2468
c146afad 2469 mutex_lock(&root->fs_info->cleaner_mutex);
24bbcf04 2470 btrfs_run_delayed_iputs(root);
a74a4b97 2471 btrfs_clean_old_snapshots(root);
c146afad 2472 mutex_unlock(&root->fs_info->cleaner_mutex);
c71bf099
YZ
2473
2474 /* wait until ongoing cleanup work done */
2475 down_write(&root->fs_info->cleanup_work_sem);
2476 up_write(&root->fs_info->cleanup_work_sem);
2477
a22285a6 2478 trans = btrfs_join_transaction(root, 1);
3612b495
TI
2479 if (IS_ERR(trans))
2480 return PTR_ERR(trans);
54aa1f4d 2481 ret = btrfs_commit_transaction(trans, root);
c146afad
YZ
2482 BUG_ON(ret);
2483 /* run commit again to drop the original snapshot */
a22285a6 2484 trans = btrfs_join_transaction(root, 1);
3612b495
TI
2485 if (IS_ERR(trans))
2486 return PTR_ERR(trans);
79154b1b
CM
2487 btrfs_commit_transaction(trans, root);
2488 ret = btrfs_write_and_wait_transaction(NULL, root);
3768f368 2489 BUG_ON(ret);
d6bfde87 2490
a512bbf8 2491 ret = write_ctree_super(NULL, root, 0);
c146afad
YZ
2492 return ret;
2493}
2494
2495int close_ctree(struct btrfs_root *root)
2496{
2497 struct btrfs_fs_info *fs_info = root->fs_info;
2498 int ret;
2499
2500 fs_info->closing = 1;
2501 smp_mb();
2502
a2de733c 2503 btrfs_scrub_cancel(root);
0af3d00b 2504 btrfs_put_block_group_cache(fs_info);
acce952b 2505
2506 /*
2507 * Here come 2 situations when btrfs is broken to flip readonly:
2508 *
2509 * 1. when btrfs flips readonly somewhere else before
2510 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2511 * and btrfs will skip to write sb directly to keep
2512 * ERROR state on disk.
2513 *
2514 * 2. when btrfs flips readonly just in btrfs_commit_super,
ae0e47f0 2515 * and in such case, btrfs cannot write sb via btrfs_commit_super,
acce952b 2516 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2517 * btrfs will cleanup all FS resources first and write sb then.
2518 */
c146afad 2519 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
acce952b 2520 ret = btrfs_commit_super(root);
2521 if (ret)
2522 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2523 }
2524
2525 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2526 ret = btrfs_error_commit_super(root);
d397712b
CM
2527 if (ret)
2528 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
c146afad 2529 }
0f7d52f4 2530
8929ecfa
YZ
2531 kthread_stop(root->fs_info->transaction_kthread);
2532 kthread_stop(root->fs_info->cleaner_kthread);
2533
f25784b3
YZ
2534 fs_info->closing = 2;
2535 smp_mb();
2536
b0c68f8b 2537 if (fs_info->delalloc_bytes) {
d397712b 2538 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
21380931 2539 (unsigned long long)fs_info->delalloc_bytes);
b0c68f8b 2540 }
31153d81 2541 if (fs_info->total_ref_cache_size) {
d397712b
CM
2542 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2543 (unsigned long long)fs_info->total_ref_cache_size);
31153d81 2544 }
bcc63abb 2545
5d4f98a2
YZ
2546 free_extent_buffer(fs_info->extent_root->node);
2547 free_extent_buffer(fs_info->extent_root->commit_root);
2548 free_extent_buffer(fs_info->tree_root->node);
2549 free_extent_buffer(fs_info->tree_root->commit_root);
2550 free_extent_buffer(root->fs_info->chunk_root->node);
2551 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2552 free_extent_buffer(root->fs_info->dev_root->node);
2553 free_extent_buffer(root->fs_info->dev_root->commit_root);
2554 free_extent_buffer(root->fs_info->csum_root->node);
2555 free_extent_buffer(root->fs_info->csum_root->commit_root);
d20f7043 2556
9078a3e1 2557 btrfs_free_block_groups(root->fs_info);
d10c5f31 2558
c146afad 2559 del_fs_roots(fs_info);
d10c5f31 2560
c146afad 2561 iput(fs_info->btree_inode);
16cdcec7 2562 kfree(fs_info->delayed_root);
9ad6b7bc 2563
61d92c32 2564 btrfs_stop_workers(&fs_info->generic_worker);
247e743c 2565 btrfs_stop_workers(&fs_info->fixup_workers);
771ed689 2566 btrfs_stop_workers(&fs_info->delalloc_workers);
8b712842
CM
2567 btrfs_stop_workers(&fs_info->workers);
2568 btrfs_stop_workers(&fs_info->endio_workers);
d20f7043 2569 btrfs_stop_workers(&fs_info->endio_meta_workers);
cad321ad 2570 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
e6dcd2dc 2571 btrfs_stop_workers(&fs_info->endio_write_workers);
0cb59c99 2572 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1cc127b5 2573 btrfs_stop_workers(&fs_info->submit_workers);
16cdcec7 2574 btrfs_stop_workers(&fs_info->delayed_workers);
d6bfde87 2575
dfe25020 2576 btrfs_close_devices(fs_info->fs_devices);
0b86a832 2577 btrfs_mapping_tree_free(&fs_info->mapping_tree);
b248a415 2578
04160088 2579 bdi_destroy(&fs_info->bdi);
76dda93c 2580 cleanup_srcu_struct(&fs_info->subvol_srcu);
0b86a832 2581
0f7d52f4 2582 kfree(fs_info->extent_root);
0f7d52f4 2583 kfree(fs_info->tree_root);
0b86a832
CM
2584 kfree(fs_info->chunk_root);
2585 kfree(fs_info->dev_root);
d20f7043 2586 kfree(fs_info->csum_root);
83a4d548
LZ
2587 kfree(fs_info);
2588
eb60ceac
CM
2589 return 0;
2590}
2591
1259ab75 2592int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
5f39d397 2593{
1259ab75 2594 int ret;
810191ff 2595 struct inode *btree_inode = buf->first_page->mapping->host;
1259ab75 2596
2ac55d41
JB
2597 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2598 NULL);
1259ab75
CM
2599 if (!ret)
2600 return ret;
2601
2602 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2603 parent_transid);
2604 return !ret;
5f39d397
CM
2605}
2606
2607int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
ccd467d6 2608{
810191ff 2609 struct inode *btree_inode = buf->first_page->mapping->host;
d1310b2e 2610 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
5f39d397
CM
2611 buf);
2612}
6702ed49 2613
5f39d397
CM
2614void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2615{
810191ff 2616 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
5f39d397
CM
2617 u64 transid = btrfs_header_generation(buf);
2618 struct inode *btree_inode = root->fs_info->btree_inode;
b9473439 2619 int was_dirty;
b4ce94de 2620
b9447ef8 2621 btrfs_assert_tree_locked(buf);
ccd467d6 2622 if (transid != root->fs_info->generation) {
d397712b
CM
2623 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2624 "found %llu running %llu\n",
db94535d 2625 (unsigned long long)buf->start,
d397712b
CM
2626 (unsigned long long)transid,
2627 (unsigned long long)root->fs_info->generation);
ccd467d6
CM
2628 WARN_ON(1);
2629 }
b9473439
CM
2630 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2631 buf);
2632 if (!was_dirty) {
2633 spin_lock(&root->fs_info->delalloc_lock);
2634 root->fs_info->dirty_metadata_bytes += buf->len;
2635 spin_unlock(&root->fs_info->delalloc_lock);
2636 }
eb60ceac
CM
2637}
2638
d3c2fdcf 2639void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
16cdcec7
MX
2640{
2641 /*
2642 * looks as though older kernels can get into trouble with
2643 * this code, they end up stuck in balance_dirty_pages forever
2644 */
2645 u64 num_dirty;
2646 unsigned long thresh = 32 * 1024 * 1024;
2647
2648 if (current->flags & PF_MEMALLOC)
2649 return;
2650
2651 btrfs_balance_delayed_items(root);
2652
2653 num_dirty = root->fs_info->dirty_metadata_bytes;
2654
2655 if (num_dirty > thresh) {
2656 balance_dirty_pages_ratelimited_nr(
2657 root->fs_info->btree_inode->i_mapping, 1);
2658 }
2659 return;
2660}
2661
2662void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
35b7e476 2663{
188de649
CM
2664 /*
2665 * looks as though older kernels can get into trouble with
2666 * this code, they end up stuck in balance_dirty_pages forever
2667 */
d6bfde87 2668 u64 num_dirty;
771ed689 2669 unsigned long thresh = 32 * 1024 * 1024;
d6bfde87 2670
6933c02e 2671 if (current->flags & PF_MEMALLOC)
d6bfde87
CM
2672 return;
2673
585ad2c3
CM
2674 num_dirty = root->fs_info->dirty_metadata_bytes;
2675
d6bfde87
CM
2676 if (num_dirty > thresh) {
2677 balance_dirty_pages_ratelimited_nr(
d7fc640e 2678 root->fs_info->btree_inode->i_mapping, 1);
d6bfde87 2679 }
188de649 2680 return;
35b7e476 2681}
6b80053d 2682
ca7a79ad 2683int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
6b80053d 2684{
810191ff 2685 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
ce9adaa5 2686 int ret;
ca7a79ad 2687 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
d397712b 2688 if (ret == 0)
b4ce94de 2689 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
ce9adaa5 2690 return ret;
6b80053d 2691}
0da5468f 2692
4bef0848
CM
2693int btree_lock_page_hook(struct page *page)
2694{
2695 struct inode *inode = page->mapping->host;
b9473439 2696 struct btrfs_root *root = BTRFS_I(inode)->root;
4bef0848
CM
2697 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2698 struct extent_buffer *eb;
2699 unsigned long len;
2700 u64 bytenr = page_offset(page);
2701
2702 if (page->private == EXTENT_PAGE_PRIVATE)
2703 goto out;
2704
2705 len = page->private >> 2;
f09d1f60 2706 eb = find_extent_buffer(io_tree, bytenr, len);
4bef0848
CM
2707 if (!eb)
2708 goto out;
2709
2710 btrfs_tree_lock(eb);
4bef0848 2711 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
b9473439
CM
2712
2713 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2714 spin_lock(&root->fs_info->delalloc_lock);
2715 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2716 root->fs_info->dirty_metadata_bytes -= eb->len;
2717 else
2718 WARN_ON(1);
2719 spin_unlock(&root->fs_info->delalloc_lock);
2720 }
2721
4bef0848
CM
2722 btrfs_tree_unlock(eb);
2723 free_extent_buffer(eb);
2724out:
2725 lock_page(page);
2726 return 0;
2727}
2728
acce952b 2729static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2730 int read_only)
2731{
2732 if (read_only)
2733 return;
2734
2735 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2736 printk(KERN_WARNING "warning: mount fs with errors, "
2737 "running btrfsck is recommended\n");
2738}
2739
2740int btrfs_error_commit_super(struct btrfs_root *root)
2741{
2742 int ret;
2743
2744 mutex_lock(&root->fs_info->cleaner_mutex);
2745 btrfs_run_delayed_iputs(root);
2746 mutex_unlock(&root->fs_info->cleaner_mutex);
2747
2748 down_write(&root->fs_info->cleanup_work_sem);
2749 up_write(&root->fs_info->cleanup_work_sem);
2750
2751 /* cleanup FS via transaction */
2752 btrfs_cleanup_transaction(root);
2753
2754 ret = write_ctree_super(NULL, root, 0);
2755
2756 return ret;
2757}
2758
2759static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2760{
2761 struct btrfs_inode *btrfs_inode;
2762 struct list_head splice;
2763
2764 INIT_LIST_HEAD(&splice);
2765
2766 mutex_lock(&root->fs_info->ordered_operations_mutex);
2767 spin_lock(&root->fs_info->ordered_extent_lock);
2768
2769 list_splice_init(&root->fs_info->ordered_operations, &splice);
2770 while (!list_empty(&splice)) {
2771 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2772 ordered_operations);
2773
2774 list_del_init(&btrfs_inode->ordered_operations);
2775
2776 btrfs_invalidate_inodes(btrfs_inode->root);
2777 }
2778
2779 spin_unlock(&root->fs_info->ordered_extent_lock);
2780 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2781
2782 return 0;
2783}
2784
2785static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2786{
2787 struct list_head splice;
2788 struct btrfs_ordered_extent *ordered;
2789 struct inode *inode;
2790
2791 INIT_LIST_HEAD(&splice);
2792
2793 spin_lock(&root->fs_info->ordered_extent_lock);
2794
2795 list_splice_init(&root->fs_info->ordered_extents, &splice);
2796 while (!list_empty(&splice)) {
2797 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2798 root_extent_list);
2799
2800 list_del_init(&ordered->root_extent_list);
2801 atomic_inc(&ordered->refs);
2802
2803 /* the inode may be getting freed (in sys_unlink path). */
2804 inode = igrab(ordered->inode);
2805
2806 spin_unlock(&root->fs_info->ordered_extent_lock);
2807 if (inode)
2808 iput(inode);
2809
2810 atomic_set(&ordered->refs, 1);
2811 btrfs_put_ordered_extent(ordered);
2812
2813 spin_lock(&root->fs_info->ordered_extent_lock);
2814 }
2815
2816 spin_unlock(&root->fs_info->ordered_extent_lock);
2817
2818 return 0;
2819}
2820
2821static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2822 struct btrfs_root *root)
2823{
2824 struct rb_node *node;
2825 struct btrfs_delayed_ref_root *delayed_refs;
2826 struct btrfs_delayed_ref_node *ref;
2827 int ret = 0;
2828
2829 delayed_refs = &trans->delayed_refs;
2830
2831 spin_lock(&delayed_refs->lock);
2832 if (delayed_refs->num_entries == 0) {
cfece4db 2833 spin_unlock(&delayed_refs->lock);
acce952b 2834 printk(KERN_INFO "delayed_refs has NO entry\n");
2835 return ret;
2836 }
2837
2838 node = rb_first(&delayed_refs->root);
2839 while (node) {
2840 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2841 node = rb_next(node);
2842
2843 ref->in_tree = 0;
2844 rb_erase(&ref->rb_node, &delayed_refs->root);
2845 delayed_refs->num_entries--;
2846
2847 atomic_set(&ref->refs, 1);
2848 if (btrfs_delayed_ref_is_head(ref)) {
2849 struct btrfs_delayed_ref_head *head;
2850
2851 head = btrfs_delayed_node_to_head(ref);
2852 mutex_lock(&head->mutex);
2853 kfree(head->extent_op);
2854 delayed_refs->num_heads--;
2855 if (list_empty(&head->cluster))
2856 delayed_refs->num_heads_ready--;
2857 list_del_init(&head->cluster);
2858 mutex_unlock(&head->mutex);
2859 }
2860
2861 spin_unlock(&delayed_refs->lock);
2862 btrfs_put_delayed_ref(ref);
2863
2864 cond_resched();
2865 spin_lock(&delayed_refs->lock);
2866 }
2867
2868 spin_unlock(&delayed_refs->lock);
2869
2870 return ret;
2871}
2872
2873static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2874{
2875 struct btrfs_pending_snapshot *snapshot;
2876 struct list_head splice;
2877
2878 INIT_LIST_HEAD(&splice);
2879
2880 list_splice_init(&t->pending_snapshots, &splice);
2881
2882 while (!list_empty(&splice)) {
2883 snapshot = list_entry(splice.next,
2884 struct btrfs_pending_snapshot,
2885 list);
2886
2887 list_del_init(&snapshot->list);
2888
2889 kfree(snapshot);
2890 }
2891
2892 return 0;
2893}
2894
2895static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2896{
2897 struct btrfs_inode *btrfs_inode;
2898 struct list_head splice;
2899
2900 INIT_LIST_HEAD(&splice);
2901
2902 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2903
2904 spin_lock(&root->fs_info->delalloc_lock);
2905
2906 while (!list_empty(&splice)) {
2907 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2908 delalloc_inodes);
2909
2910 list_del_init(&btrfs_inode->delalloc_inodes);
2911
2912 btrfs_invalidate_inodes(btrfs_inode->root);
2913 }
2914
2915 spin_unlock(&root->fs_info->delalloc_lock);
2916
2917 return 0;
2918}
2919
2920static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2921 struct extent_io_tree *dirty_pages,
2922 int mark)
2923{
2924 int ret;
2925 struct page *page;
2926 struct inode *btree_inode = root->fs_info->btree_inode;
2927 struct extent_buffer *eb;
2928 u64 start = 0;
2929 u64 end;
2930 u64 offset;
2931 unsigned long index;
2932
2933 while (1) {
2934 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2935 mark);
2936 if (ret)
2937 break;
2938
2939 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2940 while (start <= end) {
2941 index = start >> PAGE_CACHE_SHIFT;
2942 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2943 page = find_get_page(btree_inode->i_mapping, index);
2944 if (!page)
2945 continue;
2946 offset = page_offset(page);
2947
2948 spin_lock(&dirty_pages->buffer_lock);
2949 eb = radix_tree_lookup(
2950 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2951 offset >> PAGE_CACHE_SHIFT);
2952 spin_unlock(&dirty_pages->buffer_lock);
2953 if (eb) {
2954 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
2955 &eb->bflags);
2956 atomic_set(&eb->refs, 1);
2957 }
2958 if (PageWriteback(page))
2959 end_page_writeback(page);
2960
2961 lock_page(page);
2962 if (PageDirty(page)) {
2963 clear_page_dirty_for_io(page);
2964 spin_lock_irq(&page->mapping->tree_lock);
2965 radix_tree_tag_clear(&page->mapping->page_tree,
2966 page_index(page),
2967 PAGECACHE_TAG_DIRTY);
2968 spin_unlock_irq(&page->mapping->tree_lock);
2969 }
2970
2971 page->mapping->a_ops->invalidatepage(page, 0);
2972 unlock_page(page);
2973 }
2974 }
2975
2976 return ret;
2977}
2978
2979static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
2980 struct extent_io_tree *pinned_extents)
2981{
2982 struct extent_io_tree *unpin;
2983 u64 start;
2984 u64 end;
2985 int ret;
2986
2987 unpin = pinned_extents;
2988 while (1) {
2989 ret = find_first_extent_bit(unpin, 0, &start, &end,
2990 EXTENT_DIRTY);
2991 if (ret)
2992 break;
2993
2994 /* opt_discard */
5378e607
LD
2995 if (btrfs_test_opt(root, DISCARD))
2996 ret = btrfs_error_discard_extent(root, start,
2997 end + 1 - start,
2998 NULL);
acce952b 2999
3000 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3001 btrfs_error_unpin_extent_range(root, start, end);
3002 cond_resched();
3003 }
3004
3005 return 0;
3006}
3007
3008static int btrfs_cleanup_transaction(struct btrfs_root *root)
3009{
3010 struct btrfs_transaction *t;
3011 LIST_HEAD(list);
3012
3013 WARN_ON(1);
3014
3015 mutex_lock(&root->fs_info->trans_mutex);
3016 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3017
3018 list_splice_init(&root->fs_info->trans_list, &list);
3019 while (!list_empty(&list)) {
3020 t = list_entry(list.next, struct btrfs_transaction, list);
3021 if (!t)
3022 break;
3023
3024 btrfs_destroy_ordered_operations(root);
3025
3026 btrfs_destroy_ordered_extents(root);
3027
3028 btrfs_destroy_delayed_refs(t, root);
3029
3030 btrfs_block_rsv_release(root,
3031 &root->fs_info->trans_block_rsv,
3032 t->dirty_pages.dirty_bytes);
3033
3034 /* FIXME: cleanup wait for commit */
3035 t->in_commit = 1;
3036 t->blocked = 1;
3037 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3038 wake_up(&root->fs_info->transaction_blocked_wait);
3039
3040 t->blocked = 0;
3041 if (waitqueue_active(&root->fs_info->transaction_wait))
3042 wake_up(&root->fs_info->transaction_wait);
3043 mutex_unlock(&root->fs_info->trans_mutex);
3044
3045 mutex_lock(&root->fs_info->trans_mutex);
3046 t->commit_done = 1;
3047 if (waitqueue_active(&t->commit_wait))
3048 wake_up(&t->commit_wait);
3049 mutex_unlock(&root->fs_info->trans_mutex);
3050
3051 mutex_lock(&root->fs_info->trans_mutex);
3052
3053 btrfs_destroy_pending_snapshots(t);
3054
3055 btrfs_destroy_delalloc_inodes(root);
3056
3057 spin_lock(&root->fs_info->new_trans_lock);
3058 root->fs_info->running_transaction = NULL;
3059 spin_unlock(&root->fs_info->new_trans_lock);
3060
3061 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3062 EXTENT_DIRTY);
3063
3064 btrfs_destroy_pinned_extent(root,
3065 root->fs_info->pinned_extents);
3066
13c5a93e 3067 atomic_set(&t->use_count, 0);
acce952b 3068 list_del_init(&t->list);
3069 memset(t, 0, sizeof(*t));
3070 kmem_cache_free(btrfs_transaction_cachep, t);
3071 }
3072
3073 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3074 mutex_unlock(&root->fs_info->trans_mutex);
3075
3076 return 0;
3077}
3078
d1310b2e 3079static struct extent_io_ops btree_extent_io_ops = {
4bef0848 3080 .write_cache_pages_lock_hook = btree_lock_page_hook,
ce9adaa5 3081 .readpage_end_io_hook = btree_readpage_end_io_hook,
0b86a832 3082 .submit_bio_hook = btree_submit_bio_hook,
239b14b3
CM
3083 /* note we're sharing with inode.c for the merge bio hook */
3084 .merge_bio_hook = btrfs_merge_bio_hook,
0da5468f 3085};