projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: fix unblocked autodefraggers when remount
[linux-2.6-block.git] / fs / btrfs / disk-io.c
... / ...
CommitLineData
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
19#include <linux/fs.h>
20#include <linux/blkdev.h>
21#include <linux/scatterlist.h>
22#include <linux/swap.h>
23#include <linux/radix-tree.h>
24#include <linux/writeback.h>
25#include <linux/buffer_head.h>
26#include <linux/workqueue.h>
27#include <linux/kthread.h>
28#include <linux/freezer.h>
29#include <linux/crc32c.h>
30#include <linux/slab.h>
31#include <linux/migrate.h>
32#include <linux/ratelimit.h>
33#include <asm/unaligned.h>
34#include "compat.h"
35#include "ctree.h"
36#include "disk-io.h"
37#include "transaction.h"
38#include "btrfs_inode.h"
39#include "volumes.h"
40#include "print-tree.h"
41#include "async-thread.h"
42#include "locking.h"
43#include "tree-log.h"
44#include "free-space-cache.h"
45#include "inode-map.h"
46#include "check-integrity.h"
47#include "rcu-string.h"
48#include "dev-replace.h"
49#include "raid56.h"
50
51#ifdef CONFIG_X86
52#include <asm/cpufeature.h>
53#endif
54
55static struct extent_io_ops btree_extent_io_ops;
56static void end_workqueue_fn(struct btrfs_work *work);
57static void free_fs_root(struct btrfs_root *root);
58static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
59 int read_only);
60static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
61 struct btrfs_root *root);
62static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
63static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
64 struct btrfs_root *root);
65static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
66static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
67static int btrfs_destroy_marked_extents(struct btrfs_root *root,
68 struct extent_io_tree *dirty_pages,
69 int mark);
70static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
71 struct extent_io_tree *pinned_extents);
72
73/*
74 * end_io_wq structs are used to do processing in task context when an IO is
75 * complete. This is used during reads to verify checksums, and it is used
76 * by writes to insert metadata for new file extents after IO is complete.
77 */
78struct end_io_wq {
79 struct bio *bio;
80 bio_end_io_t *end_io;
81 void *private;
82 struct btrfs_fs_info *info;
83 int error;
84 int metadata;
85 struct list_head list;
86 struct btrfs_work work;
87};
88
89/*
90 * async submit bios are used to offload expensive checksumming
91 * onto the worker threads. They checksum file and metadata bios
92 * just before they are sent down the IO stack.
93 */
94struct async_submit_bio {
95 struct inode *inode;
96 struct bio *bio;
97 struct list_head list;
98 extent_submit_bio_hook_t *submit_bio_start;
99 extent_submit_bio_hook_t *submit_bio_done;
100 int rw;
101 int mirror_num;
102 unsigned long bio_flags;
103 /*
104 * bio_offset is optional, can be used if the pages in the bio
105 * can't tell us where in the file the bio should go
106 */
107 u64 bio_offset;
108 struct btrfs_work work;
109 int error;
110};
111
112/*
113 * Lockdep class keys for extent_buffer->lock's in this root. For a given
114 * eb, the lockdep key is determined by the btrfs_root it belongs to and
115 * the level the eb occupies in the tree.
116 *
117 * Different roots are used for different purposes and may nest inside each
118 * other and they require separate keysets. As lockdep keys should be
119 * static, assign keysets according to the purpose of the root as indicated
120 * by btrfs_root->objectid. This ensures that all special purpose roots
121 * have separate keysets.
122 *
123 * Lock-nesting across peer nodes is always done with the immediate parent
124 * node locked thus preventing deadlock. As lockdep doesn't know this, use
125 * subclass to avoid triggering lockdep warning in such cases.
126 *
127 * The key is set by the readpage_end_io_hook after the buffer has passed
128 * csum validation but before the pages are unlocked. It is also set by
129 * btrfs_init_new_buffer on freshly allocated blocks.
130 *
131 * We also add a check to make sure the highest level of the tree is the
132 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
133 * needs update as well.
134 */
135#ifdef CONFIG_DEBUG_LOCK_ALLOC
136# if BTRFS_MAX_LEVEL != 8
137# error
138# endif
139
140static struct btrfs_lockdep_keyset {
141 u64 id; /* root objectid */
142 const char *name_stem; /* lock name stem */
143 char names[BTRFS_MAX_LEVEL + 1][20];
144 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
145} btrfs_lockdep_keysets[] = {
146 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
147 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
148 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
149 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
150 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
151 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
152 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
153 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
154 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
155 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
156 { .id = 0, .name_stem = "tree" },
157};
158
159void __init btrfs_init_lockdep(void)
160{
161 int i, j;
162
163 /* initialize lockdep class names */
164 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
165 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
166
167 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
168 snprintf(ks->names[j], sizeof(ks->names[j]),
169 "btrfs-%s-%02d", ks->name_stem, j);
170 }
171}
172
173void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
174 int level)
175{
176 struct btrfs_lockdep_keyset *ks;
177
178 BUG_ON(level >= ARRAY_SIZE(ks->keys));
179
180 /* find the matching keyset, id 0 is the default entry */
181 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
182 if (ks->id == objectid)
183 break;
184
185 lockdep_set_class_and_name(&eb->lock,
186 &ks->keys[level], ks->names[level]);
187}
188
189#endif
190
191/*
192 * extents on the btree inode are pretty simple, there's one extent
193 * that covers the entire device
194 */
195static struct extent_map *btree_get_extent(struct inode *inode,
196 struct page *page, size_t pg_offset, u64 start, u64 len,
197 int create)
198{
199 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
200 struct extent_map *em;
201 int ret;
202
203 read_lock(&em_tree->lock);
204 em = lookup_extent_mapping(em_tree, start, len);
205 if (em) {
206 em->bdev =
207 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
208 read_unlock(&em_tree->lock);
209 goto out;
210 }
211 read_unlock(&em_tree->lock);
212
213 em = alloc_extent_map();
214 if (!em) {
215 em = ERR_PTR(-ENOMEM);
216 goto out;
217 }
218 em->start = 0;
219 em->len = (u64)-1;
220 em->block_len = (u64)-1;
221 em->block_start = 0;
222 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
223
224 write_lock(&em_tree->lock);
225 ret = add_extent_mapping(em_tree, em, 0);
226 if (ret == -EEXIST) {
227 free_extent_map(em);
228 em = lookup_extent_mapping(em_tree, start, len);
229 if (!em)
230 em = ERR_PTR(-EIO);
231 } else if (ret) {
232 free_extent_map(em);
233 em = ERR_PTR(ret);
234 }
235 write_unlock(&em_tree->lock);
236
237out:
238 return em;
239}
240
241u32 btrfs_csum_data(char *data, u32 seed, size_t len)
242{
243 return crc32c(seed, data, len);
244}
245
246void btrfs_csum_final(u32 crc, char *result)
247{
248 put_unaligned_le32(~crc, result);
249}
250
251/*
252 * compute the csum for a btree block, and either verify it or write it
253 * into the csum field of the block.
254 */
255static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
256 int verify)
257{
258 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
259 char *result = NULL;
260 unsigned long len;
261 unsigned long cur_len;
262 unsigned long offset = BTRFS_CSUM_SIZE;
263 char *kaddr;
264 unsigned long map_start;
265 unsigned long map_len;
266 int err;
267 u32 crc = ~(u32)0;
268 unsigned long inline_result;
269
270 len = buf->len - offset;
271 while (len > 0) {
272 err = map_private_extent_buffer(buf, offset, 32,
273 &kaddr, &map_start, &map_len);
274 if (err)
275 return 1;
276 cur_len = min(len, map_len - (offset - map_start));
277 crc = btrfs_csum_data(kaddr + offset - map_start,
278 crc, cur_len);
279 len -= cur_len;
280 offset += cur_len;
281 }
282 if (csum_size > sizeof(inline_result)) {
283 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
284 if (!result)
285 return 1;
286 } else {
287 result = (char *)&inline_result;
288 }
289
290 btrfs_csum_final(crc, result);
291
292 if (verify) {
293 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
294 u32 val;
295 u32 found = 0;
296 memcpy(&found, result, csum_size);
297
298 read_extent_buffer(buf, &val, 0, csum_size);
299 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
300 "failed on %llu wanted %X found %X "
301 "level %d\n",
302 root->fs_info->sb->s_id,
303 (unsigned long long)buf->start, val, found,
304 btrfs_header_level(buf));
305 if (result != (char *)&inline_result)
306 kfree(result);
307 return 1;
308 }
309 } else {
310 write_extent_buffer(buf, result, 0, csum_size);
311 }
312 if (result != (char *)&inline_result)
313 kfree(result);
314 return 0;
315}
316
317/*
318 * we can't consider a given block up to date unless the transid of the
319 * block matches the transid in the parent node's pointer. This is how we
320 * detect blocks that either didn't get written at all or got written
321 * in the wrong place.
322 */
323static int verify_parent_transid(struct extent_io_tree *io_tree,
324 struct extent_buffer *eb, u64 parent_transid,
325 int atomic)
326{
327 struct extent_state *cached_state = NULL;
328 int ret;
329
330 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
331 return 0;
332
333 if (atomic)
334 return -EAGAIN;
335
336 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
337 0, &cached_state);
338 if (extent_buffer_uptodate(eb) &&
339 btrfs_header_generation(eb) == parent_transid) {
340 ret = 0;
341 goto out;
342 }
343 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
344 "found %llu\n",
345 (unsigned long long)eb->start,
346 (unsigned long long)parent_transid,
347 (unsigned long long)btrfs_header_generation(eb));
348 ret = 1;
349 clear_extent_buffer_uptodate(eb);
350out:
351 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
352 &cached_state, GFP_NOFS);
353 return ret;
354}
355
356/*
357 * helper to read a given tree block, doing retries as required when
358 * the checksums don't match and we have alternate mirrors to try.
359 */
360static int btree_read_extent_buffer_pages(struct btrfs_root *root,
361 struct extent_buffer *eb,
362 u64 start, u64 parent_transid)
363{
364 struct extent_io_tree *io_tree;
365 int failed = 0;
366 int ret;
367 int num_copies = 0;
368 int mirror_num = 0;
369 int failed_mirror = 0;
370
371 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
372 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
373 while (1) {
374 ret = read_extent_buffer_pages(io_tree, eb, start,
375 WAIT_COMPLETE,
376 btree_get_extent, mirror_num);
377 if (!ret) {
378 if (!verify_parent_transid(io_tree, eb,
379 parent_transid, 0))
380 break;
381 else
382 ret = -EIO;
383 }
384
385 /*
386 * This buffer's crc is fine, but its contents are corrupted, so
387 * there is no reason to read the other copies, they won't be
388 * any less wrong.
389 */
390 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
391 break;
392
393 num_copies = btrfs_num_copies(root->fs_info,
394 eb->start, eb->len);
395 if (num_copies == 1)
396 break;
397
398 if (!failed_mirror) {
399 failed = 1;
400 failed_mirror = eb->read_mirror;
401 }
402
403 mirror_num++;
404 if (mirror_num == failed_mirror)
405 mirror_num++;
406
407 if (mirror_num > num_copies)
408 break;
409 }
410
411 if (failed && !ret && failed_mirror)
412 repair_eb_io_failure(root, eb, failed_mirror);
413
414 return ret;
415}
416
417/*
418 * checksum a dirty tree block before IO. This has extra checks to make sure
419 * we only fill in the checksum field in the first page of a multi-page block
420 */
421
422static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
423{
424 struct extent_io_tree *tree;
425 u64 start = page_offset(page);
426 u64 found_start;
427 struct extent_buffer *eb;
428
429 tree = &BTRFS_I(page->mapping->host)->io_tree;
430
431 eb = (struct extent_buffer *)page->private;
432 if (page != eb->pages[0])
433 return 0;
434 found_start = btrfs_header_bytenr(eb);
435 if (found_start != start) {
436 WARN_ON(1);
437 return 0;
438 }
439 if (!PageUptodate(page)) {
440 WARN_ON(1);
441 return 0;
442 }
443 csum_tree_block(root, eb, 0);
444 return 0;
445}
446
447static int check_tree_block_fsid(struct btrfs_root *root,
448 struct extent_buffer *eb)
449{
450 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
451 u8 fsid[BTRFS_UUID_SIZE];
452 int ret = 1;
453
454 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
455 BTRFS_FSID_SIZE);
456 while (fs_devices) {
457 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
458 ret = 0;
459 break;
460 }
461 fs_devices = fs_devices->seed;
462 }
463 return ret;
464}
465
466#define CORRUPT(reason, eb, root, slot) \
467 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
468 "root=%llu, slot=%d\n", reason, \
469 (unsigned long long)btrfs_header_bytenr(eb), \
470 (unsigned long long)root->objectid, slot)
471
472static noinline int check_leaf(struct btrfs_root *root,
473 struct extent_buffer *leaf)
474{
475 struct btrfs_key key;
476 struct btrfs_key leaf_key;
477 u32 nritems = btrfs_header_nritems(leaf);
478 int slot;
479
480 if (nritems == 0)
481 return 0;
482
483 /* Check the 0 item */
484 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
485 BTRFS_LEAF_DATA_SIZE(root)) {
486 CORRUPT("invalid item offset size pair", leaf, root, 0);
487 return -EIO;
488 }
489
490 /*
491 * Check to make sure each items keys are in the correct order and their
492 * offsets make sense. We only have to loop through nritems-1 because
493 * we check the current slot against the next slot, which verifies the
494 * next slot's offset+size makes sense and that the current's slot
495 * offset is correct.
496 */
497 for (slot = 0; slot < nritems - 1; slot++) {
498 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
499 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
500
501 /* Make sure the keys are in the right order */
502 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
503 CORRUPT("bad key order", leaf, root, slot);
504 return -EIO;
505 }
506
507 /*
508 * Make sure the offset and ends are right, remember that the
509 * item data starts at the end of the leaf and grows towards the
510 * front.
511 */
512 if (btrfs_item_offset_nr(leaf, slot) !=
513 btrfs_item_end_nr(leaf, slot + 1)) {
514 CORRUPT("slot offset bad", leaf, root, slot);
515 return -EIO;
516 }
517
518 /*
519 * Check to make sure that we don't point outside of the leaf,
520 * just incase all the items are consistent to eachother, but
521 * all point outside of the leaf.
522 */
523 if (btrfs_item_end_nr(leaf, slot) >
524 BTRFS_LEAF_DATA_SIZE(root)) {
525 CORRUPT("slot end outside of leaf", leaf, root, slot);
526 return -EIO;
527 }
528 }
529
530 return 0;
531}
532
533struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
534 struct page *page, int max_walk)
535{
536 struct extent_buffer *eb;
537 u64 start = page_offset(page);
538 u64 target = start;
539 u64 min_start;
540
541 if (start < max_walk)
542 min_start = 0;
543 else
544 min_start = start - max_walk;
545
546 while (start >= min_start) {
547 eb = find_extent_buffer(tree, start, 0);
548 if (eb) {
549 /*
550 * we found an extent buffer and it contains our page
551 * horray!
552 */
553 if (eb->start <= target &&
554 eb->start + eb->len > target)
555 return eb;
556
557 /* we found an extent buffer that wasn't for us */
558 free_extent_buffer(eb);
559 return NULL;
560 }
561 if (start == 0)
562 break;
563 start -= PAGE_CACHE_SIZE;
564 }
565 return NULL;
566}
567
568static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
569 struct extent_state *state, int mirror)
570{
571 struct extent_io_tree *tree;
572 u64 found_start;
573 int found_level;
574 struct extent_buffer *eb;
575 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
576 int ret = 0;
577 int reads_done;
578
579 if (!page->private)
580 goto out;
581
582 tree = &BTRFS_I(page->mapping->host)->io_tree;
583 eb = (struct extent_buffer *)page->private;
584
585 /* the pending IO might have been the only thing that kept this buffer
586 * in memory. Make sure we have a ref for all this other checks
587 */
588 extent_buffer_get(eb);
589
590 reads_done = atomic_dec_and_test(&eb->io_pages);
591 if (!reads_done)
592 goto err;
593
594 eb->read_mirror = mirror;
595 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
596 ret = -EIO;
597 goto err;
598 }
599
600 found_start = btrfs_header_bytenr(eb);
601 if (found_start != eb->start) {
602 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
603 "%llu %llu\n",
604 (unsigned long long)found_start,
605 (unsigned long long)eb->start);
606 ret = -EIO;
607 goto err;
608 }
609 if (check_tree_block_fsid(root, eb)) {
610 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
611 (unsigned long long)eb->start);
612 ret = -EIO;
613 goto err;
614 }
615 found_level = btrfs_header_level(eb);
616
617 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
618 eb, found_level);
619
620 ret = csum_tree_block(root, eb, 1);
621 if (ret) {
622 ret = -EIO;
623 goto err;
624 }
625
626 /*
627 * If this is a leaf block and it is corrupt, set the corrupt bit so
628 * that we don't try and read the other copies of this block, just
629 * return -EIO.
630 */
631 if (found_level == 0 && check_leaf(root, eb)) {
632 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
633 ret = -EIO;
634 }
635
636 if (!ret)
637 set_extent_buffer_uptodate(eb);
638err:
639 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
640 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
641 btree_readahead_hook(root, eb, eb->start, ret);
642 }
643
644 if (ret) {
645 /*
646 * our io error hook is going to dec the io pages
647 * again, we have to make sure it has something
648 * to decrement
649 */
650 atomic_inc(&eb->io_pages);
651 clear_extent_buffer_uptodate(eb);
652 }
653 free_extent_buffer(eb);
654out:
655 return ret;
656}
657
658static int btree_io_failed_hook(struct page *page, int failed_mirror)
659{
660 struct extent_buffer *eb;
661 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
662
663 eb = (struct extent_buffer *)page->private;
664 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
665 eb->read_mirror = failed_mirror;
666 atomic_dec(&eb->io_pages);
667 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
668 btree_readahead_hook(root, eb, eb->start, -EIO);
669 return -EIO; /* we fixed nothing */
670}
671
672static void end_workqueue_bio(struct bio *bio, int err)
673{
674 struct end_io_wq *end_io_wq = bio->bi_private;
675 struct btrfs_fs_info *fs_info;
676
677 fs_info = end_io_wq->info;
678 end_io_wq->error = err;
679 end_io_wq->work.func = end_workqueue_fn;
680 end_io_wq->work.flags = 0;
681
682 if (bio->bi_rw & REQ_WRITE) {
683 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
684 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
685 &end_io_wq->work);
686 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
687 btrfs_queue_worker(&fs_info->endio_freespace_worker,
688 &end_io_wq->work);
689 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
690 btrfs_queue_worker(&fs_info->endio_raid56_workers,
691 &end_io_wq->work);
692 else
693 btrfs_queue_worker(&fs_info->endio_write_workers,
694 &end_io_wq->work);
695 } else {
696 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
697 btrfs_queue_worker(&fs_info->endio_raid56_workers,
698 &end_io_wq->work);
699 else if (end_io_wq->metadata)
700 btrfs_queue_worker(&fs_info->endio_meta_workers,
701 &end_io_wq->work);
702 else
703 btrfs_queue_worker(&fs_info->endio_workers,
704 &end_io_wq->work);
705 }
706}
707
708/*
709 * For the metadata arg you want
710 *
711 * 0 - if data
712 * 1 - if normal metadta
713 * 2 - if writing to the free space cache area
714 * 3 - raid parity work
715 */
716int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
717 int metadata)
718{
719 struct end_io_wq *end_io_wq;
720 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
721 if (!end_io_wq)
722 return -ENOMEM;
723
724 end_io_wq->private = bio->bi_private;
725 end_io_wq->end_io = bio->bi_end_io;
726 end_io_wq->info = info;
727 end_io_wq->error = 0;
728 end_io_wq->bio = bio;
729 end_io_wq->metadata = metadata;
730
731 bio->bi_private = end_io_wq;
732 bio->bi_end_io = end_workqueue_bio;
733 return 0;
734}
735
736unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
737{
738 unsigned long limit = min_t(unsigned long,
739 info->workers.max_workers,
740 info->fs_devices->open_devices);
741 return 256 * limit;
742}
743
744static void run_one_async_start(struct btrfs_work *work)
745{
746 struct async_submit_bio *async;
747 int ret;
748
749 async = container_of(work, struct async_submit_bio, work);
750 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
751 async->mirror_num, async->bio_flags,
752 async->bio_offset);
753 if (ret)
754 async->error = ret;
755}
756
757static void run_one_async_done(struct btrfs_work *work)
758{
759 struct btrfs_fs_info *fs_info;
760 struct async_submit_bio *async;
761 int limit;
762
763 async = container_of(work, struct async_submit_bio, work);
764 fs_info = BTRFS_I(async->inode)->root->fs_info;
765
766 limit = btrfs_async_submit_limit(fs_info);
767 limit = limit * 2 / 3;
768
769 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
770 waitqueue_active(&fs_info->async_submit_wait))
771 wake_up(&fs_info->async_submit_wait);
772
773 /* If an error occured we just want to clean up the bio and move on */
774 if (async->error) {
775 bio_endio(async->bio, async->error);
776 return;
777 }
778
779 async->submit_bio_done(async->inode, async->rw, async->bio,
780 async->mirror_num, async->bio_flags,
781 async->bio_offset);
782}
783
784static void run_one_async_free(struct btrfs_work *work)
785{
786 struct async_submit_bio *async;
787
788 async = container_of(work, struct async_submit_bio, work);
789 kfree(async);
790}
791
792int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
793 int rw, struct bio *bio, int mirror_num,
794 unsigned long bio_flags,
795 u64 bio_offset,
796 extent_submit_bio_hook_t *submit_bio_start,
797 extent_submit_bio_hook_t *submit_bio_done)
798{
799 struct async_submit_bio *async;
800
801 async = kmalloc(sizeof(*async), GFP_NOFS);
802 if (!async)
803 return -ENOMEM;
804
805 async->inode = inode;
806 async->rw = rw;
807 async->bio = bio;
808 async->mirror_num = mirror_num;
809 async->submit_bio_start = submit_bio_start;
810 async->submit_bio_done = submit_bio_done;
811
812 async->work.func = run_one_async_start;
813 async->work.ordered_func = run_one_async_done;
814 async->work.ordered_free = run_one_async_free;
815
816 async->work.flags = 0;
817 async->bio_flags = bio_flags;
818 async->bio_offset = bio_offset;
819
820 async->error = 0;
821
822 atomic_inc(&fs_info->nr_async_submits);
823
824 if (rw & REQ_SYNC)
825 btrfs_set_work_high_prio(&async->work);
826
827 btrfs_queue_worker(&fs_info->workers, &async->work);
828
829 while (atomic_read(&fs_info->async_submit_draining) &&
830 atomic_read(&fs_info->nr_async_submits)) {
831 wait_event(fs_info->async_submit_wait,
832 (atomic_read(&fs_info->nr_async_submits) == 0));
833 }
834
835 return 0;
836}
837
838static int btree_csum_one_bio(struct bio *bio)
839{
840 struct bio_vec *bvec = bio->bi_io_vec;
841 int bio_index = 0;
842 struct btrfs_root *root;
843 int ret = 0;
844
845 WARN_ON(bio->bi_vcnt <= 0);
846 while (bio_index < bio->bi_vcnt) {
847 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
848 ret = csum_dirty_buffer(root, bvec->bv_page);
849 if (ret)
850 break;
851 bio_index++;
852 bvec++;
853 }
854 return ret;
855}
856
857static int __btree_submit_bio_start(struct inode *inode, int rw,
858 struct bio *bio, int mirror_num,
859 unsigned long bio_flags,
860 u64 bio_offset)
861{
862 /*
863 * when we're called for a write, we're already in the async
864 * submission context. Just jump into btrfs_map_bio
865 */
866 return btree_csum_one_bio(bio);
867}
868
869static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
870 int mirror_num, unsigned long bio_flags,
871 u64 bio_offset)
872{
873 int ret;
874
875 /*
876 * when we're called for a write, we're already in the async
877 * submission context. Just jump into btrfs_map_bio
878 */
879 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
880 if (ret)
881 bio_endio(bio, ret);
882 return ret;
883}
884
885static int check_async_write(struct inode *inode, unsigned long bio_flags)
886{
887 if (bio_flags & EXTENT_BIO_TREE_LOG)
888 return 0;
889#ifdef CONFIG_X86
890 if (cpu_has_xmm4_2)
891 return 0;
892#endif
893 return 1;
894}
895
896static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
897 int mirror_num, unsigned long bio_flags,
898 u64 bio_offset)
899{
900 int async = check_async_write(inode, bio_flags);
901 int ret;
902
903 if (!(rw & REQ_WRITE)) {
904 /*
905 * called for a read, do the setup so that checksum validation
906 * can happen in the async kernel threads
907 */
908 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
909 bio, 1);
910 if (ret)
911 goto out_w_error;
912 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
913 mirror_num, 0);
914 } else if (!async) {
915 ret = btree_csum_one_bio(bio);
916 if (ret)
917 goto out_w_error;
918 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
919 mirror_num, 0);
920 } else {
921 /*
922 * kthread helpers are used to submit writes so that
923 * checksumming can happen in parallel across all CPUs
924 */
925 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
926 inode, rw, bio, mirror_num, 0,
927 bio_offset,
928 __btree_submit_bio_start,
929 __btree_submit_bio_done);
930 }
931
932 if (ret) {
933out_w_error:
934 bio_endio(bio, ret);
935 }
936 return ret;
937}
938
939#ifdef CONFIG_MIGRATION
940static int btree_migratepage(struct address_space *mapping,
941 struct page *newpage, struct page *page,
942 enum migrate_mode mode)
943{
944 /*
945 * we can't safely write a btree page from here,
946 * we haven't done the locking hook
947 */
948 if (PageDirty(page))
949 return -EAGAIN;
950 /*
951 * Buffers may be managed in a filesystem specific way.
952 * We must have no buffers or drop them.
953 */
954 if (page_has_private(page) &&
955 !try_to_release_page(page, GFP_KERNEL))
956 return -EAGAIN;
957 return migrate_page(mapping, newpage, page, mode);
958}
959#endif
960
961
962static int btree_writepages(struct address_space *mapping,
963 struct writeback_control *wbc)
964{
965 struct extent_io_tree *tree;
966 struct btrfs_fs_info *fs_info;
967 int ret;
968
969 tree = &BTRFS_I(mapping->host)->io_tree;
970 if (wbc->sync_mode == WB_SYNC_NONE) {
971
972 if (wbc->for_kupdate)
973 return 0;
974
975 fs_info = BTRFS_I(mapping->host)->root->fs_info;
976 /* this is a bit racy, but that's ok */
977 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
978 BTRFS_DIRTY_METADATA_THRESH);
979 if (ret < 0)
980 return 0;
981 }
982 return btree_write_cache_pages(mapping, wbc);
983}
984
985static int btree_readpage(struct file *file, struct page *page)
986{
987 struct extent_io_tree *tree;
988 tree = &BTRFS_I(page->mapping->host)->io_tree;
989 return extent_read_full_page(tree, page, btree_get_extent, 0);
990}
991
992static int btree_releasepage(struct page *page, gfp_t gfp_flags)
993{
994 if (PageWriteback(page) || PageDirty(page))
995 return 0;
996 /*
997 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
998 * slab allocation from alloc_extent_state down the callchain where
999 * it'd hit a BUG_ON as those flags are not allowed.
1000 */
1001 gfp_flags &= ~GFP_SLAB_BUG_MASK;
1002
1003 return try_release_extent_buffer(page, gfp_flags);
1004}
1005
1006static void btree_invalidatepage(struct page *page, unsigned long offset)
1007{
1008 struct extent_io_tree *tree;
1009 tree = &BTRFS_I(page->mapping->host)->io_tree;
1010 extent_invalidatepage(tree, page, offset);
1011 btree_releasepage(page, GFP_NOFS);
1012 if (PagePrivate(page)) {
1013 printk(KERN_WARNING "btrfs warning page private not zero "
1014 "on page %llu\n", (unsigned long long)page_offset(page));
1015 ClearPagePrivate(page);
1016 set_page_private(page, 0);
1017 page_cache_release(page);
1018 }
1019}
1020
1021static int btree_set_page_dirty(struct page *page)
1022{
1023#ifdef DEBUG
1024 struct extent_buffer *eb;
1025
1026 BUG_ON(!PagePrivate(page));
1027 eb = (struct extent_buffer *)page->private;
1028 BUG_ON(!eb);
1029 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1030 BUG_ON(!atomic_read(&eb->refs));
1031 btrfs_assert_tree_locked(eb);
1032#endif
1033 return __set_page_dirty_nobuffers(page);
1034}
1035
1036static const struct address_space_operations btree_aops = {
1037 .readpage = btree_readpage,
1038 .writepages = btree_writepages,
1039 .releasepage = btree_releasepage,
1040 .invalidatepage = btree_invalidatepage,
1041#ifdef CONFIG_MIGRATION
1042 .migratepage = btree_migratepage,
1043#endif
1044 .set_page_dirty = btree_set_page_dirty,
1045};
1046
1047int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1048 u64 parent_transid)
1049{
1050 struct extent_buffer *buf = NULL;
1051 struct inode *btree_inode = root->fs_info->btree_inode;
1052 int ret = 0;
1053
1054 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1055 if (!buf)
1056 return 0;
1057 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1058 buf, 0, WAIT_NONE, btree_get_extent, 0);
1059 free_extent_buffer(buf);
1060 return ret;
1061}
1062
1063int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1064 int mirror_num, struct extent_buffer **eb)
1065{
1066 struct extent_buffer *buf = NULL;
1067 struct inode *btree_inode = root->fs_info->btree_inode;
1068 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1069 int ret;
1070
1071 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1072 if (!buf)
1073 return 0;
1074
1075 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1076
1077 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1078 btree_get_extent, mirror_num);
1079 if (ret) {
1080 free_extent_buffer(buf);
1081 return ret;
1082 }
1083
1084 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1085 free_extent_buffer(buf);
1086 return -EIO;
1087 } else if (extent_buffer_uptodate(buf)) {
1088 *eb = buf;
1089 } else {
1090 free_extent_buffer(buf);
1091 }
1092 return 0;
1093}
1094
1095struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1096 u64 bytenr, u32 blocksize)
1097{
1098 struct inode *btree_inode = root->fs_info->btree_inode;
1099 struct extent_buffer *eb;
1100 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1101 bytenr, blocksize);
1102 return eb;
1103}
1104
1105struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1106 u64 bytenr, u32 blocksize)
1107{
1108 struct inode *btree_inode = root->fs_info->btree_inode;
1109 struct extent_buffer *eb;
1110
1111 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1112 bytenr, blocksize);
1113 return eb;
1114}
1115
1116
1117int btrfs_write_tree_block(struct extent_buffer *buf)
1118{
1119 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1120 buf->start + buf->len - 1);
1121}
1122
1123int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1124{
1125 return filemap_fdatawait_range(buf->pages[0]->mapping,
1126 buf->start, buf->start + buf->len - 1);
1127}
1128
1129struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1130 u32 blocksize, u64 parent_transid)
1131{
1132 struct extent_buffer *buf = NULL;
1133 int ret;
1134
1135 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1136 if (!buf)
1137 return NULL;
1138
1139 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1140 return buf;
1141
1142}
1143
1144void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1145 struct extent_buffer *buf)
1146{
1147 struct btrfs_fs_info *fs_info = root->fs_info;
1148
1149 if (btrfs_header_generation(buf) ==
1150 fs_info->running_transaction->transid) {
1151 btrfs_assert_tree_locked(buf);
1152
1153 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1154 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1155 -buf->len,
1156 fs_info->dirty_metadata_batch);
1157 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1158 btrfs_set_lock_blocking(buf);
1159 clear_extent_buffer_dirty(buf);
1160 }
1161 }
1162}
1163
1164static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1165 u32 stripesize, struct btrfs_root *root,
1166 struct btrfs_fs_info *fs_info,
1167 u64 objectid)
1168{
1169 root->node = NULL;
1170 root->commit_root = NULL;
1171 root->sectorsize = sectorsize;
1172 root->nodesize = nodesize;
1173 root->leafsize = leafsize;
1174 root->stripesize = stripesize;
1175 root->ref_cows = 0;
1176 root->track_dirty = 0;
1177 root->in_radix = 0;
1178 root->orphan_item_inserted = 0;
1179 root->orphan_cleanup_state = 0;
1180
1181 root->objectid = objectid;
1182 root->last_trans = 0;
1183 root->highest_objectid = 0;
1184 root->name = NULL;
1185 root->inode_tree = RB_ROOT;
1186 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1187 root->block_rsv = NULL;
1188 root->orphan_block_rsv = NULL;
1189
1190 INIT_LIST_HEAD(&root->dirty_list);
1191 INIT_LIST_HEAD(&root->root_list);
1192 INIT_LIST_HEAD(&root->logged_list[0]);
1193 INIT_LIST_HEAD(&root->logged_list[1]);
1194 spin_lock_init(&root->orphan_lock);
1195 spin_lock_init(&root->inode_lock);
1196 spin_lock_init(&root->accounting_lock);
1197 spin_lock_init(&root->log_extents_lock[0]);
1198 spin_lock_init(&root->log_extents_lock[1]);
1199 mutex_init(&root->objectid_mutex);
1200 mutex_init(&root->log_mutex);
1201 init_waitqueue_head(&root->log_writer_wait);
1202 init_waitqueue_head(&root->log_commit_wait[0]);
1203 init_waitqueue_head(&root->log_commit_wait[1]);
1204 atomic_set(&root->log_commit[0], 0);
1205 atomic_set(&root->log_commit[1], 0);
1206 atomic_set(&root->log_writers, 0);
1207 atomic_set(&root->log_batch, 0);
1208 atomic_set(&root->orphan_inodes, 0);
1209 root->log_transid = 0;
1210 root->last_log_commit = 0;
1211 extent_io_tree_init(&root->dirty_log_pages,
1212 fs_info->btree_inode->i_mapping);
1213
1214 memset(&root->root_key, 0, sizeof(root->root_key));
1215 memset(&root->root_item, 0, sizeof(root->root_item));
1216 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1217 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1218 root->defrag_trans_start = fs_info->generation;
1219 init_completion(&root->kobj_unregister);
1220 root->defrag_running = 0;
1221 root->root_key.objectid = objectid;
1222 root->anon_dev = 0;
1223
1224 spin_lock_init(&root->root_item_lock);
1225}
1226
1227static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1228 struct btrfs_fs_info *fs_info,
1229 u64 objectid,
1230 struct btrfs_root *root)
1231{
1232 int ret;
1233 u32 blocksize;
1234 u64 generation;
1235
1236 __setup_root(tree_root->nodesize, tree_root->leafsize,
1237 tree_root->sectorsize, tree_root->stripesize,
1238 root, fs_info, objectid);
1239 ret = btrfs_find_last_root(tree_root, objectid,
1240 &root->root_item, &root->root_key);
1241 if (ret > 0)
1242 return -ENOENT;
1243 else if (ret < 0)
1244 return ret;
1245
1246 generation = btrfs_root_generation(&root->root_item);
1247 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1248 root->commit_root = NULL;
1249 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1250 blocksize, generation);
1251 if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1252 free_extent_buffer(root->node);
1253 root->node = NULL;
1254 return -EIO;
1255 }
1256 root->commit_root = btrfs_root_node(root);
1257 return 0;
1258}
1259
1260static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1261{
1262 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1263 if (root)
1264 root->fs_info = fs_info;
1265 return root;
1266}
1267
1268struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1269 struct btrfs_fs_info *fs_info,
1270 u64 objectid)
1271{
1272 struct extent_buffer *leaf;
1273 struct btrfs_root *tree_root = fs_info->tree_root;
1274 struct btrfs_root *root;
1275 struct btrfs_key key;
1276 int ret = 0;
1277 u64 bytenr;
1278
1279 root = btrfs_alloc_root(fs_info);
1280 if (!root)
1281 return ERR_PTR(-ENOMEM);
1282
1283 __setup_root(tree_root->nodesize, tree_root->leafsize,
1284 tree_root->sectorsize, tree_root->stripesize,
1285 root, fs_info, objectid);
1286 root->root_key.objectid = objectid;
1287 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1288 root->root_key.offset = 0;
1289
1290 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1291 0, objectid, NULL, 0, 0, 0);
1292 if (IS_ERR(leaf)) {
1293 ret = PTR_ERR(leaf);
1294 leaf = NULL;
1295 goto fail;
1296 }
1297
1298 bytenr = leaf->start;
1299 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1300 btrfs_set_header_bytenr(leaf, leaf->start);
1301 btrfs_set_header_generation(leaf, trans->transid);
1302 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1303 btrfs_set_header_owner(leaf, objectid);
1304 root->node = leaf;
1305
1306 write_extent_buffer(leaf, fs_info->fsid,
1307 (unsigned long)btrfs_header_fsid(leaf),
1308 BTRFS_FSID_SIZE);
1309 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1310 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1311 BTRFS_UUID_SIZE);
1312 btrfs_mark_buffer_dirty(leaf);
1313
1314 root->commit_root = btrfs_root_node(root);
1315 root->track_dirty = 1;
1316
1317
1318 root->root_item.flags = 0;
1319 root->root_item.byte_limit = 0;
1320 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1321 btrfs_set_root_generation(&root->root_item, trans->transid);
1322 btrfs_set_root_level(&root->root_item, 0);
1323 btrfs_set_root_refs(&root->root_item, 1);
1324 btrfs_set_root_used(&root->root_item, leaf->len);
1325 btrfs_set_root_last_snapshot(&root->root_item, 0);
1326 btrfs_set_root_dirid(&root->root_item, 0);
1327 root->root_item.drop_level = 0;
1328
1329 key.objectid = objectid;
1330 key.type = BTRFS_ROOT_ITEM_KEY;
1331 key.offset = 0;
1332 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1333 if (ret)
1334 goto fail;
1335
1336 btrfs_tree_unlock(leaf);
1337
1338 return root;
1339
1340fail:
1341 if (leaf) {
1342 btrfs_tree_unlock(leaf);
1343 free_extent_buffer(leaf);
1344 }
1345 kfree(root);
1346
1347 return ERR_PTR(ret);
1348}
1349
1350static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1351 struct btrfs_fs_info *fs_info)
1352{
1353 struct btrfs_root *root;
1354 struct btrfs_root *tree_root = fs_info->tree_root;
1355 struct extent_buffer *leaf;
1356
1357 root = btrfs_alloc_root(fs_info);
1358 if (!root)
1359 return ERR_PTR(-ENOMEM);
1360
1361 __setup_root(tree_root->nodesize, tree_root->leafsize,
1362 tree_root->sectorsize, tree_root->stripesize,
1363 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1364
1365 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1366 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1367 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1368 /*
1369 * log trees do not get reference counted because they go away
1370 * before a real commit is actually done. They do store pointers
1371 * to file data extents, and those reference counts still get
1372 * updated (along with back refs to the log tree).
1373 */
1374 root->ref_cows = 0;
1375
1376 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1377 BTRFS_TREE_LOG_OBJECTID, NULL,
1378 0, 0, 0);
1379 if (IS_ERR(leaf)) {
1380 kfree(root);
1381 return ERR_CAST(leaf);
1382 }
1383
1384 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1385 btrfs_set_header_bytenr(leaf, leaf->start);
1386 btrfs_set_header_generation(leaf, trans->transid);
1387 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1388 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1389 root->node = leaf;
1390
1391 write_extent_buffer(root->node, root->fs_info->fsid,
1392 (unsigned long)btrfs_header_fsid(root->node),
1393 BTRFS_FSID_SIZE);
1394 btrfs_mark_buffer_dirty(root->node);
1395 btrfs_tree_unlock(root->node);
1396 return root;
1397}
1398
1399int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1400 struct btrfs_fs_info *fs_info)
1401{
1402 struct btrfs_root *log_root;
1403
1404 log_root = alloc_log_tree(trans, fs_info);
1405 if (IS_ERR(log_root))
1406 return PTR_ERR(log_root);
1407 WARN_ON(fs_info->log_root_tree);
1408 fs_info->log_root_tree = log_root;
1409 return 0;
1410}
1411
1412int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1413 struct btrfs_root *root)
1414{
1415 struct btrfs_root *log_root;
1416 struct btrfs_inode_item *inode_item;
1417
1418 log_root = alloc_log_tree(trans, root->fs_info);
1419 if (IS_ERR(log_root))
1420 return PTR_ERR(log_root);
1421
1422 log_root->last_trans = trans->transid;
1423 log_root->root_key.offset = root->root_key.objectid;
1424
1425 inode_item = &log_root->root_item.inode;
1426 inode_item->generation = cpu_to_le64(1);
1427 inode_item->size = cpu_to_le64(3);
1428 inode_item->nlink = cpu_to_le32(1);
1429 inode_item->nbytes = cpu_to_le64(root->leafsize);
1430 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1431
1432 btrfs_set_root_node(&log_root->root_item, log_root->node);
1433
1434 WARN_ON(root->log_root);
1435 root->log_root = log_root;
1436 root->log_transid = 0;
1437 root->last_log_commit = 0;
1438 return 0;
1439}
1440
1441struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1442 struct btrfs_key *location)
1443{
1444 struct btrfs_root *root;
1445 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1446 struct btrfs_path *path;
1447 struct extent_buffer *l;
1448 u64 generation;
1449 u32 blocksize;
1450 int ret = 0;
1451 int slot;
1452
1453 root = btrfs_alloc_root(fs_info);
1454 if (!root)
1455 return ERR_PTR(-ENOMEM);
1456 if (location->offset == (u64)-1) {
1457 ret = find_and_setup_root(tree_root, fs_info,
1458 location->objectid, root);
1459 if (ret) {
1460 kfree(root);
1461 return ERR_PTR(ret);
1462 }
1463 goto out;
1464 }
1465
1466 __setup_root(tree_root->nodesize, tree_root->leafsize,
1467 tree_root->sectorsize, tree_root->stripesize,
1468 root, fs_info, location->objectid);
1469
1470 path = btrfs_alloc_path();
1471 if (!path) {
1472 kfree(root);
1473 return ERR_PTR(-ENOMEM);
1474 }
1475 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1476 if (ret == 0) {
1477 l = path->nodes[0];
1478 slot = path->slots[0];
1479 btrfs_read_root_item(tree_root, l, slot, &root->root_item);
1480 memcpy(&root->root_key, location, sizeof(*location));
1481 }
1482 btrfs_free_path(path);
1483 if (ret) {
1484 kfree(root);
1485 if (ret > 0)
1486 ret = -ENOENT;
1487 return ERR_PTR(ret);
1488 }
1489
1490 generation = btrfs_root_generation(&root->root_item);
1491 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1492 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1493 blocksize, generation);
1494 root->commit_root = btrfs_root_node(root);
1495 BUG_ON(!root->node); /* -ENOMEM */
1496out:
1497 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1498 root->ref_cows = 1;
1499 btrfs_check_and_init_root_item(&root->root_item);
1500 }
1501
1502 return root;
1503}
1504
1505struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1506 struct btrfs_key *location)
1507{
1508 struct btrfs_root *root;
1509 int ret;
1510
1511 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1512 return fs_info->tree_root;
1513 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1514 return fs_info->extent_root;
1515 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1516 return fs_info->chunk_root;
1517 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1518 return fs_info->dev_root;
1519 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1520 return fs_info->csum_root;
1521 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1522 return fs_info->quota_root ? fs_info->quota_root :
1523 ERR_PTR(-ENOENT);
1524again:
1525 spin_lock(&fs_info->fs_roots_radix_lock);
1526 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1527 (unsigned long)location->objectid);
1528 spin_unlock(&fs_info->fs_roots_radix_lock);
1529 if (root)
1530 return root;
1531
1532 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1533 if (IS_ERR(root))
1534 return root;
1535
1536 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1537 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1538 GFP_NOFS);
1539 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1540 ret = -ENOMEM;
1541 goto fail;
1542 }
1543
1544 btrfs_init_free_ino_ctl(root);
1545 mutex_init(&root->fs_commit_mutex);
1546 spin_lock_init(&root->cache_lock);
1547 init_waitqueue_head(&root->cache_wait);
1548
1549 ret = get_anon_bdev(&root->anon_dev);
1550 if (ret)
1551 goto fail;
1552
1553 if (btrfs_root_refs(&root->root_item) == 0) {
1554 ret = -ENOENT;
1555 goto fail;
1556 }
1557
1558 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1559 if (ret < 0)
1560 goto fail;
1561 if (ret == 0)
1562 root->orphan_item_inserted = 1;
1563
1564 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1565 if (ret)
1566 goto fail;
1567
1568 spin_lock(&fs_info->fs_roots_radix_lock);
1569 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1570 (unsigned long)root->root_key.objectid,
1571 root);
1572 if (ret == 0)
1573 root->in_radix = 1;
1574
1575 spin_unlock(&fs_info->fs_roots_radix_lock);
1576 radix_tree_preload_end();
1577 if (ret) {
1578 if (ret == -EEXIST) {
1579 free_fs_root(root);
1580 goto again;
1581 }
1582 goto fail;
1583 }
1584
1585 ret = btrfs_find_dead_roots(fs_info->tree_root,
1586 root->root_key.objectid);
1587 WARN_ON(ret);
1588 return root;
1589fail:
1590 free_fs_root(root);
1591 return ERR_PTR(ret);
1592}
1593
1594static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1595{
1596 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1597 int ret = 0;
1598 struct btrfs_device *device;
1599 struct backing_dev_info *bdi;
1600
1601 rcu_read_lock();
1602 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1603 if (!device->bdev)
1604 continue;
1605 bdi = blk_get_backing_dev_info(device->bdev);
1606 if (bdi && bdi_congested(bdi, bdi_bits)) {
1607 ret = 1;
1608 break;
1609 }
1610 }
1611 rcu_read_unlock();
1612 return ret;
1613}
1614
1615/*
1616 * If this fails, caller must call bdi_destroy() to get rid of the
1617 * bdi again.
1618 */
1619static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1620{
1621 int err;
1622
1623 bdi->capabilities = BDI_CAP_MAP_COPY;
1624 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1625 if (err)
1626 return err;
1627
1628 bdi->ra_pages = default_backing_dev_info.ra_pages;
1629 bdi->congested_fn = btrfs_congested_fn;
1630 bdi->congested_data = info;
1631 return 0;
1632}
1633
1634/*
1635 * called by the kthread helper functions to finally call the bio end_io
1636 * functions. This is where read checksum verification actually happens
1637 */
1638static void end_workqueue_fn(struct btrfs_work *work)
1639{
1640 struct bio *bio;
1641 struct end_io_wq *end_io_wq;
1642 struct btrfs_fs_info *fs_info;
1643 int error;
1644
1645 end_io_wq = container_of(work, struct end_io_wq, work);
1646 bio = end_io_wq->bio;
1647 fs_info = end_io_wq->info;
1648
1649 error = end_io_wq->error;
1650 bio->bi_private = end_io_wq->private;
1651 bio->bi_end_io = end_io_wq->end_io;
1652 kfree(end_io_wq);
1653 bio_endio(bio, error);
1654}
1655
1656static int cleaner_kthread(void *arg)
1657{
1658 struct btrfs_root *root = arg;
1659
1660 do {
1661 int again = 0;
1662
1663 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1664 down_read_trylock(&root->fs_info->sb->s_umount)) {
1665 if (mutex_trylock(&root->fs_info->cleaner_mutex)) {
1666 btrfs_run_delayed_iputs(root);
1667 again = btrfs_clean_one_deleted_snapshot(root);
1668 mutex_unlock(&root->fs_info->cleaner_mutex);
1669 }
1670 btrfs_run_defrag_inodes(root->fs_info);
1671 up_read(&root->fs_info->sb->s_umount);
1672 }
1673
1674 if (!try_to_freeze() && !again) {
1675 set_current_state(TASK_INTERRUPTIBLE);
1676 if (!kthread_should_stop())
1677 schedule();
1678 __set_current_state(TASK_RUNNING);
1679 }
1680 } while (!kthread_should_stop());
1681 return 0;
1682}
1683
1684static int transaction_kthread(void *arg)
1685{
1686 struct btrfs_root *root = arg;
1687 struct btrfs_trans_handle *trans;
1688 struct btrfs_transaction *cur;
1689 u64 transid;
1690 unsigned long now;
1691 unsigned long delay;
1692 bool cannot_commit;
1693
1694 do {
1695 cannot_commit = false;
1696 delay = HZ * 30;
1697 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1698
1699 spin_lock(&root->fs_info->trans_lock);
1700 cur = root->fs_info->running_transaction;
1701 if (!cur) {
1702 spin_unlock(&root->fs_info->trans_lock);
1703 goto sleep;
1704 }
1705
1706 now = get_seconds();
1707 if (!cur->blocked &&
1708 (now < cur->start_time || now - cur->start_time < 30)) {
1709 spin_unlock(&root->fs_info->trans_lock);
1710 delay = HZ * 5;
1711 goto sleep;
1712 }
1713 transid = cur->transid;
1714 spin_unlock(&root->fs_info->trans_lock);
1715
1716 /* If the file system is aborted, this will always fail. */
1717 trans = btrfs_attach_transaction(root);
1718 if (IS_ERR(trans)) {
1719 if (PTR_ERR(trans) != -ENOENT)
1720 cannot_commit = true;
1721 goto sleep;
1722 }
1723 if (transid == trans->transid) {
1724 btrfs_commit_transaction(trans, root);
1725 } else {
1726 btrfs_end_transaction(trans, root);
1727 }
1728sleep:
1729 wake_up_process(root->fs_info->cleaner_kthread);
1730 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1731
1732 if (!try_to_freeze()) {
1733 set_current_state(TASK_INTERRUPTIBLE);
1734 if (!kthread_should_stop() &&
1735 (!btrfs_transaction_blocked(root->fs_info) ||
1736 cannot_commit))
1737 schedule_timeout(delay);
1738 __set_current_state(TASK_RUNNING);
1739 }
1740 } while (!kthread_should_stop());
1741 return 0;
1742}
1743
1744/*
1745 * this will find the highest generation in the array of
1746 * root backups. The index of the highest array is returned,
1747 * or -1 if we can't find anything.
1748 *
1749 * We check to make sure the array is valid by comparing the
1750 * generation of the latest root in the array with the generation
1751 * in the super block. If they don't match we pitch it.
1752 */
1753static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1754{
1755 u64 cur;
1756 int newest_index = -1;
1757 struct btrfs_root_backup *root_backup;
1758 int i;
1759
1760 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1761 root_backup = info->super_copy->super_roots + i;
1762 cur = btrfs_backup_tree_root_gen(root_backup);
1763 if (cur == newest_gen)
1764 newest_index = i;
1765 }
1766
1767 /* check to see if we actually wrapped around */
1768 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1769 root_backup = info->super_copy->super_roots;
1770 cur = btrfs_backup_tree_root_gen(root_backup);
1771 if (cur == newest_gen)
1772 newest_index = 0;
1773 }
1774 return newest_index;
1775}
1776
1777
1778/*
1779 * find the oldest backup so we know where to store new entries
1780 * in the backup array. This will set the backup_root_index
1781 * field in the fs_info struct
1782 */
1783static void find_oldest_super_backup(struct btrfs_fs_info *info,
1784 u64 newest_gen)
1785{
1786 int newest_index = -1;
1787
1788 newest_index = find_newest_super_backup(info, newest_gen);
1789 /* if there was garbage in there, just move along */
1790 if (newest_index == -1) {
1791 info->backup_root_index = 0;
1792 } else {
1793 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1794 }
1795}
1796
1797/*
1798 * copy all the root pointers into the super backup array.
1799 * this will bump the backup pointer by one when it is
1800 * done
1801 */
1802static void backup_super_roots(struct btrfs_fs_info *info)
1803{
1804 int next_backup;
1805 struct btrfs_root_backup *root_backup;
1806 int last_backup;
1807
1808 next_backup = info->backup_root_index;
1809 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1810 BTRFS_NUM_BACKUP_ROOTS;
1811
1812 /*
1813 * just overwrite the last backup if we're at the same generation
1814 * this happens only at umount
1815 */
1816 root_backup = info->super_for_commit->super_roots + last_backup;
1817 if (btrfs_backup_tree_root_gen(root_backup) ==
1818 btrfs_header_generation(info->tree_root->node))
1819 next_backup = last_backup;
1820
1821 root_backup = info->super_for_commit->super_roots + next_backup;
1822
1823 /*
1824 * make sure all of our padding and empty slots get zero filled
1825 * regardless of which ones we use today
1826 */
1827 memset(root_backup, 0, sizeof(*root_backup));
1828
1829 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1830
1831 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1832 btrfs_set_backup_tree_root_gen(root_backup,
1833 btrfs_header_generation(info->tree_root->node));
1834
1835 btrfs_set_backup_tree_root_level(root_backup,
1836 btrfs_header_level(info->tree_root->node));
1837
1838 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1839 btrfs_set_backup_chunk_root_gen(root_backup,
1840 btrfs_header_generation(info->chunk_root->node));
1841 btrfs_set_backup_chunk_root_level(root_backup,
1842 btrfs_header_level(info->chunk_root->node));
1843
1844 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1845 btrfs_set_backup_extent_root_gen(root_backup,
1846 btrfs_header_generation(info->extent_root->node));
1847 btrfs_set_backup_extent_root_level(root_backup,
1848 btrfs_header_level(info->extent_root->node));
1849
1850 /*
1851 * we might commit during log recovery, which happens before we set
1852 * the fs_root. Make sure it is valid before we fill it in.
1853 */
1854 if (info->fs_root && info->fs_root->node) {
1855 btrfs_set_backup_fs_root(root_backup,
1856 info->fs_root->node->start);
1857 btrfs_set_backup_fs_root_gen(root_backup,
1858 btrfs_header_generation(info->fs_root->node));
1859 btrfs_set_backup_fs_root_level(root_backup,
1860 btrfs_header_level(info->fs_root->node));
1861 }
1862
1863 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1864 btrfs_set_backup_dev_root_gen(root_backup,
1865 btrfs_header_generation(info->dev_root->node));
1866 btrfs_set_backup_dev_root_level(root_backup,
1867 btrfs_header_level(info->dev_root->node));
1868
1869 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1870 btrfs_set_backup_csum_root_gen(root_backup,
1871 btrfs_header_generation(info->csum_root->node));
1872 btrfs_set_backup_csum_root_level(root_backup,
1873 btrfs_header_level(info->csum_root->node));
1874
1875 btrfs_set_backup_total_bytes(root_backup,
1876 btrfs_super_total_bytes(info->super_copy));
1877 btrfs_set_backup_bytes_used(root_backup,
1878 btrfs_super_bytes_used(info->super_copy));
1879 btrfs_set_backup_num_devices(root_backup,
1880 btrfs_super_num_devices(info->super_copy));
1881
1882 /*
1883 * if we don't copy this out to the super_copy, it won't get remembered
1884 * for the next commit
1885 */
1886 memcpy(&info->super_copy->super_roots,
1887 &info->super_for_commit->super_roots,
1888 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1889}
1890
1891/*
1892 * this copies info out of the root backup array and back into
1893 * the in-memory super block. It is meant to help iterate through
1894 * the array, so you send it the number of backups you've already
1895 * tried and the last backup index you used.
1896 *
1897 * this returns -1 when it has tried all the backups
1898 */
1899static noinline int next_root_backup(struct btrfs_fs_info *info,
1900 struct btrfs_super_block *super,
1901 int *num_backups_tried, int *backup_index)
1902{
1903 struct btrfs_root_backup *root_backup;
1904 int newest = *backup_index;
1905
1906 if (*num_backups_tried == 0) {
1907 u64 gen = btrfs_super_generation(super);
1908
1909 newest = find_newest_super_backup(info, gen);
1910 if (newest == -1)
1911 return -1;
1912
1913 *backup_index = newest;
1914 *num_backups_tried = 1;
1915 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1916 /* we've tried all the backups, all done */
1917 return -1;
1918 } else {
1919 /* jump to the next oldest backup */
1920 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1921 BTRFS_NUM_BACKUP_ROOTS;
1922 *backup_index = newest;
1923 *num_backups_tried += 1;
1924 }
1925 root_backup = super->super_roots + newest;
1926
1927 btrfs_set_super_generation(super,
1928 btrfs_backup_tree_root_gen(root_backup));
1929 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1930 btrfs_set_super_root_level(super,
1931 btrfs_backup_tree_root_level(root_backup));
1932 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1933
1934 /*
1935 * fixme: the total bytes and num_devices need to match or we should
1936 * need a fsck
1937 */
1938 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1939 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1940 return 0;
1941}
1942
1943/* helper to cleanup workers */
1944static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1945{
1946 btrfs_stop_workers(&fs_info->generic_worker);
1947 btrfs_stop_workers(&fs_info->fixup_workers);
1948 btrfs_stop_workers(&fs_info->delalloc_workers);
1949 btrfs_stop_workers(&fs_info->workers);
1950 btrfs_stop_workers(&fs_info->endio_workers);
1951 btrfs_stop_workers(&fs_info->endio_meta_workers);
1952 btrfs_stop_workers(&fs_info->endio_raid56_workers);
1953 btrfs_stop_workers(&fs_info->rmw_workers);
1954 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1955 btrfs_stop_workers(&fs_info->endio_write_workers);
1956 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1957 btrfs_stop_workers(&fs_info->submit_workers);
1958 btrfs_stop_workers(&fs_info->delayed_workers);
1959 btrfs_stop_workers(&fs_info->caching_workers);
1960 btrfs_stop_workers(&fs_info->readahead_workers);
1961 btrfs_stop_workers(&fs_info->flush_workers);
1962}
1963
1964/* helper to cleanup tree roots */
1965static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1966{
1967 free_extent_buffer(info->tree_root->node);
1968 free_extent_buffer(info->tree_root->commit_root);
1969 free_extent_buffer(info->dev_root->node);
1970 free_extent_buffer(info->dev_root->commit_root);
1971 free_extent_buffer(info->extent_root->node);
1972 free_extent_buffer(info->extent_root->commit_root);
1973 free_extent_buffer(info->csum_root->node);
1974 free_extent_buffer(info->csum_root->commit_root);
1975 if (info->quota_root) {
1976 free_extent_buffer(info->quota_root->node);
1977 free_extent_buffer(info->quota_root->commit_root);
1978 }
1979
1980 info->tree_root->node = NULL;
1981 info->tree_root->commit_root = NULL;
1982 info->dev_root->node = NULL;
1983 info->dev_root->commit_root = NULL;
1984 info->extent_root->node = NULL;
1985 info->extent_root->commit_root = NULL;
1986 info->csum_root->node = NULL;
1987 info->csum_root->commit_root = NULL;
1988 if (info->quota_root) {
1989 info->quota_root->node = NULL;
1990 info->quota_root->commit_root = NULL;
1991 }
1992
1993 if (chunk_root) {
1994 free_extent_buffer(info->chunk_root->node);
1995 free_extent_buffer(info->chunk_root->commit_root);
1996 info->chunk_root->node = NULL;
1997 info->chunk_root->commit_root = NULL;
1998 }
1999}
2000
2001
2002int open_ctree(struct super_block *sb,
2003 struct btrfs_fs_devices *fs_devices,
2004 char *options)
2005{
2006 u32 sectorsize;
2007 u32 nodesize;
2008 u32 leafsize;
2009 u32 blocksize;
2010 u32 stripesize;
2011 u64 generation;
2012 u64 features;
2013 struct btrfs_key location;
2014 struct buffer_head *bh;
2015 struct btrfs_super_block *disk_super;
2016 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2017 struct btrfs_root *tree_root;
2018 struct btrfs_root *extent_root;
2019 struct btrfs_root *csum_root;
2020 struct btrfs_root *chunk_root;
2021 struct btrfs_root *dev_root;
2022 struct btrfs_root *quota_root;
2023 struct btrfs_root *log_tree_root;
2024 int ret;
2025 int err = -EINVAL;
2026 int num_backups_tried = 0;
2027 int backup_index = 0;
2028
2029 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2030 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
2031 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
2032 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2033 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
2034 quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
2035
2036 if (!tree_root || !extent_root || !csum_root ||
2037 !chunk_root || !dev_root || !quota_root) {
2038 err = -ENOMEM;
2039 goto fail;
2040 }
2041
2042 ret = init_srcu_struct(&fs_info->subvol_srcu);
2043 if (ret) {
2044 err = ret;
2045 goto fail;
2046 }
2047
2048 ret = setup_bdi(fs_info, &fs_info->bdi);
2049 if (ret) {
2050 err = ret;
2051 goto fail_srcu;
2052 }
2053
2054 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2055 if (ret) {
2056 err = ret;
2057 goto fail_bdi;
2058 }
2059 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2060 (1 + ilog2(nr_cpu_ids));
2061
2062 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2063 if (ret) {
2064 err = ret;
2065 goto fail_dirty_metadata_bytes;
2066 }
2067
2068 fs_info->btree_inode = new_inode(sb);
2069 if (!fs_info->btree_inode) {
2070 err = -ENOMEM;
2071 goto fail_delalloc_bytes;
2072 }
2073
2074 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2075
2076 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2077 INIT_LIST_HEAD(&fs_info->trans_list);
2078 INIT_LIST_HEAD(&fs_info->dead_roots);
2079 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2080 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
2081 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2082 spin_lock_init(&fs_info->delalloc_lock);
2083 spin_lock_init(&fs_info->trans_lock);
2084 spin_lock_init(&fs_info->fs_roots_radix_lock);
2085 spin_lock_init(&fs_info->delayed_iput_lock);
2086 spin_lock_init(&fs_info->defrag_inodes_lock);
2087 spin_lock_init(&fs_info->free_chunk_lock);
2088 spin_lock_init(&fs_info->tree_mod_seq_lock);
2089 rwlock_init(&fs_info->tree_mod_log_lock);
2090 mutex_init(&fs_info->reloc_mutex);
2091 seqlock_init(&fs_info->profiles_lock);
2092
2093 init_completion(&fs_info->kobj_unregister);
2094 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2095 INIT_LIST_HEAD(&fs_info->space_info);
2096 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2097 btrfs_mapping_init(&fs_info->mapping_tree);
2098 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2099 BTRFS_BLOCK_RSV_GLOBAL);
2100 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2101 BTRFS_BLOCK_RSV_DELALLOC);
2102 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2103 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2104 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2105 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2106 BTRFS_BLOCK_RSV_DELOPS);
2107 atomic_set(&fs_info->nr_async_submits, 0);
2108 atomic_set(&fs_info->async_delalloc_pages, 0);
2109 atomic_set(&fs_info->async_submit_draining, 0);
2110 atomic_set(&fs_info->nr_async_bios, 0);
2111 atomic_set(&fs_info->defrag_running, 0);
2112 atomic_set(&fs_info->tree_mod_seq, 0);
2113 fs_info->sb = sb;
2114 fs_info->max_inline = 8192 * 1024;
2115 fs_info->metadata_ratio = 0;
2116 fs_info->defrag_inodes = RB_ROOT;
2117 fs_info->trans_no_join = 0;
2118 fs_info->free_chunk_space = 0;
2119 fs_info->tree_mod_log = RB_ROOT;
2120
2121 /* readahead state */
2122 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2123 spin_lock_init(&fs_info->reada_lock);
2124
2125 fs_info->thread_pool_size = min_t(unsigned long,
2126 num_online_cpus() + 2, 8);
2127
2128 INIT_LIST_HEAD(&fs_info->ordered_extents);
2129 spin_lock_init(&fs_info->ordered_extent_lock);
2130 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2131 GFP_NOFS);
2132 if (!fs_info->delayed_root) {
2133 err = -ENOMEM;
2134 goto fail_iput;
2135 }
2136 btrfs_init_delayed_root(fs_info->delayed_root);
2137
2138 mutex_init(&fs_info->scrub_lock);
2139 atomic_set(&fs_info->scrubs_running, 0);
2140 atomic_set(&fs_info->scrub_pause_req, 0);
2141 atomic_set(&fs_info->scrubs_paused, 0);
2142 atomic_set(&fs_info->scrub_cancel_req, 0);
2143 init_waitqueue_head(&fs_info->scrub_pause_wait);
2144 init_rwsem(&fs_info->scrub_super_lock);
2145 fs_info->scrub_workers_refcnt = 0;
2146#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2147 fs_info->check_integrity_print_mask = 0;
2148#endif
2149
2150 spin_lock_init(&fs_info->balance_lock);
2151 mutex_init(&fs_info->balance_mutex);
2152 atomic_set(&fs_info->balance_running, 0);
2153 atomic_set(&fs_info->balance_pause_req, 0);
2154 atomic_set(&fs_info->balance_cancel_req, 0);
2155 fs_info->balance_ctl = NULL;
2156 init_waitqueue_head(&fs_info->balance_wait_q);
2157
2158 sb->s_blocksize = 4096;
2159 sb->s_blocksize_bits = blksize_bits(4096);
2160 sb->s_bdi = &fs_info->bdi;
2161
2162 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2163 set_nlink(fs_info->btree_inode, 1);
2164 /*
2165 * we set the i_size on the btree inode to the max possible int.
2166 * the real end of the address space is determined by all of
2167 * the devices in the system
2168 */
2169 fs_info->btree_inode->i_size = OFFSET_MAX;
2170 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2171 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2172
2173 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2174 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2175 fs_info->btree_inode->i_mapping);
2176 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2177 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2178
2179 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2180
2181 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2182 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2183 sizeof(struct btrfs_key));
2184 set_bit(BTRFS_INODE_DUMMY,
2185 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2186 insert_inode_hash(fs_info->btree_inode);
2187
2188 spin_lock_init(&fs_info->block_group_cache_lock);
2189 fs_info->block_group_cache_tree = RB_ROOT;
2190 fs_info->first_logical_byte = (u64)-1;
2191
2192 extent_io_tree_init(&fs_info->freed_extents[0],
2193 fs_info->btree_inode->i_mapping);
2194 extent_io_tree_init(&fs_info->freed_extents[1],
2195 fs_info->btree_inode->i_mapping);
2196 fs_info->pinned_extents = &fs_info->freed_extents[0];
2197 fs_info->do_barriers = 1;
2198
2199
2200 mutex_init(&fs_info->ordered_operations_mutex);
2201 mutex_init(&fs_info->tree_log_mutex);
2202 mutex_init(&fs_info->chunk_mutex);
2203 mutex_init(&fs_info->transaction_kthread_mutex);
2204 mutex_init(&fs_info->cleaner_mutex);
2205 mutex_init(&fs_info->volume_mutex);
2206 init_rwsem(&fs_info->extent_commit_sem);
2207 init_rwsem(&fs_info->cleanup_work_sem);
2208 init_rwsem(&fs_info->subvol_sem);
2209 fs_info->dev_replace.lock_owner = 0;
2210 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2211 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2212 mutex_init(&fs_info->dev_replace.lock_management_lock);
2213 mutex_init(&fs_info->dev_replace.lock);
2214
2215 spin_lock_init(&fs_info->qgroup_lock);
2216 mutex_init(&fs_info->qgroup_ioctl_lock);
2217 fs_info->qgroup_tree = RB_ROOT;
2218 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2219 fs_info->qgroup_seq = 1;
2220 fs_info->quota_enabled = 0;
2221 fs_info->pending_quota_state = 0;
2222
2223 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2224 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2225
2226 init_waitqueue_head(&fs_info->transaction_throttle);
2227 init_waitqueue_head(&fs_info->transaction_wait);
2228 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2229 init_waitqueue_head(&fs_info->async_submit_wait);
2230
2231 ret = btrfs_alloc_stripe_hash_table(fs_info);
2232 if (ret) {
2233 err = ret;
2234 goto fail_alloc;
2235 }
2236
2237 __setup_root(4096, 4096, 4096, 4096, tree_root,
2238 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2239
2240 invalidate_bdev(fs_devices->latest_bdev);
2241 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2242 if (!bh) {
2243 err = -EINVAL;
2244 goto fail_alloc;
2245 }
2246
2247 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2248 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2249 sizeof(*fs_info->super_for_commit));
2250 brelse(bh);
2251
2252 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2253
2254 disk_super = fs_info->super_copy;
2255 if (!btrfs_super_root(disk_super))
2256 goto fail_alloc;
2257
2258 /* check FS state, whether FS is broken. */
2259 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2260 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2261
2262 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2263 if (ret) {
2264 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2265 err = ret;
2266 goto fail_alloc;
2267 }
2268
2269 /*
2270 * run through our array of backup supers and setup
2271 * our ring pointer to the oldest one
2272 */
2273 generation = btrfs_super_generation(disk_super);
2274 find_oldest_super_backup(fs_info, generation);
2275
2276 /*
2277 * In the long term, we'll store the compression type in the super
2278 * block, and it'll be used for per file compression control.
2279 */
2280 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2281
2282 ret = btrfs_parse_options(tree_root, options);
2283 if (ret) {
2284 err = ret;
2285 goto fail_alloc;
2286 }
2287
2288 features = btrfs_super_incompat_flags(disk_super) &
2289 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2290 if (features) {
2291 printk(KERN_ERR "BTRFS: couldn't mount because of "
2292 "unsupported optional features (%Lx).\n",
2293 (unsigned long long)features);
2294 err = -EINVAL;
2295 goto fail_alloc;
2296 }
2297
2298 if (btrfs_super_leafsize(disk_super) !=
2299 btrfs_super_nodesize(disk_super)) {
2300 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2301 "blocksizes don't match. node %d leaf %d\n",
2302 btrfs_super_nodesize(disk_super),
2303 btrfs_super_leafsize(disk_super));
2304 err = -EINVAL;
2305 goto fail_alloc;
2306 }
2307 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2308 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2309 "blocksize (%d) was too large\n",
2310 btrfs_super_leafsize(disk_super));
2311 err = -EINVAL;
2312 goto fail_alloc;
2313 }
2314
2315 features = btrfs_super_incompat_flags(disk_super);
2316 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2317 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2318 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2319
2320 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2321 printk(KERN_ERR "btrfs: has skinny extents\n");
2322
2323 /*
2324 * flag our filesystem as having big metadata blocks if
2325 * they are bigger than the page size
2326 */
2327 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2328 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2329 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2330 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2331 }
2332
2333 nodesize = btrfs_super_nodesize(disk_super);
2334 leafsize = btrfs_super_leafsize(disk_super);
2335 sectorsize = btrfs_super_sectorsize(disk_super);
2336 stripesize = btrfs_super_stripesize(disk_super);
2337 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2338 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2339
2340 /*
2341 * mixed block groups end up with duplicate but slightly offset
2342 * extent buffers for the same range. It leads to corruptions
2343 */
2344 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2345 (sectorsize != leafsize)) {
2346 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2347 "are not allowed for mixed block groups on %s\n",
2348 sb->s_id);
2349 goto fail_alloc;
2350 }
2351
2352 btrfs_set_super_incompat_flags(disk_super, features);
2353
2354 features = btrfs_super_compat_ro_flags(disk_super) &
2355 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2356 if (!(sb->s_flags & MS_RDONLY) && features) {
2357 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2358 "unsupported option features (%Lx).\n",
2359 (unsigned long long)features);
2360 err = -EINVAL;
2361 goto fail_alloc;
2362 }
2363
2364 btrfs_init_workers(&fs_info->generic_worker,
2365 "genwork", 1, NULL);
2366
2367 btrfs_init_workers(&fs_info->workers, "worker",
2368 fs_info->thread_pool_size,
2369 &fs_info->generic_worker);
2370
2371 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2372 fs_info->thread_pool_size,
2373 &fs_info->generic_worker);
2374
2375 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2376 fs_info->thread_pool_size,
2377 &fs_info->generic_worker);
2378
2379 btrfs_init_workers(&fs_info->submit_workers, "submit",
2380 min_t(u64, fs_devices->num_devices,
2381 fs_info->thread_pool_size),
2382 &fs_info->generic_worker);
2383
2384 btrfs_init_workers(&fs_info->caching_workers, "cache",
2385 2, &fs_info->generic_worker);
2386
2387 /* a higher idle thresh on the submit workers makes it much more
2388 * likely that bios will be send down in a sane order to the
2389 * devices
2390 */
2391 fs_info->submit_workers.idle_thresh = 64;
2392
2393 fs_info->workers.idle_thresh = 16;
2394 fs_info->workers.ordered = 1;
2395
2396 fs_info->delalloc_workers.idle_thresh = 2;
2397 fs_info->delalloc_workers.ordered = 1;
2398
2399 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2400 &fs_info->generic_worker);
2401 btrfs_init_workers(&fs_info->endio_workers, "endio",
2402 fs_info->thread_pool_size,
2403 &fs_info->generic_worker);
2404 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2405 fs_info->thread_pool_size,
2406 &fs_info->generic_worker);
2407 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2408 "endio-meta-write", fs_info->thread_pool_size,
2409 &fs_info->generic_worker);
2410 btrfs_init_workers(&fs_info->endio_raid56_workers,
2411 "endio-raid56", fs_info->thread_pool_size,
2412 &fs_info->generic_worker);
2413 btrfs_init_workers(&fs_info->rmw_workers,
2414 "rmw", fs_info->thread_pool_size,
2415 &fs_info->generic_worker);
2416 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2417 fs_info->thread_pool_size,
2418 &fs_info->generic_worker);
2419 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2420 1, &fs_info->generic_worker);
2421 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2422 fs_info->thread_pool_size,
2423 &fs_info->generic_worker);
2424 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2425 fs_info->thread_pool_size,
2426 &fs_info->generic_worker);
2427
2428 /*
2429 * endios are largely parallel and should have a very
2430 * low idle thresh
2431 */
2432 fs_info->endio_workers.idle_thresh = 4;
2433 fs_info->endio_meta_workers.idle_thresh = 4;
2434 fs_info->endio_raid56_workers.idle_thresh = 4;
2435 fs_info->rmw_workers.idle_thresh = 2;
2436
2437 fs_info->endio_write_workers.idle_thresh = 2;
2438 fs_info->endio_meta_write_workers.idle_thresh = 2;
2439 fs_info->readahead_workers.idle_thresh = 2;
2440
2441 /*
2442 * btrfs_start_workers can really only fail because of ENOMEM so just
2443 * return -ENOMEM if any of these fail.
2444 */
2445 ret = btrfs_start_workers(&fs_info->workers);
2446 ret |= btrfs_start_workers(&fs_info->generic_worker);
2447 ret |= btrfs_start_workers(&fs_info->submit_workers);
2448 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2449 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2450 ret |= btrfs_start_workers(&fs_info->endio_workers);
2451 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2452 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2453 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2454 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2455 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2456 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2457 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2458 ret |= btrfs_start_workers(&fs_info->caching_workers);
2459 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2460 ret |= btrfs_start_workers(&fs_info->flush_workers);
2461 if (ret) {
2462 err = -ENOMEM;
2463 goto fail_sb_buffer;
2464 }
2465
2466 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2467 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2468 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2469
2470 tree_root->nodesize = nodesize;
2471 tree_root->leafsize = leafsize;
2472 tree_root->sectorsize = sectorsize;
2473 tree_root->stripesize = stripesize;
2474
2475 sb->s_blocksize = sectorsize;
2476 sb->s_blocksize_bits = blksize_bits(sectorsize);
2477
2478 if (disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2479 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2480 goto fail_sb_buffer;
2481 }
2482
2483 if (sectorsize != PAGE_SIZE) {
2484 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2485 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2486 goto fail_sb_buffer;
2487 }
2488
2489 mutex_lock(&fs_info->chunk_mutex);
2490 ret = btrfs_read_sys_array(tree_root);
2491 mutex_unlock(&fs_info->chunk_mutex);
2492 if (ret) {
2493 printk(KERN_WARNING "btrfs: failed to read the system "
2494 "array on %s\n", sb->s_id);
2495 goto fail_sb_buffer;
2496 }
2497
2498 blocksize = btrfs_level_size(tree_root,
2499 btrfs_super_chunk_root_level(disk_super));
2500 generation = btrfs_super_chunk_root_generation(disk_super);
2501
2502 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2503 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2504
2505 chunk_root->node = read_tree_block(chunk_root,
2506 btrfs_super_chunk_root(disk_super),
2507 blocksize, generation);
2508 BUG_ON(!chunk_root->node); /* -ENOMEM */
2509 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2510 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2511 sb->s_id);
2512 goto fail_tree_roots;
2513 }
2514 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2515 chunk_root->commit_root = btrfs_root_node(chunk_root);
2516
2517 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2518 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2519 BTRFS_UUID_SIZE);
2520
2521 ret = btrfs_read_chunk_tree(chunk_root);
2522 if (ret) {
2523 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2524 sb->s_id);
2525 goto fail_tree_roots;
2526 }
2527
2528 /*
2529 * keep the device that is marked to be the target device for the
2530 * dev_replace procedure
2531 */
2532 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2533
2534 if (!fs_devices->latest_bdev) {
2535 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2536 sb->s_id);
2537 goto fail_tree_roots;
2538 }
2539
2540retry_root_backup:
2541 blocksize = btrfs_level_size(tree_root,
2542 btrfs_super_root_level(disk_super));
2543 generation = btrfs_super_generation(disk_super);
2544
2545 tree_root->node = read_tree_block(tree_root,
2546 btrfs_super_root(disk_super),
2547 blocksize, generation);
2548 if (!tree_root->node ||
2549 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2550 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2551 sb->s_id);
2552
2553 goto recovery_tree_root;
2554 }
2555
2556 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2557 tree_root->commit_root = btrfs_root_node(tree_root);
2558
2559 ret = find_and_setup_root(tree_root, fs_info,
2560 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2561 if (ret)
2562 goto recovery_tree_root;
2563 extent_root->track_dirty = 1;
2564
2565 ret = find_and_setup_root(tree_root, fs_info,
2566 BTRFS_DEV_TREE_OBJECTID, dev_root);
2567 if (ret)
2568 goto recovery_tree_root;
2569 dev_root->track_dirty = 1;
2570
2571 ret = find_and_setup_root(tree_root, fs_info,
2572 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2573 if (ret)
2574 goto recovery_tree_root;
2575 csum_root->track_dirty = 1;
2576
2577 ret = find_and_setup_root(tree_root, fs_info,
2578 BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2579 if (ret) {
2580 kfree(quota_root);
2581 quota_root = fs_info->quota_root = NULL;
2582 } else {
2583 quota_root->track_dirty = 1;
2584 fs_info->quota_enabled = 1;
2585 fs_info->pending_quota_state = 1;
2586 }
2587
2588 fs_info->generation = generation;
2589 fs_info->last_trans_committed = generation;
2590
2591 ret = btrfs_recover_balance(fs_info);
2592 if (ret) {
2593 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2594 goto fail_block_groups;
2595 }
2596
2597 ret = btrfs_init_dev_stats(fs_info);
2598 if (ret) {
2599 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2600 ret);
2601 goto fail_block_groups;
2602 }
2603
2604 ret = btrfs_init_dev_replace(fs_info);
2605 if (ret) {
2606 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2607 goto fail_block_groups;
2608 }
2609
2610 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2611
2612 ret = btrfs_init_space_info(fs_info);
2613 if (ret) {
2614 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2615 goto fail_block_groups;
2616 }
2617
2618 ret = btrfs_read_block_groups(extent_root);
2619 if (ret) {
2620 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2621 goto fail_block_groups;
2622 }
2623 fs_info->num_tolerated_disk_barrier_failures =
2624 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2625 if (fs_info->fs_devices->missing_devices >
2626 fs_info->num_tolerated_disk_barrier_failures &&
2627 !(sb->s_flags & MS_RDONLY)) {
2628 printk(KERN_WARNING
2629 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2630 goto fail_block_groups;
2631 }
2632
2633 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2634 "btrfs-cleaner");
2635 if (IS_ERR(fs_info->cleaner_kthread))
2636 goto fail_block_groups;
2637
2638 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2639 tree_root,
2640 "btrfs-transaction");
2641 if (IS_ERR(fs_info->transaction_kthread))
2642 goto fail_cleaner;
2643
2644 if (!btrfs_test_opt(tree_root, SSD) &&
2645 !btrfs_test_opt(tree_root, NOSSD) &&
2646 !fs_info->fs_devices->rotating) {
2647 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2648 "mode\n");
2649 btrfs_set_opt(fs_info->mount_opt, SSD);
2650 }
2651
2652#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2653 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2654 ret = btrfsic_mount(tree_root, fs_devices,
2655 btrfs_test_opt(tree_root,
2656 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2657 1 : 0,
2658 fs_info->check_integrity_print_mask);
2659 if (ret)
2660 printk(KERN_WARNING "btrfs: failed to initialize"
2661 " integrity check module %s\n", sb->s_id);
2662 }
2663#endif
2664 ret = btrfs_read_qgroup_config(fs_info);
2665 if (ret)
2666 goto fail_trans_kthread;
2667
2668 /* do not make disk changes in broken FS */
2669 if (btrfs_super_log_root(disk_super) != 0) {
2670 u64 bytenr = btrfs_super_log_root(disk_super);
2671
2672 if (fs_devices->rw_devices == 0) {
2673 printk(KERN_WARNING "Btrfs log replay required "
2674 "on RO media\n");
2675 err = -EIO;
2676 goto fail_qgroup;
2677 }
2678 blocksize =
2679 btrfs_level_size(tree_root,
2680 btrfs_super_log_root_level(disk_super));
2681
2682 log_tree_root = btrfs_alloc_root(fs_info);
2683 if (!log_tree_root) {
2684 err = -ENOMEM;
2685 goto fail_qgroup;
2686 }
2687
2688 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2689 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2690
2691 log_tree_root->node = read_tree_block(tree_root, bytenr,
2692 blocksize,
2693 generation + 1);
2694 /* returns with log_tree_root freed on success */
2695 ret = btrfs_recover_log_trees(log_tree_root);
2696 if (ret) {
2697 btrfs_error(tree_root->fs_info, ret,
2698 "Failed to recover log tree");
2699 free_extent_buffer(log_tree_root->node);
2700 kfree(log_tree_root);
2701 goto fail_trans_kthread;
2702 }
2703
2704 if (sb->s_flags & MS_RDONLY) {
2705 ret = btrfs_commit_super(tree_root);
2706 if (ret)
2707 goto fail_trans_kthread;
2708 }
2709 }
2710
2711 ret = btrfs_find_orphan_roots(tree_root);
2712 if (ret)
2713 goto fail_trans_kthread;
2714
2715 if (!(sb->s_flags & MS_RDONLY)) {
2716 ret = btrfs_cleanup_fs_roots(fs_info);
2717 if (ret)
2718 goto fail_trans_kthread;
2719
2720 ret = btrfs_recover_relocation(tree_root);
2721 if (ret < 0) {
2722 printk(KERN_WARNING
2723 "btrfs: failed to recover relocation\n");
2724 err = -EINVAL;
2725 goto fail_qgroup;
2726 }
2727 }
2728
2729 location.objectid = BTRFS_FS_TREE_OBJECTID;
2730 location.type = BTRFS_ROOT_ITEM_KEY;
2731 location.offset = (u64)-1;
2732
2733 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2734 if (!fs_info->fs_root)
2735 goto fail_qgroup;
2736 if (IS_ERR(fs_info->fs_root)) {
2737 err = PTR_ERR(fs_info->fs_root);
2738 goto fail_qgroup;
2739 }
2740
2741 if (sb->s_flags & MS_RDONLY)
2742 return 0;
2743
2744 down_read(&fs_info->cleanup_work_sem);
2745 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2746 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2747 up_read(&fs_info->cleanup_work_sem);
2748 close_ctree(tree_root);
2749 return ret;
2750 }
2751 up_read(&fs_info->cleanup_work_sem);
2752
2753 ret = btrfs_resume_balance_async(fs_info);
2754 if (ret) {
2755 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2756 close_ctree(tree_root);
2757 return ret;
2758 }
2759
2760 ret = btrfs_resume_dev_replace_async(fs_info);
2761 if (ret) {
2762 pr_warn("btrfs: failed to resume dev_replace\n");
2763 close_ctree(tree_root);
2764 return ret;
2765 }
2766
2767 return 0;
2768
2769fail_qgroup:
2770 btrfs_free_qgroup_config(fs_info);
2771fail_trans_kthread:
2772 kthread_stop(fs_info->transaction_kthread);
2773fail_cleaner:
2774 kthread_stop(fs_info->cleaner_kthread);
2775
2776 /*
2777 * make sure we're done with the btree inode before we stop our
2778 * kthreads
2779 */
2780 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2781
2782fail_block_groups:
2783 btrfs_free_block_groups(fs_info);
2784
2785fail_tree_roots:
2786 free_root_pointers(fs_info, 1);
2787 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2788
2789fail_sb_buffer:
2790 btrfs_stop_all_workers(fs_info);
2791fail_alloc:
2792fail_iput:
2793 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2794
2795 iput(fs_info->btree_inode);
2796fail_delalloc_bytes:
2797 percpu_counter_destroy(&fs_info->delalloc_bytes);
2798fail_dirty_metadata_bytes:
2799 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2800fail_bdi:
2801 bdi_destroy(&fs_info->bdi);
2802fail_srcu:
2803 cleanup_srcu_struct(&fs_info->subvol_srcu);
2804fail:
2805 btrfs_free_stripe_hash_table(fs_info);
2806 btrfs_close_devices(fs_info->fs_devices);
2807 return err;
2808
2809recovery_tree_root:
2810 if (!btrfs_test_opt(tree_root, RECOVERY))
2811 goto fail_tree_roots;
2812
2813 free_root_pointers(fs_info, 0);
2814
2815 /* don't use the log in recovery mode, it won't be valid */
2816 btrfs_set_super_log_root(disk_super, 0);
2817
2818 /* we can't trust the free space cache either */
2819 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2820
2821 ret = next_root_backup(fs_info, fs_info->super_copy,
2822 &num_backups_tried, &backup_index);
2823 if (ret == -1)
2824 goto fail_block_groups;
2825 goto retry_root_backup;
2826}
2827
2828static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2829{
2830 if (uptodate) {
2831 set_buffer_uptodate(bh);
2832 } else {
2833 struct btrfs_device *device = (struct btrfs_device *)
2834 bh->b_private;
2835
2836 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2837 "I/O error on %s\n",
2838 rcu_str_deref(device->name));
2839 /* note, we dont' set_buffer_write_io_error because we have
2840 * our own ways of dealing with the IO errors
2841 */
2842 clear_buffer_uptodate(bh);
2843 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2844 }
2845 unlock_buffer(bh);
2846 put_bh(bh);
2847}
2848
2849struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2850{
2851 struct buffer_head *bh;
2852 struct buffer_head *latest = NULL;
2853 struct btrfs_super_block *super;
2854 int i;
2855 u64 transid = 0;
2856 u64 bytenr;
2857
2858 /* we would like to check all the supers, but that would make
2859 * a btrfs mount succeed after a mkfs from a different FS.
2860 * So, we need to add a special mount option to scan for
2861 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2862 */
2863 for (i = 0; i < 1; i++) {
2864 bytenr = btrfs_sb_offset(i);
2865 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2866 break;
2867 bh = __bread(bdev, bytenr / 4096, 4096);
2868 if (!bh)
2869 continue;
2870
2871 super = (struct btrfs_super_block *)bh->b_data;
2872 if (btrfs_super_bytenr(super) != bytenr ||
2873 super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2874 brelse(bh);
2875 continue;
2876 }
2877
2878 if (!latest || btrfs_super_generation(super) > transid) {
2879 brelse(latest);
2880 latest = bh;
2881 transid = btrfs_super_generation(super);
2882 } else {
2883 brelse(bh);
2884 }
2885 }
2886 return latest;
2887}
2888
2889/*
2890 * this should be called twice, once with wait == 0 and
2891 * once with wait == 1. When wait == 0 is done, all the buffer heads
2892 * we write are pinned.
2893 *
2894 * They are released when wait == 1 is done.
2895 * max_mirrors must be the same for both runs, and it indicates how
2896 * many supers on this one device should be written.
2897 *
2898 * max_mirrors == 0 means to write them all.
2899 */
2900static int write_dev_supers(struct btrfs_device *device,
2901 struct btrfs_super_block *sb,
2902 int do_barriers, int wait, int max_mirrors)
2903{
2904 struct buffer_head *bh;
2905 int i;
2906 int ret;
2907 int errors = 0;
2908 u32 crc;
2909 u64 bytenr;
2910
2911 if (max_mirrors == 0)
2912 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2913
2914 for (i = 0; i < max_mirrors; i++) {
2915 bytenr = btrfs_sb_offset(i);
2916 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2917 break;
2918
2919 if (wait) {
2920 bh = __find_get_block(device->bdev, bytenr / 4096,
2921 BTRFS_SUPER_INFO_SIZE);
2922 BUG_ON(!bh);
2923 wait_on_buffer(bh);
2924 if (!buffer_uptodate(bh))
2925 errors++;
2926
2927 /* drop our reference */
2928 brelse(bh);
2929
2930 /* drop the reference from the wait == 0 run */
2931 brelse(bh);
2932 continue;
2933 } else {
2934 btrfs_set_super_bytenr(sb, bytenr);
2935
2936 crc = ~(u32)0;
2937 crc = btrfs_csum_data((char *)sb +
2938 BTRFS_CSUM_SIZE, crc,
2939 BTRFS_SUPER_INFO_SIZE -
2940 BTRFS_CSUM_SIZE);
2941 btrfs_csum_final(crc, sb->csum);
2942
2943 /*
2944 * one reference for us, and we leave it for the
2945 * caller
2946 */
2947 bh = __getblk(device->bdev, bytenr / 4096,
2948 BTRFS_SUPER_INFO_SIZE);
2949 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2950
2951 /* one reference for submit_bh */
2952 get_bh(bh);
2953
2954 set_buffer_uptodate(bh);
2955 lock_buffer(bh);
2956 bh->b_end_io = btrfs_end_buffer_write_sync;
2957 bh->b_private = device;
2958 }
2959
2960 /*
2961 * we fua the first super. The others we allow
2962 * to go down lazy.
2963 */
2964 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2965 if (ret)
2966 errors++;
2967 }
2968 return errors < i ? 0 : -1;
2969}
2970
2971/*
2972 * endio for the write_dev_flush, this will wake anyone waiting
2973 * for the barrier when it is done
2974 */
2975static void btrfs_end_empty_barrier(struct bio *bio, int err)
2976{
2977 if (err) {
2978 if (err == -EOPNOTSUPP)
2979 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2980 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2981 }
2982 if (bio->bi_private)
2983 complete(bio->bi_private);
2984 bio_put(bio);
2985}
2986
2987/*
2988 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2989 * sent down. With wait == 1, it waits for the previous flush.
2990 *
2991 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2992 * capable
2993 */
2994static int write_dev_flush(struct btrfs_device *device, int wait)
2995{
2996 struct bio *bio;
2997 int ret = 0;
2998
2999 if (device->nobarriers)
3000 return 0;
3001
3002 if (wait) {
3003 bio = device->flush_bio;
3004 if (!bio)
3005 return 0;
3006
3007 wait_for_completion(&device->flush_wait);
3008
3009 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3010 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3011 rcu_str_deref(device->name));
3012 device->nobarriers = 1;
3013 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3014 ret = -EIO;
3015 btrfs_dev_stat_inc_and_print(device,
3016 BTRFS_DEV_STAT_FLUSH_ERRS);
3017 }
3018
3019 /* drop the reference from the wait == 0 run */
3020 bio_put(bio);
3021 device->flush_bio = NULL;
3022
3023 return ret;
3024 }
3025
3026 /*
3027 * one reference for us, and we leave it for the
3028 * caller
3029 */
3030 device->flush_bio = NULL;
3031 bio = bio_alloc(GFP_NOFS, 0);
3032 if (!bio)
3033 return -ENOMEM;
3034
3035 bio->bi_end_io = btrfs_end_empty_barrier;
3036 bio->bi_bdev = device->bdev;
3037 init_completion(&device->flush_wait);
3038 bio->bi_private = &device->flush_wait;
3039 device->flush_bio = bio;
3040
3041 bio_get(bio);
3042 btrfsic_submit_bio(WRITE_FLUSH, bio);
3043
3044 return 0;
3045}
3046
3047/*
3048 * send an empty flush down to each device in parallel,
3049 * then wait for them
3050 */
3051static int barrier_all_devices(struct btrfs_fs_info *info)
3052{
3053 struct list_head *head;
3054 struct btrfs_device *dev;
3055 int errors_send = 0;
3056 int errors_wait = 0;
3057 int ret;
3058
3059 /* send down all the barriers */
3060 head = &info->fs_devices->devices;
3061 list_for_each_entry_rcu(dev, head, dev_list) {
3062 if (!dev->bdev) {
3063 errors_send++;
3064 continue;
3065 }
3066 if (!dev->in_fs_metadata || !dev->writeable)
3067 continue;
3068
3069 ret = write_dev_flush(dev, 0);
3070 if (ret)
3071 errors_send++;
3072 }
3073
3074 /* wait for all the barriers */
3075 list_for_each_entry_rcu(dev, head, dev_list) {
3076 if (!dev->bdev) {
3077 errors_wait++;
3078 continue;
3079 }
3080 if (!dev->in_fs_metadata || !dev->writeable)
3081 continue;
3082
3083 ret = write_dev_flush(dev, 1);
3084 if (ret)
3085 errors_wait++;
3086 }
3087 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3088 errors_wait > info->num_tolerated_disk_barrier_failures)
3089 return -EIO;
3090 return 0;
3091}
3092
3093int btrfs_calc_num_tolerated_disk_barrier_failures(
3094 struct btrfs_fs_info *fs_info)
3095{
3096 struct btrfs_ioctl_space_info space;
3097 struct btrfs_space_info *sinfo;
3098 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3099 BTRFS_BLOCK_GROUP_SYSTEM,
3100 BTRFS_BLOCK_GROUP_METADATA,
3101 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3102 int num_types = 4;
3103 int i;
3104 int c;
3105 int num_tolerated_disk_barrier_failures =
3106 (int)fs_info->fs_devices->num_devices;
3107
3108 for (i = 0; i < num_types; i++) {
3109 struct btrfs_space_info *tmp;
3110
3111 sinfo = NULL;
3112 rcu_read_lock();
3113 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3114 if (tmp->flags == types[i]) {
3115 sinfo = tmp;
3116 break;
3117 }
3118 }
3119 rcu_read_unlock();
3120
3121 if (!sinfo)
3122 continue;
3123
3124 down_read(&sinfo->groups_sem);
3125 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3126 if (!list_empty(&sinfo->block_groups[c])) {
3127 u64 flags;
3128
3129 btrfs_get_block_group_info(
3130 &sinfo->block_groups[c], &space);
3131 if (space.total_bytes == 0 ||
3132 space.used_bytes == 0)
3133 continue;
3134 flags = space.flags;
3135 /*
3136 * return
3137 * 0: if dup, single or RAID0 is configured for
3138 * any of metadata, system or data, else
3139 * 1: if RAID5 is configured, or if RAID1 or
3140 * RAID10 is configured and only two mirrors
3141 * are used, else
3142 * 2: if RAID6 is configured, else
3143 * num_mirrors - 1: if RAID1 or RAID10 is
3144 * configured and more than
3145 * 2 mirrors are used.
3146 */
3147 if (num_tolerated_disk_barrier_failures > 0 &&
3148 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3149 BTRFS_BLOCK_GROUP_RAID0)) ||
3150 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3151 == 0)))
3152 num_tolerated_disk_barrier_failures = 0;
3153 else if (num_tolerated_disk_barrier_failures > 1) {
3154 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3155 BTRFS_BLOCK_GROUP_RAID5 |
3156 BTRFS_BLOCK_GROUP_RAID10)) {
3157 num_tolerated_disk_barrier_failures = 1;
3158 } else if (flags &
3159 BTRFS_BLOCK_GROUP_RAID5) {
3160 num_tolerated_disk_barrier_failures = 2;
3161 }
3162 }
3163 }
3164 }
3165 up_read(&sinfo->groups_sem);
3166 }
3167
3168 return num_tolerated_disk_barrier_failures;
3169}
3170
3171int write_all_supers(struct btrfs_root *root, int max_mirrors)
3172{
3173 struct list_head *head;
3174 struct btrfs_device *dev;
3175 struct btrfs_super_block *sb;
3176 struct btrfs_dev_item *dev_item;
3177 int ret;
3178 int do_barriers;
3179 int max_errors;
3180 int total_errors = 0;
3181 u64 flags;
3182
3183 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3184 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3185 backup_super_roots(root->fs_info);
3186
3187 sb = root->fs_info->super_for_commit;
3188 dev_item = &sb->dev_item;
3189
3190 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3191 head = &root->fs_info->fs_devices->devices;
3192
3193 if (do_barriers) {
3194 ret = barrier_all_devices(root->fs_info);
3195 if (ret) {
3196 mutex_unlock(
3197 &root->fs_info->fs_devices->device_list_mutex);
3198 btrfs_error(root->fs_info, ret,
3199 "errors while submitting device barriers.");
3200 return ret;
3201 }
3202 }
3203
3204 list_for_each_entry_rcu(dev, head, dev_list) {
3205 if (!dev->bdev) {
3206 total_errors++;
3207 continue;
3208 }
3209 if (!dev->in_fs_metadata || !dev->writeable)
3210 continue;
3211
3212 btrfs_set_stack_device_generation(dev_item, 0);
3213 btrfs_set_stack_device_type(dev_item, dev->type);
3214 btrfs_set_stack_device_id(dev_item, dev->devid);
3215 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3216 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3217 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3218 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3219 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3220 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3221 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3222
3223 flags = btrfs_super_flags(sb);
3224 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3225
3226 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3227 if (ret)
3228 total_errors++;
3229 }
3230 if (total_errors > max_errors) {
3231 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3232 total_errors);
3233
3234 /* This shouldn't happen. FUA is masked off if unsupported */
3235 BUG();
3236 }
3237
3238 total_errors = 0;
3239 list_for_each_entry_rcu(dev, head, dev_list) {
3240 if (!dev->bdev)
3241 continue;
3242 if (!dev->in_fs_metadata || !dev->writeable)
3243 continue;
3244
3245 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3246 if (ret)
3247 total_errors++;
3248 }
3249 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3250 if (total_errors > max_errors) {
3251 btrfs_error(root->fs_info, -EIO,
3252 "%d errors while writing supers", total_errors);
3253 return -EIO;
3254 }
3255 return 0;
3256}
3257
3258int write_ctree_super(struct btrfs_trans_handle *trans,
3259 struct btrfs_root *root, int max_mirrors)
3260{
3261 int ret;
3262
3263 ret = write_all_supers(root, max_mirrors);
3264 return ret;
3265}
3266
3267void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3268{
3269 spin_lock(&fs_info->fs_roots_radix_lock);
3270 radix_tree_delete(&fs_info->fs_roots_radix,
3271 (unsigned long)root->root_key.objectid);
3272 spin_unlock(&fs_info->fs_roots_radix_lock);
3273
3274 if (btrfs_root_refs(&root->root_item) == 0)
3275 synchronize_srcu(&fs_info->subvol_srcu);
3276
3277 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3278 btrfs_free_log(NULL, root);
3279 btrfs_free_log_root_tree(NULL, fs_info);
3280 }
3281
3282 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3283 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3284 free_fs_root(root);
3285}
3286
3287static void free_fs_root(struct btrfs_root *root)
3288{
3289 iput(root->cache_inode);
3290 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3291 if (root->anon_dev)
3292 free_anon_bdev(root->anon_dev);
3293 free_extent_buffer(root->node);
3294 free_extent_buffer(root->commit_root);
3295 kfree(root->free_ino_ctl);
3296 kfree(root->free_ino_pinned);
3297 kfree(root->name);
3298 kfree(root);
3299}
3300
3301static void del_fs_roots(struct btrfs_fs_info *fs_info)
3302{
3303 int ret;
3304 struct btrfs_root *gang[8];
3305 int i;
3306
3307 while (!list_empty(&fs_info->dead_roots)) {
3308 gang[0] = list_entry(fs_info->dead_roots.next,
3309 struct btrfs_root, root_list);
3310 list_del(&gang[0]->root_list);
3311
3312 if (gang[0]->in_radix) {
3313 btrfs_free_fs_root(fs_info, gang[0]);
3314 } else {
3315 free_extent_buffer(gang[0]->node);
3316 free_extent_buffer(gang[0]->commit_root);
3317 kfree(gang[0]);
3318 }
3319 }
3320
3321 while (1) {
3322 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3323 (void **)gang, 0,
3324 ARRAY_SIZE(gang));
3325 if (!ret)
3326 break;
3327 for (i = 0; i < ret; i++)
3328 btrfs_free_fs_root(fs_info, gang[i]);
3329 }
3330}
3331
3332int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3333{
3334 u64 root_objectid = 0;
3335 struct btrfs_root *gang[8];
3336 int i;
3337 int ret;
3338
3339 while (1) {
3340 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3341 (void **)gang, root_objectid,
3342 ARRAY_SIZE(gang));
3343 if (!ret)
3344 break;
3345
3346 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3347 for (i = 0; i < ret; i++) {
3348 int err;
3349
3350 root_objectid = gang[i]->root_key.objectid;
3351 err = btrfs_orphan_cleanup(gang[i]);
3352 if (err)
3353 return err;
3354 }
3355 root_objectid++;
3356 }
3357 return 0;
3358}
3359
3360int btrfs_commit_super(struct btrfs_root *root)
3361{
3362 struct btrfs_trans_handle *trans;
3363 int ret;
3364
3365 mutex_lock(&root->fs_info->cleaner_mutex);
3366 btrfs_run_delayed_iputs(root);
3367 mutex_unlock(&root->fs_info->cleaner_mutex);
3368 wake_up_process(root->fs_info->cleaner_kthread);
3369
3370 /* wait until ongoing cleanup work done */
3371 down_write(&root->fs_info->cleanup_work_sem);
3372 up_write(&root->fs_info->cleanup_work_sem);
3373
3374 trans = btrfs_join_transaction(root);
3375 if (IS_ERR(trans))
3376 return PTR_ERR(trans);
3377 ret = btrfs_commit_transaction(trans, root);
3378 if (ret)
3379 return ret;
3380 /* run commit again to drop the original snapshot */
3381 trans = btrfs_join_transaction(root);
3382 if (IS_ERR(trans))
3383 return PTR_ERR(trans);
3384 ret = btrfs_commit_transaction(trans, root);
3385 if (ret)
3386 return ret;
3387 ret = btrfs_write_and_wait_transaction(NULL, root);
3388 if (ret) {
3389 btrfs_error(root->fs_info, ret,
3390 "Failed to sync btree inode to disk.");
3391 return ret;
3392 }
3393
3394 ret = write_ctree_super(NULL, root, 0);
3395 return ret;
3396}
3397
3398int close_ctree(struct btrfs_root *root)
3399{
3400 struct btrfs_fs_info *fs_info = root->fs_info;
3401 int ret;
3402
3403 fs_info->closing = 1;
3404 smp_mb();
3405
3406 /* pause restriper - we want to resume on mount */
3407 btrfs_pause_balance(fs_info);
3408
3409 btrfs_dev_replace_suspend_for_unmount(fs_info);
3410
3411 btrfs_scrub_cancel(fs_info);
3412
3413 /* wait for any defraggers to finish */
3414 wait_event(fs_info->transaction_wait,
3415 (atomic_read(&fs_info->defrag_running) == 0));
3416
3417 /* clear out the rbtree of defraggable inodes */
3418 btrfs_cleanup_defrag_inodes(fs_info);
3419
3420 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3421 ret = btrfs_commit_super(root);
3422 if (ret)
3423 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3424 }
3425
3426 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3427 btrfs_error_commit_super(root);
3428
3429 btrfs_put_block_group_cache(fs_info);
3430
3431 kthread_stop(fs_info->transaction_kthread);
3432 kthread_stop(fs_info->cleaner_kthread);
3433
3434 fs_info->closing = 2;
3435 smp_mb();
3436
3437 btrfs_free_qgroup_config(root->fs_info);
3438
3439 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3440 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3441 percpu_counter_sum(&fs_info->delalloc_bytes));
3442 }
3443
3444 free_root_pointers(fs_info, 1);
3445
3446 btrfs_free_block_groups(fs_info);
3447
3448 del_fs_roots(fs_info);
3449
3450 iput(fs_info->btree_inode);
3451
3452 btrfs_stop_all_workers(fs_info);
3453
3454#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3455 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3456 btrfsic_unmount(root, fs_info->fs_devices);
3457#endif
3458
3459 btrfs_close_devices(fs_info->fs_devices);
3460 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3461
3462 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3463 percpu_counter_destroy(&fs_info->delalloc_bytes);
3464 bdi_destroy(&fs_info->bdi);
3465 cleanup_srcu_struct(&fs_info->subvol_srcu);
3466
3467 btrfs_free_stripe_hash_table(fs_info);
3468
3469 return 0;
3470}
3471
3472int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3473 int atomic)
3474{
3475 int ret;
3476 struct inode *btree_inode = buf->pages[0]->mapping->host;
3477
3478 ret = extent_buffer_uptodate(buf);
3479 if (!ret)
3480 return ret;
3481
3482 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3483 parent_transid, atomic);
3484 if (ret == -EAGAIN)
3485 return ret;
3486 return !ret;
3487}
3488
3489int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3490{
3491 return set_extent_buffer_uptodate(buf);
3492}
3493
3494void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3495{
3496 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3497 u64 transid = btrfs_header_generation(buf);
3498 int was_dirty;
3499
3500 btrfs_assert_tree_locked(buf);
3501 if (transid != root->fs_info->generation)
3502 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3503 "found %llu running %llu\n",
3504 (unsigned long long)buf->start,
3505 (unsigned long long)transid,
3506 (unsigned long long)root->fs_info->generation);
3507 was_dirty = set_extent_buffer_dirty(buf);
3508 if (!was_dirty)
3509 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3510 buf->len,
3511 root->fs_info->dirty_metadata_batch);
3512}
3513
3514static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3515 int flush_delayed)
3516{
3517 /*
3518 * looks as though older kernels can get into trouble with
3519 * this code, they end up stuck in balance_dirty_pages forever
3520 */
3521 int ret;
3522
3523 if (current->flags & PF_MEMALLOC)
3524 return;
3525
3526 if (flush_delayed)
3527 btrfs_balance_delayed_items(root);
3528
3529 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3530 BTRFS_DIRTY_METADATA_THRESH);
3531 if (ret > 0) {
3532 balance_dirty_pages_ratelimited(
3533 root->fs_info->btree_inode->i_mapping);
3534 }
3535 return;
3536}
3537
3538void btrfs_btree_balance_dirty(struct btrfs_root *root)
3539{
3540 __btrfs_btree_balance_dirty(root, 1);
3541}
3542
3543void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3544{
3545 __btrfs_btree_balance_dirty(root, 0);
3546}
3547
3548int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3549{
3550 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3551 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3552}
3553
3554static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3555 int read_only)
3556{
3557 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3558 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3559 return -EINVAL;
3560 }
3561
3562 if (read_only)
3563 return 0;
3564
3565 return 0;
3566}
3567
3568void btrfs_error_commit_super(struct btrfs_root *root)
3569{
3570 mutex_lock(&root->fs_info->cleaner_mutex);
3571 btrfs_run_delayed_iputs(root);
3572 mutex_unlock(&root->fs_info->cleaner_mutex);
3573
3574 down_write(&root->fs_info->cleanup_work_sem);
3575 up_write(&root->fs_info->cleanup_work_sem);
3576
3577 /* cleanup FS via transaction */
3578 btrfs_cleanup_transaction(root);
3579}
3580
3581static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3582 struct btrfs_root *root)
3583{
3584 struct btrfs_inode *btrfs_inode;
3585 struct list_head splice;
3586
3587 INIT_LIST_HEAD(&splice);
3588
3589 mutex_lock(&root->fs_info->ordered_operations_mutex);
3590 spin_lock(&root->fs_info->ordered_extent_lock);
3591
3592 list_splice_init(&t->ordered_operations, &splice);
3593 while (!list_empty(&splice)) {
3594 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3595 ordered_operations);
3596
3597 list_del_init(&btrfs_inode->ordered_operations);
3598
3599 btrfs_invalidate_inodes(btrfs_inode->root);
3600 }
3601
3602 spin_unlock(&root->fs_info->ordered_extent_lock);
3603 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3604}
3605
3606static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3607{
3608 struct btrfs_ordered_extent *ordered;
3609
3610 spin_lock(&root->fs_info->ordered_extent_lock);
3611 /*
3612 * This will just short circuit the ordered completion stuff which will
3613 * make sure the ordered extent gets properly cleaned up.
3614 */
3615 list_for_each_entry(ordered, &root->fs_info->ordered_extents,
3616 root_extent_list)
3617 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3618 spin_unlock(&root->fs_info->ordered_extent_lock);
3619}
3620
3621int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3622 struct btrfs_root *root)
3623{
3624 struct rb_node *node;
3625 struct btrfs_delayed_ref_root *delayed_refs;
3626 struct btrfs_delayed_ref_node *ref;
3627 int ret = 0;
3628
3629 delayed_refs = &trans->delayed_refs;
3630
3631 spin_lock(&delayed_refs->lock);
3632 if (delayed_refs->num_entries == 0) {
3633 spin_unlock(&delayed_refs->lock);
3634 printk(KERN_INFO "delayed_refs has NO entry\n");
3635 return ret;
3636 }
3637
3638 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3639 struct btrfs_delayed_ref_head *head = NULL;
3640
3641 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3642 atomic_set(&ref->refs, 1);
3643 if (btrfs_delayed_ref_is_head(ref)) {
3644
3645 head = btrfs_delayed_node_to_head(ref);
3646 if (!mutex_trylock(&head->mutex)) {
3647 atomic_inc(&ref->refs);
3648 spin_unlock(&delayed_refs->lock);
3649
3650 /* Need to wait for the delayed ref to run */
3651 mutex_lock(&head->mutex);
3652 mutex_unlock(&head->mutex);
3653 btrfs_put_delayed_ref(ref);
3654
3655 spin_lock(&delayed_refs->lock);
3656 continue;
3657 }
3658
3659 btrfs_free_delayed_extent_op(head->extent_op);
3660 delayed_refs->num_heads--;
3661 if (list_empty(&head->cluster))
3662 delayed_refs->num_heads_ready--;
3663 list_del_init(&head->cluster);
3664 }
3665
3666 ref->in_tree = 0;
3667 rb_erase(&ref->rb_node, &delayed_refs->root);
3668 delayed_refs->num_entries--;
3669 if (head)
3670 mutex_unlock(&head->mutex);
3671 spin_unlock(&delayed_refs->lock);
3672 btrfs_put_delayed_ref(ref);
3673
3674 cond_resched();
3675 spin_lock(&delayed_refs->lock);
3676 }
3677
3678 spin_unlock(&delayed_refs->lock);
3679
3680 return ret;
3681}
3682
3683static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3684{
3685 struct btrfs_pending_snapshot *snapshot;
3686 struct list_head splice;
3687
3688 INIT_LIST_HEAD(&splice);
3689
3690 list_splice_init(&t->pending_snapshots, &splice);
3691
3692 while (!list_empty(&splice)) {
3693 snapshot = list_entry(splice.next,
3694 struct btrfs_pending_snapshot,
3695 list);
3696 snapshot->error = -ECANCELED;
3697 list_del_init(&snapshot->list);
3698 }
3699}
3700
3701static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3702{
3703 struct btrfs_inode *btrfs_inode;
3704 struct list_head splice;
3705
3706 INIT_LIST_HEAD(&splice);
3707
3708 spin_lock(&root->fs_info->delalloc_lock);
3709 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3710
3711 while (!list_empty(&splice)) {
3712 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3713 delalloc_inodes);
3714
3715 list_del_init(&btrfs_inode->delalloc_inodes);
3716 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3717 &btrfs_inode->runtime_flags);
3718
3719 btrfs_invalidate_inodes(btrfs_inode->root);
3720 }
3721
3722 spin_unlock(&root->fs_info->delalloc_lock);
3723}
3724
3725static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3726 struct extent_io_tree *dirty_pages,
3727 int mark)
3728{
3729 int ret;
3730 struct page *page;
3731 struct inode *btree_inode = root->fs_info->btree_inode;
3732 struct extent_buffer *eb;
3733 u64 start = 0;
3734 u64 end;
3735 u64 offset;
3736 unsigned long index;
3737
3738 while (1) {
3739 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3740 mark, NULL);
3741 if (ret)
3742 break;
3743
3744 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3745 while (start <= end) {
3746 index = start >> PAGE_CACHE_SHIFT;
3747 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3748 page = find_get_page(btree_inode->i_mapping, index);
3749 if (!page)
3750 continue;
3751 offset = page_offset(page);
3752
3753 spin_lock(&dirty_pages->buffer_lock);
3754 eb = radix_tree_lookup(
3755 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3756 offset >> PAGE_CACHE_SHIFT);
3757 spin_unlock(&dirty_pages->buffer_lock);
3758 if (eb)
3759 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3760 &eb->bflags);
3761 if (PageWriteback(page))
3762 end_page_writeback(page);
3763
3764 lock_page(page);
3765 if (PageDirty(page)) {
3766 clear_page_dirty_for_io(page);
3767 spin_lock_irq(&page->mapping->tree_lock);
3768 radix_tree_tag_clear(&page->mapping->page_tree,
3769 page_index(page),
3770 PAGECACHE_TAG_DIRTY);
3771 spin_unlock_irq(&page->mapping->tree_lock);
3772 }
3773
3774 unlock_page(page);
3775 page_cache_release(page);
3776 }
3777 }
3778
3779 return ret;
3780}
3781
3782static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3783 struct extent_io_tree *pinned_extents)
3784{
3785 struct extent_io_tree *unpin;
3786 u64 start;
3787 u64 end;
3788 int ret;
3789 bool loop = true;
3790
3791 unpin = pinned_extents;
3792again:
3793 while (1) {
3794 ret = find_first_extent_bit(unpin, 0, &start, &end,
3795 EXTENT_DIRTY, NULL);
3796 if (ret)
3797 break;
3798
3799 /* opt_discard */
3800 if (btrfs_test_opt(root, DISCARD))
3801 ret = btrfs_error_discard_extent(root, start,
3802 end + 1 - start,
3803 NULL);
3804
3805 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3806 btrfs_error_unpin_extent_range(root, start, end);
3807 cond_resched();
3808 }
3809
3810 if (loop) {
3811 if (unpin == &root->fs_info->freed_extents[0])
3812 unpin = &root->fs_info->freed_extents[1];
3813 else
3814 unpin = &root->fs_info->freed_extents[0];
3815 loop = false;
3816 goto again;
3817 }
3818
3819 return 0;
3820}
3821
3822void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3823 struct btrfs_root *root)
3824{
3825 btrfs_destroy_delayed_refs(cur_trans, root);
3826 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3827 cur_trans->dirty_pages.dirty_bytes);
3828
3829 /* FIXME: cleanup wait for commit */
3830 cur_trans->in_commit = 1;
3831 cur_trans->blocked = 1;
3832 wake_up(&root->fs_info->transaction_blocked_wait);
3833
3834 btrfs_evict_pending_snapshots(cur_trans);
3835
3836 cur_trans->blocked = 0;
3837 wake_up(&root->fs_info->transaction_wait);
3838
3839 cur_trans->commit_done = 1;
3840 wake_up(&cur_trans->commit_wait);
3841
3842 btrfs_destroy_delayed_inodes(root);
3843 btrfs_assert_delayed_root_empty(root);
3844
3845 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3846 EXTENT_DIRTY);
3847 btrfs_destroy_pinned_extent(root,
3848 root->fs_info->pinned_extents);
3849
3850 /*
3851 memset(cur_trans, 0, sizeof(*cur_trans));
3852 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3853 */
3854}
3855
3856int btrfs_cleanup_transaction(struct btrfs_root *root)
3857{
3858 struct btrfs_transaction *t;
3859 LIST_HEAD(list);
3860
3861 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3862
3863 spin_lock(&root->fs_info->trans_lock);
3864 list_splice_init(&root->fs_info->trans_list, &list);
3865 root->fs_info->trans_no_join = 1;
3866 spin_unlock(&root->fs_info->trans_lock);
3867
3868 while (!list_empty(&list)) {
3869 t = list_entry(list.next, struct btrfs_transaction, list);
3870
3871 btrfs_destroy_ordered_operations(t, root);
3872
3873 btrfs_destroy_ordered_extents(root);
3874
3875 btrfs_destroy_delayed_refs(t, root);
3876
3877 btrfs_block_rsv_release(root,
3878 &root->fs_info->trans_block_rsv,
3879 t->dirty_pages.dirty_bytes);
3880
3881 /* FIXME: cleanup wait for commit */
3882 t->in_commit = 1;
3883 t->blocked = 1;
3884 smp_mb();
3885 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3886 wake_up(&root->fs_info->transaction_blocked_wait);
3887
3888 btrfs_evict_pending_snapshots(t);
3889
3890 t->blocked = 0;
3891 smp_mb();
3892 if (waitqueue_active(&root->fs_info->transaction_wait))
3893 wake_up(&root->fs_info->transaction_wait);
3894
3895 t->commit_done = 1;
3896 smp_mb();
3897 if (waitqueue_active(&t->commit_wait))
3898 wake_up(&t->commit_wait);
3899
3900 btrfs_destroy_delayed_inodes(root);
3901 btrfs_assert_delayed_root_empty(root);
3902
3903 btrfs_destroy_delalloc_inodes(root);
3904
3905 spin_lock(&root->fs_info->trans_lock);
3906 root->fs_info->running_transaction = NULL;
3907 spin_unlock(&root->fs_info->trans_lock);
3908
3909 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3910 EXTENT_DIRTY);
3911
3912 btrfs_destroy_pinned_extent(root,
3913 root->fs_info->pinned_extents);
3914
3915 atomic_set(&t->use_count, 0);
3916 list_del_init(&t->list);
3917 memset(t, 0, sizeof(*t));
3918 kmem_cache_free(btrfs_transaction_cachep, t);
3919 }
3920
3921 spin_lock(&root->fs_info->trans_lock);
3922 root->fs_info->trans_no_join = 0;
3923 spin_unlock(&root->fs_info->trans_lock);
3924 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3925
3926 return 0;
3927}
3928
3929static struct extent_io_ops btree_extent_io_ops = {
3930 .readpage_end_io_hook = btree_readpage_end_io_hook,
3931 .readpage_io_failed_hook = btree_io_failed_hook,
3932 .submit_bio_hook = btree_submit_bio_hook,
3933 /* note we're sharing with inode.c for the merge bio hook */
3934 .merge_bio_hook = btrfs_merge_bio_hook,
3935};
Linux 6.x block layer and io_uring tree(s)