projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: only exclude supers in the range of our block group
[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 if (found_level >= BTRFS_MAX_LEVEL) {
617 btrfs_info(root->fs_info, "bad tree block level %d\n",
618 (int)btrfs_header_level(eb));
619 ret = -EIO;
620 goto err;
621 }
622
623 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
624 eb, found_level);
625
626 ret = csum_tree_block(root, eb, 1);
627 if (ret) {
628 ret = -EIO;
629 goto err;
630 }
631
632 /*
633 * If this is a leaf block and it is corrupt, set the corrupt bit so
634 * that we don't try and read the other copies of this block, just
635 * return -EIO.
636 */
637 if (found_level == 0 && check_leaf(root, eb)) {
638 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
639 ret = -EIO;
640 }
641
642 if (!ret)
643 set_extent_buffer_uptodate(eb);
644err:
645 if (reads_done &&
646 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
647 btree_readahead_hook(root, eb, eb->start, ret);
648
649 if (ret) {
650 /*
651 * our io error hook is going to dec the io pages
652 * again, we have to make sure it has something
653 * to decrement
654 */
655 atomic_inc(&eb->io_pages);
656 clear_extent_buffer_uptodate(eb);
657 }
658 free_extent_buffer(eb);
659out:
660 return ret;
661}
662
663static int btree_io_failed_hook(struct page *page, int failed_mirror)
664{
665 struct extent_buffer *eb;
666 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
667
668 eb = (struct extent_buffer *)page->private;
669 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
670 eb->read_mirror = failed_mirror;
671 atomic_dec(&eb->io_pages);
672 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
673 btree_readahead_hook(root, eb, eb->start, -EIO);
674 return -EIO; /* we fixed nothing */
675}
676
677static void end_workqueue_bio(struct bio *bio, int err)
678{
679 struct end_io_wq *end_io_wq = bio->bi_private;
680 struct btrfs_fs_info *fs_info;
681
682 fs_info = end_io_wq->info;
683 end_io_wq->error = err;
684 end_io_wq->work.func = end_workqueue_fn;
685 end_io_wq->work.flags = 0;
686
687 if (bio->bi_rw & REQ_WRITE) {
688 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
689 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
690 &end_io_wq->work);
691 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
692 btrfs_queue_worker(&fs_info->endio_freespace_worker,
693 &end_io_wq->work);
694 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
695 btrfs_queue_worker(&fs_info->endio_raid56_workers,
696 &end_io_wq->work);
697 else
698 btrfs_queue_worker(&fs_info->endio_write_workers,
699 &end_io_wq->work);
700 } else {
701 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
702 btrfs_queue_worker(&fs_info->endio_raid56_workers,
703 &end_io_wq->work);
704 else if (end_io_wq->metadata)
705 btrfs_queue_worker(&fs_info->endio_meta_workers,
706 &end_io_wq->work);
707 else
708 btrfs_queue_worker(&fs_info->endio_workers,
709 &end_io_wq->work);
710 }
711}
712
713/*
714 * For the metadata arg you want
715 *
716 * 0 - if data
717 * 1 - if normal metadta
718 * 2 - if writing to the free space cache area
719 * 3 - raid parity work
720 */
721int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
722 int metadata)
723{
724 struct end_io_wq *end_io_wq;
725 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
726 if (!end_io_wq)
727 return -ENOMEM;
728
729 end_io_wq->private = bio->bi_private;
730 end_io_wq->end_io = bio->bi_end_io;
731 end_io_wq->info = info;
732 end_io_wq->error = 0;
733 end_io_wq->bio = bio;
734 end_io_wq->metadata = metadata;
735
736 bio->bi_private = end_io_wq;
737 bio->bi_end_io = end_workqueue_bio;
738 return 0;
739}
740
741unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
742{
743 unsigned long limit = min_t(unsigned long,
744 info->workers.max_workers,
745 info->fs_devices->open_devices);
746 return 256 * limit;
747}
748
749static void run_one_async_start(struct btrfs_work *work)
750{
751 struct async_submit_bio *async;
752 int ret;
753
754 async = container_of(work, struct async_submit_bio, work);
755 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
756 async->mirror_num, async->bio_flags,
757 async->bio_offset);
758 if (ret)
759 async->error = ret;
760}
761
762static void run_one_async_done(struct btrfs_work *work)
763{
764 struct btrfs_fs_info *fs_info;
765 struct async_submit_bio *async;
766 int limit;
767
768 async = container_of(work, struct async_submit_bio, work);
769 fs_info = BTRFS_I(async->inode)->root->fs_info;
770
771 limit = btrfs_async_submit_limit(fs_info);
772 limit = limit * 2 / 3;
773
774 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
775 waitqueue_active(&fs_info->async_submit_wait))
776 wake_up(&fs_info->async_submit_wait);
777
778 /* If an error occured we just want to clean up the bio and move on */
779 if (async->error) {
780 bio_endio(async->bio, async->error);
781 return;
782 }
783
784 async->submit_bio_done(async->inode, async->rw, async->bio,
785 async->mirror_num, async->bio_flags,
786 async->bio_offset);
787}
788
789static void run_one_async_free(struct btrfs_work *work)
790{
791 struct async_submit_bio *async;
792
793 async = container_of(work, struct async_submit_bio, work);
794 kfree(async);
795}
796
797int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
798 int rw, struct bio *bio, int mirror_num,
799 unsigned long bio_flags,
800 u64 bio_offset,
801 extent_submit_bio_hook_t *submit_bio_start,
802 extent_submit_bio_hook_t *submit_bio_done)
803{
804 struct async_submit_bio *async;
805
806 async = kmalloc(sizeof(*async), GFP_NOFS);
807 if (!async)
808 return -ENOMEM;
809
810 async->inode = inode;
811 async->rw = rw;
812 async->bio = bio;
813 async->mirror_num = mirror_num;
814 async->submit_bio_start = submit_bio_start;
815 async->submit_bio_done = submit_bio_done;
816
817 async->work.func = run_one_async_start;
818 async->work.ordered_func = run_one_async_done;
819 async->work.ordered_free = run_one_async_free;
820
821 async->work.flags = 0;
822 async->bio_flags = bio_flags;
823 async->bio_offset = bio_offset;
824
825 async->error = 0;
826
827 atomic_inc(&fs_info->nr_async_submits);
828
829 if (rw & REQ_SYNC)
830 btrfs_set_work_high_prio(&async->work);
831
832 btrfs_queue_worker(&fs_info->workers, &async->work);
833
834 while (atomic_read(&fs_info->async_submit_draining) &&
835 atomic_read(&fs_info->nr_async_submits)) {
836 wait_event(fs_info->async_submit_wait,
837 (atomic_read(&fs_info->nr_async_submits) == 0));
838 }
839
840 return 0;
841}
842
843static int btree_csum_one_bio(struct bio *bio)
844{
845 struct bio_vec *bvec = bio->bi_io_vec;
846 int bio_index = 0;
847 struct btrfs_root *root;
848 int ret = 0;
849
850 WARN_ON(bio->bi_vcnt <= 0);
851 while (bio_index < bio->bi_vcnt) {
852 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
853 ret = csum_dirty_buffer(root, bvec->bv_page);
854 if (ret)
855 break;
856 bio_index++;
857 bvec++;
858 }
859 return ret;
860}
861
862static int __btree_submit_bio_start(struct inode *inode, int rw,
863 struct bio *bio, int mirror_num,
864 unsigned long bio_flags,
865 u64 bio_offset)
866{
867 /*
868 * when we're called for a write, we're already in the async
869 * submission context. Just jump into btrfs_map_bio
870 */
871 return btree_csum_one_bio(bio);
872}
873
874static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
875 int mirror_num, unsigned long bio_flags,
876 u64 bio_offset)
877{
878 int ret;
879
880 /*
881 * when we're called for a write, we're already in the async
882 * submission context. Just jump into btrfs_map_bio
883 */
884 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
885 if (ret)
886 bio_endio(bio, ret);
887 return ret;
888}
889
890static int check_async_write(struct inode *inode, unsigned long bio_flags)
891{
892 if (bio_flags & EXTENT_BIO_TREE_LOG)
893 return 0;
894#ifdef CONFIG_X86
895 if (cpu_has_xmm4_2)
896 return 0;
897#endif
898 return 1;
899}
900
901static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
902 int mirror_num, unsigned long bio_flags,
903 u64 bio_offset)
904{
905 int async = check_async_write(inode, bio_flags);
906 int ret;
907
908 if (!(rw & REQ_WRITE)) {
909 /*
910 * called for a read, do the setup so that checksum validation
911 * can happen in the async kernel threads
912 */
913 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
914 bio, 1);
915 if (ret)
916 goto out_w_error;
917 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
918 mirror_num, 0);
919 } else if (!async) {
920 ret = btree_csum_one_bio(bio);
921 if (ret)
922 goto out_w_error;
923 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
924 mirror_num, 0);
925 } else {
926 /*
927 * kthread helpers are used to submit writes so that
928 * checksumming can happen in parallel across all CPUs
929 */
930 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
931 inode, rw, bio, mirror_num, 0,
932 bio_offset,
933 __btree_submit_bio_start,
934 __btree_submit_bio_done);
935 }
936
937 if (ret) {
938out_w_error:
939 bio_endio(bio, ret);
940 }
941 return ret;
942}
943
944#ifdef CONFIG_MIGRATION
945static int btree_migratepage(struct address_space *mapping,
946 struct page *newpage, struct page *page,
947 enum migrate_mode mode)
948{
949 /*
950 * we can't safely write a btree page from here,
951 * we haven't done the locking hook
952 */
953 if (PageDirty(page))
954 return -EAGAIN;
955 /*
956 * Buffers may be managed in a filesystem specific way.
957 * We must have no buffers or drop them.
958 */
959 if (page_has_private(page) &&
960 !try_to_release_page(page, GFP_KERNEL))
961 return -EAGAIN;
962 return migrate_page(mapping, newpage, page, mode);
963}
964#endif
965
966
967static int btree_writepages(struct address_space *mapping,
968 struct writeback_control *wbc)
969{
970 struct extent_io_tree *tree;
971 struct btrfs_fs_info *fs_info;
972 int ret;
973
974 tree = &BTRFS_I(mapping->host)->io_tree;
975 if (wbc->sync_mode == WB_SYNC_NONE) {
976
977 if (wbc->for_kupdate)
978 return 0;
979
980 fs_info = BTRFS_I(mapping->host)->root->fs_info;
981 /* this is a bit racy, but that's ok */
982 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
983 BTRFS_DIRTY_METADATA_THRESH);
984 if (ret < 0)
985 return 0;
986 }
987 return btree_write_cache_pages(mapping, wbc);
988}
989
990static int btree_readpage(struct file *file, struct page *page)
991{
992 struct extent_io_tree *tree;
993 tree = &BTRFS_I(page->mapping->host)->io_tree;
994 return extent_read_full_page(tree, page, btree_get_extent, 0);
995}
996
997static int btree_releasepage(struct page *page, gfp_t gfp_flags)
998{
999 if (PageWriteback(page) || PageDirty(page))
1000 return 0;
1001 /*
1002 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
1003 * slab allocation from alloc_extent_state down the callchain where
1004 * it'd hit a BUG_ON as those flags are not allowed.
1005 */
1006 gfp_flags &= ~GFP_SLAB_BUG_MASK;
1007
1008 return try_release_extent_buffer(page, gfp_flags);
1009}
1010
1011static void btree_invalidatepage(struct page *page, unsigned long offset)
1012{
1013 struct extent_io_tree *tree;
1014 tree = &BTRFS_I(page->mapping->host)->io_tree;
1015 extent_invalidatepage(tree, page, offset);
1016 btree_releasepage(page, GFP_NOFS);
1017 if (PagePrivate(page)) {
1018 printk(KERN_WARNING "btrfs warning page private not zero "
1019 "on page %llu\n", (unsigned long long)page_offset(page));
1020 ClearPagePrivate(page);
1021 set_page_private(page, 0);
1022 page_cache_release(page);
1023 }
1024}
1025
1026static int btree_set_page_dirty(struct page *page)
1027{
1028#ifdef DEBUG
1029 struct extent_buffer *eb;
1030
1031 BUG_ON(!PagePrivate(page));
1032 eb = (struct extent_buffer *)page->private;
1033 BUG_ON(!eb);
1034 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1035 BUG_ON(!atomic_read(&eb->refs));
1036 btrfs_assert_tree_locked(eb);
1037#endif
1038 return __set_page_dirty_nobuffers(page);
1039}
1040
1041static const struct address_space_operations btree_aops = {
1042 .readpage = btree_readpage,
1043 .writepages = btree_writepages,
1044 .releasepage = btree_releasepage,
1045 .invalidatepage = btree_invalidatepage,
1046#ifdef CONFIG_MIGRATION
1047 .migratepage = btree_migratepage,
1048#endif
1049 .set_page_dirty = btree_set_page_dirty,
1050};
1051
1052int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1053 u64 parent_transid)
1054{
1055 struct extent_buffer *buf = NULL;
1056 struct inode *btree_inode = root->fs_info->btree_inode;
1057 int ret = 0;
1058
1059 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1060 if (!buf)
1061 return 0;
1062 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1063 buf, 0, WAIT_NONE, btree_get_extent, 0);
1064 free_extent_buffer(buf);
1065 return ret;
1066}
1067
1068int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1069 int mirror_num, struct extent_buffer **eb)
1070{
1071 struct extent_buffer *buf = NULL;
1072 struct inode *btree_inode = root->fs_info->btree_inode;
1073 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1074 int ret;
1075
1076 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1077 if (!buf)
1078 return 0;
1079
1080 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1081
1082 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1083 btree_get_extent, mirror_num);
1084 if (ret) {
1085 free_extent_buffer(buf);
1086 return ret;
1087 }
1088
1089 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1090 free_extent_buffer(buf);
1091 return -EIO;
1092 } else if (extent_buffer_uptodate(buf)) {
1093 *eb = buf;
1094 } else {
1095 free_extent_buffer(buf);
1096 }
1097 return 0;
1098}
1099
1100struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1101 u64 bytenr, u32 blocksize)
1102{
1103 struct inode *btree_inode = root->fs_info->btree_inode;
1104 struct extent_buffer *eb;
1105 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1106 bytenr, blocksize);
1107 return eb;
1108}
1109
1110struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1111 u64 bytenr, u32 blocksize)
1112{
1113 struct inode *btree_inode = root->fs_info->btree_inode;
1114 struct extent_buffer *eb;
1115
1116 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1117 bytenr, blocksize);
1118 return eb;
1119}
1120
1121
1122int btrfs_write_tree_block(struct extent_buffer *buf)
1123{
1124 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1125 buf->start + buf->len - 1);
1126}
1127
1128int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1129{
1130 return filemap_fdatawait_range(buf->pages[0]->mapping,
1131 buf->start, buf->start + buf->len - 1);
1132}
1133
1134struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1135 u32 blocksize, u64 parent_transid)
1136{
1137 struct extent_buffer *buf = NULL;
1138 int ret;
1139
1140 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1141 if (!buf)
1142 return NULL;
1143
1144 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1145 return buf;
1146
1147}
1148
1149void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1150 struct extent_buffer *buf)
1151{
1152 struct btrfs_fs_info *fs_info = root->fs_info;
1153
1154 if (btrfs_header_generation(buf) ==
1155 fs_info->running_transaction->transid) {
1156 btrfs_assert_tree_locked(buf);
1157
1158 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1159 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1160 -buf->len,
1161 fs_info->dirty_metadata_batch);
1162 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1163 btrfs_set_lock_blocking(buf);
1164 clear_extent_buffer_dirty(buf);
1165 }
1166 }
1167}
1168
1169static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1170 u32 stripesize, struct btrfs_root *root,
1171 struct btrfs_fs_info *fs_info,
1172 u64 objectid)
1173{
1174 root->node = NULL;
1175 root->commit_root = NULL;
1176 root->sectorsize = sectorsize;
1177 root->nodesize = nodesize;
1178 root->leafsize = leafsize;
1179 root->stripesize = stripesize;
1180 root->ref_cows = 0;
1181 root->track_dirty = 0;
1182 root->in_radix = 0;
1183 root->orphan_item_inserted = 0;
1184 root->orphan_cleanup_state = 0;
1185
1186 root->objectid = objectid;
1187 root->last_trans = 0;
1188 root->highest_objectid = 0;
1189 root->name = NULL;
1190 root->inode_tree = RB_ROOT;
1191 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1192 root->block_rsv = NULL;
1193 root->orphan_block_rsv = NULL;
1194
1195 INIT_LIST_HEAD(&root->dirty_list);
1196 INIT_LIST_HEAD(&root->root_list);
1197 INIT_LIST_HEAD(&root->logged_list[0]);
1198 INIT_LIST_HEAD(&root->logged_list[1]);
1199 spin_lock_init(&root->orphan_lock);
1200 spin_lock_init(&root->inode_lock);
1201 spin_lock_init(&root->accounting_lock);
1202 spin_lock_init(&root->log_extents_lock[0]);
1203 spin_lock_init(&root->log_extents_lock[1]);
1204 mutex_init(&root->objectid_mutex);
1205 mutex_init(&root->log_mutex);
1206 init_waitqueue_head(&root->log_writer_wait);
1207 init_waitqueue_head(&root->log_commit_wait[0]);
1208 init_waitqueue_head(&root->log_commit_wait[1]);
1209 atomic_set(&root->log_commit[0], 0);
1210 atomic_set(&root->log_commit[1], 0);
1211 atomic_set(&root->log_writers, 0);
1212 atomic_set(&root->log_batch, 0);
1213 atomic_set(&root->orphan_inodes, 0);
1214 root->log_transid = 0;
1215 root->last_log_commit = 0;
1216 extent_io_tree_init(&root->dirty_log_pages,
1217 fs_info->btree_inode->i_mapping);
1218
1219 memset(&root->root_key, 0, sizeof(root->root_key));
1220 memset(&root->root_item, 0, sizeof(root->root_item));
1221 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1222 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1223 root->defrag_trans_start = fs_info->generation;
1224 init_completion(&root->kobj_unregister);
1225 root->defrag_running = 0;
1226 root->root_key.objectid = objectid;
1227 root->anon_dev = 0;
1228
1229 spin_lock_init(&root->root_item_lock);
1230}
1231
1232static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1233 struct btrfs_fs_info *fs_info,
1234 u64 objectid,
1235 struct btrfs_root *root)
1236{
1237 int ret;
1238 u32 blocksize;
1239 u64 generation;
1240
1241 __setup_root(tree_root->nodesize, tree_root->leafsize,
1242 tree_root->sectorsize, tree_root->stripesize,
1243 root, fs_info, objectid);
1244 ret = btrfs_find_last_root(tree_root, objectid,
1245 &root->root_item, &root->root_key);
1246 if (ret > 0)
1247 return -ENOENT;
1248 else if (ret < 0)
1249 return ret;
1250
1251 generation = btrfs_root_generation(&root->root_item);
1252 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1253 root->commit_root = NULL;
1254 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1255 blocksize, generation);
1256 if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1257 free_extent_buffer(root->node);
1258 root->node = NULL;
1259 return -EIO;
1260 }
1261 root->commit_root = btrfs_root_node(root);
1262 return 0;
1263}
1264
1265static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1266{
1267 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1268 if (root)
1269 root->fs_info = fs_info;
1270 return root;
1271}
1272
1273struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1274 struct btrfs_fs_info *fs_info,
1275 u64 objectid)
1276{
1277 struct extent_buffer *leaf;
1278 struct btrfs_root *tree_root = fs_info->tree_root;
1279 struct btrfs_root *root;
1280 struct btrfs_key key;
1281 int ret = 0;
1282 u64 bytenr;
1283
1284 root = btrfs_alloc_root(fs_info);
1285 if (!root)
1286 return ERR_PTR(-ENOMEM);
1287
1288 __setup_root(tree_root->nodesize, tree_root->leafsize,
1289 tree_root->sectorsize, tree_root->stripesize,
1290 root, fs_info, objectid);
1291 root->root_key.objectid = objectid;
1292 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1293 root->root_key.offset = 0;
1294
1295 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1296 0, objectid, NULL, 0, 0, 0);
1297 if (IS_ERR(leaf)) {
1298 ret = PTR_ERR(leaf);
1299 leaf = NULL;
1300 goto fail;
1301 }
1302
1303 bytenr = leaf->start;
1304 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1305 btrfs_set_header_bytenr(leaf, leaf->start);
1306 btrfs_set_header_generation(leaf, trans->transid);
1307 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1308 btrfs_set_header_owner(leaf, objectid);
1309 root->node = leaf;
1310
1311 write_extent_buffer(leaf, fs_info->fsid,
1312 (unsigned long)btrfs_header_fsid(leaf),
1313 BTRFS_FSID_SIZE);
1314 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1315 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1316 BTRFS_UUID_SIZE);
1317 btrfs_mark_buffer_dirty(leaf);
1318
1319 root->commit_root = btrfs_root_node(root);
1320 root->track_dirty = 1;
1321
1322
1323 root->root_item.flags = 0;
1324 root->root_item.byte_limit = 0;
1325 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1326 btrfs_set_root_generation(&root->root_item, trans->transid);
1327 btrfs_set_root_level(&root->root_item, 0);
1328 btrfs_set_root_refs(&root->root_item, 1);
1329 btrfs_set_root_used(&root->root_item, leaf->len);
1330 btrfs_set_root_last_snapshot(&root->root_item, 0);
1331 btrfs_set_root_dirid(&root->root_item, 0);
1332 root->root_item.drop_level = 0;
1333
1334 key.objectid = objectid;
1335 key.type = BTRFS_ROOT_ITEM_KEY;
1336 key.offset = 0;
1337 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1338 if (ret)
1339 goto fail;
1340
1341 btrfs_tree_unlock(leaf);
1342
1343 return root;
1344
1345fail:
1346 if (leaf) {
1347 btrfs_tree_unlock(leaf);
1348 free_extent_buffer(leaf);
1349 }
1350 kfree(root);
1351
1352 return ERR_PTR(ret);
1353}
1354
1355static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1356 struct btrfs_fs_info *fs_info)
1357{
1358 struct btrfs_root *root;
1359 struct btrfs_root *tree_root = fs_info->tree_root;
1360 struct extent_buffer *leaf;
1361
1362 root = btrfs_alloc_root(fs_info);
1363 if (!root)
1364 return ERR_PTR(-ENOMEM);
1365
1366 __setup_root(tree_root->nodesize, tree_root->leafsize,
1367 tree_root->sectorsize, tree_root->stripesize,
1368 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1369
1370 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1371 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1372 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1373 /*
1374 * log trees do not get reference counted because they go away
1375 * before a real commit is actually done. They do store pointers
1376 * to file data extents, and those reference counts still get
1377 * updated (along with back refs to the log tree).
1378 */
1379 root->ref_cows = 0;
1380
1381 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1382 BTRFS_TREE_LOG_OBJECTID, NULL,
1383 0, 0, 0);
1384 if (IS_ERR(leaf)) {
1385 kfree(root);
1386 return ERR_CAST(leaf);
1387 }
1388
1389 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1390 btrfs_set_header_bytenr(leaf, leaf->start);
1391 btrfs_set_header_generation(leaf, trans->transid);
1392 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1393 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1394 root->node = leaf;
1395
1396 write_extent_buffer(root->node, root->fs_info->fsid,
1397 (unsigned long)btrfs_header_fsid(root->node),
1398 BTRFS_FSID_SIZE);
1399 btrfs_mark_buffer_dirty(root->node);
1400 btrfs_tree_unlock(root->node);
1401 return root;
1402}
1403
1404int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1405 struct btrfs_fs_info *fs_info)
1406{
1407 struct btrfs_root *log_root;
1408
1409 log_root = alloc_log_tree(trans, fs_info);
1410 if (IS_ERR(log_root))
1411 return PTR_ERR(log_root);
1412 WARN_ON(fs_info->log_root_tree);
1413 fs_info->log_root_tree = log_root;
1414 return 0;
1415}
1416
1417int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1418 struct btrfs_root *root)
1419{
1420 struct btrfs_root *log_root;
1421 struct btrfs_inode_item *inode_item;
1422
1423 log_root = alloc_log_tree(trans, root->fs_info);
1424 if (IS_ERR(log_root))
1425 return PTR_ERR(log_root);
1426
1427 log_root->last_trans = trans->transid;
1428 log_root->root_key.offset = root->root_key.objectid;
1429
1430 inode_item = &log_root->root_item.inode;
1431 inode_item->generation = cpu_to_le64(1);
1432 inode_item->size = cpu_to_le64(3);
1433 inode_item->nlink = cpu_to_le32(1);
1434 inode_item->nbytes = cpu_to_le64(root->leafsize);
1435 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1436
1437 btrfs_set_root_node(&log_root->root_item, log_root->node);
1438
1439 WARN_ON(root->log_root);
1440 root->log_root = log_root;
1441 root->log_transid = 0;
1442 root->last_log_commit = 0;
1443 return 0;
1444}
1445
1446struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1447 struct btrfs_key *location)
1448{
1449 struct btrfs_root *root;
1450 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1451 struct btrfs_path *path;
1452 struct extent_buffer *l;
1453 u64 generation;
1454 u32 blocksize;
1455 int ret = 0;
1456 int slot;
1457
1458 root = btrfs_alloc_root(fs_info);
1459 if (!root)
1460 return ERR_PTR(-ENOMEM);
1461 if (location->offset == (u64)-1) {
1462 ret = find_and_setup_root(tree_root, fs_info,
1463 location->objectid, root);
1464 if (ret) {
1465 kfree(root);
1466 return ERR_PTR(ret);
1467 }
1468 goto out;
1469 }
1470
1471 __setup_root(tree_root->nodesize, tree_root->leafsize,
1472 tree_root->sectorsize, tree_root->stripesize,
1473 root, fs_info, location->objectid);
1474
1475 path = btrfs_alloc_path();
1476 if (!path) {
1477 kfree(root);
1478 return ERR_PTR(-ENOMEM);
1479 }
1480 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1481 if (ret == 0) {
1482 l = path->nodes[0];
1483 slot = path->slots[0];
1484 btrfs_read_root_item(tree_root, l, slot, &root->root_item);
1485 memcpy(&root->root_key, location, sizeof(*location));
1486 }
1487 btrfs_free_path(path);
1488 if (ret) {
1489 kfree(root);
1490 if (ret > 0)
1491 ret = -ENOENT;
1492 return ERR_PTR(ret);
1493 }
1494
1495 generation = btrfs_root_generation(&root->root_item);
1496 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1497 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1498 blocksize, generation);
1499 root->commit_root = btrfs_root_node(root);
1500 BUG_ON(!root->node); /* -ENOMEM */
1501out:
1502 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1503 root->ref_cows = 1;
1504 btrfs_check_and_init_root_item(&root->root_item);
1505 }
1506
1507 return root;
1508}
1509
1510struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1511 struct btrfs_key *location)
1512{
1513 struct btrfs_root *root;
1514 int ret;
1515
1516 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1517 return fs_info->tree_root;
1518 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1519 return fs_info->extent_root;
1520 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1521 return fs_info->chunk_root;
1522 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1523 return fs_info->dev_root;
1524 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1525 return fs_info->csum_root;
1526 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1527 return fs_info->quota_root ? fs_info->quota_root :
1528 ERR_PTR(-ENOENT);
1529again:
1530 spin_lock(&fs_info->fs_roots_radix_lock);
1531 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1532 (unsigned long)location->objectid);
1533 spin_unlock(&fs_info->fs_roots_radix_lock);
1534 if (root)
1535 return root;
1536
1537 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1538 if (IS_ERR(root))
1539 return root;
1540
1541 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1542 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1543 GFP_NOFS);
1544 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1545 ret = -ENOMEM;
1546 goto fail;
1547 }
1548
1549 btrfs_init_free_ino_ctl(root);
1550 mutex_init(&root->fs_commit_mutex);
1551 spin_lock_init(&root->cache_lock);
1552 init_waitqueue_head(&root->cache_wait);
1553
1554 ret = get_anon_bdev(&root->anon_dev);
1555 if (ret)
1556 goto fail;
1557
1558 if (btrfs_root_refs(&root->root_item) == 0) {
1559 ret = -ENOENT;
1560 goto fail;
1561 }
1562
1563 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1564 if (ret < 0)
1565 goto fail;
1566 if (ret == 0)
1567 root->orphan_item_inserted = 1;
1568
1569 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1570 if (ret)
1571 goto fail;
1572
1573 spin_lock(&fs_info->fs_roots_radix_lock);
1574 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1575 (unsigned long)root->root_key.objectid,
1576 root);
1577 if (ret == 0)
1578 root->in_radix = 1;
1579
1580 spin_unlock(&fs_info->fs_roots_radix_lock);
1581 radix_tree_preload_end();
1582 if (ret) {
1583 if (ret == -EEXIST) {
1584 free_fs_root(root);
1585 goto again;
1586 }
1587 goto fail;
1588 }
1589
1590 ret = btrfs_find_dead_roots(fs_info->tree_root,
1591 root->root_key.objectid);
1592 WARN_ON(ret);
1593 return root;
1594fail:
1595 free_fs_root(root);
1596 return ERR_PTR(ret);
1597}
1598
1599static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1600{
1601 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1602 int ret = 0;
1603 struct btrfs_device *device;
1604 struct backing_dev_info *bdi;
1605
1606 rcu_read_lock();
1607 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1608 if (!device->bdev)
1609 continue;
1610 bdi = blk_get_backing_dev_info(device->bdev);
1611 if (bdi && bdi_congested(bdi, bdi_bits)) {
1612 ret = 1;
1613 break;
1614 }
1615 }
1616 rcu_read_unlock();
1617 return ret;
1618}
1619
1620/*
1621 * If this fails, caller must call bdi_destroy() to get rid of the
1622 * bdi again.
1623 */
1624static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1625{
1626 int err;
1627
1628 bdi->capabilities = BDI_CAP_MAP_COPY;
1629 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1630 if (err)
1631 return err;
1632
1633 bdi->ra_pages = default_backing_dev_info.ra_pages;
1634 bdi->congested_fn = btrfs_congested_fn;
1635 bdi->congested_data = info;
1636 return 0;
1637}
1638
1639/*
1640 * called by the kthread helper functions to finally call the bio end_io
1641 * functions. This is where read checksum verification actually happens
1642 */
1643static void end_workqueue_fn(struct btrfs_work *work)
1644{
1645 struct bio *bio;
1646 struct end_io_wq *end_io_wq;
1647 struct btrfs_fs_info *fs_info;
1648 int error;
1649
1650 end_io_wq = container_of(work, struct end_io_wq, work);
1651 bio = end_io_wq->bio;
1652 fs_info = end_io_wq->info;
1653
1654 error = end_io_wq->error;
1655 bio->bi_private = end_io_wq->private;
1656 bio->bi_end_io = end_io_wq->end_io;
1657 kfree(end_io_wq);
1658 bio_endio(bio, error);
1659}
1660
1661static int cleaner_kthread(void *arg)
1662{
1663 struct btrfs_root *root = arg;
1664
1665 do {
1666 int again = 0;
1667
1668 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1669 down_read_trylock(&root->fs_info->sb->s_umount)) {
1670 if (mutex_trylock(&root->fs_info->cleaner_mutex)) {
1671 btrfs_run_delayed_iputs(root);
1672 again = btrfs_clean_one_deleted_snapshot(root);
1673 mutex_unlock(&root->fs_info->cleaner_mutex);
1674 }
1675 btrfs_run_defrag_inodes(root->fs_info);
1676 up_read(&root->fs_info->sb->s_umount);
1677 }
1678
1679 if (!try_to_freeze() && !again) {
1680 set_current_state(TASK_INTERRUPTIBLE);
1681 if (!kthread_should_stop())
1682 schedule();
1683 __set_current_state(TASK_RUNNING);
1684 }
1685 } while (!kthread_should_stop());
1686 return 0;
1687}
1688
1689static int transaction_kthread(void *arg)
1690{
1691 struct btrfs_root *root = arg;
1692 struct btrfs_trans_handle *trans;
1693 struct btrfs_transaction *cur;
1694 u64 transid;
1695 unsigned long now;
1696 unsigned long delay;
1697 bool cannot_commit;
1698
1699 do {
1700 cannot_commit = false;
1701 delay = HZ * 30;
1702 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1703
1704 spin_lock(&root->fs_info->trans_lock);
1705 cur = root->fs_info->running_transaction;
1706 if (!cur) {
1707 spin_unlock(&root->fs_info->trans_lock);
1708 goto sleep;
1709 }
1710
1711 now = get_seconds();
1712 if (!cur->blocked &&
1713 (now < cur->start_time || now - cur->start_time < 30)) {
1714 spin_unlock(&root->fs_info->trans_lock);
1715 delay = HZ * 5;
1716 goto sleep;
1717 }
1718 transid = cur->transid;
1719 spin_unlock(&root->fs_info->trans_lock);
1720
1721 /* If the file system is aborted, this will always fail. */
1722 trans = btrfs_attach_transaction(root);
1723 if (IS_ERR(trans)) {
1724 if (PTR_ERR(trans) != -ENOENT)
1725 cannot_commit = true;
1726 goto sleep;
1727 }
1728 if (transid == trans->transid) {
1729 btrfs_commit_transaction(trans, root);
1730 } else {
1731 btrfs_end_transaction(trans, root);
1732 }
1733sleep:
1734 wake_up_process(root->fs_info->cleaner_kthread);
1735 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1736
1737 if (!try_to_freeze()) {
1738 set_current_state(TASK_INTERRUPTIBLE);
1739 if (!kthread_should_stop() &&
1740 (!btrfs_transaction_blocked(root->fs_info) ||
1741 cannot_commit))
1742 schedule_timeout(delay);
1743 __set_current_state(TASK_RUNNING);
1744 }
1745 } while (!kthread_should_stop());
1746 return 0;
1747}
1748
1749/*
1750 * this will find the highest generation in the array of
1751 * root backups. The index of the highest array is returned,
1752 * or -1 if we can't find anything.
1753 *
1754 * We check to make sure the array is valid by comparing the
1755 * generation of the latest root in the array with the generation
1756 * in the super block. If they don't match we pitch it.
1757 */
1758static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1759{
1760 u64 cur;
1761 int newest_index = -1;
1762 struct btrfs_root_backup *root_backup;
1763 int i;
1764
1765 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1766 root_backup = info->super_copy->super_roots + i;
1767 cur = btrfs_backup_tree_root_gen(root_backup);
1768 if (cur == newest_gen)
1769 newest_index = i;
1770 }
1771
1772 /* check to see if we actually wrapped around */
1773 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1774 root_backup = info->super_copy->super_roots;
1775 cur = btrfs_backup_tree_root_gen(root_backup);
1776 if (cur == newest_gen)
1777 newest_index = 0;
1778 }
1779 return newest_index;
1780}
1781
1782
1783/*
1784 * find the oldest backup so we know where to store new entries
1785 * in the backup array. This will set the backup_root_index
1786 * field in the fs_info struct
1787 */
1788static void find_oldest_super_backup(struct btrfs_fs_info *info,
1789 u64 newest_gen)
1790{
1791 int newest_index = -1;
1792
1793 newest_index = find_newest_super_backup(info, newest_gen);
1794 /* if there was garbage in there, just move along */
1795 if (newest_index == -1) {
1796 info->backup_root_index = 0;
1797 } else {
1798 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1799 }
1800}
1801
1802/*
1803 * copy all the root pointers into the super backup array.
1804 * this will bump the backup pointer by one when it is
1805 * done
1806 */
1807static void backup_super_roots(struct btrfs_fs_info *info)
1808{
1809 int next_backup;
1810 struct btrfs_root_backup *root_backup;
1811 int last_backup;
1812
1813 next_backup = info->backup_root_index;
1814 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1815 BTRFS_NUM_BACKUP_ROOTS;
1816
1817 /*
1818 * just overwrite the last backup if we're at the same generation
1819 * this happens only at umount
1820 */
1821 root_backup = info->super_for_commit->super_roots + last_backup;
1822 if (btrfs_backup_tree_root_gen(root_backup) ==
1823 btrfs_header_generation(info->tree_root->node))
1824 next_backup = last_backup;
1825
1826 root_backup = info->super_for_commit->super_roots + next_backup;
1827
1828 /*
1829 * make sure all of our padding and empty slots get zero filled
1830 * regardless of which ones we use today
1831 */
1832 memset(root_backup, 0, sizeof(*root_backup));
1833
1834 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1835
1836 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1837 btrfs_set_backup_tree_root_gen(root_backup,
1838 btrfs_header_generation(info->tree_root->node));
1839
1840 btrfs_set_backup_tree_root_level(root_backup,
1841 btrfs_header_level(info->tree_root->node));
1842
1843 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1844 btrfs_set_backup_chunk_root_gen(root_backup,
1845 btrfs_header_generation(info->chunk_root->node));
1846 btrfs_set_backup_chunk_root_level(root_backup,
1847 btrfs_header_level(info->chunk_root->node));
1848
1849 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1850 btrfs_set_backup_extent_root_gen(root_backup,
1851 btrfs_header_generation(info->extent_root->node));
1852 btrfs_set_backup_extent_root_level(root_backup,
1853 btrfs_header_level(info->extent_root->node));
1854
1855 /*
1856 * we might commit during log recovery, which happens before we set
1857 * the fs_root. Make sure it is valid before we fill it in.
1858 */
1859 if (info->fs_root && info->fs_root->node) {
1860 btrfs_set_backup_fs_root(root_backup,
1861 info->fs_root->node->start);
1862 btrfs_set_backup_fs_root_gen(root_backup,
1863 btrfs_header_generation(info->fs_root->node));
1864 btrfs_set_backup_fs_root_level(root_backup,
1865 btrfs_header_level(info->fs_root->node));
1866 }
1867
1868 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1869 btrfs_set_backup_dev_root_gen(root_backup,
1870 btrfs_header_generation(info->dev_root->node));
1871 btrfs_set_backup_dev_root_level(root_backup,
1872 btrfs_header_level(info->dev_root->node));
1873
1874 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1875 btrfs_set_backup_csum_root_gen(root_backup,
1876 btrfs_header_generation(info->csum_root->node));
1877 btrfs_set_backup_csum_root_level(root_backup,
1878 btrfs_header_level(info->csum_root->node));
1879
1880 btrfs_set_backup_total_bytes(root_backup,
1881 btrfs_super_total_bytes(info->super_copy));
1882 btrfs_set_backup_bytes_used(root_backup,
1883 btrfs_super_bytes_used(info->super_copy));
1884 btrfs_set_backup_num_devices(root_backup,
1885 btrfs_super_num_devices(info->super_copy));
1886
1887 /*
1888 * if we don't copy this out to the super_copy, it won't get remembered
1889 * for the next commit
1890 */
1891 memcpy(&info->super_copy->super_roots,
1892 &info->super_for_commit->super_roots,
1893 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1894}
1895
1896/*
1897 * this copies info out of the root backup array and back into
1898 * the in-memory super block. It is meant to help iterate through
1899 * the array, so you send it the number of backups you've already
1900 * tried and the last backup index you used.
1901 *
1902 * this returns -1 when it has tried all the backups
1903 */
1904static noinline int next_root_backup(struct btrfs_fs_info *info,
1905 struct btrfs_super_block *super,
1906 int *num_backups_tried, int *backup_index)
1907{
1908 struct btrfs_root_backup *root_backup;
1909 int newest = *backup_index;
1910
1911 if (*num_backups_tried == 0) {
1912 u64 gen = btrfs_super_generation(super);
1913
1914 newest = find_newest_super_backup(info, gen);
1915 if (newest == -1)
1916 return -1;
1917
1918 *backup_index = newest;
1919 *num_backups_tried = 1;
1920 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1921 /* we've tried all the backups, all done */
1922 return -1;
1923 } else {
1924 /* jump to the next oldest backup */
1925 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1926 BTRFS_NUM_BACKUP_ROOTS;
1927 *backup_index = newest;
1928 *num_backups_tried += 1;
1929 }
1930 root_backup = super->super_roots + newest;
1931
1932 btrfs_set_super_generation(super,
1933 btrfs_backup_tree_root_gen(root_backup));
1934 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1935 btrfs_set_super_root_level(super,
1936 btrfs_backup_tree_root_level(root_backup));
1937 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1938
1939 /*
1940 * fixme: the total bytes and num_devices need to match or we should
1941 * need a fsck
1942 */
1943 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1944 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1945 return 0;
1946}
1947
1948/* helper to cleanup workers */
1949static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1950{
1951 btrfs_stop_workers(&fs_info->generic_worker);
1952 btrfs_stop_workers(&fs_info->fixup_workers);
1953 btrfs_stop_workers(&fs_info->delalloc_workers);
1954 btrfs_stop_workers(&fs_info->workers);
1955 btrfs_stop_workers(&fs_info->endio_workers);
1956 btrfs_stop_workers(&fs_info->endio_meta_workers);
1957 btrfs_stop_workers(&fs_info->endio_raid56_workers);
1958 btrfs_stop_workers(&fs_info->rmw_workers);
1959 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1960 btrfs_stop_workers(&fs_info->endio_write_workers);
1961 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1962 btrfs_stop_workers(&fs_info->submit_workers);
1963 btrfs_stop_workers(&fs_info->delayed_workers);
1964 btrfs_stop_workers(&fs_info->caching_workers);
1965 btrfs_stop_workers(&fs_info->readahead_workers);
1966 btrfs_stop_workers(&fs_info->flush_workers);
1967}
1968
1969/* helper to cleanup tree roots */
1970static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1971{
1972 free_extent_buffer(info->tree_root->node);
1973 free_extent_buffer(info->tree_root->commit_root);
1974 free_extent_buffer(info->dev_root->node);
1975 free_extent_buffer(info->dev_root->commit_root);
1976 free_extent_buffer(info->extent_root->node);
1977 free_extent_buffer(info->extent_root->commit_root);
1978 free_extent_buffer(info->csum_root->node);
1979 free_extent_buffer(info->csum_root->commit_root);
1980 if (info->quota_root) {
1981 free_extent_buffer(info->quota_root->node);
1982 free_extent_buffer(info->quota_root->commit_root);
1983 }
1984
1985 info->tree_root->node = NULL;
1986 info->tree_root->commit_root = NULL;
1987 info->dev_root->node = NULL;
1988 info->dev_root->commit_root = NULL;
1989 info->extent_root->node = NULL;
1990 info->extent_root->commit_root = NULL;
1991 info->csum_root->node = NULL;
1992 info->csum_root->commit_root = NULL;
1993 if (info->quota_root) {
1994 info->quota_root->node = NULL;
1995 info->quota_root->commit_root = NULL;
1996 }
1997
1998 if (chunk_root) {
1999 free_extent_buffer(info->chunk_root->node);
2000 free_extent_buffer(info->chunk_root->commit_root);
2001 info->chunk_root->node = NULL;
2002 info->chunk_root->commit_root = NULL;
2003 }
2004}
2005
2006
2007int open_ctree(struct super_block *sb,
2008 struct btrfs_fs_devices *fs_devices,
2009 char *options)
2010{
2011 u32 sectorsize;
2012 u32 nodesize;
2013 u32 leafsize;
2014 u32 blocksize;
2015 u32 stripesize;
2016 u64 generation;
2017 u64 features;
2018 struct btrfs_key location;
2019 struct buffer_head *bh;
2020 struct btrfs_super_block *disk_super;
2021 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2022 struct btrfs_root *tree_root;
2023 struct btrfs_root *extent_root;
2024 struct btrfs_root *csum_root;
2025 struct btrfs_root *chunk_root;
2026 struct btrfs_root *dev_root;
2027 struct btrfs_root *quota_root;
2028 struct btrfs_root *log_tree_root;
2029 int ret;
2030 int err = -EINVAL;
2031 int num_backups_tried = 0;
2032 int backup_index = 0;
2033
2034 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2035 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
2036 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
2037 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2038 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
2039 quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
2040
2041 if (!tree_root || !extent_root || !csum_root ||
2042 !chunk_root || !dev_root || !quota_root) {
2043 err = -ENOMEM;
2044 goto fail;
2045 }
2046
2047 ret = init_srcu_struct(&fs_info->subvol_srcu);
2048 if (ret) {
2049 err = ret;
2050 goto fail;
2051 }
2052
2053 ret = setup_bdi(fs_info, &fs_info->bdi);
2054 if (ret) {
2055 err = ret;
2056 goto fail_srcu;
2057 }
2058
2059 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2060 if (ret) {
2061 err = ret;
2062 goto fail_bdi;
2063 }
2064 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2065 (1 + ilog2(nr_cpu_ids));
2066
2067 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2068 if (ret) {
2069 err = ret;
2070 goto fail_dirty_metadata_bytes;
2071 }
2072
2073 fs_info->btree_inode = new_inode(sb);
2074 if (!fs_info->btree_inode) {
2075 err = -ENOMEM;
2076 goto fail_delalloc_bytes;
2077 }
2078
2079 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2080
2081 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2082 INIT_LIST_HEAD(&fs_info->trans_list);
2083 INIT_LIST_HEAD(&fs_info->dead_roots);
2084 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2085 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
2086 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2087 spin_lock_init(&fs_info->delalloc_lock);
2088 spin_lock_init(&fs_info->trans_lock);
2089 spin_lock_init(&fs_info->fs_roots_radix_lock);
2090 spin_lock_init(&fs_info->delayed_iput_lock);
2091 spin_lock_init(&fs_info->defrag_inodes_lock);
2092 spin_lock_init(&fs_info->free_chunk_lock);
2093 spin_lock_init(&fs_info->tree_mod_seq_lock);
2094 spin_lock_init(&fs_info->super_lock);
2095 rwlock_init(&fs_info->tree_mod_log_lock);
2096 mutex_init(&fs_info->reloc_mutex);
2097 seqlock_init(&fs_info->profiles_lock);
2098
2099 init_completion(&fs_info->kobj_unregister);
2100 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2101 INIT_LIST_HEAD(&fs_info->space_info);
2102 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2103 btrfs_mapping_init(&fs_info->mapping_tree);
2104 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2105 BTRFS_BLOCK_RSV_GLOBAL);
2106 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2107 BTRFS_BLOCK_RSV_DELALLOC);
2108 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2109 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2110 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2111 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2112 BTRFS_BLOCK_RSV_DELOPS);
2113 atomic_set(&fs_info->nr_async_submits, 0);
2114 atomic_set(&fs_info->async_delalloc_pages, 0);
2115 atomic_set(&fs_info->async_submit_draining, 0);
2116 atomic_set(&fs_info->nr_async_bios, 0);
2117 atomic_set(&fs_info->defrag_running, 0);
2118 atomic_set(&fs_info->tree_mod_seq, 0);
2119 fs_info->sb = sb;
2120 fs_info->max_inline = 8192 * 1024;
2121 fs_info->metadata_ratio = 0;
2122 fs_info->defrag_inodes = RB_ROOT;
2123 fs_info->trans_no_join = 0;
2124 fs_info->free_chunk_space = 0;
2125 fs_info->tree_mod_log = RB_ROOT;
2126
2127 /* readahead state */
2128 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2129 spin_lock_init(&fs_info->reada_lock);
2130
2131 fs_info->thread_pool_size = min_t(unsigned long,
2132 num_online_cpus() + 2, 8);
2133
2134 INIT_LIST_HEAD(&fs_info->ordered_extents);
2135 spin_lock_init(&fs_info->ordered_extent_lock);
2136 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2137 GFP_NOFS);
2138 if (!fs_info->delayed_root) {
2139 err = -ENOMEM;
2140 goto fail_iput;
2141 }
2142 btrfs_init_delayed_root(fs_info->delayed_root);
2143
2144 mutex_init(&fs_info->scrub_lock);
2145 atomic_set(&fs_info->scrubs_running, 0);
2146 atomic_set(&fs_info->scrub_pause_req, 0);
2147 atomic_set(&fs_info->scrubs_paused, 0);
2148 atomic_set(&fs_info->scrub_cancel_req, 0);
2149 init_waitqueue_head(&fs_info->scrub_pause_wait);
2150 init_rwsem(&fs_info->scrub_super_lock);
2151 fs_info->scrub_workers_refcnt = 0;
2152#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2153 fs_info->check_integrity_print_mask = 0;
2154#endif
2155
2156 spin_lock_init(&fs_info->balance_lock);
2157 mutex_init(&fs_info->balance_mutex);
2158 atomic_set(&fs_info->balance_running, 0);
2159 atomic_set(&fs_info->balance_pause_req, 0);
2160 atomic_set(&fs_info->balance_cancel_req, 0);
2161 fs_info->balance_ctl = NULL;
2162 init_waitqueue_head(&fs_info->balance_wait_q);
2163
2164 sb->s_blocksize = 4096;
2165 sb->s_blocksize_bits = blksize_bits(4096);
2166 sb->s_bdi = &fs_info->bdi;
2167
2168 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2169 set_nlink(fs_info->btree_inode, 1);
2170 /*
2171 * we set the i_size on the btree inode to the max possible int.
2172 * the real end of the address space is determined by all of
2173 * the devices in the system
2174 */
2175 fs_info->btree_inode->i_size = OFFSET_MAX;
2176 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2177 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2178
2179 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2180 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2181 fs_info->btree_inode->i_mapping);
2182 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2183 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2184
2185 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2186
2187 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2188 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2189 sizeof(struct btrfs_key));
2190 set_bit(BTRFS_INODE_DUMMY,
2191 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2192 insert_inode_hash(fs_info->btree_inode);
2193
2194 spin_lock_init(&fs_info->block_group_cache_lock);
2195 fs_info->block_group_cache_tree = RB_ROOT;
2196 fs_info->first_logical_byte = (u64)-1;
2197
2198 extent_io_tree_init(&fs_info->freed_extents[0],
2199 fs_info->btree_inode->i_mapping);
2200 extent_io_tree_init(&fs_info->freed_extents[1],
2201 fs_info->btree_inode->i_mapping);
2202 fs_info->pinned_extents = &fs_info->freed_extents[0];
2203 fs_info->do_barriers = 1;
2204
2205
2206 mutex_init(&fs_info->ordered_operations_mutex);
2207 mutex_init(&fs_info->tree_log_mutex);
2208 mutex_init(&fs_info->chunk_mutex);
2209 mutex_init(&fs_info->transaction_kthread_mutex);
2210 mutex_init(&fs_info->cleaner_mutex);
2211 mutex_init(&fs_info->volume_mutex);
2212 init_rwsem(&fs_info->extent_commit_sem);
2213 init_rwsem(&fs_info->cleanup_work_sem);
2214 init_rwsem(&fs_info->subvol_sem);
2215 fs_info->dev_replace.lock_owner = 0;
2216 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2217 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2218 mutex_init(&fs_info->dev_replace.lock_management_lock);
2219 mutex_init(&fs_info->dev_replace.lock);
2220
2221 spin_lock_init(&fs_info->qgroup_lock);
2222 mutex_init(&fs_info->qgroup_ioctl_lock);
2223 fs_info->qgroup_tree = RB_ROOT;
2224 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2225 fs_info->qgroup_seq = 1;
2226 fs_info->quota_enabled = 0;
2227 fs_info->pending_quota_state = 0;
2228
2229 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2230 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2231
2232 init_waitqueue_head(&fs_info->transaction_throttle);
2233 init_waitqueue_head(&fs_info->transaction_wait);
2234 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2235 init_waitqueue_head(&fs_info->async_submit_wait);
2236
2237 ret = btrfs_alloc_stripe_hash_table(fs_info);
2238 if (ret) {
2239 err = ret;
2240 goto fail_alloc;
2241 }
2242
2243 __setup_root(4096, 4096, 4096, 4096, tree_root,
2244 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2245
2246 invalidate_bdev(fs_devices->latest_bdev);
2247 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2248 if (!bh) {
2249 err = -EINVAL;
2250 goto fail_alloc;
2251 }
2252
2253 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2254 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2255 sizeof(*fs_info->super_for_commit));
2256 brelse(bh);
2257
2258 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2259
2260 disk_super = fs_info->super_copy;
2261 if (!btrfs_super_root(disk_super))
2262 goto fail_alloc;
2263
2264 /* check FS state, whether FS is broken. */
2265 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2266 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2267
2268 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2269 if (ret) {
2270 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2271 err = ret;
2272 goto fail_alloc;
2273 }
2274
2275 /*
2276 * run through our array of backup supers and setup
2277 * our ring pointer to the oldest one
2278 */
2279 generation = btrfs_super_generation(disk_super);
2280 find_oldest_super_backup(fs_info, generation);
2281
2282 /*
2283 * In the long term, we'll store the compression type in the super
2284 * block, and it'll be used for per file compression control.
2285 */
2286 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2287
2288 ret = btrfs_parse_options(tree_root, options);
2289 if (ret) {
2290 err = ret;
2291 goto fail_alloc;
2292 }
2293
2294 features = btrfs_super_incompat_flags(disk_super) &
2295 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2296 if (features) {
2297 printk(KERN_ERR "BTRFS: couldn't mount because of "
2298 "unsupported optional features (%Lx).\n",
2299 (unsigned long long)features);
2300 err = -EINVAL;
2301 goto fail_alloc;
2302 }
2303
2304 if (btrfs_super_leafsize(disk_super) !=
2305 btrfs_super_nodesize(disk_super)) {
2306 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2307 "blocksizes don't match. node %d leaf %d\n",
2308 btrfs_super_nodesize(disk_super),
2309 btrfs_super_leafsize(disk_super));
2310 err = -EINVAL;
2311 goto fail_alloc;
2312 }
2313 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2314 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2315 "blocksize (%d) was too large\n",
2316 btrfs_super_leafsize(disk_super));
2317 err = -EINVAL;
2318 goto fail_alloc;
2319 }
2320
2321 features = btrfs_super_incompat_flags(disk_super);
2322 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2323 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2324 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2325
2326 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2327 printk(KERN_ERR "btrfs: has skinny extents\n");
2328
2329 /*
2330 * flag our filesystem as having big metadata blocks if
2331 * they are bigger than the page size
2332 */
2333 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2334 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2335 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2336 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2337 }
2338
2339 nodesize = btrfs_super_nodesize(disk_super);
2340 leafsize = btrfs_super_leafsize(disk_super);
2341 sectorsize = btrfs_super_sectorsize(disk_super);
2342 stripesize = btrfs_super_stripesize(disk_super);
2343 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2344 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2345
2346 /*
2347 * mixed block groups end up with duplicate but slightly offset
2348 * extent buffers for the same range. It leads to corruptions
2349 */
2350 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2351 (sectorsize != leafsize)) {
2352 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2353 "are not allowed for mixed block groups on %s\n",
2354 sb->s_id);
2355 goto fail_alloc;
2356 }
2357
2358 /*
2359 * Needn't use the lock because there is no other task which will
2360 * update the flag.
2361 */
2362 btrfs_set_super_incompat_flags(disk_super, features);
2363
2364 features = btrfs_super_compat_ro_flags(disk_super) &
2365 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2366 if (!(sb->s_flags & MS_RDONLY) && features) {
2367 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2368 "unsupported option features (%Lx).\n",
2369 (unsigned long long)features);
2370 err = -EINVAL;
2371 goto fail_alloc;
2372 }
2373
2374 btrfs_init_workers(&fs_info->generic_worker,
2375 "genwork", 1, NULL);
2376
2377 btrfs_init_workers(&fs_info->workers, "worker",
2378 fs_info->thread_pool_size,
2379 &fs_info->generic_worker);
2380
2381 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2382 fs_info->thread_pool_size,
2383 &fs_info->generic_worker);
2384
2385 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2386 fs_info->thread_pool_size,
2387 &fs_info->generic_worker);
2388
2389 btrfs_init_workers(&fs_info->submit_workers, "submit",
2390 min_t(u64, fs_devices->num_devices,
2391 fs_info->thread_pool_size),
2392 &fs_info->generic_worker);
2393
2394 btrfs_init_workers(&fs_info->caching_workers, "cache",
2395 2, &fs_info->generic_worker);
2396
2397 /* a higher idle thresh on the submit workers makes it much more
2398 * likely that bios will be send down in a sane order to the
2399 * devices
2400 */
2401 fs_info->submit_workers.idle_thresh = 64;
2402
2403 fs_info->workers.idle_thresh = 16;
2404 fs_info->workers.ordered = 1;
2405
2406 fs_info->delalloc_workers.idle_thresh = 2;
2407 fs_info->delalloc_workers.ordered = 1;
2408
2409 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2410 &fs_info->generic_worker);
2411 btrfs_init_workers(&fs_info->endio_workers, "endio",
2412 fs_info->thread_pool_size,
2413 &fs_info->generic_worker);
2414 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2415 fs_info->thread_pool_size,
2416 &fs_info->generic_worker);
2417 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2418 "endio-meta-write", fs_info->thread_pool_size,
2419 &fs_info->generic_worker);
2420 btrfs_init_workers(&fs_info->endio_raid56_workers,
2421 "endio-raid56", fs_info->thread_pool_size,
2422 &fs_info->generic_worker);
2423 btrfs_init_workers(&fs_info->rmw_workers,
2424 "rmw", fs_info->thread_pool_size,
2425 &fs_info->generic_worker);
2426 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2427 fs_info->thread_pool_size,
2428 &fs_info->generic_worker);
2429 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2430 1, &fs_info->generic_worker);
2431 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2432 fs_info->thread_pool_size,
2433 &fs_info->generic_worker);
2434 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2435 fs_info->thread_pool_size,
2436 &fs_info->generic_worker);
2437
2438 /*
2439 * endios are largely parallel and should have a very
2440 * low idle thresh
2441 */
2442 fs_info->endio_workers.idle_thresh = 4;
2443 fs_info->endio_meta_workers.idle_thresh = 4;
2444 fs_info->endio_raid56_workers.idle_thresh = 4;
2445 fs_info->rmw_workers.idle_thresh = 2;
2446
2447 fs_info->endio_write_workers.idle_thresh = 2;
2448 fs_info->endio_meta_write_workers.idle_thresh = 2;
2449 fs_info->readahead_workers.idle_thresh = 2;
2450
2451 /*
2452 * btrfs_start_workers can really only fail because of ENOMEM so just
2453 * return -ENOMEM if any of these fail.
2454 */
2455 ret = btrfs_start_workers(&fs_info->workers);
2456 ret |= btrfs_start_workers(&fs_info->generic_worker);
2457 ret |= btrfs_start_workers(&fs_info->submit_workers);
2458 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2459 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2460 ret |= btrfs_start_workers(&fs_info->endio_workers);
2461 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2462 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2463 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2464 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2465 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2466 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2467 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2468 ret |= btrfs_start_workers(&fs_info->caching_workers);
2469 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2470 ret |= btrfs_start_workers(&fs_info->flush_workers);
2471 if (ret) {
2472 err = -ENOMEM;
2473 goto fail_sb_buffer;
2474 }
2475
2476 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2477 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2478 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2479
2480 tree_root->nodesize = nodesize;
2481 tree_root->leafsize = leafsize;
2482 tree_root->sectorsize = sectorsize;
2483 tree_root->stripesize = stripesize;
2484
2485 sb->s_blocksize = sectorsize;
2486 sb->s_blocksize_bits = blksize_bits(sectorsize);
2487
2488 if (disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2489 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2490 goto fail_sb_buffer;
2491 }
2492
2493 if (sectorsize != PAGE_SIZE) {
2494 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2495 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2496 goto fail_sb_buffer;
2497 }
2498
2499 mutex_lock(&fs_info->chunk_mutex);
2500 ret = btrfs_read_sys_array(tree_root);
2501 mutex_unlock(&fs_info->chunk_mutex);
2502 if (ret) {
2503 printk(KERN_WARNING "btrfs: failed to read the system "
2504 "array on %s\n", sb->s_id);
2505 goto fail_sb_buffer;
2506 }
2507
2508 blocksize = btrfs_level_size(tree_root,
2509 btrfs_super_chunk_root_level(disk_super));
2510 generation = btrfs_super_chunk_root_generation(disk_super);
2511
2512 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2513 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2514
2515 chunk_root->node = read_tree_block(chunk_root,
2516 btrfs_super_chunk_root(disk_super),
2517 blocksize, generation);
2518 BUG_ON(!chunk_root->node); /* -ENOMEM */
2519 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2520 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2521 sb->s_id);
2522 goto fail_tree_roots;
2523 }
2524 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2525 chunk_root->commit_root = btrfs_root_node(chunk_root);
2526
2527 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2528 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2529 BTRFS_UUID_SIZE);
2530
2531 ret = btrfs_read_chunk_tree(chunk_root);
2532 if (ret) {
2533 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2534 sb->s_id);
2535 goto fail_tree_roots;
2536 }
2537
2538 /*
2539 * keep the device that is marked to be the target device for the
2540 * dev_replace procedure
2541 */
2542 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2543
2544 if (!fs_devices->latest_bdev) {
2545 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2546 sb->s_id);
2547 goto fail_tree_roots;
2548 }
2549
2550retry_root_backup:
2551 blocksize = btrfs_level_size(tree_root,
2552 btrfs_super_root_level(disk_super));
2553 generation = btrfs_super_generation(disk_super);
2554
2555 tree_root->node = read_tree_block(tree_root,
2556 btrfs_super_root(disk_super),
2557 blocksize, generation);
2558 if (!tree_root->node ||
2559 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2560 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2561 sb->s_id);
2562
2563 goto recovery_tree_root;
2564 }
2565
2566 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2567 tree_root->commit_root = btrfs_root_node(tree_root);
2568
2569 ret = find_and_setup_root(tree_root, fs_info,
2570 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2571 if (ret)
2572 goto recovery_tree_root;
2573 extent_root->track_dirty = 1;
2574
2575 ret = find_and_setup_root(tree_root, fs_info,
2576 BTRFS_DEV_TREE_OBJECTID, dev_root);
2577 if (ret)
2578 goto recovery_tree_root;
2579 dev_root->track_dirty = 1;
2580
2581 ret = find_and_setup_root(tree_root, fs_info,
2582 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2583 if (ret)
2584 goto recovery_tree_root;
2585 csum_root->track_dirty = 1;
2586
2587 ret = find_and_setup_root(tree_root, fs_info,
2588 BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2589 if (ret) {
2590 kfree(quota_root);
2591 quota_root = fs_info->quota_root = NULL;
2592 } else {
2593 quota_root->track_dirty = 1;
2594 fs_info->quota_enabled = 1;
2595 fs_info->pending_quota_state = 1;
2596 }
2597
2598 fs_info->generation = generation;
2599 fs_info->last_trans_committed = generation;
2600
2601 ret = btrfs_recover_balance(fs_info);
2602 if (ret) {
2603 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2604 goto fail_block_groups;
2605 }
2606
2607 ret = btrfs_init_dev_stats(fs_info);
2608 if (ret) {
2609 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2610 ret);
2611 goto fail_block_groups;
2612 }
2613
2614 ret = btrfs_init_dev_replace(fs_info);
2615 if (ret) {
2616 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2617 goto fail_block_groups;
2618 }
2619
2620 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2621
2622 ret = btrfs_init_space_info(fs_info);
2623 if (ret) {
2624 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2625 goto fail_block_groups;
2626 }
2627
2628 ret = btrfs_read_block_groups(extent_root);
2629 if (ret) {
2630 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2631 goto fail_block_groups;
2632 }
2633 fs_info->num_tolerated_disk_barrier_failures =
2634 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2635 if (fs_info->fs_devices->missing_devices >
2636 fs_info->num_tolerated_disk_barrier_failures &&
2637 !(sb->s_flags & MS_RDONLY)) {
2638 printk(KERN_WARNING
2639 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2640 goto fail_block_groups;
2641 }
2642
2643 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2644 "btrfs-cleaner");
2645 if (IS_ERR(fs_info->cleaner_kthread))
2646 goto fail_block_groups;
2647
2648 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2649 tree_root,
2650 "btrfs-transaction");
2651 if (IS_ERR(fs_info->transaction_kthread))
2652 goto fail_cleaner;
2653
2654 if (!btrfs_test_opt(tree_root, SSD) &&
2655 !btrfs_test_opt(tree_root, NOSSD) &&
2656 !fs_info->fs_devices->rotating) {
2657 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2658 "mode\n");
2659 btrfs_set_opt(fs_info->mount_opt, SSD);
2660 }
2661
2662#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2663 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2664 ret = btrfsic_mount(tree_root, fs_devices,
2665 btrfs_test_opt(tree_root,
2666 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2667 1 : 0,
2668 fs_info->check_integrity_print_mask);
2669 if (ret)
2670 printk(KERN_WARNING "btrfs: failed to initialize"
2671 " integrity check module %s\n", sb->s_id);
2672 }
2673#endif
2674 ret = btrfs_read_qgroup_config(fs_info);
2675 if (ret)
2676 goto fail_trans_kthread;
2677
2678 /* do not make disk changes in broken FS */
2679 if (btrfs_super_log_root(disk_super) != 0) {
2680 u64 bytenr = btrfs_super_log_root(disk_super);
2681
2682 if (fs_devices->rw_devices == 0) {
2683 printk(KERN_WARNING "Btrfs log replay required "
2684 "on RO media\n");
2685 err = -EIO;
2686 goto fail_qgroup;
2687 }
2688 blocksize =
2689 btrfs_level_size(tree_root,
2690 btrfs_super_log_root_level(disk_super));
2691
2692 log_tree_root = btrfs_alloc_root(fs_info);
2693 if (!log_tree_root) {
2694 err = -ENOMEM;
2695 goto fail_qgroup;
2696 }
2697
2698 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2699 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2700
2701 log_tree_root->node = read_tree_block(tree_root, bytenr,
2702 blocksize,
2703 generation + 1);
2704 /* returns with log_tree_root freed on success */
2705 ret = btrfs_recover_log_trees(log_tree_root);
2706 if (ret) {
2707 btrfs_error(tree_root->fs_info, ret,
2708 "Failed to recover log tree");
2709 free_extent_buffer(log_tree_root->node);
2710 kfree(log_tree_root);
2711 goto fail_trans_kthread;
2712 }
2713
2714 if (sb->s_flags & MS_RDONLY) {
2715 ret = btrfs_commit_super(tree_root);
2716 if (ret)
2717 goto fail_trans_kthread;
2718 }
2719 }
2720
2721 ret = btrfs_find_orphan_roots(tree_root);
2722 if (ret)
2723 goto fail_trans_kthread;
2724
2725 if (!(sb->s_flags & MS_RDONLY)) {
2726 ret = btrfs_cleanup_fs_roots(fs_info);
2727 if (ret)
2728 goto fail_trans_kthread;
2729
2730 ret = btrfs_recover_relocation(tree_root);
2731 if (ret < 0) {
2732 printk(KERN_WARNING
2733 "btrfs: failed to recover relocation\n");
2734 err = -EINVAL;
2735 goto fail_qgroup;
2736 }
2737 }
2738
2739 location.objectid = BTRFS_FS_TREE_OBJECTID;
2740 location.type = BTRFS_ROOT_ITEM_KEY;
2741 location.offset = (u64)-1;
2742
2743 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2744 if (!fs_info->fs_root)
2745 goto fail_qgroup;
2746 if (IS_ERR(fs_info->fs_root)) {
2747 err = PTR_ERR(fs_info->fs_root);
2748 goto fail_qgroup;
2749 }
2750
2751 if (sb->s_flags & MS_RDONLY)
2752 return 0;
2753
2754 down_read(&fs_info->cleanup_work_sem);
2755 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2756 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2757 up_read(&fs_info->cleanup_work_sem);
2758 close_ctree(tree_root);
2759 return ret;
2760 }
2761 up_read(&fs_info->cleanup_work_sem);
2762
2763 ret = btrfs_resume_balance_async(fs_info);
2764 if (ret) {
2765 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2766 close_ctree(tree_root);
2767 return ret;
2768 }
2769
2770 ret = btrfs_resume_dev_replace_async(fs_info);
2771 if (ret) {
2772 pr_warn("btrfs: failed to resume dev_replace\n");
2773 close_ctree(tree_root);
2774 return ret;
2775 }
2776
2777 return 0;
2778
2779fail_qgroup:
2780 btrfs_free_qgroup_config(fs_info);
2781fail_trans_kthread:
2782 kthread_stop(fs_info->transaction_kthread);
2783fail_cleaner:
2784 kthread_stop(fs_info->cleaner_kthread);
2785
2786 /*
2787 * make sure we're done with the btree inode before we stop our
2788 * kthreads
2789 */
2790 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2791
2792fail_block_groups:
2793 btrfs_free_block_groups(fs_info);
2794
2795fail_tree_roots:
2796 free_root_pointers(fs_info, 1);
2797 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2798
2799fail_sb_buffer:
2800 btrfs_stop_all_workers(fs_info);
2801fail_alloc:
2802fail_iput:
2803 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2804
2805 iput(fs_info->btree_inode);
2806fail_delalloc_bytes:
2807 percpu_counter_destroy(&fs_info->delalloc_bytes);
2808fail_dirty_metadata_bytes:
2809 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2810fail_bdi:
2811 bdi_destroy(&fs_info->bdi);
2812fail_srcu:
2813 cleanup_srcu_struct(&fs_info->subvol_srcu);
2814fail:
2815 btrfs_free_stripe_hash_table(fs_info);
2816 btrfs_close_devices(fs_info->fs_devices);
2817 return err;
2818
2819recovery_tree_root:
2820 if (!btrfs_test_opt(tree_root, RECOVERY))
2821 goto fail_tree_roots;
2822
2823 free_root_pointers(fs_info, 0);
2824
2825 /* don't use the log in recovery mode, it won't be valid */
2826 btrfs_set_super_log_root(disk_super, 0);
2827
2828 /* we can't trust the free space cache either */
2829 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2830
2831 ret = next_root_backup(fs_info, fs_info->super_copy,
2832 &num_backups_tried, &backup_index);
2833 if (ret == -1)
2834 goto fail_block_groups;
2835 goto retry_root_backup;
2836}
2837
2838static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2839{
2840 if (uptodate) {
2841 set_buffer_uptodate(bh);
2842 } else {
2843 struct btrfs_device *device = (struct btrfs_device *)
2844 bh->b_private;
2845
2846 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2847 "I/O error on %s\n",
2848 rcu_str_deref(device->name));
2849 /* note, we dont' set_buffer_write_io_error because we have
2850 * our own ways of dealing with the IO errors
2851 */
2852 clear_buffer_uptodate(bh);
2853 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2854 }
2855 unlock_buffer(bh);
2856 put_bh(bh);
2857}
2858
2859struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2860{
2861 struct buffer_head *bh;
2862 struct buffer_head *latest = NULL;
2863 struct btrfs_super_block *super;
2864 int i;
2865 u64 transid = 0;
2866 u64 bytenr;
2867
2868 /* we would like to check all the supers, but that would make
2869 * a btrfs mount succeed after a mkfs from a different FS.
2870 * So, we need to add a special mount option to scan for
2871 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2872 */
2873 for (i = 0; i < 1; i++) {
2874 bytenr = btrfs_sb_offset(i);
2875 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2876 break;
2877 bh = __bread(bdev, bytenr / 4096, 4096);
2878 if (!bh)
2879 continue;
2880
2881 super = (struct btrfs_super_block *)bh->b_data;
2882 if (btrfs_super_bytenr(super) != bytenr ||
2883 super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2884 brelse(bh);
2885 continue;
2886 }
2887
2888 if (!latest || btrfs_super_generation(super) > transid) {
2889 brelse(latest);
2890 latest = bh;
2891 transid = btrfs_super_generation(super);
2892 } else {
2893 brelse(bh);
2894 }
2895 }
2896 return latest;
2897}
2898
2899/*
2900 * this should be called twice, once with wait == 0 and
2901 * once with wait == 1. When wait == 0 is done, all the buffer heads
2902 * we write are pinned.
2903 *
2904 * They are released when wait == 1 is done.
2905 * max_mirrors must be the same for both runs, and it indicates how
2906 * many supers on this one device should be written.
2907 *
2908 * max_mirrors == 0 means to write them all.
2909 */
2910static int write_dev_supers(struct btrfs_device *device,
2911 struct btrfs_super_block *sb,
2912 int do_barriers, int wait, int max_mirrors)
2913{
2914 struct buffer_head *bh;
2915 int i;
2916 int ret;
2917 int errors = 0;
2918 u32 crc;
2919 u64 bytenr;
2920
2921 if (max_mirrors == 0)
2922 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2923
2924 for (i = 0; i < max_mirrors; i++) {
2925 bytenr = btrfs_sb_offset(i);
2926 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2927 break;
2928
2929 if (wait) {
2930 bh = __find_get_block(device->bdev, bytenr / 4096,
2931 BTRFS_SUPER_INFO_SIZE);
2932 BUG_ON(!bh);
2933 wait_on_buffer(bh);
2934 if (!buffer_uptodate(bh))
2935 errors++;
2936
2937 /* drop our reference */
2938 brelse(bh);
2939
2940 /* drop the reference from the wait == 0 run */
2941 brelse(bh);
2942 continue;
2943 } else {
2944 btrfs_set_super_bytenr(sb, bytenr);
2945
2946 crc = ~(u32)0;
2947 crc = btrfs_csum_data((char *)sb +
2948 BTRFS_CSUM_SIZE, crc,
2949 BTRFS_SUPER_INFO_SIZE -
2950 BTRFS_CSUM_SIZE);
2951 btrfs_csum_final(crc, sb->csum);
2952
2953 /*
2954 * one reference for us, and we leave it for the
2955 * caller
2956 */
2957 bh = __getblk(device->bdev, bytenr / 4096,
2958 BTRFS_SUPER_INFO_SIZE);
2959 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2960
2961 /* one reference for submit_bh */
2962 get_bh(bh);
2963
2964 set_buffer_uptodate(bh);
2965 lock_buffer(bh);
2966 bh->b_end_io = btrfs_end_buffer_write_sync;
2967 bh->b_private = device;
2968 }
2969
2970 /*
2971 * we fua the first super. The others we allow
2972 * to go down lazy.
2973 */
2974 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2975 if (ret)
2976 errors++;
2977 }
2978 return errors < i ? 0 : -1;
2979}
2980
2981/*
2982 * endio for the write_dev_flush, this will wake anyone waiting
2983 * for the barrier when it is done
2984 */
2985static void btrfs_end_empty_barrier(struct bio *bio, int err)
2986{
2987 if (err) {
2988 if (err == -EOPNOTSUPP)
2989 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2990 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2991 }
2992 if (bio->bi_private)
2993 complete(bio->bi_private);
2994 bio_put(bio);
2995}
2996
2997/*
2998 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2999 * sent down. With wait == 1, it waits for the previous flush.
3000 *
3001 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3002 * capable
3003 */
3004static int write_dev_flush(struct btrfs_device *device, int wait)
3005{
3006 struct bio *bio;
3007 int ret = 0;
3008
3009 if (device->nobarriers)
3010 return 0;
3011
3012 if (wait) {
3013 bio = device->flush_bio;
3014 if (!bio)
3015 return 0;
3016
3017 wait_for_completion(&device->flush_wait);
3018
3019 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3020 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3021 rcu_str_deref(device->name));
3022 device->nobarriers = 1;
3023 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3024 ret = -EIO;
3025 btrfs_dev_stat_inc_and_print(device,
3026 BTRFS_DEV_STAT_FLUSH_ERRS);
3027 }
3028
3029 /* drop the reference from the wait == 0 run */
3030 bio_put(bio);
3031 device->flush_bio = NULL;
3032
3033 return ret;
3034 }
3035
3036 /*
3037 * one reference for us, and we leave it for the
3038 * caller
3039 */
3040 device->flush_bio = NULL;
3041 bio = bio_alloc(GFP_NOFS, 0);
3042 if (!bio)
3043 return -ENOMEM;
3044
3045 bio->bi_end_io = btrfs_end_empty_barrier;
3046 bio->bi_bdev = device->bdev;
3047 init_completion(&device->flush_wait);
3048 bio->bi_private = &device->flush_wait;
3049 device->flush_bio = bio;
3050
3051 bio_get(bio);
3052 btrfsic_submit_bio(WRITE_FLUSH, bio);
3053
3054 return 0;
3055}
3056
3057/*
3058 * send an empty flush down to each device in parallel,
3059 * then wait for them
3060 */
3061static int barrier_all_devices(struct btrfs_fs_info *info)
3062{
3063 struct list_head *head;
3064 struct btrfs_device *dev;
3065 int errors_send = 0;
3066 int errors_wait = 0;
3067 int ret;
3068
3069 /* send down all the barriers */
3070 head = &info->fs_devices->devices;
3071 list_for_each_entry_rcu(dev, head, dev_list) {
3072 if (!dev->bdev) {
3073 errors_send++;
3074 continue;
3075 }
3076 if (!dev->in_fs_metadata || !dev->writeable)
3077 continue;
3078
3079 ret = write_dev_flush(dev, 0);
3080 if (ret)
3081 errors_send++;
3082 }
3083
3084 /* wait for all the barriers */
3085 list_for_each_entry_rcu(dev, head, dev_list) {
3086 if (!dev->bdev) {
3087 errors_wait++;
3088 continue;
3089 }
3090 if (!dev->in_fs_metadata || !dev->writeable)
3091 continue;
3092
3093 ret = write_dev_flush(dev, 1);
3094 if (ret)
3095 errors_wait++;
3096 }
3097 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3098 errors_wait > info->num_tolerated_disk_barrier_failures)
3099 return -EIO;
3100 return 0;
3101}
3102
3103int btrfs_calc_num_tolerated_disk_barrier_failures(
3104 struct btrfs_fs_info *fs_info)
3105{
3106 struct btrfs_ioctl_space_info space;
3107 struct btrfs_space_info *sinfo;
3108 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3109 BTRFS_BLOCK_GROUP_SYSTEM,
3110 BTRFS_BLOCK_GROUP_METADATA,
3111 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3112 int num_types = 4;
3113 int i;
3114 int c;
3115 int num_tolerated_disk_barrier_failures =
3116 (int)fs_info->fs_devices->num_devices;
3117
3118 for (i = 0; i < num_types; i++) {
3119 struct btrfs_space_info *tmp;
3120
3121 sinfo = NULL;
3122 rcu_read_lock();
3123 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3124 if (tmp->flags == types[i]) {
3125 sinfo = tmp;
3126 break;
3127 }
3128 }
3129 rcu_read_unlock();
3130
3131 if (!sinfo)
3132 continue;
3133
3134 down_read(&sinfo->groups_sem);
3135 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3136 if (!list_empty(&sinfo->block_groups[c])) {
3137 u64 flags;
3138
3139 btrfs_get_block_group_info(
3140 &sinfo->block_groups[c], &space);
3141 if (space.total_bytes == 0 ||
3142 space.used_bytes == 0)
3143 continue;
3144 flags = space.flags;
3145 /*
3146 * return
3147 * 0: if dup, single or RAID0 is configured for
3148 * any of metadata, system or data, else
3149 * 1: if RAID5 is configured, or if RAID1 or
3150 * RAID10 is configured and only two mirrors
3151 * are used, else
3152 * 2: if RAID6 is configured, else
3153 * num_mirrors - 1: if RAID1 or RAID10 is
3154 * configured and more than
3155 * 2 mirrors are used.
3156 */
3157 if (num_tolerated_disk_barrier_failures > 0 &&
3158 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3159 BTRFS_BLOCK_GROUP_RAID0)) ||
3160 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3161 == 0)))
3162 num_tolerated_disk_barrier_failures = 0;
3163 else if (num_tolerated_disk_barrier_failures > 1) {
3164 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3165 BTRFS_BLOCK_GROUP_RAID5 |
3166 BTRFS_BLOCK_GROUP_RAID10)) {
3167 num_tolerated_disk_barrier_failures = 1;
3168 } else if (flags &
3169 BTRFS_BLOCK_GROUP_RAID5) {
3170 num_tolerated_disk_barrier_failures = 2;
3171 }
3172 }
3173 }
3174 }
3175 up_read(&sinfo->groups_sem);
3176 }
3177
3178 return num_tolerated_disk_barrier_failures;
3179}
3180
3181int write_all_supers(struct btrfs_root *root, int max_mirrors)
3182{
3183 struct list_head *head;
3184 struct btrfs_device *dev;
3185 struct btrfs_super_block *sb;
3186 struct btrfs_dev_item *dev_item;
3187 int ret;
3188 int do_barriers;
3189 int max_errors;
3190 int total_errors = 0;
3191 u64 flags;
3192
3193 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3194 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3195 backup_super_roots(root->fs_info);
3196
3197 sb = root->fs_info->super_for_commit;
3198 dev_item = &sb->dev_item;
3199
3200 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3201 head = &root->fs_info->fs_devices->devices;
3202
3203 if (do_barriers) {
3204 ret = barrier_all_devices(root->fs_info);
3205 if (ret) {
3206 mutex_unlock(
3207 &root->fs_info->fs_devices->device_list_mutex);
3208 btrfs_error(root->fs_info, ret,
3209 "errors while submitting device barriers.");
3210 return ret;
3211 }
3212 }
3213
3214 list_for_each_entry_rcu(dev, head, dev_list) {
3215 if (!dev->bdev) {
3216 total_errors++;
3217 continue;
3218 }
3219 if (!dev->in_fs_metadata || !dev->writeable)
3220 continue;
3221
3222 btrfs_set_stack_device_generation(dev_item, 0);
3223 btrfs_set_stack_device_type(dev_item, dev->type);
3224 btrfs_set_stack_device_id(dev_item, dev->devid);
3225 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3226 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3227 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3228 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3229 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3230 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3231 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3232
3233 flags = btrfs_super_flags(sb);
3234 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3235
3236 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3237 if (ret)
3238 total_errors++;
3239 }
3240 if (total_errors > max_errors) {
3241 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3242 total_errors);
3243
3244 /* This shouldn't happen. FUA is masked off if unsupported */
3245 BUG();
3246 }
3247
3248 total_errors = 0;
3249 list_for_each_entry_rcu(dev, head, dev_list) {
3250 if (!dev->bdev)
3251 continue;
3252 if (!dev->in_fs_metadata || !dev->writeable)
3253 continue;
3254
3255 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3256 if (ret)
3257 total_errors++;
3258 }
3259 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3260 if (total_errors > max_errors) {
3261 btrfs_error(root->fs_info, -EIO,
3262 "%d errors while writing supers", total_errors);
3263 return -EIO;
3264 }
3265 return 0;
3266}
3267
3268int write_ctree_super(struct btrfs_trans_handle *trans,
3269 struct btrfs_root *root, int max_mirrors)
3270{
3271 int ret;
3272
3273 ret = write_all_supers(root, max_mirrors);
3274 return ret;
3275}
3276
3277void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3278{
3279 spin_lock(&fs_info->fs_roots_radix_lock);
3280 radix_tree_delete(&fs_info->fs_roots_radix,
3281 (unsigned long)root->root_key.objectid);
3282 spin_unlock(&fs_info->fs_roots_radix_lock);
3283
3284 if (btrfs_root_refs(&root->root_item) == 0)
3285 synchronize_srcu(&fs_info->subvol_srcu);
3286
3287 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3288 btrfs_free_log(NULL, root);
3289 btrfs_free_log_root_tree(NULL, fs_info);
3290 }
3291
3292 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3293 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3294 free_fs_root(root);
3295}
3296
3297static void free_fs_root(struct btrfs_root *root)
3298{
3299 iput(root->cache_inode);
3300 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3301 if (root->anon_dev)
3302 free_anon_bdev(root->anon_dev);
3303 free_extent_buffer(root->node);
3304 free_extent_buffer(root->commit_root);
3305 kfree(root->free_ino_ctl);
3306 kfree(root->free_ino_pinned);
3307 kfree(root->name);
3308 kfree(root);
3309}
3310
3311static void del_fs_roots(struct btrfs_fs_info *fs_info)
3312{
3313 int ret;
3314 struct btrfs_root *gang[8];
3315 int i;
3316
3317 while (!list_empty(&fs_info->dead_roots)) {
3318 gang[0] = list_entry(fs_info->dead_roots.next,
3319 struct btrfs_root, root_list);
3320 list_del(&gang[0]->root_list);
3321
3322 if (gang[0]->in_radix) {
3323 btrfs_free_fs_root(fs_info, gang[0]);
3324 } else {
3325 free_extent_buffer(gang[0]->node);
3326 free_extent_buffer(gang[0]->commit_root);
3327 kfree(gang[0]);
3328 }
3329 }
3330
3331 while (1) {
3332 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3333 (void **)gang, 0,
3334 ARRAY_SIZE(gang));
3335 if (!ret)
3336 break;
3337 for (i = 0; i < ret; i++)
3338 btrfs_free_fs_root(fs_info, gang[i]);
3339 }
3340}
3341
3342int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3343{
3344 u64 root_objectid = 0;
3345 struct btrfs_root *gang[8];
3346 int i;
3347 int ret;
3348
3349 while (1) {
3350 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3351 (void **)gang, root_objectid,
3352 ARRAY_SIZE(gang));
3353 if (!ret)
3354 break;
3355
3356 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3357 for (i = 0; i < ret; i++) {
3358 int err;
3359
3360 root_objectid = gang[i]->root_key.objectid;
3361 err = btrfs_orphan_cleanup(gang[i]);
3362 if (err)
3363 return err;
3364 }
3365 root_objectid++;
3366 }
3367 return 0;
3368}
3369
3370int btrfs_commit_super(struct btrfs_root *root)
3371{
3372 struct btrfs_trans_handle *trans;
3373 int ret;
3374
3375 mutex_lock(&root->fs_info->cleaner_mutex);
3376 btrfs_run_delayed_iputs(root);
3377 mutex_unlock(&root->fs_info->cleaner_mutex);
3378 wake_up_process(root->fs_info->cleaner_kthread);
3379
3380 /* wait until ongoing cleanup work done */
3381 down_write(&root->fs_info->cleanup_work_sem);
3382 up_write(&root->fs_info->cleanup_work_sem);
3383
3384 trans = btrfs_join_transaction(root);
3385 if (IS_ERR(trans))
3386 return PTR_ERR(trans);
3387 ret = btrfs_commit_transaction(trans, root);
3388 if (ret)
3389 return ret;
3390 /* run commit again to drop the original snapshot */
3391 trans = btrfs_join_transaction(root);
3392 if (IS_ERR(trans))
3393 return PTR_ERR(trans);
3394 ret = btrfs_commit_transaction(trans, root);
3395 if (ret)
3396 return ret;
3397 ret = btrfs_write_and_wait_transaction(NULL, root);
3398 if (ret) {
3399 btrfs_error(root->fs_info, ret,
3400 "Failed to sync btree inode to disk.");
3401 return ret;
3402 }
3403
3404 ret = write_ctree_super(NULL, root, 0);
3405 return ret;
3406}
3407
3408int close_ctree(struct btrfs_root *root)
3409{
3410 struct btrfs_fs_info *fs_info = root->fs_info;
3411 int ret;
3412
3413 fs_info->closing = 1;
3414 smp_mb();
3415
3416 /* pause restriper - we want to resume on mount */
3417 btrfs_pause_balance(fs_info);
3418
3419 btrfs_dev_replace_suspend_for_unmount(fs_info);
3420
3421 btrfs_scrub_cancel(fs_info);
3422
3423 /* wait for any defraggers to finish */
3424 wait_event(fs_info->transaction_wait,
3425 (atomic_read(&fs_info->defrag_running) == 0));
3426
3427 /* clear out the rbtree of defraggable inodes */
3428 btrfs_cleanup_defrag_inodes(fs_info);
3429
3430 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3431 ret = btrfs_commit_super(root);
3432 if (ret)
3433 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3434 }
3435
3436 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3437 btrfs_error_commit_super(root);
3438
3439 btrfs_put_block_group_cache(fs_info);
3440
3441 kthread_stop(fs_info->transaction_kthread);
3442 kthread_stop(fs_info->cleaner_kthread);
3443
3444 fs_info->closing = 2;
3445 smp_mb();
3446
3447 btrfs_free_qgroup_config(root->fs_info);
3448
3449 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3450 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3451 percpu_counter_sum(&fs_info->delalloc_bytes));
3452 }
3453
3454 free_root_pointers(fs_info, 1);
3455
3456 btrfs_free_block_groups(fs_info);
3457
3458 del_fs_roots(fs_info);
3459
3460 iput(fs_info->btree_inode);
3461
3462 btrfs_stop_all_workers(fs_info);
3463
3464#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3465 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3466 btrfsic_unmount(root, fs_info->fs_devices);
3467#endif
3468
3469 btrfs_close_devices(fs_info->fs_devices);
3470 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3471
3472 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3473 percpu_counter_destroy(&fs_info->delalloc_bytes);
3474 bdi_destroy(&fs_info->bdi);
3475 cleanup_srcu_struct(&fs_info->subvol_srcu);
3476
3477 btrfs_free_stripe_hash_table(fs_info);
3478
3479 return 0;
3480}
3481
3482int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3483 int atomic)
3484{
3485 int ret;
3486 struct inode *btree_inode = buf->pages[0]->mapping->host;
3487
3488 ret = extent_buffer_uptodate(buf);
3489 if (!ret)
3490 return ret;
3491
3492 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3493 parent_transid, atomic);
3494 if (ret == -EAGAIN)
3495 return ret;
3496 return !ret;
3497}
3498
3499int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3500{
3501 return set_extent_buffer_uptodate(buf);
3502}
3503
3504void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3505{
3506 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3507 u64 transid = btrfs_header_generation(buf);
3508 int was_dirty;
3509
3510 btrfs_assert_tree_locked(buf);
3511 if (transid != root->fs_info->generation)
3512 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3513 "found %llu running %llu\n",
3514 (unsigned long long)buf->start,
3515 (unsigned long long)transid,
3516 (unsigned long long)root->fs_info->generation);
3517 was_dirty = set_extent_buffer_dirty(buf);
3518 if (!was_dirty)
3519 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3520 buf->len,
3521 root->fs_info->dirty_metadata_batch);
3522}
3523
3524static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3525 int flush_delayed)
3526{
3527 /*
3528 * looks as though older kernels can get into trouble with
3529 * this code, they end up stuck in balance_dirty_pages forever
3530 */
3531 int ret;
3532
3533 if (current->flags & PF_MEMALLOC)
3534 return;
3535
3536 if (flush_delayed)
3537 btrfs_balance_delayed_items(root);
3538
3539 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3540 BTRFS_DIRTY_METADATA_THRESH);
3541 if (ret > 0) {
3542 balance_dirty_pages_ratelimited(
3543 root->fs_info->btree_inode->i_mapping);
3544 }
3545 return;
3546}
3547
3548void btrfs_btree_balance_dirty(struct btrfs_root *root)
3549{
3550 __btrfs_btree_balance_dirty(root, 1);
3551}
3552
3553void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3554{
3555 __btrfs_btree_balance_dirty(root, 0);
3556}
3557
3558int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3559{
3560 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3561 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3562}
3563
3564static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3565 int read_only)
3566{
3567 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3568 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3569 return -EINVAL;
3570 }
3571
3572 if (read_only)
3573 return 0;
3574
3575 return 0;
3576}
3577
3578void btrfs_error_commit_super(struct btrfs_root *root)
3579{
3580 mutex_lock(&root->fs_info->cleaner_mutex);
3581 btrfs_run_delayed_iputs(root);
3582 mutex_unlock(&root->fs_info->cleaner_mutex);
3583
3584 down_write(&root->fs_info->cleanup_work_sem);
3585 up_write(&root->fs_info->cleanup_work_sem);
3586
3587 /* cleanup FS via transaction */
3588 btrfs_cleanup_transaction(root);
3589}
3590
3591static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3592 struct btrfs_root *root)
3593{
3594 struct btrfs_inode *btrfs_inode;
3595 struct list_head splice;
3596
3597 INIT_LIST_HEAD(&splice);
3598
3599 mutex_lock(&root->fs_info->ordered_operations_mutex);
3600 spin_lock(&root->fs_info->ordered_extent_lock);
3601
3602 list_splice_init(&t->ordered_operations, &splice);
3603 while (!list_empty(&splice)) {
3604 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3605 ordered_operations);
3606
3607 list_del_init(&btrfs_inode->ordered_operations);
3608
3609 btrfs_invalidate_inodes(btrfs_inode->root);
3610 }
3611
3612 spin_unlock(&root->fs_info->ordered_extent_lock);
3613 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3614}
3615
3616static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3617{
3618 struct btrfs_ordered_extent *ordered;
3619
3620 spin_lock(&root->fs_info->ordered_extent_lock);
3621 /*
3622 * This will just short circuit the ordered completion stuff which will
3623 * make sure the ordered extent gets properly cleaned up.
3624 */
3625 list_for_each_entry(ordered, &root->fs_info->ordered_extents,
3626 root_extent_list)
3627 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3628 spin_unlock(&root->fs_info->ordered_extent_lock);
3629}
3630
3631int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3632 struct btrfs_root *root)
3633{
3634 struct rb_node *node;
3635 struct btrfs_delayed_ref_root *delayed_refs;
3636 struct btrfs_delayed_ref_node *ref;
3637 int ret = 0;
3638
3639 delayed_refs = &trans->delayed_refs;
3640
3641 spin_lock(&delayed_refs->lock);
3642 if (delayed_refs->num_entries == 0) {
3643 spin_unlock(&delayed_refs->lock);
3644 printk(KERN_INFO "delayed_refs has NO entry\n");
3645 return ret;
3646 }
3647
3648 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3649 struct btrfs_delayed_ref_head *head = NULL;
3650
3651 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3652 atomic_set(&ref->refs, 1);
3653 if (btrfs_delayed_ref_is_head(ref)) {
3654
3655 head = btrfs_delayed_node_to_head(ref);
3656 if (!mutex_trylock(&head->mutex)) {
3657 atomic_inc(&ref->refs);
3658 spin_unlock(&delayed_refs->lock);
3659
3660 /* Need to wait for the delayed ref to run */
3661 mutex_lock(&head->mutex);
3662 mutex_unlock(&head->mutex);
3663 btrfs_put_delayed_ref(ref);
3664
3665 spin_lock(&delayed_refs->lock);
3666 continue;
3667 }
3668
3669 btrfs_free_delayed_extent_op(head->extent_op);
3670 delayed_refs->num_heads--;
3671 if (list_empty(&head->cluster))
3672 delayed_refs->num_heads_ready--;
3673 list_del_init(&head->cluster);
3674 }
3675
3676 ref->in_tree = 0;
3677 rb_erase(&ref->rb_node, &delayed_refs->root);
3678 delayed_refs->num_entries--;
3679 if (head)
3680 mutex_unlock(&head->mutex);
3681 spin_unlock(&delayed_refs->lock);
3682 btrfs_put_delayed_ref(ref);
3683
3684 cond_resched();
3685 spin_lock(&delayed_refs->lock);
3686 }
3687
3688 spin_unlock(&delayed_refs->lock);
3689
3690 return ret;
3691}
3692
3693static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3694{
3695 struct btrfs_pending_snapshot *snapshot;
3696 struct list_head splice;
3697
3698 INIT_LIST_HEAD(&splice);
3699
3700 list_splice_init(&t->pending_snapshots, &splice);
3701
3702 while (!list_empty(&splice)) {
3703 snapshot = list_entry(splice.next,
3704 struct btrfs_pending_snapshot,
3705 list);
3706 snapshot->error = -ECANCELED;
3707 list_del_init(&snapshot->list);
3708 }
3709}
3710
3711static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3712{
3713 struct btrfs_inode *btrfs_inode;
3714 struct list_head splice;
3715
3716 INIT_LIST_HEAD(&splice);
3717
3718 spin_lock(&root->fs_info->delalloc_lock);
3719 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3720
3721 while (!list_empty(&splice)) {
3722 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3723 delalloc_inodes);
3724
3725 list_del_init(&btrfs_inode->delalloc_inodes);
3726 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3727 &btrfs_inode->runtime_flags);
3728
3729 btrfs_invalidate_inodes(btrfs_inode->root);
3730 }
3731
3732 spin_unlock(&root->fs_info->delalloc_lock);
3733}
3734
3735static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3736 struct extent_io_tree *dirty_pages,
3737 int mark)
3738{
3739 int ret;
3740 struct page *page;
3741 struct inode *btree_inode = root->fs_info->btree_inode;
3742 struct extent_buffer *eb;
3743 u64 start = 0;
3744 u64 end;
3745 u64 offset;
3746 unsigned long index;
3747
3748 while (1) {
3749 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3750 mark, NULL);
3751 if (ret)
3752 break;
3753
3754 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3755 while (start <= end) {
3756 index = start >> PAGE_CACHE_SHIFT;
3757 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3758 page = find_get_page(btree_inode->i_mapping, index);
3759 if (!page)
3760 continue;
3761 offset = page_offset(page);
3762
3763 spin_lock(&dirty_pages->buffer_lock);
3764 eb = radix_tree_lookup(
3765 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3766 offset >> PAGE_CACHE_SHIFT);
3767 spin_unlock(&dirty_pages->buffer_lock);
3768 if (eb)
3769 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3770 &eb->bflags);
3771 lock_page(page);
3772
3773 wait_on_page_writeback(page);
3774 if (PageDirty(page)) {
3775 clear_page_dirty_for_io(page);
3776 spin_lock_irq(&page->mapping->tree_lock);
3777 radix_tree_tag_clear(&page->mapping->page_tree,
3778 page_index(page),
3779 PAGECACHE_TAG_DIRTY);
3780 spin_unlock_irq(&page->mapping->tree_lock);
3781 }
3782
3783 unlock_page(page);
3784 page_cache_release(page);
3785 }
3786 }
3787
3788 return ret;
3789}
3790
3791static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3792 struct extent_io_tree *pinned_extents)
3793{
3794 struct extent_io_tree *unpin;
3795 u64 start;
3796 u64 end;
3797 int ret;
3798 bool loop = true;
3799
3800 unpin = pinned_extents;
3801again:
3802 while (1) {
3803 ret = find_first_extent_bit(unpin, 0, &start, &end,
3804 EXTENT_DIRTY, NULL);
3805 if (ret)
3806 break;
3807
3808 /* opt_discard */
3809 if (btrfs_test_opt(root, DISCARD))
3810 ret = btrfs_error_discard_extent(root, start,
3811 end + 1 - start,
3812 NULL);
3813
3814 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3815 btrfs_error_unpin_extent_range(root, start, end);
3816 cond_resched();
3817 }
3818
3819 if (loop) {
3820 if (unpin == &root->fs_info->freed_extents[0])
3821 unpin = &root->fs_info->freed_extents[1];
3822 else
3823 unpin = &root->fs_info->freed_extents[0];
3824 loop = false;
3825 goto again;
3826 }
3827
3828 return 0;
3829}
3830
3831void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3832 struct btrfs_root *root)
3833{
3834 btrfs_destroy_delayed_refs(cur_trans, root);
3835 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3836 cur_trans->dirty_pages.dirty_bytes);
3837
3838 /* FIXME: cleanup wait for commit */
3839 cur_trans->in_commit = 1;
3840 cur_trans->blocked = 1;
3841 wake_up(&root->fs_info->transaction_blocked_wait);
3842
3843 btrfs_evict_pending_snapshots(cur_trans);
3844
3845 cur_trans->blocked = 0;
3846 wake_up(&root->fs_info->transaction_wait);
3847
3848 cur_trans->commit_done = 1;
3849 wake_up(&cur_trans->commit_wait);
3850
3851 btrfs_destroy_delayed_inodes(root);
3852 btrfs_assert_delayed_root_empty(root);
3853
3854 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3855 EXTENT_DIRTY);
3856 btrfs_destroy_pinned_extent(root,
3857 root->fs_info->pinned_extents);
3858
3859 /*
3860 memset(cur_trans, 0, sizeof(*cur_trans));
3861 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3862 */
3863}
3864
3865int btrfs_cleanup_transaction(struct btrfs_root *root)
3866{
3867 struct btrfs_transaction *t;
3868 LIST_HEAD(list);
3869
3870 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3871
3872 spin_lock(&root->fs_info->trans_lock);
3873 list_splice_init(&root->fs_info->trans_list, &list);
3874 root->fs_info->trans_no_join = 1;
3875 spin_unlock(&root->fs_info->trans_lock);
3876
3877 while (!list_empty(&list)) {
3878 t = list_entry(list.next, struct btrfs_transaction, list);
3879
3880 btrfs_destroy_ordered_operations(t, root);
3881
3882 btrfs_destroy_ordered_extents(root);
3883
3884 btrfs_destroy_delayed_refs(t, root);
3885
3886 btrfs_block_rsv_release(root,
3887 &root->fs_info->trans_block_rsv,
3888 t->dirty_pages.dirty_bytes);
3889
3890 /* FIXME: cleanup wait for commit */
3891 t->in_commit = 1;
3892 t->blocked = 1;
3893 smp_mb();
3894 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3895 wake_up(&root->fs_info->transaction_blocked_wait);
3896
3897 btrfs_evict_pending_snapshots(t);
3898
3899 t->blocked = 0;
3900 smp_mb();
3901 if (waitqueue_active(&root->fs_info->transaction_wait))
3902 wake_up(&root->fs_info->transaction_wait);
3903
3904 t->commit_done = 1;
3905 smp_mb();
3906 if (waitqueue_active(&t->commit_wait))
3907 wake_up(&t->commit_wait);
3908
3909 btrfs_destroy_delayed_inodes(root);
3910 btrfs_assert_delayed_root_empty(root);
3911
3912 btrfs_destroy_delalloc_inodes(root);
3913
3914 spin_lock(&root->fs_info->trans_lock);
3915 root->fs_info->running_transaction = NULL;
3916 spin_unlock(&root->fs_info->trans_lock);
3917
3918 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3919 EXTENT_DIRTY);
3920
3921 btrfs_destroy_pinned_extent(root,
3922 root->fs_info->pinned_extents);
3923
3924 atomic_set(&t->use_count, 0);
3925 list_del_init(&t->list);
3926 memset(t, 0, sizeof(*t));
3927 kmem_cache_free(btrfs_transaction_cachep, t);
3928 }
3929
3930 spin_lock(&root->fs_info->trans_lock);
3931 root->fs_info->trans_no_join = 0;
3932 spin_unlock(&root->fs_info->trans_lock);
3933 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3934
3935 return 0;
3936}
3937
3938static struct extent_io_ops btree_extent_io_ops = {
3939 .readpage_end_io_hook = btree_readpage_end_io_hook,
3940 .readpage_io_failed_hook = btree_io_failed_hook,
3941 .submit_bio_hook = btree_submit_bio_hook,
3942 /* note we're sharing with inode.c for the merge bio hook */
3943 .merge_bio_hook = btrfs_merge_bio_hook,
3944};
Linux 6.x block layer and io_uring tree(s)