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