projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: Cocci spatch "memdup.spatch"
[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 <linux/uuid.h>
34#include <asm/unaligned.h>
35#include "compat.h"
36#include "ctree.h"
37#include "disk-io.h"
38#include "transaction.h"
39#include "btrfs_inode.h"
40#include "volumes.h"
41#include "print-tree.h"
42#include "async-thread.h"
43#include "locking.h"
44#include "tree-log.h"
45#include "free-space-cache.h"
46#include "inode-map.h"
47#include "check-integrity.h"
48#include "rcu-string.h"
49#include "dev-replace.h"
50#include "raid56.h"
51
52#ifdef CONFIG_X86
53#include <asm/cpufeature.h>
54#endif
55
56static struct extent_io_ops btree_extent_io_ops;
57static void end_workqueue_fn(struct btrfs_work *work);
58static void free_fs_root(struct btrfs_root *root);
59static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
60 int read_only);
61static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
62 struct btrfs_root *root);
63static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
64static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
65 struct btrfs_root *root);
66static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
67static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
68static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 struct extent_io_tree *dirty_pages,
70 int mark);
71static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 struct extent_io_tree *pinned_extents);
73static int btrfs_cleanup_transaction(struct btrfs_root *root);
74static void btrfs_error_commit_super(struct btrfs_root *root);
75
76/*
77 * end_io_wq structs are used to do processing in task context when an IO is
78 * complete. This is used during reads to verify checksums, and it is used
79 * by writes to insert metadata for new file extents after IO is complete.
80 */
81struct end_io_wq {
82 struct bio *bio;
83 bio_end_io_t *end_io;
84 void *private;
85 struct btrfs_fs_info *info;
86 int error;
87 int metadata;
88 struct list_head list;
89 struct btrfs_work work;
90};
91
92/*
93 * async submit bios are used to offload expensive checksumming
94 * onto the worker threads. They checksum file and metadata bios
95 * just before they are sent down the IO stack.
96 */
97struct async_submit_bio {
98 struct inode *inode;
99 struct bio *bio;
100 struct list_head list;
101 extent_submit_bio_hook_t *submit_bio_start;
102 extent_submit_bio_hook_t *submit_bio_done;
103 int rw;
104 int mirror_num;
105 unsigned long bio_flags;
106 /*
107 * bio_offset is optional, can be used if the pages in the bio
108 * can't tell us where in the file the bio should go
109 */
110 u64 bio_offset;
111 struct btrfs_work work;
112 int error;
113};
114
115/*
116 * Lockdep class keys for extent_buffer->lock's in this root. For a given
117 * eb, the lockdep key is determined by the btrfs_root it belongs to and
118 * the level the eb occupies in the tree.
119 *
120 * Different roots are used for different purposes and may nest inside each
121 * other and they require separate keysets. As lockdep keys should be
122 * static, assign keysets according to the purpose of the root as indicated
123 * by btrfs_root->objectid. This ensures that all special purpose roots
124 * have separate keysets.
125 *
126 * Lock-nesting across peer nodes is always done with the immediate parent
127 * node locked thus preventing deadlock. As lockdep doesn't know this, use
128 * subclass to avoid triggering lockdep warning in such cases.
129 *
130 * The key is set by the readpage_end_io_hook after the buffer has passed
131 * csum validation but before the pages are unlocked. It is also set by
132 * btrfs_init_new_buffer on freshly allocated blocks.
133 *
134 * We also add a check to make sure the highest level of the tree is the
135 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
136 * needs update as well.
137 */
138#ifdef CONFIG_DEBUG_LOCK_ALLOC
139# if BTRFS_MAX_LEVEL != 8
140# error
141# endif
142
143static struct btrfs_lockdep_keyset {
144 u64 id; /* root objectid */
145 const char *name_stem; /* lock name stem */
146 char names[BTRFS_MAX_LEVEL + 1][20];
147 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
148} btrfs_lockdep_keysets[] = {
149 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
150 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
151 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
152 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
153 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
154 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
155 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
156 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
157 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
158 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
159 { .id = 0, .name_stem = "tree" },
160};
161
162void __init btrfs_init_lockdep(void)
163{
164 int i, j;
165
166 /* initialize lockdep class names */
167 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
168 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
169
170 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
171 snprintf(ks->names[j], sizeof(ks->names[j]),
172 "btrfs-%s-%02d", ks->name_stem, j);
173 }
174}
175
176void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
177 int level)
178{
179 struct btrfs_lockdep_keyset *ks;
180
181 BUG_ON(level >= ARRAY_SIZE(ks->keys));
182
183 /* find the matching keyset, id 0 is the default entry */
184 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
185 if (ks->id == objectid)
186 break;
187
188 lockdep_set_class_and_name(&eb->lock,
189 &ks->keys[level], ks->names[level]);
190}
191
192#endif
193
194/*
195 * extents on the btree inode are pretty simple, there's one extent
196 * that covers the entire device
197 */
198static struct extent_map *btree_get_extent(struct inode *inode,
199 struct page *page, size_t pg_offset, u64 start, u64 len,
200 int create)
201{
202 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
203 struct extent_map *em;
204 int ret;
205
206 read_lock(&em_tree->lock);
207 em = lookup_extent_mapping(em_tree, start, len);
208 if (em) {
209 em->bdev =
210 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
211 read_unlock(&em_tree->lock);
212 goto out;
213 }
214 read_unlock(&em_tree->lock);
215
216 em = alloc_extent_map();
217 if (!em) {
218 em = ERR_PTR(-ENOMEM);
219 goto out;
220 }
221 em->start = 0;
222 em->len = (u64)-1;
223 em->block_len = (u64)-1;
224 em->block_start = 0;
225 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
226
227 write_lock(&em_tree->lock);
228 ret = add_extent_mapping(em_tree, em, 0);
229 if (ret == -EEXIST) {
230 free_extent_map(em);
231 em = lookup_extent_mapping(em_tree, start, len);
232 if (!em)
233 em = ERR_PTR(-EIO);
234 } else if (ret) {
235 free_extent_map(em);
236 em = ERR_PTR(ret);
237 }
238 write_unlock(&em_tree->lock);
239
240out:
241 return em;
242}
243
244u32 btrfs_csum_data(char *data, u32 seed, size_t len)
245{
246 return crc32c(seed, data, len);
247}
248
249void btrfs_csum_final(u32 crc, char *result)
250{
251 put_unaligned_le32(~crc, result);
252}
253
254/*
255 * compute the csum for a btree block, and either verify it or write it
256 * into the csum field of the block.
257 */
258static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
259 int verify)
260{
261 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
262 char *result = NULL;
263 unsigned long len;
264 unsigned long cur_len;
265 unsigned long offset = BTRFS_CSUM_SIZE;
266 char *kaddr;
267 unsigned long map_start;
268 unsigned long map_len;
269 int err;
270 u32 crc = ~(u32)0;
271 unsigned long inline_result;
272
273 len = buf->len - offset;
274 while (len > 0) {
275 err = map_private_extent_buffer(buf, offset, 32,
276 &kaddr, &map_start, &map_len);
277 if (err)
278 return 1;
279 cur_len = min(len, map_len - (offset - map_start));
280 crc = btrfs_csum_data(kaddr + offset - map_start,
281 crc, cur_len);
282 len -= cur_len;
283 offset += cur_len;
284 }
285 if (csum_size > sizeof(inline_result)) {
286 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
287 if (!result)
288 return 1;
289 } else {
290 result = (char *)&inline_result;
291 }
292
293 btrfs_csum_final(crc, result);
294
295 if (verify) {
296 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
297 u32 val;
298 u32 found = 0;
299 memcpy(&found, result, csum_size);
300
301 read_extent_buffer(buf, &val, 0, csum_size);
302 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
303 "failed on %llu wanted %X found %X "
304 "level %d\n",
305 root->fs_info->sb->s_id,
306 (unsigned long long)buf->start, val, found,
307 btrfs_header_level(buf));
308 if (result != (char *)&inline_result)
309 kfree(result);
310 return 1;
311 }
312 } else {
313 write_extent_buffer(buf, result, 0, csum_size);
314 }
315 if (result != (char *)&inline_result)
316 kfree(result);
317 return 0;
318}
319
320/*
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
325 */
326static int verify_parent_transid(struct extent_io_tree *io_tree,
327 struct extent_buffer *eb, u64 parent_transid,
328 int atomic)
329{
330 struct extent_state *cached_state = NULL;
331 int ret;
332
333 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334 return 0;
335
336 if (atomic)
337 return -EAGAIN;
338
339 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
340 0, &cached_state);
341 if (extent_buffer_uptodate(eb) &&
342 btrfs_header_generation(eb) == parent_transid) {
343 ret = 0;
344 goto out;
345 }
346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
347 "found %llu\n",
348 (unsigned long long)eb->start,
349 (unsigned long long)parent_transid,
350 (unsigned long long)btrfs_header_generation(eb));
351 ret = 1;
352 clear_extent_buffer_uptodate(eb);
353out:
354 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
355 &cached_state, GFP_NOFS);
356 return ret;
357}
358
359/*
360 * Return 0 if the superblock checksum type matches the checksum value of that
361 * algorithm. Pass the raw disk superblock data.
362 */
363static int btrfs_check_super_csum(char *raw_disk_sb)
364{
365 struct btrfs_super_block *disk_sb =
366 (struct btrfs_super_block *)raw_disk_sb;
367 u16 csum_type = btrfs_super_csum_type(disk_sb);
368 int ret = 0;
369
370 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
371 u32 crc = ~(u32)0;
372 const int csum_size = sizeof(crc);
373 char result[csum_size];
374
375 /*
376 * The super_block structure does not span the whole
377 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
378 * is filled with zeros and is included in the checkum.
379 */
380 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
381 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
382 btrfs_csum_final(crc, result);
383
384 if (memcmp(raw_disk_sb, result, csum_size))
385 ret = 1;
386
387 if (ret && btrfs_super_generation(disk_sb) < 10) {
388 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
389 ret = 0;
390 }
391 }
392
393 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
394 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
395 csum_type);
396 ret = 1;
397 }
398
399 return ret;
400}
401
402/*
403 * helper to read a given tree block, doing retries as required when
404 * the checksums don't match and we have alternate mirrors to try.
405 */
406static int btree_read_extent_buffer_pages(struct btrfs_root *root,
407 struct extent_buffer *eb,
408 u64 start, u64 parent_transid)
409{
410 struct extent_io_tree *io_tree;
411 int failed = 0;
412 int ret;
413 int num_copies = 0;
414 int mirror_num = 0;
415 int failed_mirror = 0;
416
417 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
418 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
419 while (1) {
420 ret = read_extent_buffer_pages(io_tree, eb, start,
421 WAIT_COMPLETE,
422 btree_get_extent, mirror_num);
423 if (!ret) {
424 if (!verify_parent_transid(io_tree, eb,
425 parent_transid, 0))
426 break;
427 else
428 ret = -EIO;
429 }
430
431 /*
432 * This buffer's crc is fine, but its contents are corrupted, so
433 * there is no reason to read the other copies, they won't be
434 * any less wrong.
435 */
436 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
437 break;
438
439 num_copies = btrfs_num_copies(root->fs_info,
440 eb->start, eb->len);
441 if (num_copies == 1)
442 break;
443
444 if (!failed_mirror) {
445 failed = 1;
446 failed_mirror = eb->read_mirror;
447 }
448
449 mirror_num++;
450 if (mirror_num == failed_mirror)
451 mirror_num++;
452
453 if (mirror_num > num_copies)
454 break;
455 }
456
457 if (failed && !ret && failed_mirror)
458 repair_eb_io_failure(root, eb, failed_mirror);
459
460 return ret;
461}
462
463/*
464 * checksum a dirty tree block before IO. This has extra checks to make sure
465 * we only fill in the checksum field in the first page of a multi-page block
466 */
467
468static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
469{
470 struct extent_io_tree *tree;
471 u64 start = page_offset(page);
472 u64 found_start;
473 struct extent_buffer *eb;
474
475 tree = &BTRFS_I(page->mapping->host)->io_tree;
476
477 eb = (struct extent_buffer *)page->private;
478 if (page != eb->pages[0])
479 return 0;
480 found_start = btrfs_header_bytenr(eb);
481 if (found_start != start) {
482 WARN_ON(1);
483 return 0;
484 }
485 if (!PageUptodate(page)) {
486 WARN_ON(1);
487 return 0;
488 }
489 csum_tree_block(root, eb, 0);
490 return 0;
491}
492
493static int check_tree_block_fsid(struct btrfs_root *root,
494 struct extent_buffer *eb)
495{
496 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
497 u8 fsid[BTRFS_UUID_SIZE];
498 int ret = 1;
499
500 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
501 BTRFS_FSID_SIZE);
502 while (fs_devices) {
503 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
504 ret = 0;
505 break;
506 }
507 fs_devices = fs_devices->seed;
508 }
509 return ret;
510}
511
512#define CORRUPT(reason, eb, root, slot) \
513 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
514 "root=%llu, slot=%d\n", reason, \
515 (unsigned long long)btrfs_header_bytenr(eb), \
516 (unsigned long long)root->objectid, slot)
517
518static noinline int check_leaf(struct btrfs_root *root,
519 struct extent_buffer *leaf)
520{
521 struct btrfs_key key;
522 struct btrfs_key leaf_key;
523 u32 nritems = btrfs_header_nritems(leaf);
524 int slot;
525
526 if (nritems == 0)
527 return 0;
528
529 /* Check the 0 item */
530 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
531 BTRFS_LEAF_DATA_SIZE(root)) {
532 CORRUPT("invalid item offset size pair", leaf, root, 0);
533 return -EIO;
534 }
535
536 /*
537 * Check to make sure each items keys are in the correct order and their
538 * offsets make sense. We only have to loop through nritems-1 because
539 * we check the current slot against the next slot, which verifies the
540 * next slot's offset+size makes sense and that the current's slot
541 * offset is correct.
542 */
543 for (slot = 0; slot < nritems - 1; slot++) {
544 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
545 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
546
547 /* Make sure the keys are in the right order */
548 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
549 CORRUPT("bad key order", leaf, root, slot);
550 return -EIO;
551 }
552
553 /*
554 * Make sure the offset and ends are right, remember that the
555 * item data starts at the end of the leaf and grows towards the
556 * front.
557 */
558 if (btrfs_item_offset_nr(leaf, slot) !=
559 btrfs_item_end_nr(leaf, slot + 1)) {
560 CORRUPT("slot offset bad", leaf, root, slot);
561 return -EIO;
562 }
563
564 /*
565 * Check to make sure that we don't point outside of the leaf,
566 * just incase all the items are consistent to eachother, but
567 * all point outside of the leaf.
568 */
569 if (btrfs_item_end_nr(leaf, slot) >
570 BTRFS_LEAF_DATA_SIZE(root)) {
571 CORRUPT("slot end outside of leaf", leaf, root, slot);
572 return -EIO;
573 }
574 }
575
576 return 0;
577}
578
579static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
580 struct extent_state *state, int mirror)
581{
582 struct extent_io_tree *tree;
583 u64 found_start;
584 int found_level;
585 struct extent_buffer *eb;
586 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
587 int ret = 0;
588 int reads_done;
589
590 if (!page->private)
591 goto out;
592
593 tree = &BTRFS_I(page->mapping->host)->io_tree;
594 eb = (struct extent_buffer *)page->private;
595
596 /* the pending IO might have been the only thing that kept this buffer
597 * in memory. Make sure we have a ref for all this other checks
598 */
599 extent_buffer_get(eb);
600
601 reads_done = atomic_dec_and_test(&eb->io_pages);
602 if (!reads_done)
603 goto err;
604
605 eb->read_mirror = mirror;
606 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
607 ret = -EIO;
608 goto err;
609 }
610
611 found_start = btrfs_header_bytenr(eb);
612 if (found_start != eb->start) {
613 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
614 "%llu %llu\n",
615 (unsigned long long)found_start,
616 (unsigned long long)eb->start);
617 ret = -EIO;
618 goto err;
619 }
620 if (check_tree_block_fsid(root, eb)) {
621 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
622 (unsigned long long)eb->start);
623 ret = -EIO;
624 goto err;
625 }
626 found_level = btrfs_header_level(eb);
627 if (found_level >= BTRFS_MAX_LEVEL) {
628 btrfs_info(root->fs_info, "bad tree block level %d\n",
629 (int)btrfs_header_level(eb));
630 ret = -EIO;
631 goto err;
632 }
633
634 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
635 eb, found_level);
636
637 ret = csum_tree_block(root, eb, 1);
638 if (ret) {
639 ret = -EIO;
640 goto err;
641 }
642
643 /*
644 * If this is a leaf block and it is corrupt, set the corrupt bit so
645 * that we don't try and read the other copies of this block, just
646 * return -EIO.
647 */
648 if (found_level == 0 && check_leaf(root, eb)) {
649 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
650 ret = -EIO;
651 }
652
653 if (!ret)
654 set_extent_buffer_uptodate(eb);
655err:
656 if (reads_done &&
657 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
658 btree_readahead_hook(root, eb, eb->start, ret);
659
660 if (ret) {
661 /*
662 * our io error hook is going to dec the io pages
663 * again, we have to make sure it has something
664 * to decrement
665 */
666 atomic_inc(&eb->io_pages);
667 clear_extent_buffer_uptodate(eb);
668 }
669 free_extent_buffer(eb);
670out:
671 return ret;
672}
673
674static int btree_io_failed_hook(struct page *page, int failed_mirror)
675{
676 struct extent_buffer *eb;
677 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
678
679 eb = (struct extent_buffer *)page->private;
680 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
681 eb->read_mirror = failed_mirror;
682 atomic_dec(&eb->io_pages);
683 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
684 btree_readahead_hook(root, eb, eb->start, -EIO);
685 return -EIO; /* we fixed nothing */
686}
687
688static void end_workqueue_bio(struct bio *bio, int err)
689{
690 struct end_io_wq *end_io_wq = bio->bi_private;
691 struct btrfs_fs_info *fs_info;
692
693 fs_info = end_io_wq->info;
694 end_io_wq->error = err;
695 end_io_wq->work.func = end_workqueue_fn;
696 end_io_wq->work.flags = 0;
697
698 if (bio->bi_rw & REQ_WRITE) {
699 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
700 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
701 &end_io_wq->work);
702 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
703 btrfs_queue_worker(&fs_info->endio_freespace_worker,
704 &end_io_wq->work);
705 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
706 btrfs_queue_worker(&fs_info->endio_raid56_workers,
707 &end_io_wq->work);
708 else
709 btrfs_queue_worker(&fs_info->endio_write_workers,
710 &end_io_wq->work);
711 } else {
712 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
713 btrfs_queue_worker(&fs_info->endio_raid56_workers,
714 &end_io_wq->work);
715 else if (end_io_wq->metadata)
716 btrfs_queue_worker(&fs_info->endio_meta_workers,
717 &end_io_wq->work);
718 else
719 btrfs_queue_worker(&fs_info->endio_workers,
720 &end_io_wq->work);
721 }
722}
723
724/*
725 * For the metadata arg you want
726 *
727 * 0 - if data
728 * 1 - if normal metadta
729 * 2 - if writing to the free space cache area
730 * 3 - raid parity work
731 */
732int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
733 int metadata)
734{
735 struct end_io_wq *end_io_wq;
736 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
737 if (!end_io_wq)
738 return -ENOMEM;
739
740 end_io_wq->private = bio->bi_private;
741 end_io_wq->end_io = bio->bi_end_io;
742 end_io_wq->info = info;
743 end_io_wq->error = 0;
744 end_io_wq->bio = bio;
745 end_io_wq->metadata = metadata;
746
747 bio->bi_private = end_io_wq;
748 bio->bi_end_io = end_workqueue_bio;
749 return 0;
750}
751
752unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
753{
754 unsigned long limit = min_t(unsigned long,
755 info->workers.max_workers,
756 info->fs_devices->open_devices);
757 return 256 * limit;
758}
759
760static void run_one_async_start(struct btrfs_work *work)
761{
762 struct async_submit_bio *async;
763 int ret;
764
765 async = container_of(work, struct async_submit_bio, work);
766 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
767 async->mirror_num, async->bio_flags,
768 async->bio_offset);
769 if (ret)
770 async->error = ret;
771}
772
773static void run_one_async_done(struct btrfs_work *work)
774{
775 struct btrfs_fs_info *fs_info;
776 struct async_submit_bio *async;
777 int limit;
778
779 async = container_of(work, struct async_submit_bio, work);
780 fs_info = BTRFS_I(async->inode)->root->fs_info;
781
782 limit = btrfs_async_submit_limit(fs_info);
783 limit = limit * 2 / 3;
784
785 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
786 waitqueue_active(&fs_info->async_submit_wait))
787 wake_up(&fs_info->async_submit_wait);
788
789 /* If an error occured we just want to clean up the bio and move on */
790 if (async->error) {
791 bio_endio(async->bio, async->error);
792 return;
793 }
794
795 async->submit_bio_done(async->inode, async->rw, async->bio,
796 async->mirror_num, async->bio_flags,
797 async->bio_offset);
798}
799
800static void run_one_async_free(struct btrfs_work *work)
801{
802 struct async_submit_bio *async;
803
804 async = container_of(work, struct async_submit_bio, work);
805 kfree(async);
806}
807
808int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
809 int rw, struct bio *bio, int mirror_num,
810 unsigned long bio_flags,
811 u64 bio_offset,
812 extent_submit_bio_hook_t *submit_bio_start,
813 extent_submit_bio_hook_t *submit_bio_done)
814{
815 struct async_submit_bio *async;
816
817 async = kmalloc(sizeof(*async), GFP_NOFS);
818 if (!async)
819 return -ENOMEM;
820
821 async->inode = inode;
822 async->rw = rw;
823 async->bio = bio;
824 async->mirror_num = mirror_num;
825 async->submit_bio_start = submit_bio_start;
826 async->submit_bio_done = submit_bio_done;
827
828 async->work.func = run_one_async_start;
829 async->work.ordered_func = run_one_async_done;
830 async->work.ordered_free = run_one_async_free;
831
832 async->work.flags = 0;
833 async->bio_flags = bio_flags;
834 async->bio_offset = bio_offset;
835
836 async->error = 0;
837
838 atomic_inc(&fs_info->nr_async_submits);
839
840 if (rw & REQ_SYNC)
841 btrfs_set_work_high_prio(&async->work);
842
843 btrfs_queue_worker(&fs_info->workers, &async->work);
844
845 while (atomic_read(&fs_info->async_submit_draining) &&
846 atomic_read(&fs_info->nr_async_submits)) {
847 wait_event(fs_info->async_submit_wait,
848 (atomic_read(&fs_info->nr_async_submits) == 0));
849 }
850
851 return 0;
852}
853
854static int btree_csum_one_bio(struct bio *bio)
855{
856 struct bio_vec *bvec = bio->bi_io_vec;
857 int bio_index = 0;
858 struct btrfs_root *root;
859 int ret = 0;
860
861 WARN_ON(bio->bi_vcnt <= 0);
862 while (bio_index < bio->bi_vcnt) {
863 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
864 ret = csum_dirty_buffer(root, bvec->bv_page);
865 if (ret)
866 break;
867 bio_index++;
868 bvec++;
869 }
870 return ret;
871}
872
873static int __btree_submit_bio_start(struct inode *inode, int rw,
874 struct bio *bio, int mirror_num,
875 unsigned long bio_flags,
876 u64 bio_offset)
877{
878 /*
879 * when we're called for a write, we're already in the async
880 * submission context. Just jump into btrfs_map_bio
881 */
882 return btree_csum_one_bio(bio);
883}
884
885static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
886 int mirror_num, unsigned long bio_flags,
887 u64 bio_offset)
888{
889 int ret;
890
891 /*
892 * when we're called for a write, we're already in the async
893 * submission context. Just jump into btrfs_map_bio
894 */
895 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
896 if (ret)
897 bio_endio(bio, ret);
898 return ret;
899}
900
901static int check_async_write(struct inode *inode, unsigned long bio_flags)
902{
903 if (bio_flags & EXTENT_BIO_TREE_LOG)
904 return 0;
905#ifdef CONFIG_X86
906 if (cpu_has_xmm4_2)
907 return 0;
908#endif
909 return 1;
910}
911
912static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
913 int mirror_num, unsigned long bio_flags,
914 u64 bio_offset)
915{
916 int async = check_async_write(inode, bio_flags);
917 int ret;
918
919 if (!(rw & REQ_WRITE)) {
920 /*
921 * called for a read, do the setup so that checksum validation
922 * can happen in the async kernel threads
923 */
924 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
925 bio, 1);
926 if (ret)
927 goto out_w_error;
928 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
929 mirror_num, 0);
930 } else if (!async) {
931 ret = btree_csum_one_bio(bio);
932 if (ret)
933 goto out_w_error;
934 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
935 mirror_num, 0);
936 } else {
937 /*
938 * kthread helpers are used to submit writes so that
939 * checksumming can happen in parallel across all CPUs
940 */
941 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
942 inode, rw, bio, mirror_num, 0,
943 bio_offset,
944 __btree_submit_bio_start,
945 __btree_submit_bio_done);
946 }
947
948 if (ret) {
949out_w_error:
950 bio_endio(bio, ret);
951 }
952 return ret;
953}
954
955#ifdef CONFIG_MIGRATION
956static int btree_migratepage(struct address_space *mapping,
957 struct page *newpage, struct page *page,
958 enum migrate_mode mode)
959{
960 /*
961 * we can't safely write a btree page from here,
962 * we haven't done the locking hook
963 */
964 if (PageDirty(page))
965 return -EAGAIN;
966 /*
967 * Buffers may be managed in a filesystem specific way.
968 * We must have no buffers or drop them.
969 */
970 if (page_has_private(page) &&
971 !try_to_release_page(page, GFP_KERNEL))
972 return -EAGAIN;
973 return migrate_page(mapping, newpage, page, mode);
974}
975#endif
976
977
978static int btree_writepages(struct address_space *mapping,
979 struct writeback_control *wbc)
980{
981 struct extent_io_tree *tree;
982 struct btrfs_fs_info *fs_info;
983 int ret;
984
985 tree = &BTRFS_I(mapping->host)->io_tree;
986 if (wbc->sync_mode == WB_SYNC_NONE) {
987
988 if (wbc->for_kupdate)
989 return 0;
990
991 fs_info = BTRFS_I(mapping->host)->root->fs_info;
992 /* this is a bit racy, but that's ok */
993 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
994 BTRFS_DIRTY_METADATA_THRESH);
995 if (ret < 0)
996 return 0;
997 }
998 return btree_write_cache_pages(mapping, wbc);
999}
1000
1001static int btree_readpage(struct file *file, struct page *page)
1002{
1003 struct extent_io_tree *tree;
1004 tree = &BTRFS_I(page->mapping->host)->io_tree;
1005 return extent_read_full_page(tree, page, btree_get_extent, 0);
1006}
1007
1008static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1009{
1010 if (PageWriteback(page) || PageDirty(page))
1011 return 0;
1012
1013 return try_release_extent_buffer(page);
1014}
1015
1016static void btree_invalidatepage(struct page *page, unsigned long offset)
1017{
1018 struct extent_io_tree *tree;
1019 tree = &BTRFS_I(page->mapping->host)->io_tree;
1020 extent_invalidatepage(tree, page, offset);
1021 btree_releasepage(page, GFP_NOFS);
1022 if (PagePrivate(page)) {
1023 printk(KERN_WARNING "btrfs warning page private not zero "
1024 "on page %llu\n", (unsigned long long)page_offset(page));
1025 ClearPagePrivate(page);
1026 set_page_private(page, 0);
1027 page_cache_release(page);
1028 }
1029}
1030
1031static int btree_set_page_dirty(struct page *page)
1032{
1033#ifdef DEBUG
1034 struct extent_buffer *eb;
1035
1036 BUG_ON(!PagePrivate(page));
1037 eb = (struct extent_buffer *)page->private;
1038 BUG_ON(!eb);
1039 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1040 BUG_ON(!atomic_read(&eb->refs));
1041 btrfs_assert_tree_locked(eb);
1042#endif
1043 return __set_page_dirty_nobuffers(page);
1044}
1045
1046static const struct address_space_operations btree_aops = {
1047 .readpage = btree_readpage,
1048 .writepages = btree_writepages,
1049 .releasepage = btree_releasepage,
1050 .invalidatepage = btree_invalidatepage,
1051#ifdef CONFIG_MIGRATION
1052 .migratepage = btree_migratepage,
1053#endif
1054 .set_page_dirty = btree_set_page_dirty,
1055};
1056
1057int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1058 u64 parent_transid)
1059{
1060 struct extent_buffer *buf = NULL;
1061 struct inode *btree_inode = root->fs_info->btree_inode;
1062 int ret = 0;
1063
1064 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1065 if (!buf)
1066 return 0;
1067 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1068 buf, 0, WAIT_NONE, btree_get_extent, 0);
1069 free_extent_buffer(buf);
1070 return ret;
1071}
1072
1073int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1074 int mirror_num, struct extent_buffer **eb)
1075{
1076 struct extent_buffer *buf = NULL;
1077 struct inode *btree_inode = root->fs_info->btree_inode;
1078 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1079 int ret;
1080
1081 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1082 if (!buf)
1083 return 0;
1084
1085 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1086
1087 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1088 btree_get_extent, mirror_num);
1089 if (ret) {
1090 free_extent_buffer(buf);
1091 return ret;
1092 }
1093
1094 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1095 free_extent_buffer(buf);
1096 return -EIO;
1097 } else if (extent_buffer_uptodate(buf)) {
1098 *eb = buf;
1099 } else {
1100 free_extent_buffer(buf);
1101 }
1102 return 0;
1103}
1104
1105struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1106 u64 bytenr, u32 blocksize)
1107{
1108 struct inode *btree_inode = root->fs_info->btree_inode;
1109 struct extent_buffer *eb;
1110 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1111 bytenr, blocksize);
1112 return eb;
1113}
1114
1115struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1116 u64 bytenr, u32 blocksize)
1117{
1118 struct inode *btree_inode = root->fs_info->btree_inode;
1119 struct extent_buffer *eb;
1120
1121 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1122 bytenr, blocksize);
1123 return eb;
1124}
1125
1126
1127int btrfs_write_tree_block(struct extent_buffer *buf)
1128{
1129 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1130 buf->start + buf->len - 1);
1131}
1132
1133int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1134{
1135 return filemap_fdatawait_range(buf->pages[0]->mapping,
1136 buf->start, buf->start + buf->len - 1);
1137}
1138
1139struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1140 u32 blocksize, u64 parent_transid)
1141{
1142 struct extent_buffer *buf = NULL;
1143 int ret;
1144
1145 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1146 if (!buf)
1147 return NULL;
1148
1149 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1150 return buf;
1151
1152}
1153
1154void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1155 struct extent_buffer *buf)
1156{
1157 struct btrfs_fs_info *fs_info = root->fs_info;
1158
1159 if (btrfs_header_generation(buf) ==
1160 fs_info->running_transaction->transid) {
1161 btrfs_assert_tree_locked(buf);
1162
1163 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1164 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1165 -buf->len,
1166 fs_info->dirty_metadata_batch);
1167 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1168 btrfs_set_lock_blocking(buf);
1169 clear_extent_buffer_dirty(buf);
1170 }
1171 }
1172}
1173
1174static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1175 u32 stripesize, struct btrfs_root *root,
1176 struct btrfs_fs_info *fs_info,
1177 u64 objectid)
1178{
1179 root->node = NULL;
1180 root->commit_root = NULL;
1181 root->sectorsize = sectorsize;
1182 root->nodesize = nodesize;
1183 root->leafsize = leafsize;
1184 root->stripesize = stripesize;
1185 root->ref_cows = 0;
1186 root->track_dirty = 0;
1187 root->in_radix = 0;
1188 root->orphan_item_inserted = 0;
1189 root->orphan_cleanup_state = 0;
1190
1191 root->objectid = objectid;
1192 root->last_trans = 0;
1193 root->highest_objectid = 0;
1194 root->nr_delalloc_inodes = 0;
1195 root->nr_ordered_extents = 0;
1196 root->name = NULL;
1197 root->inode_tree = RB_ROOT;
1198 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1199 root->block_rsv = NULL;
1200 root->orphan_block_rsv = NULL;
1201
1202 INIT_LIST_HEAD(&root->dirty_list);
1203 INIT_LIST_HEAD(&root->root_list);
1204 INIT_LIST_HEAD(&root->delalloc_inodes);
1205 INIT_LIST_HEAD(&root->delalloc_root);
1206 INIT_LIST_HEAD(&root->ordered_extents);
1207 INIT_LIST_HEAD(&root->ordered_root);
1208 INIT_LIST_HEAD(&root->logged_list[0]);
1209 INIT_LIST_HEAD(&root->logged_list[1]);
1210 spin_lock_init(&root->orphan_lock);
1211 spin_lock_init(&root->inode_lock);
1212 spin_lock_init(&root->delalloc_lock);
1213 spin_lock_init(&root->ordered_extent_lock);
1214 spin_lock_init(&root->accounting_lock);
1215 spin_lock_init(&root->log_extents_lock[0]);
1216 spin_lock_init(&root->log_extents_lock[1]);
1217 mutex_init(&root->objectid_mutex);
1218 mutex_init(&root->log_mutex);
1219 init_waitqueue_head(&root->log_writer_wait);
1220 init_waitqueue_head(&root->log_commit_wait[0]);
1221 init_waitqueue_head(&root->log_commit_wait[1]);
1222 atomic_set(&root->log_commit[0], 0);
1223 atomic_set(&root->log_commit[1], 0);
1224 atomic_set(&root->log_writers, 0);
1225 atomic_set(&root->log_batch, 0);
1226 atomic_set(&root->orphan_inodes, 0);
1227 atomic_set(&root->refs, 1);
1228 root->log_transid = 0;
1229 root->last_log_commit = 0;
1230 extent_io_tree_init(&root->dirty_log_pages,
1231 fs_info->btree_inode->i_mapping);
1232
1233 memset(&root->root_key, 0, sizeof(root->root_key));
1234 memset(&root->root_item, 0, sizeof(root->root_item));
1235 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1236 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1237 root->defrag_trans_start = fs_info->generation;
1238 init_completion(&root->kobj_unregister);
1239 root->defrag_running = 0;
1240 root->root_key.objectid = objectid;
1241 root->anon_dev = 0;
1242
1243 spin_lock_init(&root->root_item_lock);
1244}
1245
1246static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1247{
1248 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1249 if (root)
1250 root->fs_info = fs_info;
1251 return root;
1252}
1253
1254struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1255 struct btrfs_fs_info *fs_info,
1256 u64 objectid)
1257{
1258 struct extent_buffer *leaf;
1259 struct btrfs_root *tree_root = fs_info->tree_root;
1260 struct btrfs_root *root;
1261 struct btrfs_key key;
1262 int ret = 0;
1263 u64 bytenr;
1264 uuid_le uuid;
1265
1266 root = btrfs_alloc_root(fs_info);
1267 if (!root)
1268 return ERR_PTR(-ENOMEM);
1269
1270 __setup_root(tree_root->nodesize, tree_root->leafsize,
1271 tree_root->sectorsize, tree_root->stripesize,
1272 root, fs_info, objectid);
1273 root->root_key.objectid = objectid;
1274 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1275 root->root_key.offset = 0;
1276
1277 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1278 0, objectid, NULL, 0, 0, 0);
1279 if (IS_ERR(leaf)) {
1280 ret = PTR_ERR(leaf);
1281 leaf = NULL;
1282 goto fail;
1283 }
1284
1285 bytenr = leaf->start;
1286 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1287 btrfs_set_header_bytenr(leaf, leaf->start);
1288 btrfs_set_header_generation(leaf, trans->transid);
1289 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1290 btrfs_set_header_owner(leaf, objectid);
1291 root->node = leaf;
1292
1293 write_extent_buffer(leaf, fs_info->fsid,
1294 (unsigned long)btrfs_header_fsid(leaf),
1295 BTRFS_FSID_SIZE);
1296 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1297 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1298 BTRFS_UUID_SIZE);
1299 btrfs_mark_buffer_dirty(leaf);
1300
1301 root->commit_root = btrfs_root_node(root);
1302 root->track_dirty = 1;
1303
1304
1305 root->root_item.flags = 0;
1306 root->root_item.byte_limit = 0;
1307 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1308 btrfs_set_root_generation(&root->root_item, trans->transid);
1309 btrfs_set_root_level(&root->root_item, 0);
1310 btrfs_set_root_refs(&root->root_item, 1);
1311 btrfs_set_root_used(&root->root_item, leaf->len);
1312 btrfs_set_root_last_snapshot(&root->root_item, 0);
1313 btrfs_set_root_dirid(&root->root_item, 0);
1314 uuid_le_gen(&uuid);
1315 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1316 root->root_item.drop_level = 0;
1317
1318 key.objectid = objectid;
1319 key.type = BTRFS_ROOT_ITEM_KEY;
1320 key.offset = 0;
1321 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1322 if (ret)
1323 goto fail;
1324
1325 btrfs_tree_unlock(leaf);
1326
1327 return root;
1328
1329fail:
1330 if (leaf) {
1331 btrfs_tree_unlock(leaf);
1332 free_extent_buffer(leaf);
1333 }
1334 kfree(root);
1335
1336 return ERR_PTR(ret);
1337}
1338
1339static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1340 struct btrfs_fs_info *fs_info)
1341{
1342 struct btrfs_root *root;
1343 struct btrfs_root *tree_root = fs_info->tree_root;
1344 struct extent_buffer *leaf;
1345
1346 root = btrfs_alloc_root(fs_info);
1347 if (!root)
1348 return ERR_PTR(-ENOMEM);
1349
1350 __setup_root(tree_root->nodesize, tree_root->leafsize,
1351 tree_root->sectorsize, tree_root->stripesize,
1352 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1353
1354 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1355 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1356 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1357 /*
1358 * log trees do not get reference counted because they go away
1359 * before a real commit is actually done. They do store pointers
1360 * to file data extents, and those reference counts still get
1361 * updated (along with back refs to the log tree).
1362 */
1363 root->ref_cows = 0;
1364
1365 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1366 BTRFS_TREE_LOG_OBJECTID, NULL,
1367 0, 0, 0);
1368 if (IS_ERR(leaf)) {
1369 kfree(root);
1370 return ERR_CAST(leaf);
1371 }
1372
1373 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1374 btrfs_set_header_bytenr(leaf, leaf->start);
1375 btrfs_set_header_generation(leaf, trans->transid);
1376 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1377 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1378 root->node = leaf;
1379
1380 write_extent_buffer(root->node, root->fs_info->fsid,
1381 (unsigned long)btrfs_header_fsid(root->node),
1382 BTRFS_FSID_SIZE);
1383 btrfs_mark_buffer_dirty(root->node);
1384 btrfs_tree_unlock(root->node);
1385 return root;
1386}
1387
1388int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1389 struct btrfs_fs_info *fs_info)
1390{
1391 struct btrfs_root *log_root;
1392
1393 log_root = alloc_log_tree(trans, fs_info);
1394 if (IS_ERR(log_root))
1395 return PTR_ERR(log_root);
1396 WARN_ON(fs_info->log_root_tree);
1397 fs_info->log_root_tree = log_root;
1398 return 0;
1399}
1400
1401int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1402 struct btrfs_root *root)
1403{
1404 struct btrfs_root *log_root;
1405 struct btrfs_inode_item *inode_item;
1406
1407 log_root = alloc_log_tree(trans, root->fs_info);
1408 if (IS_ERR(log_root))
1409 return PTR_ERR(log_root);
1410
1411 log_root->last_trans = trans->transid;
1412 log_root->root_key.offset = root->root_key.objectid;
1413
1414 inode_item = &log_root->root_item.inode;
1415 inode_item->generation = cpu_to_le64(1);
1416 inode_item->size = cpu_to_le64(3);
1417 inode_item->nlink = cpu_to_le32(1);
1418 inode_item->nbytes = cpu_to_le64(root->leafsize);
1419 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1420
1421 btrfs_set_root_node(&log_root->root_item, log_root->node);
1422
1423 WARN_ON(root->log_root);
1424 root->log_root = log_root;
1425 root->log_transid = 0;
1426 root->last_log_commit = 0;
1427 return 0;
1428}
1429
1430struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1431 struct btrfs_key *key)
1432{
1433 struct btrfs_root *root;
1434 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1435 struct btrfs_path *path;
1436 u64 generation;
1437 u32 blocksize;
1438 int ret;
1439
1440 path = btrfs_alloc_path();
1441 if (!path)
1442 return ERR_PTR(-ENOMEM);
1443
1444 root = btrfs_alloc_root(fs_info);
1445 if (!root) {
1446 ret = -ENOMEM;
1447 goto alloc_fail;
1448 }
1449
1450 __setup_root(tree_root->nodesize, tree_root->leafsize,
1451 tree_root->sectorsize, tree_root->stripesize,
1452 root, fs_info, key->objectid);
1453
1454 ret = btrfs_find_root(tree_root, key, path,
1455 &root->root_item, &root->root_key);
1456 if (ret) {
1457 if (ret > 0)
1458 ret = -ENOENT;
1459 goto find_fail;
1460 }
1461
1462 generation = btrfs_root_generation(&root->root_item);
1463 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1464 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1465 blocksize, generation);
1466 if (!root->node) {
1467 ret = -ENOMEM;
1468 goto find_fail;
1469 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1470 ret = -EIO;
1471 goto read_fail;
1472 }
1473 root->commit_root = btrfs_root_node(root);
1474out:
1475 btrfs_free_path(path);
1476 return root;
1477
1478read_fail:
1479 free_extent_buffer(root->node);
1480find_fail:
1481 kfree(root);
1482alloc_fail:
1483 root = ERR_PTR(ret);
1484 goto out;
1485}
1486
1487struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1488 struct btrfs_key *location)
1489{
1490 struct btrfs_root *root;
1491
1492 root = btrfs_read_tree_root(tree_root, location);
1493 if (IS_ERR(root))
1494 return root;
1495
1496 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1497 root->ref_cows = 1;
1498 btrfs_check_and_init_root_item(&root->root_item);
1499 }
1500
1501 return root;
1502}
1503
1504int btrfs_init_fs_root(struct btrfs_root *root)
1505{
1506 int ret;
1507
1508 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1509 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1510 GFP_NOFS);
1511 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1512 ret = -ENOMEM;
1513 goto fail;
1514 }
1515
1516 btrfs_init_free_ino_ctl(root);
1517 mutex_init(&root->fs_commit_mutex);
1518 spin_lock_init(&root->cache_lock);
1519 init_waitqueue_head(&root->cache_wait);
1520
1521 ret = get_anon_bdev(&root->anon_dev);
1522 if (ret)
1523 goto fail;
1524 return 0;
1525fail:
1526 kfree(root->free_ino_ctl);
1527 kfree(root->free_ino_pinned);
1528 return ret;
1529}
1530
1531struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1532 u64 root_id)
1533{
1534 struct btrfs_root *root;
1535
1536 spin_lock(&fs_info->fs_roots_radix_lock);
1537 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1538 (unsigned long)root_id);
1539 spin_unlock(&fs_info->fs_roots_radix_lock);
1540 return root;
1541}
1542
1543int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1544 struct btrfs_root *root)
1545{
1546 int ret;
1547
1548 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1549 if (ret)
1550 return ret;
1551
1552 spin_lock(&fs_info->fs_roots_radix_lock);
1553 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1554 (unsigned long)root->root_key.objectid,
1555 root);
1556 if (ret == 0)
1557 root->in_radix = 1;
1558 spin_unlock(&fs_info->fs_roots_radix_lock);
1559 radix_tree_preload_end();
1560
1561 return ret;
1562}
1563
1564struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1565 struct btrfs_key *location)
1566{
1567 struct btrfs_root *root;
1568 int ret;
1569
1570 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1571 return fs_info->tree_root;
1572 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1573 return fs_info->extent_root;
1574 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1575 return fs_info->chunk_root;
1576 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1577 return fs_info->dev_root;
1578 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1579 return fs_info->csum_root;
1580 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1581 return fs_info->quota_root ? fs_info->quota_root :
1582 ERR_PTR(-ENOENT);
1583again:
1584 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1585 if (root)
1586 return root;
1587
1588 root = btrfs_read_fs_root(fs_info->tree_root, location);
1589 if (IS_ERR(root))
1590 return root;
1591
1592 if (btrfs_root_refs(&root->root_item) == 0) {
1593 ret = -ENOENT;
1594 goto fail;
1595 }
1596
1597 ret = btrfs_init_fs_root(root);
1598 if (ret)
1599 goto fail;
1600
1601 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1602 if (ret < 0)
1603 goto fail;
1604 if (ret == 0)
1605 root->orphan_item_inserted = 1;
1606
1607 ret = btrfs_insert_fs_root(fs_info, root);
1608 if (ret) {
1609 if (ret == -EEXIST) {
1610 free_fs_root(root);
1611 goto again;
1612 }
1613 goto fail;
1614 }
1615 return root;
1616fail:
1617 free_fs_root(root);
1618 return ERR_PTR(ret);
1619}
1620
1621static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1622{
1623 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1624 int ret = 0;
1625 struct btrfs_device *device;
1626 struct backing_dev_info *bdi;
1627
1628 rcu_read_lock();
1629 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1630 if (!device->bdev)
1631 continue;
1632 bdi = blk_get_backing_dev_info(device->bdev);
1633 if (bdi && bdi_congested(bdi, bdi_bits)) {
1634 ret = 1;
1635 break;
1636 }
1637 }
1638 rcu_read_unlock();
1639 return ret;
1640}
1641
1642/*
1643 * If this fails, caller must call bdi_destroy() to get rid of the
1644 * bdi again.
1645 */
1646static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1647{
1648 int err;
1649
1650 bdi->capabilities = BDI_CAP_MAP_COPY;
1651 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1652 if (err)
1653 return err;
1654
1655 bdi->ra_pages = default_backing_dev_info.ra_pages;
1656 bdi->congested_fn = btrfs_congested_fn;
1657 bdi->congested_data = info;
1658 return 0;
1659}
1660
1661/*
1662 * called by the kthread helper functions to finally call the bio end_io
1663 * functions. This is where read checksum verification actually happens
1664 */
1665static void end_workqueue_fn(struct btrfs_work *work)
1666{
1667 struct bio *bio;
1668 struct end_io_wq *end_io_wq;
1669 struct btrfs_fs_info *fs_info;
1670 int error;
1671
1672 end_io_wq = container_of(work, struct end_io_wq, work);
1673 bio = end_io_wq->bio;
1674 fs_info = end_io_wq->info;
1675
1676 error = end_io_wq->error;
1677 bio->bi_private = end_io_wq->private;
1678 bio->bi_end_io = end_io_wq->end_io;
1679 kfree(end_io_wq);
1680 bio_endio(bio, error);
1681}
1682
1683static int cleaner_kthread(void *arg)
1684{
1685 struct btrfs_root *root = arg;
1686 int again;
1687
1688 do {
1689 again = 0;
1690
1691 /* Make the cleaner go to sleep early. */
1692 if (btrfs_need_cleaner_sleep(root))
1693 goto sleep;
1694
1695 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1696 goto sleep;
1697
1698 /*
1699 * Avoid the problem that we change the status of the fs
1700 * during the above check and trylock.
1701 */
1702 if (btrfs_need_cleaner_sleep(root)) {
1703 mutex_unlock(&root->fs_info->cleaner_mutex);
1704 goto sleep;
1705 }
1706
1707 btrfs_run_delayed_iputs(root);
1708 again = btrfs_clean_one_deleted_snapshot(root);
1709 mutex_unlock(&root->fs_info->cleaner_mutex);
1710
1711 /*
1712 * The defragger has dealt with the R/O remount and umount,
1713 * needn't do anything special here.
1714 */
1715 btrfs_run_defrag_inodes(root->fs_info);
1716sleep:
1717 if (!try_to_freeze() && !again) {
1718 set_current_state(TASK_INTERRUPTIBLE);
1719 if (!kthread_should_stop())
1720 schedule();
1721 __set_current_state(TASK_RUNNING);
1722 }
1723 } while (!kthread_should_stop());
1724 return 0;
1725}
1726
1727static int transaction_kthread(void *arg)
1728{
1729 struct btrfs_root *root = arg;
1730 struct btrfs_trans_handle *trans;
1731 struct btrfs_transaction *cur;
1732 u64 transid;
1733 unsigned long now;
1734 unsigned long delay;
1735 bool cannot_commit;
1736
1737 do {
1738 cannot_commit = false;
1739 delay = HZ * 30;
1740 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1741
1742 spin_lock(&root->fs_info->trans_lock);
1743 cur = root->fs_info->running_transaction;
1744 if (!cur) {
1745 spin_unlock(&root->fs_info->trans_lock);
1746 goto sleep;
1747 }
1748
1749 now = get_seconds();
1750 if (cur->state < TRANS_STATE_BLOCKED &&
1751 (now < cur->start_time || now - cur->start_time < 30)) {
1752 spin_unlock(&root->fs_info->trans_lock);
1753 delay = HZ * 5;
1754 goto sleep;
1755 }
1756 transid = cur->transid;
1757 spin_unlock(&root->fs_info->trans_lock);
1758
1759 /* If the file system is aborted, this will always fail. */
1760 trans = btrfs_attach_transaction(root);
1761 if (IS_ERR(trans)) {
1762 if (PTR_ERR(trans) != -ENOENT)
1763 cannot_commit = true;
1764 goto sleep;
1765 }
1766 if (transid == trans->transid) {
1767 btrfs_commit_transaction(trans, root);
1768 } else {
1769 btrfs_end_transaction(trans, root);
1770 }
1771sleep:
1772 wake_up_process(root->fs_info->cleaner_kthread);
1773 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1774
1775 if (!try_to_freeze()) {
1776 set_current_state(TASK_INTERRUPTIBLE);
1777 if (!kthread_should_stop() &&
1778 (!btrfs_transaction_blocked(root->fs_info) ||
1779 cannot_commit))
1780 schedule_timeout(delay);
1781 __set_current_state(TASK_RUNNING);
1782 }
1783 } while (!kthread_should_stop());
1784 return 0;
1785}
1786
1787/*
1788 * this will find the highest generation in the array of
1789 * root backups. The index of the highest array is returned,
1790 * or -1 if we can't find anything.
1791 *
1792 * We check to make sure the array is valid by comparing the
1793 * generation of the latest root in the array with the generation
1794 * in the super block. If they don't match we pitch it.
1795 */
1796static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1797{
1798 u64 cur;
1799 int newest_index = -1;
1800 struct btrfs_root_backup *root_backup;
1801 int i;
1802
1803 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1804 root_backup = info->super_copy->super_roots + i;
1805 cur = btrfs_backup_tree_root_gen(root_backup);
1806 if (cur == newest_gen)
1807 newest_index = i;
1808 }
1809
1810 /* check to see if we actually wrapped around */
1811 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1812 root_backup = info->super_copy->super_roots;
1813 cur = btrfs_backup_tree_root_gen(root_backup);
1814 if (cur == newest_gen)
1815 newest_index = 0;
1816 }
1817 return newest_index;
1818}
1819
1820
1821/*
1822 * find the oldest backup so we know where to store new entries
1823 * in the backup array. This will set the backup_root_index
1824 * field in the fs_info struct
1825 */
1826static void find_oldest_super_backup(struct btrfs_fs_info *info,
1827 u64 newest_gen)
1828{
1829 int newest_index = -1;
1830
1831 newest_index = find_newest_super_backup(info, newest_gen);
1832 /* if there was garbage in there, just move along */
1833 if (newest_index == -1) {
1834 info->backup_root_index = 0;
1835 } else {
1836 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1837 }
1838}
1839
1840/*
1841 * copy all the root pointers into the super backup array.
1842 * this will bump the backup pointer by one when it is
1843 * done
1844 */
1845static void backup_super_roots(struct btrfs_fs_info *info)
1846{
1847 int next_backup;
1848 struct btrfs_root_backup *root_backup;
1849 int last_backup;
1850
1851 next_backup = info->backup_root_index;
1852 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1853 BTRFS_NUM_BACKUP_ROOTS;
1854
1855 /*
1856 * just overwrite the last backup if we're at the same generation
1857 * this happens only at umount
1858 */
1859 root_backup = info->super_for_commit->super_roots + last_backup;
1860 if (btrfs_backup_tree_root_gen(root_backup) ==
1861 btrfs_header_generation(info->tree_root->node))
1862 next_backup = last_backup;
1863
1864 root_backup = info->super_for_commit->super_roots + next_backup;
1865
1866 /*
1867 * make sure all of our padding and empty slots get zero filled
1868 * regardless of which ones we use today
1869 */
1870 memset(root_backup, 0, sizeof(*root_backup));
1871
1872 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1873
1874 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1875 btrfs_set_backup_tree_root_gen(root_backup,
1876 btrfs_header_generation(info->tree_root->node));
1877
1878 btrfs_set_backup_tree_root_level(root_backup,
1879 btrfs_header_level(info->tree_root->node));
1880
1881 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1882 btrfs_set_backup_chunk_root_gen(root_backup,
1883 btrfs_header_generation(info->chunk_root->node));
1884 btrfs_set_backup_chunk_root_level(root_backup,
1885 btrfs_header_level(info->chunk_root->node));
1886
1887 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1888 btrfs_set_backup_extent_root_gen(root_backup,
1889 btrfs_header_generation(info->extent_root->node));
1890 btrfs_set_backup_extent_root_level(root_backup,
1891 btrfs_header_level(info->extent_root->node));
1892
1893 /*
1894 * we might commit during log recovery, which happens before we set
1895 * the fs_root. Make sure it is valid before we fill it in.
1896 */
1897 if (info->fs_root && info->fs_root->node) {
1898 btrfs_set_backup_fs_root(root_backup,
1899 info->fs_root->node->start);
1900 btrfs_set_backup_fs_root_gen(root_backup,
1901 btrfs_header_generation(info->fs_root->node));
1902 btrfs_set_backup_fs_root_level(root_backup,
1903 btrfs_header_level(info->fs_root->node));
1904 }
1905
1906 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1907 btrfs_set_backup_dev_root_gen(root_backup,
1908 btrfs_header_generation(info->dev_root->node));
1909 btrfs_set_backup_dev_root_level(root_backup,
1910 btrfs_header_level(info->dev_root->node));
1911
1912 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1913 btrfs_set_backup_csum_root_gen(root_backup,
1914 btrfs_header_generation(info->csum_root->node));
1915 btrfs_set_backup_csum_root_level(root_backup,
1916 btrfs_header_level(info->csum_root->node));
1917
1918 btrfs_set_backup_total_bytes(root_backup,
1919 btrfs_super_total_bytes(info->super_copy));
1920 btrfs_set_backup_bytes_used(root_backup,
1921 btrfs_super_bytes_used(info->super_copy));
1922 btrfs_set_backup_num_devices(root_backup,
1923 btrfs_super_num_devices(info->super_copy));
1924
1925 /*
1926 * if we don't copy this out to the super_copy, it won't get remembered
1927 * for the next commit
1928 */
1929 memcpy(&info->super_copy->super_roots,
1930 &info->super_for_commit->super_roots,
1931 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1932}
1933
1934/*
1935 * this copies info out of the root backup array and back into
1936 * the in-memory super block. It is meant to help iterate through
1937 * the array, so you send it the number of backups you've already
1938 * tried and the last backup index you used.
1939 *
1940 * this returns -1 when it has tried all the backups
1941 */
1942static noinline int next_root_backup(struct btrfs_fs_info *info,
1943 struct btrfs_super_block *super,
1944 int *num_backups_tried, int *backup_index)
1945{
1946 struct btrfs_root_backup *root_backup;
1947 int newest = *backup_index;
1948
1949 if (*num_backups_tried == 0) {
1950 u64 gen = btrfs_super_generation(super);
1951
1952 newest = find_newest_super_backup(info, gen);
1953 if (newest == -1)
1954 return -1;
1955
1956 *backup_index = newest;
1957 *num_backups_tried = 1;
1958 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1959 /* we've tried all the backups, all done */
1960 return -1;
1961 } else {
1962 /* jump to the next oldest backup */
1963 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1964 BTRFS_NUM_BACKUP_ROOTS;
1965 *backup_index = newest;
1966 *num_backups_tried += 1;
1967 }
1968 root_backup = super->super_roots + newest;
1969
1970 btrfs_set_super_generation(super,
1971 btrfs_backup_tree_root_gen(root_backup));
1972 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1973 btrfs_set_super_root_level(super,
1974 btrfs_backup_tree_root_level(root_backup));
1975 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1976
1977 /*
1978 * fixme: the total bytes and num_devices need to match or we should
1979 * need a fsck
1980 */
1981 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1982 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1983 return 0;
1984}
1985
1986/* helper to cleanup workers */
1987static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1988{
1989 btrfs_stop_workers(&fs_info->generic_worker);
1990 btrfs_stop_workers(&fs_info->fixup_workers);
1991 btrfs_stop_workers(&fs_info->delalloc_workers);
1992 btrfs_stop_workers(&fs_info->workers);
1993 btrfs_stop_workers(&fs_info->endio_workers);
1994 btrfs_stop_workers(&fs_info->endio_meta_workers);
1995 btrfs_stop_workers(&fs_info->endio_raid56_workers);
1996 btrfs_stop_workers(&fs_info->rmw_workers);
1997 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1998 btrfs_stop_workers(&fs_info->endio_write_workers);
1999 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2000 btrfs_stop_workers(&fs_info->submit_workers);
2001 btrfs_stop_workers(&fs_info->delayed_workers);
2002 btrfs_stop_workers(&fs_info->caching_workers);
2003 btrfs_stop_workers(&fs_info->readahead_workers);
2004 btrfs_stop_workers(&fs_info->flush_workers);
2005 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2006}
2007
2008/* helper to cleanup tree roots */
2009static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2010{
2011 free_extent_buffer(info->tree_root->node);
2012 free_extent_buffer(info->tree_root->commit_root);
2013 info->tree_root->node = NULL;
2014 info->tree_root->commit_root = NULL;
2015
2016 if (info->dev_root) {
2017 free_extent_buffer(info->dev_root->node);
2018 free_extent_buffer(info->dev_root->commit_root);
2019 info->dev_root->node = NULL;
2020 info->dev_root->commit_root = NULL;
2021 }
2022 if (info->extent_root) {
2023 free_extent_buffer(info->extent_root->node);
2024 free_extent_buffer(info->extent_root->commit_root);
2025 info->extent_root->node = NULL;
2026 info->extent_root->commit_root = NULL;
2027 }
2028 if (info->csum_root) {
2029 free_extent_buffer(info->csum_root->node);
2030 free_extent_buffer(info->csum_root->commit_root);
2031 info->csum_root->node = NULL;
2032 info->csum_root->commit_root = NULL;
2033 }
2034 if (info->quota_root) {
2035 free_extent_buffer(info->quota_root->node);
2036 free_extent_buffer(info->quota_root->commit_root);
2037 info->quota_root->node = NULL;
2038 info->quota_root->commit_root = NULL;
2039 }
2040 if (chunk_root) {
2041 free_extent_buffer(info->chunk_root->node);
2042 free_extent_buffer(info->chunk_root->commit_root);
2043 info->chunk_root->node = NULL;
2044 info->chunk_root->commit_root = NULL;
2045 }
2046}
2047
2048static void del_fs_roots(struct btrfs_fs_info *fs_info)
2049{
2050 int ret;
2051 struct btrfs_root *gang[8];
2052 int i;
2053
2054 while (!list_empty(&fs_info->dead_roots)) {
2055 gang[0] = list_entry(fs_info->dead_roots.next,
2056 struct btrfs_root, root_list);
2057 list_del(&gang[0]->root_list);
2058
2059 if (gang[0]->in_radix) {
2060 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2061 } else {
2062 free_extent_buffer(gang[0]->node);
2063 free_extent_buffer(gang[0]->commit_root);
2064 btrfs_put_fs_root(gang[0]);
2065 }
2066 }
2067
2068 while (1) {
2069 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2070 (void **)gang, 0,
2071 ARRAY_SIZE(gang));
2072 if (!ret)
2073 break;
2074 for (i = 0; i < ret; i++)
2075 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2076 }
2077}
2078
2079int open_ctree(struct super_block *sb,
2080 struct btrfs_fs_devices *fs_devices,
2081 char *options)
2082{
2083 u32 sectorsize;
2084 u32 nodesize;
2085 u32 leafsize;
2086 u32 blocksize;
2087 u32 stripesize;
2088 u64 generation;
2089 u64 features;
2090 struct btrfs_key location;
2091 struct buffer_head *bh;
2092 struct btrfs_super_block *disk_super;
2093 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2094 struct btrfs_root *tree_root;
2095 struct btrfs_root *extent_root;
2096 struct btrfs_root *csum_root;
2097 struct btrfs_root *chunk_root;
2098 struct btrfs_root *dev_root;
2099 struct btrfs_root *quota_root;
2100 struct btrfs_root *log_tree_root;
2101 int ret;
2102 int err = -EINVAL;
2103 int num_backups_tried = 0;
2104 int backup_index = 0;
2105
2106 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2107 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2108 if (!tree_root || !chunk_root) {
2109 err = -ENOMEM;
2110 goto fail;
2111 }
2112
2113 ret = init_srcu_struct(&fs_info->subvol_srcu);
2114 if (ret) {
2115 err = ret;
2116 goto fail;
2117 }
2118
2119 ret = setup_bdi(fs_info, &fs_info->bdi);
2120 if (ret) {
2121 err = ret;
2122 goto fail_srcu;
2123 }
2124
2125 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2126 if (ret) {
2127 err = ret;
2128 goto fail_bdi;
2129 }
2130 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2131 (1 + ilog2(nr_cpu_ids));
2132
2133 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2134 if (ret) {
2135 err = ret;
2136 goto fail_dirty_metadata_bytes;
2137 }
2138
2139 fs_info->btree_inode = new_inode(sb);
2140 if (!fs_info->btree_inode) {
2141 err = -ENOMEM;
2142 goto fail_delalloc_bytes;
2143 }
2144
2145 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2146
2147 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2148 INIT_LIST_HEAD(&fs_info->trans_list);
2149 INIT_LIST_HEAD(&fs_info->dead_roots);
2150 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2151 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2152 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2153 spin_lock_init(&fs_info->delalloc_root_lock);
2154 spin_lock_init(&fs_info->trans_lock);
2155 spin_lock_init(&fs_info->fs_roots_radix_lock);
2156 spin_lock_init(&fs_info->delayed_iput_lock);
2157 spin_lock_init(&fs_info->defrag_inodes_lock);
2158 spin_lock_init(&fs_info->free_chunk_lock);
2159 spin_lock_init(&fs_info->tree_mod_seq_lock);
2160 spin_lock_init(&fs_info->super_lock);
2161 rwlock_init(&fs_info->tree_mod_log_lock);
2162 mutex_init(&fs_info->reloc_mutex);
2163 seqlock_init(&fs_info->profiles_lock);
2164
2165 init_completion(&fs_info->kobj_unregister);
2166 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2167 INIT_LIST_HEAD(&fs_info->space_info);
2168 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2169 btrfs_mapping_init(&fs_info->mapping_tree);
2170 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2171 BTRFS_BLOCK_RSV_GLOBAL);
2172 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2173 BTRFS_BLOCK_RSV_DELALLOC);
2174 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2175 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2176 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2177 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2178 BTRFS_BLOCK_RSV_DELOPS);
2179 atomic_set(&fs_info->nr_async_submits, 0);
2180 atomic_set(&fs_info->async_delalloc_pages, 0);
2181 atomic_set(&fs_info->async_submit_draining, 0);
2182 atomic_set(&fs_info->nr_async_bios, 0);
2183 atomic_set(&fs_info->defrag_running, 0);
2184 atomic64_set(&fs_info->tree_mod_seq, 0);
2185 fs_info->sb = sb;
2186 fs_info->max_inline = 8192 * 1024;
2187 fs_info->metadata_ratio = 0;
2188 fs_info->defrag_inodes = RB_ROOT;
2189 fs_info->free_chunk_space = 0;
2190 fs_info->tree_mod_log = RB_ROOT;
2191
2192 /* readahead state */
2193 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2194 spin_lock_init(&fs_info->reada_lock);
2195
2196 fs_info->thread_pool_size = min_t(unsigned long,
2197 num_online_cpus() + 2, 8);
2198
2199 INIT_LIST_HEAD(&fs_info->ordered_roots);
2200 spin_lock_init(&fs_info->ordered_root_lock);
2201 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2202 GFP_NOFS);
2203 if (!fs_info->delayed_root) {
2204 err = -ENOMEM;
2205 goto fail_iput;
2206 }
2207 btrfs_init_delayed_root(fs_info->delayed_root);
2208
2209 mutex_init(&fs_info->scrub_lock);
2210 atomic_set(&fs_info->scrubs_running, 0);
2211 atomic_set(&fs_info->scrub_pause_req, 0);
2212 atomic_set(&fs_info->scrubs_paused, 0);
2213 atomic_set(&fs_info->scrub_cancel_req, 0);
2214 init_waitqueue_head(&fs_info->scrub_pause_wait);
2215 init_rwsem(&fs_info->scrub_super_lock);
2216 fs_info->scrub_workers_refcnt = 0;
2217#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2218 fs_info->check_integrity_print_mask = 0;
2219#endif
2220
2221 spin_lock_init(&fs_info->balance_lock);
2222 mutex_init(&fs_info->balance_mutex);
2223 atomic_set(&fs_info->balance_running, 0);
2224 atomic_set(&fs_info->balance_pause_req, 0);
2225 atomic_set(&fs_info->balance_cancel_req, 0);
2226 fs_info->balance_ctl = NULL;
2227 init_waitqueue_head(&fs_info->balance_wait_q);
2228
2229 sb->s_blocksize = 4096;
2230 sb->s_blocksize_bits = blksize_bits(4096);
2231 sb->s_bdi = &fs_info->bdi;
2232
2233 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2234 set_nlink(fs_info->btree_inode, 1);
2235 /*
2236 * we set the i_size on the btree inode to the max possible int.
2237 * the real end of the address space is determined by all of
2238 * the devices in the system
2239 */
2240 fs_info->btree_inode->i_size = OFFSET_MAX;
2241 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2242 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2243
2244 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2245 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2246 fs_info->btree_inode->i_mapping);
2247 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2248 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2249
2250 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2251
2252 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2253 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2254 sizeof(struct btrfs_key));
2255 set_bit(BTRFS_INODE_DUMMY,
2256 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2257 insert_inode_hash(fs_info->btree_inode);
2258
2259 spin_lock_init(&fs_info->block_group_cache_lock);
2260 fs_info->block_group_cache_tree = RB_ROOT;
2261 fs_info->first_logical_byte = (u64)-1;
2262
2263 extent_io_tree_init(&fs_info->freed_extents[0],
2264 fs_info->btree_inode->i_mapping);
2265 extent_io_tree_init(&fs_info->freed_extents[1],
2266 fs_info->btree_inode->i_mapping);
2267 fs_info->pinned_extents = &fs_info->freed_extents[0];
2268 fs_info->do_barriers = 1;
2269
2270
2271 mutex_init(&fs_info->ordered_operations_mutex);
2272 mutex_init(&fs_info->tree_log_mutex);
2273 mutex_init(&fs_info->chunk_mutex);
2274 mutex_init(&fs_info->transaction_kthread_mutex);
2275 mutex_init(&fs_info->cleaner_mutex);
2276 mutex_init(&fs_info->volume_mutex);
2277 init_rwsem(&fs_info->extent_commit_sem);
2278 init_rwsem(&fs_info->cleanup_work_sem);
2279 init_rwsem(&fs_info->subvol_sem);
2280 fs_info->dev_replace.lock_owner = 0;
2281 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2282 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2283 mutex_init(&fs_info->dev_replace.lock_management_lock);
2284 mutex_init(&fs_info->dev_replace.lock);
2285
2286 spin_lock_init(&fs_info->qgroup_lock);
2287 mutex_init(&fs_info->qgroup_ioctl_lock);
2288 fs_info->qgroup_tree = RB_ROOT;
2289 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2290 fs_info->qgroup_seq = 1;
2291 fs_info->quota_enabled = 0;
2292 fs_info->pending_quota_state = 0;
2293 fs_info->qgroup_ulist = NULL;
2294 mutex_init(&fs_info->qgroup_rescan_lock);
2295
2296 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2297 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2298
2299 init_waitqueue_head(&fs_info->transaction_throttle);
2300 init_waitqueue_head(&fs_info->transaction_wait);
2301 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2302 init_waitqueue_head(&fs_info->async_submit_wait);
2303
2304 ret = btrfs_alloc_stripe_hash_table(fs_info);
2305 if (ret) {
2306 err = ret;
2307 goto fail_alloc;
2308 }
2309
2310 __setup_root(4096, 4096, 4096, 4096, tree_root,
2311 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2312
2313 invalidate_bdev(fs_devices->latest_bdev);
2314
2315 /*
2316 * Read super block and check the signature bytes only
2317 */
2318 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2319 if (!bh) {
2320 err = -EINVAL;
2321 goto fail_alloc;
2322 }
2323
2324 /*
2325 * We want to check superblock checksum, the type is stored inside.
2326 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2327 */
2328 if (btrfs_check_super_csum(bh->b_data)) {
2329 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2330 err = -EINVAL;
2331 goto fail_alloc;
2332 }
2333
2334 /*
2335 * super_copy is zeroed at allocation time and we never touch the
2336 * following bytes up to INFO_SIZE, the checksum is calculated from
2337 * the whole block of INFO_SIZE
2338 */
2339 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2340 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2341 sizeof(*fs_info->super_for_commit));
2342 brelse(bh);
2343
2344 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2345
2346 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2347 if (ret) {
2348 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2349 err = -EINVAL;
2350 goto fail_alloc;
2351 }
2352
2353 disk_super = fs_info->super_copy;
2354 if (!btrfs_super_root(disk_super))
2355 goto fail_alloc;
2356
2357 /* check FS state, whether FS is broken. */
2358 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2359 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2360
2361 /*
2362 * run through our array of backup supers and setup
2363 * our ring pointer to the oldest one
2364 */
2365 generation = btrfs_super_generation(disk_super);
2366 find_oldest_super_backup(fs_info, generation);
2367
2368 /*
2369 * In the long term, we'll store the compression type in the super
2370 * block, and it'll be used for per file compression control.
2371 */
2372 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2373
2374 ret = btrfs_parse_options(tree_root, options);
2375 if (ret) {
2376 err = ret;
2377 goto fail_alloc;
2378 }
2379
2380 features = btrfs_super_incompat_flags(disk_super) &
2381 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2382 if (features) {
2383 printk(KERN_ERR "BTRFS: couldn't mount because of "
2384 "unsupported optional features (%Lx).\n",
2385 (unsigned long long)features);
2386 err = -EINVAL;
2387 goto fail_alloc;
2388 }
2389
2390 if (btrfs_super_leafsize(disk_super) !=
2391 btrfs_super_nodesize(disk_super)) {
2392 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2393 "blocksizes don't match. node %d leaf %d\n",
2394 btrfs_super_nodesize(disk_super),
2395 btrfs_super_leafsize(disk_super));
2396 err = -EINVAL;
2397 goto fail_alloc;
2398 }
2399 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2400 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2401 "blocksize (%d) was too large\n",
2402 btrfs_super_leafsize(disk_super));
2403 err = -EINVAL;
2404 goto fail_alloc;
2405 }
2406
2407 features = btrfs_super_incompat_flags(disk_super);
2408 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2409 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2410 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2411
2412 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2413 printk(KERN_ERR "btrfs: has skinny extents\n");
2414
2415 /*
2416 * flag our filesystem as having big metadata blocks if
2417 * they are bigger than the page size
2418 */
2419 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2420 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2421 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2422 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2423 }
2424
2425 nodesize = btrfs_super_nodesize(disk_super);
2426 leafsize = btrfs_super_leafsize(disk_super);
2427 sectorsize = btrfs_super_sectorsize(disk_super);
2428 stripesize = btrfs_super_stripesize(disk_super);
2429 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2430 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2431
2432 /*
2433 * mixed block groups end up with duplicate but slightly offset
2434 * extent buffers for the same range. It leads to corruptions
2435 */
2436 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2437 (sectorsize != leafsize)) {
2438 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2439 "are not allowed for mixed block groups on %s\n",
2440 sb->s_id);
2441 goto fail_alloc;
2442 }
2443
2444 /*
2445 * Needn't use the lock because there is no other task which will
2446 * update the flag.
2447 */
2448 btrfs_set_super_incompat_flags(disk_super, features);
2449
2450 features = btrfs_super_compat_ro_flags(disk_super) &
2451 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2452 if (!(sb->s_flags & MS_RDONLY) && features) {
2453 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2454 "unsupported option features (%Lx).\n",
2455 (unsigned long long)features);
2456 err = -EINVAL;
2457 goto fail_alloc;
2458 }
2459
2460 btrfs_init_workers(&fs_info->generic_worker,
2461 "genwork", 1, NULL);
2462
2463 btrfs_init_workers(&fs_info->workers, "worker",
2464 fs_info->thread_pool_size,
2465 &fs_info->generic_worker);
2466
2467 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2468 fs_info->thread_pool_size,
2469 &fs_info->generic_worker);
2470
2471 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2472 fs_info->thread_pool_size,
2473 &fs_info->generic_worker);
2474
2475 btrfs_init_workers(&fs_info->submit_workers, "submit",
2476 min_t(u64, fs_devices->num_devices,
2477 fs_info->thread_pool_size),
2478 &fs_info->generic_worker);
2479
2480 btrfs_init_workers(&fs_info->caching_workers, "cache",
2481 2, &fs_info->generic_worker);
2482
2483 /* a higher idle thresh on the submit workers makes it much more
2484 * likely that bios will be send down in a sane order to the
2485 * devices
2486 */
2487 fs_info->submit_workers.idle_thresh = 64;
2488
2489 fs_info->workers.idle_thresh = 16;
2490 fs_info->workers.ordered = 1;
2491
2492 fs_info->delalloc_workers.idle_thresh = 2;
2493 fs_info->delalloc_workers.ordered = 1;
2494
2495 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2496 &fs_info->generic_worker);
2497 btrfs_init_workers(&fs_info->endio_workers, "endio",
2498 fs_info->thread_pool_size,
2499 &fs_info->generic_worker);
2500 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2501 fs_info->thread_pool_size,
2502 &fs_info->generic_worker);
2503 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2504 "endio-meta-write", fs_info->thread_pool_size,
2505 &fs_info->generic_worker);
2506 btrfs_init_workers(&fs_info->endio_raid56_workers,
2507 "endio-raid56", fs_info->thread_pool_size,
2508 &fs_info->generic_worker);
2509 btrfs_init_workers(&fs_info->rmw_workers,
2510 "rmw", fs_info->thread_pool_size,
2511 &fs_info->generic_worker);
2512 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2513 fs_info->thread_pool_size,
2514 &fs_info->generic_worker);
2515 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2516 1, &fs_info->generic_worker);
2517 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2518 fs_info->thread_pool_size,
2519 &fs_info->generic_worker);
2520 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2521 fs_info->thread_pool_size,
2522 &fs_info->generic_worker);
2523 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2524 &fs_info->generic_worker);
2525
2526 /*
2527 * endios are largely parallel and should have a very
2528 * low idle thresh
2529 */
2530 fs_info->endio_workers.idle_thresh = 4;
2531 fs_info->endio_meta_workers.idle_thresh = 4;
2532 fs_info->endio_raid56_workers.idle_thresh = 4;
2533 fs_info->rmw_workers.idle_thresh = 2;
2534
2535 fs_info->endio_write_workers.idle_thresh = 2;
2536 fs_info->endio_meta_write_workers.idle_thresh = 2;
2537 fs_info->readahead_workers.idle_thresh = 2;
2538
2539 /*
2540 * btrfs_start_workers can really only fail because of ENOMEM so just
2541 * return -ENOMEM if any of these fail.
2542 */
2543 ret = btrfs_start_workers(&fs_info->workers);
2544 ret |= btrfs_start_workers(&fs_info->generic_worker);
2545 ret |= btrfs_start_workers(&fs_info->submit_workers);
2546 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2547 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2548 ret |= btrfs_start_workers(&fs_info->endio_workers);
2549 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2550 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2551 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2552 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2553 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2554 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2555 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2556 ret |= btrfs_start_workers(&fs_info->caching_workers);
2557 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2558 ret |= btrfs_start_workers(&fs_info->flush_workers);
2559 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2560 if (ret) {
2561 err = -ENOMEM;
2562 goto fail_sb_buffer;
2563 }
2564
2565 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2566 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2567 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2568
2569 tree_root->nodesize = nodesize;
2570 tree_root->leafsize = leafsize;
2571 tree_root->sectorsize = sectorsize;
2572 tree_root->stripesize = stripesize;
2573
2574 sb->s_blocksize = sectorsize;
2575 sb->s_blocksize_bits = blksize_bits(sectorsize);
2576
2577 if (disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2578 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2579 goto fail_sb_buffer;
2580 }
2581
2582 if (sectorsize != PAGE_SIZE) {
2583 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2584 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2585 goto fail_sb_buffer;
2586 }
2587
2588 mutex_lock(&fs_info->chunk_mutex);
2589 ret = btrfs_read_sys_array(tree_root);
2590 mutex_unlock(&fs_info->chunk_mutex);
2591 if (ret) {
2592 printk(KERN_WARNING "btrfs: failed to read the system "
2593 "array on %s\n", sb->s_id);
2594 goto fail_sb_buffer;
2595 }
2596
2597 blocksize = btrfs_level_size(tree_root,
2598 btrfs_super_chunk_root_level(disk_super));
2599 generation = btrfs_super_chunk_root_generation(disk_super);
2600
2601 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2602 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2603
2604 chunk_root->node = read_tree_block(chunk_root,
2605 btrfs_super_chunk_root(disk_super),
2606 blocksize, generation);
2607 if (!chunk_root->node ||
2608 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2609 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2610 sb->s_id);
2611 goto fail_tree_roots;
2612 }
2613 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2614 chunk_root->commit_root = btrfs_root_node(chunk_root);
2615
2616 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2617 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2618 BTRFS_UUID_SIZE);
2619
2620 ret = btrfs_read_chunk_tree(chunk_root);
2621 if (ret) {
2622 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2623 sb->s_id);
2624 goto fail_tree_roots;
2625 }
2626
2627 /*
2628 * keep the device that is marked to be the target device for the
2629 * dev_replace procedure
2630 */
2631 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2632
2633 if (!fs_devices->latest_bdev) {
2634 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2635 sb->s_id);
2636 goto fail_tree_roots;
2637 }
2638
2639retry_root_backup:
2640 blocksize = btrfs_level_size(tree_root,
2641 btrfs_super_root_level(disk_super));
2642 generation = btrfs_super_generation(disk_super);
2643
2644 tree_root->node = read_tree_block(tree_root,
2645 btrfs_super_root(disk_super),
2646 blocksize, generation);
2647 if (!tree_root->node ||
2648 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2649 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2650 sb->s_id);
2651
2652 goto recovery_tree_root;
2653 }
2654
2655 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2656 tree_root->commit_root = btrfs_root_node(tree_root);
2657
2658 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2659 location.type = BTRFS_ROOT_ITEM_KEY;
2660 location.offset = 0;
2661
2662 extent_root = btrfs_read_tree_root(tree_root, &location);
2663 if (IS_ERR(extent_root)) {
2664 ret = PTR_ERR(extent_root);
2665 goto recovery_tree_root;
2666 }
2667 extent_root->track_dirty = 1;
2668 fs_info->extent_root = extent_root;
2669
2670 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2671 dev_root = btrfs_read_tree_root(tree_root, &location);
2672 if (IS_ERR(dev_root)) {
2673 ret = PTR_ERR(dev_root);
2674 goto recovery_tree_root;
2675 }
2676 dev_root->track_dirty = 1;
2677 fs_info->dev_root = dev_root;
2678 btrfs_init_devices_late(fs_info);
2679
2680 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2681 csum_root = btrfs_read_tree_root(tree_root, &location);
2682 if (IS_ERR(csum_root)) {
2683 ret = PTR_ERR(csum_root);
2684 goto recovery_tree_root;
2685 }
2686 csum_root->track_dirty = 1;
2687 fs_info->csum_root = csum_root;
2688
2689 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2690 quota_root = btrfs_read_tree_root(tree_root, &location);
2691 if (!IS_ERR(quota_root)) {
2692 quota_root->track_dirty = 1;
2693 fs_info->quota_enabled = 1;
2694 fs_info->pending_quota_state = 1;
2695 fs_info->quota_root = quota_root;
2696 }
2697
2698 fs_info->generation = generation;
2699 fs_info->last_trans_committed = generation;
2700
2701 ret = btrfs_recover_balance(fs_info);
2702 if (ret) {
2703 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2704 goto fail_block_groups;
2705 }
2706
2707 ret = btrfs_init_dev_stats(fs_info);
2708 if (ret) {
2709 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2710 ret);
2711 goto fail_block_groups;
2712 }
2713
2714 ret = btrfs_init_dev_replace(fs_info);
2715 if (ret) {
2716 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2717 goto fail_block_groups;
2718 }
2719
2720 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2721
2722 ret = btrfs_init_space_info(fs_info);
2723 if (ret) {
2724 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2725 goto fail_block_groups;
2726 }
2727
2728 ret = btrfs_read_block_groups(extent_root);
2729 if (ret) {
2730 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2731 goto fail_block_groups;
2732 }
2733 fs_info->num_tolerated_disk_barrier_failures =
2734 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2735 if (fs_info->fs_devices->missing_devices >
2736 fs_info->num_tolerated_disk_barrier_failures &&
2737 !(sb->s_flags & MS_RDONLY)) {
2738 printk(KERN_WARNING
2739 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2740 goto fail_block_groups;
2741 }
2742
2743 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2744 "btrfs-cleaner");
2745 if (IS_ERR(fs_info->cleaner_kthread))
2746 goto fail_block_groups;
2747
2748 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2749 tree_root,
2750 "btrfs-transaction");
2751 if (IS_ERR(fs_info->transaction_kthread))
2752 goto fail_cleaner;
2753
2754 if (!btrfs_test_opt(tree_root, SSD) &&
2755 !btrfs_test_opt(tree_root, NOSSD) &&
2756 !fs_info->fs_devices->rotating) {
2757 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2758 "mode\n");
2759 btrfs_set_opt(fs_info->mount_opt, SSD);
2760 }
2761
2762#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2763 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2764 ret = btrfsic_mount(tree_root, fs_devices,
2765 btrfs_test_opt(tree_root,
2766 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2767 1 : 0,
2768 fs_info->check_integrity_print_mask);
2769 if (ret)
2770 printk(KERN_WARNING "btrfs: failed to initialize"
2771 " integrity check module %s\n", sb->s_id);
2772 }
2773#endif
2774 ret = btrfs_read_qgroup_config(fs_info);
2775 if (ret)
2776 goto fail_trans_kthread;
2777
2778 /* do not make disk changes in broken FS */
2779 if (btrfs_super_log_root(disk_super) != 0) {
2780 u64 bytenr = btrfs_super_log_root(disk_super);
2781
2782 if (fs_devices->rw_devices == 0) {
2783 printk(KERN_WARNING "Btrfs log replay required "
2784 "on RO media\n");
2785 err = -EIO;
2786 goto fail_qgroup;
2787 }
2788 blocksize =
2789 btrfs_level_size(tree_root,
2790 btrfs_super_log_root_level(disk_super));
2791
2792 log_tree_root = btrfs_alloc_root(fs_info);
2793 if (!log_tree_root) {
2794 err = -ENOMEM;
2795 goto fail_qgroup;
2796 }
2797
2798 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2799 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2800
2801 log_tree_root->node = read_tree_block(tree_root, bytenr,
2802 blocksize,
2803 generation + 1);
2804 if (!log_tree_root->node ||
2805 !extent_buffer_uptodate(log_tree_root->node)) {
2806 printk(KERN_ERR "btrfs: failed to read log tree\n");
2807 free_extent_buffer(log_tree_root->node);
2808 kfree(log_tree_root);
2809 goto fail_trans_kthread;
2810 }
2811 /* returns with log_tree_root freed on success */
2812 ret = btrfs_recover_log_trees(log_tree_root);
2813 if (ret) {
2814 btrfs_error(tree_root->fs_info, ret,
2815 "Failed to recover log tree");
2816 free_extent_buffer(log_tree_root->node);
2817 kfree(log_tree_root);
2818 goto fail_trans_kthread;
2819 }
2820
2821 if (sb->s_flags & MS_RDONLY) {
2822 ret = btrfs_commit_super(tree_root);
2823 if (ret)
2824 goto fail_trans_kthread;
2825 }
2826 }
2827
2828 ret = btrfs_find_orphan_roots(tree_root);
2829 if (ret)
2830 goto fail_trans_kthread;
2831
2832 if (!(sb->s_flags & MS_RDONLY)) {
2833 ret = btrfs_cleanup_fs_roots(fs_info);
2834 if (ret)
2835 goto fail_trans_kthread;
2836
2837 ret = btrfs_recover_relocation(tree_root);
2838 if (ret < 0) {
2839 printk(KERN_WARNING
2840 "btrfs: failed to recover relocation\n");
2841 err = -EINVAL;
2842 goto fail_qgroup;
2843 }
2844 }
2845
2846 location.objectid = BTRFS_FS_TREE_OBJECTID;
2847 location.type = BTRFS_ROOT_ITEM_KEY;
2848 location.offset = 0;
2849
2850 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2851 if (IS_ERR(fs_info->fs_root)) {
2852 err = PTR_ERR(fs_info->fs_root);
2853 goto fail_qgroup;
2854 }
2855
2856 if (sb->s_flags & MS_RDONLY)
2857 return 0;
2858
2859 down_read(&fs_info->cleanup_work_sem);
2860 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2861 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2862 up_read(&fs_info->cleanup_work_sem);
2863 close_ctree(tree_root);
2864 return ret;
2865 }
2866 up_read(&fs_info->cleanup_work_sem);
2867
2868 ret = btrfs_resume_balance_async(fs_info);
2869 if (ret) {
2870 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2871 close_ctree(tree_root);
2872 return ret;
2873 }
2874
2875 ret = btrfs_resume_dev_replace_async(fs_info);
2876 if (ret) {
2877 pr_warn("btrfs: failed to resume dev_replace\n");
2878 close_ctree(tree_root);
2879 return ret;
2880 }
2881
2882 btrfs_qgroup_rescan_resume(fs_info);
2883
2884 return 0;
2885
2886fail_qgroup:
2887 btrfs_free_qgroup_config(fs_info);
2888fail_trans_kthread:
2889 kthread_stop(fs_info->transaction_kthread);
2890 btrfs_cleanup_transaction(fs_info->tree_root);
2891 del_fs_roots(fs_info);
2892fail_cleaner:
2893 kthread_stop(fs_info->cleaner_kthread);
2894
2895 /*
2896 * make sure we're done with the btree inode before we stop our
2897 * kthreads
2898 */
2899 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2900
2901fail_block_groups:
2902 btrfs_put_block_group_cache(fs_info);
2903 btrfs_free_block_groups(fs_info);
2904
2905fail_tree_roots:
2906 free_root_pointers(fs_info, 1);
2907 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2908
2909fail_sb_buffer:
2910 btrfs_stop_all_workers(fs_info);
2911fail_alloc:
2912fail_iput:
2913 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2914
2915 iput(fs_info->btree_inode);
2916fail_delalloc_bytes:
2917 percpu_counter_destroy(&fs_info->delalloc_bytes);
2918fail_dirty_metadata_bytes:
2919 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2920fail_bdi:
2921 bdi_destroy(&fs_info->bdi);
2922fail_srcu:
2923 cleanup_srcu_struct(&fs_info->subvol_srcu);
2924fail:
2925 btrfs_free_stripe_hash_table(fs_info);
2926 btrfs_close_devices(fs_info->fs_devices);
2927 return err;
2928
2929recovery_tree_root:
2930 if (!btrfs_test_opt(tree_root, RECOVERY))
2931 goto fail_tree_roots;
2932
2933 free_root_pointers(fs_info, 0);
2934
2935 /* don't use the log in recovery mode, it won't be valid */
2936 btrfs_set_super_log_root(disk_super, 0);
2937
2938 /* we can't trust the free space cache either */
2939 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2940
2941 ret = next_root_backup(fs_info, fs_info->super_copy,
2942 &num_backups_tried, &backup_index);
2943 if (ret == -1)
2944 goto fail_block_groups;
2945 goto retry_root_backup;
2946}
2947
2948static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2949{
2950 if (uptodate) {
2951 set_buffer_uptodate(bh);
2952 } else {
2953 struct btrfs_device *device = (struct btrfs_device *)
2954 bh->b_private;
2955
2956 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2957 "I/O error on %s\n",
2958 rcu_str_deref(device->name));
2959 /* note, we dont' set_buffer_write_io_error because we have
2960 * our own ways of dealing with the IO errors
2961 */
2962 clear_buffer_uptodate(bh);
2963 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2964 }
2965 unlock_buffer(bh);
2966 put_bh(bh);
2967}
2968
2969struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2970{
2971 struct buffer_head *bh;
2972 struct buffer_head *latest = NULL;
2973 struct btrfs_super_block *super;
2974 int i;
2975 u64 transid = 0;
2976 u64 bytenr;
2977
2978 /* we would like to check all the supers, but that would make
2979 * a btrfs mount succeed after a mkfs from a different FS.
2980 * So, we need to add a special mount option to scan for
2981 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2982 */
2983 for (i = 0; i < 1; i++) {
2984 bytenr = btrfs_sb_offset(i);
2985 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2986 break;
2987 bh = __bread(bdev, bytenr / 4096, 4096);
2988 if (!bh)
2989 continue;
2990
2991 super = (struct btrfs_super_block *)bh->b_data;
2992 if (btrfs_super_bytenr(super) != bytenr ||
2993 super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2994 brelse(bh);
2995 continue;
2996 }
2997
2998 if (!latest || btrfs_super_generation(super) > transid) {
2999 brelse(latest);
3000 latest = bh;
3001 transid = btrfs_super_generation(super);
3002 } else {
3003 brelse(bh);
3004 }
3005 }
3006 return latest;
3007}
3008
3009/*
3010 * this should be called twice, once with wait == 0 and
3011 * once with wait == 1. When wait == 0 is done, all the buffer heads
3012 * we write are pinned.
3013 *
3014 * They are released when wait == 1 is done.
3015 * max_mirrors must be the same for both runs, and it indicates how
3016 * many supers on this one device should be written.
3017 *
3018 * max_mirrors == 0 means to write them all.
3019 */
3020static int write_dev_supers(struct btrfs_device *device,
3021 struct btrfs_super_block *sb,
3022 int do_barriers, int wait, int max_mirrors)
3023{
3024 struct buffer_head *bh;
3025 int i;
3026 int ret;
3027 int errors = 0;
3028 u32 crc;
3029 u64 bytenr;
3030
3031 if (max_mirrors == 0)
3032 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3033
3034 for (i = 0; i < max_mirrors; i++) {
3035 bytenr = btrfs_sb_offset(i);
3036 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3037 break;
3038
3039 if (wait) {
3040 bh = __find_get_block(device->bdev, bytenr / 4096,
3041 BTRFS_SUPER_INFO_SIZE);
3042 if (!bh) {
3043 errors++;
3044 continue;
3045 }
3046 wait_on_buffer(bh);
3047 if (!buffer_uptodate(bh))
3048 errors++;
3049
3050 /* drop our reference */
3051 brelse(bh);
3052
3053 /* drop the reference from the wait == 0 run */
3054 brelse(bh);
3055 continue;
3056 } else {
3057 btrfs_set_super_bytenr(sb, bytenr);
3058
3059 crc = ~(u32)0;
3060 crc = btrfs_csum_data((char *)sb +
3061 BTRFS_CSUM_SIZE, crc,
3062 BTRFS_SUPER_INFO_SIZE -
3063 BTRFS_CSUM_SIZE);
3064 btrfs_csum_final(crc, sb->csum);
3065
3066 /*
3067 * one reference for us, and we leave it for the
3068 * caller
3069 */
3070 bh = __getblk(device->bdev, bytenr / 4096,
3071 BTRFS_SUPER_INFO_SIZE);
3072 if (!bh) {
3073 printk(KERN_ERR "btrfs: couldn't get super "
3074 "buffer head for bytenr %Lu\n", bytenr);
3075 errors++;
3076 continue;
3077 }
3078
3079 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3080
3081 /* one reference for submit_bh */
3082 get_bh(bh);
3083
3084 set_buffer_uptodate(bh);
3085 lock_buffer(bh);
3086 bh->b_end_io = btrfs_end_buffer_write_sync;
3087 bh->b_private = device;
3088 }
3089
3090 /*
3091 * we fua the first super. The others we allow
3092 * to go down lazy.
3093 */
3094 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3095 if (ret)
3096 errors++;
3097 }
3098 return errors < i ? 0 : -1;
3099}
3100
3101/*
3102 * endio for the write_dev_flush, this will wake anyone waiting
3103 * for the barrier when it is done
3104 */
3105static void btrfs_end_empty_barrier(struct bio *bio, int err)
3106{
3107 if (err) {
3108 if (err == -EOPNOTSUPP)
3109 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3110 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3111 }
3112 if (bio->bi_private)
3113 complete(bio->bi_private);
3114 bio_put(bio);
3115}
3116
3117/*
3118 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3119 * sent down. With wait == 1, it waits for the previous flush.
3120 *
3121 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3122 * capable
3123 */
3124static int write_dev_flush(struct btrfs_device *device, int wait)
3125{
3126 struct bio *bio;
3127 int ret = 0;
3128
3129 if (device->nobarriers)
3130 return 0;
3131
3132 if (wait) {
3133 bio = device->flush_bio;
3134 if (!bio)
3135 return 0;
3136
3137 wait_for_completion(&device->flush_wait);
3138
3139 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3140 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3141 rcu_str_deref(device->name));
3142 device->nobarriers = 1;
3143 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3144 ret = -EIO;
3145 btrfs_dev_stat_inc_and_print(device,
3146 BTRFS_DEV_STAT_FLUSH_ERRS);
3147 }
3148
3149 /* drop the reference from the wait == 0 run */
3150 bio_put(bio);
3151 device->flush_bio = NULL;
3152
3153 return ret;
3154 }
3155
3156 /*
3157 * one reference for us, and we leave it for the
3158 * caller
3159 */
3160 device->flush_bio = NULL;
3161 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3162 if (!bio)
3163 return -ENOMEM;
3164
3165 bio->bi_end_io = btrfs_end_empty_barrier;
3166 bio->bi_bdev = device->bdev;
3167 init_completion(&device->flush_wait);
3168 bio->bi_private = &device->flush_wait;
3169 device->flush_bio = bio;
3170
3171 bio_get(bio);
3172 btrfsic_submit_bio(WRITE_FLUSH, bio);
3173
3174 return 0;
3175}
3176
3177/*
3178 * send an empty flush down to each device in parallel,
3179 * then wait for them
3180 */
3181static int barrier_all_devices(struct btrfs_fs_info *info)
3182{
3183 struct list_head *head;
3184 struct btrfs_device *dev;
3185 int errors_send = 0;
3186 int errors_wait = 0;
3187 int ret;
3188
3189 /* send down all the barriers */
3190 head = &info->fs_devices->devices;
3191 list_for_each_entry_rcu(dev, head, dev_list) {
3192 if (!dev->bdev) {
3193 errors_send++;
3194 continue;
3195 }
3196 if (!dev->in_fs_metadata || !dev->writeable)
3197 continue;
3198
3199 ret = write_dev_flush(dev, 0);
3200 if (ret)
3201 errors_send++;
3202 }
3203
3204 /* wait for all the barriers */
3205 list_for_each_entry_rcu(dev, head, dev_list) {
3206 if (!dev->bdev) {
3207 errors_wait++;
3208 continue;
3209 }
3210 if (!dev->in_fs_metadata || !dev->writeable)
3211 continue;
3212
3213 ret = write_dev_flush(dev, 1);
3214 if (ret)
3215 errors_wait++;
3216 }
3217 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3218 errors_wait > info->num_tolerated_disk_barrier_failures)
3219 return -EIO;
3220 return 0;
3221}
3222
3223int btrfs_calc_num_tolerated_disk_barrier_failures(
3224 struct btrfs_fs_info *fs_info)
3225{
3226 struct btrfs_ioctl_space_info space;
3227 struct btrfs_space_info *sinfo;
3228 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3229 BTRFS_BLOCK_GROUP_SYSTEM,
3230 BTRFS_BLOCK_GROUP_METADATA,
3231 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3232 int num_types = 4;
3233 int i;
3234 int c;
3235 int num_tolerated_disk_barrier_failures =
3236 (int)fs_info->fs_devices->num_devices;
3237
3238 for (i = 0; i < num_types; i++) {
3239 struct btrfs_space_info *tmp;
3240
3241 sinfo = NULL;
3242 rcu_read_lock();
3243 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3244 if (tmp->flags == types[i]) {
3245 sinfo = tmp;
3246 break;
3247 }
3248 }
3249 rcu_read_unlock();
3250
3251 if (!sinfo)
3252 continue;
3253
3254 down_read(&sinfo->groups_sem);
3255 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3256 if (!list_empty(&sinfo->block_groups[c])) {
3257 u64 flags;
3258
3259 btrfs_get_block_group_info(
3260 &sinfo->block_groups[c], &space);
3261 if (space.total_bytes == 0 ||
3262 space.used_bytes == 0)
3263 continue;
3264 flags = space.flags;
3265 /*
3266 * return
3267 * 0: if dup, single or RAID0 is configured for
3268 * any of metadata, system or data, else
3269 * 1: if RAID5 is configured, or if RAID1 or
3270 * RAID10 is configured and only two mirrors
3271 * are used, else
3272 * 2: if RAID6 is configured, else
3273 * num_mirrors - 1: if RAID1 or RAID10 is
3274 * configured and more than
3275 * 2 mirrors are used.
3276 */
3277 if (num_tolerated_disk_barrier_failures > 0 &&
3278 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3279 BTRFS_BLOCK_GROUP_RAID0)) ||
3280 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3281 == 0)))
3282 num_tolerated_disk_barrier_failures = 0;
3283 else if (num_tolerated_disk_barrier_failures > 1) {
3284 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3285 BTRFS_BLOCK_GROUP_RAID5 |
3286 BTRFS_BLOCK_GROUP_RAID10)) {
3287 num_tolerated_disk_barrier_failures = 1;
3288 } else if (flags &
3289 BTRFS_BLOCK_GROUP_RAID6) {
3290 num_tolerated_disk_barrier_failures = 2;
3291 }
3292 }
3293 }
3294 }
3295 up_read(&sinfo->groups_sem);
3296 }
3297
3298 return num_tolerated_disk_barrier_failures;
3299}
3300
3301static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3302{
3303 struct list_head *head;
3304 struct btrfs_device *dev;
3305 struct btrfs_super_block *sb;
3306 struct btrfs_dev_item *dev_item;
3307 int ret;
3308 int do_barriers;
3309 int max_errors;
3310 int total_errors = 0;
3311 u64 flags;
3312
3313 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3314 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3315 backup_super_roots(root->fs_info);
3316
3317 sb = root->fs_info->super_for_commit;
3318 dev_item = &sb->dev_item;
3319
3320 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3321 head = &root->fs_info->fs_devices->devices;
3322
3323 if (do_barriers) {
3324 ret = barrier_all_devices(root->fs_info);
3325 if (ret) {
3326 mutex_unlock(
3327 &root->fs_info->fs_devices->device_list_mutex);
3328 btrfs_error(root->fs_info, ret,
3329 "errors while submitting device barriers.");
3330 return ret;
3331 }
3332 }
3333
3334 list_for_each_entry_rcu(dev, head, dev_list) {
3335 if (!dev->bdev) {
3336 total_errors++;
3337 continue;
3338 }
3339 if (!dev->in_fs_metadata || !dev->writeable)
3340 continue;
3341
3342 btrfs_set_stack_device_generation(dev_item, 0);
3343 btrfs_set_stack_device_type(dev_item, dev->type);
3344 btrfs_set_stack_device_id(dev_item, dev->devid);
3345 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3346 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3347 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3348 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3349 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3350 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3351 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3352
3353 flags = btrfs_super_flags(sb);
3354 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3355
3356 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3357 if (ret)
3358 total_errors++;
3359 }
3360 if (total_errors > max_errors) {
3361 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3362 total_errors);
3363
3364 /* This shouldn't happen. FUA is masked off if unsupported */
3365 BUG();
3366 }
3367
3368 total_errors = 0;
3369 list_for_each_entry_rcu(dev, head, dev_list) {
3370 if (!dev->bdev)
3371 continue;
3372 if (!dev->in_fs_metadata || !dev->writeable)
3373 continue;
3374
3375 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3376 if (ret)
3377 total_errors++;
3378 }
3379 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3380 if (total_errors > max_errors) {
3381 btrfs_error(root->fs_info, -EIO,
3382 "%d errors while writing supers", total_errors);
3383 return -EIO;
3384 }
3385 return 0;
3386}
3387
3388int write_ctree_super(struct btrfs_trans_handle *trans,
3389 struct btrfs_root *root, int max_mirrors)
3390{
3391 int ret;
3392
3393 ret = write_all_supers(root, max_mirrors);
3394 return ret;
3395}
3396
3397/* Drop a fs root from the radix tree and free it. */
3398void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3399 struct btrfs_root *root)
3400{
3401 spin_lock(&fs_info->fs_roots_radix_lock);
3402 radix_tree_delete(&fs_info->fs_roots_radix,
3403 (unsigned long)root->root_key.objectid);
3404 spin_unlock(&fs_info->fs_roots_radix_lock);
3405
3406 if (btrfs_root_refs(&root->root_item) == 0)
3407 synchronize_srcu(&fs_info->subvol_srcu);
3408
3409 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3410 btrfs_free_log(NULL, root);
3411 btrfs_free_log_root_tree(NULL, fs_info);
3412 }
3413
3414 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3415 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3416 free_fs_root(root);
3417}
3418
3419static void free_fs_root(struct btrfs_root *root)
3420{
3421 iput(root->cache_inode);
3422 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3423 if (root->anon_dev)
3424 free_anon_bdev(root->anon_dev);
3425 free_extent_buffer(root->node);
3426 free_extent_buffer(root->commit_root);
3427 kfree(root->free_ino_ctl);
3428 kfree(root->free_ino_pinned);
3429 kfree(root->name);
3430 btrfs_put_fs_root(root);
3431}
3432
3433void btrfs_free_fs_root(struct btrfs_root *root)
3434{
3435 free_fs_root(root);
3436}
3437
3438int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3439{
3440 u64 root_objectid = 0;
3441 struct btrfs_root *gang[8];
3442 int i;
3443 int ret;
3444
3445 while (1) {
3446 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3447 (void **)gang, root_objectid,
3448 ARRAY_SIZE(gang));
3449 if (!ret)
3450 break;
3451
3452 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3453 for (i = 0; i < ret; i++) {
3454 int err;
3455
3456 root_objectid = gang[i]->root_key.objectid;
3457 err = btrfs_orphan_cleanup(gang[i]);
3458 if (err)
3459 return err;
3460 }
3461 root_objectid++;
3462 }
3463 return 0;
3464}
3465
3466int btrfs_commit_super(struct btrfs_root *root)
3467{
3468 struct btrfs_trans_handle *trans;
3469 int ret;
3470
3471 mutex_lock(&root->fs_info->cleaner_mutex);
3472 btrfs_run_delayed_iputs(root);
3473 mutex_unlock(&root->fs_info->cleaner_mutex);
3474 wake_up_process(root->fs_info->cleaner_kthread);
3475
3476 /* wait until ongoing cleanup work done */
3477 down_write(&root->fs_info->cleanup_work_sem);
3478 up_write(&root->fs_info->cleanup_work_sem);
3479
3480 trans = btrfs_join_transaction(root);
3481 if (IS_ERR(trans))
3482 return PTR_ERR(trans);
3483 ret = btrfs_commit_transaction(trans, root);
3484 if (ret)
3485 return ret;
3486 /* run commit again to drop the original snapshot */
3487 trans = btrfs_join_transaction(root);
3488 if (IS_ERR(trans))
3489 return PTR_ERR(trans);
3490 ret = btrfs_commit_transaction(trans, root);
3491 if (ret)
3492 return ret;
3493 ret = btrfs_write_and_wait_transaction(NULL, root);
3494 if (ret) {
3495 btrfs_error(root->fs_info, ret,
3496 "Failed to sync btree inode to disk.");
3497 return ret;
3498 }
3499
3500 ret = write_ctree_super(NULL, root, 0);
3501 return ret;
3502}
3503
3504int close_ctree(struct btrfs_root *root)
3505{
3506 struct btrfs_fs_info *fs_info = root->fs_info;
3507 int ret;
3508
3509 fs_info->closing = 1;
3510 smp_mb();
3511
3512 /* pause restriper - we want to resume on mount */
3513 btrfs_pause_balance(fs_info);
3514
3515 btrfs_dev_replace_suspend_for_unmount(fs_info);
3516
3517 btrfs_scrub_cancel(fs_info);
3518
3519 /* wait for any defraggers to finish */
3520 wait_event(fs_info->transaction_wait,
3521 (atomic_read(&fs_info->defrag_running) == 0));
3522
3523 /* clear out the rbtree of defraggable inodes */
3524 btrfs_cleanup_defrag_inodes(fs_info);
3525
3526 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3527 ret = btrfs_commit_super(root);
3528 if (ret)
3529 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3530 }
3531
3532 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3533 btrfs_error_commit_super(root);
3534
3535 btrfs_put_block_group_cache(fs_info);
3536
3537 kthread_stop(fs_info->transaction_kthread);
3538 kthread_stop(fs_info->cleaner_kthread);
3539
3540 fs_info->closing = 2;
3541 smp_mb();
3542
3543 btrfs_free_qgroup_config(root->fs_info);
3544
3545 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3546 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3547 percpu_counter_sum(&fs_info->delalloc_bytes));
3548 }
3549
3550 btrfs_free_block_groups(fs_info);
3551
3552 btrfs_stop_all_workers(fs_info);
3553
3554 del_fs_roots(fs_info);
3555
3556 free_root_pointers(fs_info, 1);
3557
3558 iput(fs_info->btree_inode);
3559
3560#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3561 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3562 btrfsic_unmount(root, fs_info->fs_devices);
3563#endif
3564
3565 btrfs_close_devices(fs_info->fs_devices);
3566 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3567
3568 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3569 percpu_counter_destroy(&fs_info->delalloc_bytes);
3570 bdi_destroy(&fs_info->bdi);
3571 cleanup_srcu_struct(&fs_info->subvol_srcu);
3572
3573 btrfs_free_stripe_hash_table(fs_info);
3574
3575 return 0;
3576}
3577
3578int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3579 int atomic)
3580{
3581 int ret;
3582 struct inode *btree_inode = buf->pages[0]->mapping->host;
3583
3584 ret = extent_buffer_uptodate(buf);
3585 if (!ret)
3586 return ret;
3587
3588 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3589 parent_transid, atomic);
3590 if (ret == -EAGAIN)
3591 return ret;
3592 return !ret;
3593}
3594
3595int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3596{
3597 return set_extent_buffer_uptodate(buf);
3598}
3599
3600void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3601{
3602 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3603 u64 transid = btrfs_header_generation(buf);
3604 int was_dirty;
3605
3606 btrfs_assert_tree_locked(buf);
3607 if (transid != root->fs_info->generation)
3608 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3609 "found %llu running %llu\n",
3610 (unsigned long long)buf->start,
3611 (unsigned long long)transid,
3612 (unsigned long long)root->fs_info->generation);
3613 was_dirty = set_extent_buffer_dirty(buf);
3614 if (!was_dirty)
3615 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3616 buf->len,
3617 root->fs_info->dirty_metadata_batch);
3618}
3619
3620static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3621 int flush_delayed)
3622{
3623 /*
3624 * looks as though older kernels can get into trouble with
3625 * this code, they end up stuck in balance_dirty_pages forever
3626 */
3627 int ret;
3628
3629 if (current->flags & PF_MEMALLOC)
3630 return;
3631
3632 if (flush_delayed)
3633 btrfs_balance_delayed_items(root);
3634
3635 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3636 BTRFS_DIRTY_METADATA_THRESH);
3637 if (ret > 0) {
3638 balance_dirty_pages_ratelimited(
3639 root->fs_info->btree_inode->i_mapping);
3640 }
3641 return;
3642}
3643
3644void btrfs_btree_balance_dirty(struct btrfs_root *root)
3645{
3646 __btrfs_btree_balance_dirty(root, 1);
3647}
3648
3649void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3650{
3651 __btrfs_btree_balance_dirty(root, 0);
3652}
3653
3654int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3655{
3656 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3657 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3658}
3659
3660static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3661 int read_only)
3662{
3663 /*
3664 * Placeholder for checks
3665 */
3666 return 0;
3667}
3668
3669static void btrfs_error_commit_super(struct btrfs_root *root)
3670{
3671 mutex_lock(&root->fs_info->cleaner_mutex);
3672 btrfs_run_delayed_iputs(root);
3673 mutex_unlock(&root->fs_info->cleaner_mutex);
3674
3675 down_write(&root->fs_info->cleanup_work_sem);
3676 up_write(&root->fs_info->cleanup_work_sem);
3677
3678 /* cleanup FS via transaction */
3679 btrfs_cleanup_transaction(root);
3680}
3681
3682static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3683 struct btrfs_root *root)
3684{
3685 struct btrfs_inode *btrfs_inode;
3686 struct list_head splice;
3687
3688 INIT_LIST_HEAD(&splice);
3689
3690 mutex_lock(&root->fs_info->ordered_operations_mutex);
3691 spin_lock(&root->fs_info->ordered_root_lock);
3692
3693 list_splice_init(&t->ordered_operations, &splice);
3694 while (!list_empty(&splice)) {
3695 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3696 ordered_operations);
3697
3698 list_del_init(&btrfs_inode->ordered_operations);
3699 spin_unlock(&root->fs_info->ordered_root_lock);
3700
3701 btrfs_invalidate_inodes(btrfs_inode->root);
3702
3703 spin_lock(&root->fs_info->ordered_root_lock);
3704 }
3705
3706 spin_unlock(&root->fs_info->ordered_root_lock);
3707 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3708}
3709
3710static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3711{
3712 struct btrfs_ordered_extent *ordered;
3713
3714 spin_lock(&root->ordered_extent_lock);
3715 /*
3716 * This will just short circuit the ordered completion stuff which will
3717 * make sure the ordered extent gets properly cleaned up.
3718 */
3719 list_for_each_entry(ordered, &root->ordered_extents,
3720 root_extent_list)
3721 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3722 spin_unlock(&root->ordered_extent_lock);
3723}
3724
3725static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3726{
3727 struct btrfs_root *root;
3728 struct list_head splice;
3729
3730 INIT_LIST_HEAD(&splice);
3731
3732 spin_lock(&fs_info->ordered_root_lock);
3733 list_splice_init(&fs_info->ordered_roots, &splice);
3734 while (!list_empty(&splice)) {
3735 root = list_first_entry(&splice, struct btrfs_root,
3736 ordered_root);
3737 list_del_init(&root->ordered_root);
3738
3739 btrfs_destroy_ordered_extents(root);
3740
3741 cond_resched_lock(&fs_info->ordered_root_lock);
3742 }
3743 spin_unlock(&fs_info->ordered_root_lock);
3744}
3745
3746int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3747 struct btrfs_root *root)
3748{
3749 struct rb_node *node;
3750 struct btrfs_delayed_ref_root *delayed_refs;
3751 struct btrfs_delayed_ref_node *ref;
3752 int ret = 0;
3753
3754 delayed_refs = &trans->delayed_refs;
3755
3756 spin_lock(&delayed_refs->lock);
3757 if (delayed_refs->num_entries == 0) {
3758 spin_unlock(&delayed_refs->lock);
3759 printk(KERN_INFO "delayed_refs has NO entry\n");
3760 return ret;
3761 }
3762
3763 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3764 struct btrfs_delayed_ref_head *head = NULL;
3765
3766 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3767 atomic_set(&ref->refs, 1);
3768 if (btrfs_delayed_ref_is_head(ref)) {
3769
3770 head = btrfs_delayed_node_to_head(ref);
3771 if (!mutex_trylock(&head->mutex)) {
3772 atomic_inc(&ref->refs);
3773 spin_unlock(&delayed_refs->lock);
3774
3775 /* Need to wait for the delayed ref to run */
3776 mutex_lock(&head->mutex);
3777 mutex_unlock(&head->mutex);
3778 btrfs_put_delayed_ref(ref);
3779
3780 spin_lock(&delayed_refs->lock);
3781 continue;
3782 }
3783
3784 if (head->must_insert_reserved)
3785 btrfs_pin_extent(root, ref->bytenr,
3786 ref->num_bytes, 1);
3787 btrfs_free_delayed_extent_op(head->extent_op);
3788 delayed_refs->num_heads--;
3789 if (list_empty(&head->cluster))
3790 delayed_refs->num_heads_ready--;
3791 list_del_init(&head->cluster);
3792 }
3793
3794 ref->in_tree = 0;
3795 rb_erase(&ref->rb_node, &delayed_refs->root);
3796 delayed_refs->num_entries--;
3797 if (head)
3798 mutex_unlock(&head->mutex);
3799 spin_unlock(&delayed_refs->lock);
3800 btrfs_put_delayed_ref(ref);
3801
3802 cond_resched();
3803 spin_lock(&delayed_refs->lock);
3804 }
3805
3806 spin_unlock(&delayed_refs->lock);
3807
3808 return ret;
3809}
3810
3811static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3812{
3813 struct btrfs_pending_snapshot *snapshot;
3814 struct list_head splice;
3815
3816 INIT_LIST_HEAD(&splice);
3817
3818 list_splice_init(&t->pending_snapshots, &splice);
3819
3820 while (!list_empty(&splice)) {
3821 snapshot = list_entry(splice.next,
3822 struct btrfs_pending_snapshot,
3823 list);
3824 snapshot->error = -ECANCELED;
3825 list_del_init(&snapshot->list);
3826 }
3827}
3828
3829static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3830{
3831 struct btrfs_inode *btrfs_inode;
3832 struct list_head splice;
3833
3834 INIT_LIST_HEAD(&splice);
3835
3836 spin_lock(&root->delalloc_lock);
3837 list_splice_init(&root->delalloc_inodes, &splice);
3838
3839 while (!list_empty(&splice)) {
3840 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3841 delalloc_inodes);
3842
3843 list_del_init(&btrfs_inode->delalloc_inodes);
3844 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3845 &btrfs_inode->runtime_flags);
3846 spin_unlock(&root->delalloc_lock);
3847
3848 btrfs_invalidate_inodes(btrfs_inode->root);
3849
3850 spin_lock(&root->delalloc_lock);
3851 }
3852
3853 spin_unlock(&root->delalloc_lock);
3854}
3855
3856static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3857{
3858 struct btrfs_root *root;
3859 struct list_head splice;
3860
3861 INIT_LIST_HEAD(&splice);
3862
3863 spin_lock(&fs_info->delalloc_root_lock);
3864 list_splice_init(&fs_info->delalloc_roots, &splice);
3865 while (!list_empty(&splice)) {
3866 root = list_first_entry(&splice, struct btrfs_root,
3867 delalloc_root);
3868 list_del_init(&root->delalloc_root);
3869 root = btrfs_grab_fs_root(root);
3870 BUG_ON(!root);
3871 spin_unlock(&fs_info->delalloc_root_lock);
3872
3873 btrfs_destroy_delalloc_inodes(root);
3874 btrfs_put_fs_root(root);
3875
3876 spin_lock(&fs_info->delalloc_root_lock);
3877 }
3878 spin_unlock(&fs_info->delalloc_root_lock);
3879}
3880
3881static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3882 struct extent_io_tree *dirty_pages,
3883 int mark)
3884{
3885 int ret;
3886 struct extent_buffer *eb;
3887 u64 start = 0;
3888 u64 end;
3889
3890 while (1) {
3891 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3892 mark, NULL);
3893 if (ret)
3894 break;
3895
3896 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3897 while (start <= end) {
3898 eb = btrfs_find_tree_block(root, start,
3899 root->leafsize);
3900 start += root->leafsize;
3901 if (!eb)
3902 continue;
3903 wait_on_extent_buffer_writeback(eb);
3904
3905 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3906 &eb->bflags))
3907 clear_extent_buffer_dirty(eb);
3908 free_extent_buffer_stale(eb);
3909 }
3910 }
3911
3912 return ret;
3913}
3914
3915static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3916 struct extent_io_tree *pinned_extents)
3917{
3918 struct extent_io_tree *unpin;
3919 u64 start;
3920 u64 end;
3921 int ret;
3922 bool loop = true;
3923
3924 unpin = pinned_extents;
3925again:
3926 while (1) {
3927 ret = find_first_extent_bit(unpin, 0, &start, &end,
3928 EXTENT_DIRTY, NULL);
3929 if (ret)
3930 break;
3931
3932 /* opt_discard */
3933 if (btrfs_test_opt(root, DISCARD))
3934 ret = btrfs_error_discard_extent(root, start,
3935 end + 1 - start,
3936 NULL);
3937
3938 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3939 btrfs_error_unpin_extent_range(root, start, end);
3940 cond_resched();
3941 }
3942
3943 if (loop) {
3944 if (unpin == &root->fs_info->freed_extents[0])
3945 unpin = &root->fs_info->freed_extents[1];
3946 else
3947 unpin = &root->fs_info->freed_extents[0];
3948 loop = false;
3949 goto again;
3950 }
3951
3952 return 0;
3953}
3954
3955void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3956 struct btrfs_root *root)
3957{
3958 btrfs_destroy_delayed_refs(cur_trans, root);
3959 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3960 cur_trans->dirty_pages.dirty_bytes);
3961
3962 cur_trans->state = TRANS_STATE_COMMIT_START;
3963 wake_up(&root->fs_info->transaction_blocked_wait);
3964
3965 btrfs_evict_pending_snapshots(cur_trans);
3966
3967 cur_trans->state = TRANS_STATE_UNBLOCKED;
3968 wake_up(&root->fs_info->transaction_wait);
3969
3970 btrfs_destroy_delayed_inodes(root);
3971 btrfs_assert_delayed_root_empty(root);
3972
3973 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3974 EXTENT_DIRTY);
3975 btrfs_destroy_pinned_extent(root,
3976 root->fs_info->pinned_extents);
3977
3978 cur_trans->state =TRANS_STATE_COMPLETED;
3979 wake_up(&cur_trans->commit_wait);
3980
3981 /*
3982 memset(cur_trans, 0, sizeof(*cur_trans));
3983 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3984 */
3985}
3986
3987static int btrfs_cleanup_transaction(struct btrfs_root *root)
3988{
3989 struct btrfs_transaction *t;
3990 LIST_HEAD(list);
3991
3992 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3993
3994 spin_lock(&root->fs_info->trans_lock);
3995 list_splice_init(&root->fs_info->trans_list, &list);
3996 root->fs_info->running_transaction = NULL;
3997 spin_unlock(&root->fs_info->trans_lock);
3998
3999 while (!list_empty(&list)) {
4000 t = list_entry(list.next, struct btrfs_transaction, list);
4001
4002 btrfs_destroy_ordered_operations(t, root);
4003
4004 btrfs_destroy_all_ordered_extents(root->fs_info);
4005
4006 btrfs_destroy_delayed_refs(t, root);
4007
4008 /*
4009 * FIXME: cleanup wait for commit
4010 * We needn't acquire the lock here, because we are during
4011 * the umount, there is no other task which will change it.
4012 */
4013 t->state = TRANS_STATE_COMMIT_START;
4014 smp_mb();
4015 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4016 wake_up(&root->fs_info->transaction_blocked_wait);
4017
4018 btrfs_evict_pending_snapshots(t);
4019
4020 t->state = TRANS_STATE_UNBLOCKED;
4021 smp_mb();
4022 if (waitqueue_active(&root->fs_info->transaction_wait))
4023 wake_up(&root->fs_info->transaction_wait);
4024
4025 btrfs_destroy_delayed_inodes(root);
4026 btrfs_assert_delayed_root_empty(root);
4027
4028 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4029
4030 btrfs_destroy_marked_extents(root, &t->dirty_pages,
4031 EXTENT_DIRTY);
4032
4033 btrfs_destroy_pinned_extent(root,
4034 root->fs_info->pinned_extents);
4035
4036 t->state = TRANS_STATE_COMPLETED;
4037 smp_mb();
4038 if (waitqueue_active(&t->commit_wait))
4039 wake_up(&t->commit_wait);
4040
4041 atomic_set(&t->use_count, 0);
4042 list_del_init(&t->list);
4043 memset(t, 0, sizeof(*t));
4044 kmem_cache_free(btrfs_transaction_cachep, t);
4045 }
4046
4047 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4048
4049 return 0;
4050}
4051
4052static struct extent_io_ops btree_extent_io_ops = {
4053 .readpage_end_io_hook = btree_readpage_end_io_hook,
4054 .readpage_io_failed_hook = btree_io_failed_hook,
4055 .submit_bio_hook = btree_submit_bio_hook,
4056 /* note we're sharing with inode.c for the merge bio hook */
4057 .merge_bio_hook = btrfs_merge_bio_hook,
4058};
Linux 6.x block layer and io_uring tree(s)