projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: cleanup the similar code of the fs root read
[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->name = NULL;
1195 root->inode_tree = RB_ROOT;
1196 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1197 root->block_rsv = NULL;
1198 root->orphan_block_rsv = NULL;
1199
1200 INIT_LIST_HEAD(&root->dirty_list);
1201 INIT_LIST_HEAD(&root->root_list);
1202 INIT_LIST_HEAD(&root->logged_list[0]);
1203 INIT_LIST_HEAD(&root->logged_list[1]);
1204 spin_lock_init(&root->orphan_lock);
1205 spin_lock_init(&root->inode_lock);
1206 spin_lock_init(&root->accounting_lock);
1207 spin_lock_init(&root->log_extents_lock[0]);
1208 spin_lock_init(&root->log_extents_lock[1]);
1209 mutex_init(&root->objectid_mutex);
1210 mutex_init(&root->log_mutex);
1211 init_waitqueue_head(&root->log_writer_wait);
1212 init_waitqueue_head(&root->log_commit_wait[0]);
1213 init_waitqueue_head(&root->log_commit_wait[1]);
1214 atomic_set(&root->log_commit[0], 0);
1215 atomic_set(&root->log_commit[1], 0);
1216 atomic_set(&root->log_writers, 0);
1217 atomic_set(&root->log_batch, 0);
1218 atomic_set(&root->orphan_inodes, 0);
1219 root->log_transid = 0;
1220 root->last_log_commit = 0;
1221 extent_io_tree_init(&root->dirty_log_pages,
1222 fs_info->btree_inode->i_mapping);
1223
1224 memset(&root->root_key, 0, sizeof(root->root_key));
1225 memset(&root->root_item, 0, sizeof(root->root_item));
1226 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1227 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1228 root->defrag_trans_start = fs_info->generation;
1229 init_completion(&root->kobj_unregister);
1230 root->defrag_running = 0;
1231 root->root_key.objectid = objectid;
1232 root->anon_dev = 0;
1233
1234 spin_lock_init(&root->root_item_lock);
1235}
1236
1237static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1238{
1239 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1240 if (root)
1241 root->fs_info = fs_info;
1242 return root;
1243}
1244
1245struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1246 struct btrfs_fs_info *fs_info,
1247 u64 objectid)
1248{
1249 struct extent_buffer *leaf;
1250 struct btrfs_root *tree_root = fs_info->tree_root;
1251 struct btrfs_root *root;
1252 struct btrfs_key key;
1253 int ret = 0;
1254 u64 bytenr;
1255 uuid_le uuid;
1256
1257 root = btrfs_alloc_root(fs_info);
1258 if (!root)
1259 return ERR_PTR(-ENOMEM);
1260
1261 __setup_root(tree_root->nodesize, tree_root->leafsize,
1262 tree_root->sectorsize, tree_root->stripesize,
1263 root, fs_info, objectid);
1264 root->root_key.objectid = objectid;
1265 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1266 root->root_key.offset = 0;
1267
1268 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1269 0, objectid, NULL, 0, 0, 0);
1270 if (IS_ERR(leaf)) {
1271 ret = PTR_ERR(leaf);
1272 leaf = NULL;
1273 goto fail;
1274 }
1275
1276 bytenr = leaf->start;
1277 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1278 btrfs_set_header_bytenr(leaf, leaf->start);
1279 btrfs_set_header_generation(leaf, trans->transid);
1280 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1281 btrfs_set_header_owner(leaf, objectid);
1282 root->node = leaf;
1283
1284 write_extent_buffer(leaf, fs_info->fsid,
1285 (unsigned long)btrfs_header_fsid(leaf),
1286 BTRFS_FSID_SIZE);
1287 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1288 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1289 BTRFS_UUID_SIZE);
1290 btrfs_mark_buffer_dirty(leaf);
1291
1292 root->commit_root = btrfs_root_node(root);
1293 root->track_dirty = 1;
1294
1295
1296 root->root_item.flags = 0;
1297 root->root_item.byte_limit = 0;
1298 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1299 btrfs_set_root_generation(&root->root_item, trans->transid);
1300 btrfs_set_root_level(&root->root_item, 0);
1301 btrfs_set_root_refs(&root->root_item, 1);
1302 btrfs_set_root_used(&root->root_item, leaf->len);
1303 btrfs_set_root_last_snapshot(&root->root_item, 0);
1304 btrfs_set_root_dirid(&root->root_item, 0);
1305 uuid_le_gen(&uuid);
1306 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1307 root->root_item.drop_level = 0;
1308
1309 key.objectid = objectid;
1310 key.type = BTRFS_ROOT_ITEM_KEY;
1311 key.offset = 0;
1312 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1313 if (ret)
1314 goto fail;
1315
1316 btrfs_tree_unlock(leaf);
1317
1318 return root;
1319
1320fail:
1321 if (leaf) {
1322 btrfs_tree_unlock(leaf);
1323 free_extent_buffer(leaf);
1324 }
1325 kfree(root);
1326
1327 return ERR_PTR(ret);
1328}
1329
1330static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1331 struct btrfs_fs_info *fs_info)
1332{
1333 struct btrfs_root *root;
1334 struct btrfs_root *tree_root = fs_info->tree_root;
1335 struct extent_buffer *leaf;
1336
1337 root = btrfs_alloc_root(fs_info);
1338 if (!root)
1339 return ERR_PTR(-ENOMEM);
1340
1341 __setup_root(tree_root->nodesize, tree_root->leafsize,
1342 tree_root->sectorsize, tree_root->stripesize,
1343 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1344
1345 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1346 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1347 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1348 /*
1349 * log trees do not get reference counted because they go away
1350 * before a real commit is actually done. They do store pointers
1351 * to file data extents, and those reference counts still get
1352 * updated (along with back refs to the log tree).
1353 */
1354 root->ref_cows = 0;
1355
1356 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1357 BTRFS_TREE_LOG_OBJECTID, NULL,
1358 0, 0, 0);
1359 if (IS_ERR(leaf)) {
1360 kfree(root);
1361 return ERR_CAST(leaf);
1362 }
1363
1364 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1365 btrfs_set_header_bytenr(leaf, leaf->start);
1366 btrfs_set_header_generation(leaf, trans->transid);
1367 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1368 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1369 root->node = leaf;
1370
1371 write_extent_buffer(root->node, root->fs_info->fsid,
1372 (unsigned long)btrfs_header_fsid(root->node),
1373 BTRFS_FSID_SIZE);
1374 btrfs_mark_buffer_dirty(root->node);
1375 btrfs_tree_unlock(root->node);
1376 return root;
1377}
1378
1379int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1380 struct btrfs_fs_info *fs_info)
1381{
1382 struct btrfs_root *log_root;
1383
1384 log_root = alloc_log_tree(trans, fs_info);
1385 if (IS_ERR(log_root))
1386 return PTR_ERR(log_root);
1387 WARN_ON(fs_info->log_root_tree);
1388 fs_info->log_root_tree = log_root;
1389 return 0;
1390}
1391
1392int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1393 struct btrfs_root *root)
1394{
1395 struct btrfs_root *log_root;
1396 struct btrfs_inode_item *inode_item;
1397
1398 log_root = alloc_log_tree(trans, root->fs_info);
1399 if (IS_ERR(log_root))
1400 return PTR_ERR(log_root);
1401
1402 log_root->last_trans = trans->transid;
1403 log_root->root_key.offset = root->root_key.objectid;
1404
1405 inode_item = &log_root->root_item.inode;
1406 inode_item->generation = cpu_to_le64(1);
1407 inode_item->size = cpu_to_le64(3);
1408 inode_item->nlink = cpu_to_le32(1);
1409 inode_item->nbytes = cpu_to_le64(root->leafsize);
1410 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1411
1412 btrfs_set_root_node(&log_root->root_item, log_root->node);
1413
1414 WARN_ON(root->log_root);
1415 root->log_root = log_root;
1416 root->log_transid = 0;
1417 root->last_log_commit = 0;
1418 return 0;
1419}
1420
1421struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1422 struct btrfs_key *key)
1423{
1424 struct btrfs_root *root;
1425 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1426 struct btrfs_path *path;
1427 u64 generation;
1428 u32 blocksize;
1429 int ret;
1430
1431 path = btrfs_alloc_path();
1432 if (!path)
1433 return ERR_PTR(-ENOMEM);
1434
1435 root = btrfs_alloc_root(fs_info);
1436 if (!root) {
1437 ret = -ENOMEM;
1438 goto alloc_fail;
1439 }
1440
1441 __setup_root(tree_root->nodesize, tree_root->leafsize,
1442 tree_root->sectorsize, tree_root->stripesize,
1443 root, fs_info, key->objectid);
1444
1445 ret = btrfs_find_root(tree_root, key, path,
1446 &root->root_item, &root->root_key);
1447 if (ret) {
1448 if (ret > 0)
1449 ret = -ENOENT;
1450 goto find_fail;
1451 }
1452
1453 generation = btrfs_root_generation(&root->root_item);
1454 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1455 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1456 blocksize, generation);
1457 if (!root->node) {
1458 ret = -ENOMEM;
1459 goto find_fail;
1460 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1461 ret = -EIO;
1462 goto read_fail;
1463 }
1464 root->commit_root = btrfs_root_node(root);
1465out:
1466 btrfs_free_path(path);
1467 return root;
1468
1469read_fail:
1470 free_extent_buffer(root->node);
1471find_fail:
1472 kfree(root);
1473alloc_fail:
1474 root = ERR_PTR(ret);
1475 goto out;
1476}
1477
1478struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1479 struct btrfs_key *location)
1480{
1481 struct btrfs_root *root;
1482
1483 root = btrfs_read_tree_root(tree_root, location);
1484 if (IS_ERR(root))
1485 return root;
1486
1487 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1488 root->ref_cows = 1;
1489 btrfs_check_and_init_root_item(&root->root_item);
1490 }
1491
1492 return root;
1493}
1494
1495int btrfs_init_fs_root(struct btrfs_root *root)
1496{
1497 int ret;
1498
1499 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1500 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1501 GFP_NOFS);
1502 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1503 ret = -ENOMEM;
1504 goto fail;
1505 }
1506
1507 btrfs_init_free_ino_ctl(root);
1508 mutex_init(&root->fs_commit_mutex);
1509 spin_lock_init(&root->cache_lock);
1510 init_waitqueue_head(&root->cache_wait);
1511
1512 ret = get_anon_bdev(&root->anon_dev);
1513 if (ret)
1514 goto fail;
1515 return 0;
1516fail:
1517 kfree(root->free_ino_ctl);
1518 kfree(root->free_ino_pinned);
1519 return ret;
1520}
1521
1522struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1523 u64 root_id)
1524{
1525 struct btrfs_root *root;
1526
1527 spin_lock(&fs_info->fs_roots_radix_lock);
1528 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1529 (unsigned long)root_id);
1530 spin_unlock(&fs_info->fs_roots_radix_lock);
1531 return root;
1532}
1533
1534int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1535 struct btrfs_root *root)
1536{
1537 int ret;
1538
1539 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1540 if (ret)
1541 return ret;
1542
1543 spin_lock(&fs_info->fs_roots_radix_lock);
1544 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1545 (unsigned long)root->root_key.objectid,
1546 root);
1547 if (ret == 0)
1548 root->in_radix = 1;
1549 spin_unlock(&fs_info->fs_roots_radix_lock);
1550 radix_tree_preload_end();
1551
1552 return ret;
1553}
1554
1555struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1556 struct btrfs_key *location)
1557{
1558 struct btrfs_root *root;
1559 int ret;
1560
1561 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1562 return fs_info->tree_root;
1563 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1564 return fs_info->extent_root;
1565 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1566 return fs_info->chunk_root;
1567 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1568 return fs_info->dev_root;
1569 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1570 return fs_info->csum_root;
1571 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1572 return fs_info->quota_root ? fs_info->quota_root :
1573 ERR_PTR(-ENOENT);
1574again:
1575 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1576 if (root)
1577 return root;
1578
1579 root = btrfs_read_fs_root(fs_info->tree_root, location);
1580 if (IS_ERR(root))
1581 return root;
1582
1583 if (btrfs_root_refs(&root->root_item) == 0) {
1584 ret = -ENOENT;
1585 goto fail;
1586 }
1587
1588 ret = btrfs_init_fs_root(root);
1589 if (ret)
1590 goto fail;
1591
1592 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1593 if (ret < 0)
1594 goto fail;
1595 if (ret == 0)
1596 root->orphan_item_inserted = 1;
1597
1598 ret = btrfs_insert_fs_root(fs_info, root);
1599 if (ret) {
1600 if (ret == -EEXIST) {
1601 free_fs_root(root);
1602 goto again;
1603 }
1604 goto fail;
1605 }
1606 return root;
1607fail:
1608 free_fs_root(root);
1609 return ERR_PTR(ret);
1610}
1611
1612static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1613{
1614 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1615 int ret = 0;
1616 struct btrfs_device *device;
1617 struct backing_dev_info *bdi;
1618
1619 rcu_read_lock();
1620 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1621 if (!device->bdev)
1622 continue;
1623 bdi = blk_get_backing_dev_info(device->bdev);
1624 if (bdi && bdi_congested(bdi, bdi_bits)) {
1625 ret = 1;
1626 break;
1627 }
1628 }
1629 rcu_read_unlock();
1630 return ret;
1631}
1632
1633/*
1634 * If this fails, caller must call bdi_destroy() to get rid of the
1635 * bdi again.
1636 */
1637static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1638{
1639 int err;
1640
1641 bdi->capabilities = BDI_CAP_MAP_COPY;
1642 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1643 if (err)
1644 return err;
1645
1646 bdi->ra_pages = default_backing_dev_info.ra_pages;
1647 bdi->congested_fn = btrfs_congested_fn;
1648 bdi->congested_data = info;
1649 return 0;
1650}
1651
1652/*
1653 * called by the kthread helper functions to finally call the bio end_io
1654 * functions. This is where read checksum verification actually happens
1655 */
1656static void end_workqueue_fn(struct btrfs_work *work)
1657{
1658 struct bio *bio;
1659 struct end_io_wq *end_io_wq;
1660 struct btrfs_fs_info *fs_info;
1661 int error;
1662
1663 end_io_wq = container_of(work, struct end_io_wq, work);
1664 bio = end_io_wq->bio;
1665 fs_info = end_io_wq->info;
1666
1667 error = end_io_wq->error;
1668 bio->bi_private = end_io_wq->private;
1669 bio->bi_end_io = end_io_wq->end_io;
1670 kfree(end_io_wq);
1671 bio_endio(bio, error);
1672}
1673
1674static int cleaner_kthread(void *arg)
1675{
1676 struct btrfs_root *root = arg;
1677 int again;
1678
1679 do {
1680 again = 0;
1681
1682 /* Make the cleaner go to sleep early. */
1683 if (btrfs_need_cleaner_sleep(root))
1684 goto sleep;
1685
1686 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1687 goto sleep;
1688
1689 /*
1690 * Avoid the problem that we change the status of the fs
1691 * during the above check and trylock.
1692 */
1693 if (btrfs_need_cleaner_sleep(root)) {
1694 mutex_unlock(&root->fs_info->cleaner_mutex);
1695 goto sleep;
1696 }
1697
1698 btrfs_run_delayed_iputs(root);
1699 again = btrfs_clean_one_deleted_snapshot(root);
1700 mutex_unlock(&root->fs_info->cleaner_mutex);
1701
1702 /*
1703 * The defragger has dealt with the R/O remount and umount,
1704 * needn't do anything special here.
1705 */
1706 btrfs_run_defrag_inodes(root->fs_info);
1707sleep:
1708 if (!try_to_freeze() && !again) {
1709 set_current_state(TASK_INTERRUPTIBLE);
1710 if (!kthread_should_stop())
1711 schedule();
1712 __set_current_state(TASK_RUNNING);
1713 }
1714 } while (!kthread_should_stop());
1715 return 0;
1716}
1717
1718static int transaction_kthread(void *arg)
1719{
1720 struct btrfs_root *root = arg;
1721 struct btrfs_trans_handle *trans;
1722 struct btrfs_transaction *cur;
1723 u64 transid;
1724 unsigned long now;
1725 unsigned long delay;
1726 bool cannot_commit;
1727
1728 do {
1729 cannot_commit = false;
1730 delay = HZ * 30;
1731 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1732
1733 spin_lock(&root->fs_info->trans_lock);
1734 cur = root->fs_info->running_transaction;
1735 if (!cur) {
1736 spin_unlock(&root->fs_info->trans_lock);
1737 goto sleep;
1738 }
1739
1740 now = get_seconds();
1741 if (!cur->blocked &&
1742 (now < cur->start_time || now - cur->start_time < 30)) {
1743 spin_unlock(&root->fs_info->trans_lock);
1744 delay = HZ * 5;
1745 goto sleep;
1746 }
1747 transid = cur->transid;
1748 spin_unlock(&root->fs_info->trans_lock);
1749
1750 /* If the file system is aborted, this will always fail. */
1751 trans = btrfs_attach_transaction(root);
1752 if (IS_ERR(trans)) {
1753 if (PTR_ERR(trans) != -ENOENT)
1754 cannot_commit = true;
1755 goto sleep;
1756 }
1757 if (transid == trans->transid) {
1758 btrfs_commit_transaction(trans, root);
1759 } else {
1760 btrfs_end_transaction(trans, root);
1761 }
1762sleep:
1763 wake_up_process(root->fs_info->cleaner_kthread);
1764 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1765
1766 if (!try_to_freeze()) {
1767 set_current_state(TASK_INTERRUPTIBLE);
1768 if (!kthread_should_stop() &&
1769 (!btrfs_transaction_blocked(root->fs_info) ||
1770 cannot_commit))
1771 schedule_timeout(delay);
1772 __set_current_state(TASK_RUNNING);
1773 }
1774 } while (!kthread_should_stop());
1775 return 0;
1776}
1777
1778/*
1779 * this will find the highest generation in the array of
1780 * root backups. The index of the highest array is returned,
1781 * or -1 if we can't find anything.
1782 *
1783 * We check to make sure the array is valid by comparing the
1784 * generation of the latest root in the array with the generation
1785 * in the super block. If they don't match we pitch it.
1786 */
1787static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1788{
1789 u64 cur;
1790 int newest_index = -1;
1791 struct btrfs_root_backup *root_backup;
1792 int i;
1793
1794 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1795 root_backup = info->super_copy->super_roots + i;
1796 cur = btrfs_backup_tree_root_gen(root_backup);
1797 if (cur == newest_gen)
1798 newest_index = i;
1799 }
1800
1801 /* check to see if we actually wrapped around */
1802 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1803 root_backup = info->super_copy->super_roots;
1804 cur = btrfs_backup_tree_root_gen(root_backup);
1805 if (cur == newest_gen)
1806 newest_index = 0;
1807 }
1808 return newest_index;
1809}
1810
1811
1812/*
1813 * find the oldest backup so we know where to store new entries
1814 * in the backup array. This will set the backup_root_index
1815 * field in the fs_info struct
1816 */
1817static void find_oldest_super_backup(struct btrfs_fs_info *info,
1818 u64 newest_gen)
1819{
1820 int newest_index = -1;
1821
1822 newest_index = find_newest_super_backup(info, newest_gen);
1823 /* if there was garbage in there, just move along */
1824 if (newest_index == -1) {
1825 info->backup_root_index = 0;
1826 } else {
1827 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1828 }
1829}
1830
1831/*
1832 * copy all the root pointers into the super backup array.
1833 * this will bump the backup pointer by one when it is
1834 * done
1835 */
1836static void backup_super_roots(struct btrfs_fs_info *info)
1837{
1838 int next_backup;
1839 struct btrfs_root_backup *root_backup;
1840 int last_backup;
1841
1842 next_backup = info->backup_root_index;
1843 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1844 BTRFS_NUM_BACKUP_ROOTS;
1845
1846 /*
1847 * just overwrite the last backup if we're at the same generation
1848 * this happens only at umount
1849 */
1850 root_backup = info->super_for_commit->super_roots + last_backup;
1851 if (btrfs_backup_tree_root_gen(root_backup) ==
1852 btrfs_header_generation(info->tree_root->node))
1853 next_backup = last_backup;
1854
1855 root_backup = info->super_for_commit->super_roots + next_backup;
1856
1857 /*
1858 * make sure all of our padding and empty slots get zero filled
1859 * regardless of which ones we use today
1860 */
1861 memset(root_backup, 0, sizeof(*root_backup));
1862
1863 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1864
1865 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1866 btrfs_set_backup_tree_root_gen(root_backup,
1867 btrfs_header_generation(info->tree_root->node));
1868
1869 btrfs_set_backup_tree_root_level(root_backup,
1870 btrfs_header_level(info->tree_root->node));
1871
1872 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1873 btrfs_set_backup_chunk_root_gen(root_backup,
1874 btrfs_header_generation(info->chunk_root->node));
1875 btrfs_set_backup_chunk_root_level(root_backup,
1876 btrfs_header_level(info->chunk_root->node));
1877
1878 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1879 btrfs_set_backup_extent_root_gen(root_backup,
1880 btrfs_header_generation(info->extent_root->node));
1881 btrfs_set_backup_extent_root_level(root_backup,
1882 btrfs_header_level(info->extent_root->node));
1883
1884 /*
1885 * we might commit during log recovery, which happens before we set
1886 * the fs_root. Make sure it is valid before we fill it in.
1887 */
1888 if (info->fs_root && info->fs_root->node) {
1889 btrfs_set_backup_fs_root(root_backup,
1890 info->fs_root->node->start);
1891 btrfs_set_backup_fs_root_gen(root_backup,
1892 btrfs_header_generation(info->fs_root->node));
1893 btrfs_set_backup_fs_root_level(root_backup,
1894 btrfs_header_level(info->fs_root->node));
1895 }
1896
1897 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1898 btrfs_set_backup_dev_root_gen(root_backup,
1899 btrfs_header_generation(info->dev_root->node));
1900 btrfs_set_backup_dev_root_level(root_backup,
1901 btrfs_header_level(info->dev_root->node));
1902
1903 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1904 btrfs_set_backup_csum_root_gen(root_backup,
1905 btrfs_header_generation(info->csum_root->node));
1906 btrfs_set_backup_csum_root_level(root_backup,
1907 btrfs_header_level(info->csum_root->node));
1908
1909 btrfs_set_backup_total_bytes(root_backup,
1910 btrfs_super_total_bytes(info->super_copy));
1911 btrfs_set_backup_bytes_used(root_backup,
1912 btrfs_super_bytes_used(info->super_copy));
1913 btrfs_set_backup_num_devices(root_backup,
1914 btrfs_super_num_devices(info->super_copy));
1915
1916 /*
1917 * if we don't copy this out to the super_copy, it won't get remembered
1918 * for the next commit
1919 */
1920 memcpy(&info->super_copy->super_roots,
1921 &info->super_for_commit->super_roots,
1922 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1923}
1924
1925/*
1926 * this copies info out of the root backup array and back into
1927 * the in-memory super block. It is meant to help iterate through
1928 * the array, so you send it the number of backups you've already
1929 * tried and the last backup index you used.
1930 *
1931 * this returns -1 when it has tried all the backups
1932 */
1933static noinline int next_root_backup(struct btrfs_fs_info *info,
1934 struct btrfs_super_block *super,
1935 int *num_backups_tried, int *backup_index)
1936{
1937 struct btrfs_root_backup *root_backup;
1938 int newest = *backup_index;
1939
1940 if (*num_backups_tried == 0) {
1941 u64 gen = btrfs_super_generation(super);
1942
1943 newest = find_newest_super_backup(info, gen);
1944 if (newest == -1)
1945 return -1;
1946
1947 *backup_index = newest;
1948 *num_backups_tried = 1;
1949 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1950 /* we've tried all the backups, all done */
1951 return -1;
1952 } else {
1953 /* jump to the next oldest backup */
1954 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1955 BTRFS_NUM_BACKUP_ROOTS;
1956 *backup_index = newest;
1957 *num_backups_tried += 1;
1958 }
1959 root_backup = super->super_roots + newest;
1960
1961 btrfs_set_super_generation(super,
1962 btrfs_backup_tree_root_gen(root_backup));
1963 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1964 btrfs_set_super_root_level(super,
1965 btrfs_backup_tree_root_level(root_backup));
1966 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1967
1968 /*
1969 * fixme: the total bytes and num_devices need to match or we should
1970 * need a fsck
1971 */
1972 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1973 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1974 return 0;
1975}
1976
1977/* helper to cleanup workers */
1978static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1979{
1980 btrfs_stop_workers(&fs_info->generic_worker);
1981 btrfs_stop_workers(&fs_info->fixup_workers);
1982 btrfs_stop_workers(&fs_info->delalloc_workers);
1983 btrfs_stop_workers(&fs_info->workers);
1984 btrfs_stop_workers(&fs_info->endio_workers);
1985 btrfs_stop_workers(&fs_info->endio_meta_workers);
1986 btrfs_stop_workers(&fs_info->endio_raid56_workers);
1987 btrfs_stop_workers(&fs_info->rmw_workers);
1988 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1989 btrfs_stop_workers(&fs_info->endio_write_workers);
1990 btrfs_stop_workers(&fs_info->endio_freespace_worker);
1991 btrfs_stop_workers(&fs_info->submit_workers);
1992 btrfs_stop_workers(&fs_info->delayed_workers);
1993 btrfs_stop_workers(&fs_info->caching_workers);
1994 btrfs_stop_workers(&fs_info->readahead_workers);
1995 btrfs_stop_workers(&fs_info->flush_workers);
1996 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
1997}
1998
1999/* helper to cleanup tree roots */
2000static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2001{
2002 free_extent_buffer(info->tree_root->node);
2003 free_extent_buffer(info->tree_root->commit_root);
2004 info->tree_root->node = NULL;
2005 info->tree_root->commit_root = NULL;
2006
2007 if (info->dev_root) {
2008 free_extent_buffer(info->dev_root->node);
2009 free_extent_buffer(info->dev_root->commit_root);
2010 info->dev_root->node = NULL;
2011 info->dev_root->commit_root = NULL;
2012 }
2013 if (info->extent_root) {
2014 free_extent_buffer(info->extent_root->node);
2015 free_extent_buffer(info->extent_root->commit_root);
2016 info->extent_root->node = NULL;
2017 info->extent_root->commit_root = NULL;
2018 }
2019 if (info->csum_root) {
2020 free_extent_buffer(info->csum_root->node);
2021 free_extent_buffer(info->csum_root->commit_root);
2022 info->csum_root->node = NULL;
2023 info->csum_root->commit_root = NULL;
2024 }
2025 if (info->quota_root) {
2026 free_extent_buffer(info->quota_root->node);
2027 free_extent_buffer(info->quota_root->commit_root);
2028 info->quota_root->node = NULL;
2029 info->quota_root->commit_root = NULL;
2030 }
2031 if (chunk_root) {
2032 free_extent_buffer(info->chunk_root->node);
2033 free_extent_buffer(info->chunk_root->commit_root);
2034 info->chunk_root->node = NULL;
2035 info->chunk_root->commit_root = NULL;
2036 }
2037}
2038
2039static void del_fs_roots(struct btrfs_fs_info *fs_info)
2040{
2041 int ret;
2042 struct btrfs_root *gang[8];
2043 int i;
2044
2045 while (!list_empty(&fs_info->dead_roots)) {
2046 gang[0] = list_entry(fs_info->dead_roots.next,
2047 struct btrfs_root, root_list);
2048 list_del(&gang[0]->root_list);
2049
2050 if (gang[0]->in_radix) {
2051 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2052 } else {
2053 free_extent_buffer(gang[0]->node);
2054 free_extent_buffer(gang[0]->commit_root);
2055 kfree(gang[0]);
2056 }
2057 }
2058
2059 while (1) {
2060 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2061 (void **)gang, 0,
2062 ARRAY_SIZE(gang));
2063 if (!ret)
2064 break;
2065 for (i = 0; i < ret; i++)
2066 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2067 }
2068}
2069
2070int open_ctree(struct super_block *sb,
2071 struct btrfs_fs_devices *fs_devices,
2072 char *options)
2073{
2074 u32 sectorsize;
2075 u32 nodesize;
2076 u32 leafsize;
2077 u32 blocksize;
2078 u32 stripesize;
2079 u64 generation;
2080 u64 features;
2081 struct btrfs_key location;
2082 struct buffer_head *bh;
2083 struct btrfs_super_block *disk_super;
2084 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2085 struct btrfs_root *tree_root;
2086 struct btrfs_root *extent_root;
2087 struct btrfs_root *csum_root;
2088 struct btrfs_root *chunk_root;
2089 struct btrfs_root *dev_root;
2090 struct btrfs_root *quota_root;
2091 struct btrfs_root *log_tree_root;
2092 int ret;
2093 int err = -EINVAL;
2094 int num_backups_tried = 0;
2095 int backup_index = 0;
2096
2097 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2098 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2099 if (!tree_root || !chunk_root) {
2100 err = -ENOMEM;
2101 goto fail;
2102 }
2103
2104 ret = init_srcu_struct(&fs_info->subvol_srcu);
2105 if (ret) {
2106 err = ret;
2107 goto fail;
2108 }
2109
2110 ret = setup_bdi(fs_info, &fs_info->bdi);
2111 if (ret) {
2112 err = ret;
2113 goto fail_srcu;
2114 }
2115
2116 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2117 if (ret) {
2118 err = ret;
2119 goto fail_bdi;
2120 }
2121 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2122 (1 + ilog2(nr_cpu_ids));
2123
2124 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2125 if (ret) {
2126 err = ret;
2127 goto fail_dirty_metadata_bytes;
2128 }
2129
2130 fs_info->btree_inode = new_inode(sb);
2131 if (!fs_info->btree_inode) {
2132 err = -ENOMEM;
2133 goto fail_delalloc_bytes;
2134 }
2135
2136 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2137
2138 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2139 INIT_LIST_HEAD(&fs_info->trans_list);
2140 INIT_LIST_HEAD(&fs_info->dead_roots);
2141 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2142 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
2143 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2144 spin_lock_init(&fs_info->delalloc_lock);
2145 spin_lock_init(&fs_info->trans_lock);
2146 spin_lock_init(&fs_info->fs_roots_radix_lock);
2147 spin_lock_init(&fs_info->delayed_iput_lock);
2148 spin_lock_init(&fs_info->defrag_inodes_lock);
2149 spin_lock_init(&fs_info->free_chunk_lock);
2150 spin_lock_init(&fs_info->tree_mod_seq_lock);
2151 spin_lock_init(&fs_info->super_lock);
2152 rwlock_init(&fs_info->tree_mod_log_lock);
2153 mutex_init(&fs_info->reloc_mutex);
2154 seqlock_init(&fs_info->profiles_lock);
2155
2156 init_completion(&fs_info->kobj_unregister);
2157 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2158 INIT_LIST_HEAD(&fs_info->space_info);
2159 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2160 btrfs_mapping_init(&fs_info->mapping_tree);
2161 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2162 BTRFS_BLOCK_RSV_GLOBAL);
2163 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2164 BTRFS_BLOCK_RSV_DELALLOC);
2165 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2166 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2167 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2168 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2169 BTRFS_BLOCK_RSV_DELOPS);
2170 atomic_set(&fs_info->nr_async_submits, 0);
2171 atomic_set(&fs_info->async_delalloc_pages, 0);
2172 atomic_set(&fs_info->async_submit_draining, 0);
2173 atomic_set(&fs_info->nr_async_bios, 0);
2174 atomic_set(&fs_info->defrag_running, 0);
2175 atomic64_set(&fs_info->tree_mod_seq, 0);
2176 fs_info->sb = sb;
2177 fs_info->max_inline = 8192 * 1024;
2178 fs_info->metadata_ratio = 0;
2179 fs_info->defrag_inodes = RB_ROOT;
2180 fs_info->trans_no_join = 0;
2181 fs_info->free_chunk_space = 0;
2182 fs_info->tree_mod_log = RB_ROOT;
2183
2184 /* readahead state */
2185 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2186 spin_lock_init(&fs_info->reada_lock);
2187
2188 fs_info->thread_pool_size = min_t(unsigned long,
2189 num_online_cpus() + 2, 8);
2190
2191 INIT_LIST_HEAD(&fs_info->ordered_extents);
2192 spin_lock_init(&fs_info->ordered_extent_lock);
2193 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2194 GFP_NOFS);
2195 if (!fs_info->delayed_root) {
2196 err = -ENOMEM;
2197 goto fail_iput;
2198 }
2199 btrfs_init_delayed_root(fs_info->delayed_root);
2200
2201 mutex_init(&fs_info->scrub_lock);
2202 atomic_set(&fs_info->scrubs_running, 0);
2203 atomic_set(&fs_info->scrub_pause_req, 0);
2204 atomic_set(&fs_info->scrubs_paused, 0);
2205 atomic_set(&fs_info->scrub_cancel_req, 0);
2206 init_waitqueue_head(&fs_info->scrub_pause_wait);
2207 init_rwsem(&fs_info->scrub_super_lock);
2208 fs_info->scrub_workers_refcnt = 0;
2209#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2210 fs_info->check_integrity_print_mask = 0;
2211#endif
2212
2213 spin_lock_init(&fs_info->balance_lock);
2214 mutex_init(&fs_info->balance_mutex);
2215 atomic_set(&fs_info->balance_running, 0);
2216 atomic_set(&fs_info->balance_pause_req, 0);
2217 atomic_set(&fs_info->balance_cancel_req, 0);
2218 fs_info->balance_ctl = NULL;
2219 init_waitqueue_head(&fs_info->balance_wait_q);
2220
2221 sb->s_blocksize = 4096;
2222 sb->s_blocksize_bits = blksize_bits(4096);
2223 sb->s_bdi = &fs_info->bdi;
2224
2225 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2226 set_nlink(fs_info->btree_inode, 1);
2227 /*
2228 * we set the i_size on the btree inode to the max possible int.
2229 * the real end of the address space is determined by all of
2230 * the devices in the system
2231 */
2232 fs_info->btree_inode->i_size = OFFSET_MAX;
2233 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2234 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2235
2236 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2237 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2238 fs_info->btree_inode->i_mapping);
2239 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2240 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2241
2242 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2243
2244 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2245 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2246 sizeof(struct btrfs_key));
2247 set_bit(BTRFS_INODE_DUMMY,
2248 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2249 insert_inode_hash(fs_info->btree_inode);
2250
2251 spin_lock_init(&fs_info->block_group_cache_lock);
2252 fs_info->block_group_cache_tree = RB_ROOT;
2253 fs_info->first_logical_byte = (u64)-1;
2254
2255 extent_io_tree_init(&fs_info->freed_extents[0],
2256 fs_info->btree_inode->i_mapping);
2257 extent_io_tree_init(&fs_info->freed_extents[1],
2258 fs_info->btree_inode->i_mapping);
2259 fs_info->pinned_extents = &fs_info->freed_extents[0];
2260 fs_info->do_barriers = 1;
2261
2262
2263 mutex_init(&fs_info->ordered_operations_mutex);
2264 mutex_init(&fs_info->tree_log_mutex);
2265 mutex_init(&fs_info->chunk_mutex);
2266 mutex_init(&fs_info->transaction_kthread_mutex);
2267 mutex_init(&fs_info->cleaner_mutex);
2268 mutex_init(&fs_info->volume_mutex);
2269 init_rwsem(&fs_info->extent_commit_sem);
2270 init_rwsem(&fs_info->cleanup_work_sem);
2271 init_rwsem(&fs_info->subvol_sem);
2272 fs_info->dev_replace.lock_owner = 0;
2273 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2274 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2275 mutex_init(&fs_info->dev_replace.lock_management_lock);
2276 mutex_init(&fs_info->dev_replace.lock);
2277
2278 spin_lock_init(&fs_info->qgroup_lock);
2279 mutex_init(&fs_info->qgroup_ioctl_lock);
2280 fs_info->qgroup_tree = RB_ROOT;
2281 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2282 fs_info->qgroup_seq = 1;
2283 fs_info->quota_enabled = 0;
2284 fs_info->pending_quota_state = 0;
2285 fs_info->qgroup_ulist = NULL;
2286 mutex_init(&fs_info->qgroup_rescan_lock);
2287
2288 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2289 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2290
2291 init_waitqueue_head(&fs_info->transaction_throttle);
2292 init_waitqueue_head(&fs_info->transaction_wait);
2293 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2294 init_waitqueue_head(&fs_info->async_submit_wait);
2295
2296 ret = btrfs_alloc_stripe_hash_table(fs_info);
2297 if (ret) {
2298 err = ret;
2299 goto fail_alloc;
2300 }
2301
2302 __setup_root(4096, 4096, 4096, 4096, tree_root,
2303 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2304
2305 invalidate_bdev(fs_devices->latest_bdev);
2306
2307 /*
2308 * Read super block and check the signature bytes only
2309 */
2310 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2311 if (!bh) {
2312 err = -EINVAL;
2313 goto fail_alloc;
2314 }
2315
2316 /*
2317 * We want to check superblock checksum, the type is stored inside.
2318 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2319 */
2320 if (btrfs_check_super_csum(bh->b_data)) {
2321 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2322 err = -EINVAL;
2323 goto fail_alloc;
2324 }
2325
2326 /*
2327 * super_copy is zeroed at allocation time and we never touch the
2328 * following bytes up to INFO_SIZE, the checksum is calculated from
2329 * the whole block of INFO_SIZE
2330 */
2331 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2332 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2333 sizeof(*fs_info->super_for_commit));
2334 brelse(bh);
2335
2336 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2337
2338 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2339 if (ret) {
2340 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2341 err = -EINVAL;
2342 goto fail_alloc;
2343 }
2344
2345 disk_super = fs_info->super_copy;
2346 if (!btrfs_super_root(disk_super))
2347 goto fail_alloc;
2348
2349 /* check FS state, whether FS is broken. */
2350 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2351 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2352
2353 /*
2354 * run through our array of backup supers and setup
2355 * our ring pointer to the oldest one
2356 */
2357 generation = btrfs_super_generation(disk_super);
2358 find_oldest_super_backup(fs_info, generation);
2359
2360 /*
2361 * In the long term, we'll store the compression type in the super
2362 * block, and it'll be used for per file compression control.
2363 */
2364 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2365
2366 ret = btrfs_parse_options(tree_root, options);
2367 if (ret) {
2368 err = ret;
2369 goto fail_alloc;
2370 }
2371
2372 features = btrfs_super_incompat_flags(disk_super) &
2373 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2374 if (features) {
2375 printk(KERN_ERR "BTRFS: couldn't mount because of "
2376 "unsupported optional features (%Lx).\n",
2377 (unsigned long long)features);
2378 err = -EINVAL;
2379 goto fail_alloc;
2380 }
2381
2382 if (btrfs_super_leafsize(disk_super) !=
2383 btrfs_super_nodesize(disk_super)) {
2384 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2385 "blocksizes don't match. node %d leaf %d\n",
2386 btrfs_super_nodesize(disk_super),
2387 btrfs_super_leafsize(disk_super));
2388 err = -EINVAL;
2389 goto fail_alloc;
2390 }
2391 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2392 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2393 "blocksize (%d) was too large\n",
2394 btrfs_super_leafsize(disk_super));
2395 err = -EINVAL;
2396 goto fail_alloc;
2397 }
2398
2399 features = btrfs_super_incompat_flags(disk_super);
2400 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2401 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2402 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2403
2404 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2405 printk(KERN_ERR "btrfs: has skinny extents\n");
2406
2407 /*
2408 * flag our filesystem as having big metadata blocks if
2409 * they are bigger than the page size
2410 */
2411 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2412 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2413 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2414 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2415 }
2416
2417 nodesize = btrfs_super_nodesize(disk_super);
2418 leafsize = btrfs_super_leafsize(disk_super);
2419 sectorsize = btrfs_super_sectorsize(disk_super);
2420 stripesize = btrfs_super_stripesize(disk_super);
2421 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2422 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2423
2424 /*
2425 * mixed block groups end up with duplicate but slightly offset
2426 * extent buffers for the same range. It leads to corruptions
2427 */
2428 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2429 (sectorsize != leafsize)) {
2430 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2431 "are not allowed for mixed block groups on %s\n",
2432 sb->s_id);
2433 goto fail_alloc;
2434 }
2435
2436 /*
2437 * Needn't use the lock because there is no other task which will
2438 * update the flag.
2439 */
2440 btrfs_set_super_incompat_flags(disk_super, features);
2441
2442 features = btrfs_super_compat_ro_flags(disk_super) &
2443 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2444 if (!(sb->s_flags & MS_RDONLY) && features) {
2445 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2446 "unsupported option features (%Lx).\n",
2447 (unsigned long long)features);
2448 err = -EINVAL;
2449 goto fail_alloc;
2450 }
2451
2452 btrfs_init_workers(&fs_info->generic_worker,
2453 "genwork", 1, NULL);
2454
2455 btrfs_init_workers(&fs_info->workers, "worker",
2456 fs_info->thread_pool_size,
2457 &fs_info->generic_worker);
2458
2459 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2460 fs_info->thread_pool_size,
2461 &fs_info->generic_worker);
2462
2463 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2464 fs_info->thread_pool_size,
2465 &fs_info->generic_worker);
2466
2467 btrfs_init_workers(&fs_info->submit_workers, "submit",
2468 min_t(u64, fs_devices->num_devices,
2469 fs_info->thread_pool_size),
2470 &fs_info->generic_worker);
2471
2472 btrfs_init_workers(&fs_info->caching_workers, "cache",
2473 2, &fs_info->generic_worker);
2474
2475 /* a higher idle thresh on the submit workers makes it much more
2476 * likely that bios will be send down in a sane order to the
2477 * devices
2478 */
2479 fs_info->submit_workers.idle_thresh = 64;
2480
2481 fs_info->workers.idle_thresh = 16;
2482 fs_info->workers.ordered = 1;
2483
2484 fs_info->delalloc_workers.idle_thresh = 2;
2485 fs_info->delalloc_workers.ordered = 1;
2486
2487 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2488 &fs_info->generic_worker);
2489 btrfs_init_workers(&fs_info->endio_workers, "endio",
2490 fs_info->thread_pool_size,
2491 &fs_info->generic_worker);
2492 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2493 fs_info->thread_pool_size,
2494 &fs_info->generic_worker);
2495 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2496 "endio-meta-write", fs_info->thread_pool_size,
2497 &fs_info->generic_worker);
2498 btrfs_init_workers(&fs_info->endio_raid56_workers,
2499 "endio-raid56", fs_info->thread_pool_size,
2500 &fs_info->generic_worker);
2501 btrfs_init_workers(&fs_info->rmw_workers,
2502 "rmw", fs_info->thread_pool_size,
2503 &fs_info->generic_worker);
2504 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2505 fs_info->thread_pool_size,
2506 &fs_info->generic_worker);
2507 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2508 1, &fs_info->generic_worker);
2509 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2510 fs_info->thread_pool_size,
2511 &fs_info->generic_worker);
2512 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2513 fs_info->thread_pool_size,
2514 &fs_info->generic_worker);
2515 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2516 &fs_info->generic_worker);
2517
2518 /*
2519 * endios are largely parallel and should have a very
2520 * low idle thresh
2521 */
2522 fs_info->endio_workers.idle_thresh = 4;
2523 fs_info->endio_meta_workers.idle_thresh = 4;
2524 fs_info->endio_raid56_workers.idle_thresh = 4;
2525 fs_info->rmw_workers.idle_thresh = 2;
2526
2527 fs_info->endio_write_workers.idle_thresh = 2;
2528 fs_info->endio_meta_write_workers.idle_thresh = 2;
2529 fs_info->readahead_workers.idle_thresh = 2;
2530
2531 /*
2532 * btrfs_start_workers can really only fail because of ENOMEM so just
2533 * return -ENOMEM if any of these fail.
2534 */
2535 ret = btrfs_start_workers(&fs_info->workers);
2536 ret |= btrfs_start_workers(&fs_info->generic_worker);
2537 ret |= btrfs_start_workers(&fs_info->submit_workers);
2538 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2539 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2540 ret |= btrfs_start_workers(&fs_info->endio_workers);
2541 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2542 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2543 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2544 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2545 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2546 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2547 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2548 ret |= btrfs_start_workers(&fs_info->caching_workers);
2549 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2550 ret |= btrfs_start_workers(&fs_info->flush_workers);
2551 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2552 if (ret) {
2553 err = -ENOMEM;
2554 goto fail_sb_buffer;
2555 }
2556
2557 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2558 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2559 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2560
2561 tree_root->nodesize = nodesize;
2562 tree_root->leafsize = leafsize;
2563 tree_root->sectorsize = sectorsize;
2564 tree_root->stripesize = stripesize;
2565
2566 sb->s_blocksize = sectorsize;
2567 sb->s_blocksize_bits = blksize_bits(sectorsize);
2568
2569 if (disk_super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2570 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2571 goto fail_sb_buffer;
2572 }
2573
2574 if (sectorsize != PAGE_SIZE) {
2575 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2576 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2577 goto fail_sb_buffer;
2578 }
2579
2580 mutex_lock(&fs_info->chunk_mutex);
2581 ret = btrfs_read_sys_array(tree_root);
2582 mutex_unlock(&fs_info->chunk_mutex);
2583 if (ret) {
2584 printk(KERN_WARNING "btrfs: failed to read the system "
2585 "array on %s\n", sb->s_id);
2586 goto fail_sb_buffer;
2587 }
2588
2589 blocksize = btrfs_level_size(tree_root,
2590 btrfs_super_chunk_root_level(disk_super));
2591 generation = btrfs_super_chunk_root_generation(disk_super);
2592
2593 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2594 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2595
2596 chunk_root->node = read_tree_block(chunk_root,
2597 btrfs_super_chunk_root(disk_super),
2598 blocksize, generation);
2599 if (!chunk_root->node ||
2600 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2601 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2602 sb->s_id);
2603 goto fail_tree_roots;
2604 }
2605 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2606 chunk_root->commit_root = btrfs_root_node(chunk_root);
2607
2608 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2609 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2610 BTRFS_UUID_SIZE);
2611
2612 ret = btrfs_read_chunk_tree(chunk_root);
2613 if (ret) {
2614 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2615 sb->s_id);
2616 goto fail_tree_roots;
2617 }
2618
2619 /*
2620 * keep the device that is marked to be the target device for the
2621 * dev_replace procedure
2622 */
2623 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2624
2625 if (!fs_devices->latest_bdev) {
2626 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2627 sb->s_id);
2628 goto fail_tree_roots;
2629 }
2630
2631retry_root_backup:
2632 blocksize = btrfs_level_size(tree_root,
2633 btrfs_super_root_level(disk_super));
2634 generation = btrfs_super_generation(disk_super);
2635
2636 tree_root->node = read_tree_block(tree_root,
2637 btrfs_super_root(disk_super),
2638 blocksize, generation);
2639 if (!tree_root->node ||
2640 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2641 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2642 sb->s_id);
2643
2644 goto recovery_tree_root;
2645 }
2646
2647 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2648 tree_root->commit_root = btrfs_root_node(tree_root);
2649
2650 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2651 location.type = BTRFS_ROOT_ITEM_KEY;
2652 location.offset = 0;
2653
2654 extent_root = btrfs_read_tree_root(tree_root, &location);
2655 if (IS_ERR(extent_root)) {
2656 ret = PTR_ERR(extent_root);
2657 goto recovery_tree_root;
2658 }
2659 extent_root->track_dirty = 1;
2660 fs_info->extent_root = extent_root;
2661
2662 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2663 dev_root = btrfs_read_tree_root(tree_root, &location);
2664 if (IS_ERR(dev_root)) {
2665 ret = PTR_ERR(dev_root);
2666 goto recovery_tree_root;
2667 }
2668 dev_root->track_dirty = 1;
2669 fs_info->dev_root = dev_root;
2670 btrfs_init_devices_late(fs_info);
2671
2672 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2673 csum_root = btrfs_read_tree_root(tree_root, &location);
2674 if (IS_ERR(csum_root)) {
2675 ret = PTR_ERR(csum_root);
2676 goto recovery_tree_root;
2677 }
2678 csum_root->track_dirty = 1;
2679 fs_info->csum_root = csum_root;
2680
2681 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2682 quota_root = btrfs_read_tree_root(tree_root, &location);
2683 if (!IS_ERR(quota_root)) {
2684 quota_root->track_dirty = 1;
2685 fs_info->quota_enabled = 1;
2686 fs_info->pending_quota_state = 1;
2687 fs_info->quota_root = quota_root;
2688 }
2689
2690 fs_info->generation = generation;
2691 fs_info->last_trans_committed = generation;
2692
2693 ret = btrfs_recover_balance(fs_info);
2694 if (ret) {
2695 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2696 goto fail_block_groups;
2697 }
2698
2699 ret = btrfs_init_dev_stats(fs_info);
2700 if (ret) {
2701 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2702 ret);
2703 goto fail_block_groups;
2704 }
2705
2706 ret = btrfs_init_dev_replace(fs_info);
2707 if (ret) {
2708 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2709 goto fail_block_groups;
2710 }
2711
2712 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2713
2714 ret = btrfs_init_space_info(fs_info);
2715 if (ret) {
2716 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2717 goto fail_block_groups;
2718 }
2719
2720 ret = btrfs_read_block_groups(extent_root);
2721 if (ret) {
2722 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2723 goto fail_block_groups;
2724 }
2725 fs_info->num_tolerated_disk_barrier_failures =
2726 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2727 if (fs_info->fs_devices->missing_devices >
2728 fs_info->num_tolerated_disk_barrier_failures &&
2729 !(sb->s_flags & MS_RDONLY)) {
2730 printk(KERN_WARNING
2731 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2732 goto fail_block_groups;
2733 }
2734
2735 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2736 "btrfs-cleaner");
2737 if (IS_ERR(fs_info->cleaner_kthread))
2738 goto fail_block_groups;
2739
2740 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2741 tree_root,
2742 "btrfs-transaction");
2743 if (IS_ERR(fs_info->transaction_kthread))
2744 goto fail_cleaner;
2745
2746 if (!btrfs_test_opt(tree_root, SSD) &&
2747 !btrfs_test_opt(tree_root, NOSSD) &&
2748 !fs_info->fs_devices->rotating) {
2749 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2750 "mode\n");
2751 btrfs_set_opt(fs_info->mount_opt, SSD);
2752 }
2753
2754#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2755 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2756 ret = btrfsic_mount(tree_root, fs_devices,
2757 btrfs_test_opt(tree_root,
2758 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2759 1 : 0,
2760 fs_info->check_integrity_print_mask);
2761 if (ret)
2762 printk(KERN_WARNING "btrfs: failed to initialize"
2763 " integrity check module %s\n", sb->s_id);
2764 }
2765#endif
2766 ret = btrfs_read_qgroup_config(fs_info);
2767 if (ret)
2768 goto fail_trans_kthread;
2769
2770 /* do not make disk changes in broken FS */
2771 if (btrfs_super_log_root(disk_super) != 0) {
2772 u64 bytenr = btrfs_super_log_root(disk_super);
2773
2774 if (fs_devices->rw_devices == 0) {
2775 printk(KERN_WARNING "Btrfs log replay required "
2776 "on RO media\n");
2777 err = -EIO;
2778 goto fail_qgroup;
2779 }
2780 blocksize =
2781 btrfs_level_size(tree_root,
2782 btrfs_super_log_root_level(disk_super));
2783
2784 log_tree_root = btrfs_alloc_root(fs_info);
2785 if (!log_tree_root) {
2786 err = -ENOMEM;
2787 goto fail_qgroup;
2788 }
2789
2790 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2791 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2792
2793 log_tree_root->node = read_tree_block(tree_root, bytenr,
2794 blocksize,
2795 generation + 1);
2796 if (!log_tree_root->node ||
2797 !extent_buffer_uptodate(log_tree_root->node)) {
2798 printk(KERN_ERR "btrfs: failed to read log tree\n");
2799 free_extent_buffer(log_tree_root->node);
2800 kfree(log_tree_root);
2801 goto fail_trans_kthread;
2802 }
2803 /* returns with log_tree_root freed on success */
2804 ret = btrfs_recover_log_trees(log_tree_root);
2805 if (ret) {
2806 btrfs_error(tree_root->fs_info, ret,
2807 "Failed to recover log tree");
2808 free_extent_buffer(log_tree_root->node);
2809 kfree(log_tree_root);
2810 goto fail_trans_kthread;
2811 }
2812
2813 if (sb->s_flags & MS_RDONLY) {
2814 ret = btrfs_commit_super(tree_root);
2815 if (ret)
2816 goto fail_trans_kthread;
2817 }
2818 }
2819
2820 ret = btrfs_find_orphan_roots(tree_root);
2821 if (ret)
2822 goto fail_trans_kthread;
2823
2824 if (!(sb->s_flags & MS_RDONLY)) {
2825 ret = btrfs_cleanup_fs_roots(fs_info);
2826 if (ret)
2827 goto fail_trans_kthread;
2828
2829 ret = btrfs_recover_relocation(tree_root);
2830 if (ret < 0) {
2831 printk(KERN_WARNING
2832 "btrfs: failed to recover relocation\n");
2833 err = -EINVAL;
2834 goto fail_qgroup;
2835 }
2836 }
2837
2838 location.objectid = BTRFS_FS_TREE_OBJECTID;
2839 location.type = BTRFS_ROOT_ITEM_KEY;
2840 location.offset = 0;
2841
2842 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2843 if (IS_ERR(fs_info->fs_root)) {
2844 err = PTR_ERR(fs_info->fs_root);
2845 goto fail_qgroup;
2846 }
2847
2848 if (sb->s_flags & MS_RDONLY)
2849 return 0;
2850
2851 down_read(&fs_info->cleanup_work_sem);
2852 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2853 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2854 up_read(&fs_info->cleanup_work_sem);
2855 close_ctree(tree_root);
2856 return ret;
2857 }
2858 up_read(&fs_info->cleanup_work_sem);
2859
2860 ret = btrfs_resume_balance_async(fs_info);
2861 if (ret) {
2862 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2863 close_ctree(tree_root);
2864 return ret;
2865 }
2866
2867 ret = btrfs_resume_dev_replace_async(fs_info);
2868 if (ret) {
2869 pr_warn("btrfs: failed to resume dev_replace\n");
2870 close_ctree(tree_root);
2871 return ret;
2872 }
2873
2874 return 0;
2875
2876fail_qgroup:
2877 btrfs_free_qgroup_config(fs_info);
2878fail_trans_kthread:
2879 kthread_stop(fs_info->transaction_kthread);
2880 btrfs_cleanup_transaction(fs_info->tree_root);
2881 del_fs_roots(fs_info);
2882fail_cleaner:
2883 kthread_stop(fs_info->cleaner_kthread);
2884
2885 /*
2886 * make sure we're done with the btree inode before we stop our
2887 * kthreads
2888 */
2889 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2890
2891fail_block_groups:
2892 btrfs_put_block_group_cache(fs_info);
2893 btrfs_free_block_groups(fs_info);
2894
2895fail_tree_roots:
2896 free_root_pointers(fs_info, 1);
2897 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2898
2899fail_sb_buffer:
2900 btrfs_stop_all_workers(fs_info);
2901fail_alloc:
2902fail_iput:
2903 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2904
2905 iput(fs_info->btree_inode);
2906fail_delalloc_bytes:
2907 percpu_counter_destroy(&fs_info->delalloc_bytes);
2908fail_dirty_metadata_bytes:
2909 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2910fail_bdi:
2911 bdi_destroy(&fs_info->bdi);
2912fail_srcu:
2913 cleanup_srcu_struct(&fs_info->subvol_srcu);
2914fail:
2915 btrfs_free_stripe_hash_table(fs_info);
2916 btrfs_close_devices(fs_info->fs_devices);
2917 return err;
2918
2919recovery_tree_root:
2920 if (!btrfs_test_opt(tree_root, RECOVERY))
2921 goto fail_tree_roots;
2922
2923 free_root_pointers(fs_info, 0);
2924
2925 /* don't use the log in recovery mode, it won't be valid */
2926 btrfs_set_super_log_root(disk_super, 0);
2927
2928 /* we can't trust the free space cache either */
2929 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2930
2931 ret = next_root_backup(fs_info, fs_info->super_copy,
2932 &num_backups_tried, &backup_index);
2933 if (ret == -1)
2934 goto fail_block_groups;
2935 goto retry_root_backup;
2936}
2937
2938static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2939{
2940 if (uptodate) {
2941 set_buffer_uptodate(bh);
2942 } else {
2943 struct btrfs_device *device = (struct btrfs_device *)
2944 bh->b_private;
2945
2946 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2947 "I/O error on %s\n",
2948 rcu_str_deref(device->name));
2949 /* note, we dont' set_buffer_write_io_error because we have
2950 * our own ways of dealing with the IO errors
2951 */
2952 clear_buffer_uptodate(bh);
2953 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2954 }
2955 unlock_buffer(bh);
2956 put_bh(bh);
2957}
2958
2959struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2960{
2961 struct buffer_head *bh;
2962 struct buffer_head *latest = NULL;
2963 struct btrfs_super_block *super;
2964 int i;
2965 u64 transid = 0;
2966 u64 bytenr;
2967
2968 /* we would like to check all the supers, but that would make
2969 * a btrfs mount succeed after a mkfs from a different FS.
2970 * So, we need to add a special mount option to scan for
2971 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2972 */
2973 for (i = 0; i < 1; i++) {
2974 bytenr = btrfs_sb_offset(i);
2975 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2976 break;
2977 bh = __bread(bdev, bytenr / 4096, 4096);
2978 if (!bh)
2979 continue;
2980
2981 super = (struct btrfs_super_block *)bh->b_data;
2982 if (btrfs_super_bytenr(super) != bytenr ||
2983 super->magic != cpu_to_le64(BTRFS_MAGIC)) {
2984 brelse(bh);
2985 continue;
2986 }
2987
2988 if (!latest || btrfs_super_generation(super) > transid) {
2989 brelse(latest);
2990 latest = bh;
2991 transid = btrfs_super_generation(super);
2992 } else {
2993 brelse(bh);
2994 }
2995 }
2996 return latest;
2997}
2998
2999/*
3000 * this should be called twice, once with wait == 0 and
3001 * once with wait == 1. When wait == 0 is done, all the buffer heads
3002 * we write are pinned.
3003 *
3004 * They are released when wait == 1 is done.
3005 * max_mirrors must be the same for both runs, and it indicates how
3006 * many supers on this one device should be written.
3007 *
3008 * max_mirrors == 0 means to write them all.
3009 */
3010static int write_dev_supers(struct btrfs_device *device,
3011 struct btrfs_super_block *sb,
3012 int do_barriers, int wait, int max_mirrors)
3013{
3014 struct buffer_head *bh;
3015 int i;
3016 int ret;
3017 int errors = 0;
3018 u32 crc;
3019 u64 bytenr;
3020
3021 if (max_mirrors == 0)
3022 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3023
3024 for (i = 0; i < max_mirrors; i++) {
3025 bytenr = btrfs_sb_offset(i);
3026 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3027 break;
3028
3029 if (wait) {
3030 bh = __find_get_block(device->bdev, bytenr / 4096,
3031 BTRFS_SUPER_INFO_SIZE);
3032 if (!bh) {
3033 errors++;
3034 continue;
3035 }
3036 wait_on_buffer(bh);
3037 if (!buffer_uptodate(bh))
3038 errors++;
3039
3040 /* drop our reference */
3041 brelse(bh);
3042
3043 /* drop the reference from the wait == 0 run */
3044 brelse(bh);
3045 continue;
3046 } else {
3047 btrfs_set_super_bytenr(sb, bytenr);
3048
3049 crc = ~(u32)0;
3050 crc = btrfs_csum_data((char *)sb +
3051 BTRFS_CSUM_SIZE, crc,
3052 BTRFS_SUPER_INFO_SIZE -
3053 BTRFS_CSUM_SIZE);
3054 btrfs_csum_final(crc, sb->csum);
3055
3056 /*
3057 * one reference for us, and we leave it for the
3058 * caller
3059 */
3060 bh = __getblk(device->bdev, bytenr / 4096,
3061 BTRFS_SUPER_INFO_SIZE);
3062 if (!bh) {
3063 printk(KERN_ERR "btrfs: couldn't get super "
3064 "buffer head for bytenr %Lu\n", bytenr);
3065 errors++;
3066 continue;
3067 }
3068
3069 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3070
3071 /* one reference for submit_bh */
3072 get_bh(bh);
3073
3074 set_buffer_uptodate(bh);
3075 lock_buffer(bh);
3076 bh->b_end_io = btrfs_end_buffer_write_sync;
3077 bh->b_private = device;
3078 }
3079
3080 /*
3081 * we fua the first super. The others we allow
3082 * to go down lazy.
3083 */
3084 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3085 if (ret)
3086 errors++;
3087 }
3088 return errors < i ? 0 : -1;
3089}
3090
3091/*
3092 * endio for the write_dev_flush, this will wake anyone waiting
3093 * for the barrier when it is done
3094 */
3095static void btrfs_end_empty_barrier(struct bio *bio, int err)
3096{
3097 if (err) {
3098 if (err == -EOPNOTSUPP)
3099 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3100 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3101 }
3102 if (bio->bi_private)
3103 complete(bio->bi_private);
3104 bio_put(bio);
3105}
3106
3107/*
3108 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3109 * sent down. With wait == 1, it waits for the previous flush.
3110 *
3111 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3112 * capable
3113 */
3114static int write_dev_flush(struct btrfs_device *device, int wait)
3115{
3116 struct bio *bio;
3117 int ret = 0;
3118
3119 if (device->nobarriers)
3120 return 0;
3121
3122 if (wait) {
3123 bio = device->flush_bio;
3124 if (!bio)
3125 return 0;
3126
3127 wait_for_completion(&device->flush_wait);
3128
3129 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3130 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3131 rcu_str_deref(device->name));
3132 device->nobarriers = 1;
3133 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3134 ret = -EIO;
3135 btrfs_dev_stat_inc_and_print(device,
3136 BTRFS_DEV_STAT_FLUSH_ERRS);
3137 }
3138
3139 /* drop the reference from the wait == 0 run */
3140 bio_put(bio);
3141 device->flush_bio = NULL;
3142
3143 return ret;
3144 }
3145
3146 /*
3147 * one reference for us, and we leave it for the
3148 * caller
3149 */
3150 device->flush_bio = NULL;
3151 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3152 if (!bio)
3153 return -ENOMEM;
3154
3155 bio->bi_end_io = btrfs_end_empty_barrier;
3156 bio->bi_bdev = device->bdev;
3157 init_completion(&device->flush_wait);
3158 bio->bi_private = &device->flush_wait;
3159 device->flush_bio = bio;
3160
3161 bio_get(bio);
3162 btrfsic_submit_bio(WRITE_FLUSH, bio);
3163
3164 return 0;
3165}
3166
3167/*
3168 * send an empty flush down to each device in parallel,
3169 * then wait for them
3170 */
3171static int barrier_all_devices(struct btrfs_fs_info *info)
3172{
3173 struct list_head *head;
3174 struct btrfs_device *dev;
3175 int errors_send = 0;
3176 int errors_wait = 0;
3177 int ret;
3178
3179 /* send down all the barriers */
3180 head = &info->fs_devices->devices;
3181 list_for_each_entry_rcu(dev, head, dev_list) {
3182 if (!dev->bdev) {
3183 errors_send++;
3184 continue;
3185 }
3186 if (!dev->in_fs_metadata || !dev->writeable)
3187 continue;
3188
3189 ret = write_dev_flush(dev, 0);
3190 if (ret)
3191 errors_send++;
3192 }
3193
3194 /* wait for all the barriers */
3195 list_for_each_entry_rcu(dev, head, dev_list) {
3196 if (!dev->bdev) {
3197 errors_wait++;
3198 continue;
3199 }
3200 if (!dev->in_fs_metadata || !dev->writeable)
3201 continue;
3202
3203 ret = write_dev_flush(dev, 1);
3204 if (ret)
3205 errors_wait++;
3206 }
3207 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3208 errors_wait > info->num_tolerated_disk_barrier_failures)
3209 return -EIO;
3210 return 0;
3211}
3212
3213int btrfs_calc_num_tolerated_disk_barrier_failures(
3214 struct btrfs_fs_info *fs_info)
3215{
3216 struct btrfs_ioctl_space_info space;
3217 struct btrfs_space_info *sinfo;
3218 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3219 BTRFS_BLOCK_GROUP_SYSTEM,
3220 BTRFS_BLOCK_GROUP_METADATA,
3221 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3222 int num_types = 4;
3223 int i;
3224 int c;
3225 int num_tolerated_disk_barrier_failures =
3226 (int)fs_info->fs_devices->num_devices;
3227
3228 for (i = 0; i < num_types; i++) {
3229 struct btrfs_space_info *tmp;
3230
3231 sinfo = NULL;
3232 rcu_read_lock();
3233 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3234 if (tmp->flags == types[i]) {
3235 sinfo = tmp;
3236 break;
3237 }
3238 }
3239 rcu_read_unlock();
3240
3241 if (!sinfo)
3242 continue;
3243
3244 down_read(&sinfo->groups_sem);
3245 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3246 if (!list_empty(&sinfo->block_groups[c])) {
3247 u64 flags;
3248
3249 btrfs_get_block_group_info(
3250 &sinfo->block_groups[c], &space);
3251 if (space.total_bytes == 0 ||
3252 space.used_bytes == 0)
3253 continue;
3254 flags = space.flags;
3255 /*
3256 * return
3257 * 0: if dup, single or RAID0 is configured for
3258 * any of metadata, system or data, else
3259 * 1: if RAID5 is configured, or if RAID1 or
3260 * RAID10 is configured and only two mirrors
3261 * are used, else
3262 * 2: if RAID6 is configured, else
3263 * num_mirrors - 1: if RAID1 or RAID10 is
3264 * configured and more than
3265 * 2 mirrors are used.
3266 */
3267 if (num_tolerated_disk_barrier_failures > 0 &&
3268 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3269 BTRFS_BLOCK_GROUP_RAID0)) ||
3270 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3271 == 0)))
3272 num_tolerated_disk_barrier_failures = 0;
3273 else if (num_tolerated_disk_barrier_failures > 1) {
3274 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3275 BTRFS_BLOCK_GROUP_RAID5 |
3276 BTRFS_BLOCK_GROUP_RAID10)) {
3277 num_tolerated_disk_barrier_failures = 1;
3278 } else if (flags &
3279 BTRFS_BLOCK_GROUP_RAID6) {
3280 num_tolerated_disk_barrier_failures = 2;
3281 }
3282 }
3283 }
3284 }
3285 up_read(&sinfo->groups_sem);
3286 }
3287
3288 return num_tolerated_disk_barrier_failures;
3289}
3290
3291static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3292{
3293 struct list_head *head;
3294 struct btrfs_device *dev;
3295 struct btrfs_super_block *sb;
3296 struct btrfs_dev_item *dev_item;
3297 int ret;
3298 int do_barriers;
3299 int max_errors;
3300 int total_errors = 0;
3301 u64 flags;
3302
3303 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3304 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3305 backup_super_roots(root->fs_info);
3306
3307 sb = root->fs_info->super_for_commit;
3308 dev_item = &sb->dev_item;
3309
3310 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3311 head = &root->fs_info->fs_devices->devices;
3312
3313 if (do_barriers) {
3314 ret = barrier_all_devices(root->fs_info);
3315 if (ret) {
3316 mutex_unlock(
3317 &root->fs_info->fs_devices->device_list_mutex);
3318 btrfs_error(root->fs_info, ret,
3319 "errors while submitting device barriers.");
3320 return ret;
3321 }
3322 }
3323
3324 list_for_each_entry_rcu(dev, head, dev_list) {
3325 if (!dev->bdev) {
3326 total_errors++;
3327 continue;
3328 }
3329 if (!dev->in_fs_metadata || !dev->writeable)
3330 continue;
3331
3332 btrfs_set_stack_device_generation(dev_item, 0);
3333 btrfs_set_stack_device_type(dev_item, dev->type);
3334 btrfs_set_stack_device_id(dev_item, dev->devid);
3335 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3336 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3337 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3338 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3339 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3340 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3341 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3342
3343 flags = btrfs_super_flags(sb);
3344 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3345
3346 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3347 if (ret)
3348 total_errors++;
3349 }
3350 if (total_errors > max_errors) {
3351 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3352 total_errors);
3353
3354 /* This shouldn't happen. FUA is masked off if unsupported */
3355 BUG();
3356 }
3357
3358 total_errors = 0;
3359 list_for_each_entry_rcu(dev, head, dev_list) {
3360 if (!dev->bdev)
3361 continue;
3362 if (!dev->in_fs_metadata || !dev->writeable)
3363 continue;
3364
3365 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3366 if (ret)
3367 total_errors++;
3368 }
3369 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3370 if (total_errors > max_errors) {
3371 btrfs_error(root->fs_info, -EIO,
3372 "%d errors while writing supers", total_errors);
3373 return -EIO;
3374 }
3375 return 0;
3376}
3377
3378int write_ctree_super(struct btrfs_trans_handle *trans,
3379 struct btrfs_root *root, int max_mirrors)
3380{
3381 int ret;
3382
3383 ret = write_all_supers(root, max_mirrors);
3384 return ret;
3385}
3386
3387/* Drop a fs root from the radix tree and free it. */
3388void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3389 struct btrfs_root *root)
3390{
3391 spin_lock(&fs_info->fs_roots_radix_lock);
3392 radix_tree_delete(&fs_info->fs_roots_radix,
3393 (unsigned long)root->root_key.objectid);
3394 spin_unlock(&fs_info->fs_roots_radix_lock);
3395
3396 if (btrfs_root_refs(&root->root_item) == 0)
3397 synchronize_srcu(&fs_info->subvol_srcu);
3398
3399 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3400 btrfs_free_log(NULL, root);
3401 btrfs_free_log_root_tree(NULL, fs_info);
3402 }
3403
3404 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3405 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3406 free_fs_root(root);
3407}
3408
3409static void free_fs_root(struct btrfs_root *root)
3410{
3411 iput(root->cache_inode);
3412 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3413 if (root->anon_dev)
3414 free_anon_bdev(root->anon_dev);
3415 free_extent_buffer(root->node);
3416 free_extent_buffer(root->commit_root);
3417 kfree(root->free_ino_ctl);
3418 kfree(root->free_ino_pinned);
3419 kfree(root->name);
3420 kfree(root);
3421}
3422
3423void btrfs_free_fs_root(struct btrfs_root *root)
3424{
3425 free_fs_root(root);
3426}
3427
3428int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3429{
3430 u64 root_objectid = 0;
3431 struct btrfs_root *gang[8];
3432 int i;
3433 int ret;
3434
3435 while (1) {
3436 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3437 (void **)gang, root_objectid,
3438 ARRAY_SIZE(gang));
3439 if (!ret)
3440 break;
3441
3442 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3443 for (i = 0; i < ret; i++) {
3444 int err;
3445
3446 root_objectid = gang[i]->root_key.objectid;
3447 err = btrfs_orphan_cleanup(gang[i]);
3448 if (err)
3449 return err;
3450 }
3451 root_objectid++;
3452 }
3453 return 0;
3454}
3455
3456int btrfs_commit_super(struct btrfs_root *root)
3457{
3458 struct btrfs_trans_handle *trans;
3459 int ret;
3460
3461 mutex_lock(&root->fs_info->cleaner_mutex);
3462 btrfs_run_delayed_iputs(root);
3463 mutex_unlock(&root->fs_info->cleaner_mutex);
3464 wake_up_process(root->fs_info->cleaner_kthread);
3465
3466 /* wait until ongoing cleanup work done */
3467 down_write(&root->fs_info->cleanup_work_sem);
3468 up_write(&root->fs_info->cleanup_work_sem);
3469
3470 trans = btrfs_join_transaction(root);
3471 if (IS_ERR(trans))
3472 return PTR_ERR(trans);
3473 ret = btrfs_commit_transaction(trans, root);
3474 if (ret)
3475 return ret;
3476 /* run commit again to drop the original snapshot */
3477 trans = btrfs_join_transaction(root);
3478 if (IS_ERR(trans))
3479 return PTR_ERR(trans);
3480 ret = btrfs_commit_transaction(trans, root);
3481 if (ret)
3482 return ret;
3483 ret = btrfs_write_and_wait_transaction(NULL, root);
3484 if (ret) {
3485 btrfs_error(root->fs_info, ret,
3486 "Failed to sync btree inode to disk.");
3487 return ret;
3488 }
3489
3490 ret = write_ctree_super(NULL, root, 0);
3491 return ret;
3492}
3493
3494int close_ctree(struct btrfs_root *root)
3495{
3496 struct btrfs_fs_info *fs_info = root->fs_info;
3497 int ret;
3498
3499 fs_info->closing = 1;
3500 smp_mb();
3501
3502 /* pause restriper - we want to resume on mount */
3503 btrfs_pause_balance(fs_info);
3504
3505 btrfs_dev_replace_suspend_for_unmount(fs_info);
3506
3507 btrfs_scrub_cancel(fs_info);
3508
3509 /* wait for any defraggers to finish */
3510 wait_event(fs_info->transaction_wait,
3511 (atomic_read(&fs_info->defrag_running) == 0));
3512
3513 /* clear out the rbtree of defraggable inodes */
3514 btrfs_cleanup_defrag_inodes(fs_info);
3515
3516 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3517 ret = btrfs_commit_super(root);
3518 if (ret)
3519 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3520 }
3521
3522 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3523 btrfs_error_commit_super(root);
3524
3525 btrfs_put_block_group_cache(fs_info);
3526
3527 kthread_stop(fs_info->transaction_kthread);
3528 kthread_stop(fs_info->cleaner_kthread);
3529
3530 fs_info->closing = 2;
3531 smp_mb();
3532
3533 btrfs_free_qgroup_config(root->fs_info);
3534
3535 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3536 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3537 percpu_counter_sum(&fs_info->delalloc_bytes));
3538 }
3539
3540 btrfs_free_block_groups(fs_info);
3541
3542 btrfs_stop_all_workers(fs_info);
3543
3544 del_fs_roots(fs_info);
3545
3546 free_root_pointers(fs_info, 1);
3547
3548 iput(fs_info->btree_inode);
3549
3550#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3551 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3552 btrfsic_unmount(root, fs_info->fs_devices);
3553#endif
3554
3555 btrfs_close_devices(fs_info->fs_devices);
3556 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3557
3558 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3559 percpu_counter_destroy(&fs_info->delalloc_bytes);
3560 bdi_destroy(&fs_info->bdi);
3561 cleanup_srcu_struct(&fs_info->subvol_srcu);
3562
3563 btrfs_free_stripe_hash_table(fs_info);
3564
3565 return 0;
3566}
3567
3568int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3569 int atomic)
3570{
3571 int ret;
3572 struct inode *btree_inode = buf->pages[0]->mapping->host;
3573
3574 ret = extent_buffer_uptodate(buf);
3575 if (!ret)
3576 return ret;
3577
3578 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3579 parent_transid, atomic);
3580 if (ret == -EAGAIN)
3581 return ret;
3582 return !ret;
3583}
3584
3585int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3586{
3587 return set_extent_buffer_uptodate(buf);
3588}
3589
3590void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3591{
3592 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3593 u64 transid = btrfs_header_generation(buf);
3594 int was_dirty;
3595
3596 btrfs_assert_tree_locked(buf);
3597 if (transid != root->fs_info->generation)
3598 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3599 "found %llu running %llu\n",
3600 (unsigned long long)buf->start,
3601 (unsigned long long)transid,
3602 (unsigned long long)root->fs_info->generation);
3603 was_dirty = set_extent_buffer_dirty(buf);
3604 if (!was_dirty)
3605 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3606 buf->len,
3607 root->fs_info->dirty_metadata_batch);
3608}
3609
3610static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3611 int flush_delayed)
3612{
3613 /*
3614 * looks as though older kernels can get into trouble with
3615 * this code, they end up stuck in balance_dirty_pages forever
3616 */
3617 int ret;
3618
3619 if (current->flags & PF_MEMALLOC)
3620 return;
3621
3622 if (flush_delayed)
3623 btrfs_balance_delayed_items(root);
3624
3625 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3626 BTRFS_DIRTY_METADATA_THRESH);
3627 if (ret > 0) {
3628 balance_dirty_pages_ratelimited(
3629 root->fs_info->btree_inode->i_mapping);
3630 }
3631 return;
3632}
3633
3634void btrfs_btree_balance_dirty(struct btrfs_root *root)
3635{
3636 __btrfs_btree_balance_dirty(root, 1);
3637}
3638
3639void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3640{
3641 __btrfs_btree_balance_dirty(root, 0);
3642}
3643
3644int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3645{
3646 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3647 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3648}
3649
3650static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3651 int read_only)
3652{
3653 /*
3654 * Placeholder for checks
3655 */
3656 return 0;
3657}
3658
3659static void btrfs_error_commit_super(struct btrfs_root *root)
3660{
3661 mutex_lock(&root->fs_info->cleaner_mutex);
3662 btrfs_run_delayed_iputs(root);
3663 mutex_unlock(&root->fs_info->cleaner_mutex);
3664
3665 down_write(&root->fs_info->cleanup_work_sem);
3666 up_write(&root->fs_info->cleanup_work_sem);
3667
3668 /* cleanup FS via transaction */
3669 btrfs_cleanup_transaction(root);
3670}
3671
3672static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3673 struct btrfs_root *root)
3674{
3675 struct btrfs_inode *btrfs_inode;
3676 struct list_head splice;
3677
3678 INIT_LIST_HEAD(&splice);
3679
3680 mutex_lock(&root->fs_info->ordered_operations_mutex);
3681 spin_lock(&root->fs_info->ordered_extent_lock);
3682
3683 list_splice_init(&t->ordered_operations, &splice);
3684 while (!list_empty(&splice)) {
3685 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3686 ordered_operations);
3687
3688 list_del_init(&btrfs_inode->ordered_operations);
3689 spin_unlock(&root->fs_info->ordered_extent_lock);
3690
3691 btrfs_invalidate_inodes(btrfs_inode->root);
3692
3693 spin_lock(&root->fs_info->ordered_extent_lock);
3694 }
3695
3696 spin_unlock(&root->fs_info->ordered_extent_lock);
3697 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3698}
3699
3700static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3701{
3702 struct btrfs_ordered_extent *ordered;
3703
3704 spin_lock(&root->fs_info->ordered_extent_lock);
3705 /*
3706 * This will just short circuit the ordered completion stuff which will
3707 * make sure the ordered extent gets properly cleaned up.
3708 */
3709 list_for_each_entry(ordered, &root->fs_info->ordered_extents,
3710 root_extent_list)
3711 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3712 spin_unlock(&root->fs_info->ordered_extent_lock);
3713}
3714
3715int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3716 struct btrfs_root *root)
3717{
3718 struct rb_node *node;
3719 struct btrfs_delayed_ref_root *delayed_refs;
3720 struct btrfs_delayed_ref_node *ref;
3721 int ret = 0;
3722
3723 delayed_refs = &trans->delayed_refs;
3724
3725 spin_lock(&delayed_refs->lock);
3726 if (delayed_refs->num_entries == 0) {
3727 spin_unlock(&delayed_refs->lock);
3728 printk(KERN_INFO "delayed_refs has NO entry\n");
3729 return ret;
3730 }
3731
3732 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3733 struct btrfs_delayed_ref_head *head = NULL;
3734
3735 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3736 atomic_set(&ref->refs, 1);
3737 if (btrfs_delayed_ref_is_head(ref)) {
3738
3739 head = btrfs_delayed_node_to_head(ref);
3740 if (!mutex_trylock(&head->mutex)) {
3741 atomic_inc(&ref->refs);
3742 spin_unlock(&delayed_refs->lock);
3743
3744 /* Need to wait for the delayed ref to run */
3745 mutex_lock(&head->mutex);
3746 mutex_unlock(&head->mutex);
3747 btrfs_put_delayed_ref(ref);
3748
3749 spin_lock(&delayed_refs->lock);
3750 continue;
3751 }
3752
3753 if (head->must_insert_reserved)
3754 btrfs_pin_extent(root, ref->bytenr,
3755 ref->num_bytes, 1);
3756 btrfs_free_delayed_extent_op(head->extent_op);
3757 delayed_refs->num_heads--;
3758 if (list_empty(&head->cluster))
3759 delayed_refs->num_heads_ready--;
3760 list_del_init(&head->cluster);
3761 }
3762
3763 ref->in_tree = 0;
3764 rb_erase(&ref->rb_node, &delayed_refs->root);
3765 delayed_refs->num_entries--;
3766 if (head)
3767 mutex_unlock(&head->mutex);
3768 spin_unlock(&delayed_refs->lock);
3769 btrfs_put_delayed_ref(ref);
3770
3771 cond_resched();
3772 spin_lock(&delayed_refs->lock);
3773 }
3774
3775 spin_unlock(&delayed_refs->lock);
3776
3777 return ret;
3778}
3779
3780static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3781{
3782 struct btrfs_pending_snapshot *snapshot;
3783 struct list_head splice;
3784
3785 INIT_LIST_HEAD(&splice);
3786
3787 list_splice_init(&t->pending_snapshots, &splice);
3788
3789 while (!list_empty(&splice)) {
3790 snapshot = list_entry(splice.next,
3791 struct btrfs_pending_snapshot,
3792 list);
3793 snapshot->error = -ECANCELED;
3794 list_del_init(&snapshot->list);
3795 }
3796}
3797
3798static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3799{
3800 struct btrfs_inode *btrfs_inode;
3801 struct list_head splice;
3802
3803 INIT_LIST_HEAD(&splice);
3804
3805 spin_lock(&root->fs_info->delalloc_lock);
3806 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3807
3808 while (!list_empty(&splice)) {
3809 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3810 delalloc_inodes);
3811
3812 list_del_init(&btrfs_inode->delalloc_inodes);
3813 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3814 &btrfs_inode->runtime_flags);
3815 spin_unlock(&root->fs_info->delalloc_lock);
3816
3817 btrfs_invalidate_inodes(btrfs_inode->root);
3818
3819 spin_lock(&root->fs_info->delalloc_lock);
3820 }
3821
3822 spin_unlock(&root->fs_info->delalloc_lock);
3823}
3824
3825static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3826 struct extent_io_tree *dirty_pages,
3827 int mark)
3828{
3829 int ret;
3830 struct extent_buffer *eb;
3831 u64 start = 0;
3832 u64 end;
3833
3834 while (1) {
3835 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3836 mark, NULL);
3837 if (ret)
3838 break;
3839
3840 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3841 while (start <= end) {
3842 eb = btrfs_find_tree_block(root, start,
3843 root->leafsize);
3844 start += root->leafsize;
3845 if (!eb)
3846 continue;
3847 wait_on_extent_buffer_writeback(eb);
3848
3849 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3850 &eb->bflags))
3851 clear_extent_buffer_dirty(eb);
3852 free_extent_buffer_stale(eb);
3853 }
3854 }
3855
3856 return ret;
3857}
3858
3859static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3860 struct extent_io_tree *pinned_extents)
3861{
3862 struct extent_io_tree *unpin;
3863 u64 start;
3864 u64 end;
3865 int ret;
3866 bool loop = true;
3867
3868 unpin = pinned_extents;
3869again:
3870 while (1) {
3871 ret = find_first_extent_bit(unpin, 0, &start, &end,
3872 EXTENT_DIRTY, NULL);
3873 if (ret)
3874 break;
3875
3876 /* opt_discard */
3877 if (btrfs_test_opt(root, DISCARD))
3878 ret = btrfs_error_discard_extent(root, start,
3879 end + 1 - start,
3880 NULL);
3881
3882 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3883 btrfs_error_unpin_extent_range(root, start, end);
3884 cond_resched();
3885 }
3886
3887 if (loop) {
3888 if (unpin == &root->fs_info->freed_extents[0])
3889 unpin = &root->fs_info->freed_extents[1];
3890 else
3891 unpin = &root->fs_info->freed_extents[0];
3892 loop = false;
3893 goto again;
3894 }
3895
3896 return 0;
3897}
3898
3899void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3900 struct btrfs_root *root)
3901{
3902 btrfs_destroy_delayed_refs(cur_trans, root);
3903 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3904 cur_trans->dirty_pages.dirty_bytes);
3905
3906 /* FIXME: cleanup wait for commit */
3907 cur_trans->in_commit = 1;
3908 cur_trans->blocked = 1;
3909 wake_up(&root->fs_info->transaction_blocked_wait);
3910
3911 btrfs_evict_pending_snapshots(cur_trans);
3912
3913 cur_trans->blocked = 0;
3914 wake_up(&root->fs_info->transaction_wait);
3915
3916 cur_trans->commit_done = 1;
3917 wake_up(&cur_trans->commit_wait);
3918
3919 btrfs_destroy_delayed_inodes(root);
3920 btrfs_assert_delayed_root_empty(root);
3921
3922 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3923 EXTENT_DIRTY);
3924 btrfs_destroy_pinned_extent(root,
3925 root->fs_info->pinned_extents);
3926
3927 /*
3928 memset(cur_trans, 0, sizeof(*cur_trans));
3929 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3930 */
3931}
3932
3933static int btrfs_cleanup_transaction(struct btrfs_root *root)
3934{
3935 struct btrfs_transaction *t;
3936 LIST_HEAD(list);
3937
3938 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3939
3940 spin_lock(&root->fs_info->trans_lock);
3941 list_splice_init(&root->fs_info->trans_list, &list);
3942 root->fs_info->trans_no_join = 1;
3943 spin_unlock(&root->fs_info->trans_lock);
3944
3945 while (!list_empty(&list)) {
3946 t = list_entry(list.next, struct btrfs_transaction, list);
3947
3948 btrfs_destroy_ordered_operations(t, root);
3949
3950 btrfs_destroy_ordered_extents(root);
3951
3952 btrfs_destroy_delayed_refs(t, root);
3953
3954 /* FIXME: cleanup wait for commit */
3955 t->in_commit = 1;
3956 t->blocked = 1;
3957 smp_mb();
3958 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3959 wake_up(&root->fs_info->transaction_blocked_wait);
3960
3961 btrfs_evict_pending_snapshots(t);
3962
3963 t->blocked = 0;
3964 smp_mb();
3965 if (waitqueue_active(&root->fs_info->transaction_wait))
3966 wake_up(&root->fs_info->transaction_wait);
3967
3968 t->commit_done = 1;
3969 smp_mb();
3970 if (waitqueue_active(&t->commit_wait))
3971 wake_up(&t->commit_wait);
3972
3973 btrfs_destroy_delayed_inodes(root);
3974 btrfs_assert_delayed_root_empty(root);
3975
3976 btrfs_destroy_delalloc_inodes(root);
3977
3978 spin_lock(&root->fs_info->trans_lock);
3979 root->fs_info->running_transaction = NULL;
3980 spin_unlock(&root->fs_info->trans_lock);
3981
3982 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3983 EXTENT_DIRTY);
3984
3985 btrfs_destroy_pinned_extent(root,
3986 root->fs_info->pinned_extents);
3987
3988 atomic_set(&t->use_count, 0);
3989 list_del_init(&t->list);
3990 memset(t, 0, sizeof(*t));
3991 kmem_cache_free(btrfs_transaction_cachep, t);
3992 }
3993
3994 spin_lock(&root->fs_info->trans_lock);
3995 root->fs_info->trans_no_join = 0;
3996 spin_unlock(&root->fs_info->trans_lock);
3997 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3998
3999 return 0;
4000}
4001
4002static struct extent_io_ops btree_extent_io_ops = {
4003 .readpage_end_io_hook = btree_readpage_end_io_hook,
4004 .readpage_io_failed_hook = btree_io_failed_hook,
4005 .submit_bio_hook = btree_submit_bio_hook,
4006 /* note we're sharing with inode.c for the merge bio hook */
4007 .merge_bio_hook = btrfs_merge_bio_hook,
4008};
Linux 6.x block layer and io_uring tree(s)