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