projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Fix leak in __btrfs_map_block error path
[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 btrfs_io_bio *io_bio,
580 u64 phy_offset, struct page *page,
581 u64 start, u64 end, int mirror)
582{
583 struct extent_io_tree *tree;
584 u64 found_start;
585 int found_level;
586 struct extent_buffer *eb;
587 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
588 int ret = 0;
589 int reads_done;
590
591 if (!page->private)
592 goto out;
593
594 tree = &BTRFS_I(page->mapping->host)->io_tree;
595 eb = (struct extent_buffer *)page->private;
596
597 /* the pending IO might have been the only thing that kept this buffer
598 * in memory. Make sure we have a ref for all this other checks
599 */
600 extent_buffer_get(eb);
601
602 reads_done = atomic_dec_and_test(&eb->io_pages);
603 if (!reads_done)
604 goto err;
605
606 eb->read_mirror = mirror;
607 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
608 ret = -EIO;
609 goto err;
610 }
611
612 found_start = btrfs_header_bytenr(eb);
613 if (found_start != eb->start) {
614 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
615 "%llu %llu\n",
616 (unsigned long long)found_start,
617 (unsigned long long)eb->start);
618 ret = -EIO;
619 goto err;
620 }
621 if (check_tree_block_fsid(root, eb)) {
622 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
623 (unsigned long long)eb->start);
624 ret = -EIO;
625 goto err;
626 }
627 found_level = btrfs_header_level(eb);
628 if (found_level >= BTRFS_MAX_LEVEL) {
629 btrfs_info(root->fs_info, "bad tree block level %d\n",
630 (int)btrfs_header_level(eb));
631 ret = -EIO;
632 goto err;
633 }
634
635 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
636 eb, found_level);
637
638 ret = csum_tree_block(root, eb, 1);
639 if (ret) {
640 ret = -EIO;
641 goto err;
642 }
643
644 /*
645 * If this is a leaf block and it is corrupt, set the corrupt bit so
646 * that we don't try and read the other copies of this block, just
647 * return -EIO.
648 */
649 if (found_level == 0 && check_leaf(root, eb)) {
650 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
651 ret = -EIO;
652 }
653
654 if (!ret)
655 set_extent_buffer_uptodate(eb);
656err:
657 if (reads_done &&
658 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
659 btree_readahead_hook(root, eb, eb->start, ret);
660
661 if (ret) {
662 /*
663 * our io error hook is going to dec the io pages
664 * again, we have to make sure it has something
665 * to decrement
666 */
667 atomic_inc(&eb->io_pages);
668 clear_extent_buffer_uptodate(eb);
669 }
670 free_extent_buffer(eb);
671out:
672 return ret;
673}
674
675static int btree_io_failed_hook(struct page *page, int failed_mirror)
676{
677 struct extent_buffer *eb;
678 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
679
680 eb = (struct extent_buffer *)page->private;
681 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
682 eb->read_mirror = failed_mirror;
683 atomic_dec(&eb->io_pages);
684 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
685 btree_readahead_hook(root, eb, eb->start, -EIO);
686 return -EIO; /* we fixed nothing */
687}
688
689static void end_workqueue_bio(struct bio *bio, int err)
690{
691 struct end_io_wq *end_io_wq = bio->bi_private;
692 struct btrfs_fs_info *fs_info;
693
694 fs_info = end_io_wq->info;
695 end_io_wq->error = err;
696 end_io_wq->work.func = end_workqueue_fn;
697 end_io_wq->work.flags = 0;
698
699 if (bio->bi_rw & REQ_WRITE) {
700 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
701 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
702 &end_io_wq->work);
703 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
704 btrfs_queue_worker(&fs_info->endio_freespace_worker,
705 &end_io_wq->work);
706 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
707 btrfs_queue_worker(&fs_info->endio_raid56_workers,
708 &end_io_wq->work);
709 else
710 btrfs_queue_worker(&fs_info->endio_write_workers,
711 &end_io_wq->work);
712 } else {
713 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
714 btrfs_queue_worker(&fs_info->endio_raid56_workers,
715 &end_io_wq->work);
716 else if (end_io_wq->metadata)
717 btrfs_queue_worker(&fs_info->endio_meta_workers,
718 &end_io_wq->work);
719 else
720 btrfs_queue_worker(&fs_info->endio_workers,
721 &end_io_wq->work);
722 }
723}
724
725/*
726 * For the metadata arg you want
727 *
728 * 0 - if data
729 * 1 - if normal metadta
730 * 2 - if writing to the free space cache area
731 * 3 - raid parity work
732 */
733int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
734 int metadata)
735{
736 struct end_io_wq *end_io_wq;
737 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
738 if (!end_io_wq)
739 return -ENOMEM;
740
741 end_io_wq->private = bio->bi_private;
742 end_io_wq->end_io = bio->bi_end_io;
743 end_io_wq->info = info;
744 end_io_wq->error = 0;
745 end_io_wq->bio = bio;
746 end_io_wq->metadata = metadata;
747
748 bio->bi_private = end_io_wq;
749 bio->bi_end_io = end_workqueue_bio;
750 return 0;
751}
752
753unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
754{
755 unsigned long limit = min_t(unsigned long,
756 info->workers.max_workers,
757 info->fs_devices->open_devices);
758 return 256 * limit;
759}
760
761static void run_one_async_start(struct btrfs_work *work)
762{
763 struct async_submit_bio *async;
764 int ret;
765
766 async = container_of(work, struct async_submit_bio, work);
767 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
768 async->mirror_num, async->bio_flags,
769 async->bio_offset);
770 if (ret)
771 async->error = ret;
772}
773
774static void run_one_async_done(struct btrfs_work *work)
775{
776 struct btrfs_fs_info *fs_info;
777 struct async_submit_bio *async;
778 int limit;
779
780 async = container_of(work, struct async_submit_bio, work);
781 fs_info = BTRFS_I(async->inode)->root->fs_info;
782
783 limit = btrfs_async_submit_limit(fs_info);
784 limit = limit * 2 / 3;
785
786 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
787 waitqueue_active(&fs_info->async_submit_wait))
788 wake_up(&fs_info->async_submit_wait);
789
790 /* If an error occured we just want to clean up the bio and move on */
791 if (async->error) {
792 bio_endio(async->bio, async->error);
793 return;
794 }
795
796 async->submit_bio_done(async->inode, async->rw, async->bio,
797 async->mirror_num, async->bio_flags,
798 async->bio_offset);
799}
800
801static void run_one_async_free(struct btrfs_work *work)
802{
803 struct async_submit_bio *async;
804
805 async = container_of(work, struct async_submit_bio, work);
806 kfree(async);
807}
808
809int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
810 int rw, struct bio *bio, int mirror_num,
811 unsigned long bio_flags,
812 u64 bio_offset,
813 extent_submit_bio_hook_t *submit_bio_start,
814 extent_submit_bio_hook_t *submit_bio_done)
815{
816 struct async_submit_bio *async;
817
818 async = kmalloc(sizeof(*async), GFP_NOFS);
819 if (!async)
820 return -ENOMEM;
821
822 async->inode = inode;
823 async->rw = rw;
824 async->bio = bio;
825 async->mirror_num = mirror_num;
826 async->submit_bio_start = submit_bio_start;
827 async->submit_bio_done = submit_bio_done;
828
829 async->work.func = run_one_async_start;
830 async->work.ordered_func = run_one_async_done;
831 async->work.ordered_free = run_one_async_free;
832
833 async->work.flags = 0;
834 async->bio_flags = bio_flags;
835 async->bio_offset = bio_offset;
836
837 async->error = 0;
838
839 atomic_inc(&fs_info->nr_async_submits);
840
841 if (rw & REQ_SYNC)
842 btrfs_set_work_high_prio(&async->work);
843
844 btrfs_queue_worker(&fs_info->workers, &async->work);
845
846 while (atomic_read(&fs_info->async_submit_draining) &&
847 atomic_read(&fs_info->nr_async_submits)) {
848 wait_event(fs_info->async_submit_wait,
849 (atomic_read(&fs_info->nr_async_submits) == 0));
850 }
851
852 return 0;
853}
854
855static int btree_csum_one_bio(struct bio *bio)
856{
857 struct bio_vec *bvec = bio->bi_io_vec;
858 int bio_index = 0;
859 struct btrfs_root *root;
860 int ret = 0;
861
862 WARN_ON(bio->bi_vcnt <= 0);
863 while (bio_index < bio->bi_vcnt) {
864 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
865 ret = csum_dirty_buffer(root, bvec->bv_page);
866 if (ret)
867 break;
868 bio_index++;
869 bvec++;
870 }
871 return ret;
872}
873
874static int __btree_submit_bio_start(struct inode *inode, int rw,
875 struct bio *bio, int mirror_num,
876 unsigned long bio_flags,
877 u64 bio_offset)
878{
879 /*
880 * when we're called for a write, we're already in the async
881 * submission context. Just jump into btrfs_map_bio
882 */
883 return btree_csum_one_bio(bio);
884}
885
886static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
887 int mirror_num, unsigned long bio_flags,
888 u64 bio_offset)
889{
890 int ret;
891
892 /*
893 * when we're called for a write, we're already in the async
894 * submission context. Just jump into btrfs_map_bio
895 */
896 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
897 if (ret)
898 bio_endio(bio, ret);
899 return ret;
900}
901
902static int check_async_write(struct inode *inode, unsigned long bio_flags)
903{
904 if (bio_flags & EXTENT_BIO_TREE_LOG)
905 return 0;
906#ifdef CONFIG_X86
907 if (cpu_has_xmm4_2)
908 return 0;
909#endif
910 return 1;
911}
912
913static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
914 int mirror_num, unsigned long bio_flags,
915 u64 bio_offset)
916{
917 int async = check_async_write(inode, bio_flags);
918 int ret;
919
920 if (!(rw & REQ_WRITE)) {
921 /*
922 * called for a read, do the setup so that checksum validation
923 * can happen in the async kernel threads
924 */
925 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
926 bio, 1);
927 if (ret)
928 goto out_w_error;
929 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
930 mirror_num, 0);
931 } else if (!async) {
932 ret = btree_csum_one_bio(bio);
933 if (ret)
934 goto out_w_error;
935 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
936 mirror_num, 0);
937 } else {
938 /*
939 * kthread helpers are used to submit writes so that
940 * checksumming can happen in parallel across all CPUs
941 */
942 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
943 inode, rw, bio, mirror_num, 0,
944 bio_offset,
945 __btree_submit_bio_start,
946 __btree_submit_bio_done);
947 }
948
949 if (ret) {
950out_w_error:
951 bio_endio(bio, ret);
952 }
953 return ret;
954}
955
956#ifdef CONFIG_MIGRATION
957static int btree_migratepage(struct address_space *mapping,
958 struct page *newpage, struct page *page,
959 enum migrate_mode mode)
960{
961 /*
962 * we can't safely write a btree page from here,
963 * we haven't done the locking hook
964 */
965 if (PageDirty(page))
966 return -EAGAIN;
967 /*
968 * Buffers may be managed in a filesystem specific way.
969 * We must have no buffers or drop them.
970 */
971 if (page_has_private(page) &&
972 !try_to_release_page(page, GFP_KERNEL))
973 return -EAGAIN;
974 return migrate_page(mapping, newpage, page, mode);
975}
976#endif
977
978
979static int btree_writepages(struct address_space *mapping,
980 struct writeback_control *wbc)
981{
982 struct extent_io_tree *tree;
983 struct btrfs_fs_info *fs_info;
984 int ret;
985
986 tree = &BTRFS_I(mapping->host)->io_tree;
987 if (wbc->sync_mode == WB_SYNC_NONE) {
988
989 if (wbc->for_kupdate)
990 return 0;
991
992 fs_info = BTRFS_I(mapping->host)->root->fs_info;
993 /* this is a bit racy, but that's ok */
994 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
995 BTRFS_DIRTY_METADATA_THRESH);
996 if (ret < 0)
997 return 0;
998 }
999 return btree_write_cache_pages(mapping, wbc);
1000}
1001
1002static int btree_readpage(struct file *file, struct page *page)
1003{
1004 struct extent_io_tree *tree;
1005 tree = &BTRFS_I(page->mapping->host)->io_tree;
1006 return extent_read_full_page(tree, page, btree_get_extent, 0);
1007}
1008
1009static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1010{
1011 if (PageWriteback(page) || PageDirty(page))
1012 return 0;
1013
1014 return try_release_extent_buffer(page);
1015}
1016
1017static void btree_invalidatepage(struct page *page, unsigned int offset,
1018 unsigned int length)
1019{
1020 struct extent_io_tree *tree;
1021 tree = &BTRFS_I(page->mapping->host)->io_tree;
1022 extent_invalidatepage(tree, page, offset);
1023 btree_releasepage(page, GFP_NOFS);
1024 if (PagePrivate(page)) {
1025 printk(KERN_WARNING "btrfs warning page private not zero "
1026 "on page %llu\n", (unsigned long long)page_offset(page));
1027 ClearPagePrivate(page);
1028 set_page_private(page, 0);
1029 page_cache_release(page);
1030 }
1031}
1032
1033static int btree_set_page_dirty(struct page *page)
1034{
1035#ifdef DEBUG
1036 struct extent_buffer *eb;
1037
1038 BUG_ON(!PagePrivate(page));
1039 eb = (struct extent_buffer *)page->private;
1040 BUG_ON(!eb);
1041 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1042 BUG_ON(!atomic_read(&eb->refs));
1043 btrfs_assert_tree_locked(eb);
1044#endif
1045 return __set_page_dirty_nobuffers(page);
1046}
1047
1048static const struct address_space_operations btree_aops = {
1049 .readpage = btree_readpage,
1050 .writepages = btree_writepages,
1051 .releasepage = btree_releasepage,
1052 .invalidatepage = btree_invalidatepage,
1053#ifdef CONFIG_MIGRATION
1054 .migratepage = btree_migratepage,
1055#endif
1056 .set_page_dirty = btree_set_page_dirty,
1057};
1058
1059int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1060 u64 parent_transid)
1061{
1062 struct extent_buffer *buf = NULL;
1063 struct inode *btree_inode = root->fs_info->btree_inode;
1064 int ret = 0;
1065
1066 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1067 if (!buf)
1068 return 0;
1069 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1070 buf, 0, WAIT_NONE, btree_get_extent, 0);
1071 free_extent_buffer(buf);
1072 return ret;
1073}
1074
1075int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1076 int mirror_num, struct extent_buffer **eb)
1077{
1078 struct extent_buffer *buf = NULL;
1079 struct inode *btree_inode = root->fs_info->btree_inode;
1080 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1081 int ret;
1082
1083 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1084 if (!buf)
1085 return 0;
1086
1087 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1088
1089 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1090 btree_get_extent, mirror_num);
1091 if (ret) {
1092 free_extent_buffer(buf);
1093 return ret;
1094 }
1095
1096 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1097 free_extent_buffer(buf);
1098 return -EIO;
1099 } else if (extent_buffer_uptodate(buf)) {
1100 *eb = buf;
1101 } else {
1102 free_extent_buffer(buf);
1103 }
1104 return 0;
1105}
1106
1107struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1108 u64 bytenr, u32 blocksize)
1109{
1110 struct inode *btree_inode = root->fs_info->btree_inode;
1111 struct extent_buffer *eb;
1112 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1113 bytenr, blocksize);
1114 return eb;
1115}
1116
1117struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1118 u64 bytenr, u32 blocksize)
1119{
1120 struct inode *btree_inode = root->fs_info->btree_inode;
1121 struct extent_buffer *eb;
1122
1123 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1124 bytenr, blocksize);
1125 return eb;
1126}
1127
1128
1129int btrfs_write_tree_block(struct extent_buffer *buf)
1130{
1131 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1132 buf->start + buf->len - 1);
1133}
1134
1135int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1136{
1137 return filemap_fdatawait_range(buf->pages[0]->mapping,
1138 buf->start, buf->start + buf->len - 1);
1139}
1140
1141struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1142 u32 blocksize, u64 parent_transid)
1143{
1144 struct extent_buffer *buf = NULL;
1145 int ret;
1146
1147 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1148 if (!buf)
1149 return NULL;
1150
1151 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1152 return buf;
1153
1154}
1155
1156void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1157 struct extent_buffer *buf)
1158{
1159 struct btrfs_fs_info *fs_info = root->fs_info;
1160
1161 if (btrfs_header_generation(buf) ==
1162 fs_info->running_transaction->transid) {
1163 btrfs_assert_tree_locked(buf);
1164
1165 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1166 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1167 -buf->len,
1168 fs_info->dirty_metadata_batch);
1169 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1170 btrfs_set_lock_blocking(buf);
1171 clear_extent_buffer_dirty(buf);
1172 }
1173 }
1174}
1175
1176static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1177 u32 stripesize, struct btrfs_root *root,
1178 struct btrfs_fs_info *fs_info,
1179 u64 objectid)
1180{
1181 root->node = NULL;
1182 root->commit_root = NULL;
1183 root->sectorsize = sectorsize;
1184 root->nodesize = nodesize;
1185 root->leafsize = leafsize;
1186 root->stripesize = stripesize;
1187 root->ref_cows = 0;
1188 root->track_dirty = 0;
1189 root->in_radix = 0;
1190 root->orphan_item_inserted = 0;
1191 root->orphan_cleanup_state = 0;
1192
1193 root->objectid = objectid;
1194 root->last_trans = 0;
1195 root->highest_objectid = 0;
1196 root->nr_delalloc_inodes = 0;
1197 root->nr_ordered_extents = 0;
1198 root->name = NULL;
1199 root->inode_tree = RB_ROOT;
1200 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1201 root->block_rsv = NULL;
1202 root->orphan_block_rsv = NULL;
1203
1204 INIT_LIST_HEAD(&root->dirty_list);
1205 INIT_LIST_HEAD(&root->root_list);
1206 INIT_LIST_HEAD(&root->delalloc_inodes);
1207 INIT_LIST_HEAD(&root->delalloc_root);
1208 INIT_LIST_HEAD(&root->ordered_extents);
1209 INIT_LIST_HEAD(&root->ordered_root);
1210 INIT_LIST_HEAD(&root->logged_list[0]);
1211 INIT_LIST_HEAD(&root->logged_list[1]);
1212 spin_lock_init(&root->orphan_lock);
1213 spin_lock_init(&root->inode_lock);
1214 spin_lock_init(&root->delalloc_lock);
1215 spin_lock_init(&root->ordered_extent_lock);
1216 spin_lock_init(&root->accounting_lock);
1217 spin_lock_init(&root->log_extents_lock[0]);
1218 spin_lock_init(&root->log_extents_lock[1]);
1219 mutex_init(&root->objectid_mutex);
1220 mutex_init(&root->log_mutex);
1221 init_waitqueue_head(&root->log_writer_wait);
1222 init_waitqueue_head(&root->log_commit_wait[0]);
1223 init_waitqueue_head(&root->log_commit_wait[1]);
1224 atomic_set(&root->log_commit[0], 0);
1225 atomic_set(&root->log_commit[1], 0);
1226 atomic_set(&root->log_writers, 0);
1227 atomic_set(&root->log_batch, 0);
1228 atomic_set(&root->orphan_inodes, 0);
1229 atomic_set(&root->refs, 1);
1230 root->log_transid = 0;
1231 root->last_log_commit = 0;
1232 extent_io_tree_init(&root->dirty_log_pages,
1233 fs_info->btree_inode->i_mapping);
1234
1235 memset(&root->root_key, 0, sizeof(root->root_key));
1236 memset(&root->root_item, 0, sizeof(root->root_item));
1237 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1238 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1239 root->defrag_trans_start = fs_info->generation;
1240 init_completion(&root->kobj_unregister);
1241 root->defrag_running = 0;
1242 root->root_key.objectid = objectid;
1243 root->anon_dev = 0;
1244
1245 spin_lock_init(&root->root_item_lock);
1246}
1247
1248static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1249{
1250 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1251 if (root)
1252 root->fs_info = fs_info;
1253 return root;
1254}
1255
1256struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1257 struct btrfs_fs_info *fs_info,
1258 u64 objectid)
1259{
1260 struct extent_buffer *leaf;
1261 struct btrfs_root *tree_root = fs_info->tree_root;
1262 struct btrfs_root *root;
1263 struct btrfs_key key;
1264 int ret = 0;
1265 u64 bytenr;
1266 uuid_le uuid;
1267
1268 root = btrfs_alloc_root(fs_info);
1269 if (!root)
1270 return ERR_PTR(-ENOMEM);
1271
1272 __setup_root(tree_root->nodesize, tree_root->leafsize,
1273 tree_root->sectorsize, tree_root->stripesize,
1274 root, fs_info, objectid);
1275 root->root_key.objectid = objectid;
1276 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1277 root->root_key.offset = 0;
1278
1279 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1280 0, objectid, NULL, 0, 0, 0);
1281 if (IS_ERR(leaf)) {
1282 ret = PTR_ERR(leaf);
1283 leaf = NULL;
1284 goto fail;
1285 }
1286
1287 bytenr = leaf->start;
1288 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1289 btrfs_set_header_bytenr(leaf, leaf->start);
1290 btrfs_set_header_generation(leaf, trans->transid);
1291 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1292 btrfs_set_header_owner(leaf, objectid);
1293 root->node = leaf;
1294
1295 write_extent_buffer(leaf, fs_info->fsid,
1296 (unsigned long)btrfs_header_fsid(leaf),
1297 BTRFS_FSID_SIZE);
1298 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1299 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1300 BTRFS_UUID_SIZE);
1301 btrfs_mark_buffer_dirty(leaf);
1302
1303 root->commit_root = btrfs_root_node(root);
1304 root->track_dirty = 1;
1305
1306
1307 root->root_item.flags = 0;
1308 root->root_item.byte_limit = 0;
1309 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1310 btrfs_set_root_generation(&root->root_item, trans->transid);
1311 btrfs_set_root_level(&root->root_item, 0);
1312 btrfs_set_root_refs(&root->root_item, 1);
1313 btrfs_set_root_used(&root->root_item, leaf->len);
1314 btrfs_set_root_last_snapshot(&root->root_item, 0);
1315 btrfs_set_root_dirid(&root->root_item, 0);
1316 uuid_le_gen(&uuid);
1317 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1318 root->root_item.drop_level = 0;
1319
1320 key.objectid = objectid;
1321 key.type = BTRFS_ROOT_ITEM_KEY;
1322 key.offset = 0;
1323 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1324 if (ret)
1325 goto fail;
1326
1327 btrfs_tree_unlock(leaf);
1328
1329 return root;
1330
1331fail:
1332 if (leaf) {
1333 btrfs_tree_unlock(leaf);
1334 free_extent_buffer(leaf);
1335 }
1336 kfree(root);
1337
1338 return ERR_PTR(ret);
1339}
1340
1341static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1342 struct btrfs_fs_info *fs_info)
1343{
1344 struct btrfs_root *root;
1345 struct btrfs_root *tree_root = fs_info->tree_root;
1346 struct extent_buffer *leaf;
1347
1348 root = btrfs_alloc_root(fs_info);
1349 if (!root)
1350 return ERR_PTR(-ENOMEM);
1351
1352 __setup_root(tree_root->nodesize, tree_root->leafsize,
1353 tree_root->sectorsize, tree_root->stripesize,
1354 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1355
1356 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1357 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1358 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1359 /*
1360 * log trees do not get reference counted because they go away
1361 * before a real commit is actually done. They do store pointers
1362 * to file data extents, and those reference counts still get
1363 * updated (along with back refs to the log tree).
1364 */
1365 root->ref_cows = 0;
1366
1367 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1368 BTRFS_TREE_LOG_OBJECTID, NULL,
1369 0, 0, 0);
1370 if (IS_ERR(leaf)) {
1371 kfree(root);
1372 return ERR_CAST(leaf);
1373 }
1374
1375 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1376 btrfs_set_header_bytenr(leaf, leaf->start);
1377 btrfs_set_header_generation(leaf, trans->transid);
1378 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1379 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1380 root->node = leaf;
1381
1382 write_extent_buffer(root->node, root->fs_info->fsid,
1383 (unsigned long)btrfs_header_fsid(root->node),
1384 BTRFS_FSID_SIZE);
1385 btrfs_mark_buffer_dirty(root->node);
1386 btrfs_tree_unlock(root->node);
1387 return root;
1388}
1389
1390int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1391 struct btrfs_fs_info *fs_info)
1392{
1393 struct btrfs_root *log_root;
1394
1395 log_root = alloc_log_tree(trans, fs_info);
1396 if (IS_ERR(log_root))
1397 return PTR_ERR(log_root);
1398 WARN_ON(fs_info->log_root_tree);
1399 fs_info->log_root_tree = log_root;
1400 return 0;
1401}
1402
1403int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1404 struct btrfs_root *root)
1405{
1406 struct btrfs_root *log_root;
1407 struct btrfs_inode_item *inode_item;
1408
1409 log_root = alloc_log_tree(trans, root->fs_info);
1410 if (IS_ERR(log_root))
1411 return PTR_ERR(log_root);
1412
1413 log_root->last_trans = trans->transid;
1414 log_root->root_key.offset = root->root_key.objectid;
1415
1416 inode_item = &log_root->root_item.inode;
1417 btrfs_set_stack_inode_generation(inode_item, 1);
1418 btrfs_set_stack_inode_size(inode_item, 3);
1419 btrfs_set_stack_inode_nlink(inode_item, 1);
1420 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1421 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1422
1423 btrfs_set_root_node(&log_root->root_item, log_root->node);
1424
1425 WARN_ON(root->log_root);
1426 root->log_root = log_root;
1427 root->log_transid = 0;
1428 root->last_log_commit = 0;
1429 return 0;
1430}
1431
1432struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1433 struct btrfs_key *key)
1434{
1435 struct btrfs_root *root;
1436 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1437 struct btrfs_path *path;
1438 u64 generation;
1439 u32 blocksize;
1440 int ret;
1441
1442 path = btrfs_alloc_path();
1443 if (!path)
1444 return ERR_PTR(-ENOMEM);
1445
1446 root = btrfs_alloc_root(fs_info);
1447 if (!root) {
1448 ret = -ENOMEM;
1449 goto alloc_fail;
1450 }
1451
1452 __setup_root(tree_root->nodesize, tree_root->leafsize,
1453 tree_root->sectorsize, tree_root->stripesize,
1454 root, fs_info, key->objectid);
1455
1456 ret = btrfs_find_root(tree_root, key, path,
1457 &root->root_item, &root->root_key);
1458 if (ret) {
1459 if (ret > 0)
1460 ret = -ENOENT;
1461 goto find_fail;
1462 }
1463
1464 generation = btrfs_root_generation(&root->root_item);
1465 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1466 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1467 blocksize, generation);
1468 if (!root->node) {
1469 ret = -ENOMEM;
1470 goto find_fail;
1471 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1472 ret = -EIO;
1473 goto read_fail;
1474 }
1475 root->commit_root = btrfs_root_node(root);
1476out:
1477 btrfs_free_path(path);
1478 return root;
1479
1480read_fail:
1481 free_extent_buffer(root->node);
1482find_fail:
1483 kfree(root);
1484alloc_fail:
1485 root = ERR_PTR(ret);
1486 goto out;
1487}
1488
1489struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1490 struct btrfs_key *location)
1491{
1492 struct btrfs_root *root;
1493
1494 root = btrfs_read_tree_root(tree_root, location);
1495 if (IS_ERR(root))
1496 return root;
1497
1498 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1499 root->ref_cows = 1;
1500 btrfs_check_and_init_root_item(&root->root_item);
1501 }
1502
1503 return root;
1504}
1505
1506int btrfs_init_fs_root(struct btrfs_root *root)
1507{
1508 int ret;
1509
1510 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1511 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1512 GFP_NOFS);
1513 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1514 ret = -ENOMEM;
1515 goto fail;
1516 }
1517
1518 btrfs_init_free_ino_ctl(root);
1519 mutex_init(&root->fs_commit_mutex);
1520 spin_lock_init(&root->cache_lock);
1521 init_waitqueue_head(&root->cache_wait);
1522
1523 ret = get_anon_bdev(&root->anon_dev);
1524 if (ret)
1525 goto fail;
1526 return 0;
1527fail:
1528 kfree(root->free_ino_ctl);
1529 kfree(root->free_ino_pinned);
1530 return ret;
1531}
1532
1533struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1534 u64 root_id)
1535{
1536 struct btrfs_root *root;
1537
1538 spin_lock(&fs_info->fs_roots_radix_lock);
1539 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1540 (unsigned long)root_id);
1541 spin_unlock(&fs_info->fs_roots_radix_lock);
1542 return root;
1543}
1544
1545int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1546 struct btrfs_root *root)
1547{
1548 int ret;
1549
1550 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1551 if (ret)
1552 return ret;
1553
1554 spin_lock(&fs_info->fs_roots_radix_lock);
1555 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1556 (unsigned long)root->root_key.objectid,
1557 root);
1558 if (ret == 0)
1559 root->in_radix = 1;
1560 spin_unlock(&fs_info->fs_roots_radix_lock);
1561 radix_tree_preload_end();
1562
1563 return ret;
1564}
1565
1566struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1567 struct btrfs_key *location)
1568{
1569 struct btrfs_root *root;
1570 int ret;
1571
1572 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1573 return fs_info->tree_root;
1574 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1575 return fs_info->extent_root;
1576 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1577 return fs_info->chunk_root;
1578 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1579 return fs_info->dev_root;
1580 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1581 return fs_info->csum_root;
1582 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1583 return fs_info->quota_root ? fs_info->quota_root :
1584 ERR_PTR(-ENOENT);
1585again:
1586 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1587 if (root)
1588 return root;
1589
1590 root = btrfs_read_fs_root(fs_info->tree_root, location);
1591 if (IS_ERR(root))
1592 return root;
1593
1594 if (btrfs_root_refs(&root->root_item) == 0) {
1595 ret = -ENOENT;
1596 goto fail;
1597 }
1598
1599 ret = btrfs_init_fs_root(root);
1600 if (ret)
1601 goto fail;
1602
1603 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1604 if (ret < 0)
1605 goto fail;
1606 if (ret == 0)
1607 root->orphan_item_inserted = 1;
1608
1609 ret = btrfs_insert_fs_root(fs_info, root);
1610 if (ret) {
1611 if (ret == -EEXIST) {
1612 free_fs_root(root);
1613 goto again;
1614 }
1615 goto fail;
1616 }
1617 return root;
1618fail:
1619 free_fs_root(root);
1620 return ERR_PTR(ret);
1621}
1622
1623static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1624{
1625 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1626 int ret = 0;
1627 struct btrfs_device *device;
1628 struct backing_dev_info *bdi;
1629
1630 rcu_read_lock();
1631 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1632 if (!device->bdev)
1633 continue;
1634 bdi = blk_get_backing_dev_info(device->bdev);
1635 if (bdi && bdi_congested(bdi, bdi_bits)) {
1636 ret = 1;
1637 break;
1638 }
1639 }
1640 rcu_read_unlock();
1641 return ret;
1642}
1643
1644/*
1645 * If this fails, caller must call bdi_destroy() to get rid of the
1646 * bdi again.
1647 */
1648static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1649{
1650 int err;
1651
1652 bdi->capabilities = BDI_CAP_MAP_COPY;
1653 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1654 if (err)
1655 return err;
1656
1657 bdi->ra_pages = default_backing_dev_info.ra_pages;
1658 bdi->congested_fn = btrfs_congested_fn;
1659 bdi->congested_data = info;
1660 return 0;
1661}
1662
1663/*
1664 * called by the kthread helper functions to finally call the bio end_io
1665 * functions. This is where read checksum verification actually happens
1666 */
1667static void end_workqueue_fn(struct btrfs_work *work)
1668{
1669 struct bio *bio;
1670 struct end_io_wq *end_io_wq;
1671 struct btrfs_fs_info *fs_info;
1672 int error;
1673
1674 end_io_wq = container_of(work, struct end_io_wq, work);
1675 bio = end_io_wq->bio;
1676 fs_info = end_io_wq->info;
1677
1678 error = end_io_wq->error;
1679 bio->bi_private = end_io_wq->private;
1680 bio->bi_end_io = end_io_wq->end_io;
1681 kfree(end_io_wq);
1682 bio_endio(bio, error);
1683}
1684
1685static int cleaner_kthread(void *arg)
1686{
1687 struct btrfs_root *root = arg;
1688 int again;
1689
1690 do {
1691 again = 0;
1692
1693 /* Make the cleaner go to sleep early. */
1694 if (btrfs_need_cleaner_sleep(root))
1695 goto sleep;
1696
1697 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1698 goto sleep;
1699
1700 /*
1701 * Avoid the problem that we change the status of the fs
1702 * during the above check and trylock.
1703 */
1704 if (btrfs_need_cleaner_sleep(root)) {
1705 mutex_unlock(&root->fs_info->cleaner_mutex);
1706 goto sleep;
1707 }
1708
1709 btrfs_run_delayed_iputs(root);
1710 again = btrfs_clean_one_deleted_snapshot(root);
1711 mutex_unlock(&root->fs_info->cleaner_mutex);
1712
1713 /*
1714 * The defragger has dealt with the R/O remount and umount,
1715 * needn't do anything special here.
1716 */
1717 btrfs_run_defrag_inodes(root->fs_info);
1718sleep:
1719 if (!try_to_freeze() && !again) {
1720 set_current_state(TASK_INTERRUPTIBLE);
1721 if (!kthread_should_stop())
1722 schedule();
1723 __set_current_state(TASK_RUNNING);
1724 }
1725 } while (!kthread_should_stop());
1726 return 0;
1727}
1728
1729static int transaction_kthread(void *arg)
1730{
1731 struct btrfs_root *root = arg;
1732 struct btrfs_trans_handle *trans;
1733 struct btrfs_transaction *cur;
1734 u64 transid;
1735 unsigned long now;
1736 unsigned long delay;
1737 bool cannot_commit;
1738
1739 do {
1740 cannot_commit = false;
1741 delay = HZ * 30;
1742 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1743
1744 spin_lock(&root->fs_info->trans_lock);
1745 cur = root->fs_info->running_transaction;
1746 if (!cur) {
1747 spin_unlock(&root->fs_info->trans_lock);
1748 goto sleep;
1749 }
1750
1751 now = get_seconds();
1752 if (cur->state < TRANS_STATE_BLOCKED &&
1753 (now < cur->start_time || now - cur->start_time < 30)) {
1754 spin_unlock(&root->fs_info->trans_lock);
1755 delay = HZ * 5;
1756 goto sleep;
1757 }
1758 transid = cur->transid;
1759 spin_unlock(&root->fs_info->trans_lock);
1760
1761 /* If the file system is aborted, this will always fail. */
1762 trans = btrfs_attach_transaction(root);
1763 if (IS_ERR(trans)) {
1764 if (PTR_ERR(trans) != -ENOENT)
1765 cannot_commit = true;
1766 goto sleep;
1767 }
1768 if (transid == trans->transid) {
1769 btrfs_commit_transaction(trans, root);
1770 } else {
1771 btrfs_end_transaction(trans, root);
1772 }
1773sleep:
1774 wake_up_process(root->fs_info->cleaner_kthread);
1775 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1776
1777 if (!try_to_freeze()) {
1778 set_current_state(TASK_INTERRUPTIBLE);
1779 if (!kthread_should_stop() &&
1780 (!btrfs_transaction_blocked(root->fs_info) ||
1781 cannot_commit))
1782 schedule_timeout(delay);
1783 __set_current_state(TASK_RUNNING);
1784 }
1785 } while (!kthread_should_stop());
1786 return 0;
1787}
1788
1789/*
1790 * this will find the highest generation in the array of
1791 * root backups. The index of the highest array is returned,
1792 * or -1 if we can't find anything.
1793 *
1794 * We check to make sure the array is valid by comparing the
1795 * generation of the latest root in the array with the generation
1796 * in the super block. If they don't match we pitch it.
1797 */
1798static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1799{
1800 u64 cur;
1801 int newest_index = -1;
1802 struct btrfs_root_backup *root_backup;
1803 int i;
1804
1805 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1806 root_backup = info->super_copy->super_roots + i;
1807 cur = btrfs_backup_tree_root_gen(root_backup);
1808 if (cur == newest_gen)
1809 newest_index = i;
1810 }
1811
1812 /* check to see if we actually wrapped around */
1813 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1814 root_backup = info->super_copy->super_roots;
1815 cur = btrfs_backup_tree_root_gen(root_backup);
1816 if (cur == newest_gen)
1817 newest_index = 0;
1818 }
1819 return newest_index;
1820}
1821
1822
1823/*
1824 * find the oldest backup so we know where to store new entries
1825 * in the backup array. This will set the backup_root_index
1826 * field in the fs_info struct
1827 */
1828static void find_oldest_super_backup(struct btrfs_fs_info *info,
1829 u64 newest_gen)
1830{
1831 int newest_index = -1;
1832
1833 newest_index = find_newest_super_backup(info, newest_gen);
1834 /* if there was garbage in there, just move along */
1835 if (newest_index == -1) {
1836 info->backup_root_index = 0;
1837 } else {
1838 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1839 }
1840}
1841
1842/*
1843 * copy all the root pointers into the super backup array.
1844 * this will bump the backup pointer by one when it is
1845 * done
1846 */
1847static void backup_super_roots(struct btrfs_fs_info *info)
1848{
1849 int next_backup;
1850 struct btrfs_root_backup *root_backup;
1851 int last_backup;
1852
1853 next_backup = info->backup_root_index;
1854 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1855 BTRFS_NUM_BACKUP_ROOTS;
1856
1857 /*
1858 * just overwrite the last backup if we're at the same generation
1859 * this happens only at umount
1860 */
1861 root_backup = info->super_for_commit->super_roots + last_backup;
1862 if (btrfs_backup_tree_root_gen(root_backup) ==
1863 btrfs_header_generation(info->tree_root->node))
1864 next_backup = last_backup;
1865
1866 root_backup = info->super_for_commit->super_roots + next_backup;
1867
1868 /*
1869 * make sure all of our padding and empty slots get zero filled
1870 * regardless of which ones we use today
1871 */
1872 memset(root_backup, 0, sizeof(*root_backup));
1873
1874 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1875
1876 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1877 btrfs_set_backup_tree_root_gen(root_backup,
1878 btrfs_header_generation(info->tree_root->node));
1879
1880 btrfs_set_backup_tree_root_level(root_backup,
1881 btrfs_header_level(info->tree_root->node));
1882
1883 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1884 btrfs_set_backup_chunk_root_gen(root_backup,
1885 btrfs_header_generation(info->chunk_root->node));
1886 btrfs_set_backup_chunk_root_level(root_backup,
1887 btrfs_header_level(info->chunk_root->node));
1888
1889 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1890 btrfs_set_backup_extent_root_gen(root_backup,
1891 btrfs_header_generation(info->extent_root->node));
1892 btrfs_set_backup_extent_root_level(root_backup,
1893 btrfs_header_level(info->extent_root->node));
1894
1895 /*
1896 * we might commit during log recovery, which happens before we set
1897 * the fs_root. Make sure it is valid before we fill it in.
1898 */
1899 if (info->fs_root && info->fs_root->node) {
1900 btrfs_set_backup_fs_root(root_backup,
1901 info->fs_root->node->start);
1902 btrfs_set_backup_fs_root_gen(root_backup,
1903 btrfs_header_generation(info->fs_root->node));
1904 btrfs_set_backup_fs_root_level(root_backup,
1905 btrfs_header_level(info->fs_root->node));
1906 }
1907
1908 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1909 btrfs_set_backup_dev_root_gen(root_backup,
1910 btrfs_header_generation(info->dev_root->node));
1911 btrfs_set_backup_dev_root_level(root_backup,
1912 btrfs_header_level(info->dev_root->node));
1913
1914 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1915 btrfs_set_backup_csum_root_gen(root_backup,
1916 btrfs_header_generation(info->csum_root->node));
1917 btrfs_set_backup_csum_root_level(root_backup,
1918 btrfs_header_level(info->csum_root->node));
1919
1920 btrfs_set_backup_total_bytes(root_backup,
1921 btrfs_super_total_bytes(info->super_copy));
1922 btrfs_set_backup_bytes_used(root_backup,
1923 btrfs_super_bytes_used(info->super_copy));
1924 btrfs_set_backup_num_devices(root_backup,
1925 btrfs_super_num_devices(info->super_copy));
1926
1927 /*
1928 * if we don't copy this out to the super_copy, it won't get remembered
1929 * for the next commit
1930 */
1931 memcpy(&info->super_copy->super_roots,
1932 &info->super_for_commit->super_roots,
1933 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1934}
1935
1936/*
1937 * this copies info out of the root backup array and back into
1938 * the in-memory super block. It is meant to help iterate through
1939 * the array, so you send it the number of backups you've already
1940 * tried and the last backup index you used.
1941 *
1942 * this returns -1 when it has tried all the backups
1943 */
1944static noinline int next_root_backup(struct btrfs_fs_info *info,
1945 struct btrfs_super_block *super,
1946 int *num_backups_tried, int *backup_index)
1947{
1948 struct btrfs_root_backup *root_backup;
1949 int newest = *backup_index;
1950
1951 if (*num_backups_tried == 0) {
1952 u64 gen = btrfs_super_generation(super);
1953
1954 newest = find_newest_super_backup(info, gen);
1955 if (newest == -1)
1956 return -1;
1957
1958 *backup_index = newest;
1959 *num_backups_tried = 1;
1960 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1961 /* we've tried all the backups, all done */
1962 return -1;
1963 } else {
1964 /* jump to the next oldest backup */
1965 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1966 BTRFS_NUM_BACKUP_ROOTS;
1967 *backup_index = newest;
1968 *num_backups_tried += 1;
1969 }
1970 root_backup = super->super_roots + newest;
1971
1972 btrfs_set_super_generation(super,
1973 btrfs_backup_tree_root_gen(root_backup));
1974 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1975 btrfs_set_super_root_level(super,
1976 btrfs_backup_tree_root_level(root_backup));
1977 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1978
1979 /*
1980 * fixme: the total bytes and num_devices need to match or we should
1981 * need a fsck
1982 */
1983 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1984 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1985 return 0;
1986}
1987
1988/* helper to cleanup workers */
1989static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1990{
1991 btrfs_stop_workers(&fs_info->generic_worker);
1992 btrfs_stop_workers(&fs_info->fixup_workers);
1993 btrfs_stop_workers(&fs_info->delalloc_workers);
1994 btrfs_stop_workers(&fs_info->workers);
1995 btrfs_stop_workers(&fs_info->endio_workers);
1996 btrfs_stop_workers(&fs_info->endio_meta_workers);
1997 btrfs_stop_workers(&fs_info->endio_raid56_workers);
1998 btrfs_stop_workers(&fs_info->rmw_workers);
1999 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2000 btrfs_stop_workers(&fs_info->endio_write_workers);
2001 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2002 btrfs_stop_workers(&fs_info->submit_workers);
2003 btrfs_stop_workers(&fs_info->delayed_workers);
2004 btrfs_stop_workers(&fs_info->caching_workers);
2005 btrfs_stop_workers(&fs_info->readahead_workers);
2006 btrfs_stop_workers(&fs_info->flush_workers);
2007 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2008}
2009
2010/* helper to cleanup tree roots */
2011static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2012{
2013 free_extent_buffer(info->tree_root->node);
2014 free_extent_buffer(info->tree_root->commit_root);
2015 info->tree_root->node = NULL;
2016 info->tree_root->commit_root = NULL;
2017
2018 if (info->dev_root) {
2019 free_extent_buffer(info->dev_root->node);
2020 free_extent_buffer(info->dev_root->commit_root);
2021 info->dev_root->node = NULL;
2022 info->dev_root->commit_root = NULL;
2023 }
2024 if (info->extent_root) {
2025 free_extent_buffer(info->extent_root->node);
2026 free_extent_buffer(info->extent_root->commit_root);
2027 info->extent_root->node = NULL;
2028 info->extent_root->commit_root = NULL;
2029 }
2030 if (info->csum_root) {
2031 free_extent_buffer(info->csum_root->node);
2032 free_extent_buffer(info->csum_root->commit_root);
2033 info->csum_root->node = NULL;
2034 info->csum_root->commit_root = NULL;
2035 }
2036 if (info->quota_root) {
2037 free_extent_buffer(info->quota_root->node);
2038 free_extent_buffer(info->quota_root->commit_root);
2039 info->quota_root->node = NULL;
2040 info->quota_root->commit_root = NULL;
2041 }
2042 if (chunk_root) {
2043 free_extent_buffer(info->chunk_root->node);
2044 free_extent_buffer(info->chunk_root->commit_root);
2045 info->chunk_root->node = NULL;
2046 info->chunk_root->commit_root = NULL;
2047 }
2048}
2049
2050static void del_fs_roots(struct btrfs_fs_info *fs_info)
2051{
2052 int ret;
2053 struct btrfs_root *gang[8];
2054 int i;
2055
2056 while (!list_empty(&fs_info->dead_roots)) {
2057 gang[0] = list_entry(fs_info->dead_roots.next,
2058 struct btrfs_root, root_list);
2059 list_del(&gang[0]->root_list);
2060
2061 if (gang[0]->in_radix) {
2062 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2063 } else {
2064 free_extent_buffer(gang[0]->node);
2065 free_extent_buffer(gang[0]->commit_root);
2066 btrfs_put_fs_root(gang[0]);
2067 }
2068 }
2069
2070 while (1) {
2071 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2072 (void **)gang, 0,
2073 ARRAY_SIZE(gang));
2074 if (!ret)
2075 break;
2076 for (i = 0; i < ret; i++)
2077 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2078 }
2079}
2080
2081int open_ctree(struct super_block *sb,
2082 struct btrfs_fs_devices *fs_devices,
2083 char *options)
2084{
2085 u32 sectorsize;
2086 u32 nodesize;
2087 u32 leafsize;
2088 u32 blocksize;
2089 u32 stripesize;
2090 u64 generation;
2091 u64 features;
2092 struct btrfs_key location;
2093 struct buffer_head *bh;
2094 struct btrfs_super_block *disk_super;
2095 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2096 struct btrfs_root *tree_root;
2097 struct btrfs_root *extent_root;
2098 struct btrfs_root *csum_root;
2099 struct btrfs_root *chunk_root;
2100 struct btrfs_root *dev_root;
2101 struct btrfs_root *quota_root;
2102 struct btrfs_root *log_tree_root;
2103 int ret;
2104 int err = -EINVAL;
2105 int num_backups_tried = 0;
2106 int backup_index = 0;
2107
2108 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2109 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2110 if (!tree_root || !chunk_root) {
2111 err = -ENOMEM;
2112 goto fail;
2113 }
2114
2115 ret = init_srcu_struct(&fs_info->subvol_srcu);
2116 if (ret) {
2117 err = ret;
2118 goto fail;
2119 }
2120
2121 ret = setup_bdi(fs_info, &fs_info->bdi);
2122 if (ret) {
2123 err = ret;
2124 goto fail_srcu;
2125 }
2126
2127 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2128 if (ret) {
2129 err = ret;
2130 goto fail_bdi;
2131 }
2132 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2133 (1 + ilog2(nr_cpu_ids));
2134
2135 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2136 if (ret) {
2137 err = ret;
2138 goto fail_dirty_metadata_bytes;
2139 }
2140
2141 fs_info->btree_inode = new_inode(sb);
2142 if (!fs_info->btree_inode) {
2143 err = -ENOMEM;
2144 goto fail_delalloc_bytes;
2145 }
2146
2147 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2148
2149 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2150 INIT_LIST_HEAD(&fs_info->trans_list);
2151 INIT_LIST_HEAD(&fs_info->dead_roots);
2152 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2153 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2154 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2155 spin_lock_init(&fs_info->delalloc_root_lock);
2156 spin_lock_init(&fs_info->trans_lock);
2157 spin_lock_init(&fs_info->fs_roots_radix_lock);
2158 spin_lock_init(&fs_info->delayed_iput_lock);
2159 spin_lock_init(&fs_info->defrag_inodes_lock);
2160 spin_lock_init(&fs_info->free_chunk_lock);
2161 spin_lock_init(&fs_info->tree_mod_seq_lock);
2162 spin_lock_init(&fs_info->super_lock);
2163 rwlock_init(&fs_info->tree_mod_log_lock);
2164 mutex_init(&fs_info->reloc_mutex);
2165 seqlock_init(&fs_info->profiles_lock);
2166
2167 init_completion(&fs_info->kobj_unregister);
2168 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2169 INIT_LIST_HEAD(&fs_info->space_info);
2170 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2171 btrfs_mapping_init(&fs_info->mapping_tree);
2172 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2173 BTRFS_BLOCK_RSV_GLOBAL);
2174 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2175 BTRFS_BLOCK_RSV_DELALLOC);
2176 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2177 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2178 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2179 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2180 BTRFS_BLOCK_RSV_DELOPS);
2181 atomic_set(&fs_info->nr_async_submits, 0);
2182 atomic_set(&fs_info->async_delalloc_pages, 0);
2183 atomic_set(&fs_info->async_submit_draining, 0);
2184 atomic_set(&fs_info->nr_async_bios, 0);
2185 atomic_set(&fs_info->defrag_running, 0);
2186 atomic64_set(&fs_info->tree_mod_seq, 0);
2187 fs_info->sb = sb;
2188 fs_info->max_inline = 8192 * 1024;
2189 fs_info->metadata_ratio = 0;
2190 fs_info->defrag_inodes = RB_ROOT;
2191 fs_info->free_chunk_space = 0;
2192 fs_info->tree_mod_log = RB_ROOT;
2193
2194 /* readahead state */
2195 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2196 spin_lock_init(&fs_info->reada_lock);
2197
2198 fs_info->thread_pool_size = min_t(unsigned long,
2199 num_online_cpus() + 2, 8);
2200
2201 INIT_LIST_HEAD(&fs_info->ordered_roots);
2202 spin_lock_init(&fs_info->ordered_root_lock);
2203 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2204 GFP_NOFS);
2205 if (!fs_info->delayed_root) {
2206 err = -ENOMEM;
2207 goto fail_iput;
2208 }
2209 btrfs_init_delayed_root(fs_info->delayed_root);
2210
2211 mutex_init(&fs_info->scrub_lock);
2212 atomic_set(&fs_info->scrubs_running, 0);
2213 atomic_set(&fs_info->scrub_pause_req, 0);
2214 atomic_set(&fs_info->scrubs_paused, 0);
2215 atomic_set(&fs_info->scrub_cancel_req, 0);
2216 init_waitqueue_head(&fs_info->scrub_pause_wait);
2217 init_rwsem(&fs_info->scrub_super_lock);
2218 fs_info->scrub_workers_refcnt = 0;
2219#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2220 fs_info->check_integrity_print_mask = 0;
2221#endif
2222
2223 spin_lock_init(&fs_info->balance_lock);
2224 mutex_init(&fs_info->balance_mutex);
2225 atomic_set(&fs_info->balance_running, 0);
2226 atomic_set(&fs_info->balance_pause_req, 0);
2227 atomic_set(&fs_info->balance_cancel_req, 0);
2228 fs_info->balance_ctl = NULL;
2229 init_waitqueue_head(&fs_info->balance_wait_q);
2230
2231 sb->s_blocksize = 4096;
2232 sb->s_blocksize_bits = blksize_bits(4096);
2233 sb->s_bdi = &fs_info->bdi;
2234
2235 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2236 set_nlink(fs_info->btree_inode, 1);
2237 /*
2238 * we set the i_size on the btree inode to the max possible int.
2239 * the real end of the address space is determined by all of
2240 * the devices in the system
2241 */
2242 fs_info->btree_inode->i_size = OFFSET_MAX;
2243 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2244 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2245
2246 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2247 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2248 fs_info->btree_inode->i_mapping);
2249 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2250 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2251
2252 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2253
2254 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2255 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2256 sizeof(struct btrfs_key));
2257 set_bit(BTRFS_INODE_DUMMY,
2258 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2259 insert_inode_hash(fs_info->btree_inode);
2260
2261 spin_lock_init(&fs_info->block_group_cache_lock);
2262 fs_info->block_group_cache_tree = RB_ROOT;
2263 fs_info->first_logical_byte = (u64)-1;
2264
2265 extent_io_tree_init(&fs_info->freed_extents[0],
2266 fs_info->btree_inode->i_mapping);
2267 extent_io_tree_init(&fs_info->freed_extents[1],
2268 fs_info->btree_inode->i_mapping);
2269 fs_info->pinned_extents = &fs_info->freed_extents[0];
2270 fs_info->do_barriers = 1;
2271
2272
2273 mutex_init(&fs_info->ordered_operations_mutex);
2274 mutex_init(&fs_info->tree_log_mutex);
2275 mutex_init(&fs_info->chunk_mutex);
2276 mutex_init(&fs_info->transaction_kthread_mutex);
2277 mutex_init(&fs_info->cleaner_mutex);
2278 mutex_init(&fs_info->volume_mutex);
2279 init_rwsem(&fs_info->extent_commit_sem);
2280 init_rwsem(&fs_info->cleanup_work_sem);
2281 init_rwsem(&fs_info->subvol_sem);
2282 fs_info->dev_replace.lock_owner = 0;
2283 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2284 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2285 mutex_init(&fs_info->dev_replace.lock_management_lock);
2286 mutex_init(&fs_info->dev_replace.lock);
2287
2288 spin_lock_init(&fs_info->qgroup_lock);
2289 mutex_init(&fs_info->qgroup_ioctl_lock);
2290 fs_info->qgroup_tree = RB_ROOT;
2291 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2292 fs_info->qgroup_seq = 1;
2293 fs_info->quota_enabled = 0;
2294 fs_info->pending_quota_state = 0;
2295 fs_info->qgroup_ulist = NULL;
2296 mutex_init(&fs_info->qgroup_rescan_lock);
2297
2298 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2299 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2300
2301 init_waitqueue_head(&fs_info->transaction_throttle);
2302 init_waitqueue_head(&fs_info->transaction_wait);
2303 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2304 init_waitqueue_head(&fs_info->async_submit_wait);
2305
2306 ret = btrfs_alloc_stripe_hash_table(fs_info);
2307 if (ret) {
2308 err = ret;
2309 goto fail_alloc;
2310 }
2311
2312 __setup_root(4096, 4096, 4096, 4096, tree_root,
2313 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2314
2315 invalidate_bdev(fs_devices->latest_bdev);
2316
2317 /*
2318 * Read super block and check the signature bytes only
2319 */
2320 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2321 if (!bh) {
2322 err = -EINVAL;
2323 goto fail_alloc;
2324 }
2325
2326 /*
2327 * We want to check superblock checksum, the type is stored inside.
2328 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2329 */
2330 if (btrfs_check_super_csum(bh->b_data)) {
2331 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2332 err = -EINVAL;
2333 goto fail_alloc;
2334 }
2335
2336 /*
2337 * super_copy is zeroed at allocation time and we never touch the
2338 * following bytes up to INFO_SIZE, the checksum is calculated from
2339 * the whole block of INFO_SIZE
2340 */
2341 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2342 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2343 sizeof(*fs_info->super_for_commit));
2344 brelse(bh);
2345
2346 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2347
2348 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2349 if (ret) {
2350 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2351 err = -EINVAL;
2352 goto fail_alloc;
2353 }
2354
2355 disk_super = fs_info->super_copy;
2356 if (!btrfs_super_root(disk_super))
2357 goto fail_alloc;
2358
2359 /* check FS state, whether FS is broken. */
2360 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2361 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2362
2363 /*
2364 * run through our array of backup supers and setup
2365 * our ring pointer to the oldest one
2366 */
2367 generation = btrfs_super_generation(disk_super);
2368 find_oldest_super_backup(fs_info, generation);
2369
2370 /*
2371 * In the long term, we'll store the compression type in the super
2372 * block, and it'll be used for per file compression control.
2373 */
2374 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2375
2376 ret = btrfs_parse_options(tree_root, options);
2377 if (ret) {
2378 err = ret;
2379 goto fail_alloc;
2380 }
2381
2382 features = btrfs_super_incompat_flags(disk_super) &
2383 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2384 if (features) {
2385 printk(KERN_ERR "BTRFS: couldn't mount because of "
2386 "unsupported optional features (%Lx).\n",
2387 (unsigned long long)features);
2388 err = -EINVAL;
2389 goto fail_alloc;
2390 }
2391
2392 if (btrfs_super_leafsize(disk_super) !=
2393 btrfs_super_nodesize(disk_super)) {
2394 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2395 "blocksizes don't match. node %d leaf %d\n",
2396 btrfs_super_nodesize(disk_super),
2397 btrfs_super_leafsize(disk_super));
2398 err = -EINVAL;
2399 goto fail_alloc;
2400 }
2401 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2402 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2403 "blocksize (%d) was too large\n",
2404 btrfs_super_leafsize(disk_super));
2405 err = -EINVAL;
2406 goto fail_alloc;
2407 }
2408
2409 features = btrfs_super_incompat_flags(disk_super);
2410 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2411 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2412 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2413
2414 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2415 printk(KERN_ERR "btrfs: has skinny extents\n");
2416
2417 /*
2418 * flag our filesystem as having big metadata blocks if
2419 * they are bigger than the page size
2420 */
2421 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2422 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2423 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2424 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2425 }
2426
2427 nodesize = btrfs_super_nodesize(disk_super);
2428 leafsize = btrfs_super_leafsize(disk_super);
2429 sectorsize = btrfs_super_sectorsize(disk_super);
2430 stripesize = btrfs_super_stripesize(disk_super);
2431 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2432 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2433
2434 /*
2435 * mixed block groups end up with duplicate but slightly offset
2436 * extent buffers for the same range. It leads to corruptions
2437 */
2438 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2439 (sectorsize != leafsize)) {
2440 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2441 "are not allowed for mixed block groups on %s\n",
2442 sb->s_id);
2443 goto fail_alloc;
2444 }
2445
2446 /*
2447 * Needn't use the lock because there is no other task which will
2448 * update the flag.
2449 */
2450 btrfs_set_super_incompat_flags(disk_super, features);
2451
2452 features = btrfs_super_compat_ro_flags(disk_super) &
2453 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2454 if (!(sb->s_flags & MS_RDONLY) && features) {
2455 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2456 "unsupported option features (%Lx).\n",
2457 (unsigned long long)features);
2458 err = -EINVAL;
2459 goto fail_alloc;
2460 }
2461
2462 btrfs_init_workers(&fs_info->generic_worker,
2463 "genwork", 1, NULL);
2464
2465 btrfs_init_workers(&fs_info->workers, "worker",
2466 fs_info->thread_pool_size,
2467 &fs_info->generic_worker);
2468
2469 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2470 fs_info->thread_pool_size,
2471 &fs_info->generic_worker);
2472
2473 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2474 fs_info->thread_pool_size,
2475 &fs_info->generic_worker);
2476
2477 btrfs_init_workers(&fs_info->submit_workers, "submit",
2478 min_t(u64, fs_devices->num_devices,
2479 fs_info->thread_pool_size),
2480 &fs_info->generic_worker);
2481
2482 btrfs_init_workers(&fs_info->caching_workers, "cache",
2483 2, &fs_info->generic_worker);
2484
2485 /* a higher idle thresh on the submit workers makes it much more
2486 * likely that bios will be send down in a sane order to the
2487 * devices
2488 */
2489 fs_info->submit_workers.idle_thresh = 64;
2490
2491 fs_info->workers.idle_thresh = 16;
2492 fs_info->workers.ordered = 1;
2493
2494 fs_info->delalloc_workers.idle_thresh = 2;
2495 fs_info->delalloc_workers.ordered = 1;
2496
2497 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2498 &fs_info->generic_worker);
2499 btrfs_init_workers(&fs_info->endio_workers, "endio",
2500 fs_info->thread_pool_size,
2501 &fs_info->generic_worker);
2502 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2503 fs_info->thread_pool_size,
2504 &fs_info->generic_worker);
2505 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2506 "endio-meta-write", fs_info->thread_pool_size,
2507 &fs_info->generic_worker);
2508 btrfs_init_workers(&fs_info->endio_raid56_workers,
2509 "endio-raid56", fs_info->thread_pool_size,
2510 &fs_info->generic_worker);
2511 btrfs_init_workers(&fs_info->rmw_workers,
2512 "rmw", fs_info->thread_pool_size,
2513 &fs_info->generic_worker);
2514 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2515 fs_info->thread_pool_size,
2516 &fs_info->generic_worker);
2517 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2518 1, &fs_info->generic_worker);
2519 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2520 fs_info->thread_pool_size,
2521 &fs_info->generic_worker);
2522 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2523 fs_info->thread_pool_size,
2524 &fs_info->generic_worker);
2525 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2526 &fs_info->generic_worker);
2527
2528 /*
2529 * endios are largely parallel and should have a very
2530 * low idle thresh
2531 */
2532 fs_info->endio_workers.idle_thresh = 4;
2533 fs_info->endio_meta_workers.idle_thresh = 4;
2534 fs_info->endio_raid56_workers.idle_thresh = 4;
2535 fs_info->rmw_workers.idle_thresh = 2;
2536
2537 fs_info->endio_write_workers.idle_thresh = 2;
2538 fs_info->endio_meta_write_workers.idle_thresh = 2;
2539 fs_info->readahead_workers.idle_thresh = 2;
2540
2541 /*
2542 * btrfs_start_workers can really only fail because of ENOMEM so just
2543 * return -ENOMEM if any of these fail.
2544 */
2545 ret = btrfs_start_workers(&fs_info->workers);
2546 ret |= btrfs_start_workers(&fs_info->generic_worker);
2547 ret |= btrfs_start_workers(&fs_info->submit_workers);
2548 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2549 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2550 ret |= btrfs_start_workers(&fs_info->endio_workers);
2551 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2552 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2553 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2554 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2555 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2556 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2557 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2558 ret |= btrfs_start_workers(&fs_info->caching_workers);
2559 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2560 ret |= btrfs_start_workers(&fs_info->flush_workers);
2561 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2562 if (ret) {
2563 err = -ENOMEM;
2564 goto fail_sb_buffer;
2565 }
2566
2567 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2568 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2569 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2570
2571 tree_root->nodesize = nodesize;
2572 tree_root->leafsize = leafsize;
2573 tree_root->sectorsize = sectorsize;
2574 tree_root->stripesize = stripesize;
2575
2576 sb->s_blocksize = sectorsize;
2577 sb->s_blocksize_bits = blksize_bits(sectorsize);
2578
2579 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2580 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2581 goto fail_sb_buffer;
2582 }
2583
2584 if (sectorsize != PAGE_SIZE) {
2585 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2586 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2587 goto fail_sb_buffer;
2588 }
2589
2590 mutex_lock(&fs_info->chunk_mutex);
2591 ret = btrfs_read_sys_array(tree_root);
2592 mutex_unlock(&fs_info->chunk_mutex);
2593 if (ret) {
2594 printk(KERN_WARNING "btrfs: failed to read the system "
2595 "array on %s\n", sb->s_id);
2596 goto fail_sb_buffer;
2597 }
2598
2599 blocksize = btrfs_level_size(tree_root,
2600 btrfs_super_chunk_root_level(disk_super));
2601 generation = btrfs_super_chunk_root_generation(disk_super);
2602
2603 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2604 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2605
2606 chunk_root->node = read_tree_block(chunk_root,
2607 btrfs_super_chunk_root(disk_super),
2608 blocksize, generation);
2609 if (!chunk_root->node ||
2610 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2611 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2612 sb->s_id);
2613 goto fail_tree_roots;
2614 }
2615 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2616 chunk_root->commit_root = btrfs_root_node(chunk_root);
2617
2618 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2619 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2620 BTRFS_UUID_SIZE);
2621
2622 ret = btrfs_read_chunk_tree(chunk_root);
2623 if (ret) {
2624 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2625 sb->s_id);
2626 goto fail_tree_roots;
2627 }
2628
2629 /*
2630 * keep the device that is marked to be the target device for the
2631 * dev_replace procedure
2632 */
2633 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2634
2635 if (!fs_devices->latest_bdev) {
2636 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2637 sb->s_id);
2638 goto fail_tree_roots;
2639 }
2640
2641retry_root_backup:
2642 blocksize = btrfs_level_size(tree_root,
2643 btrfs_super_root_level(disk_super));
2644 generation = btrfs_super_generation(disk_super);
2645
2646 tree_root->node = read_tree_block(tree_root,
2647 btrfs_super_root(disk_super),
2648 blocksize, generation);
2649 if (!tree_root->node ||
2650 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2651 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2652 sb->s_id);
2653
2654 goto recovery_tree_root;
2655 }
2656
2657 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2658 tree_root->commit_root = btrfs_root_node(tree_root);
2659
2660 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2661 location.type = BTRFS_ROOT_ITEM_KEY;
2662 location.offset = 0;
2663
2664 extent_root = btrfs_read_tree_root(tree_root, &location);
2665 if (IS_ERR(extent_root)) {
2666 ret = PTR_ERR(extent_root);
2667 goto recovery_tree_root;
2668 }
2669 extent_root->track_dirty = 1;
2670 fs_info->extent_root = extent_root;
2671
2672 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2673 dev_root = btrfs_read_tree_root(tree_root, &location);
2674 if (IS_ERR(dev_root)) {
2675 ret = PTR_ERR(dev_root);
2676 goto recovery_tree_root;
2677 }
2678 dev_root->track_dirty = 1;
2679 fs_info->dev_root = dev_root;
2680 btrfs_init_devices_late(fs_info);
2681
2682 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2683 csum_root = btrfs_read_tree_root(tree_root, &location);
2684 if (IS_ERR(csum_root)) {
2685 ret = PTR_ERR(csum_root);
2686 goto recovery_tree_root;
2687 }
2688 csum_root->track_dirty = 1;
2689 fs_info->csum_root = csum_root;
2690
2691 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2692 quota_root = btrfs_read_tree_root(tree_root, &location);
2693 if (!IS_ERR(quota_root)) {
2694 quota_root->track_dirty = 1;
2695 fs_info->quota_enabled = 1;
2696 fs_info->pending_quota_state = 1;
2697 fs_info->quota_root = quota_root;
2698 }
2699
2700 fs_info->generation = generation;
2701 fs_info->last_trans_committed = generation;
2702
2703 ret = btrfs_recover_balance(fs_info);
2704 if (ret) {
2705 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2706 goto fail_block_groups;
2707 }
2708
2709 ret = btrfs_init_dev_stats(fs_info);
2710 if (ret) {
2711 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2712 ret);
2713 goto fail_block_groups;
2714 }
2715
2716 ret = btrfs_init_dev_replace(fs_info);
2717 if (ret) {
2718 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2719 goto fail_block_groups;
2720 }
2721
2722 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2723
2724 ret = btrfs_init_space_info(fs_info);
2725 if (ret) {
2726 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2727 goto fail_block_groups;
2728 }
2729
2730 ret = btrfs_read_block_groups(extent_root);
2731 if (ret) {
2732 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2733 goto fail_block_groups;
2734 }
2735 fs_info->num_tolerated_disk_barrier_failures =
2736 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2737 if (fs_info->fs_devices->missing_devices >
2738 fs_info->num_tolerated_disk_barrier_failures &&
2739 !(sb->s_flags & MS_RDONLY)) {
2740 printk(KERN_WARNING
2741 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2742 goto fail_block_groups;
2743 }
2744
2745 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2746 "btrfs-cleaner");
2747 if (IS_ERR(fs_info->cleaner_kthread))
2748 goto fail_block_groups;
2749
2750 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2751 tree_root,
2752 "btrfs-transaction");
2753 if (IS_ERR(fs_info->transaction_kthread))
2754 goto fail_cleaner;
2755
2756 if (!btrfs_test_opt(tree_root, SSD) &&
2757 !btrfs_test_opt(tree_root, NOSSD) &&
2758 !fs_info->fs_devices->rotating) {
2759 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2760 "mode\n");
2761 btrfs_set_opt(fs_info->mount_opt, SSD);
2762 }
2763
2764#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2765 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2766 ret = btrfsic_mount(tree_root, fs_devices,
2767 btrfs_test_opt(tree_root,
2768 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2769 1 : 0,
2770 fs_info->check_integrity_print_mask);
2771 if (ret)
2772 printk(KERN_WARNING "btrfs: failed to initialize"
2773 " integrity check module %s\n", sb->s_id);
2774 }
2775#endif
2776 ret = btrfs_read_qgroup_config(fs_info);
2777 if (ret)
2778 goto fail_trans_kthread;
2779
2780 /* do not make disk changes in broken FS */
2781 if (btrfs_super_log_root(disk_super) != 0) {
2782 u64 bytenr = btrfs_super_log_root(disk_super);
2783
2784 if (fs_devices->rw_devices == 0) {
2785 printk(KERN_WARNING "Btrfs log replay required "
2786 "on RO media\n");
2787 err = -EIO;
2788 goto fail_qgroup;
2789 }
2790 blocksize =
2791 btrfs_level_size(tree_root,
2792 btrfs_super_log_root_level(disk_super));
2793
2794 log_tree_root = btrfs_alloc_root(fs_info);
2795 if (!log_tree_root) {
2796 err = -ENOMEM;
2797 goto fail_qgroup;
2798 }
2799
2800 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2801 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2802
2803 log_tree_root->node = read_tree_block(tree_root, bytenr,
2804 blocksize,
2805 generation + 1);
2806 if (!log_tree_root->node ||
2807 !extent_buffer_uptodate(log_tree_root->node)) {
2808 printk(KERN_ERR "btrfs: failed to read log tree\n");
2809 free_extent_buffer(log_tree_root->node);
2810 kfree(log_tree_root);
2811 goto fail_trans_kthread;
2812 }
2813 /* returns with log_tree_root freed on success */
2814 ret = btrfs_recover_log_trees(log_tree_root);
2815 if (ret) {
2816 btrfs_error(tree_root->fs_info, ret,
2817 "Failed to recover log tree");
2818 free_extent_buffer(log_tree_root->node);
2819 kfree(log_tree_root);
2820 goto fail_trans_kthread;
2821 }
2822
2823 if (sb->s_flags & MS_RDONLY) {
2824 ret = btrfs_commit_super(tree_root);
2825 if (ret)
2826 goto fail_trans_kthread;
2827 }
2828 }
2829
2830 ret = btrfs_find_orphan_roots(tree_root);
2831 if (ret)
2832 goto fail_trans_kthread;
2833
2834 if (!(sb->s_flags & MS_RDONLY)) {
2835 ret = btrfs_cleanup_fs_roots(fs_info);
2836 if (ret)
2837 goto fail_trans_kthread;
2838
2839 ret = btrfs_recover_relocation(tree_root);
2840 if (ret < 0) {
2841 printk(KERN_WARNING
2842 "btrfs: failed to recover relocation\n");
2843 err = -EINVAL;
2844 goto fail_qgroup;
2845 }
2846 }
2847
2848 location.objectid = BTRFS_FS_TREE_OBJECTID;
2849 location.type = BTRFS_ROOT_ITEM_KEY;
2850 location.offset = 0;
2851
2852 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2853 if (IS_ERR(fs_info->fs_root)) {
2854 err = PTR_ERR(fs_info->fs_root);
2855 goto fail_qgroup;
2856 }
2857
2858 if (sb->s_flags & MS_RDONLY)
2859 return 0;
2860
2861 down_read(&fs_info->cleanup_work_sem);
2862 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2863 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2864 up_read(&fs_info->cleanup_work_sem);
2865 close_ctree(tree_root);
2866 return ret;
2867 }
2868 up_read(&fs_info->cleanup_work_sem);
2869
2870 ret = btrfs_resume_balance_async(fs_info);
2871 if (ret) {
2872 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2873 close_ctree(tree_root);
2874 return ret;
2875 }
2876
2877 ret = btrfs_resume_dev_replace_async(fs_info);
2878 if (ret) {
2879 pr_warn("btrfs: failed to resume dev_replace\n");
2880 close_ctree(tree_root);
2881 return ret;
2882 }
2883
2884 btrfs_qgroup_rescan_resume(fs_info);
2885
2886 return 0;
2887
2888fail_qgroup:
2889 btrfs_free_qgroup_config(fs_info);
2890fail_trans_kthread:
2891 kthread_stop(fs_info->transaction_kthread);
2892 btrfs_cleanup_transaction(fs_info->tree_root);
2893 del_fs_roots(fs_info);
2894fail_cleaner:
2895 kthread_stop(fs_info->cleaner_kthread);
2896
2897 /*
2898 * make sure we're done with the btree inode before we stop our
2899 * kthreads
2900 */
2901 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2902
2903fail_block_groups:
2904 btrfs_put_block_group_cache(fs_info);
2905 btrfs_free_block_groups(fs_info);
2906
2907fail_tree_roots:
2908 free_root_pointers(fs_info, 1);
2909 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2910
2911fail_sb_buffer:
2912 btrfs_stop_all_workers(fs_info);
2913fail_alloc:
2914fail_iput:
2915 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2916
2917 iput(fs_info->btree_inode);
2918fail_delalloc_bytes:
2919 percpu_counter_destroy(&fs_info->delalloc_bytes);
2920fail_dirty_metadata_bytes:
2921 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2922fail_bdi:
2923 bdi_destroy(&fs_info->bdi);
2924fail_srcu:
2925 cleanup_srcu_struct(&fs_info->subvol_srcu);
2926fail:
2927 btrfs_free_stripe_hash_table(fs_info);
2928 btrfs_close_devices(fs_info->fs_devices);
2929 return err;
2930
2931recovery_tree_root:
2932 if (!btrfs_test_opt(tree_root, RECOVERY))
2933 goto fail_tree_roots;
2934
2935 free_root_pointers(fs_info, 0);
2936
2937 /* don't use the log in recovery mode, it won't be valid */
2938 btrfs_set_super_log_root(disk_super, 0);
2939
2940 /* we can't trust the free space cache either */
2941 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2942
2943 ret = next_root_backup(fs_info, fs_info->super_copy,
2944 &num_backups_tried, &backup_index);
2945 if (ret == -1)
2946 goto fail_block_groups;
2947 goto retry_root_backup;
2948}
2949
2950static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2951{
2952 if (uptodate) {
2953 set_buffer_uptodate(bh);
2954 } else {
2955 struct btrfs_device *device = (struct btrfs_device *)
2956 bh->b_private;
2957
2958 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2959 "I/O error on %s\n",
2960 rcu_str_deref(device->name));
2961 /* note, we dont' set_buffer_write_io_error because we have
2962 * our own ways of dealing with the IO errors
2963 */
2964 clear_buffer_uptodate(bh);
2965 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2966 }
2967 unlock_buffer(bh);
2968 put_bh(bh);
2969}
2970
2971struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2972{
2973 struct buffer_head *bh;
2974 struct buffer_head *latest = NULL;
2975 struct btrfs_super_block *super;
2976 int i;
2977 u64 transid = 0;
2978 u64 bytenr;
2979
2980 /* we would like to check all the supers, but that would make
2981 * a btrfs mount succeed after a mkfs from a different FS.
2982 * So, we need to add a special mount option to scan for
2983 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2984 */
2985 for (i = 0; i < 1; i++) {
2986 bytenr = btrfs_sb_offset(i);
2987 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2988 i_size_read(bdev->bd_inode))
2989 break;
2990 bh = __bread(bdev, bytenr / 4096,
2991 BTRFS_SUPER_INFO_SIZE);
2992 if (!bh)
2993 continue;
2994
2995 super = (struct btrfs_super_block *)bh->b_data;
2996 if (btrfs_super_bytenr(super) != bytenr ||
2997 btrfs_super_magic(super) != BTRFS_MAGIC) {
2998 brelse(bh);
2999 continue;
3000 }
3001
3002 if (!latest || btrfs_super_generation(super) > transid) {
3003 brelse(latest);
3004 latest = bh;
3005 transid = btrfs_super_generation(super);
3006 } else {
3007 brelse(bh);
3008 }
3009 }
3010 return latest;
3011}
3012
3013/*
3014 * this should be called twice, once with wait == 0 and
3015 * once with wait == 1. When wait == 0 is done, all the buffer heads
3016 * we write are pinned.
3017 *
3018 * They are released when wait == 1 is done.
3019 * max_mirrors must be the same for both runs, and it indicates how
3020 * many supers on this one device should be written.
3021 *
3022 * max_mirrors == 0 means to write them all.
3023 */
3024static int write_dev_supers(struct btrfs_device *device,
3025 struct btrfs_super_block *sb,
3026 int do_barriers, int wait, int max_mirrors)
3027{
3028 struct buffer_head *bh;
3029 int i;
3030 int ret;
3031 int errors = 0;
3032 u32 crc;
3033 u64 bytenr;
3034
3035 if (max_mirrors == 0)
3036 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3037
3038 for (i = 0; i < max_mirrors; i++) {
3039 bytenr = btrfs_sb_offset(i);
3040 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3041 break;
3042
3043 if (wait) {
3044 bh = __find_get_block(device->bdev, bytenr / 4096,
3045 BTRFS_SUPER_INFO_SIZE);
3046 if (!bh) {
3047 errors++;
3048 continue;
3049 }
3050 wait_on_buffer(bh);
3051 if (!buffer_uptodate(bh))
3052 errors++;
3053
3054 /* drop our reference */
3055 brelse(bh);
3056
3057 /* drop the reference from the wait == 0 run */
3058 brelse(bh);
3059 continue;
3060 } else {
3061 btrfs_set_super_bytenr(sb, bytenr);
3062
3063 crc = ~(u32)0;
3064 crc = btrfs_csum_data((char *)sb +
3065 BTRFS_CSUM_SIZE, crc,
3066 BTRFS_SUPER_INFO_SIZE -
3067 BTRFS_CSUM_SIZE);
3068 btrfs_csum_final(crc, sb->csum);
3069
3070 /*
3071 * one reference for us, and we leave it for the
3072 * caller
3073 */
3074 bh = __getblk(device->bdev, bytenr / 4096,
3075 BTRFS_SUPER_INFO_SIZE);
3076 if (!bh) {
3077 printk(KERN_ERR "btrfs: couldn't get super "
3078 "buffer head for bytenr %Lu\n", bytenr);
3079 errors++;
3080 continue;
3081 }
3082
3083 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3084
3085 /* one reference for submit_bh */
3086 get_bh(bh);
3087
3088 set_buffer_uptodate(bh);
3089 lock_buffer(bh);
3090 bh->b_end_io = btrfs_end_buffer_write_sync;
3091 bh->b_private = device;
3092 }
3093
3094 /*
3095 * we fua the first super. The others we allow
3096 * to go down lazy.
3097 */
3098 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3099 if (ret)
3100 errors++;
3101 }
3102 return errors < i ? 0 : -1;
3103}
3104
3105/*
3106 * endio for the write_dev_flush, this will wake anyone waiting
3107 * for the barrier when it is done
3108 */
3109static void btrfs_end_empty_barrier(struct bio *bio, int err)
3110{
3111 if (err) {
3112 if (err == -EOPNOTSUPP)
3113 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3114 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3115 }
3116 if (bio->bi_private)
3117 complete(bio->bi_private);
3118 bio_put(bio);
3119}
3120
3121/*
3122 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3123 * sent down. With wait == 1, it waits for the previous flush.
3124 *
3125 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3126 * capable
3127 */
3128static int write_dev_flush(struct btrfs_device *device, int wait)
3129{
3130 struct bio *bio;
3131 int ret = 0;
3132
3133 if (device->nobarriers)
3134 return 0;
3135
3136 if (wait) {
3137 bio = device->flush_bio;
3138 if (!bio)
3139 return 0;
3140
3141 wait_for_completion(&device->flush_wait);
3142
3143 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3144 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3145 rcu_str_deref(device->name));
3146 device->nobarriers = 1;
3147 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3148 ret = -EIO;
3149 btrfs_dev_stat_inc_and_print(device,
3150 BTRFS_DEV_STAT_FLUSH_ERRS);
3151 }
3152
3153 /* drop the reference from the wait == 0 run */
3154 bio_put(bio);
3155 device->flush_bio = NULL;
3156
3157 return ret;
3158 }
3159
3160 /*
3161 * one reference for us, and we leave it for the
3162 * caller
3163 */
3164 device->flush_bio = NULL;
3165 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3166 if (!bio)
3167 return -ENOMEM;
3168
3169 bio->bi_end_io = btrfs_end_empty_barrier;
3170 bio->bi_bdev = device->bdev;
3171 init_completion(&device->flush_wait);
3172 bio->bi_private = &device->flush_wait;
3173 device->flush_bio = bio;
3174
3175 bio_get(bio);
3176 btrfsic_submit_bio(WRITE_FLUSH, bio);
3177
3178 return 0;
3179}
3180
3181/*
3182 * send an empty flush down to each device in parallel,
3183 * then wait for them
3184 */
3185static int barrier_all_devices(struct btrfs_fs_info *info)
3186{
3187 struct list_head *head;
3188 struct btrfs_device *dev;
3189 int errors_send = 0;
3190 int errors_wait = 0;
3191 int ret;
3192
3193 /* send down all the barriers */
3194 head = &info->fs_devices->devices;
3195 list_for_each_entry_rcu(dev, head, dev_list) {
3196 if (!dev->bdev) {
3197 errors_send++;
3198 continue;
3199 }
3200 if (!dev->in_fs_metadata || !dev->writeable)
3201 continue;
3202
3203 ret = write_dev_flush(dev, 0);
3204 if (ret)
3205 errors_send++;
3206 }
3207
3208 /* wait for all the barriers */
3209 list_for_each_entry_rcu(dev, head, dev_list) {
3210 if (!dev->bdev) {
3211 errors_wait++;
3212 continue;
3213 }
3214 if (!dev->in_fs_metadata || !dev->writeable)
3215 continue;
3216
3217 ret = write_dev_flush(dev, 1);
3218 if (ret)
3219 errors_wait++;
3220 }
3221 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3222 errors_wait > info->num_tolerated_disk_barrier_failures)
3223 return -EIO;
3224 return 0;
3225}
3226
3227int btrfs_calc_num_tolerated_disk_barrier_failures(
3228 struct btrfs_fs_info *fs_info)
3229{
3230 struct btrfs_ioctl_space_info space;
3231 struct btrfs_space_info *sinfo;
3232 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3233 BTRFS_BLOCK_GROUP_SYSTEM,
3234 BTRFS_BLOCK_GROUP_METADATA,
3235 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3236 int num_types = 4;
3237 int i;
3238 int c;
3239 int num_tolerated_disk_barrier_failures =
3240 (int)fs_info->fs_devices->num_devices;
3241
3242 for (i = 0; i < num_types; i++) {
3243 struct btrfs_space_info *tmp;
3244
3245 sinfo = NULL;
3246 rcu_read_lock();
3247 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3248 if (tmp->flags == types[i]) {
3249 sinfo = tmp;
3250 break;
3251 }
3252 }
3253 rcu_read_unlock();
3254
3255 if (!sinfo)
3256 continue;
3257
3258 down_read(&sinfo->groups_sem);
3259 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3260 if (!list_empty(&sinfo->block_groups[c])) {
3261 u64 flags;
3262
3263 btrfs_get_block_group_info(
3264 &sinfo->block_groups[c], &space);
3265 if (space.total_bytes == 0 ||
3266 space.used_bytes == 0)
3267 continue;
3268 flags = space.flags;
3269 /*
3270 * return
3271 * 0: if dup, single or RAID0 is configured for
3272 * any of metadata, system or data, else
3273 * 1: if RAID5 is configured, or if RAID1 or
3274 * RAID10 is configured and only two mirrors
3275 * are used, else
3276 * 2: if RAID6 is configured, else
3277 * num_mirrors - 1: if RAID1 or RAID10 is
3278 * configured and more than
3279 * 2 mirrors are used.
3280 */
3281 if (num_tolerated_disk_barrier_failures > 0 &&
3282 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3283 BTRFS_BLOCK_GROUP_RAID0)) ||
3284 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3285 == 0)))
3286 num_tolerated_disk_barrier_failures = 0;
3287 else if (num_tolerated_disk_barrier_failures > 1) {
3288 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3289 BTRFS_BLOCK_GROUP_RAID5 |
3290 BTRFS_BLOCK_GROUP_RAID10)) {
3291 num_tolerated_disk_barrier_failures = 1;
3292 } else if (flags &
3293 BTRFS_BLOCK_GROUP_RAID6) {
3294 num_tolerated_disk_barrier_failures = 2;
3295 }
3296 }
3297 }
3298 }
3299 up_read(&sinfo->groups_sem);
3300 }
3301
3302 return num_tolerated_disk_barrier_failures;
3303}
3304
3305static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3306{
3307 struct list_head *head;
3308 struct btrfs_device *dev;
3309 struct btrfs_super_block *sb;
3310 struct btrfs_dev_item *dev_item;
3311 int ret;
3312 int do_barriers;
3313 int max_errors;
3314 int total_errors = 0;
3315 u64 flags;
3316
3317 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3318 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3319 backup_super_roots(root->fs_info);
3320
3321 sb = root->fs_info->super_for_commit;
3322 dev_item = &sb->dev_item;
3323
3324 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3325 head = &root->fs_info->fs_devices->devices;
3326
3327 if (do_barriers) {
3328 ret = barrier_all_devices(root->fs_info);
3329 if (ret) {
3330 mutex_unlock(
3331 &root->fs_info->fs_devices->device_list_mutex);
3332 btrfs_error(root->fs_info, ret,
3333 "errors while submitting device barriers.");
3334 return ret;
3335 }
3336 }
3337
3338 list_for_each_entry_rcu(dev, head, dev_list) {
3339 if (!dev->bdev) {
3340 total_errors++;
3341 continue;
3342 }
3343 if (!dev->in_fs_metadata || !dev->writeable)
3344 continue;
3345
3346 btrfs_set_stack_device_generation(dev_item, 0);
3347 btrfs_set_stack_device_type(dev_item, dev->type);
3348 btrfs_set_stack_device_id(dev_item, dev->devid);
3349 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3350 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3351 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3352 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3353 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3354 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3355 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3356
3357 flags = btrfs_super_flags(sb);
3358 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3359
3360 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3361 if (ret)
3362 total_errors++;
3363 }
3364 if (total_errors > max_errors) {
3365 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3366 total_errors);
3367
3368 /* This shouldn't happen. FUA is masked off if unsupported */
3369 BUG();
3370 }
3371
3372 total_errors = 0;
3373 list_for_each_entry_rcu(dev, head, dev_list) {
3374 if (!dev->bdev)
3375 continue;
3376 if (!dev->in_fs_metadata || !dev->writeable)
3377 continue;
3378
3379 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3380 if (ret)
3381 total_errors++;
3382 }
3383 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3384 if (total_errors > max_errors) {
3385 btrfs_error(root->fs_info, -EIO,
3386 "%d errors while writing supers", total_errors);
3387 return -EIO;
3388 }
3389 return 0;
3390}
3391
3392int write_ctree_super(struct btrfs_trans_handle *trans,
3393 struct btrfs_root *root, int max_mirrors)
3394{
3395 int ret;
3396
3397 ret = write_all_supers(root, max_mirrors);
3398 return ret;
3399}
3400
3401/* Drop a fs root from the radix tree and free it. */
3402void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3403 struct btrfs_root *root)
3404{
3405 spin_lock(&fs_info->fs_roots_radix_lock);
3406 radix_tree_delete(&fs_info->fs_roots_radix,
3407 (unsigned long)root->root_key.objectid);
3408 spin_unlock(&fs_info->fs_roots_radix_lock);
3409
3410 if (btrfs_root_refs(&root->root_item) == 0)
3411 synchronize_srcu(&fs_info->subvol_srcu);
3412
3413 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3414 btrfs_free_log(NULL, root);
3415 btrfs_free_log_root_tree(NULL, fs_info);
3416 }
3417
3418 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3419 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3420 free_fs_root(root);
3421}
3422
3423static void free_fs_root(struct btrfs_root *root)
3424{
3425 iput(root->cache_inode);
3426 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3427 if (root->anon_dev)
3428 free_anon_bdev(root->anon_dev);
3429 free_extent_buffer(root->node);
3430 free_extent_buffer(root->commit_root);
3431 kfree(root->free_ino_ctl);
3432 kfree(root->free_ino_pinned);
3433 kfree(root->name);
3434 btrfs_put_fs_root(root);
3435}
3436
3437void btrfs_free_fs_root(struct btrfs_root *root)
3438{
3439 free_fs_root(root);
3440}
3441
3442int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3443{
3444 u64 root_objectid = 0;
3445 struct btrfs_root *gang[8];
3446 int i;
3447 int ret;
3448
3449 while (1) {
3450 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3451 (void **)gang, root_objectid,
3452 ARRAY_SIZE(gang));
3453 if (!ret)
3454 break;
3455
3456 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3457 for (i = 0; i < ret; i++) {
3458 int err;
3459
3460 root_objectid = gang[i]->root_key.objectid;
3461 err = btrfs_orphan_cleanup(gang[i]);
3462 if (err)
3463 return err;
3464 }
3465 root_objectid++;
3466 }
3467 return 0;
3468}
3469
3470int btrfs_commit_super(struct btrfs_root *root)
3471{
3472 struct btrfs_trans_handle *trans;
3473 int ret;
3474
3475 mutex_lock(&root->fs_info->cleaner_mutex);
3476 btrfs_run_delayed_iputs(root);
3477 mutex_unlock(&root->fs_info->cleaner_mutex);
3478 wake_up_process(root->fs_info->cleaner_kthread);
3479
3480 /* wait until ongoing cleanup work done */
3481 down_write(&root->fs_info->cleanup_work_sem);
3482 up_write(&root->fs_info->cleanup_work_sem);
3483
3484 trans = btrfs_join_transaction(root);
3485 if (IS_ERR(trans))
3486 return PTR_ERR(trans);
3487 ret = btrfs_commit_transaction(trans, root);
3488 if (ret)
3489 return ret;
3490 /* run commit again to drop the original snapshot */
3491 trans = btrfs_join_transaction(root);
3492 if (IS_ERR(trans))
3493 return PTR_ERR(trans);
3494 ret = btrfs_commit_transaction(trans, root);
3495 if (ret)
3496 return ret;
3497 ret = btrfs_write_and_wait_transaction(NULL, root);
3498 if (ret) {
3499 btrfs_error(root->fs_info, ret,
3500 "Failed to sync btree inode to disk.");
3501 return ret;
3502 }
3503
3504 ret = write_ctree_super(NULL, root, 0);
3505 return ret;
3506}
3507
3508int close_ctree(struct btrfs_root *root)
3509{
3510 struct btrfs_fs_info *fs_info = root->fs_info;
3511 int ret;
3512
3513 fs_info->closing = 1;
3514 smp_mb();
3515
3516 /* pause restriper - we want to resume on mount */
3517 btrfs_pause_balance(fs_info);
3518
3519 btrfs_dev_replace_suspend_for_unmount(fs_info);
3520
3521 btrfs_scrub_cancel(fs_info);
3522
3523 /* wait for any defraggers to finish */
3524 wait_event(fs_info->transaction_wait,
3525 (atomic_read(&fs_info->defrag_running) == 0));
3526
3527 /* clear out the rbtree of defraggable inodes */
3528 btrfs_cleanup_defrag_inodes(fs_info);
3529
3530 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3531 ret = btrfs_commit_super(root);
3532 if (ret)
3533 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3534 }
3535
3536 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3537 btrfs_error_commit_super(root);
3538
3539 btrfs_put_block_group_cache(fs_info);
3540
3541 kthread_stop(fs_info->transaction_kthread);
3542 kthread_stop(fs_info->cleaner_kthread);
3543
3544 fs_info->closing = 2;
3545 smp_mb();
3546
3547 btrfs_free_qgroup_config(root->fs_info);
3548
3549 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3550 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3551 percpu_counter_sum(&fs_info->delalloc_bytes));
3552 }
3553
3554 btrfs_free_block_groups(fs_info);
3555
3556 btrfs_stop_all_workers(fs_info);
3557
3558 del_fs_roots(fs_info);
3559
3560 free_root_pointers(fs_info, 1);
3561
3562 iput(fs_info->btree_inode);
3563
3564#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3565 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3566 btrfsic_unmount(root, fs_info->fs_devices);
3567#endif
3568
3569 btrfs_close_devices(fs_info->fs_devices);
3570 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3571
3572 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3573 percpu_counter_destroy(&fs_info->delalloc_bytes);
3574 bdi_destroy(&fs_info->bdi);
3575 cleanup_srcu_struct(&fs_info->subvol_srcu);
3576
3577 btrfs_free_stripe_hash_table(fs_info);
3578
3579 return 0;
3580}
3581
3582int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3583 int atomic)
3584{
3585 int ret;
3586 struct inode *btree_inode = buf->pages[0]->mapping->host;
3587
3588 ret = extent_buffer_uptodate(buf);
3589 if (!ret)
3590 return ret;
3591
3592 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3593 parent_transid, atomic);
3594 if (ret == -EAGAIN)
3595 return ret;
3596 return !ret;
3597}
3598
3599int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3600{
3601 return set_extent_buffer_uptodate(buf);
3602}
3603
3604void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3605{
3606 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3607 u64 transid = btrfs_header_generation(buf);
3608 int was_dirty;
3609
3610 btrfs_assert_tree_locked(buf);
3611 if (transid != root->fs_info->generation)
3612 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3613 "found %llu running %llu\n",
3614 (unsigned long long)buf->start,
3615 (unsigned long long)transid,
3616 (unsigned long long)root->fs_info->generation);
3617 was_dirty = set_extent_buffer_dirty(buf);
3618 if (!was_dirty)
3619 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3620 buf->len,
3621 root->fs_info->dirty_metadata_batch);
3622}
3623
3624static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3625 int flush_delayed)
3626{
3627 /*
3628 * looks as though older kernels can get into trouble with
3629 * this code, they end up stuck in balance_dirty_pages forever
3630 */
3631 int ret;
3632
3633 if (current->flags & PF_MEMALLOC)
3634 return;
3635
3636 if (flush_delayed)
3637 btrfs_balance_delayed_items(root);
3638
3639 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3640 BTRFS_DIRTY_METADATA_THRESH);
3641 if (ret > 0) {
3642 balance_dirty_pages_ratelimited(
3643 root->fs_info->btree_inode->i_mapping);
3644 }
3645 return;
3646}
3647
3648void btrfs_btree_balance_dirty(struct btrfs_root *root)
3649{
3650 __btrfs_btree_balance_dirty(root, 1);
3651}
3652
3653void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3654{
3655 __btrfs_btree_balance_dirty(root, 0);
3656}
3657
3658int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3659{
3660 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3661 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3662}
3663
3664static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3665 int read_only)
3666{
3667 /*
3668 * Placeholder for checks
3669 */
3670 return 0;
3671}
3672
3673static void btrfs_error_commit_super(struct btrfs_root *root)
3674{
3675 mutex_lock(&root->fs_info->cleaner_mutex);
3676 btrfs_run_delayed_iputs(root);
3677 mutex_unlock(&root->fs_info->cleaner_mutex);
3678
3679 down_write(&root->fs_info->cleanup_work_sem);
3680 up_write(&root->fs_info->cleanup_work_sem);
3681
3682 /* cleanup FS via transaction */
3683 btrfs_cleanup_transaction(root);
3684}
3685
3686static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3687 struct btrfs_root *root)
3688{
3689 struct btrfs_inode *btrfs_inode;
3690 struct list_head splice;
3691
3692 INIT_LIST_HEAD(&splice);
3693
3694 mutex_lock(&root->fs_info->ordered_operations_mutex);
3695 spin_lock(&root->fs_info->ordered_root_lock);
3696
3697 list_splice_init(&t->ordered_operations, &splice);
3698 while (!list_empty(&splice)) {
3699 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3700 ordered_operations);
3701
3702 list_del_init(&btrfs_inode->ordered_operations);
3703 spin_unlock(&root->fs_info->ordered_root_lock);
3704
3705 btrfs_invalidate_inodes(btrfs_inode->root);
3706
3707 spin_lock(&root->fs_info->ordered_root_lock);
3708 }
3709
3710 spin_unlock(&root->fs_info->ordered_root_lock);
3711 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3712}
3713
3714static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3715{
3716 struct btrfs_ordered_extent *ordered;
3717
3718 spin_lock(&root->ordered_extent_lock);
3719 /*
3720 * This will just short circuit the ordered completion stuff which will
3721 * make sure the ordered extent gets properly cleaned up.
3722 */
3723 list_for_each_entry(ordered, &root->ordered_extents,
3724 root_extent_list)
3725 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3726 spin_unlock(&root->ordered_extent_lock);
3727}
3728
3729static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3730{
3731 struct btrfs_root *root;
3732 struct list_head splice;
3733
3734 INIT_LIST_HEAD(&splice);
3735
3736 spin_lock(&fs_info->ordered_root_lock);
3737 list_splice_init(&fs_info->ordered_roots, &splice);
3738 while (!list_empty(&splice)) {
3739 root = list_first_entry(&splice, struct btrfs_root,
3740 ordered_root);
3741 list_del_init(&root->ordered_root);
3742
3743 btrfs_destroy_ordered_extents(root);
3744
3745 cond_resched_lock(&fs_info->ordered_root_lock);
3746 }
3747 spin_unlock(&fs_info->ordered_root_lock);
3748}
3749
3750int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3751 struct btrfs_root *root)
3752{
3753 struct rb_node *node;
3754 struct btrfs_delayed_ref_root *delayed_refs;
3755 struct btrfs_delayed_ref_node *ref;
3756 int ret = 0;
3757
3758 delayed_refs = &trans->delayed_refs;
3759
3760 spin_lock(&delayed_refs->lock);
3761 if (delayed_refs->num_entries == 0) {
3762 spin_unlock(&delayed_refs->lock);
3763 printk(KERN_INFO "delayed_refs has NO entry\n");
3764 return ret;
3765 }
3766
3767 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3768 struct btrfs_delayed_ref_head *head = NULL;
3769 bool pin_bytes = false;
3770
3771 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3772 atomic_set(&ref->refs, 1);
3773 if (btrfs_delayed_ref_is_head(ref)) {
3774
3775 head = btrfs_delayed_node_to_head(ref);
3776 if (!mutex_trylock(&head->mutex)) {
3777 atomic_inc(&ref->refs);
3778 spin_unlock(&delayed_refs->lock);
3779
3780 /* Need to wait for the delayed ref to run */
3781 mutex_lock(&head->mutex);
3782 mutex_unlock(&head->mutex);
3783 btrfs_put_delayed_ref(ref);
3784
3785 spin_lock(&delayed_refs->lock);
3786 continue;
3787 }
3788
3789 if (head->must_insert_reserved)
3790 pin_bytes = true;
3791 btrfs_free_delayed_extent_op(head->extent_op);
3792 delayed_refs->num_heads--;
3793 if (list_empty(&head->cluster))
3794 delayed_refs->num_heads_ready--;
3795 list_del_init(&head->cluster);
3796 }
3797
3798 ref->in_tree = 0;
3799 rb_erase(&ref->rb_node, &delayed_refs->root);
3800 delayed_refs->num_entries--;
3801 spin_unlock(&delayed_refs->lock);
3802 if (head) {
3803 if (pin_bytes)
3804 btrfs_pin_extent(root, ref->bytenr,
3805 ref->num_bytes, 1);
3806 mutex_unlock(&head->mutex);
3807 }
3808 btrfs_put_delayed_ref(ref);
3809
3810 cond_resched();
3811 spin_lock(&delayed_refs->lock);
3812 }
3813
3814 spin_unlock(&delayed_refs->lock);
3815
3816 return ret;
3817}
3818
3819static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3820{
3821 struct btrfs_pending_snapshot *snapshot;
3822 struct list_head splice;
3823
3824 INIT_LIST_HEAD(&splice);
3825
3826 list_splice_init(&t->pending_snapshots, &splice);
3827
3828 while (!list_empty(&splice)) {
3829 snapshot = list_entry(splice.next,
3830 struct btrfs_pending_snapshot,
3831 list);
3832 snapshot->error = -ECANCELED;
3833 list_del_init(&snapshot->list);
3834 }
3835}
3836
3837static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3838{
3839 struct btrfs_inode *btrfs_inode;
3840 struct list_head splice;
3841
3842 INIT_LIST_HEAD(&splice);
3843
3844 spin_lock(&root->delalloc_lock);
3845 list_splice_init(&root->delalloc_inodes, &splice);
3846
3847 while (!list_empty(&splice)) {
3848 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3849 delalloc_inodes);
3850
3851 list_del_init(&btrfs_inode->delalloc_inodes);
3852 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3853 &btrfs_inode->runtime_flags);
3854 spin_unlock(&root->delalloc_lock);
3855
3856 btrfs_invalidate_inodes(btrfs_inode->root);
3857
3858 spin_lock(&root->delalloc_lock);
3859 }
3860
3861 spin_unlock(&root->delalloc_lock);
3862}
3863
3864static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3865{
3866 struct btrfs_root *root;
3867 struct list_head splice;
3868
3869 INIT_LIST_HEAD(&splice);
3870
3871 spin_lock(&fs_info->delalloc_root_lock);
3872 list_splice_init(&fs_info->delalloc_roots, &splice);
3873 while (!list_empty(&splice)) {
3874 root = list_first_entry(&splice, struct btrfs_root,
3875 delalloc_root);
3876 list_del_init(&root->delalloc_root);
3877 root = btrfs_grab_fs_root(root);
3878 BUG_ON(!root);
3879 spin_unlock(&fs_info->delalloc_root_lock);
3880
3881 btrfs_destroy_delalloc_inodes(root);
3882 btrfs_put_fs_root(root);
3883
3884 spin_lock(&fs_info->delalloc_root_lock);
3885 }
3886 spin_unlock(&fs_info->delalloc_root_lock);
3887}
3888
3889static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3890 struct extent_io_tree *dirty_pages,
3891 int mark)
3892{
3893 int ret;
3894 struct extent_buffer *eb;
3895 u64 start = 0;
3896 u64 end;
3897
3898 while (1) {
3899 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3900 mark, NULL);
3901 if (ret)
3902 break;
3903
3904 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3905 while (start <= end) {
3906 eb = btrfs_find_tree_block(root, start,
3907 root->leafsize);
3908 start += root->leafsize;
3909 if (!eb)
3910 continue;
3911 wait_on_extent_buffer_writeback(eb);
3912
3913 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3914 &eb->bflags))
3915 clear_extent_buffer_dirty(eb);
3916 free_extent_buffer_stale(eb);
3917 }
3918 }
3919
3920 return ret;
3921}
3922
3923static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3924 struct extent_io_tree *pinned_extents)
3925{
3926 struct extent_io_tree *unpin;
3927 u64 start;
3928 u64 end;
3929 int ret;
3930 bool loop = true;
3931
3932 unpin = pinned_extents;
3933again:
3934 while (1) {
3935 ret = find_first_extent_bit(unpin, 0, &start, &end,
3936 EXTENT_DIRTY, NULL);
3937 if (ret)
3938 break;
3939
3940 /* opt_discard */
3941 if (btrfs_test_opt(root, DISCARD))
3942 ret = btrfs_error_discard_extent(root, start,
3943 end + 1 - start,
3944 NULL);
3945
3946 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3947 btrfs_error_unpin_extent_range(root, start, end);
3948 cond_resched();
3949 }
3950
3951 if (loop) {
3952 if (unpin == &root->fs_info->freed_extents[0])
3953 unpin = &root->fs_info->freed_extents[1];
3954 else
3955 unpin = &root->fs_info->freed_extents[0];
3956 loop = false;
3957 goto again;
3958 }
3959
3960 return 0;
3961}
3962
3963void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3964 struct btrfs_root *root)
3965{
3966 btrfs_destroy_delayed_refs(cur_trans, root);
3967 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3968 cur_trans->dirty_pages.dirty_bytes);
3969
3970 cur_trans->state = TRANS_STATE_COMMIT_START;
3971 wake_up(&root->fs_info->transaction_blocked_wait);
3972
3973 btrfs_evict_pending_snapshots(cur_trans);
3974
3975 cur_trans->state = TRANS_STATE_UNBLOCKED;
3976 wake_up(&root->fs_info->transaction_wait);
3977
3978 btrfs_destroy_delayed_inodes(root);
3979 btrfs_assert_delayed_root_empty(root);
3980
3981 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3982 EXTENT_DIRTY);
3983 btrfs_destroy_pinned_extent(root,
3984 root->fs_info->pinned_extents);
3985
3986 cur_trans->state =TRANS_STATE_COMPLETED;
3987 wake_up(&cur_trans->commit_wait);
3988
3989 /*
3990 memset(cur_trans, 0, sizeof(*cur_trans));
3991 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3992 */
3993}
3994
3995static int btrfs_cleanup_transaction(struct btrfs_root *root)
3996{
3997 struct btrfs_transaction *t;
3998 LIST_HEAD(list);
3999
4000 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4001
4002 spin_lock(&root->fs_info->trans_lock);
4003 list_splice_init(&root->fs_info->trans_list, &list);
4004 root->fs_info->running_transaction = NULL;
4005 spin_unlock(&root->fs_info->trans_lock);
4006
4007 while (!list_empty(&list)) {
4008 t = list_entry(list.next, struct btrfs_transaction, list);
4009
4010 btrfs_destroy_ordered_operations(t, root);
4011
4012 btrfs_destroy_all_ordered_extents(root->fs_info);
4013
4014 btrfs_destroy_delayed_refs(t, root);
4015
4016 /*
4017 * FIXME: cleanup wait for commit
4018 * We needn't acquire the lock here, because we are during
4019 * the umount, there is no other task which will change it.
4020 */
4021 t->state = TRANS_STATE_COMMIT_START;
4022 smp_mb();
4023 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4024 wake_up(&root->fs_info->transaction_blocked_wait);
4025
4026 btrfs_evict_pending_snapshots(t);
4027
4028 t->state = TRANS_STATE_UNBLOCKED;
4029 smp_mb();
4030 if (waitqueue_active(&root->fs_info->transaction_wait))
4031 wake_up(&root->fs_info->transaction_wait);
4032
4033 btrfs_destroy_delayed_inodes(root);
4034 btrfs_assert_delayed_root_empty(root);
4035
4036 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4037
4038 btrfs_destroy_marked_extents(root, &t->dirty_pages,
4039 EXTENT_DIRTY);
4040
4041 btrfs_destroy_pinned_extent(root,
4042 root->fs_info->pinned_extents);
4043
4044 t->state = TRANS_STATE_COMPLETED;
4045 smp_mb();
4046 if (waitqueue_active(&t->commit_wait))
4047 wake_up(&t->commit_wait);
4048
4049 atomic_set(&t->use_count, 0);
4050 list_del_init(&t->list);
4051 memset(t, 0, sizeof(*t));
4052 kmem_cache_free(btrfs_transaction_cachep, t);
4053 }
4054
4055 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4056
4057 return 0;
4058}
4059
4060static struct extent_io_ops btree_extent_io_ops = {
4061 .readpage_end_io_hook = btree_readpage_end_io_hook,
4062 .readpage_io_failed_hook = btree_io_failed_hook,
4063 .submit_bio_hook = btree_submit_bio_hook,
4064 /* note we're sharing with inode.c for the merge bio hook */
4065 .merge_bio_hook = btrfs_merge_bio_hook,
4066};
Linux 6.x block layer and io_uring tree(s)