projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: split the global ordered extents mutex
[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/slab.h>
30#include <linux/migrate.h>
31#include <linux/ratelimit.h>
32#include <linux/uuid.h>
33#include <linux/semaphore.h>
34#include <asm/unaligned.h>
35#include "ctree.h"
36#include "disk-io.h"
37#include "hash.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#include "sysfs.h"
52
53#ifdef CONFIG_X86
54#include <asm/cpufeature.h>
55#endif
56
57static struct extent_io_ops btree_extent_io_ops;
58static void end_workqueue_fn(struct btrfs_work *work);
59static void free_fs_root(struct btrfs_root *root);
60static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
61 int read_only);
62static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
63 struct btrfs_root *root);
64static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
65static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
66 struct btrfs_root *root);
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 = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
160 { .id = 0, .name_stem = "tree" },
161};
162
163void __init btrfs_init_lockdep(void)
164{
165 int i, j;
166
167 /* initialize lockdep class names */
168 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
169 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
170
171 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
172 snprintf(ks->names[j], sizeof(ks->names[j]),
173 "btrfs-%s-%02d", ks->name_stem, j);
174 }
175}
176
177void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
178 int level)
179{
180 struct btrfs_lockdep_keyset *ks;
181
182 BUG_ON(level >= ARRAY_SIZE(ks->keys));
183
184 /* find the matching keyset, id 0 is the default entry */
185 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
186 if (ks->id == objectid)
187 break;
188
189 lockdep_set_class_and_name(&eb->lock,
190 &ks->keys[level], ks->names[level]);
191}
192
193#endif
194
195/*
196 * extents on the btree inode are pretty simple, there's one extent
197 * that covers the entire device
198 */
199static struct extent_map *btree_get_extent(struct inode *inode,
200 struct page *page, size_t pg_offset, u64 start, u64 len,
201 int create)
202{
203 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
204 struct extent_map *em;
205 int ret;
206
207 read_lock(&em_tree->lock);
208 em = lookup_extent_mapping(em_tree, start, len);
209 if (em) {
210 em->bdev =
211 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
212 read_unlock(&em_tree->lock);
213 goto out;
214 }
215 read_unlock(&em_tree->lock);
216
217 em = alloc_extent_map();
218 if (!em) {
219 em = ERR_PTR(-ENOMEM);
220 goto out;
221 }
222 em->start = 0;
223 em->len = (u64)-1;
224 em->block_len = (u64)-1;
225 em->block_start = 0;
226 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
227
228 write_lock(&em_tree->lock);
229 ret = add_extent_mapping(em_tree, em, 0);
230 if (ret == -EEXIST) {
231 free_extent_map(em);
232 em = lookup_extent_mapping(em_tree, start, len);
233 if (!em)
234 em = ERR_PTR(-EIO);
235 } else if (ret) {
236 free_extent_map(em);
237 em = ERR_PTR(ret);
238 }
239 write_unlock(&em_tree->lock);
240
241out:
242 return em;
243}
244
245u32 btrfs_csum_data(char *data, u32 seed, size_t len)
246{
247 return btrfs_crc32c(seed, data, len);
248}
249
250void btrfs_csum_final(u32 crc, char *result)
251{
252 put_unaligned_le32(~crc, result);
253}
254
255/*
256 * compute the csum for a btree block, and either verify it or write it
257 * into the csum field of the block.
258 */
259static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
260 int verify)
261{
262 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
263 char *result = NULL;
264 unsigned long len;
265 unsigned long cur_len;
266 unsigned long offset = BTRFS_CSUM_SIZE;
267 char *kaddr;
268 unsigned long map_start;
269 unsigned long map_len;
270 int err;
271 u32 crc = ~(u32)0;
272 unsigned long inline_result;
273
274 len = buf->len - offset;
275 while (len > 0) {
276 err = map_private_extent_buffer(buf, offset, 32,
277 &kaddr, &map_start, &map_len);
278 if (err)
279 return 1;
280 cur_len = min(len, map_len - (offset - map_start));
281 crc = btrfs_csum_data(kaddr + offset - map_start,
282 crc, cur_len);
283 len -= cur_len;
284 offset += cur_len;
285 }
286 if (csum_size > sizeof(inline_result)) {
287 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
288 if (!result)
289 return 1;
290 } else {
291 result = (char *)&inline_result;
292 }
293
294 btrfs_csum_final(crc, result);
295
296 if (verify) {
297 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
298 u32 val;
299 u32 found = 0;
300 memcpy(&found, result, csum_size);
301
302 read_extent_buffer(buf, &val, 0, csum_size);
303 printk_ratelimited(KERN_INFO
304 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
305 "level %d\n",
306 root->fs_info->sb->s_id, buf->start,
307 val, found, 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 eb->start, parent_transid, btrfs_header_generation(eb));
349 ret = 1;
350 clear_extent_buffer_uptodate(eb);
351out:
352 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
353 &cached_state, GFP_NOFS);
354 return ret;
355}
356
357/*
358 * Return 0 if the superblock checksum type matches the checksum value of that
359 * algorithm. Pass the raw disk superblock data.
360 */
361static int btrfs_check_super_csum(char *raw_disk_sb)
362{
363 struct btrfs_super_block *disk_sb =
364 (struct btrfs_super_block *)raw_disk_sb;
365 u16 csum_type = btrfs_super_csum_type(disk_sb);
366 int ret = 0;
367
368 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
369 u32 crc = ~(u32)0;
370 const int csum_size = sizeof(crc);
371 char result[csum_size];
372
373 /*
374 * The super_block structure does not span the whole
375 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
376 * is filled with zeros and is included in the checkum.
377 */
378 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
379 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
380 btrfs_csum_final(crc, result);
381
382 if (memcmp(raw_disk_sb, result, csum_size))
383 ret = 1;
384
385 if (ret && btrfs_super_generation(disk_sb) < 10) {
386 printk(KERN_WARNING
387 "BTRFS: super block crcs don't match, older mkfs detected\n");
388 ret = 0;
389 }
390 }
391
392 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
393 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
394 csum_type);
395 ret = 1;
396 }
397
398 return ret;
399}
400
401/*
402 * helper to read a given tree block, doing retries as required when
403 * the checksums don't match and we have alternate mirrors to try.
404 */
405static int btree_read_extent_buffer_pages(struct btrfs_root *root,
406 struct extent_buffer *eb,
407 u64 start, u64 parent_transid)
408{
409 struct extent_io_tree *io_tree;
410 int failed = 0;
411 int ret;
412 int num_copies = 0;
413 int mirror_num = 0;
414 int failed_mirror = 0;
415
416 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
417 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
418 while (1) {
419 ret = read_extent_buffer_pages(io_tree, eb, start,
420 WAIT_COMPLETE,
421 btree_get_extent, mirror_num);
422 if (!ret) {
423 if (!verify_parent_transid(io_tree, eb,
424 parent_transid, 0))
425 break;
426 else
427 ret = -EIO;
428 }
429
430 /*
431 * This buffer's crc is fine, but its contents are corrupted, so
432 * there is no reason to read the other copies, they won't be
433 * any less wrong.
434 */
435 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
436 break;
437
438 num_copies = btrfs_num_copies(root->fs_info,
439 eb->start, eb->len);
440 if (num_copies == 1)
441 break;
442
443 if (!failed_mirror) {
444 failed = 1;
445 failed_mirror = eb->read_mirror;
446 }
447
448 mirror_num++;
449 if (mirror_num == failed_mirror)
450 mirror_num++;
451
452 if (mirror_num > num_copies)
453 break;
454 }
455
456 if (failed && !ret && failed_mirror)
457 repair_eb_io_failure(root, eb, failed_mirror);
458
459 return ret;
460}
461
462/*
463 * checksum a dirty tree block before IO. This has extra checks to make sure
464 * we only fill in the checksum field in the first page of a multi-page block
465 */
466
467static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
468{
469 u64 start = page_offset(page);
470 u64 found_start;
471 struct extent_buffer *eb;
472
473 eb = (struct extent_buffer *)page->private;
474 if (page != eb->pages[0])
475 return 0;
476 found_start = btrfs_header_bytenr(eb);
477 if (WARN_ON(found_start != start || !PageUptodate(page)))
478 return 0;
479 csum_tree_block(root, eb, 0);
480 return 0;
481}
482
483static int check_tree_block_fsid(struct btrfs_root *root,
484 struct extent_buffer *eb)
485{
486 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
487 u8 fsid[BTRFS_UUID_SIZE];
488 int ret = 1;
489
490 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
491 while (fs_devices) {
492 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
493 ret = 0;
494 break;
495 }
496 fs_devices = fs_devices->seed;
497 }
498 return ret;
499}
500
501#define CORRUPT(reason, eb, root, slot) \
502 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
503 "root=%llu, slot=%d", reason, \
504 btrfs_header_bytenr(eb), root->objectid, slot)
505
506static noinline int check_leaf(struct btrfs_root *root,
507 struct extent_buffer *leaf)
508{
509 struct btrfs_key key;
510 struct btrfs_key leaf_key;
511 u32 nritems = btrfs_header_nritems(leaf);
512 int slot;
513
514 if (nritems == 0)
515 return 0;
516
517 /* Check the 0 item */
518 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
519 BTRFS_LEAF_DATA_SIZE(root)) {
520 CORRUPT("invalid item offset size pair", leaf, root, 0);
521 return -EIO;
522 }
523
524 /*
525 * Check to make sure each items keys are in the correct order and their
526 * offsets make sense. We only have to loop through nritems-1 because
527 * we check the current slot against the next slot, which verifies the
528 * next slot's offset+size makes sense and that the current's slot
529 * offset is correct.
530 */
531 for (slot = 0; slot < nritems - 1; slot++) {
532 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
533 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
534
535 /* Make sure the keys are in the right order */
536 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
537 CORRUPT("bad key order", leaf, root, slot);
538 return -EIO;
539 }
540
541 /*
542 * Make sure the offset and ends are right, remember that the
543 * item data starts at the end of the leaf and grows towards the
544 * front.
545 */
546 if (btrfs_item_offset_nr(leaf, slot) !=
547 btrfs_item_end_nr(leaf, slot + 1)) {
548 CORRUPT("slot offset bad", leaf, root, slot);
549 return -EIO;
550 }
551
552 /*
553 * Check to make sure that we don't point outside of the leaf,
554 * just incase all the items are consistent to eachother, but
555 * all point outside of the leaf.
556 */
557 if (btrfs_item_end_nr(leaf, slot) >
558 BTRFS_LEAF_DATA_SIZE(root)) {
559 CORRUPT("slot end outside of leaf", leaf, root, slot);
560 return -EIO;
561 }
562 }
563
564 return 0;
565}
566
567static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
568 u64 phy_offset, struct page *page,
569 u64 start, u64 end, int mirror)
570{
571 u64 found_start;
572 int found_level;
573 struct extent_buffer *eb;
574 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
575 int ret = 0;
576 int reads_done;
577
578 if (!page->private)
579 goto out;
580
581 eb = (struct extent_buffer *)page->private;
582
583 /* the pending IO might have been the only thing that kept this buffer
584 * in memory. Make sure we have a ref for all this other checks
585 */
586 extent_buffer_get(eb);
587
588 reads_done = atomic_dec_and_test(&eb->io_pages);
589 if (!reads_done)
590 goto err;
591
592 eb->read_mirror = mirror;
593 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
594 ret = -EIO;
595 goto err;
596 }
597
598 found_start = btrfs_header_bytenr(eb);
599 if (found_start != eb->start) {
600 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
601 "%llu %llu\n",
602 found_start, eb->start);
603 ret = -EIO;
604 goto err;
605 }
606 if (check_tree_block_fsid(root, eb)) {
607 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
608 eb->start);
609 ret = -EIO;
610 goto err;
611 }
612 found_level = btrfs_header_level(eb);
613 if (found_level >= BTRFS_MAX_LEVEL) {
614 btrfs_info(root->fs_info, "bad tree block level %d",
615 (int)btrfs_header_level(eb));
616 ret = -EIO;
617 goto err;
618 }
619
620 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
621 eb, found_level);
622
623 ret = csum_tree_block(root, eb, 1);
624 if (ret) {
625 ret = -EIO;
626 goto err;
627 }
628
629 /*
630 * If this is a leaf block and it is corrupt, set the corrupt bit so
631 * that we don't try and read the other copies of this block, just
632 * return -EIO.
633 */
634 if (found_level == 0 && check_leaf(root, eb)) {
635 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
636 ret = -EIO;
637 }
638
639 if (!ret)
640 set_extent_buffer_uptodate(eb);
641err:
642 if (reads_done &&
643 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
644 btree_readahead_hook(root, eb, eb->start, ret);
645
646 if (ret) {
647 /*
648 * our io error hook is going to dec the io pages
649 * again, we have to make sure it has something
650 * to decrement
651 */
652 atomic_inc(&eb->io_pages);
653 clear_extent_buffer_uptodate(eb);
654 }
655 free_extent_buffer(eb);
656out:
657 return ret;
658}
659
660static int btree_io_failed_hook(struct page *page, int failed_mirror)
661{
662 struct extent_buffer *eb;
663 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
664
665 eb = (struct extent_buffer *)page->private;
666 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
667 eb->read_mirror = failed_mirror;
668 atomic_dec(&eb->io_pages);
669 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
670 btree_readahead_hook(root, eb, eb->start, -EIO);
671 return -EIO; /* we fixed nothing */
672}
673
674static void end_workqueue_bio(struct bio *bio, int err)
675{
676 struct end_io_wq *end_io_wq = bio->bi_private;
677 struct btrfs_fs_info *fs_info;
678
679 fs_info = end_io_wq->info;
680 end_io_wq->error = err;
681 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
682
683 if (bio->bi_rw & REQ_WRITE) {
684 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
685 btrfs_queue_work(fs_info->endio_meta_write_workers,
686 &end_io_wq->work);
687 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
688 btrfs_queue_work(fs_info->endio_freespace_worker,
689 &end_io_wq->work);
690 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
691 btrfs_queue_work(fs_info->endio_raid56_workers,
692 &end_io_wq->work);
693 else
694 btrfs_queue_work(fs_info->endio_write_workers,
695 &end_io_wq->work);
696 } else {
697 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
698 btrfs_queue_work(fs_info->endio_raid56_workers,
699 &end_io_wq->work);
700 else if (end_io_wq->metadata)
701 btrfs_queue_work(fs_info->endio_meta_workers,
702 &end_io_wq->work);
703 else
704 btrfs_queue_work(fs_info->endio_workers,
705 &end_io_wq->work);
706 }
707}
708
709/*
710 * For the metadata arg you want
711 *
712 * 0 - if data
713 * 1 - if normal metadta
714 * 2 - if writing to the free space cache area
715 * 3 - raid parity work
716 */
717int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
718 int metadata)
719{
720 struct end_io_wq *end_io_wq;
721 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
722 if (!end_io_wq)
723 return -ENOMEM;
724
725 end_io_wq->private = bio->bi_private;
726 end_io_wq->end_io = bio->bi_end_io;
727 end_io_wq->info = info;
728 end_io_wq->error = 0;
729 end_io_wq->bio = bio;
730 end_io_wq->metadata = metadata;
731
732 bio->bi_private = end_io_wq;
733 bio->bi_end_io = end_workqueue_bio;
734 return 0;
735}
736
737unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
738{
739 unsigned long limit = min_t(unsigned long,
740 info->thread_pool_size,
741 info->fs_devices->open_devices);
742 return 256 * limit;
743}
744
745static void run_one_async_start(struct btrfs_work *work)
746{
747 struct async_submit_bio *async;
748 int ret;
749
750 async = container_of(work, struct async_submit_bio, work);
751 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
752 async->mirror_num, async->bio_flags,
753 async->bio_offset);
754 if (ret)
755 async->error = ret;
756}
757
758static void run_one_async_done(struct btrfs_work *work)
759{
760 struct btrfs_fs_info *fs_info;
761 struct async_submit_bio *async;
762 int limit;
763
764 async = container_of(work, struct async_submit_bio, work);
765 fs_info = BTRFS_I(async->inode)->root->fs_info;
766
767 limit = btrfs_async_submit_limit(fs_info);
768 limit = limit * 2 / 3;
769
770 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
771 waitqueue_active(&fs_info->async_submit_wait))
772 wake_up(&fs_info->async_submit_wait);
773
774 /* If an error occured we just want to clean up the bio and move on */
775 if (async->error) {
776 bio_endio(async->bio, async->error);
777 return;
778 }
779
780 async->submit_bio_done(async->inode, async->rw, async->bio,
781 async->mirror_num, async->bio_flags,
782 async->bio_offset);
783}
784
785static void run_one_async_free(struct btrfs_work *work)
786{
787 struct async_submit_bio *async;
788
789 async = container_of(work, struct async_submit_bio, work);
790 kfree(async);
791}
792
793int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
794 int rw, struct bio *bio, int mirror_num,
795 unsigned long bio_flags,
796 u64 bio_offset,
797 extent_submit_bio_hook_t *submit_bio_start,
798 extent_submit_bio_hook_t *submit_bio_done)
799{
800 struct async_submit_bio *async;
801
802 async = kmalloc(sizeof(*async), GFP_NOFS);
803 if (!async)
804 return -ENOMEM;
805
806 async->inode = inode;
807 async->rw = rw;
808 async->bio = bio;
809 async->mirror_num = mirror_num;
810 async->submit_bio_start = submit_bio_start;
811 async->submit_bio_done = submit_bio_done;
812
813 btrfs_init_work(&async->work, run_one_async_start,
814 run_one_async_done, run_one_async_free);
815
816 async->bio_flags = bio_flags;
817 async->bio_offset = bio_offset;
818
819 async->error = 0;
820
821 atomic_inc(&fs_info->nr_async_submits);
822
823 if (rw & REQ_SYNC)
824 btrfs_set_work_high_priority(&async->work);
825
826 btrfs_queue_work(fs_info->workers, &async->work);
827
828 while (atomic_read(&fs_info->async_submit_draining) &&
829 atomic_read(&fs_info->nr_async_submits)) {
830 wait_event(fs_info->async_submit_wait,
831 (atomic_read(&fs_info->nr_async_submits) == 0));
832 }
833
834 return 0;
835}
836
837static int btree_csum_one_bio(struct bio *bio)
838{
839 struct bio_vec *bvec = bio->bi_io_vec;
840 int bio_index = 0;
841 struct btrfs_root *root;
842 int ret = 0;
843
844 WARN_ON(bio->bi_vcnt <= 0);
845 while (bio_index < bio->bi_vcnt) {
846 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
847 ret = csum_dirty_buffer(root, bvec->bv_page);
848 if (ret)
849 break;
850 bio_index++;
851 bvec++;
852 }
853 return ret;
854}
855
856static int __btree_submit_bio_start(struct inode *inode, int rw,
857 struct bio *bio, int mirror_num,
858 unsigned long bio_flags,
859 u64 bio_offset)
860{
861 /*
862 * when we're called for a write, we're already in the async
863 * submission context. Just jump into btrfs_map_bio
864 */
865 return btree_csum_one_bio(bio);
866}
867
868static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
869 int mirror_num, unsigned long bio_flags,
870 u64 bio_offset)
871{
872 int ret;
873
874 /*
875 * when we're called for a write, we're already in the async
876 * submission context. Just jump into btrfs_map_bio
877 */
878 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
879 if (ret)
880 bio_endio(bio, ret);
881 return ret;
882}
883
884static int check_async_write(struct inode *inode, unsigned long bio_flags)
885{
886 if (bio_flags & EXTENT_BIO_TREE_LOG)
887 return 0;
888#ifdef CONFIG_X86
889 if (cpu_has_xmm4_2)
890 return 0;
891#endif
892 return 1;
893}
894
895static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
896 int mirror_num, unsigned long bio_flags,
897 u64 bio_offset)
898{
899 int async = check_async_write(inode, bio_flags);
900 int ret;
901
902 if (!(rw & REQ_WRITE)) {
903 /*
904 * called for a read, do the setup so that checksum validation
905 * can happen in the async kernel threads
906 */
907 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
908 bio, 1);
909 if (ret)
910 goto out_w_error;
911 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
912 mirror_num, 0);
913 } else if (!async) {
914 ret = btree_csum_one_bio(bio);
915 if (ret)
916 goto out_w_error;
917 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
918 mirror_num, 0);
919 } else {
920 /*
921 * kthread helpers are used to submit writes so that
922 * checksumming can happen in parallel across all CPUs
923 */
924 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
925 inode, rw, bio, mirror_num, 0,
926 bio_offset,
927 __btree_submit_bio_start,
928 __btree_submit_bio_done);
929 }
930
931 if (ret) {
932out_w_error:
933 bio_endio(bio, ret);
934 }
935 return ret;
936}
937
938#ifdef CONFIG_MIGRATION
939static int btree_migratepage(struct address_space *mapping,
940 struct page *newpage, struct page *page,
941 enum migrate_mode mode)
942{
943 /*
944 * we can't safely write a btree page from here,
945 * we haven't done the locking hook
946 */
947 if (PageDirty(page))
948 return -EAGAIN;
949 /*
950 * Buffers may be managed in a filesystem specific way.
951 * We must have no buffers or drop them.
952 */
953 if (page_has_private(page) &&
954 !try_to_release_page(page, GFP_KERNEL))
955 return -EAGAIN;
956 return migrate_page(mapping, newpage, page, mode);
957}
958#endif
959
960
961static int btree_writepages(struct address_space *mapping,
962 struct writeback_control *wbc)
963{
964 struct btrfs_fs_info *fs_info;
965 int ret;
966
967 if (wbc->sync_mode == WB_SYNC_NONE) {
968
969 if (wbc->for_kupdate)
970 return 0;
971
972 fs_info = BTRFS_I(mapping->host)->root->fs_info;
973 /* this is a bit racy, but that's ok */
974 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
975 BTRFS_DIRTY_METADATA_THRESH);
976 if (ret < 0)
977 return 0;
978 }
979 return btree_write_cache_pages(mapping, wbc);
980}
981
982static int btree_readpage(struct file *file, struct page *page)
983{
984 struct extent_io_tree *tree;
985 tree = &BTRFS_I(page->mapping->host)->io_tree;
986 return extent_read_full_page(tree, page, btree_get_extent, 0);
987}
988
989static int btree_releasepage(struct page *page, gfp_t gfp_flags)
990{
991 if (PageWriteback(page) || PageDirty(page))
992 return 0;
993
994 return try_release_extent_buffer(page);
995}
996
997static void btree_invalidatepage(struct page *page, unsigned int offset,
998 unsigned int length)
999{
1000 struct extent_io_tree *tree;
1001 tree = &BTRFS_I(page->mapping->host)->io_tree;
1002 extent_invalidatepage(tree, page, offset);
1003 btree_releasepage(page, GFP_NOFS);
1004 if (PagePrivate(page)) {
1005 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1006 "page private not zero on page %llu",
1007 (unsigned long long)page_offset(page));
1008 ClearPagePrivate(page);
1009 set_page_private(page, 0);
1010 page_cache_release(page);
1011 }
1012}
1013
1014static int btree_set_page_dirty(struct page *page)
1015{
1016#ifdef DEBUG
1017 struct extent_buffer *eb;
1018
1019 BUG_ON(!PagePrivate(page));
1020 eb = (struct extent_buffer *)page->private;
1021 BUG_ON(!eb);
1022 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1023 BUG_ON(!atomic_read(&eb->refs));
1024 btrfs_assert_tree_locked(eb);
1025#endif
1026 return __set_page_dirty_nobuffers(page);
1027}
1028
1029static const struct address_space_operations btree_aops = {
1030 .readpage = btree_readpage,
1031 .writepages = btree_writepages,
1032 .releasepage = btree_releasepage,
1033 .invalidatepage = btree_invalidatepage,
1034#ifdef CONFIG_MIGRATION
1035 .migratepage = btree_migratepage,
1036#endif
1037 .set_page_dirty = btree_set_page_dirty,
1038};
1039
1040int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1041 u64 parent_transid)
1042{
1043 struct extent_buffer *buf = NULL;
1044 struct inode *btree_inode = root->fs_info->btree_inode;
1045 int ret = 0;
1046
1047 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1048 if (!buf)
1049 return 0;
1050 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1051 buf, 0, WAIT_NONE, btree_get_extent, 0);
1052 free_extent_buffer(buf);
1053 return ret;
1054}
1055
1056int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1057 int mirror_num, struct extent_buffer **eb)
1058{
1059 struct extent_buffer *buf = NULL;
1060 struct inode *btree_inode = root->fs_info->btree_inode;
1061 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1062 int ret;
1063
1064 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1065 if (!buf)
1066 return 0;
1067
1068 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1069
1070 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1071 btree_get_extent, mirror_num);
1072 if (ret) {
1073 free_extent_buffer(buf);
1074 return ret;
1075 }
1076
1077 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1078 free_extent_buffer(buf);
1079 return -EIO;
1080 } else if (extent_buffer_uptodate(buf)) {
1081 *eb = buf;
1082 } else {
1083 free_extent_buffer(buf);
1084 }
1085 return 0;
1086}
1087
1088struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1089 u64 bytenr, u32 blocksize)
1090{
1091 return find_extent_buffer(root->fs_info, bytenr);
1092}
1093
1094struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1095 u64 bytenr, u32 blocksize)
1096{
1097 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1098}
1099
1100
1101int btrfs_write_tree_block(struct extent_buffer *buf)
1102{
1103 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1104 buf->start + buf->len - 1);
1105}
1106
1107int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1108{
1109 return filemap_fdatawait_range(buf->pages[0]->mapping,
1110 buf->start, buf->start + buf->len - 1);
1111}
1112
1113struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1114 u32 blocksize, u64 parent_transid)
1115{
1116 struct extent_buffer *buf = NULL;
1117 int ret;
1118
1119 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1120 if (!buf)
1121 return NULL;
1122
1123 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1124 if (ret) {
1125 free_extent_buffer(buf);
1126 return NULL;
1127 }
1128 return buf;
1129
1130}
1131
1132void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1133 struct extent_buffer *buf)
1134{
1135 struct btrfs_fs_info *fs_info = root->fs_info;
1136
1137 if (btrfs_header_generation(buf) ==
1138 fs_info->running_transaction->transid) {
1139 btrfs_assert_tree_locked(buf);
1140
1141 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1142 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1143 -buf->len,
1144 fs_info->dirty_metadata_batch);
1145 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1146 btrfs_set_lock_blocking(buf);
1147 clear_extent_buffer_dirty(buf);
1148 }
1149 }
1150}
1151
1152static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1153{
1154 struct btrfs_subvolume_writers *writers;
1155 int ret;
1156
1157 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1158 if (!writers)
1159 return ERR_PTR(-ENOMEM);
1160
1161 ret = percpu_counter_init(&writers->counter, 0);
1162 if (ret < 0) {
1163 kfree(writers);
1164 return ERR_PTR(ret);
1165 }
1166
1167 init_waitqueue_head(&writers->wait);
1168 return writers;
1169}
1170
1171static void
1172btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1173{
1174 percpu_counter_destroy(&writers->counter);
1175 kfree(writers);
1176}
1177
1178static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1179 u32 stripesize, struct btrfs_root *root,
1180 struct btrfs_fs_info *fs_info,
1181 u64 objectid)
1182{
1183 root->node = NULL;
1184 root->commit_root = NULL;
1185 root->sectorsize = sectorsize;
1186 root->nodesize = nodesize;
1187 root->leafsize = leafsize;
1188 root->stripesize = stripesize;
1189 root->ref_cows = 0;
1190 root->track_dirty = 0;
1191 root->in_radix = 0;
1192 root->orphan_item_inserted = 0;
1193 root->orphan_cleanup_state = 0;
1194
1195 root->objectid = objectid;
1196 root->last_trans = 0;
1197 root->highest_objectid = 0;
1198 root->nr_delalloc_inodes = 0;
1199 root->nr_ordered_extents = 0;
1200 root->name = NULL;
1201 root->inode_tree = RB_ROOT;
1202 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1203 root->block_rsv = NULL;
1204 root->orphan_block_rsv = NULL;
1205
1206 INIT_LIST_HEAD(&root->dirty_list);
1207 INIT_LIST_HEAD(&root->root_list);
1208 INIT_LIST_HEAD(&root->delalloc_inodes);
1209 INIT_LIST_HEAD(&root->delalloc_root);
1210 INIT_LIST_HEAD(&root->ordered_extents);
1211 INIT_LIST_HEAD(&root->ordered_root);
1212 INIT_LIST_HEAD(&root->logged_list[0]);
1213 INIT_LIST_HEAD(&root->logged_list[1]);
1214 spin_lock_init(&root->orphan_lock);
1215 spin_lock_init(&root->inode_lock);
1216 spin_lock_init(&root->delalloc_lock);
1217 spin_lock_init(&root->ordered_extent_lock);
1218 spin_lock_init(&root->accounting_lock);
1219 spin_lock_init(&root->log_extents_lock[0]);
1220 spin_lock_init(&root->log_extents_lock[1]);
1221 mutex_init(&root->objectid_mutex);
1222 mutex_init(&root->log_mutex);
1223 mutex_init(&root->ordered_extent_mutex);
1224 init_waitqueue_head(&root->log_writer_wait);
1225 init_waitqueue_head(&root->log_commit_wait[0]);
1226 init_waitqueue_head(&root->log_commit_wait[1]);
1227 INIT_LIST_HEAD(&root->log_ctxs[0]);
1228 INIT_LIST_HEAD(&root->log_ctxs[1]);
1229 atomic_set(&root->log_commit[0], 0);
1230 atomic_set(&root->log_commit[1], 0);
1231 atomic_set(&root->log_writers, 0);
1232 atomic_set(&root->log_batch, 0);
1233 atomic_set(&root->orphan_inodes, 0);
1234 atomic_set(&root->refs, 1);
1235 atomic_set(&root->will_be_snapshoted, 0);
1236 root->log_transid = 0;
1237 root->log_transid_committed = -1;
1238 root->last_log_commit = 0;
1239 if (fs_info)
1240 extent_io_tree_init(&root->dirty_log_pages,
1241 fs_info->btree_inode->i_mapping);
1242
1243 memset(&root->root_key, 0, sizeof(root->root_key));
1244 memset(&root->root_item, 0, sizeof(root->root_item));
1245 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1246 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1247 if (fs_info)
1248 root->defrag_trans_start = fs_info->generation;
1249 else
1250 root->defrag_trans_start = 0;
1251 init_completion(&root->kobj_unregister);
1252 root->defrag_running = 0;
1253 root->root_key.objectid = objectid;
1254 root->anon_dev = 0;
1255
1256 spin_lock_init(&root->root_item_lock);
1257}
1258
1259static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1260{
1261 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1262 if (root)
1263 root->fs_info = fs_info;
1264 return root;
1265}
1266
1267#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1268/* Should only be used by the testing infrastructure */
1269struct btrfs_root *btrfs_alloc_dummy_root(void)
1270{
1271 struct btrfs_root *root;
1272
1273 root = btrfs_alloc_root(NULL);
1274 if (!root)
1275 return ERR_PTR(-ENOMEM);
1276 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1277 root->dummy_root = 1;
1278
1279 return root;
1280}
1281#endif
1282
1283struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1284 struct btrfs_fs_info *fs_info,
1285 u64 objectid)
1286{
1287 struct extent_buffer *leaf;
1288 struct btrfs_root *tree_root = fs_info->tree_root;
1289 struct btrfs_root *root;
1290 struct btrfs_key key;
1291 int ret = 0;
1292 uuid_le uuid;
1293
1294 root = btrfs_alloc_root(fs_info);
1295 if (!root)
1296 return ERR_PTR(-ENOMEM);
1297
1298 __setup_root(tree_root->nodesize, tree_root->leafsize,
1299 tree_root->sectorsize, tree_root->stripesize,
1300 root, fs_info, objectid);
1301 root->root_key.objectid = objectid;
1302 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1303 root->root_key.offset = 0;
1304
1305 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1306 0, objectid, NULL, 0, 0, 0);
1307 if (IS_ERR(leaf)) {
1308 ret = PTR_ERR(leaf);
1309 leaf = NULL;
1310 goto fail;
1311 }
1312
1313 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1314 btrfs_set_header_bytenr(leaf, leaf->start);
1315 btrfs_set_header_generation(leaf, trans->transid);
1316 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1317 btrfs_set_header_owner(leaf, objectid);
1318 root->node = leaf;
1319
1320 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1321 BTRFS_FSID_SIZE);
1322 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1323 btrfs_header_chunk_tree_uuid(leaf),
1324 BTRFS_UUID_SIZE);
1325 btrfs_mark_buffer_dirty(leaf);
1326
1327 root->commit_root = btrfs_root_node(root);
1328 root->track_dirty = 1;
1329
1330
1331 root->root_item.flags = 0;
1332 root->root_item.byte_limit = 0;
1333 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1334 btrfs_set_root_generation(&root->root_item, trans->transid);
1335 btrfs_set_root_level(&root->root_item, 0);
1336 btrfs_set_root_refs(&root->root_item, 1);
1337 btrfs_set_root_used(&root->root_item, leaf->len);
1338 btrfs_set_root_last_snapshot(&root->root_item, 0);
1339 btrfs_set_root_dirid(&root->root_item, 0);
1340 uuid_le_gen(&uuid);
1341 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1342 root->root_item.drop_level = 0;
1343
1344 key.objectid = objectid;
1345 key.type = BTRFS_ROOT_ITEM_KEY;
1346 key.offset = 0;
1347 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1348 if (ret)
1349 goto fail;
1350
1351 btrfs_tree_unlock(leaf);
1352
1353 return root;
1354
1355fail:
1356 if (leaf) {
1357 btrfs_tree_unlock(leaf);
1358 free_extent_buffer(leaf);
1359 }
1360 kfree(root);
1361
1362 return ERR_PTR(ret);
1363}
1364
1365static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1366 struct btrfs_fs_info *fs_info)
1367{
1368 struct btrfs_root *root;
1369 struct btrfs_root *tree_root = fs_info->tree_root;
1370 struct extent_buffer *leaf;
1371
1372 root = btrfs_alloc_root(fs_info);
1373 if (!root)
1374 return ERR_PTR(-ENOMEM);
1375
1376 __setup_root(tree_root->nodesize, tree_root->leafsize,
1377 tree_root->sectorsize, tree_root->stripesize,
1378 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1379
1380 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1381 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1382 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1383 /*
1384 * log trees do not get reference counted because they go away
1385 * before a real commit is actually done. They do store pointers
1386 * to file data extents, and those reference counts still get
1387 * updated (along with back refs to the log tree).
1388 */
1389 root->ref_cows = 0;
1390
1391 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1392 BTRFS_TREE_LOG_OBJECTID, NULL,
1393 0, 0, 0);
1394 if (IS_ERR(leaf)) {
1395 kfree(root);
1396 return ERR_CAST(leaf);
1397 }
1398
1399 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1400 btrfs_set_header_bytenr(leaf, leaf->start);
1401 btrfs_set_header_generation(leaf, trans->transid);
1402 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1403 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1404 root->node = leaf;
1405
1406 write_extent_buffer(root->node, root->fs_info->fsid,
1407 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1408 btrfs_mark_buffer_dirty(root->node);
1409 btrfs_tree_unlock(root->node);
1410 return root;
1411}
1412
1413int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1414 struct btrfs_fs_info *fs_info)
1415{
1416 struct btrfs_root *log_root;
1417
1418 log_root = alloc_log_tree(trans, fs_info);
1419 if (IS_ERR(log_root))
1420 return PTR_ERR(log_root);
1421 WARN_ON(fs_info->log_root_tree);
1422 fs_info->log_root_tree = log_root;
1423 return 0;
1424}
1425
1426int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1427 struct btrfs_root *root)
1428{
1429 struct btrfs_root *log_root;
1430 struct btrfs_inode_item *inode_item;
1431
1432 log_root = alloc_log_tree(trans, root->fs_info);
1433 if (IS_ERR(log_root))
1434 return PTR_ERR(log_root);
1435
1436 log_root->last_trans = trans->transid;
1437 log_root->root_key.offset = root->root_key.objectid;
1438
1439 inode_item = &log_root->root_item.inode;
1440 btrfs_set_stack_inode_generation(inode_item, 1);
1441 btrfs_set_stack_inode_size(inode_item, 3);
1442 btrfs_set_stack_inode_nlink(inode_item, 1);
1443 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1444 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1445
1446 btrfs_set_root_node(&log_root->root_item, log_root->node);
1447
1448 WARN_ON(root->log_root);
1449 root->log_root = log_root;
1450 root->log_transid = 0;
1451 root->log_transid_committed = -1;
1452 root->last_log_commit = 0;
1453 return 0;
1454}
1455
1456static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1457 struct btrfs_key *key)
1458{
1459 struct btrfs_root *root;
1460 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1461 struct btrfs_path *path;
1462 u64 generation;
1463 u32 blocksize;
1464 int ret;
1465
1466 path = btrfs_alloc_path();
1467 if (!path)
1468 return ERR_PTR(-ENOMEM);
1469
1470 root = btrfs_alloc_root(fs_info);
1471 if (!root) {
1472 ret = -ENOMEM;
1473 goto alloc_fail;
1474 }
1475
1476 __setup_root(tree_root->nodesize, tree_root->leafsize,
1477 tree_root->sectorsize, tree_root->stripesize,
1478 root, fs_info, key->objectid);
1479
1480 ret = btrfs_find_root(tree_root, key, path,
1481 &root->root_item, &root->root_key);
1482 if (ret) {
1483 if (ret > 0)
1484 ret = -ENOENT;
1485 goto find_fail;
1486 }
1487
1488 generation = btrfs_root_generation(&root->root_item);
1489 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1490 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1491 blocksize, generation);
1492 if (!root->node) {
1493 ret = -ENOMEM;
1494 goto find_fail;
1495 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1496 ret = -EIO;
1497 goto read_fail;
1498 }
1499 root->commit_root = btrfs_root_node(root);
1500out:
1501 btrfs_free_path(path);
1502 return root;
1503
1504read_fail:
1505 free_extent_buffer(root->node);
1506find_fail:
1507 kfree(root);
1508alloc_fail:
1509 root = ERR_PTR(ret);
1510 goto out;
1511}
1512
1513struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1514 struct btrfs_key *location)
1515{
1516 struct btrfs_root *root;
1517
1518 root = btrfs_read_tree_root(tree_root, location);
1519 if (IS_ERR(root))
1520 return root;
1521
1522 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1523 root->ref_cows = 1;
1524 btrfs_check_and_init_root_item(&root->root_item);
1525 }
1526
1527 return root;
1528}
1529
1530int btrfs_init_fs_root(struct btrfs_root *root)
1531{
1532 int ret;
1533 struct btrfs_subvolume_writers *writers;
1534
1535 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1536 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1537 GFP_NOFS);
1538 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1539 ret = -ENOMEM;
1540 goto fail;
1541 }
1542
1543 writers = btrfs_alloc_subvolume_writers();
1544 if (IS_ERR(writers)) {
1545 ret = PTR_ERR(writers);
1546 goto fail;
1547 }
1548 root->subv_writers = writers;
1549
1550 btrfs_init_free_ino_ctl(root);
1551 mutex_init(&root->fs_commit_mutex);
1552 spin_lock_init(&root->cache_lock);
1553 init_waitqueue_head(&root->cache_wait);
1554
1555 ret = get_anon_bdev(&root->anon_dev);
1556 if (ret)
1557 goto free_writers;
1558 return 0;
1559
1560free_writers:
1561 btrfs_free_subvolume_writers(root->subv_writers);
1562fail:
1563 kfree(root->free_ino_ctl);
1564 kfree(root->free_ino_pinned);
1565 return ret;
1566}
1567
1568static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1569 u64 root_id)
1570{
1571 struct btrfs_root *root;
1572
1573 spin_lock(&fs_info->fs_roots_radix_lock);
1574 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1575 (unsigned long)root_id);
1576 spin_unlock(&fs_info->fs_roots_radix_lock);
1577 return root;
1578}
1579
1580int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1581 struct btrfs_root *root)
1582{
1583 int ret;
1584
1585 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1586 if (ret)
1587 return ret;
1588
1589 spin_lock(&fs_info->fs_roots_radix_lock);
1590 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1591 (unsigned long)root->root_key.objectid,
1592 root);
1593 if (ret == 0)
1594 root->in_radix = 1;
1595 spin_unlock(&fs_info->fs_roots_radix_lock);
1596 radix_tree_preload_end();
1597
1598 return ret;
1599}
1600
1601struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1602 struct btrfs_key *location,
1603 bool check_ref)
1604{
1605 struct btrfs_root *root;
1606 int ret;
1607
1608 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1609 return fs_info->tree_root;
1610 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1611 return fs_info->extent_root;
1612 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1613 return fs_info->chunk_root;
1614 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1615 return fs_info->dev_root;
1616 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1617 return fs_info->csum_root;
1618 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1619 return fs_info->quota_root ? fs_info->quota_root :
1620 ERR_PTR(-ENOENT);
1621 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1622 return fs_info->uuid_root ? fs_info->uuid_root :
1623 ERR_PTR(-ENOENT);
1624again:
1625 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1626 if (root) {
1627 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1628 return ERR_PTR(-ENOENT);
1629 return root;
1630 }
1631
1632 root = btrfs_read_fs_root(fs_info->tree_root, location);
1633 if (IS_ERR(root))
1634 return root;
1635
1636 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1637 ret = -ENOENT;
1638 goto fail;
1639 }
1640
1641 ret = btrfs_init_fs_root(root);
1642 if (ret)
1643 goto fail;
1644
1645 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1646 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1647 if (ret < 0)
1648 goto fail;
1649 if (ret == 0)
1650 root->orphan_item_inserted = 1;
1651
1652 ret = btrfs_insert_fs_root(fs_info, root);
1653 if (ret) {
1654 if (ret == -EEXIST) {
1655 free_fs_root(root);
1656 goto again;
1657 }
1658 goto fail;
1659 }
1660 return root;
1661fail:
1662 free_fs_root(root);
1663 return ERR_PTR(ret);
1664}
1665
1666static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1667{
1668 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1669 int ret = 0;
1670 struct btrfs_device *device;
1671 struct backing_dev_info *bdi;
1672
1673 rcu_read_lock();
1674 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1675 if (!device->bdev)
1676 continue;
1677 bdi = blk_get_backing_dev_info(device->bdev);
1678 if (bdi && bdi_congested(bdi, bdi_bits)) {
1679 ret = 1;
1680 break;
1681 }
1682 }
1683 rcu_read_unlock();
1684 return ret;
1685}
1686
1687/*
1688 * If this fails, caller must call bdi_destroy() to get rid of the
1689 * bdi again.
1690 */
1691static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1692{
1693 int err;
1694
1695 bdi->capabilities = BDI_CAP_MAP_COPY;
1696 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1697 if (err)
1698 return err;
1699
1700 bdi->ra_pages = default_backing_dev_info.ra_pages;
1701 bdi->congested_fn = btrfs_congested_fn;
1702 bdi->congested_data = info;
1703 return 0;
1704}
1705
1706/*
1707 * called by the kthread helper functions to finally call the bio end_io
1708 * functions. This is where read checksum verification actually happens
1709 */
1710static void end_workqueue_fn(struct btrfs_work *work)
1711{
1712 struct bio *bio;
1713 struct end_io_wq *end_io_wq;
1714 int error;
1715
1716 end_io_wq = container_of(work, struct end_io_wq, work);
1717 bio = end_io_wq->bio;
1718
1719 error = end_io_wq->error;
1720 bio->bi_private = end_io_wq->private;
1721 bio->bi_end_io = end_io_wq->end_io;
1722 kfree(end_io_wq);
1723 bio_endio(bio, error);
1724}
1725
1726static int cleaner_kthread(void *arg)
1727{
1728 struct btrfs_root *root = arg;
1729 int again;
1730
1731 do {
1732 again = 0;
1733
1734 /* Make the cleaner go to sleep early. */
1735 if (btrfs_need_cleaner_sleep(root))
1736 goto sleep;
1737
1738 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1739 goto sleep;
1740
1741 /*
1742 * Avoid the problem that we change the status of the fs
1743 * during the above check and trylock.
1744 */
1745 if (btrfs_need_cleaner_sleep(root)) {
1746 mutex_unlock(&root->fs_info->cleaner_mutex);
1747 goto sleep;
1748 }
1749
1750 btrfs_run_delayed_iputs(root);
1751 again = btrfs_clean_one_deleted_snapshot(root);
1752 mutex_unlock(&root->fs_info->cleaner_mutex);
1753
1754 /*
1755 * The defragger has dealt with the R/O remount and umount,
1756 * needn't do anything special here.
1757 */
1758 btrfs_run_defrag_inodes(root->fs_info);
1759sleep:
1760 if (!try_to_freeze() && !again) {
1761 set_current_state(TASK_INTERRUPTIBLE);
1762 if (!kthread_should_stop())
1763 schedule();
1764 __set_current_state(TASK_RUNNING);
1765 }
1766 } while (!kthread_should_stop());
1767 return 0;
1768}
1769
1770static int transaction_kthread(void *arg)
1771{
1772 struct btrfs_root *root = arg;
1773 struct btrfs_trans_handle *trans;
1774 struct btrfs_transaction *cur;
1775 u64 transid;
1776 unsigned long now;
1777 unsigned long delay;
1778 bool cannot_commit;
1779
1780 do {
1781 cannot_commit = false;
1782 delay = HZ * root->fs_info->commit_interval;
1783 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1784
1785 spin_lock(&root->fs_info->trans_lock);
1786 cur = root->fs_info->running_transaction;
1787 if (!cur) {
1788 spin_unlock(&root->fs_info->trans_lock);
1789 goto sleep;
1790 }
1791
1792 now = get_seconds();
1793 if (cur->state < TRANS_STATE_BLOCKED &&
1794 (now < cur->start_time ||
1795 now - cur->start_time < root->fs_info->commit_interval)) {
1796 spin_unlock(&root->fs_info->trans_lock);
1797 delay = HZ * 5;
1798 goto sleep;
1799 }
1800 transid = cur->transid;
1801 spin_unlock(&root->fs_info->trans_lock);
1802
1803 /* If the file system is aborted, this will always fail. */
1804 trans = btrfs_attach_transaction(root);
1805 if (IS_ERR(trans)) {
1806 if (PTR_ERR(trans) != -ENOENT)
1807 cannot_commit = true;
1808 goto sleep;
1809 }
1810 if (transid == trans->transid) {
1811 btrfs_commit_transaction(trans, root);
1812 } else {
1813 btrfs_end_transaction(trans, root);
1814 }
1815sleep:
1816 wake_up_process(root->fs_info->cleaner_kthread);
1817 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1818
1819 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1820 &root->fs_info->fs_state)))
1821 btrfs_cleanup_transaction(root);
1822 if (!try_to_freeze()) {
1823 set_current_state(TASK_INTERRUPTIBLE);
1824 if (!kthread_should_stop() &&
1825 (!btrfs_transaction_blocked(root->fs_info) ||
1826 cannot_commit))
1827 schedule_timeout(delay);
1828 __set_current_state(TASK_RUNNING);
1829 }
1830 } while (!kthread_should_stop());
1831 return 0;
1832}
1833
1834/*
1835 * this will find the highest generation in the array of
1836 * root backups. The index of the highest array is returned,
1837 * or -1 if we can't find anything.
1838 *
1839 * We check to make sure the array is valid by comparing the
1840 * generation of the latest root in the array with the generation
1841 * in the super block. If they don't match we pitch it.
1842 */
1843static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1844{
1845 u64 cur;
1846 int newest_index = -1;
1847 struct btrfs_root_backup *root_backup;
1848 int i;
1849
1850 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1851 root_backup = info->super_copy->super_roots + i;
1852 cur = btrfs_backup_tree_root_gen(root_backup);
1853 if (cur == newest_gen)
1854 newest_index = i;
1855 }
1856
1857 /* check to see if we actually wrapped around */
1858 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1859 root_backup = info->super_copy->super_roots;
1860 cur = btrfs_backup_tree_root_gen(root_backup);
1861 if (cur == newest_gen)
1862 newest_index = 0;
1863 }
1864 return newest_index;
1865}
1866
1867
1868/*
1869 * find the oldest backup so we know where to store new entries
1870 * in the backup array. This will set the backup_root_index
1871 * field in the fs_info struct
1872 */
1873static void find_oldest_super_backup(struct btrfs_fs_info *info,
1874 u64 newest_gen)
1875{
1876 int newest_index = -1;
1877
1878 newest_index = find_newest_super_backup(info, newest_gen);
1879 /* if there was garbage in there, just move along */
1880 if (newest_index == -1) {
1881 info->backup_root_index = 0;
1882 } else {
1883 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1884 }
1885}
1886
1887/*
1888 * copy all the root pointers into the super backup array.
1889 * this will bump the backup pointer by one when it is
1890 * done
1891 */
1892static void backup_super_roots(struct btrfs_fs_info *info)
1893{
1894 int next_backup;
1895 struct btrfs_root_backup *root_backup;
1896 int last_backup;
1897
1898 next_backup = info->backup_root_index;
1899 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1900 BTRFS_NUM_BACKUP_ROOTS;
1901
1902 /*
1903 * just overwrite the last backup if we're at the same generation
1904 * this happens only at umount
1905 */
1906 root_backup = info->super_for_commit->super_roots + last_backup;
1907 if (btrfs_backup_tree_root_gen(root_backup) ==
1908 btrfs_header_generation(info->tree_root->node))
1909 next_backup = last_backup;
1910
1911 root_backup = info->super_for_commit->super_roots + next_backup;
1912
1913 /*
1914 * make sure all of our padding and empty slots get zero filled
1915 * regardless of which ones we use today
1916 */
1917 memset(root_backup, 0, sizeof(*root_backup));
1918
1919 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1920
1921 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1922 btrfs_set_backup_tree_root_gen(root_backup,
1923 btrfs_header_generation(info->tree_root->node));
1924
1925 btrfs_set_backup_tree_root_level(root_backup,
1926 btrfs_header_level(info->tree_root->node));
1927
1928 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1929 btrfs_set_backup_chunk_root_gen(root_backup,
1930 btrfs_header_generation(info->chunk_root->node));
1931 btrfs_set_backup_chunk_root_level(root_backup,
1932 btrfs_header_level(info->chunk_root->node));
1933
1934 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1935 btrfs_set_backup_extent_root_gen(root_backup,
1936 btrfs_header_generation(info->extent_root->node));
1937 btrfs_set_backup_extent_root_level(root_backup,
1938 btrfs_header_level(info->extent_root->node));
1939
1940 /*
1941 * we might commit during log recovery, which happens before we set
1942 * the fs_root. Make sure it is valid before we fill it in.
1943 */
1944 if (info->fs_root && info->fs_root->node) {
1945 btrfs_set_backup_fs_root(root_backup,
1946 info->fs_root->node->start);
1947 btrfs_set_backup_fs_root_gen(root_backup,
1948 btrfs_header_generation(info->fs_root->node));
1949 btrfs_set_backup_fs_root_level(root_backup,
1950 btrfs_header_level(info->fs_root->node));
1951 }
1952
1953 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1954 btrfs_set_backup_dev_root_gen(root_backup,
1955 btrfs_header_generation(info->dev_root->node));
1956 btrfs_set_backup_dev_root_level(root_backup,
1957 btrfs_header_level(info->dev_root->node));
1958
1959 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1960 btrfs_set_backup_csum_root_gen(root_backup,
1961 btrfs_header_generation(info->csum_root->node));
1962 btrfs_set_backup_csum_root_level(root_backup,
1963 btrfs_header_level(info->csum_root->node));
1964
1965 btrfs_set_backup_total_bytes(root_backup,
1966 btrfs_super_total_bytes(info->super_copy));
1967 btrfs_set_backup_bytes_used(root_backup,
1968 btrfs_super_bytes_used(info->super_copy));
1969 btrfs_set_backup_num_devices(root_backup,
1970 btrfs_super_num_devices(info->super_copy));
1971
1972 /*
1973 * if we don't copy this out to the super_copy, it won't get remembered
1974 * for the next commit
1975 */
1976 memcpy(&info->super_copy->super_roots,
1977 &info->super_for_commit->super_roots,
1978 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1979}
1980
1981/*
1982 * this copies info out of the root backup array and back into
1983 * the in-memory super block. It is meant to help iterate through
1984 * the array, so you send it the number of backups you've already
1985 * tried and the last backup index you used.
1986 *
1987 * this returns -1 when it has tried all the backups
1988 */
1989static noinline int next_root_backup(struct btrfs_fs_info *info,
1990 struct btrfs_super_block *super,
1991 int *num_backups_tried, int *backup_index)
1992{
1993 struct btrfs_root_backup *root_backup;
1994 int newest = *backup_index;
1995
1996 if (*num_backups_tried == 0) {
1997 u64 gen = btrfs_super_generation(super);
1998
1999 newest = find_newest_super_backup(info, gen);
2000 if (newest == -1)
2001 return -1;
2002
2003 *backup_index = newest;
2004 *num_backups_tried = 1;
2005 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2006 /* we've tried all the backups, all done */
2007 return -1;
2008 } else {
2009 /* jump to the next oldest backup */
2010 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2011 BTRFS_NUM_BACKUP_ROOTS;
2012 *backup_index = newest;
2013 *num_backups_tried += 1;
2014 }
2015 root_backup = super->super_roots + newest;
2016
2017 btrfs_set_super_generation(super,
2018 btrfs_backup_tree_root_gen(root_backup));
2019 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2020 btrfs_set_super_root_level(super,
2021 btrfs_backup_tree_root_level(root_backup));
2022 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2023
2024 /*
2025 * fixme: the total bytes and num_devices need to match or we should
2026 * need a fsck
2027 */
2028 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2029 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2030 return 0;
2031}
2032
2033/* helper to cleanup workers */
2034static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2035{
2036 btrfs_destroy_workqueue(fs_info->fixup_workers);
2037 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2038 btrfs_destroy_workqueue(fs_info->workers);
2039 btrfs_destroy_workqueue(fs_info->endio_workers);
2040 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2041 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2042 btrfs_destroy_workqueue(fs_info->rmw_workers);
2043 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2044 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2045 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2046 btrfs_destroy_workqueue(fs_info->submit_workers);
2047 btrfs_destroy_workqueue(fs_info->delayed_workers);
2048 btrfs_destroy_workqueue(fs_info->caching_workers);
2049 btrfs_destroy_workqueue(fs_info->readahead_workers);
2050 btrfs_destroy_workqueue(fs_info->flush_workers);
2051 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2052}
2053
2054static void free_root_extent_buffers(struct btrfs_root *root)
2055{
2056 if (root) {
2057 free_extent_buffer(root->node);
2058 free_extent_buffer(root->commit_root);
2059 root->node = NULL;
2060 root->commit_root = NULL;
2061 }
2062}
2063
2064/* helper to cleanup tree roots */
2065static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2066{
2067 free_root_extent_buffers(info->tree_root);
2068
2069 free_root_extent_buffers(info->dev_root);
2070 free_root_extent_buffers(info->extent_root);
2071 free_root_extent_buffers(info->csum_root);
2072 free_root_extent_buffers(info->quota_root);
2073 free_root_extent_buffers(info->uuid_root);
2074 if (chunk_root)
2075 free_root_extent_buffers(info->chunk_root);
2076}
2077
2078static void del_fs_roots(struct btrfs_fs_info *fs_info)
2079{
2080 int ret;
2081 struct btrfs_root *gang[8];
2082 int i;
2083
2084 while (!list_empty(&fs_info->dead_roots)) {
2085 gang[0] = list_entry(fs_info->dead_roots.next,
2086 struct btrfs_root, root_list);
2087 list_del(&gang[0]->root_list);
2088
2089 if (gang[0]->in_radix) {
2090 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2091 } else {
2092 free_extent_buffer(gang[0]->node);
2093 free_extent_buffer(gang[0]->commit_root);
2094 btrfs_put_fs_root(gang[0]);
2095 }
2096 }
2097
2098 while (1) {
2099 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2100 (void **)gang, 0,
2101 ARRAY_SIZE(gang));
2102 if (!ret)
2103 break;
2104 for (i = 0; i < ret; i++)
2105 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2106 }
2107
2108 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2109 btrfs_free_log_root_tree(NULL, fs_info);
2110 btrfs_destroy_pinned_extent(fs_info->tree_root,
2111 fs_info->pinned_extents);
2112 }
2113}
2114
2115int open_ctree(struct super_block *sb,
2116 struct btrfs_fs_devices *fs_devices,
2117 char *options)
2118{
2119 u32 sectorsize;
2120 u32 nodesize;
2121 u32 leafsize;
2122 u32 blocksize;
2123 u32 stripesize;
2124 u64 generation;
2125 u64 features;
2126 struct btrfs_key location;
2127 struct buffer_head *bh;
2128 struct btrfs_super_block *disk_super;
2129 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2130 struct btrfs_root *tree_root;
2131 struct btrfs_root *extent_root;
2132 struct btrfs_root *csum_root;
2133 struct btrfs_root *chunk_root;
2134 struct btrfs_root *dev_root;
2135 struct btrfs_root *quota_root;
2136 struct btrfs_root *uuid_root;
2137 struct btrfs_root *log_tree_root;
2138 int ret;
2139 int err = -EINVAL;
2140 int num_backups_tried = 0;
2141 int backup_index = 0;
2142 int max_active;
2143 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2144 bool create_uuid_tree;
2145 bool check_uuid_tree;
2146
2147 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2148 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2149 if (!tree_root || !chunk_root) {
2150 err = -ENOMEM;
2151 goto fail;
2152 }
2153
2154 ret = init_srcu_struct(&fs_info->subvol_srcu);
2155 if (ret) {
2156 err = ret;
2157 goto fail;
2158 }
2159
2160 ret = setup_bdi(fs_info, &fs_info->bdi);
2161 if (ret) {
2162 err = ret;
2163 goto fail_srcu;
2164 }
2165
2166 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2167 if (ret) {
2168 err = ret;
2169 goto fail_bdi;
2170 }
2171 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2172 (1 + ilog2(nr_cpu_ids));
2173
2174 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2175 if (ret) {
2176 err = ret;
2177 goto fail_dirty_metadata_bytes;
2178 }
2179
2180 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2181 if (ret) {
2182 err = ret;
2183 goto fail_delalloc_bytes;
2184 }
2185
2186 fs_info->btree_inode = new_inode(sb);
2187 if (!fs_info->btree_inode) {
2188 err = -ENOMEM;
2189 goto fail_bio_counter;
2190 }
2191
2192 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2193
2194 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2195 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2196 INIT_LIST_HEAD(&fs_info->trans_list);
2197 INIT_LIST_HEAD(&fs_info->dead_roots);
2198 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2199 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2200 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2201 spin_lock_init(&fs_info->delalloc_root_lock);
2202 spin_lock_init(&fs_info->trans_lock);
2203 spin_lock_init(&fs_info->fs_roots_radix_lock);
2204 spin_lock_init(&fs_info->delayed_iput_lock);
2205 spin_lock_init(&fs_info->defrag_inodes_lock);
2206 spin_lock_init(&fs_info->free_chunk_lock);
2207 spin_lock_init(&fs_info->tree_mod_seq_lock);
2208 spin_lock_init(&fs_info->super_lock);
2209 spin_lock_init(&fs_info->buffer_lock);
2210 rwlock_init(&fs_info->tree_mod_log_lock);
2211 mutex_init(&fs_info->reloc_mutex);
2212 seqlock_init(&fs_info->profiles_lock);
2213
2214 init_completion(&fs_info->kobj_unregister);
2215 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2216 INIT_LIST_HEAD(&fs_info->space_info);
2217 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2218 btrfs_mapping_init(&fs_info->mapping_tree);
2219 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2220 BTRFS_BLOCK_RSV_GLOBAL);
2221 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2222 BTRFS_BLOCK_RSV_DELALLOC);
2223 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2224 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2225 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2226 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2227 BTRFS_BLOCK_RSV_DELOPS);
2228 atomic_set(&fs_info->nr_async_submits, 0);
2229 atomic_set(&fs_info->async_delalloc_pages, 0);
2230 atomic_set(&fs_info->async_submit_draining, 0);
2231 atomic_set(&fs_info->nr_async_bios, 0);
2232 atomic_set(&fs_info->defrag_running, 0);
2233 atomic64_set(&fs_info->tree_mod_seq, 0);
2234 fs_info->sb = sb;
2235 fs_info->max_inline = 8192 * 1024;
2236 fs_info->metadata_ratio = 0;
2237 fs_info->defrag_inodes = RB_ROOT;
2238 fs_info->free_chunk_space = 0;
2239 fs_info->tree_mod_log = RB_ROOT;
2240 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2241 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2242 /* readahead state */
2243 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2244 spin_lock_init(&fs_info->reada_lock);
2245
2246 fs_info->thread_pool_size = min_t(unsigned long,
2247 num_online_cpus() + 2, 8);
2248
2249 INIT_LIST_HEAD(&fs_info->ordered_roots);
2250 spin_lock_init(&fs_info->ordered_root_lock);
2251 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2252 GFP_NOFS);
2253 if (!fs_info->delayed_root) {
2254 err = -ENOMEM;
2255 goto fail_iput;
2256 }
2257 btrfs_init_delayed_root(fs_info->delayed_root);
2258
2259 mutex_init(&fs_info->scrub_lock);
2260 atomic_set(&fs_info->scrubs_running, 0);
2261 atomic_set(&fs_info->scrub_pause_req, 0);
2262 atomic_set(&fs_info->scrubs_paused, 0);
2263 atomic_set(&fs_info->scrub_cancel_req, 0);
2264 init_waitqueue_head(&fs_info->replace_wait);
2265 init_waitqueue_head(&fs_info->scrub_pause_wait);
2266 fs_info->scrub_workers_refcnt = 0;
2267#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2268 fs_info->check_integrity_print_mask = 0;
2269#endif
2270
2271 spin_lock_init(&fs_info->balance_lock);
2272 mutex_init(&fs_info->balance_mutex);
2273 atomic_set(&fs_info->balance_running, 0);
2274 atomic_set(&fs_info->balance_pause_req, 0);
2275 atomic_set(&fs_info->balance_cancel_req, 0);
2276 fs_info->balance_ctl = NULL;
2277 init_waitqueue_head(&fs_info->balance_wait_q);
2278
2279 sb->s_blocksize = 4096;
2280 sb->s_blocksize_bits = blksize_bits(4096);
2281 sb->s_bdi = &fs_info->bdi;
2282
2283 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2284 set_nlink(fs_info->btree_inode, 1);
2285 /*
2286 * we set the i_size on the btree inode to the max possible int.
2287 * the real end of the address space is determined by all of
2288 * the devices in the system
2289 */
2290 fs_info->btree_inode->i_size = OFFSET_MAX;
2291 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2292 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2293
2294 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2295 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2296 fs_info->btree_inode->i_mapping);
2297 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2298 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2299
2300 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2301
2302 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2303 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2304 sizeof(struct btrfs_key));
2305 set_bit(BTRFS_INODE_DUMMY,
2306 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2307 btrfs_insert_inode_hash(fs_info->btree_inode);
2308
2309 spin_lock_init(&fs_info->block_group_cache_lock);
2310 fs_info->block_group_cache_tree = RB_ROOT;
2311 fs_info->first_logical_byte = (u64)-1;
2312
2313 extent_io_tree_init(&fs_info->freed_extents[0],
2314 fs_info->btree_inode->i_mapping);
2315 extent_io_tree_init(&fs_info->freed_extents[1],
2316 fs_info->btree_inode->i_mapping);
2317 fs_info->pinned_extents = &fs_info->freed_extents[0];
2318 fs_info->do_barriers = 1;
2319
2320
2321 mutex_init(&fs_info->ordered_operations_mutex);
2322 mutex_init(&fs_info->ordered_extent_flush_mutex);
2323 mutex_init(&fs_info->tree_log_mutex);
2324 mutex_init(&fs_info->chunk_mutex);
2325 mutex_init(&fs_info->transaction_kthread_mutex);
2326 mutex_init(&fs_info->cleaner_mutex);
2327 mutex_init(&fs_info->volume_mutex);
2328 init_rwsem(&fs_info->extent_commit_sem);
2329 init_rwsem(&fs_info->cleanup_work_sem);
2330 init_rwsem(&fs_info->subvol_sem);
2331 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2332 fs_info->dev_replace.lock_owner = 0;
2333 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2334 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2335 mutex_init(&fs_info->dev_replace.lock_management_lock);
2336 mutex_init(&fs_info->dev_replace.lock);
2337
2338 spin_lock_init(&fs_info->qgroup_lock);
2339 mutex_init(&fs_info->qgroup_ioctl_lock);
2340 fs_info->qgroup_tree = RB_ROOT;
2341 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2342 fs_info->qgroup_seq = 1;
2343 fs_info->quota_enabled = 0;
2344 fs_info->pending_quota_state = 0;
2345 fs_info->qgroup_ulist = NULL;
2346 mutex_init(&fs_info->qgroup_rescan_lock);
2347
2348 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2349 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2350
2351 init_waitqueue_head(&fs_info->transaction_throttle);
2352 init_waitqueue_head(&fs_info->transaction_wait);
2353 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2354 init_waitqueue_head(&fs_info->async_submit_wait);
2355
2356 ret = btrfs_alloc_stripe_hash_table(fs_info);
2357 if (ret) {
2358 err = ret;
2359 goto fail_alloc;
2360 }
2361
2362 __setup_root(4096, 4096, 4096, 4096, tree_root,
2363 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2364
2365 invalidate_bdev(fs_devices->latest_bdev);
2366
2367 /*
2368 * Read super block and check the signature bytes only
2369 */
2370 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2371 if (!bh) {
2372 err = -EINVAL;
2373 goto fail_alloc;
2374 }
2375
2376 /*
2377 * We want to check superblock checksum, the type is stored inside.
2378 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2379 */
2380 if (btrfs_check_super_csum(bh->b_data)) {
2381 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2382 err = -EINVAL;
2383 goto fail_alloc;
2384 }
2385
2386 /*
2387 * super_copy is zeroed at allocation time and we never touch the
2388 * following bytes up to INFO_SIZE, the checksum is calculated from
2389 * the whole block of INFO_SIZE
2390 */
2391 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2392 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2393 sizeof(*fs_info->super_for_commit));
2394 brelse(bh);
2395
2396 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2397
2398 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2399 if (ret) {
2400 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2401 err = -EINVAL;
2402 goto fail_alloc;
2403 }
2404
2405 disk_super = fs_info->super_copy;
2406 if (!btrfs_super_root(disk_super))
2407 goto fail_alloc;
2408
2409 /* check FS state, whether FS is broken. */
2410 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2411 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2412
2413 /*
2414 * run through our array of backup supers and setup
2415 * our ring pointer to the oldest one
2416 */
2417 generation = btrfs_super_generation(disk_super);
2418 find_oldest_super_backup(fs_info, generation);
2419
2420 /*
2421 * In the long term, we'll store the compression type in the super
2422 * block, and it'll be used for per file compression control.
2423 */
2424 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2425
2426 ret = btrfs_parse_options(tree_root, options);
2427 if (ret) {
2428 err = ret;
2429 goto fail_alloc;
2430 }
2431
2432 features = btrfs_super_incompat_flags(disk_super) &
2433 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2434 if (features) {
2435 printk(KERN_ERR "BTRFS: couldn't mount because of "
2436 "unsupported optional features (%Lx).\n",
2437 features);
2438 err = -EINVAL;
2439 goto fail_alloc;
2440 }
2441
2442 if (btrfs_super_leafsize(disk_super) !=
2443 btrfs_super_nodesize(disk_super)) {
2444 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2445 "blocksizes don't match. node %d leaf %d\n",
2446 btrfs_super_nodesize(disk_super),
2447 btrfs_super_leafsize(disk_super));
2448 err = -EINVAL;
2449 goto fail_alloc;
2450 }
2451 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2452 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2453 "blocksize (%d) was too large\n",
2454 btrfs_super_leafsize(disk_super));
2455 err = -EINVAL;
2456 goto fail_alloc;
2457 }
2458
2459 features = btrfs_super_incompat_flags(disk_super);
2460 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2461 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2462 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2463
2464 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2465 printk(KERN_ERR "BTRFS: has skinny extents\n");
2466
2467 /*
2468 * flag our filesystem as having big metadata blocks if
2469 * they are bigger than the page size
2470 */
2471 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2472 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2473 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2474 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2475 }
2476
2477 nodesize = btrfs_super_nodesize(disk_super);
2478 leafsize = btrfs_super_leafsize(disk_super);
2479 sectorsize = btrfs_super_sectorsize(disk_super);
2480 stripesize = btrfs_super_stripesize(disk_super);
2481 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2482 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2483
2484 /*
2485 * mixed block groups end up with duplicate but slightly offset
2486 * extent buffers for the same range. It leads to corruptions
2487 */
2488 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2489 (sectorsize != leafsize)) {
2490 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2491 "are not allowed for mixed block groups on %s\n",
2492 sb->s_id);
2493 goto fail_alloc;
2494 }
2495
2496 /*
2497 * Needn't use the lock because there is no other task which will
2498 * update the flag.
2499 */
2500 btrfs_set_super_incompat_flags(disk_super, features);
2501
2502 features = btrfs_super_compat_ro_flags(disk_super) &
2503 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2504 if (!(sb->s_flags & MS_RDONLY) && features) {
2505 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2506 "unsupported option features (%Lx).\n",
2507 features);
2508 err = -EINVAL;
2509 goto fail_alloc;
2510 }
2511
2512 max_active = fs_info->thread_pool_size;
2513
2514 fs_info->workers =
2515 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2516 max_active, 16);
2517
2518 fs_info->delalloc_workers =
2519 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2520
2521 fs_info->flush_workers =
2522 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2523
2524 fs_info->caching_workers =
2525 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2526
2527 /*
2528 * a higher idle thresh on the submit workers makes it much more
2529 * likely that bios will be send down in a sane order to the
2530 * devices
2531 */
2532 fs_info->submit_workers =
2533 btrfs_alloc_workqueue("submit", flags,
2534 min_t(u64, fs_devices->num_devices,
2535 max_active), 64);
2536
2537 fs_info->fixup_workers =
2538 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2539
2540 /*
2541 * endios are largely parallel and should have a very
2542 * low idle thresh
2543 */
2544 fs_info->endio_workers =
2545 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2546 fs_info->endio_meta_workers =
2547 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2548 fs_info->endio_meta_write_workers =
2549 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2550 fs_info->endio_raid56_workers =
2551 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2552 fs_info->rmw_workers =
2553 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2554 fs_info->endio_write_workers =
2555 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2556 fs_info->endio_freespace_worker =
2557 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2558 fs_info->delayed_workers =
2559 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2560 fs_info->readahead_workers =
2561 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2562 fs_info->qgroup_rescan_workers =
2563 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2564
2565 if (!(fs_info->workers && fs_info->delalloc_workers &&
2566 fs_info->submit_workers && fs_info->flush_workers &&
2567 fs_info->endio_workers && fs_info->endio_meta_workers &&
2568 fs_info->endio_meta_write_workers &&
2569 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2570 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2571 fs_info->caching_workers && fs_info->readahead_workers &&
2572 fs_info->fixup_workers && fs_info->delayed_workers &&
2573 fs_info->qgroup_rescan_workers)) {
2574 err = -ENOMEM;
2575 goto fail_sb_buffer;
2576 }
2577
2578 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2579 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2580 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2581
2582 tree_root->nodesize = nodesize;
2583 tree_root->leafsize = leafsize;
2584 tree_root->sectorsize = sectorsize;
2585 tree_root->stripesize = stripesize;
2586
2587 sb->s_blocksize = sectorsize;
2588 sb->s_blocksize_bits = blksize_bits(sectorsize);
2589
2590 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2591 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2592 goto fail_sb_buffer;
2593 }
2594
2595 if (sectorsize != PAGE_SIZE) {
2596 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2597 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2598 goto fail_sb_buffer;
2599 }
2600
2601 mutex_lock(&fs_info->chunk_mutex);
2602 ret = btrfs_read_sys_array(tree_root);
2603 mutex_unlock(&fs_info->chunk_mutex);
2604 if (ret) {
2605 printk(KERN_WARNING "BTRFS: failed to read the system "
2606 "array on %s\n", sb->s_id);
2607 goto fail_sb_buffer;
2608 }
2609
2610 blocksize = btrfs_level_size(tree_root,
2611 btrfs_super_chunk_root_level(disk_super));
2612 generation = btrfs_super_chunk_root_generation(disk_super);
2613
2614 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2615 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2616
2617 chunk_root->node = read_tree_block(chunk_root,
2618 btrfs_super_chunk_root(disk_super),
2619 blocksize, generation);
2620 if (!chunk_root->node ||
2621 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2622 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2623 sb->s_id);
2624 goto fail_tree_roots;
2625 }
2626 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2627 chunk_root->commit_root = btrfs_root_node(chunk_root);
2628
2629 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2630 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2631
2632 ret = btrfs_read_chunk_tree(chunk_root);
2633 if (ret) {
2634 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2635 sb->s_id);
2636 goto fail_tree_roots;
2637 }
2638
2639 /*
2640 * keep the device that is marked to be the target device for the
2641 * dev_replace procedure
2642 */
2643 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2644
2645 if (!fs_devices->latest_bdev) {
2646 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2647 sb->s_id);
2648 goto fail_tree_roots;
2649 }
2650
2651retry_root_backup:
2652 blocksize = btrfs_level_size(tree_root,
2653 btrfs_super_root_level(disk_super));
2654 generation = btrfs_super_generation(disk_super);
2655
2656 tree_root->node = read_tree_block(tree_root,
2657 btrfs_super_root(disk_super),
2658 blocksize, generation);
2659 if (!tree_root->node ||
2660 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2661 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2662 sb->s_id);
2663
2664 goto recovery_tree_root;
2665 }
2666
2667 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2668 tree_root->commit_root = btrfs_root_node(tree_root);
2669 btrfs_set_root_refs(&tree_root->root_item, 1);
2670
2671 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2672 location.type = BTRFS_ROOT_ITEM_KEY;
2673 location.offset = 0;
2674
2675 extent_root = btrfs_read_tree_root(tree_root, &location);
2676 if (IS_ERR(extent_root)) {
2677 ret = PTR_ERR(extent_root);
2678 goto recovery_tree_root;
2679 }
2680 extent_root->track_dirty = 1;
2681 fs_info->extent_root = extent_root;
2682
2683 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2684 dev_root = btrfs_read_tree_root(tree_root, &location);
2685 if (IS_ERR(dev_root)) {
2686 ret = PTR_ERR(dev_root);
2687 goto recovery_tree_root;
2688 }
2689 dev_root->track_dirty = 1;
2690 fs_info->dev_root = dev_root;
2691 btrfs_init_devices_late(fs_info);
2692
2693 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2694 csum_root = btrfs_read_tree_root(tree_root, &location);
2695 if (IS_ERR(csum_root)) {
2696 ret = PTR_ERR(csum_root);
2697 goto recovery_tree_root;
2698 }
2699 csum_root->track_dirty = 1;
2700 fs_info->csum_root = csum_root;
2701
2702 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2703 quota_root = btrfs_read_tree_root(tree_root, &location);
2704 if (!IS_ERR(quota_root)) {
2705 quota_root->track_dirty = 1;
2706 fs_info->quota_enabled = 1;
2707 fs_info->pending_quota_state = 1;
2708 fs_info->quota_root = quota_root;
2709 }
2710
2711 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2712 uuid_root = btrfs_read_tree_root(tree_root, &location);
2713 if (IS_ERR(uuid_root)) {
2714 ret = PTR_ERR(uuid_root);
2715 if (ret != -ENOENT)
2716 goto recovery_tree_root;
2717 create_uuid_tree = true;
2718 check_uuid_tree = false;
2719 } else {
2720 uuid_root->track_dirty = 1;
2721 fs_info->uuid_root = uuid_root;
2722 create_uuid_tree = false;
2723 check_uuid_tree =
2724 generation != btrfs_super_uuid_tree_generation(disk_super);
2725 }
2726
2727 fs_info->generation = generation;
2728 fs_info->last_trans_committed = generation;
2729
2730 ret = btrfs_recover_balance(fs_info);
2731 if (ret) {
2732 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2733 goto fail_block_groups;
2734 }
2735
2736 ret = btrfs_init_dev_stats(fs_info);
2737 if (ret) {
2738 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2739 ret);
2740 goto fail_block_groups;
2741 }
2742
2743 ret = btrfs_init_dev_replace(fs_info);
2744 if (ret) {
2745 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2746 goto fail_block_groups;
2747 }
2748
2749 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2750
2751 ret = btrfs_sysfs_add_one(fs_info);
2752 if (ret) {
2753 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2754 goto fail_block_groups;
2755 }
2756
2757 ret = btrfs_init_space_info(fs_info);
2758 if (ret) {
2759 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2760 goto fail_sysfs;
2761 }
2762
2763 ret = btrfs_read_block_groups(extent_root);
2764 if (ret) {
2765 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2766 goto fail_sysfs;
2767 }
2768 fs_info->num_tolerated_disk_barrier_failures =
2769 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2770 if (fs_info->fs_devices->missing_devices >
2771 fs_info->num_tolerated_disk_barrier_failures &&
2772 !(sb->s_flags & MS_RDONLY)) {
2773 printk(KERN_WARNING "BTRFS: "
2774 "too many missing devices, writeable mount is not allowed\n");
2775 goto fail_sysfs;
2776 }
2777
2778 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2779 "btrfs-cleaner");
2780 if (IS_ERR(fs_info->cleaner_kthread))
2781 goto fail_sysfs;
2782
2783 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2784 tree_root,
2785 "btrfs-transaction");
2786 if (IS_ERR(fs_info->transaction_kthread))
2787 goto fail_cleaner;
2788
2789 if (!btrfs_test_opt(tree_root, SSD) &&
2790 !btrfs_test_opt(tree_root, NOSSD) &&
2791 !fs_info->fs_devices->rotating) {
2792 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2793 "mode\n");
2794 btrfs_set_opt(fs_info->mount_opt, SSD);
2795 }
2796
2797 /* Set the real inode map cache flag */
2798 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2799 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2800
2801#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2802 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2803 ret = btrfsic_mount(tree_root, fs_devices,
2804 btrfs_test_opt(tree_root,
2805 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2806 1 : 0,
2807 fs_info->check_integrity_print_mask);
2808 if (ret)
2809 printk(KERN_WARNING "BTRFS: failed to initialize"
2810 " integrity check module %s\n", sb->s_id);
2811 }
2812#endif
2813 ret = btrfs_read_qgroup_config(fs_info);
2814 if (ret)
2815 goto fail_trans_kthread;
2816
2817 /* do not make disk changes in broken FS */
2818 if (btrfs_super_log_root(disk_super) != 0) {
2819 u64 bytenr = btrfs_super_log_root(disk_super);
2820
2821 if (fs_devices->rw_devices == 0) {
2822 printk(KERN_WARNING "BTRFS: log replay required "
2823 "on RO media\n");
2824 err = -EIO;
2825 goto fail_qgroup;
2826 }
2827 blocksize =
2828 btrfs_level_size(tree_root,
2829 btrfs_super_log_root_level(disk_super));
2830
2831 log_tree_root = btrfs_alloc_root(fs_info);
2832 if (!log_tree_root) {
2833 err = -ENOMEM;
2834 goto fail_qgroup;
2835 }
2836
2837 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2838 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2839
2840 log_tree_root->node = read_tree_block(tree_root, bytenr,
2841 blocksize,
2842 generation + 1);
2843 if (!log_tree_root->node ||
2844 !extent_buffer_uptodate(log_tree_root->node)) {
2845 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2846 free_extent_buffer(log_tree_root->node);
2847 kfree(log_tree_root);
2848 goto fail_trans_kthread;
2849 }
2850 /* returns with log_tree_root freed on success */
2851 ret = btrfs_recover_log_trees(log_tree_root);
2852 if (ret) {
2853 btrfs_error(tree_root->fs_info, ret,
2854 "Failed to recover log tree");
2855 free_extent_buffer(log_tree_root->node);
2856 kfree(log_tree_root);
2857 goto fail_trans_kthread;
2858 }
2859
2860 if (sb->s_flags & MS_RDONLY) {
2861 ret = btrfs_commit_super(tree_root);
2862 if (ret)
2863 goto fail_trans_kthread;
2864 }
2865 }
2866
2867 ret = btrfs_find_orphan_roots(tree_root);
2868 if (ret)
2869 goto fail_trans_kthread;
2870
2871 if (!(sb->s_flags & MS_RDONLY)) {
2872 ret = btrfs_cleanup_fs_roots(fs_info);
2873 if (ret)
2874 goto fail_trans_kthread;
2875
2876 ret = btrfs_recover_relocation(tree_root);
2877 if (ret < 0) {
2878 printk(KERN_WARNING
2879 "BTRFS: failed to recover relocation\n");
2880 err = -EINVAL;
2881 goto fail_qgroup;
2882 }
2883 }
2884
2885 location.objectid = BTRFS_FS_TREE_OBJECTID;
2886 location.type = BTRFS_ROOT_ITEM_KEY;
2887 location.offset = 0;
2888
2889 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2890 if (IS_ERR(fs_info->fs_root)) {
2891 err = PTR_ERR(fs_info->fs_root);
2892 goto fail_qgroup;
2893 }
2894
2895 if (sb->s_flags & MS_RDONLY)
2896 return 0;
2897
2898 down_read(&fs_info->cleanup_work_sem);
2899 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2900 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2901 up_read(&fs_info->cleanup_work_sem);
2902 close_ctree(tree_root);
2903 return ret;
2904 }
2905 up_read(&fs_info->cleanup_work_sem);
2906
2907 ret = btrfs_resume_balance_async(fs_info);
2908 if (ret) {
2909 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2910 close_ctree(tree_root);
2911 return ret;
2912 }
2913
2914 ret = btrfs_resume_dev_replace_async(fs_info);
2915 if (ret) {
2916 pr_warn("BTRFS: failed to resume dev_replace\n");
2917 close_ctree(tree_root);
2918 return ret;
2919 }
2920
2921 btrfs_qgroup_rescan_resume(fs_info);
2922
2923 if (create_uuid_tree) {
2924 pr_info("BTRFS: creating UUID tree\n");
2925 ret = btrfs_create_uuid_tree(fs_info);
2926 if (ret) {
2927 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2928 ret);
2929 close_ctree(tree_root);
2930 return ret;
2931 }
2932 } else if (check_uuid_tree ||
2933 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2934 pr_info("BTRFS: checking UUID tree\n");
2935 ret = btrfs_check_uuid_tree(fs_info);
2936 if (ret) {
2937 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2938 ret);
2939 close_ctree(tree_root);
2940 return ret;
2941 }
2942 } else {
2943 fs_info->update_uuid_tree_gen = 1;
2944 }
2945
2946 return 0;
2947
2948fail_qgroup:
2949 btrfs_free_qgroup_config(fs_info);
2950fail_trans_kthread:
2951 kthread_stop(fs_info->transaction_kthread);
2952 btrfs_cleanup_transaction(fs_info->tree_root);
2953 del_fs_roots(fs_info);
2954fail_cleaner:
2955 kthread_stop(fs_info->cleaner_kthread);
2956
2957 /*
2958 * make sure we're done with the btree inode before we stop our
2959 * kthreads
2960 */
2961 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2962
2963fail_sysfs:
2964 btrfs_sysfs_remove_one(fs_info);
2965
2966fail_block_groups:
2967 btrfs_put_block_group_cache(fs_info);
2968 btrfs_free_block_groups(fs_info);
2969
2970fail_tree_roots:
2971 free_root_pointers(fs_info, 1);
2972 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2973
2974fail_sb_buffer:
2975 btrfs_stop_all_workers(fs_info);
2976fail_alloc:
2977fail_iput:
2978 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2979
2980 iput(fs_info->btree_inode);
2981fail_bio_counter:
2982 percpu_counter_destroy(&fs_info->bio_counter);
2983fail_delalloc_bytes:
2984 percpu_counter_destroy(&fs_info->delalloc_bytes);
2985fail_dirty_metadata_bytes:
2986 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2987fail_bdi:
2988 bdi_destroy(&fs_info->bdi);
2989fail_srcu:
2990 cleanup_srcu_struct(&fs_info->subvol_srcu);
2991fail:
2992 btrfs_free_stripe_hash_table(fs_info);
2993 btrfs_close_devices(fs_info->fs_devices);
2994 return err;
2995
2996recovery_tree_root:
2997 if (!btrfs_test_opt(tree_root, RECOVERY))
2998 goto fail_tree_roots;
2999
3000 free_root_pointers(fs_info, 0);
3001
3002 /* don't use the log in recovery mode, it won't be valid */
3003 btrfs_set_super_log_root(disk_super, 0);
3004
3005 /* we can't trust the free space cache either */
3006 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3007
3008 ret = next_root_backup(fs_info, fs_info->super_copy,
3009 &num_backups_tried, &backup_index);
3010 if (ret == -1)
3011 goto fail_block_groups;
3012 goto retry_root_backup;
3013}
3014
3015static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3016{
3017 if (uptodate) {
3018 set_buffer_uptodate(bh);
3019 } else {
3020 struct btrfs_device *device = (struct btrfs_device *)
3021 bh->b_private;
3022
3023 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3024 "I/O error on %s\n",
3025 rcu_str_deref(device->name));
3026 /* note, we dont' set_buffer_write_io_error because we have
3027 * our own ways of dealing with the IO errors
3028 */
3029 clear_buffer_uptodate(bh);
3030 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3031 }
3032 unlock_buffer(bh);
3033 put_bh(bh);
3034}
3035
3036struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3037{
3038 struct buffer_head *bh;
3039 struct buffer_head *latest = NULL;
3040 struct btrfs_super_block *super;
3041 int i;
3042 u64 transid = 0;
3043 u64 bytenr;
3044
3045 /* we would like to check all the supers, but that would make
3046 * a btrfs mount succeed after a mkfs from a different FS.
3047 * So, we need to add a special mount option to scan for
3048 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3049 */
3050 for (i = 0; i < 1; i++) {
3051 bytenr = btrfs_sb_offset(i);
3052 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3053 i_size_read(bdev->bd_inode))
3054 break;
3055 bh = __bread(bdev, bytenr / 4096,
3056 BTRFS_SUPER_INFO_SIZE);
3057 if (!bh)
3058 continue;
3059
3060 super = (struct btrfs_super_block *)bh->b_data;
3061 if (btrfs_super_bytenr(super) != bytenr ||
3062 btrfs_super_magic(super) != BTRFS_MAGIC) {
3063 brelse(bh);
3064 continue;
3065 }
3066
3067 if (!latest || btrfs_super_generation(super) > transid) {
3068 brelse(latest);
3069 latest = bh;
3070 transid = btrfs_super_generation(super);
3071 } else {
3072 brelse(bh);
3073 }
3074 }
3075 return latest;
3076}
3077
3078/*
3079 * this should be called twice, once with wait == 0 and
3080 * once with wait == 1. When wait == 0 is done, all the buffer heads
3081 * we write are pinned.
3082 *
3083 * They are released when wait == 1 is done.
3084 * max_mirrors must be the same for both runs, and it indicates how
3085 * many supers on this one device should be written.
3086 *
3087 * max_mirrors == 0 means to write them all.
3088 */
3089static int write_dev_supers(struct btrfs_device *device,
3090 struct btrfs_super_block *sb,
3091 int do_barriers, int wait, int max_mirrors)
3092{
3093 struct buffer_head *bh;
3094 int i;
3095 int ret;
3096 int errors = 0;
3097 u32 crc;
3098 u64 bytenr;
3099
3100 if (max_mirrors == 0)
3101 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3102
3103 for (i = 0; i < max_mirrors; i++) {
3104 bytenr = btrfs_sb_offset(i);
3105 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3106 break;
3107
3108 if (wait) {
3109 bh = __find_get_block(device->bdev, bytenr / 4096,
3110 BTRFS_SUPER_INFO_SIZE);
3111 if (!bh) {
3112 errors++;
3113 continue;
3114 }
3115 wait_on_buffer(bh);
3116 if (!buffer_uptodate(bh))
3117 errors++;
3118
3119 /* drop our reference */
3120 brelse(bh);
3121
3122 /* drop the reference from the wait == 0 run */
3123 brelse(bh);
3124 continue;
3125 } else {
3126 btrfs_set_super_bytenr(sb, bytenr);
3127
3128 crc = ~(u32)0;
3129 crc = btrfs_csum_data((char *)sb +
3130 BTRFS_CSUM_SIZE, crc,
3131 BTRFS_SUPER_INFO_SIZE -
3132 BTRFS_CSUM_SIZE);
3133 btrfs_csum_final(crc, sb->csum);
3134
3135 /*
3136 * one reference for us, and we leave it for the
3137 * caller
3138 */
3139 bh = __getblk(device->bdev, bytenr / 4096,
3140 BTRFS_SUPER_INFO_SIZE);
3141 if (!bh) {
3142 printk(KERN_ERR "BTRFS: couldn't get super "
3143 "buffer head for bytenr %Lu\n", bytenr);
3144 errors++;
3145 continue;
3146 }
3147
3148 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3149
3150 /* one reference for submit_bh */
3151 get_bh(bh);
3152
3153 set_buffer_uptodate(bh);
3154 lock_buffer(bh);
3155 bh->b_end_io = btrfs_end_buffer_write_sync;
3156 bh->b_private = device;
3157 }
3158
3159 /*
3160 * we fua the first super. The others we allow
3161 * to go down lazy.
3162 */
3163 if (i == 0)
3164 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3165 else
3166 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3167 if (ret)
3168 errors++;
3169 }
3170 return errors < i ? 0 : -1;
3171}
3172
3173/*
3174 * endio for the write_dev_flush, this will wake anyone waiting
3175 * for the barrier when it is done
3176 */
3177static void btrfs_end_empty_barrier(struct bio *bio, int err)
3178{
3179 if (err) {
3180 if (err == -EOPNOTSUPP)
3181 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3182 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3183 }
3184 if (bio->bi_private)
3185 complete(bio->bi_private);
3186 bio_put(bio);
3187}
3188
3189/*
3190 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3191 * sent down. With wait == 1, it waits for the previous flush.
3192 *
3193 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3194 * capable
3195 */
3196static int write_dev_flush(struct btrfs_device *device, int wait)
3197{
3198 struct bio *bio;
3199 int ret = 0;
3200
3201 if (device->nobarriers)
3202 return 0;
3203
3204 if (wait) {
3205 bio = device->flush_bio;
3206 if (!bio)
3207 return 0;
3208
3209 wait_for_completion(&device->flush_wait);
3210
3211 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3212 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3213 rcu_str_deref(device->name));
3214 device->nobarriers = 1;
3215 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3216 ret = -EIO;
3217 btrfs_dev_stat_inc_and_print(device,
3218 BTRFS_DEV_STAT_FLUSH_ERRS);
3219 }
3220
3221 /* drop the reference from the wait == 0 run */
3222 bio_put(bio);
3223 device->flush_bio = NULL;
3224
3225 return ret;
3226 }
3227
3228 /*
3229 * one reference for us, and we leave it for the
3230 * caller
3231 */
3232 device->flush_bio = NULL;
3233 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3234 if (!bio)
3235 return -ENOMEM;
3236
3237 bio->bi_end_io = btrfs_end_empty_barrier;
3238 bio->bi_bdev = device->bdev;
3239 init_completion(&device->flush_wait);
3240 bio->bi_private = &device->flush_wait;
3241 device->flush_bio = bio;
3242
3243 bio_get(bio);
3244 btrfsic_submit_bio(WRITE_FLUSH, bio);
3245
3246 return 0;
3247}
3248
3249/*
3250 * send an empty flush down to each device in parallel,
3251 * then wait for them
3252 */
3253static int barrier_all_devices(struct btrfs_fs_info *info)
3254{
3255 struct list_head *head;
3256 struct btrfs_device *dev;
3257 int errors_send = 0;
3258 int errors_wait = 0;
3259 int ret;
3260
3261 /* send down all the barriers */
3262 head = &info->fs_devices->devices;
3263 list_for_each_entry_rcu(dev, head, dev_list) {
3264 if (dev->missing)
3265 continue;
3266 if (!dev->bdev) {
3267 errors_send++;
3268 continue;
3269 }
3270 if (!dev->in_fs_metadata || !dev->writeable)
3271 continue;
3272
3273 ret = write_dev_flush(dev, 0);
3274 if (ret)
3275 errors_send++;
3276 }
3277
3278 /* wait for all the barriers */
3279 list_for_each_entry_rcu(dev, head, dev_list) {
3280 if (dev->missing)
3281 continue;
3282 if (!dev->bdev) {
3283 errors_wait++;
3284 continue;
3285 }
3286 if (!dev->in_fs_metadata || !dev->writeable)
3287 continue;
3288
3289 ret = write_dev_flush(dev, 1);
3290 if (ret)
3291 errors_wait++;
3292 }
3293 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3294 errors_wait > info->num_tolerated_disk_barrier_failures)
3295 return -EIO;
3296 return 0;
3297}
3298
3299int btrfs_calc_num_tolerated_disk_barrier_failures(
3300 struct btrfs_fs_info *fs_info)
3301{
3302 struct btrfs_ioctl_space_info space;
3303 struct btrfs_space_info *sinfo;
3304 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3305 BTRFS_BLOCK_GROUP_SYSTEM,
3306 BTRFS_BLOCK_GROUP_METADATA,
3307 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3308 int num_types = 4;
3309 int i;
3310 int c;
3311 int num_tolerated_disk_barrier_failures =
3312 (int)fs_info->fs_devices->num_devices;
3313
3314 for (i = 0; i < num_types; i++) {
3315 struct btrfs_space_info *tmp;
3316
3317 sinfo = NULL;
3318 rcu_read_lock();
3319 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3320 if (tmp->flags == types[i]) {
3321 sinfo = tmp;
3322 break;
3323 }
3324 }
3325 rcu_read_unlock();
3326
3327 if (!sinfo)
3328 continue;
3329
3330 down_read(&sinfo->groups_sem);
3331 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3332 if (!list_empty(&sinfo->block_groups[c])) {
3333 u64 flags;
3334
3335 btrfs_get_block_group_info(
3336 &sinfo->block_groups[c], &space);
3337 if (space.total_bytes == 0 ||
3338 space.used_bytes == 0)
3339 continue;
3340 flags = space.flags;
3341 /*
3342 * return
3343 * 0: if dup, single or RAID0 is configured for
3344 * any of metadata, system or data, else
3345 * 1: if RAID5 is configured, or if RAID1 or
3346 * RAID10 is configured and only two mirrors
3347 * are used, else
3348 * 2: if RAID6 is configured, else
3349 * num_mirrors - 1: if RAID1 or RAID10 is
3350 * configured and more than
3351 * 2 mirrors are used.
3352 */
3353 if (num_tolerated_disk_barrier_failures > 0 &&
3354 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3355 BTRFS_BLOCK_GROUP_RAID0)) ||
3356 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3357 == 0)))
3358 num_tolerated_disk_barrier_failures = 0;
3359 else if (num_tolerated_disk_barrier_failures > 1) {
3360 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3361 BTRFS_BLOCK_GROUP_RAID5 |
3362 BTRFS_BLOCK_GROUP_RAID10)) {
3363 num_tolerated_disk_barrier_failures = 1;
3364 } else if (flags &
3365 BTRFS_BLOCK_GROUP_RAID6) {
3366 num_tolerated_disk_barrier_failures = 2;
3367 }
3368 }
3369 }
3370 }
3371 up_read(&sinfo->groups_sem);
3372 }
3373
3374 return num_tolerated_disk_barrier_failures;
3375}
3376
3377static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3378{
3379 struct list_head *head;
3380 struct btrfs_device *dev;
3381 struct btrfs_super_block *sb;
3382 struct btrfs_dev_item *dev_item;
3383 int ret;
3384 int do_barriers;
3385 int max_errors;
3386 int total_errors = 0;
3387 u64 flags;
3388
3389 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3390 backup_super_roots(root->fs_info);
3391
3392 sb = root->fs_info->super_for_commit;
3393 dev_item = &sb->dev_item;
3394
3395 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3396 head = &root->fs_info->fs_devices->devices;
3397 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3398
3399 if (do_barriers) {
3400 ret = barrier_all_devices(root->fs_info);
3401 if (ret) {
3402 mutex_unlock(
3403 &root->fs_info->fs_devices->device_list_mutex);
3404 btrfs_error(root->fs_info, ret,
3405 "errors while submitting device barriers.");
3406 return ret;
3407 }
3408 }
3409
3410 list_for_each_entry_rcu(dev, head, dev_list) {
3411 if (!dev->bdev) {
3412 total_errors++;
3413 continue;
3414 }
3415 if (!dev->in_fs_metadata || !dev->writeable)
3416 continue;
3417
3418 btrfs_set_stack_device_generation(dev_item, 0);
3419 btrfs_set_stack_device_type(dev_item, dev->type);
3420 btrfs_set_stack_device_id(dev_item, dev->devid);
3421 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3422 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3423 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3424 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3425 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3426 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3427 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3428
3429 flags = btrfs_super_flags(sb);
3430 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3431
3432 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3433 if (ret)
3434 total_errors++;
3435 }
3436 if (total_errors > max_errors) {
3437 btrfs_err(root->fs_info, "%d errors while writing supers",
3438 total_errors);
3439 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3440
3441 /* FUA is masked off if unsupported and can't be the reason */
3442 btrfs_error(root->fs_info, -EIO,
3443 "%d errors while writing supers", total_errors);
3444 return -EIO;
3445 }
3446
3447 total_errors = 0;
3448 list_for_each_entry_rcu(dev, head, dev_list) {
3449 if (!dev->bdev)
3450 continue;
3451 if (!dev->in_fs_metadata || !dev->writeable)
3452 continue;
3453
3454 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3455 if (ret)
3456 total_errors++;
3457 }
3458 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3459 if (total_errors > max_errors) {
3460 btrfs_error(root->fs_info, -EIO,
3461 "%d errors while writing supers", total_errors);
3462 return -EIO;
3463 }
3464 return 0;
3465}
3466
3467int write_ctree_super(struct btrfs_trans_handle *trans,
3468 struct btrfs_root *root, int max_mirrors)
3469{
3470 return write_all_supers(root, max_mirrors);
3471}
3472
3473/* Drop a fs root from the radix tree and free it. */
3474void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3475 struct btrfs_root *root)
3476{
3477 spin_lock(&fs_info->fs_roots_radix_lock);
3478 radix_tree_delete(&fs_info->fs_roots_radix,
3479 (unsigned long)root->root_key.objectid);
3480 spin_unlock(&fs_info->fs_roots_radix_lock);
3481
3482 if (btrfs_root_refs(&root->root_item) == 0)
3483 synchronize_srcu(&fs_info->subvol_srcu);
3484
3485 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3486 btrfs_free_log(NULL, root);
3487
3488 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3489 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3490 free_fs_root(root);
3491}
3492
3493static void free_fs_root(struct btrfs_root *root)
3494{
3495 iput(root->cache_inode);
3496 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3497 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3498 root->orphan_block_rsv = NULL;
3499 if (root->anon_dev)
3500 free_anon_bdev(root->anon_dev);
3501 if (root->subv_writers)
3502 btrfs_free_subvolume_writers(root->subv_writers);
3503 free_extent_buffer(root->node);
3504 free_extent_buffer(root->commit_root);
3505 kfree(root->free_ino_ctl);
3506 kfree(root->free_ino_pinned);
3507 kfree(root->name);
3508 btrfs_put_fs_root(root);
3509}
3510
3511void btrfs_free_fs_root(struct btrfs_root *root)
3512{
3513 free_fs_root(root);
3514}
3515
3516int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3517{
3518 u64 root_objectid = 0;
3519 struct btrfs_root *gang[8];
3520 int i;
3521 int ret;
3522
3523 while (1) {
3524 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3525 (void **)gang, root_objectid,
3526 ARRAY_SIZE(gang));
3527 if (!ret)
3528 break;
3529
3530 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3531 for (i = 0; i < ret; i++) {
3532 int err;
3533
3534 root_objectid = gang[i]->root_key.objectid;
3535 err = btrfs_orphan_cleanup(gang[i]);
3536 if (err)
3537 return err;
3538 }
3539 root_objectid++;
3540 }
3541 return 0;
3542}
3543
3544int btrfs_commit_super(struct btrfs_root *root)
3545{
3546 struct btrfs_trans_handle *trans;
3547
3548 mutex_lock(&root->fs_info->cleaner_mutex);
3549 btrfs_run_delayed_iputs(root);
3550 mutex_unlock(&root->fs_info->cleaner_mutex);
3551 wake_up_process(root->fs_info->cleaner_kthread);
3552
3553 /* wait until ongoing cleanup work done */
3554 down_write(&root->fs_info->cleanup_work_sem);
3555 up_write(&root->fs_info->cleanup_work_sem);
3556
3557 trans = btrfs_join_transaction(root);
3558 if (IS_ERR(trans))
3559 return PTR_ERR(trans);
3560 return btrfs_commit_transaction(trans, root);
3561}
3562
3563int close_ctree(struct btrfs_root *root)
3564{
3565 struct btrfs_fs_info *fs_info = root->fs_info;
3566 int ret;
3567
3568 fs_info->closing = 1;
3569 smp_mb();
3570
3571 /* wait for the uuid_scan task to finish */
3572 down(&fs_info->uuid_tree_rescan_sem);
3573 /* avoid complains from lockdep et al., set sem back to initial state */
3574 up(&fs_info->uuid_tree_rescan_sem);
3575
3576 /* pause restriper - we want to resume on mount */
3577 btrfs_pause_balance(fs_info);
3578
3579 btrfs_dev_replace_suspend_for_unmount(fs_info);
3580
3581 btrfs_scrub_cancel(fs_info);
3582
3583 /* wait for any defraggers to finish */
3584 wait_event(fs_info->transaction_wait,
3585 (atomic_read(&fs_info->defrag_running) == 0));
3586
3587 /* clear out the rbtree of defraggable inodes */
3588 btrfs_cleanup_defrag_inodes(fs_info);
3589
3590 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3591 ret = btrfs_commit_super(root);
3592 if (ret)
3593 btrfs_err(root->fs_info, "commit super ret %d", ret);
3594 }
3595
3596 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3597 btrfs_error_commit_super(root);
3598
3599 kthread_stop(fs_info->transaction_kthread);
3600 kthread_stop(fs_info->cleaner_kthread);
3601
3602 fs_info->closing = 2;
3603 smp_mb();
3604
3605 btrfs_free_qgroup_config(root->fs_info);
3606
3607 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3608 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3609 percpu_counter_sum(&fs_info->delalloc_bytes));
3610 }
3611
3612 btrfs_sysfs_remove_one(fs_info);
3613
3614 del_fs_roots(fs_info);
3615
3616 btrfs_put_block_group_cache(fs_info);
3617
3618 btrfs_free_block_groups(fs_info);
3619
3620 btrfs_stop_all_workers(fs_info);
3621
3622 free_root_pointers(fs_info, 1);
3623
3624 iput(fs_info->btree_inode);
3625
3626#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3627 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3628 btrfsic_unmount(root, fs_info->fs_devices);
3629#endif
3630
3631 btrfs_close_devices(fs_info->fs_devices);
3632 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3633
3634 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3635 percpu_counter_destroy(&fs_info->delalloc_bytes);
3636 percpu_counter_destroy(&fs_info->bio_counter);
3637 bdi_destroy(&fs_info->bdi);
3638 cleanup_srcu_struct(&fs_info->subvol_srcu);
3639
3640 btrfs_free_stripe_hash_table(fs_info);
3641
3642 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3643 root->orphan_block_rsv = NULL;
3644
3645 return 0;
3646}
3647
3648int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3649 int atomic)
3650{
3651 int ret;
3652 struct inode *btree_inode = buf->pages[0]->mapping->host;
3653
3654 ret = extent_buffer_uptodate(buf);
3655 if (!ret)
3656 return ret;
3657
3658 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3659 parent_transid, atomic);
3660 if (ret == -EAGAIN)
3661 return ret;
3662 return !ret;
3663}
3664
3665int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3666{
3667 return set_extent_buffer_uptodate(buf);
3668}
3669
3670void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3671{
3672 struct btrfs_root *root;
3673 u64 transid = btrfs_header_generation(buf);
3674 int was_dirty;
3675
3676#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3677 /*
3678 * This is a fast path so only do this check if we have sanity tests
3679 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3680 * outside of the sanity tests.
3681 */
3682 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3683 return;
3684#endif
3685 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3686 btrfs_assert_tree_locked(buf);
3687 if (transid != root->fs_info->generation)
3688 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3689 "found %llu running %llu\n",
3690 buf->start, transid, root->fs_info->generation);
3691 was_dirty = set_extent_buffer_dirty(buf);
3692 if (!was_dirty)
3693 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3694 buf->len,
3695 root->fs_info->dirty_metadata_batch);
3696}
3697
3698static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3699 int flush_delayed)
3700{
3701 /*
3702 * looks as though older kernels can get into trouble with
3703 * this code, they end up stuck in balance_dirty_pages forever
3704 */
3705 int ret;
3706
3707 if (current->flags & PF_MEMALLOC)
3708 return;
3709
3710 if (flush_delayed)
3711 btrfs_balance_delayed_items(root);
3712
3713 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3714 BTRFS_DIRTY_METADATA_THRESH);
3715 if (ret > 0) {
3716 balance_dirty_pages_ratelimited(
3717 root->fs_info->btree_inode->i_mapping);
3718 }
3719 return;
3720}
3721
3722void btrfs_btree_balance_dirty(struct btrfs_root *root)
3723{
3724 __btrfs_btree_balance_dirty(root, 1);
3725}
3726
3727void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3728{
3729 __btrfs_btree_balance_dirty(root, 0);
3730}
3731
3732int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3733{
3734 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3735 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3736}
3737
3738static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3739 int read_only)
3740{
3741 /*
3742 * Placeholder for checks
3743 */
3744 return 0;
3745}
3746
3747static void btrfs_error_commit_super(struct btrfs_root *root)
3748{
3749 mutex_lock(&root->fs_info->cleaner_mutex);
3750 btrfs_run_delayed_iputs(root);
3751 mutex_unlock(&root->fs_info->cleaner_mutex);
3752
3753 down_write(&root->fs_info->cleanup_work_sem);
3754 up_write(&root->fs_info->cleanup_work_sem);
3755
3756 /* cleanup FS via transaction */
3757 btrfs_cleanup_transaction(root);
3758}
3759
3760static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3761 struct btrfs_root *root)
3762{
3763 struct btrfs_inode *btrfs_inode;
3764 struct list_head splice;
3765
3766 INIT_LIST_HEAD(&splice);
3767
3768 mutex_lock(&root->fs_info->ordered_operations_mutex);
3769 spin_lock(&root->fs_info->ordered_root_lock);
3770
3771 list_splice_init(&t->ordered_operations, &splice);
3772 while (!list_empty(&splice)) {
3773 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3774 ordered_operations);
3775
3776 list_del_init(&btrfs_inode->ordered_operations);
3777 spin_unlock(&root->fs_info->ordered_root_lock);
3778
3779 btrfs_invalidate_inodes(btrfs_inode->root);
3780
3781 spin_lock(&root->fs_info->ordered_root_lock);
3782 }
3783
3784 spin_unlock(&root->fs_info->ordered_root_lock);
3785 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3786}
3787
3788static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3789{
3790 struct btrfs_ordered_extent *ordered;
3791
3792 spin_lock(&root->ordered_extent_lock);
3793 /*
3794 * This will just short circuit the ordered completion stuff which will
3795 * make sure the ordered extent gets properly cleaned up.
3796 */
3797 list_for_each_entry(ordered, &root->ordered_extents,
3798 root_extent_list)
3799 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3800 spin_unlock(&root->ordered_extent_lock);
3801}
3802
3803static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3804{
3805 struct btrfs_root *root;
3806 struct list_head splice;
3807
3808 INIT_LIST_HEAD(&splice);
3809
3810 spin_lock(&fs_info->ordered_root_lock);
3811 list_splice_init(&fs_info->ordered_roots, &splice);
3812 while (!list_empty(&splice)) {
3813 root = list_first_entry(&splice, struct btrfs_root,
3814 ordered_root);
3815 list_move_tail(&root->ordered_root,
3816 &fs_info->ordered_roots);
3817
3818 spin_unlock(&fs_info->ordered_root_lock);
3819 btrfs_destroy_ordered_extents(root);
3820
3821 cond_resched();
3822 spin_lock(&fs_info->ordered_root_lock);
3823 }
3824 spin_unlock(&fs_info->ordered_root_lock);
3825}
3826
3827static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3828 struct btrfs_root *root)
3829{
3830 struct rb_node *node;
3831 struct btrfs_delayed_ref_root *delayed_refs;
3832 struct btrfs_delayed_ref_node *ref;
3833 int ret = 0;
3834
3835 delayed_refs = &trans->delayed_refs;
3836
3837 spin_lock(&delayed_refs->lock);
3838 if (atomic_read(&delayed_refs->num_entries) == 0) {
3839 spin_unlock(&delayed_refs->lock);
3840 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3841 return ret;
3842 }
3843
3844 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3845 struct btrfs_delayed_ref_head *head;
3846 bool pin_bytes = false;
3847
3848 head = rb_entry(node, struct btrfs_delayed_ref_head,
3849 href_node);
3850 if (!mutex_trylock(&head->mutex)) {
3851 atomic_inc(&head->node.refs);
3852 spin_unlock(&delayed_refs->lock);
3853
3854 mutex_lock(&head->mutex);
3855 mutex_unlock(&head->mutex);
3856 btrfs_put_delayed_ref(&head->node);
3857 spin_lock(&delayed_refs->lock);
3858 continue;
3859 }
3860 spin_lock(&head->lock);
3861 while ((node = rb_first(&head->ref_root)) != NULL) {
3862 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3863 rb_node);
3864 ref->in_tree = 0;
3865 rb_erase(&ref->rb_node, &head->ref_root);
3866 atomic_dec(&delayed_refs->num_entries);
3867 btrfs_put_delayed_ref(ref);
3868 }
3869 if (head->must_insert_reserved)
3870 pin_bytes = true;
3871 btrfs_free_delayed_extent_op(head->extent_op);
3872 delayed_refs->num_heads--;
3873 if (head->processing == 0)
3874 delayed_refs->num_heads_ready--;
3875 atomic_dec(&delayed_refs->num_entries);
3876 head->node.in_tree = 0;
3877 rb_erase(&head->href_node, &delayed_refs->href_root);
3878 spin_unlock(&head->lock);
3879 spin_unlock(&delayed_refs->lock);
3880 mutex_unlock(&head->mutex);
3881
3882 if (pin_bytes)
3883 btrfs_pin_extent(root, head->node.bytenr,
3884 head->node.num_bytes, 1);
3885 btrfs_put_delayed_ref(&head->node);
3886 cond_resched();
3887 spin_lock(&delayed_refs->lock);
3888 }
3889
3890 spin_unlock(&delayed_refs->lock);
3891
3892 return ret;
3893}
3894
3895static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3896{
3897 struct btrfs_inode *btrfs_inode;
3898 struct list_head splice;
3899
3900 INIT_LIST_HEAD(&splice);
3901
3902 spin_lock(&root->delalloc_lock);
3903 list_splice_init(&root->delalloc_inodes, &splice);
3904
3905 while (!list_empty(&splice)) {
3906 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3907 delalloc_inodes);
3908
3909 list_del_init(&btrfs_inode->delalloc_inodes);
3910 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3911 &btrfs_inode->runtime_flags);
3912 spin_unlock(&root->delalloc_lock);
3913
3914 btrfs_invalidate_inodes(btrfs_inode->root);
3915
3916 spin_lock(&root->delalloc_lock);
3917 }
3918
3919 spin_unlock(&root->delalloc_lock);
3920}
3921
3922static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3923{
3924 struct btrfs_root *root;
3925 struct list_head splice;
3926
3927 INIT_LIST_HEAD(&splice);
3928
3929 spin_lock(&fs_info->delalloc_root_lock);
3930 list_splice_init(&fs_info->delalloc_roots, &splice);
3931 while (!list_empty(&splice)) {
3932 root = list_first_entry(&splice, struct btrfs_root,
3933 delalloc_root);
3934 list_del_init(&root->delalloc_root);
3935 root = btrfs_grab_fs_root(root);
3936 BUG_ON(!root);
3937 spin_unlock(&fs_info->delalloc_root_lock);
3938
3939 btrfs_destroy_delalloc_inodes(root);
3940 btrfs_put_fs_root(root);
3941
3942 spin_lock(&fs_info->delalloc_root_lock);
3943 }
3944 spin_unlock(&fs_info->delalloc_root_lock);
3945}
3946
3947static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3948 struct extent_io_tree *dirty_pages,
3949 int mark)
3950{
3951 int ret;
3952 struct extent_buffer *eb;
3953 u64 start = 0;
3954 u64 end;
3955
3956 while (1) {
3957 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3958 mark, NULL);
3959 if (ret)
3960 break;
3961
3962 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3963 while (start <= end) {
3964 eb = btrfs_find_tree_block(root, start,
3965 root->leafsize);
3966 start += root->leafsize;
3967 if (!eb)
3968 continue;
3969 wait_on_extent_buffer_writeback(eb);
3970
3971 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3972 &eb->bflags))
3973 clear_extent_buffer_dirty(eb);
3974 free_extent_buffer_stale(eb);
3975 }
3976 }
3977
3978 return ret;
3979}
3980
3981static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3982 struct extent_io_tree *pinned_extents)
3983{
3984 struct extent_io_tree *unpin;
3985 u64 start;
3986 u64 end;
3987 int ret;
3988 bool loop = true;
3989
3990 unpin = pinned_extents;
3991again:
3992 while (1) {
3993 ret = find_first_extent_bit(unpin, 0, &start, &end,
3994 EXTENT_DIRTY, NULL);
3995 if (ret)
3996 break;
3997
3998 /* opt_discard */
3999 if (btrfs_test_opt(root, DISCARD))
4000 ret = btrfs_error_discard_extent(root, start,
4001 end + 1 - start,
4002 NULL);
4003
4004 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4005 btrfs_error_unpin_extent_range(root, start, end);
4006 cond_resched();
4007 }
4008
4009 if (loop) {
4010 if (unpin == &root->fs_info->freed_extents[0])
4011 unpin = &root->fs_info->freed_extents[1];
4012 else
4013 unpin = &root->fs_info->freed_extents[0];
4014 loop = false;
4015 goto again;
4016 }
4017
4018 return 0;
4019}
4020
4021void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4022 struct btrfs_root *root)
4023{
4024 btrfs_destroy_ordered_operations(cur_trans, root);
4025
4026 btrfs_destroy_delayed_refs(cur_trans, root);
4027
4028 cur_trans->state = TRANS_STATE_COMMIT_START;
4029 wake_up(&root->fs_info->transaction_blocked_wait);
4030
4031 cur_trans->state = TRANS_STATE_UNBLOCKED;
4032 wake_up(&root->fs_info->transaction_wait);
4033
4034 btrfs_destroy_delayed_inodes(root);
4035 btrfs_assert_delayed_root_empty(root);
4036
4037 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4038 EXTENT_DIRTY);
4039 btrfs_destroy_pinned_extent(root,
4040 root->fs_info->pinned_extents);
4041
4042 cur_trans->state =TRANS_STATE_COMPLETED;
4043 wake_up(&cur_trans->commit_wait);
4044
4045 /*
4046 memset(cur_trans, 0, sizeof(*cur_trans));
4047 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4048 */
4049}
4050
4051static int btrfs_cleanup_transaction(struct btrfs_root *root)
4052{
4053 struct btrfs_transaction *t;
4054
4055 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4056
4057 spin_lock(&root->fs_info->trans_lock);
4058 while (!list_empty(&root->fs_info->trans_list)) {
4059 t = list_first_entry(&root->fs_info->trans_list,
4060 struct btrfs_transaction, list);
4061 if (t->state >= TRANS_STATE_COMMIT_START) {
4062 atomic_inc(&t->use_count);
4063 spin_unlock(&root->fs_info->trans_lock);
4064 btrfs_wait_for_commit(root, t->transid);
4065 btrfs_put_transaction(t);
4066 spin_lock(&root->fs_info->trans_lock);
4067 continue;
4068 }
4069 if (t == root->fs_info->running_transaction) {
4070 t->state = TRANS_STATE_COMMIT_DOING;
4071 spin_unlock(&root->fs_info->trans_lock);
4072 /*
4073 * We wait for 0 num_writers since we don't hold a trans
4074 * handle open currently for this transaction.
4075 */
4076 wait_event(t->writer_wait,
4077 atomic_read(&t->num_writers) == 0);
4078 } else {
4079 spin_unlock(&root->fs_info->trans_lock);
4080 }
4081 btrfs_cleanup_one_transaction(t, root);
4082
4083 spin_lock(&root->fs_info->trans_lock);
4084 if (t == root->fs_info->running_transaction)
4085 root->fs_info->running_transaction = NULL;
4086 list_del_init(&t->list);
4087 spin_unlock(&root->fs_info->trans_lock);
4088
4089 btrfs_put_transaction(t);
4090 trace_btrfs_transaction_commit(root);
4091 spin_lock(&root->fs_info->trans_lock);
4092 }
4093 spin_unlock(&root->fs_info->trans_lock);
4094 btrfs_destroy_all_ordered_extents(root->fs_info);
4095 btrfs_destroy_delayed_inodes(root);
4096 btrfs_assert_delayed_root_empty(root);
4097 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4098 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4099 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4100
4101 return 0;
4102}
4103
4104static struct extent_io_ops btree_extent_io_ops = {
4105 .readpage_end_io_hook = btree_readpage_end_io_hook,
4106 .readpage_io_failed_hook = btree_io_failed_hook,
4107 .submit_bio_hook = btree_submit_bio_hook,
4108 /* note we're sharing with inode.c for the merge bio hook */
4109 .merge_bio_hook = btrfs_merge_bio_hook,
4110};
Linux 6.x block layer and io_uring tree(s)