projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
btrfs: factor btrfs_init_workqueues() out of open_ctree()
[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 "locking.h"
43#include "tree-log.h"
44#include "free-space-cache.h"
45#include "inode-map.h"
46#include "check-integrity.h"
47#include "rcu-string.h"
48#include "dev-replace.h"
49#include "raid56.h"
50#include "sysfs.h"
51#include "qgroup.h"
52
53#ifdef CONFIG_X86
54#include <asm/cpufeature.h>
55#endif
56
57static const 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_extents(struct btrfs_root *root);
63static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
64 struct btrfs_root *root);
65static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
66static int btrfs_destroy_marked_extents(struct btrfs_root *root,
67 struct extent_io_tree *dirty_pages,
68 int mark);
69static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
70 struct extent_io_tree *pinned_extents);
71static int btrfs_cleanup_transaction(struct btrfs_root *root);
72static void btrfs_error_commit_super(struct btrfs_root *root);
73
74/*
75 * btrfs_end_io_wq structs are used to do processing in task context when an IO
76 * is complete. This is used during reads to verify checksums, and it is used
77 * by writes to insert metadata for new file extents after IO is complete.
78 */
79struct btrfs_end_io_wq {
80 struct bio *bio;
81 bio_end_io_t *end_io;
82 void *private;
83 struct btrfs_fs_info *info;
84 int error;
85 enum btrfs_wq_endio_type metadata;
86 struct list_head list;
87 struct btrfs_work work;
88};
89
90static struct kmem_cache *btrfs_end_io_wq_cache;
91
92int __init btrfs_end_io_wq_init(void)
93{
94 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
95 sizeof(struct btrfs_end_io_wq),
96 0,
97 SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
98 NULL);
99 if (!btrfs_end_io_wq_cache)
100 return -ENOMEM;
101 return 0;
102}
103
104void btrfs_end_io_wq_exit(void)
105{
106 if (btrfs_end_io_wq_cache)
107 kmem_cache_destroy(btrfs_end_io_wq_cache);
108}
109
110/*
111 * async submit bios are used to offload expensive checksumming
112 * onto the worker threads. They checksum file and metadata bios
113 * just before they are sent down the IO stack.
114 */
115struct async_submit_bio {
116 struct inode *inode;
117 struct bio *bio;
118 struct list_head list;
119 extent_submit_bio_hook_t *submit_bio_start;
120 extent_submit_bio_hook_t *submit_bio_done;
121 int rw;
122 int mirror_num;
123 unsigned long bio_flags;
124 /*
125 * bio_offset is optional, can be used if the pages in the bio
126 * can't tell us where in the file the bio should go
127 */
128 u64 bio_offset;
129 struct btrfs_work work;
130 int error;
131};
132
133/*
134 * Lockdep class keys for extent_buffer->lock's in this root. For a given
135 * eb, the lockdep key is determined by the btrfs_root it belongs to and
136 * the level the eb occupies in the tree.
137 *
138 * Different roots are used for different purposes and may nest inside each
139 * other and they require separate keysets. As lockdep keys should be
140 * static, assign keysets according to the purpose of the root as indicated
141 * by btrfs_root->objectid. This ensures that all special purpose roots
142 * have separate keysets.
143 *
144 * Lock-nesting across peer nodes is always done with the immediate parent
145 * node locked thus preventing deadlock. As lockdep doesn't know this, use
146 * subclass to avoid triggering lockdep warning in such cases.
147 *
148 * The key is set by the readpage_end_io_hook after the buffer has passed
149 * csum validation but before the pages are unlocked. It is also set by
150 * btrfs_init_new_buffer on freshly allocated blocks.
151 *
152 * We also add a check to make sure the highest level of the tree is the
153 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
154 * needs update as well.
155 */
156#ifdef CONFIG_DEBUG_LOCK_ALLOC
157# if BTRFS_MAX_LEVEL != 8
158# error
159# endif
160
161static struct btrfs_lockdep_keyset {
162 u64 id; /* root objectid */
163 const char *name_stem; /* lock name stem */
164 char names[BTRFS_MAX_LEVEL + 1][20];
165 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
166} btrfs_lockdep_keysets[] = {
167 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
168 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
169 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
170 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
171 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
172 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
173 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
174 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
175 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
176 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
177 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
178 { .id = 0, .name_stem = "tree" },
179};
180
181void __init btrfs_init_lockdep(void)
182{
183 int i, j;
184
185 /* initialize lockdep class names */
186 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
187 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
188
189 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
190 snprintf(ks->names[j], sizeof(ks->names[j]),
191 "btrfs-%s-%02d", ks->name_stem, j);
192 }
193}
194
195void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
196 int level)
197{
198 struct btrfs_lockdep_keyset *ks;
199
200 BUG_ON(level >= ARRAY_SIZE(ks->keys));
201
202 /* find the matching keyset, id 0 is the default entry */
203 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
204 if (ks->id == objectid)
205 break;
206
207 lockdep_set_class_and_name(&eb->lock,
208 &ks->keys[level], ks->names[level]);
209}
210
211#endif
212
213/*
214 * extents on the btree inode are pretty simple, there's one extent
215 * that covers the entire device
216 */
217static struct extent_map *btree_get_extent(struct inode *inode,
218 struct page *page, size_t pg_offset, u64 start, u64 len,
219 int create)
220{
221 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
222 struct extent_map *em;
223 int ret;
224
225 read_lock(&em_tree->lock);
226 em = lookup_extent_mapping(em_tree, start, len);
227 if (em) {
228 em->bdev =
229 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
230 read_unlock(&em_tree->lock);
231 goto out;
232 }
233 read_unlock(&em_tree->lock);
234
235 em = alloc_extent_map();
236 if (!em) {
237 em = ERR_PTR(-ENOMEM);
238 goto out;
239 }
240 em->start = 0;
241 em->len = (u64)-1;
242 em->block_len = (u64)-1;
243 em->block_start = 0;
244 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
245
246 write_lock(&em_tree->lock);
247 ret = add_extent_mapping(em_tree, em, 0);
248 if (ret == -EEXIST) {
249 free_extent_map(em);
250 em = lookup_extent_mapping(em_tree, start, len);
251 if (!em)
252 em = ERR_PTR(-EIO);
253 } else if (ret) {
254 free_extent_map(em);
255 em = ERR_PTR(ret);
256 }
257 write_unlock(&em_tree->lock);
258
259out:
260 return em;
261}
262
263u32 btrfs_csum_data(char *data, u32 seed, size_t len)
264{
265 return btrfs_crc32c(seed, data, len);
266}
267
268void btrfs_csum_final(u32 crc, char *result)
269{
270 put_unaligned_le32(~crc, result);
271}
272
273/*
274 * compute the csum for a btree block, and either verify it or write it
275 * into the csum field of the block.
276 */
277static int csum_tree_block(struct btrfs_fs_info *fs_info,
278 struct extent_buffer *buf,
279 int verify)
280{
281 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
282 char *result = NULL;
283 unsigned long len;
284 unsigned long cur_len;
285 unsigned long offset = BTRFS_CSUM_SIZE;
286 char *kaddr;
287 unsigned long map_start;
288 unsigned long map_len;
289 int err;
290 u32 crc = ~(u32)0;
291 unsigned long inline_result;
292
293 len = buf->len - offset;
294 while (len > 0) {
295 err = map_private_extent_buffer(buf, offset, 32,
296 &kaddr, &map_start, &map_len);
297 if (err)
298 return 1;
299 cur_len = min(len, map_len - (offset - map_start));
300 crc = btrfs_csum_data(kaddr + offset - map_start,
301 crc, cur_len);
302 len -= cur_len;
303 offset += cur_len;
304 }
305 if (csum_size > sizeof(inline_result)) {
306 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
307 if (!result)
308 return 1;
309 } else {
310 result = (char *)&inline_result;
311 }
312
313 btrfs_csum_final(crc, result);
314
315 if (verify) {
316 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
317 u32 val;
318 u32 found = 0;
319 memcpy(&found, result, csum_size);
320
321 read_extent_buffer(buf, &val, 0, csum_size);
322 printk_ratelimited(KERN_WARNING
323 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
324 "level %d\n",
325 fs_info->sb->s_id, buf->start,
326 val, found, btrfs_header_level(buf));
327 if (result != (char *)&inline_result)
328 kfree(result);
329 return 1;
330 }
331 } else {
332 write_extent_buffer(buf, result, 0, csum_size);
333 }
334 if (result != (char *)&inline_result)
335 kfree(result);
336 return 0;
337}
338
339/*
340 * we can't consider a given block up to date unless the transid of the
341 * block matches the transid in the parent node's pointer. This is how we
342 * detect blocks that either didn't get written at all or got written
343 * in the wrong place.
344 */
345static int verify_parent_transid(struct extent_io_tree *io_tree,
346 struct extent_buffer *eb, u64 parent_transid,
347 int atomic)
348{
349 struct extent_state *cached_state = NULL;
350 int ret;
351 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
352
353 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
354 return 0;
355
356 if (atomic)
357 return -EAGAIN;
358
359 if (need_lock) {
360 btrfs_tree_read_lock(eb);
361 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
362 }
363
364 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
365 0, &cached_state);
366 if (extent_buffer_uptodate(eb) &&
367 btrfs_header_generation(eb) == parent_transid) {
368 ret = 0;
369 goto out;
370 }
371 printk_ratelimited(KERN_ERR
372 "BTRFS (device %s): parent transid verify failed on %llu wanted %llu found %llu\n",
373 eb->fs_info->sb->s_id, eb->start,
374 parent_transid, btrfs_header_generation(eb));
375 ret = 1;
376
377 /*
378 * Things reading via commit roots that don't have normal protection,
379 * like send, can have a really old block in cache that may point at a
380 * block that has been free'd and re-allocated. So don't clear uptodate
381 * if we find an eb that is under IO (dirty/writeback) because we could
382 * end up reading in the stale data and then writing it back out and
383 * making everybody very sad.
384 */
385 if (!extent_buffer_under_io(eb))
386 clear_extent_buffer_uptodate(eb);
387out:
388 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
389 &cached_state, GFP_NOFS);
390 if (need_lock)
391 btrfs_tree_read_unlock_blocking(eb);
392 return ret;
393}
394
395/*
396 * Return 0 if the superblock checksum type matches the checksum value of that
397 * algorithm. Pass the raw disk superblock data.
398 */
399static int btrfs_check_super_csum(char *raw_disk_sb)
400{
401 struct btrfs_super_block *disk_sb =
402 (struct btrfs_super_block *)raw_disk_sb;
403 u16 csum_type = btrfs_super_csum_type(disk_sb);
404 int ret = 0;
405
406 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
407 u32 crc = ~(u32)0;
408 const int csum_size = sizeof(crc);
409 char result[csum_size];
410
411 /*
412 * The super_block structure does not span the whole
413 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
414 * is filled with zeros and is included in the checkum.
415 */
416 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
417 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
418 btrfs_csum_final(crc, result);
419
420 if (memcmp(raw_disk_sb, result, csum_size))
421 ret = 1;
422
423 if (ret && btrfs_super_generation(disk_sb) < 10) {
424 printk(KERN_WARNING
425 "BTRFS: super block crcs don't match, older mkfs detected\n");
426 ret = 0;
427 }
428 }
429
430 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
431 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
432 csum_type);
433 ret = 1;
434 }
435
436 return ret;
437}
438
439/*
440 * helper to read a given tree block, doing retries as required when
441 * the checksums don't match and we have alternate mirrors to try.
442 */
443static int btree_read_extent_buffer_pages(struct btrfs_root *root,
444 struct extent_buffer *eb,
445 u64 start, u64 parent_transid)
446{
447 struct extent_io_tree *io_tree;
448 int failed = 0;
449 int ret;
450 int num_copies = 0;
451 int mirror_num = 0;
452 int failed_mirror = 0;
453
454 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
455 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
456 while (1) {
457 ret = read_extent_buffer_pages(io_tree, eb, start,
458 WAIT_COMPLETE,
459 btree_get_extent, mirror_num);
460 if (!ret) {
461 if (!verify_parent_transid(io_tree, eb,
462 parent_transid, 0))
463 break;
464 else
465 ret = -EIO;
466 }
467
468 /*
469 * This buffer's crc is fine, but its contents are corrupted, so
470 * there is no reason to read the other copies, they won't be
471 * any less wrong.
472 */
473 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
474 break;
475
476 num_copies = btrfs_num_copies(root->fs_info,
477 eb->start, eb->len);
478 if (num_copies == 1)
479 break;
480
481 if (!failed_mirror) {
482 failed = 1;
483 failed_mirror = eb->read_mirror;
484 }
485
486 mirror_num++;
487 if (mirror_num == failed_mirror)
488 mirror_num++;
489
490 if (mirror_num > num_copies)
491 break;
492 }
493
494 if (failed && !ret && failed_mirror)
495 repair_eb_io_failure(root, eb, failed_mirror);
496
497 return ret;
498}
499
500/*
501 * checksum a dirty tree block before IO. This has extra checks to make sure
502 * we only fill in the checksum field in the first page of a multi-page block
503 */
504
505static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
506{
507 u64 start = page_offset(page);
508 u64 found_start;
509 struct extent_buffer *eb;
510
511 eb = (struct extent_buffer *)page->private;
512 if (page != eb->pages[0])
513 return 0;
514 found_start = btrfs_header_bytenr(eb);
515 if (WARN_ON(found_start != start || !PageUptodate(page)))
516 return 0;
517 csum_tree_block(fs_info, eb, 0);
518 return 0;
519}
520
521static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
522 struct extent_buffer *eb)
523{
524 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
525 u8 fsid[BTRFS_UUID_SIZE];
526 int ret = 1;
527
528 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
529 while (fs_devices) {
530 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
531 ret = 0;
532 break;
533 }
534 fs_devices = fs_devices->seed;
535 }
536 return ret;
537}
538
539#define CORRUPT(reason, eb, root, slot) \
540 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
541 "root=%llu, slot=%d", reason, \
542 btrfs_header_bytenr(eb), root->objectid, slot)
543
544static noinline int check_leaf(struct btrfs_root *root,
545 struct extent_buffer *leaf)
546{
547 struct btrfs_key key;
548 struct btrfs_key leaf_key;
549 u32 nritems = btrfs_header_nritems(leaf);
550 int slot;
551
552 if (nritems == 0)
553 return 0;
554
555 /* Check the 0 item */
556 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
557 BTRFS_LEAF_DATA_SIZE(root)) {
558 CORRUPT("invalid item offset size pair", leaf, root, 0);
559 return -EIO;
560 }
561
562 /*
563 * Check to make sure each items keys are in the correct order and their
564 * offsets make sense. We only have to loop through nritems-1 because
565 * we check the current slot against the next slot, which verifies the
566 * next slot's offset+size makes sense and that the current's slot
567 * offset is correct.
568 */
569 for (slot = 0; slot < nritems - 1; slot++) {
570 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
571 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
572
573 /* Make sure the keys are in the right order */
574 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
575 CORRUPT("bad key order", leaf, root, slot);
576 return -EIO;
577 }
578
579 /*
580 * Make sure the offset and ends are right, remember that the
581 * item data starts at the end of the leaf and grows towards the
582 * front.
583 */
584 if (btrfs_item_offset_nr(leaf, slot) !=
585 btrfs_item_end_nr(leaf, slot + 1)) {
586 CORRUPT("slot offset bad", leaf, root, slot);
587 return -EIO;
588 }
589
590 /*
591 * Check to make sure that we don't point outside of the leaf,
592 * just incase all the items are consistent to eachother, but
593 * all point outside of the leaf.
594 */
595 if (btrfs_item_end_nr(leaf, slot) >
596 BTRFS_LEAF_DATA_SIZE(root)) {
597 CORRUPT("slot end outside of leaf", leaf, root, slot);
598 return -EIO;
599 }
600 }
601
602 return 0;
603}
604
605static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
606 u64 phy_offset, struct page *page,
607 u64 start, u64 end, int mirror)
608{
609 u64 found_start;
610 int found_level;
611 struct extent_buffer *eb;
612 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
613 int ret = 0;
614 int reads_done;
615
616 if (!page->private)
617 goto out;
618
619 eb = (struct extent_buffer *)page->private;
620
621 /* the pending IO might have been the only thing that kept this buffer
622 * in memory. Make sure we have a ref for all this other checks
623 */
624 extent_buffer_get(eb);
625
626 reads_done = atomic_dec_and_test(&eb->io_pages);
627 if (!reads_done)
628 goto err;
629
630 eb->read_mirror = mirror;
631 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
632 ret = -EIO;
633 goto err;
634 }
635
636 found_start = btrfs_header_bytenr(eb);
637 if (found_start != eb->start) {
638 printk_ratelimited(KERN_ERR "BTRFS (device %s): bad tree block start "
639 "%llu %llu\n",
640 eb->fs_info->sb->s_id, found_start, eb->start);
641 ret = -EIO;
642 goto err;
643 }
644 if (check_tree_block_fsid(root->fs_info, eb)) {
645 printk_ratelimited(KERN_ERR "BTRFS (device %s): bad fsid on block %llu\n",
646 eb->fs_info->sb->s_id, eb->start);
647 ret = -EIO;
648 goto err;
649 }
650 found_level = btrfs_header_level(eb);
651 if (found_level >= BTRFS_MAX_LEVEL) {
652 btrfs_err(root->fs_info, "bad tree block level %d",
653 (int)btrfs_header_level(eb));
654 ret = -EIO;
655 goto err;
656 }
657
658 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
659 eb, found_level);
660
661 ret = csum_tree_block(root->fs_info, eb, 1);
662 if (ret) {
663 ret = -EIO;
664 goto err;
665 }
666
667 /*
668 * If this is a leaf block and it is corrupt, set the corrupt bit so
669 * that we don't try and read the other copies of this block, just
670 * return -EIO.
671 */
672 if (found_level == 0 && check_leaf(root, eb)) {
673 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
674 ret = -EIO;
675 }
676
677 if (!ret)
678 set_extent_buffer_uptodate(eb);
679err:
680 if (reads_done &&
681 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
682 btree_readahead_hook(root, eb, eb->start, ret);
683
684 if (ret) {
685 /*
686 * our io error hook is going to dec the io pages
687 * again, we have to make sure it has something
688 * to decrement
689 */
690 atomic_inc(&eb->io_pages);
691 clear_extent_buffer_uptodate(eb);
692 }
693 free_extent_buffer(eb);
694out:
695 return ret;
696}
697
698static int btree_io_failed_hook(struct page *page, int failed_mirror)
699{
700 struct extent_buffer *eb;
701 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
702
703 eb = (struct extent_buffer *)page->private;
704 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
705 eb->read_mirror = failed_mirror;
706 atomic_dec(&eb->io_pages);
707 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
708 btree_readahead_hook(root, eb, eb->start, -EIO);
709 return -EIO; /* we fixed nothing */
710}
711
712static void end_workqueue_bio(struct bio *bio, int err)
713{
714 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
715 struct btrfs_fs_info *fs_info;
716 struct btrfs_workqueue *wq;
717 btrfs_work_func_t func;
718
719 fs_info = end_io_wq->info;
720 end_io_wq->error = err;
721
722 if (bio->bi_rw & REQ_WRITE) {
723 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
724 wq = fs_info->endio_meta_write_workers;
725 func = btrfs_endio_meta_write_helper;
726 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
727 wq = fs_info->endio_freespace_worker;
728 func = btrfs_freespace_write_helper;
729 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
730 wq = fs_info->endio_raid56_workers;
731 func = btrfs_endio_raid56_helper;
732 } else {
733 wq = fs_info->endio_write_workers;
734 func = btrfs_endio_write_helper;
735 }
736 } else {
737 if (unlikely(end_io_wq->metadata ==
738 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
739 wq = fs_info->endio_repair_workers;
740 func = btrfs_endio_repair_helper;
741 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
742 wq = fs_info->endio_raid56_workers;
743 func = btrfs_endio_raid56_helper;
744 } else if (end_io_wq->metadata) {
745 wq = fs_info->endio_meta_workers;
746 func = btrfs_endio_meta_helper;
747 } else {
748 wq = fs_info->endio_workers;
749 func = btrfs_endio_helper;
750 }
751 }
752
753 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
754 btrfs_queue_work(wq, &end_io_wq->work);
755}
756
757int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
758 enum btrfs_wq_endio_type metadata)
759{
760 struct btrfs_end_io_wq *end_io_wq;
761
762 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
763 if (!end_io_wq)
764 return -ENOMEM;
765
766 end_io_wq->private = bio->bi_private;
767 end_io_wq->end_io = bio->bi_end_io;
768 end_io_wq->info = info;
769 end_io_wq->error = 0;
770 end_io_wq->bio = bio;
771 end_io_wq->metadata = metadata;
772
773 bio->bi_private = end_io_wq;
774 bio->bi_end_io = end_workqueue_bio;
775 return 0;
776}
777
778unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
779{
780 unsigned long limit = min_t(unsigned long,
781 info->thread_pool_size,
782 info->fs_devices->open_devices);
783 return 256 * limit;
784}
785
786static void run_one_async_start(struct btrfs_work *work)
787{
788 struct async_submit_bio *async;
789 int ret;
790
791 async = container_of(work, struct async_submit_bio, work);
792 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
793 async->mirror_num, async->bio_flags,
794 async->bio_offset);
795 if (ret)
796 async->error = ret;
797}
798
799static void run_one_async_done(struct btrfs_work *work)
800{
801 struct btrfs_fs_info *fs_info;
802 struct async_submit_bio *async;
803 int limit;
804
805 async = container_of(work, struct async_submit_bio, work);
806 fs_info = BTRFS_I(async->inode)->root->fs_info;
807
808 limit = btrfs_async_submit_limit(fs_info);
809 limit = limit * 2 / 3;
810
811 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
812 waitqueue_active(&fs_info->async_submit_wait))
813 wake_up(&fs_info->async_submit_wait);
814
815 /* If an error occured we just want to clean up the bio and move on */
816 if (async->error) {
817 bio_endio(async->bio, async->error);
818 return;
819 }
820
821 async->submit_bio_done(async->inode, async->rw, async->bio,
822 async->mirror_num, async->bio_flags,
823 async->bio_offset);
824}
825
826static void run_one_async_free(struct btrfs_work *work)
827{
828 struct async_submit_bio *async;
829
830 async = container_of(work, struct async_submit_bio, work);
831 kfree(async);
832}
833
834int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
835 int rw, struct bio *bio, int mirror_num,
836 unsigned long bio_flags,
837 u64 bio_offset,
838 extent_submit_bio_hook_t *submit_bio_start,
839 extent_submit_bio_hook_t *submit_bio_done)
840{
841 struct async_submit_bio *async;
842
843 async = kmalloc(sizeof(*async), GFP_NOFS);
844 if (!async)
845 return -ENOMEM;
846
847 async->inode = inode;
848 async->rw = rw;
849 async->bio = bio;
850 async->mirror_num = mirror_num;
851 async->submit_bio_start = submit_bio_start;
852 async->submit_bio_done = submit_bio_done;
853
854 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
855 run_one_async_done, run_one_async_free);
856
857 async->bio_flags = bio_flags;
858 async->bio_offset = bio_offset;
859
860 async->error = 0;
861
862 atomic_inc(&fs_info->nr_async_submits);
863
864 if (rw & REQ_SYNC)
865 btrfs_set_work_high_priority(&async->work);
866
867 btrfs_queue_work(fs_info->workers, &async->work);
868
869 while (atomic_read(&fs_info->async_submit_draining) &&
870 atomic_read(&fs_info->nr_async_submits)) {
871 wait_event(fs_info->async_submit_wait,
872 (atomic_read(&fs_info->nr_async_submits) == 0));
873 }
874
875 return 0;
876}
877
878static int btree_csum_one_bio(struct bio *bio)
879{
880 struct bio_vec *bvec;
881 struct btrfs_root *root;
882 int i, ret = 0;
883
884 bio_for_each_segment_all(bvec, bio, i) {
885 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
886 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
887 if (ret)
888 break;
889 }
890
891 return ret;
892}
893
894static int __btree_submit_bio_start(struct inode *inode, int rw,
895 struct bio *bio, int mirror_num,
896 unsigned long bio_flags,
897 u64 bio_offset)
898{
899 /*
900 * when we're called for a write, we're already in the async
901 * submission context. Just jump into btrfs_map_bio
902 */
903 return btree_csum_one_bio(bio);
904}
905
906static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
907 int mirror_num, unsigned long bio_flags,
908 u64 bio_offset)
909{
910 int ret;
911
912 /*
913 * when we're called for a write, we're already in the async
914 * submission context. Just jump into btrfs_map_bio
915 */
916 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
917 if (ret)
918 bio_endio(bio, ret);
919 return ret;
920}
921
922static int check_async_write(struct inode *inode, unsigned long bio_flags)
923{
924 if (bio_flags & EXTENT_BIO_TREE_LOG)
925 return 0;
926#ifdef CONFIG_X86
927 if (cpu_has_xmm4_2)
928 return 0;
929#endif
930 return 1;
931}
932
933static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
934 int mirror_num, unsigned long bio_flags,
935 u64 bio_offset)
936{
937 int async = check_async_write(inode, bio_flags);
938 int ret;
939
940 if (!(rw & REQ_WRITE)) {
941 /*
942 * called for a read, do the setup so that checksum validation
943 * can happen in the async kernel threads
944 */
945 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
946 bio, BTRFS_WQ_ENDIO_METADATA);
947 if (ret)
948 goto out_w_error;
949 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
950 mirror_num, 0);
951 } else if (!async) {
952 ret = btree_csum_one_bio(bio);
953 if (ret)
954 goto out_w_error;
955 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
956 mirror_num, 0);
957 } else {
958 /*
959 * kthread helpers are used to submit writes so that
960 * checksumming can happen in parallel across all CPUs
961 */
962 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
963 inode, rw, bio, mirror_num, 0,
964 bio_offset,
965 __btree_submit_bio_start,
966 __btree_submit_bio_done);
967 }
968
969 if (ret) {
970out_w_error:
971 bio_endio(bio, ret);
972 }
973 return ret;
974}
975
976#ifdef CONFIG_MIGRATION
977static int btree_migratepage(struct address_space *mapping,
978 struct page *newpage, struct page *page,
979 enum migrate_mode mode)
980{
981 /*
982 * we can't safely write a btree page from here,
983 * we haven't done the locking hook
984 */
985 if (PageDirty(page))
986 return -EAGAIN;
987 /*
988 * Buffers may be managed in a filesystem specific way.
989 * We must have no buffers or drop them.
990 */
991 if (page_has_private(page) &&
992 !try_to_release_page(page, GFP_KERNEL))
993 return -EAGAIN;
994 return migrate_page(mapping, newpage, page, mode);
995}
996#endif
997
998
999static int btree_writepages(struct address_space *mapping,
1000 struct writeback_control *wbc)
1001{
1002 struct btrfs_fs_info *fs_info;
1003 int ret;
1004
1005 if (wbc->sync_mode == WB_SYNC_NONE) {
1006
1007 if (wbc->for_kupdate)
1008 return 0;
1009
1010 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1011 /* this is a bit racy, but that's ok */
1012 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1013 BTRFS_DIRTY_METADATA_THRESH);
1014 if (ret < 0)
1015 return 0;
1016 }
1017 return btree_write_cache_pages(mapping, wbc);
1018}
1019
1020static int btree_readpage(struct file *file, struct page *page)
1021{
1022 struct extent_io_tree *tree;
1023 tree = &BTRFS_I(page->mapping->host)->io_tree;
1024 return extent_read_full_page(tree, page, btree_get_extent, 0);
1025}
1026
1027static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1028{
1029 if (PageWriteback(page) || PageDirty(page))
1030 return 0;
1031
1032 return try_release_extent_buffer(page);
1033}
1034
1035static void btree_invalidatepage(struct page *page, unsigned int offset,
1036 unsigned int length)
1037{
1038 struct extent_io_tree *tree;
1039 tree = &BTRFS_I(page->mapping->host)->io_tree;
1040 extent_invalidatepage(tree, page, offset);
1041 btree_releasepage(page, GFP_NOFS);
1042 if (PagePrivate(page)) {
1043 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1044 "page private not zero on page %llu",
1045 (unsigned long long)page_offset(page));
1046 ClearPagePrivate(page);
1047 set_page_private(page, 0);
1048 page_cache_release(page);
1049 }
1050}
1051
1052static int btree_set_page_dirty(struct page *page)
1053{
1054#ifdef DEBUG
1055 struct extent_buffer *eb;
1056
1057 BUG_ON(!PagePrivate(page));
1058 eb = (struct extent_buffer *)page->private;
1059 BUG_ON(!eb);
1060 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1061 BUG_ON(!atomic_read(&eb->refs));
1062 btrfs_assert_tree_locked(eb);
1063#endif
1064 return __set_page_dirty_nobuffers(page);
1065}
1066
1067static const struct address_space_operations btree_aops = {
1068 .readpage = btree_readpage,
1069 .writepages = btree_writepages,
1070 .releasepage = btree_releasepage,
1071 .invalidatepage = btree_invalidatepage,
1072#ifdef CONFIG_MIGRATION
1073 .migratepage = btree_migratepage,
1074#endif
1075 .set_page_dirty = btree_set_page_dirty,
1076};
1077
1078void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1079{
1080 struct extent_buffer *buf = NULL;
1081 struct inode *btree_inode = root->fs_info->btree_inode;
1082
1083 buf = btrfs_find_create_tree_block(root, bytenr);
1084 if (!buf)
1085 return;
1086 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1087 buf, 0, WAIT_NONE, btree_get_extent, 0);
1088 free_extent_buffer(buf);
1089}
1090
1091int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1092 int mirror_num, struct extent_buffer **eb)
1093{
1094 struct extent_buffer *buf = NULL;
1095 struct inode *btree_inode = root->fs_info->btree_inode;
1096 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1097 int ret;
1098
1099 buf = btrfs_find_create_tree_block(root, bytenr);
1100 if (!buf)
1101 return 0;
1102
1103 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1104
1105 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1106 btree_get_extent, mirror_num);
1107 if (ret) {
1108 free_extent_buffer(buf);
1109 return ret;
1110 }
1111
1112 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1113 free_extent_buffer(buf);
1114 return -EIO;
1115 } else if (extent_buffer_uptodate(buf)) {
1116 *eb = buf;
1117 } else {
1118 free_extent_buffer(buf);
1119 }
1120 return 0;
1121}
1122
1123struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1124 u64 bytenr)
1125{
1126 return find_extent_buffer(fs_info, bytenr);
1127}
1128
1129struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1130 u64 bytenr)
1131{
1132 if (btrfs_test_is_dummy_root(root))
1133 return alloc_test_extent_buffer(root->fs_info, bytenr);
1134 return alloc_extent_buffer(root->fs_info, bytenr);
1135}
1136
1137
1138int btrfs_write_tree_block(struct extent_buffer *buf)
1139{
1140 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1141 buf->start + buf->len - 1);
1142}
1143
1144int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1145{
1146 return filemap_fdatawait_range(buf->pages[0]->mapping,
1147 buf->start, buf->start + buf->len - 1);
1148}
1149
1150struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1151 u64 parent_transid)
1152{
1153 struct extent_buffer *buf = NULL;
1154 int ret;
1155
1156 buf = btrfs_find_create_tree_block(root, bytenr);
1157 if (!buf)
1158 return NULL;
1159
1160 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1161 if (ret) {
1162 free_extent_buffer(buf);
1163 return NULL;
1164 }
1165 return buf;
1166
1167}
1168
1169void clean_tree_block(struct btrfs_trans_handle *trans,
1170 struct btrfs_fs_info *fs_info,
1171 struct extent_buffer *buf)
1172{
1173 if (btrfs_header_generation(buf) ==
1174 fs_info->running_transaction->transid) {
1175 btrfs_assert_tree_locked(buf);
1176
1177 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1178 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1179 -buf->len,
1180 fs_info->dirty_metadata_batch);
1181 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1182 btrfs_set_lock_blocking(buf);
1183 clear_extent_buffer_dirty(buf);
1184 }
1185 }
1186}
1187
1188static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1189{
1190 struct btrfs_subvolume_writers *writers;
1191 int ret;
1192
1193 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1194 if (!writers)
1195 return ERR_PTR(-ENOMEM);
1196
1197 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1198 if (ret < 0) {
1199 kfree(writers);
1200 return ERR_PTR(ret);
1201 }
1202
1203 init_waitqueue_head(&writers->wait);
1204 return writers;
1205}
1206
1207static void
1208btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1209{
1210 percpu_counter_destroy(&writers->counter);
1211 kfree(writers);
1212}
1213
1214static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1215 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1216 u64 objectid)
1217{
1218 root->node = NULL;
1219 root->commit_root = NULL;
1220 root->sectorsize = sectorsize;
1221 root->nodesize = nodesize;
1222 root->stripesize = stripesize;
1223 root->state = 0;
1224 root->orphan_cleanup_state = 0;
1225
1226 root->objectid = objectid;
1227 root->last_trans = 0;
1228 root->highest_objectid = 0;
1229 root->nr_delalloc_inodes = 0;
1230 root->nr_ordered_extents = 0;
1231 root->name = NULL;
1232 root->inode_tree = RB_ROOT;
1233 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1234 root->block_rsv = NULL;
1235 root->orphan_block_rsv = NULL;
1236
1237 INIT_LIST_HEAD(&root->dirty_list);
1238 INIT_LIST_HEAD(&root->root_list);
1239 INIT_LIST_HEAD(&root->delalloc_inodes);
1240 INIT_LIST_HEAD(&root->delalloc_root);
1241 INIT_LIST_HEAD(&root->ordered_extents);
1242 INIT_LIST_HEAD(&root->ordered_root);
1243 INIT_LIST_HEAD(&root->logged_list[0]);
1244 INIT_LIST_HEAD(&root->logged_list[1]);
1245 spin_lock_init(&root->orphan_lock);
1246 spin_lock_init(&root->inode_lock);
1247 spin_lock_init(&root->delalloc_lock);
1248 spin_lock_init(&root->ordered_extent_lock);
1249 spin_lock_init(&root->accounting_lock);
1250 spin_lock_init(&root->log_extents_lock[0]);
1251 spin_lock_init(&root->log_extents_lock[1]);
1252 mutex_init(&root->objectid_mutex);
1253 mutex_init(&root->log_mutex);
1254 mutex_init(&root->ordered_extent_mutex);
1255 mutex_init(&root->delalloc_mutex);
1256 init_waitqueue_head(&root->log_writer_wait);
1257 init_waitqueue_head(&root->log_commit_wait[0]);
1258 init_waitqueue_head(&root->log_commit_wait[1]);
1259 INIT_LIST_HEAD(&root->log_ctxs[0]);
1260 INIT_LIST_HEAD(&root->log_ctxs[1]);
1261 atomic_set(&root->log_commit[0], 0);
1262 atomic_set(&root->log_commit[1], 0);
1263 atomic_set(&root->log_writers, 0);
1264 atomic_set(&root->log_batch, 0);
1265 atomic_set(&root->orphan_inodes, 0);
1266 atomic_set(&root->refs, 1);
1267 atomic_set(&root->will_be_snapshoted, 0);
1268 root->log_transid = 0;
1269 root->log_transid_committed = -1;
1270 root->last_log_commit = 0;
1271 if (fs_info)
1272 extent_io_tree_init(&root->dirty_log_pages,
1273 fs_info->btree_inode->i_mapping);
1274
1275 memset(&root->root_key, 0, sizeof(root->root_key));
1276 memset(&root->root_item, 0, sizeof(root->root_item));
1277 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1278 if (fs_info)
1279 root->defrag_trans_start = fs_info->generation;
1280 else
1281 root->defrag_trans_start = 0;
1282 root->root_key.objectid = objectid;
1283 root->anon_dev = 0;
1284
1285 spin_lock_init(&root->root_item_lock);
1286}
1287
1288static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1289{
1290 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1291 if (root)
1292 root->fs_info = fs_info;
1293 return root;
1294}
1295
1296#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1297/* Should only be used by the testing infrastructure */
1298struct btrfs_root *btrfs_alloc_dummy_root(void)
1299{
1300 struct btrfs_root *root;
1301
1302 root = btrfs_alloc_root(NULL);
1303 if (!root)
1304 return ERR_PTR(-ENOMEM);
1305 __setup_root(4096, 4096, 4096, root, NULL, 1);
1306 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1307 root->alloc_bytenr = 0;
1308
1309 return root;
1310}
1311#endif
1312
1313struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1314 struct btrfs_fs_info *fs_info,
1315 u64 objectid)
1316{
1317 struct extent_buffer *leaf;
1318 struct btrfs_root *tree_root = fs_info->tree_root;
1319 struct btrfs_root *root;
1320 struct btrfs_key key;
1321 int ret = 0;
1322 uuid_le uuid;
1323
1324 root = btrfs_alloc_root(fs_info);
1325 if (!root)
1326 return ERR_PTR(-ENOMEM);
1327
1328 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1329 tree_root->stripesize, root, fs_info, objectid);
1330 root->root_key.objectid = objectid;
1331 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1332 root->root_key.offset = 0;
1333
1334 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1335 if (IS_ERR(leaf)) {
1336 ret = PTR_ERR(leaf);
1337 leaf = NULL;
1338 goto fail;
1339 }
1340
1341 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1342 btrfs_set_header_bytenr(leaf, leaf->start);
1343 btrfs_set_header_generation(leaf, trans->transid);
1344 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1345 btrfs_set_header_owner(leaf, objectid);
1346 root->node = leaf;
1347
1348 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1349 BTRFS_FSID_SIZE);
1350 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1351 btrfs_header_chunk_tree_uuid(leaf),
1352 BTRFS_UUID_SIZE);
1353 btrfs_mark_buffer_dirty(leaf);
1354
1355 root->commit_root = btrfs_root_node(root);
1356 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1357
1358 root->root_item.flags = 0;
1359 root->root_item.byte_limit = 0;
1360 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1361 btrfs_set_root_generation(&root->root_item, trans->transid);
1362 btrfs_set_root_level(&root->root_item, 0);
1363 btrfs_set_root_refs(&root->root_item, 1);
1364 btrfs_set_root_used(&root->root_item, leaf->len);
1365 btrfs_set_root_last_snapshot(&root->root_item, 0);
1366 btrfs_set_root_dirid(&root->root_item, 0);
1367 uuid_le_gen(&uuid);
1368 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1369 root->root_item.drop_level = 0;
1370
1371 key.objectid = objectid;
1372 key.type = BTRFS_ROOT_ITEM_KEY;
1373 key.offset = 0;
1374 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1375 if (ret)
1376 goto fail;
1377
1378 btrfs_tree_unlock(leaf);
1379
1380 return root;
1381
1382fail:
1383 if (leaf) {
1384 btrfs_tree_unlock(leaf);
1385 free_extent_buffer(root->commit_root);
1386 free_extent_buffer(leaf);
1387 }
1388 kfree(root);
1389
1390 return ERR_PTR(ret);
1391}
1392
1393static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1394 struct btrfs_fs_info *fs_info)
1395{
1396 struct btrfs_root *root;
1397 struct btrfs_root *tree_root = fs_info->tree_root;
1398 struct extent_buffer *leaf;
1399
1400 root = btrfs_alloc_root(fs_info);
1401 if (!root)
1402 return ERR_PTR(-ENOMEM);
1403
1404 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1405 tree_root->stripesize, root, fs_info,
1406 BTRFS_TREE_LOG_OBJECTID);
1407
1408 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1409 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1410 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1411
1412 /*
1413 * DON'T set REF_COWS for log trees
1414 *
1415 * log trees do not get reference counted because they go away
1416 * before a real commit is actually done. They do store pointers
1417 * to file data extents, and those reference counts still get
1418 * updated (along with back refs to the log tree).
1419 */
1420
1421 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1422 NULL, 0, 0, 0);
1423 if (IS_ERR(leaf)) {
1424 kfree(root);
1425 return ERR_CAST(leaf);
1426 }
1427
1428 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1429 btrfs_set_header_bytenr(leaf, leaf->start);
1430 btrfs_set_header_generation(leaf, trans->transid);
1431 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1432 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1433 root->node = leaf;
1434
1435 write_extent_buffer(root->node, root->fs_info->fsid,
1436 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1437 btrfs_mark_buffer_dirty(root->node);
1438 btrfs_tree_unlock(root->node);
1439 return root;
1440}
1441
1442int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1443 struct btrfs_fs_info *fs_info)
1444{
1445 struct btrfs_root *log_root;
1446
1447 log_root = alloc_log_tree(trans, fs_info);
1448 if (IS_ERR(log_root))
1449 return PTR_ERR(log_root);
1450 WARN_ON(fs_info->log_root_tree);
1451 fs_info->log_root_tree = log_root;
1452 return 0;
1453}
1454
1455int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1456 struct btrfs_root *root)
1457{
1458 struct btrfs_root *log_root;
1459 struct btrfs_inode_item *inode_item;
1460
1461 log_root = alloc_log_tree(trans, root->fs_info);
1462 if (IS_ERR(log_root))
1463 return PTR_ERR(log_root);
1464
1465 log_root->last_trans = trans->transid;
1466 log_root->root_key.offset = root->root_key.objectid;
1467
1468 inode_item = &log_root->root_item.inode;
1469 btrfs_set_stack_inode_generation(inode_item, 1);
1470 btrfs_set_stack_inode_size(inode_item, 3);
1471 btrfs_set_stack_inode_nlink(inode_item, 1);
1472 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1473 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1474
1475 btrfs_set_root_node(&log_root->root_item, log_root->node);
1476
1477 WARN_ON(root->log_root);
1478 root->log_root = log_root;
1479 root->log_transid = 0;
1480 root->log_transid_committed = -1;
1481 root->last_log_commit = 0;
1482 return 0;
1483}
1484
1485static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1486 struct btrfs_key *key)
1487{
1488 struct btrfs_root *root;
1489 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1490 struct btrfs_path *path;
1491 u64 generation;
1492 int ret;
1493
1494 path = btrfs_alloc_path();
1495 if (!path)
1496 return ERR_PTR(-ENOMEM);
1497
1498 root = btrfs_alloc_root(fs_info);
1499 if (!root) {
1500 ret = -ENOMEM;
1501 goto alloc_fail;
1502 }
1503
1504 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1505 tree_root->stripesize, root, fs_info, key->objectid);
1506
1507 ret = btrfs_find_root(tree_root, key, path,
1508 &root->root_item, &root->root_key);
1509 if (ret) {
1510 if (ret > 0)
1511 ret = -ENOENT;
1512 goto find_fail;
1513 }
1514
1515 generation = btrfs_root_generation(&root->root_item);
1516 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1517 generation);
1518 if (!root->node) {
1519 ret = -ENOMEM;
1520 goto find_fail;
1521 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1522 ret = -EIO;
1523 goto read_fail;
1524 }
1525 root->commit_root = btrfs_root_node(root);
1526out:
1527 btrfs_free_path(path);
1528 return root;
1529
1530read_fail:
1531 free_extent_buffer(root->node);
1532find_fail:
1533 kfree(root);
1534alloc_fail:
1535 root = ERR_PTR(ret);
1536 goto out;
1537}
1538
1539struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1540 struct btrfs_key *location)
1541{
1542 struct btrfs_root *root;
1543
1544 root = btrfs_read_tree_root(tree_root, location);
1545 if (IS_ERR(root))
1546 return root;
1547
1548 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1549 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1550 btrfs_check_and_init_root_item(&root->root_item);
1551 }
1552
1553 return root;
1554}
1555
1556int btrfs_init_fs_root(struct btrfs_root *root)
1557{
1558 int ret;
1559 struct btrfs_subvolume_writers *writers;
1560
1561 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1562 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1563 GFP_NOFS);
1564 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1565 ret = -ENOMEM;
1566 goto fail;
1567 }
1568
1569 writers = btrfs_alloc_subvolume_writers();
1570 if (IS_ERR(writers)) {
1571 ret = PTR_ERR(writers);
1572 goto fail;
1573 }
1574 root->subv_writers = writers;
1575
1576 btrfs_init_free_ino_ctl(root);
1577 spin_lock_init(&root->ino_cache_lock);
1578 init_waitqueue_head(&root->ino_cache_wait);
1579
1580 ret = get_anon_bdev(&root->anon_dev);
1581 if (ret)
1582 goto free_writers;
1583 return 0;
1584
1585free_writers:
1586 btrfs_free_subvolume_writers(root->subv_writers);
1587fail:
1588 kfree(root->free_ino_ctl);
1589 kfree(root->free_ino_pinned);
1590 return ret;
1591}
1592
1593static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1594 u64 root_id)
1595{
1596 struct btrfs_root *root;
1597
1598 spin_lock(&fs_info->fs_roots_radix_lock);
1599 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1600 (unsigned long)root_id);
1601 spin_unlock(&fs_info->fs_roots_radix_lock);
1602 return root;
1603}
1604
1605int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1606 struct btrfs_root *root)
1607{
1608 int ret;
1609
1610 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1611 if (ret)
1612 return ret;
1613
1614 spin_lock(&fs_info->fs_roots_radix_lock);
1615 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1616 (unsigned long)root->root_key.objectid,
1617 root);
1618 if (ret == 0)
1619 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1620 spin_unlock(&fs_info->fs_roots_radix_lock);
1621 radix_tree_preload_end();
1622
1623 return ret;
1624}
1625
1626struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1627 struct btrfs_key *location,
1628 bool check_ref)
1629{
1630 struct btrfs_root *root;
1631 struct btrfs_path *path;
1632 struct btrfs_key key;
1633 int ret;
1634
1635 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1636 return fs_info->tree_root;
1637 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1638 return fs_info->extent_root;
1639 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1640 return fs_info->chunk_root;
1641 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1642 return fs_info->dev_root;
1643 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1644 return fs_info->csum_root;
1645 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1646 return fs_info->quota_root ? fs_info->quota_root :
1647 ERR_PTR(-ENOENT);
1648 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1649 return fs_info->uuid_root ? fs_info->uuid_root :
1650 ERR_PTR(-ENOENT);
1651again:
1652 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1653 if (root) {
1654 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1655 return ERR_PTR(-ENOENT);
1656 return root;
1657 }
1658
1659 root = btrfs_read_fs_root(fs_info->tree_root, location);
1660 if (IS_ERR(root))
1661 return root;
1662
1663 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1664 ret = -ENOENT;
1665 goto fail;
1666 }
1667
1668 ret = btrfs_init_fs_root(root);
1669 if (ret)
1670 goto fail;
1671
1672 path = btrfs_alloc_path();
1673 if (!path) {
1674 ret = -ENOMEM;
1675 goto fail;
1676 }
1677 key.objectid = BTRFS_ORPHAN_OBJECTID;
1678 key.type = BTRFS_ORPHAN_ITEM_KEY;
1679 key.offset = location->objectid;
1680
1681 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1682 btrfs_free_path(path);
1683 if (ret < 0)
1684 goto fail;
1685 if (ret == 0)
1686 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1687
1688 ret = btrfs_insert_fs_root(fs_info, root);
1689 if (ret) {
1690 if (ret == -EEXIST) {
1691 free_fs_root(root);
1692 goto again;
1693 }
1694 goto fail;
1695 }
1696 return root;
1697fail:
1698 free_fs_root(root);
1699 return ERR_PTR(ret);
1700}
1701
1702static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1703{
1704 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1705 int ret = 0;
1706 struct btrfs_device *device;
1707 struct backing_dev_info *bdi;
1708
1709 rcu_read_lock();
1710 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1711 if (!device->bdev)
1712 continue;
1713 bdi = blk_get_backing_dev_info(device->bdev);
1714 if (bdi_congested(bdi, bdi_bits)) {
1715 ret = 1;
1716 break;
1717 }
1718 }
1719 rcu_read_unlock();
1720 return ret;
1721}
1722
1723static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1724{
1725 int err;
1726
1727 bdi->capabilities = BDI_CAP_MAP_COPY;
1728 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1729 if (err)
1730 return err;
1731
1732 bdi->ra_pages = default_backing_dev_info.ra_pages;
1733 bdi->congested_fn = btrfs_congested_fn;
1734 bdi->congested_data = info;
1735 return 0;
1736}
1737
1738/*
1739 * called by the kthread helper functions to finally call the bio end_io
1740 * functions. This is where read checksum verification actually happens
1741 */
1742static void end_workqueue_fn(struct btrfs_work *work)
1743{
1744 struct bio *bio;
1745 struct btrfs_end_io_wq *end_io_wq;
1746 int error;
1747
1748 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1749 bio = end_io_wq->bio;
1750
1751 error = end_io_wq->error;
1752 bio->bi_private = end_io_wq->private;
1753 bio->bi_end_io = end_io_wq->end_io;
1754 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1755 bio_endio_nodec(bio, error);
1756}
1757
1758static int cleaner_kthread(void *arg)
1759{
1760 struct btrfs_root *root = arg;
1761 int again;
1762
1763 do {
1764 again = 0;
1765
1766 /* Make the cleaner go to sleep early. */
1767 if (btrfs_need_cleaner_sleep(root))
1768 goto sleep;
1769
1770 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1771 goto sleep;
1772
1773 /*
1774 * Avoid the problem that we change the status of the fs
1775 * during the above check and trylock.
1776 */
1777 if (btrfs_need_cleaner_sleep(root)) {
1778 mutex_unlock(&root->fs_info->cleaner_mutex);
1779 goto sleep;
1780 }
1781
1782 btrfs_run_delayed_iputs(root);
1783 btrfs_delete_unused_bgs(root->fs_info);
1784 again = btrfs_clean_one_deleted_snapshot(root);
1785 mutex_unlock(&root->fs_info->cleaner_mutex);
1786
1787 /*
1788 * The defragger has dealt with the R/O remount and umount,
1789 * needn't do anything special here.
1790 */
1791 btrfs_run_defrag_inodes(root->fs_info);
1792sleep:
1793 if (!try_to_freeze() && !again) {
1794 set_current_state(TASK_INTERRUPTIBLE);
1795 if (!kthread_should_stop())
1796 schedule();
1797 __set_current_state(TASK_RUNNING);
1798 }
1799 } while (!kthread_should_stop());
1800 return 0;
1801}
1802
1803static int transaction_kthread(void *arg)
1804{
1805 struct btrfs_root *root = arg;
1806 struct btrfs_trans_handle *trans;
1807 struct btrfs_transaction *cur;
1808 u64 transid;
1809 unsigned long now;
1810 unsigned long delay;
1811 bool cannot_commit;
1812
1813 do {
1814 cannot_commit = false;
1815 delay = HZ * root->fs_info->commit_interval;
1816 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1817
1818 spin_lock(&root->fs_info->trans_lock);
1819 cur = root->fs_info->running_transaction;
1820 if (!cur) {
1821 spin_unlock(&root->fs_info->trans_lock);
1822 goto sleep;
1823 }
1824
1825 now = get_seconds();
1826 if (cur->state < TRANS_STATE_BLOCKED &&
1827 (now < cur->start_time ||
1828 now - cur->start_time < root->fs_info->commit_interval)) {
1829 spin_unlock(&root->fs_info->trans_lock);
1830 delay = HZ * 5;
1831 goto sleep;
1832 }
1833 transid = cur->transid;
1834 spin_unlock(&root->fs_info->trans_lock);
1835
1836 /* If the file system is aborted, this will always fail. */
1837 trans = btrfs_attach_transaction(root);
1838 if (IS_ERR(trans)) {
1839 if (PTR_ERR(trans) != -ENOENT)
1840 cannot_commit = true;
1841 goto sleep;
1842 }
1843 if (transid == trans->transid) {
1844 btrfs_commit_transaction(trans, root);
1845 } else {
1846 btrfs_end_transaction(trans, root);
1847 }
1848sleep:
1849 wake_up_process(root->fs_info->cleaner_kthread);
1850 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1851
1852 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1853 &root->fs_info->fs_state)))
1854 btrfs_cleanup_transaction(root);
1855 if (!try_to_freeze()) {
1856 set_current_state(TASK_INTERRUPTIBLE);
1857 if (!kthread_should_stop() &&
1858 (!btrfs_transaction_blocked(root->fs_info) ||
1859 cannot_commit))
1860 schedule_timeout(delay);
1861 __set_current_state(TASK_RUNNING);
1862 }
1863 } while (!kthread_should_stop());
1864 return 0;
1865}
1866
1867/*
1868 * this will find the highest generation in the array of
1869 * root backups. The index of the highest array is returned,
1870 * or -1 if we can't find anything.
1871 *
1872 * We check to make sure the array is valid by comparing the
1873 * generation of the latest root in the array with the generation
1874 * in the super block. If they don't match we pitch it.
1875 */
1876static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1877{
1878 u64 cur;
1879 int newest_index = -1;
1880 struct btrfs_root_backup *root_backup;
1881 int i;
1882
1883 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1884 root_backup = info->super_copy->super_roots + i;
1885 cur = btrfs_backup_tree_root_gen(root_backup);
1886 if (cur == newest_gen)
1887 newest_index = i;
1888 }
1889
1890 /* check to see if we actually wrapped around */
1891 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1892 root_backup = info->super_copy->super_roots;
1893 cur = btrfs_backup_tree_root_gen(root_backup);
1894 if (cur == newest_gen)
1895 newest_index = 0;
1896 }
1897 return newest_index;
1898}
1899
1900
1901/*
1902 * find the oldest backup so we know where to store new entries
1903 * in the backup array. This will set the backup_root_index
1904 * field in the fs_info struct
1905 */
1906static void find_oldest_super_backup(struct btrfs_fs_info *info,
1907 u64 newest_gen)
1908{
1909 int newest_index = -1;
1910
1911 newest_index = find_newest_super_backup(info, newest_gen);
1912 /* if there was garbage in there, just move along */
1913 if (newest_index == -1) {
1914 info->backup_root_index = 0;
1915 } else {
1916 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1917 }
1918}
1919
1920/*
1921 * copy all the root pointers into the super backup array.
1922 * this will bump the backup pointer by one when it is
1923 * done
1924 */
1925static void backup_super_roots(struct btrfs_fs_info *info)
1926{
1927 int next_backup;
1928 struct btrfs_root_backup *root_backup;
1929 int last_backup;
1930
1931 next_backup = info->backup_root_index;
1932 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1933 BTRFS_NUM_BACKUP_ROOTS;
1934
1935 /*
1936 * just overwrite the last backup if we're at the same generation
1937 * this happens only at umount
1938 */
1939 root_backup = info->super_for_commit->super_roots + last_backup;
1940 if (btrfs_backup_tree_root_gen(root_backup) ==
1941 btrfs_header_generation(info->tree_root->node))
1942 next_backup = last_backup;
1943
1944 root_backup = info->super_for_commit->super_roots + next_backup;
1945
1946 /*
1947 * make sure all of our padding and empty slots get zero filled
1948 * regardless of which ones we use today
1949 */
1950 memset(root_backup, 0, sizeof(*root_backup));
1951
1952 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1953
1954 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1955 btrfs_set_backup_tree_root_gen(root_backup,
1956 btrfs_header_generation(info->tree_root->node));
1957
1958 btrfs_set_backup_tree_root_level(root_backup,
1959 btrfs_header_level(info->tree_root->node));
1960
1961 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1962 btrfs_set_backup_chunk_root_gen(root_backup,
1963 btrfs_header_generation(info->chunk_root->node));
1964 btrfs_set_backup_chunk_root_level(root_backup,
1965 btrfs_header_level(info->chunk_root->node));
1966
1967 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1968 btrfs_set_backup_extent_root_gen(root_backup,
1969 btrfs_header_generation(info->extent_root->node));
1970 btrfs_set_backup_extent_root_level(root_backup,
1971 btrfs_header_level(info->extent_root->node));
1972
1973 /*
1974 * we might commit during log recovery, which happens before we set
1975 * the fs_root. Make sure it is valid before we fill it in.
1976 */
1977 if (info->fs_root && info->fs_root->node) {
1978 btrfs_set_backup_fs_root(root_backup,
1979 info->fs_root->node->start);
1980 btrfs_set_backup_fs_root_gen(root_backup,
1981 btrfs_header_generation(info->fs_root->node));
1982 btrfs_set_backup_fs_root_level(root_backup,
1983 btrfs_header_level(info->fs_root->node));
1984 }
1985
1986 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1987 btrfs_set_backup_dev_root_gen(root_backup,
1988 btrfs_header_generation(info->dev_root->node));
1989 btrfs_set_backup_dev_root_level(root_backup,
1990 btrfs_header_level(info->dev_root->node));
1991
1992 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1993 btrfs_set_backup_csum_root_gen(root_backup,
1994 btrfs_header_generation(info->csum_root->node));
1995 btrfs_set_backup_csum_root_level(root_backup,
1996 btrfs_header_level(info->csum_root->node));
1997
1998 btrfs_set_backup_total_bytes(root_backup,
1999 btrfs_super_total_bytes(info->super_copy));
2000 btrfs_set_backup_bytes_used(root_backup,
2001 btrfs_super_bytes_used(info->super_copy));
2002 btrfs_set_backup_num_devices(root_backup,
2003 btrfs_super_num_devices(info->super_copy));
2004
2005 /*
2006 * if we don't copy this out to the super_copy, it won't get remembered
2007 * for the next commit
2008 */
2009 memcpy(&info->super_copy->super_roots,
2010 &info->super_for_commit->super_roots,
2011 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2012}
2013
2014/*
2015 * this copies info out of the root backup array and back into
2016 * the in-memory super block. It is meant to help iterate through
2017 * the array, so you send it the number of backups you've already
2018 * tried and the last backup index you used.
2019 *
2020 * this returns -1 when it has tried all the backups
2021 */
2022static noinline int next_root_backup(struct btrfs_fs_info *info,
2023 struct btrfs_super_block *super,
2024 int *num_backups_tried, int *backup_index)
2025{
2026 struct btrfs_root_backup *root_backup;
2027 int newest = *backup_index;
2028
2029 if (*num_backups_tried == 0) {
2030 u64 gen = btrfs_super_generation(super);
2031
2032 newest = find_newest_super_backup(info, gen);
2033 if (newest == -1)
2034 return -1;
2035
2036 *backup_index = newest;
2037 *num_backups_tried = 1;
2038 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2039 /* we've tried all the backups, all done */
2040 return -1;
2041 } else {
2042 /* jump to the next oldest backup */
2043 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2044 BTRFS_NUM_BACKUP_ROOTS;
2045 *backup_index = newest;
2046 *num_backups_tried += 1;
2047 }
2048 root_backup = super->super_roots + newest;
2049
2050 btrfs_set_super_generation(super,
2051 btrfs_backup_tree_root_gen(root_backup));
2052 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2053 btrfs_set_super_root_level(super,
2054 btrfs_backup_tree_root_level(root_backup));
2055 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2056
2057 /*
2058 * fixme: the total bytes and num_devices need to match or we should
2059 * need a fsck
2060 */
2061 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2062 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2063 return 0;
2064}
2065
2066/* helper to cleanup workers */
2067static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2068{
2069 btrfs_destroy_workqueue(fs_info->fixup_workers);
2070 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2071 btrfs_destroy_workqueue(fs_info->workers);
2072 btrfs_destroy_workqueue(fs_info->endio_workers);
2073 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2074 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2075 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2076 btrfs_destroy_workqueue(fs_info->rmw_workers);
2077 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2078 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2079 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2080 btrfs_destroy_workqueue(fs_info->submit_workers);
2081 btrfs_destroy_workqueue(fs_info->delayed_workers);
2082 btrfs_destroy_workqueue(fs_info->caching_workers);
2083 btrfs_destroy_workqueue(fs_info->readahead_workers);
2084 btrfs_destroy_workqueue(fs_info->flush_workers);
2085 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2086 btrfs_destroy_workqueue(fs_info->extent_workers);
2087}
2088
2089static void free_root_extent_buffers(struct btrfs_root *root)
2090{
2091 if (root) {
2092 free_extent_buffer(root->node);
2093 free_extent_buffer(root->commit_root);
2094 root->node = NULL;
2095 root->commit_root = NULL;
2096 }
2097}
2098
2099/* helper to cleanup tree roots */
2100static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2101{
2102 free_root_extent_buffers(info->tree_root);
2103
2104 free_root_extent_buffers(info->dev_root);
2105 free_root_extent_buffers(info->extent_root);
2106 free_root_extent_buffers(info->csum_root);
2107 free_root_extent_buffers(info->quota_root);
2108 free_root_extent_buffers(info->uuid_root);
2109 if (chunk_root)
2110 free_root_extent_buffers(info->chunk_root);
2111}
2112
2113void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2114{
2115 int ret;
2116 struct btrfs_root *gang[8];
2117 int i;
2118
2119 while (!list_empty(&fs_info->dead_roots)) {
2120 gang[0] = list_entry(fs_info->dead_roots.next,
2121 struct btrfs_root, root_list);
2122 list_del(&gang[0]->root_list);
2123
2124 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2125 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2126 } else {
2127 free_extent_buffer(gang[0]->node);
2128 free_extent_buffer(gang[0]->commit_root);
2129 btrfs_put_fs_root(gang[0]);
2130 }
2131 }
2132
2133 while (1) {
2134 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2135 (void **)gang, 0,
2136 ARRAY_SIZE(gang));
2137 if (!ret)
2138 break;
2139 for (i = 0; i < ret; i++)
2140 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2141 }
2142
2143 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2144 btrfs_free_log_root_tree(NULL, fs_info);
2145 btrfs_destroy_pinned_extent(fs_info->tree_root,
2146 fs_info->pinned_extents);
2147 }
2148}
2149
2150static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2151{
2152 mutex_init(&fs_info->scrub_lock);
2153 atomic_set(&fs_info->scrubs_running, 0);
2154 atomic_set(&fs_info->scrub_pause_req, 0);
2155 atomic_set(&fs_info->scrubs_paused, 0);
2156 atomic_set(&fs_info->scrub_cancel_req, 0);
2157 init_waitqueue_head(&fs_info->scrub_pause_wait);
2158 fs_info->scrub_workers_refcnt = 0;
2159}
2160
2161static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2162{
2163 spin_lock_init(&fs_info->balance_lock);
2164 mutex_init(&fs_info->balance_mutex);
2165 atomic_set(&fs_info->balance_running, 0);
2166 atomic_set(&fs_info->balance_pause_req, 0);
2167 atomic_set(&fs_info->balance_cancel_req, 0);
2168 fs_info->balance_ctl = NULL;
2169 init_waitqueue_head(&fs_info->balance_wait_q);
2170}
2171
2172static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2173 struct btrfs_root *tree_root)
2174{
2175 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2176 set_nlink(fs_info->btree_inode, 1);
2177 /*
2178 * we set the i_size on the btree inode to the max possible int.
2179 * the real end of the address space is determined by all of
2180 * the devices in the system
2181 */
2182 fs_info->btree_inode->i_size = OFFSET_MAX;
2183 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2184 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2185
2186 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2187 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2188 fs_info->btree_inode->i_mapping);
2189 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2190 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2191
2192 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2193
2194 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2195 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2196 sizeof(struct btrfs_key));
2197 set_bit(BTRFS_INODE_DUMMY,
2198 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2199 btrfs_insert_inode_hash(fs_info->btree_inode);
2200}
2201
2202static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2203{
2204 fs_info->dev_replace.lock_owner = 0;
2205 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2206 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2207 mutex_init(&fs_info->dev_replace.lock_management_lock);
2208 mutex_init(&fs_info->dev_replace.lock);
2209 init_waitqueue_head(&fs_info->replace_wait);
2210}
2211
2212static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2213{
2214 spin_lock_init(&fs_info->qgroup_lock);
2215 mutex_init(&fs_info->qgroup_ioctl_lock);
2216 fs_info->qgroup_tree = RB_ROOT;
2217 fs_info->qgroup_op_tree = RB_ROOT;
2218 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2219 fs_info->qgroup_seq = 1;
2220 fs_info->quota_enabled = 0;
2221 fs_info->pending_quota_state = 0;
2222 fs_info->qgroup_ulist = NULL;
2223 mutex_init(&fs_info->qgroup_rescan_lock);
2224}
2225
2226static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2227 struct btrfs_fs_devices *fs_devices)
2228{
2229 int max_active = fs_info->thread_pool_size;
2230 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2231
2232 fs_info->workers =
2233 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2234 max_active, 16);
2235
2236 fs_info->delalloc_workers =
2237 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2238
2239 fs_info->flush_workers =
2240 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2241
2242 fs_info->caching_workers =
2243 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2244
2245 /*
2246 * a higher idle thresh on the submit workers makes it much more
2247 * likely that bios will be send down in a sane order to the
2248 * devices
2249 */
2250 fs_info->submit_workers =
2251 btrfs_alloc_workqueue("submit", flags,
2252 min_t(u64, fs_devices->num_devices,
2253 max_active), 64);
2254
2255 fs_info->fixup_workers =
2256 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2257
2258 /*
2259 * endios are largely parallel and should have a very
2260 * low idle thresh
2261 */
2262 fs_info->endio_workers =
2263 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2264 fs_info->endio_meta_workers =
2265 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2266 fs_info->endio_meta_write_workers =
2267 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2268 fs_info->endio_raid56_workers =
2269 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2270 fs_info->endio_repair_workers =
2271 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2272 fs_info->rmw_workers =
2273 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2274 fs_info->endio_write_workers =
2275 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2276 fs_info->endio_freespace_worker =
2277 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2278 fs_info->delayed_workers =
2279 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2280 fs_info->readahead_workers =
2281 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2282 fs_info->qgroup_rescan_workers =
2283 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2284 fs_info->extent_workers =
2285 btrfs_alloc_workqueue("extent-refs", flags,
2286 min_t(u64, fs_devices->num_devices,
2287 max_active), 8);
2288
2289 if (!(fs_info->workers && fs_info->delalloc_workers &&
2290 fs_info->submit_workers && fs_info->flush_workers &&
2291 fs_info->endio_workers && fs_info->endio_meta_workers &&
2292 fs_info->endio_meta_write_workers &&
2293 fs_info->endio_repair_workers &&
2294 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2295 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2296 fs_info->caching_workers && fs_info->readahead_workers &&
2297 fs_info->fixup_workers && fs_info->delayed_workers &&
2298 fs_info->extent_workers &&
2299 fs_info->qgroup_rescan_workers)) {
2300 return -ENOMEM;
2301 }
2302
2303 return 0;
2304}
2305
2306int open_ctree(struct super_block *sb,
2307 struct btrfs_fs_devices *fs_devices,
2308 char *options)
2309{
2310 u32 sectorsize;
2311 u32 nodesize;
2312 u32 stripesize;
2313 u64 generation;
2314 u64 features;
2315 struct btrfs_key location;
2316 struct buffer_head *bh;
2317 struct btrfs_super_block *disk_super;
2318 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2319 struct btrfs_root *tree_root;
2320 struct btrfs_root *extent_root;
2321 struct btrfs_root *csum_root;
2322 struct btrfs_root *chunk_root;
2323 struct btrfs_root *dev_root;
2324 struct btrfs_root *quota_root;
2325 struct btrfs_root *uuid_root;
2326 struct btrfs_root *log_tree_root;
2327 int ret;
2328 int err = -EINVAL;
2329 int num_backups_tried = 0;
2330 int backup_index = 0;
2331 int max_active;
2332 bool create_uuid_tree;
2333 bool check_uuid_tree;
2334
2335 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2336 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2337 if (!tree_root || !chunk_root) {
2338 err = -ENOMEM;
2339 goto fail;
2340 }
2341
2342 ret = init_srcu_struct(&fs_info->subvol_srcu);
2343 if (ret) {
2344 err = ret;
2345 goto fail;
2346 }
2347
2348 ret = setup_bdi(fs_info, &fs_info->bdi);
2349 if (ret) {
2350 err = ret;
2351 goto fail_srcu;
2352 }
2353
2354 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2355 if (ret) {
2356 err = ret;
2357 goto fail_bdi;
2358 }
2359 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2360 (1 + ilog2(nr_cpu_ids));
2361
2362 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2363 if (ret) {
2364 err = ret;
2365 goto fail_dirty_metadata_bytes;
2366 }
2367
2368 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2369 if (ret) {
2370 err = ret;
2371 goto fail_delalloc_bytes;
2372 }
2373
2374 fs_info->btree_inode = new_inode(sb);
2375 if (!fs_info->btree_inode) {
2376 err = -ENOMEM;
2377 goto fail_bio_counter;
2378 }
2379
2380 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2381
2382 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2383 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2384 INIT_LIST_HEAD(&fs_info->trans_list);
2385 INIT_LIST_HEAD(&fs_info->dead_roots);
2386 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2387 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2388 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2389 spin_lock_init(&fs_info->delalloc_root_lock);
2390 spin_lock_init(&fs_info->trans_lock);
2391 spin_lock_init(&fs_info->fs_roots_radix_lock);
2392 spin_lock_init(&fs_info->delayed_iput_lock);
2393 spin_lock_init(&fs_info->defrag_inodes_lock);
2394 spin_lock_init(&fs_info->free_chunk_lock);
2395 spin_lock_init(&fs_info->tree_mod_seq_lock);
2396 spin_lock_init(&fs_info->super_lock);
2397 spin_lock_init(&fs_info->qgroup_op_lock);
2398 spin_lock_init(&fs_info->buffer_lock);
2399 spin_lock_init(&fs_info->unused_bgs_lock);
2400 mutex_init(&fs_info->unused_bg_unpin_mutex);
2401 rwlock_init(&fs_info->tree_mod_log_lock);
2402 mutex_init(&fs_info->reloc_mutex);
2403 mutex_init(&fs_info->delalloc_root_mutex);
2404 seqlock_init(&fs_info->profiles_lock);
2405
2406 init_completion(&fs_info->kobj_unregister);
2407 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2408 INIT_LIST_HEAD(&fs_info->space_info);
2409 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2410 INIT_LIST_HEAD(&fs_info->unused_bgs);
2411 btrfs_mapping_init(&fs_info->mapping_tree);
2412 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2413 BTRFS_BLOCK_RSV_GLOBAL);
2414 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2415 BTRFS_BLOCK_RSV_DELALLOC);
2416 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2417 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2418 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2419 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2420 BTRFS_BLOCK_RSV_DELOPS);
2421 atomic_set(&fs_info->nr_async_submits, 0);
2422 atomic_set(&fs_info->async_delalloc_pages, 0);
2423 atomic_set(&fs_info->async_submit_draining, 0);
2424 atomic_set(&fs_info->nr_async_bios, 0);
2425 atomic_set(&fs_info->defrag_running, 0);
2426 atomic_set(&fs_info->qgroup_op_seq, 0);
2427 atomic64_set(&fs_info->tree_mod_seq, 0);
2428 fs_info->sb = sb;
2429 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2430 fs_info->metadata_ratio = 0;
2431 fs_info->defrag_inodes = RB_ROOT;
2432 fs_info->free_chunk_space = 0;
2433 fs_info->tree_mod_log = RB_ROOT;
2434 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2435 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2436 /* readahead state */
2437 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2438 spin_lock_init(&fs_info->reada_lock);
2439
2440 fs_info->thread_pool_size = min_t(unsigned long,
2441 num_online_cpus() + 2, 8);
2442
2443 INIT_LIST_HEAD(&fs_info->ordered_roots);
2444 spin_lock_init(&fs_info->ordered_root_lock);
2445 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2446 GFP_NOFS);
2447 if (!fs_info->delayed_root) {
2448 err = -ENOMEM;
2449 goto fail_iput;
2450 }
2451 btrfs_init_delayed_root(fs_info->delayed_root);
2452
2453 btrfs_init_scrub(fs_info);
2454#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2455 fs_info->check_integrity_print_mask = 0;
2456#endif
2457 btrfs_init_balance(fs_info);
2458 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2459
2460 sb->s_blocksize = 4096;
2461 sb->s_blocksize_bits = blksize_bits(4096);
2462 sb->s_bdi = &fs_info->bdi;
2463
2464 btrfs_init_btree_inode(fs_info, tree_root);
2465
2466 spin_lock_init(&fs_info->block_group_cache_lock);
2467 fs_info->block_group_cache_tree = RB_ROOT;
2468 fs_info->first_logical_byte = (u64)-1;
2469
2470 extent_io_tree_init(&fs_info->freed_extents[0],
2471 fs_info->btree_inode->i_mapping);
2472 extent_io_tree_init(&fs_info->freed_extents[1],
2473 fs_info->btree_inode->i_mapping);
2474 fs_info->pinned_extents = &fs_info->freed_extents[0];
2475 fs_info->do_barriers = 1;
2476
2477
2478 mutex_init(&fs_info->ordered_operations_mutex);
2479 mutex_init(&fs_info->ordered_extent_flush_mutex);
2480 mutex_init(&fs_info->tree_log_mutex);
2481 mutex_init(&fs_info->chunk_mutex);
2482 mutex_init(&fs_info->transaction_kthread_mutex);
2483 mutex_init(&fs_info->cleaner_mutex);
2484 mutex_init(&fs_info->volume_mutex);
2485 init_rwsem(&fs_info->commit_root_sem);
2486 init_rwsem(&fs_info->cleanup_work_sem);
2487 init_rwsem(&fs_info->subvol_sem);
2488 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2489
2490 btrfs_init_dev_replace_locks(fs_info);
2491 btrfs_init_qgroup(fs_info);
2492
2493 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2494 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2495
2496 init_waitqueue_head(&fs_info->transaction_throttle);
2497 init_waitqueue_head(&fs_info->transaction_wait);
2498 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2499 init_waitqueue_head(&fs_info->async_submit_wait);
2500
2501 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2502
2503 ret = btrfs_alloc_stripe_hash_table(fs_info);
2504 if (ret) {
2505 err = ret;
2506 goto fail_alloc;
2507 }
2508
2509 __setup_root(4096, 4096, 4096, tree_root,
2510 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2511
2512 invalidate_bdev(fs_devices->latest_bdev);
2513
2514 /*
2515 * Read super block and check the signature bytes only
2516 */
2517 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2518 if (!bh) {
2519 err = -EINVAL;
2520 goto fail_alloc;
2521 }
2522
2523 /*
2524 * We want to check superblock checksum, the type is stored inside.
2525 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2526 */
2527 if (btrfs_check_super_csum(bh->b_data)) {
2528 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2529 err = -EINVAL;
2530 goto fail_alloc;
2531 }
2532
2533 /*
2534 * super_copy is zeroed at allocation time and we never touch the
2535 * following bytes up to INFO_SIZE, the checksum is calculated from
2536 * the whole block of INFO_SIZE
2537 */
2538 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2539 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2540 sizeof(*fs_info->super_for_commit));
2541 brelse(bh);
2542
2543 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2544
2545 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2546 if (ret) {
2547 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2548 err = -EINVAL;
2549 goto fail_alloc;
2550 }
2551
2552 disk_super = fs_info->super_copy;
2553 if (!btrfs_super_root(disk_super))
2554 goto fail_alloc;
2555
2556 /* check FS state, whether FS is broken. */
2557 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2558 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2559
2560 /*
2561 * run through our array of backup supers and setup
2562 * our ring pointer to the oldest one
2563 */
2564 generation = btrfs_super_generation(disk_super);
2565 find_oldest_super_backup(fs_info, generation);
2566
2567 /*
2568 * In the long term, we'll store the compression type in the super
2569 * block, and it'll be used for per file compression control.
2570 */
2571 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2572
2573 ret = btrfs_parse_options(tree_root, options);
2574 if (ret) {
2575 err = ret;
2576 goto fail_alloc;
2577 }
2578
2579 features = btrfs_super_incompat_flags(disk_super) &
2580 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2581 if (features) {
2582 printk(KERN_ERR "BTRFS: couldn't mount because of "
2583 "unsupported optional features (%Lx).\n",
2584 features);
2585 err = -EINVAL;
2586 goto fail_alloc;
2587 }
2588
2589 /*
2590 * Leafsize and nodesize were always equal, this is only a sanity check.
2591 */
2592 if (le32_to_cpu(disk_super->__unused_leafsize) !=
2593 btrfs_super_nodesize(disk_super)) {
2594 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2595 "blocksizes don't match. node %d leaf %d\n",
2596 btrfs_super_nodesize(disk_super),
2597 le32_to_cpu(disk_super->__unused_leafsize));
2598 err = -EINVAL;
2599 goto fail_alloc;
2600 }
2601 if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2602 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2603 "blocksize (%d) was too large\n",
2604 btrfs_super_nodesize(disk_super));
2605 err = -EINVAL;
2606 goto fail_alloc;
2607 }
2608
2609 features = btrfs_super_incompat_flags(disk_super);
2610 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2611 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2612 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2613
2614 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2615 printk(KERN_INFO "BTRFS: has skinny extents\n");
2616
2617 /*
2618 * flag our filesystem as having big metadata blocks if
2619 * they are bigger than the page size
2620 */
2621 if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2622 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2623 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2624 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2625 }
2626
2627 nodesize = btrfs_super_nodesize(disk_super);
2628 sectorsize = btrfs_super_sectorsize(disk_super);
2629 stripesize = btrfs_super_stripesize(disk_super);
2630 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2631 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2632
2633 /*
2634 * mixed block groups end up with duplicate but slightly offset
2635 * extent buffers for the same range. It leads to corruptions
2636 */
2637 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2638 (sectorsize != nodesize)) {
2639 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2640 "are not allowed for mixed block groups on %s\n",
2641 sb->s_id);
2642 goto fail_alloc;
2643 }
2644
2645 /*
2646 * Needn't use the lock because there is no other task which will
2647 * update the flag.
2648 */
2649 btrfs_set_super_incompat_flags(disk_super, features);
2650
2651 features = btrfs_super_compat_ro_flags(disk_super) &
2652 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2653 if (!(sb->s_flags & MS_RDONLY) && features) {
2654 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2655 "unsupported option features (%Lx).\n",
2656 features);
2657 err = -EINVAL;
2658 goto fail_alloc;
2659 }
2660
2661 max_active = fs_info->thread_pool_size;
2662
2663 ret = btrfs_init_workqueues(fs_info, fs_devices);
2664 if (ret) {
2665 err = ret;
2666 goto fail_sb_buffer;
2667 }
2668
2669 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2670 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2671 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2672
2673 tree_root->nodesize = nodesize;
2674 tree_root->sectorsize = sectorsize;
2675 tree_root->stripesize = stripesize;
2676
2677 sb->s_blocksize = sectorsize;
2678 sb->s_blocksize_bits = blksize_bits(sectorsize);
2679
2680 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2681 printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id);
2682 goto fail_sb_buffer;
2683 }
2684
2685 if (sectorsize != PAGE_SIZE) {
2686 printk(KERN_ERR "BTRFS: incompatible sector size (%lu) "
2687 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2688 goto fail_sb_buffer;
2689 }
2690
2691 mutex_lock(&fs_info->chunk_mutex);
2692 ret = btrfs_read_sys_array(tree_root);
2693 mutex_unlock(&fs_info->chunk_mutex);
2694 if (ret) {
2695 printk(KERN_ERR "BTRFS: failed to read the system "
2696 "array on %s\n", sb->s_id);
2697 goto fail_sb_buffer;
2698 }
2699
2700 generation = btrfs_super_chunk_root_generation(disk_super);
2701
2702 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2703 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2704
2705 chunk_root->node = read_tree_block(chunk_root,
2706 btrfs_super_chunk_root(disk_super),
2707 generation);
2708 if (!chunk_root->node ||
2709 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2710 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2711 sb->s_id);
2712 goto fail_tree_roots;
2713 }
2714 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2715 chunk_root->commit_root = btrfs_root_node(chunk_root);
2716
2717 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2718 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2719
2720 ret = btrfs_read_chunk_tree(chunk_root);
2721 if (ret) {
2722 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2723 sb->s_id);
2724 goto fail_tree_roots;
2725 }
2726
2727 /*
2728 * keep the device that is marked to be the target device for the
2729 * dev_replace procedure
2730 */
2731 btrfs_close_extra_devices(fs_devices, 0);
2732
2733 if (!fs_devices->latest_bdev) {
2734 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2735 sb->s_id);
2736 goto fail_tree_roots;
2737 }
2738
2739retry_root_backup:
2740 generation = btrfs_super_generation(disk_super);
2741
2742 tree_root->node = read_tree_block(tree_root,
2743 btrfs_super_root(disk_super),
2744 generation);
2745 if (!tree_root->node ||
2746 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2747 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2748 sb->s_id);
2749
2750 goto recovery_tree_root;
2751 }
2752
2753 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2754 tree_root->commit_root = btrfs_root_node(tree_root);
2755 btrfs_set_root_refs(&tree_root->root_item, 1);
2756
2757 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2758 location.type = BTRFS_ROOT_ITEM_KEY;
2759 location.offset = 0;
2760
2761 extent_root = btrfs_read_tree_root(tree_root, &location);
2762 if (IS_ERR(extent_root)) {
2763 ret = PTR_ERR(extent_root);
2764 goto recovery_tree_root;
2765 }
2766 set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2767 fs_info->extent_root = extent_root;
2768
2769 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2770 dev_root = btrfs_read_tree_root(tree_root, &location);
2771 if (IS_ERR(dev_root)) {
2772 ret = PTR_ERR(dev_root);
2773 goto recovery_tree_root;
2774 }
2775 set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2776 fs_info->dev_root = dev_root;
2777 btrfs_init_devices_late(fs_info);
2778
2779 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2780 csum_root = btrfs_read_tree_root(tree_root, &location);
2781 if (IS_ERR(csum_root)) {
2782 ret = PTR_ERR(csum_root);
2783 goto recovery_tree_root;
2784 }
2785 set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2786 fs_info->csum_root = csum_root;
2787
2788 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2789 quota_root = btrfs_read_tree_root(tree_root, &location);
2790 if (!IS_ERR(quota_root)) {
2791 set_bit(BTRFS_ROOT_TRACK_DIRTY, &quota_root->state);
2792 fs_info->quota_enabled = 1;
2793 fs_info->pending_quota_state = 1;
2794 fs_info->quota_root = quota_root;
2795 }
2796
2797 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2798 uuid_root = btrfs_read_tree_root(tree_root, &location);
2799 if (IS_ERR(uuid_root)) {
2800 ret = PTR_ERR(uuid_root);
2801 if (ret != -ENOENT)
2802 goto recovery_tree_root;
2803 create_uuid_tree = true;
2804 check_uuid_tree = false;
2805 } else {
2806 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2807 fs_info->uuid_root = uuid_root;
2808 create_uuid_tree = false;
2809 check_uuid_tree =
2810 generation != btrfs_super_uuid_tree_generation(disk_super);
2811 }
2812
2813 fs_info->generation = generation;
2814 fs_info->last_trans_committed = generation;
2815
2816 ret = btrfs_recover_balance(fs_info);
2817 if (ret) {
2818 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2819 goto fail_block_groups;
2820 }
2821
2822 ret = btrfs_init_dev_stats(fs_info);
2823 if (ret) {
2824 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2825 ret);
2826 goto fail_block_groups;
2827 }
2828
2829 ret = btrfs_init_dev_replace(fs_info);
2830 if (ret) {
2831 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2832 goto fail_block_groups;
2833 }
2834
2835 btrfs_close_extra_devices(fs_devices, 1);
2836
2837 ret = btrfs_sysfs_add_one(fs_info);
2838 if (ret) {
2839 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2840 goto fail_block_groups;
2841 }
2842
2843 ret = btrfs_init_space_info(fs_info);
2844 if (ret) {
2845 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2846 goto fail_sysfs;
2847 }
2848
2849 ret = btrfs_read_block_groups(extent_root);
2850 if (ret) {
2851 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2852 goto fail_sysfs;
2853 }
2854 fs_info->num_tolerated_disk_barrier_failures =
2855 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2856 if (fs_info->fs_devices->missing_devices >
2857 fs_info->num_tolerated_disk_barrier_failures &&
2858 !(sb->s_flags & MS_RDONLY)) {
2859 printk(KERN_WARNING "BTRFS: "
2860 "too many missing devices, writeable mount is not allowed\n");
2861 goto fail_sysfs;
2862 }
2863
2864 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2865 "btrfs-cleaner");
2866 if (IS_ERR(fs_info->cleaner_kthread))
2867 goto fail_sysfs;
2868
2869 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2870 tree_root,
2871 "btrfs-transaction");
2872 if (IS_ERR(fs_info->transaction_kthread))
2873 goto fail_cleaner;
2874
2875 if (!btrfs_test_opt(tree_root, SSD) &&
2876 !btrfs_test_opt(tree_root, NOSSD) &&
2877 !fs_info->fs_devices->rotating) {
2878 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2879 "mode\n");
2880 btrfs_set_opt(fs_info->mount_opt, SSD);
2881 }
2882
2883 /*
2884 * Mount does not set all options immediatelly, we can do it now and do
2885 * not have to wait for transaction commit
2886 */
2887 btrfs_apply_pending_changes(fs_info);
2888
2889#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2890 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2891 ret = btrfsic_mount(tree_root, fs_devices,
2892 btrfs_test_opt(tree_root,
2893 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2894 1 : 0,
2895 fs_info->check_integrity_print_mask);
2896 if (ret)
2897 printk(KERN_WARNING "BTRFS: failed to initialize"
2898 " integrity check module %s\n", sb->s_id);
2899 }
2900#endif
2901 ret = btrfs_read_qgroup_config(fs_info);
2902 if (ret)
2903 goto fail_trans_kthread;
2904
2905 /* do not make disk changes in broken FS */
2906 if (btrfs_super_log_root(disk_super) != 0) {
2907 u64 bytenr = btrfs_super_log_root(disk_super);
2908
2909 if (fs_devices->rw_devices == 0) {
2910 printk(KERN_WARNING "BTRFS: log replay required "
2911 "on RO media\n");
2912 err = -EIO;
2913 goto fail_qgroup;
2914 }
2915
2916 log_tree_root = btrfs_alloc_root(fs_info);
2917 if (!log_tree_root) {
2918 err = -ENOMEM;
2919 goto fail_qgroup;
2920 }
2921
2922 __setup_root(nodesize, sectorsize, stripesize,
2923 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2924
2925 log_tree_root->node = read_tree_block(tree_root, bytenr,
2926 generation + 1);
2927 if (!log_tree_root->node ||
2928 !extent_buffer_uptodate(log_tree_root->node)) {
2929 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2930 free_extent_buffer(log_tree_root->node);
2931 kfree(log_tree_root);
2932 goto fail_qgroup;
2933 }
2934 /* returns with log_tree_root freed on success */
2935 ret = btrfs_recover_log_trees(log_tree_root);
2936 if (ret) {
2937 btrfs_error(tree_root->fs_info, ret,
2938 "Failed to recover log tree");
2939 free_extent_buffer(log_tree_root->node);
2940 kfree(log_tree_root);
2941 goto fail_qgroup;
2942 }
2943
2944 if (sb->s_flags & MS_RDONLY) {
2945 ret = btrfs_commit_super(tree_root);
2946 if (ret)
2947 goto fail_qgroup;
2948 }
2949 }
2950
2951 ret = btrfs_find_orphan_roots(tree_root);
2952 if (ret)
2953 goto fail_qgroup;
2954
2955 if (!(sb->s_flags & MS_RDONLY)) {
2956 ret = btrfs_cleanup_fs_roots(fs_info);
2957 if (ret)
2958 goto fail_qgroup;
2959
2960 mutex_lock(&fs_info->cleaner_mutex);
2961 ret = btrfs_recover_relocation(tree_root);
2962 mutex_unlock(&fs_info->cleaner_mutex);
2963 if (ret < 0) {
2964 printk(KERN_WARNING
2965 "BTRFS: failed to recover relocation\n");
2966 err = -EINVAL;
2967 goto fail_qgroup;
2968 }
2969 }
2970
2971 location.objectid = BTRFS_FS_TREE_OBJECTID;
2972 location.type = BTRFS_ROOT_ITEM_KEY;
2973 location.offset = 0;
2974
2975 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2976 if (IS_ERR(fs_info->fs_root)) {
2977 err = PTR_ERR(fs_info->fs_root);
2978 goto fail_qgroup;
2979 }
2980
2981 if (sb->s_flags & MS_RDONLY)
2982 return 0;
2983
2984 down_read(&fs_info->cleanup_work_sem);
2985 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2986 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2987 up_read(&fs_info->cleanup_work_sem);
2988 close_ctree(tree_root);
2989 return ret;
2990 }
2991 up_read(&fs_info->cleanup_work_sem);
2992
2993 ret = btrfs_resume_balance_async(fs_info);
2994 if (ret) {
2995 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2996 close_ctree(tree_root);
2997 return ret;
2998 }
2999
3000 ret = btrfs_resume_dev_replace_async(fs_info);
3001 if (ret) {
3002 pr_warn("BTRFS: failed to resume dev_replace\n");
3003 close_ctree(tree_root);
3004 return ret;
3005 }
3006
3007 btrfs_qgroup_rescan_resume(fs_info);
3008
3009 if (create_uuid_tree) {
3010 pr_info("BTRFS: creating UUID tree\n");
3011 ret = btrfs_create_uuid_tree(fs_info);
3012 if (ret) {
3013 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3014 ret);
3015 close_ctree(tree_root);
3016 return ret;
3017 }
3018 } else if (check_uuid_tree ||
3019 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
3020 pr_info("BTRFS: checking UUID tree\n");
3021 ret = btrfs_check_uuid_tree(fs_info);
3022 if (ret) {
3023 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3024 ret);
3025 close_ctree(tree_root);
3026 return ret;
3027 }
3028 } else {
3029 fs_info->update_uuid_tree_gen = 1;
3030 }
3031
3032 fs_info->open = 1;
3033
3034 return 0;
3035
3036fail_qgroup:
3037 btrfs_free_qgroup_config(fs_info);
3038fail_trans_kthread:
3039 kthread_stop(fs_info->transaction_kthread);
3040 btrfs_cleanup_transaction(fs_info->tree_root);
3041 btrfs_free_fs_roots(fs_info);
3042fail_cleaner:
3043 kthread_stop(fs_info->cleaner_kthread);
3044
3045 /*
3046 * make sure we're done with the btree inode before we stop our
3047 * kthreads
3048 */
3049 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3050
3051fail_sysfs:
3052 btrfs_sysfs_remove_one(fs_info);
3053
3054fail_block_groups:
3055 btrfs_put_block_group_cache(fs_info);
3056 btrfs_free_block_groups(fs_info);
3057
3058fail_tree_roots:
3059 free_root_pointers(fs_info, 1);
3060 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3061
3062fail_sb_buffer:
3063 btrfs_stop_all_workers(fs_info);
3064fail_alloc:
3065fail_iput:
3066 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3067
3068 iput(fs_info->btree_inode);
3069fail_bio_counter:
3070 percpu_counter_destroy(&fs_info->bio_counter);
3071fail_delalloc_bytes:
3072 percpu_counter_destroy(&fs_info->delalloc_bytes);
3073fail_dirty_metadata_bytes:
3074 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3075fail_bdi:
3076 bdi_destroy(&fs_info->bdi);
3077fail_srcu:
3078 cleanup_srcu_struct(&fs_info->subvol_srcu);
3079fail:
3080 btrfs_free_stripe_hash_table(fs_info);
3081 btrfs_close_devices(fs_info->fs_devices);
3082 return err;
3083
3084recovery_tree_root:
3085 if (!btrfs_test_opt(tree_root, RECOVERY))
3086 goto fail_tree_roots;
3087
3088 free_root_pointers(fs_info, 0);
3089
3090 /* don't use the log in recovery mode, it won't be valid */
3091 btrfs_set_super_log_root(disk_super, 0);
3092
3093 /* we can't trust the free space cache either */
3094 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3095
3096 ret = next_root_backup(fs_info, fs_info->super_copy,
3097 &num_backups_tried, &backup_index);
3098 if (ret == -1)
3099 goto fail_block_groups;
3100 goto retry_root_backup;
3101}
3102
3103static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3104{
3105 if (uptodate) {
3106 set_buffer_uptodate(bh);
3107 } else {
3108 struct btrfs_device *device = (struct btrfs_device *)
3109 bh->b_private;
3110
3111 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3112 "I/O error on %s\n",
3113 rcu_str_deref(device->name));
3114 /* note, we dont' set_buffer_write_io_error because we have
3115 * our own ways of dealing with the IO errors
3116 */
3117 clear_buffer_uptodate(bh);
3118 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3119 }
3120 unlock_buffer(bh);
3121 put_bh(bh);
3122}
3123
3124struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3125{
3126 struct buffer_head *bh;
3127 struct buffer_head *latest = NULL;
3128 struct btrfs_super_block *super;
3129 int i;
3130 u64 transid = 0;
3131 u64 bytenr;
3132
3133 /* we would like to check all the supers, but that would make
3134 * a btrfs mount succeed after a mkfs from a different FS.
3135 * So, we need to add a special mount option to scan for
3136 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3137 */
3138 for (i = 0; i < 1; i++) {
3139 bytenr = btrfs_sb_offset(i);
3140 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3141 i_size_read(bdev->bd_inode))
3142 break;
3143 bh = __bread(bdev, bytenr / 4096,
3144 BTRFS_SUPER_INFO_SIZE);
3145 if (!bh)
3146 continue;
3147
3148 super = (struct btrfs_super_block *)bh->b_data;
3149 if (btrfs_super_bytenr(super) != bytenr ||
3150 btrfs_super_magic(super) != BTRFS_MAGIC) {
3151 brelse(bh);
3152 continue;
3153 }
3154
3155 if (!latest || btrfs_super_generation(super) > transid) {
3156 brelse(latest);
3157 latest = bh;
3158 transid = btrfs_super_generation(super);
3159 } else {
3160 brelse(bh);
3161 }
3162 }
3163 return latest;
3164}
3165
3166/*
3167 * this should be called twice, once with wait == 0 and
3168 * once with wait == 1. When wait == 0 is done, all the buffer heads
3169 * we write are pinned.
3170 *
3171 * They are released when wait == 1 is done.
3172 * max_mirrors must be the same for both runs, and it indicates how
3173 * many supers on this one device should be written.
3174 *
3175 * max_mirrors == 0 means to write them all.
3176 */
3177static int write_dev_supers(struct btrfs_device *device,
3178 struct btrfs_super_block *sb,
3179 int do_barriers, int wait, int max_mirrors)
3180{
3181 struct buffer_head *bh;
3182 int i;
3183 int ret;
3184 int errors = 0;
3185 u32 crc;
3186 u64 bytenr;
3187
3188 if (max_mirrors == 0)
3189 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3190
3191 for (i = 0; i < max_mirrors; i++) {
3192 bytenr = btrfs_sb_offset(i);
3193 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3194 device->commit_total_bytes)
3195 break;
3196
3197 if (wait) {
3198 bh = __find_get_block(device->bdev, bytenr / 4096,
3199 BTRFS_SUPER_INFO_SIZE);
3200 if (!bh) {
3201 errors++;
3202 continue;
3203 }
3204 wait_on_buffer(bh);
3205 if (!buffer_uptodate(bh))
3206 errors++;
3207
3208 /* drop our reference */
3209 brelse(bh);
3210
3211 /* drop the reference from the wait == 0 run */
3212 brelse(bh);
3213 continue;
3214 } else {
3215 btrfs_set_super_bytenr(sb, bytenr);
3216
3217 crc = ~(u32)0;
3218 crc = btrfs_csum_data((char *)sb +
3219 BTRFS_CSUM_SIZE, crc,
3220 BTRFS_SUPER_INFO_SIZE -
3221 BTRFS_CSUM_SIZE);
3222 btrfs_csum_final(crc, sb->csum);
3223
3224 /*
3225 * one reference for us, and we leave it for the
3226 * caller
3227 */
3228 bh = __getblk(device->bdev, bytenr / 4096,
3229 BTRFS_SUPER_INFO_SIZE);
3230 if (!bh) {
3231 printk(KERN_ERR "BTRFS: couldn't get super "
3232 "buffer head for bytenr %Lu\n", bytenr);
3233 errors++;
3234 continue;
3235 }
3236
3237 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3238
3239 /* one reference for submit_bh */
3240 get_bh(bh);
3241
3242 set_buffer_uptodate(bh);
3243 lock_buffer(bh);
3244 bh->b_end_io = btrfs_end_buffer_write_sync;
3245 bh->b_private = device;
3246 }
3247
3248 /*
3249 * we fua the first super. The others we allow
3250 * to go down lazy.
3251 */
3252 if (i == 0)
3253 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3254 else
3255 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3256 if (ret)
3257 errors++;
3258 }
3259 return errors < i ? 0 : -1;
3260}
3261
3262/*
3263 * endio for the write_dev_flush, this will wake anyone waiting
3264 * for the barrier when it is done
3265 */
3266static void btrfs_end_empty_barrier(struct bio *bio, int err)
3267{
3268 if (err) {
3269 if (err == -EOPNOTSUPP)
3270 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3271 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3272 }
3273 if (bio->bi_private)
3274 complete(bio->bi_private);
3275 bio_put(bio);
3276}
3277
3278/*
3279 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3280 * sent down. With wait == 1, it waits for the previous flush.
3281 *
3282 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3283 * capable
3284 */
3285static int write_dev_flush(struct btrfs_device *device, int wait)
3286{
3287 struct bio *bio;
3288 int ret = 0;
3289
3290 if (device->nobarriers)
3291 return 0;
3292
3293 if (wait) {
3294 bio = device->flush_bio;
3295 if (!bio)
3296 return 0;
3297
3298 wait_for_completion(&device->flush_wait);
3299
3300 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3301 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3302 rcu_str_deref(device->name));
3303 device->nobarriers = 1;
3304 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3305 ret = -EIO;
3306 btrfs_dev_stat_inc_and_print(device,
3307 BTRFS_DEV_STAT_FLUSH_ERRS);
3308 }
3309
3310 /* drop the reference from the wait == 0 run */
3311 bio_put(bio);
3312 device->flush_bio = NULL;
3313
3314 return ret;
3315 }
3316
3317 /*
3318 * one reference for us, and we leave it for the
3319 * caller
3320 */
3321 device->flush_bio = NULL;
3322 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3323 if (!bio)
3324 return -ENOMEM;
3325
3326 bio->bi_end_io = btrfs_end_empty_barrier;
3327 bio->bi_bdev = device->bdev;
3328 init_completion(&device->flush_wait);
3329 bio->bi_private = &device->flush_wait;
3330 device->flush_bio = bio;
3331
3332 bio_get(bio);
3333 btrfsic_submit_bio(WRITE_FLUSH, bio);
3334
3335 return 0;
3336}
3337
3338/*
3339 * send an empty flush down to each device in parallel,
3340 * then wait for them
3341 */
3342static int barrier_all_devices(struct btrfs_fs_info *info)
3343{
3344 struct list_head *head;
3345 struct btrfs_device *dev;
3346 int errors_send = 0;
3347 int errors_wait = 0;
3348 int ret;
3349
3350 /* send down all the barriers */
3351 head = &info->fs_devices->devices;
3352 list_for_each_entry_rcu(dev, head, dev_list) {
3353 if (dev->missing)
3354 continue;
3355 if (!dev->bdev) {
3356 errors_send++;
3357 continue;
3358 }
3359 if (!dev->in_fs_metadata || !dev->writeable)
3360 continue;
3361
3362 ret = write_dev_flush(dev, 0);
3363 if (ret)
3364 errors_send++;
3365 }
3366
3367 /* wait for all the barriers */
3368 list_for_each_entry_rcu(dev, head, dev_list) {
3369 if (dev->missing)
3370 continue;
3371 if (!dev->bdev) {
3372 errors_wait++;
3373 continue;
3374 }
3375 if (!dev->in_fs_metadata || !dev->writeable)
3376 continue;
3377
3378 ret = write_dev_flush(dev, 1);
3379 if (ret)
3380 errors_wait++;
3381 }
3382 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3383 errors_wait > info->num_tolerated_disk_barrier_failures)
3384 return -EIO;
3385 return 0;
3386}
3387
3388int btrfs_calc_num_tolerated_disk_barrier_failures(
3389 struct btrfs_fs_info *fs_info)
3390{
3391 struct btrfs_ioctl_space_info space;
3392 struct btrfs_space_info *sinfo;
3393 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3394 BTRFS_BLOCK_GROUP_SYSTEM,
3395 BTRFS_BLOCK_GROUP_METADATA,
3396 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3397 int num_types = 4;
3398 int i;
3399 int c;
3400 int num_tolerated_disk_barrier_failures =
3401 (int)fs_info->fs_devices->num_devices;
3402
3403 for (i = 0; i < num_types; i++) {
3404 struct btrfs_space_info *tmp;
3405
3406 sinfo = NULL;
3407 rcu_read_lock();
3408 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3409 if (tmp->flags == types[i]) {
3410 sinfo = tmp;
3411 break;
3412 }
3413 }
3414 rcu_read_unlock();
3415
3416 if (!sinfo)
3417 continue;
3418
3419 down_read(&sinfo->groups_sem);
3420 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3421 if (!list_empty(&sinfo->block_groups[c])) {
3422 u64 flags;
3423
3424 btrfs_get_block_group_info(
3425 &sinfo->block_groups[c], &space);
3426 if (space.total_bytes == 0 ||
3427 space.used_bytes == 0)
3428 continue;
3429 flags = space.flags;
3430 /*
3431 * return
3432 * 0: if dup, single or RAID0 is configured for
3433 * any of metadata, system or data, else
3434 * 1: if RAID5 is configured, or if RAID1 or
3435 * RAID10 is configured and only two mirrors
3436 * are used, else
3437 * 2: if RAID6 is configured, else
3438 * num_mirrors - 1: if RAID1 or RAID10 is
3439 * configured and more than
3440 * 2 mirrors are used.
3441 */
3442 if (num_tolerated_disk_barrier_failures > 0 &&
3443 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3444 BTRFS_BLOCK_GROUP_RAID0)) ||
3445 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3446 == 0)))
3447 num_tolerated_disk_barrier_failures = 0;
3448 else if (num_tolerated_disk_barrier_failures > 1) {
3449 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3450 BTRFS_BLOCK_GROUP_RAID5 |
3451 BTRFS_BLOCK_GROUP_RAID10)) {
3452 num_tolerated_disk_barrier_failures = 1;
3453 } else if (flags &
3454 BTRFS_BLOCK_GROUP_RAID6) {
3455 num_tolerated_disk_barrier_failures = 2;
3456 }
3457 }
3458 }
3459 }
3460 up_read(&sinfo->groups_sem);
3461 }
3462
3463 return num_tolerated_disk_barrier_failures;
3464}
3465
3466static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3467{
3468 struct list_head *head;
3469 struct btrfs_device *dev;
3470 struct btrfs_super_block *sb;
3471 struct btrfs_dev_item *dev_item;
3472 int ret;
3473 int do_barriers;
3474 int max_errors;
3475 int total_errors = 0;
3476 u64 flags;
3477
3478 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3479 backup_super_roots(root->fs_info);
3480
3481 sb = root->fs_info->super_for_commit;
3482 dev_item = &sb->dev_item;
3483
3484 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3485 head = &root->fs_info->fs_devices->devices;
3486 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3487
3488 if (do_barriers) {
3489 ret = barrier_all_devices(root->fs_info);
3490 if (ret) {
3491 mutex_unlock(
3492 &root->fs_info->fs_devices->device_list_mutex);
3493 btrfs_error(root->fs_info, ret,
3494 "errors while submitting device barriers.");
3495 return ret;
3496 }
3497 }
3498
3499 list_for_each_entry_rcu(dev, head, dev_list) {
3500 if (!dev->bdev) {
3501 total_errors++;
3502 continue;
3503 }
3504 if (!dev->in_fs_metadata || !dev->writeable)
3505 continue;
3506
3507 btrfs_set_stack_device_generation(dev_item, 0);
3508 btrfs_set_stack_device_type(dev_item, dev->type);
3509 btrfs_set_stack_device_id(dev_item, dev->devid);
3510 btrfs_set_stack_device_total_bytes(dev_item,
3511 dev->commit_total_bytes);
3512 btrfs_set_stack_device_bytes_used(dev_item,
3513 dev->commit_bytes_used);
3514 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3515 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3516 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3517 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3518 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3519
3520 flags = btrfs_super_flags(sb);
3521 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3522
3523 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3524 if (ret)
3525 total_errors++;
3526 }
3527 if (total_errors > max_errors) {
3528 btrfs_err(root->fs_info, "%d errors while writing supers",
3529 total_errors);
3530 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3531
3532 /* FUA is masked off if unsupported and can't be the reason */
3533 btrfs_error(root->fs_info, -EIO,
3534 "%d errors while writing supers", total_errors);
3535 return -EIO;
3536 }
3537
3538 total_errors = 0;
3539 list_for_each_entry_rcu(dev, head, dev_list) {
3540 if (!dev->bdev)
3541 continue;
3542 if (!dev->in_fs_metadata || !dev->writeable)
3543 continue;
3544
3545 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3546 if (ret)
3547 total_errors++;
3548 }
3549 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3550 if (total_errors > max_errors) {
3551 btrfs_error(root->fs_info, -EIO,
3552 "%d errors while writing supers", total_errors);
3553 return -EIO;
3554 }
3555 return 0;
3556}
3557
3558int write_ctree_super(struct btrfs_trans_handle *trans,
3559 struct btrfs_root *root, int max_mirrors)
3560{
3561 return write_all_supers(root, max_mirrors);
3562}
3563
3564/* Drop a fs root from the radix tree and free it. */
3565void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3566 struct btrfs_root *root)
3567{
3568 spin_lock(&fs_info->fs_roots_radix_lock);
3569 radix_tree_delete(&fs_info->fs_roots_radix,
3570 (unsigned long)root->root_key.objectid);
3571 spin_unlock(&fs_info->fs_roots_radix_lock);
3572
3573 if (btrfs_root_refs(&root->root_item) == 0)
3574 synchronize_srcu(&fs_info->subvol_srcu);
3575
3576 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3577 btrfs_free_log(NULL, root);
3578
3579 if (root->free_ino_pinned)
3580 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3581 if (root->free_ino_ctl)
3582 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3583 free_fs_root(root);
3584}
3585
3586static void free_fs_root(struct btrfs_root *root)
3587{
3588 iput(root->ino_cache_inode);
3589 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3590 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3591 root->orphan_block_rsv = NULL;
3592 if (root->anon_dev)
3593 free_anon_bdev(root->anon_dev);
3594 if (root->subv_writers)
3595 btrfs_free_subvolume_writers(root->subv_writers);
3596 free_extent_buffer(root->node);
3597 free_extent_buffer(root->commit_root);
3598 kfree(root->free_ino_ctl);
3599 kfree(root->free_ino_pinned);
3600 kfree(root->name);
3601 btrfs_put_fs_root(root);
3602}
3603
3604void btrfs_free_fs_root(struct btrfs_root *root)
3605{
3606 free_fs_root(root);
3607}
3608
3609int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3610{
3611 u64 root_objectid = 0;
3612 struct btrfs_root *gang[8];
3613 int i = 0;
3614 int err = 0;
3615 unsigned int ret = 0;
3616 int index;
3617
3618 while (1) {
3619 index = srcu_read_lock(&fs_info->subvol_srcu);
3620 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3621 (void **)gang, root_objectid,
3622 ARRAY_SIZE(gang));
3623 if (!ret) {
3624 srcu_read_unlock(&fs_info->subvol_srcu, index);
3625 break;
3626 }
3627 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3628
3629 for (i = 0; i < ret; i++) {
3630 /* Avoid to grab roots in dead_roots */
3631 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3632 gang[i] = NULL;
3633 continue;
3634 }
3635 /* grab all the search result for later use */
3636 gang[i] = btrfs_grab_fs_root(gang[i]);
3637 }
3638 srcu_read_unlock(&fs_info->subvol_srcu, index);
3639
3640 for (i = 0; i < ret; i++) {
3641 if (!gang[i])
3642 continue;
3643 root_objectid = gang[i]->root_key.objectid;
3644 err = btrfs_orphan_cleanup(gang[i]);
3645 if (err)
3646 break;
3647 btrfs_put_fs_root(gang[i]);
3648 }
3649 root_objectid++;
3650 }
3651
3652 /* release the uncleaned roots due to error */
3653 for (; i < ret; i++) {
3654 if (gang[i])
3655 btrfs_put_fs_root(gang[i]);
3656 }
3657 return err;
3658}
3659
3660int btrfs_commit_super(struct btrfs_root *root)
3661{
3662 struct btrfs_trans_handle *trans;
3663
3664 mutex_lock(&root->fs_info->cleaner_mutex);
3665 btrfs_run_delayed_iputs(root);
3666 mutex_unlock(&root->fs_info->cleaner_mutex);
3667 wake_up_process(root->fs_info->cleaner_kthread);
3668
3669 /* wait until ongoing cleanup work done */
3670 down_write(&root->fs_info->cleanup_work_sem);
3671 up_write(&root->fs_info->cleanup_work_sem);
3672
3673 trans = btrfs_join_transaction(root);
3674 if (IS_ERR(trans))
3675 return PTR_ERR(trans);
3676 return btrfs_commit_transaction(trans, root);
3677}
3678
3679void close_ctree(struct btrfs_root *root)
3680{
3681 struct btrfs_fs_info *fs_info = root->fs_info;
3682 int ret;
3683
3684 fs_info->closing = 1;
3685 smp_mb();
3686
3687 /* wait for the uuid_scan task to finish */
3688 down(&fs_info->uuid_tree_rescan_sem);
3689 /* avoid complains from lockdep et al., set sem back to initial state */
3690 up(&fs_info->uuid_tree_rescan_sem);
3691
3692 /* pause restriper - we want to resume on mount */
3693 btrfs_pause_balance(fs_info);
3694
3695 btrfs_dev_replace_suspend_for_unmount(fs_info);
3696
3697 btrfs_scrub_cancel(fs_info);
3698
3699 /* wait for any defraggers to finish */
3700 wait_event(fs_info->transaction_wait,
3701 (atomic_read(&fs_info->defrag_running) == 0));
3702
3703 /* clear out the rbtree of defraggable inodes */
3704 btrfs_cleanup_defrag_inodes(fs_info);
3705
3706 cancel_work_sync(&fs_info->async_reclaim_work);
3707
3708 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3709 ret = btrfs_commit_super(root);
3710 if (ret)
3711 btrfs_err(fs_info, "commit super ret %d", ret);
3712 }
3713
3714 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3715 btrfs_error_commit_super(root);
3716
3717 kthread_stop(fs_info->transaction_kthread);
3718 kthread_stop(fs_info->cleaner_kthread);
3719
3720 fs_info->closing = 2;
3721 smp_mb();
3722
3723 btrfs_free_qgroup_config(fs_info);
3724
3725 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3726 btrfs_info(fs_info, "at unmount delalloc count %lld",
3727 percpu_counter_sum(&fs_info->delalloc_bytes));
3728 }
3729
3730 btrfs_sysfs_remove_one(fs_info);
3731
3732 btrfs_free_fs_roots(fs_info);
3733
3734 btrfs_put_block_group_cache(fs_info);
3735
3736 btrfs_free_block_groups(fs_info);
3737
3738 /*
3739 * we must make sure there is not any read request to
3740 * submit after we stopping all workers.
3741 */
3742 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3743 btrfs_stop_all_workers(fs_info);
3744
3745 fs_info->open = 0;
3746 free_root_pointers(fs_info, 1);
3747
3748 iput(fs_info->btree_inode);
3749
3750#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3751 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3752 btrfsic_unmount(root, fs_info->fs_devices);
3753#endif
3754
3755 btrfs_close_devices(fs_info->fs_devices);
3756 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3757
3758 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3759 percpu_counter_destroy(&fs_info->delalloc_bytes);
3760 percpu_counter_destroy(&fs_info->bio_counter);
3761 bdi_destroy(&fs_info->bdi);
3762 cleanup_srcu_struct(&fs_info->subvol_srcu);
3763
3764 btrfs_free_stripe_hash_table(fs_info);
3765
3766 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3767 root->orphan_block_rsv = NULL;
3768
3769 lock_chunks(root);
3770 while (!list_empty(&fs_info->pinned_chunks)) {
3771 struct extent_map *em;
3772
3773 em = list_first_entry(&fs_info->pinned_chunks,
3774 struct extent_map, list);
3775 list_del_init(&em->list);
3776 free_extent_map(em);
3777 }
3778 unlock_chunks(root);
3779}
3780
3781int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3782 int atomic)
3783{
3784 int ret;
3785 struct inode *btree_inode = buf->pages[0]->mapping->host;
3786
3787 ret = extent_buffer_uptodate(buf);
3788 if (!ret)
3789 return ret;
3790
3791 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3792 parent_transid, atomic);
3793 if (ret == -EAGAIN)
3794 return ret;
3795 return !ret;
3796}
3797
3798int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3799{
3800 return set_extent_buffer_uptodate(buf);
3801}
3802
3803void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3804{
3805 struct btrfs_root *root;
3806 u64 transid = btrfs_header_generation(buf);
3807 int was_dirty;
3808
3809#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3810 /*
3811 * This is a fast path so only do this check if we have sanity tests
3812 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3813 * outside of the sanity tests.
3814 */
3815 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3816 return;
3817#endif
3818 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3819 btrfs_assert_tree_locked(buf);
3820 if (transid != root->fs_info->generation)
3821 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3822 "found %llu running %llu\n",
3823 buf->start, transid, root->fs_info->generation);
3824 was_dirty = set_extent_buffer_dirty(buf);
3825 if (!was_dirty)
3826 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3827 buf->len,
3828 root->fs_info->dirty_metadata_batch);
3829#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3830 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3831 btrfs_print_leaf(root, buf);
3832 ASSERT(0);
3833 }
3834#endif
3835}
3836
3837static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3838 int flush_delayed)
3839{
3840 /*
3841 * looks as though older kernels can get into trouble with
3842 * this code, they end up stuck in balance_dirty_pages forever
3843 */
3844 int ret;
3845
3846 if (current->flags & PF_MEMALLOC)
3847 return;
3848
3849 if (flush_delayed)
3850 btrfs_balance_delayed_items(root);
3851
3852 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3853 BTRFS_DIRTY_METADATA_THRESH);
3854 if (ret > 0) {
3855 balance_dirty_pages_ratelimited(
3856 root->fs_info->btree_inode->i_mapping);
3857 }
3858 return;
3859}
3860
3861void btrfs_btree_balance_dirty(struct btrfs_root *root)
3862{
3863 __btrfs_btree_balance_dirty(root, 1);
3864}
3865
3866void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3867{
3868 __btrfs_btree_balance_dirty(root, 0);
3869}
3870
3871int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3872{
3873 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3874 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3875}
3876
3877static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3878 int read_only)
3879{
3880 struct btrfs_super_block *sb = fs_info->super_copy;
3881 int ret = 0;
3882
3883 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3884 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3885 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3886 ret = -EINVAL;
3887 }
3888 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3889 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3890 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3891 ret = -EINVAL;
3892 }
3893 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3894 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3895 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3896 ret = -EINVAL;
3897 }
3898
3899 /*
3900 * The common minimum, we don't know if we can trust the nodesize/sectorsize
3901 * items yet, they'll be verified later. Issue just a warning.
3902 */
3903 if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
3904 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
3905 btrfs_super_root(sb));
3906 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
3907 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
3908 btrfs_super_chunk_root(sb));
3909 if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
3910 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
3911 btrfs_super_log_root(sb));
3912
3913 /*
3914 * Check the lower bound, the alignment and other constraints are
3915 * checked later.
3916 */
3917 if (btrfs_super_nodesize(sb) < 4096) {
3918 printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n",
3919 btrfs_super_nodesize(sb));
3920 ret = -EINVAL;
3921 }
3922 if (btrfs_super_sectorsize(sb) < 4096) {
3923 printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n",
3924 btrfs_super_sectorsize(sb));
3925 ret = -EINVAL;
3926 }
3927
3928 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
3929 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
3930 fs_info->fsid, sb->dev_item.fsid);
3931 ret = -EINVAL;
3932 }
3933
3934 /*
3935 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3936 * done later
3937 */
3938 if (btrfs_super_num_devices(sb) > (1UL << 31))
3939 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
3940 btrfs_super_num_devices(sb));
3941 if (btrfs_super_num_devices(sb) == 0) {
3942 printk(KERN_ERR "BTRFS: number of devices is 0\n");
3943 ret = -EINVAL;
3944 }
3945
3946 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
3947 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
3948 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
3949 ret = -EINVAL;
3950 }
3951
3952 /*
3953 * Obvious sys_chunk_array corruptions, it must hold at least one key
3954 * and one chunk
3955 */
3956 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
3957 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
3958 btrfs_super_sys_array_size(sb),
3959 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
3960 ret = -EINVAL;
3961 }
3962 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
3963 + sizeof(struct btrfs_chunk)) {
3964 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
3965 btrfs_super_sys_array_size(sb),
3966 sizeof(struct btrfs_disk_key)
3967 + sizeof(struct btrfs_chunk));
3968 ret = -EINVAL;
3969 }
3970
3971 /*
3972 * The generation is a global counter, we'll trust it more than the others
3973 * but it's still possible that it's the one that's wrong.
3974 */
3975 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
3976 printk(KERN_WARNING
3977 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
3978 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
3979 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
3980 && btrfs_super_cache_generation(sb) != (u64)-1)
3981 printk(KERN_WARNING
3982 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
3983 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
3984
3985 return ret;
3986}
3987
3988static void btrfs_error_commit_super(struct btrfs_root *root)
3989{
3990 mutex_lock(&root->fs_info->cleaner_mutex);
3991 btrfs_run_delayed_iputs(root);
3992 mutex_unlock(&root->fs_info->cleaner_mutex);
3993
3994 down_write(&root->fs_info->cleanup_work_sem);
3995 up_write(&root->fs_info->cleanup_work_sem);
3996
3997 /* cleanup FS via transaction */
3998 btrfs_cleanup_transaction(root);
3999}
4000
4001static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4002{
4003 struct btrfs_ordered_extent *ordered;
4004
4005 spin_lock(&root->ordered_extent_lock);
4006 /*
4007 * This will just short circuit the ordered completion stuff which will
4008 * make sure the ordered extent gets properly cleaned up.
4009 */
4010 list_for_each_entry(ordered, &root->ordered_extents,
4011 root_extent_list)
4012 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4013 spin_unlock(&root->ordered_extent_lock);
4014}
4015
4016static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4017{
4018 struct btrfs_root *root;
4019 struct list_head splice;
4020
4021 INIT_LIST_HEAD(&splice);
4022
4023 spin_lock(&fs_info->ordered_root_lock);
4024 list_splice_init(&fs_info->ordered_roots, &splice);
4025 while (!list_empty(&splice)) {
4026 root = list_first_entry(&splice, struct btrfs_root,
4027 ordered_root);
4028 list_move_tail(&root->ordered_root,
4029 &fs_info->ordered_roots);
4030
4031 spin_unlock(&fs_info->ordered_root_lock);
4032 btrfs_destroy_ordered_extents(root);
4033
4034 cond_resched();
4035 spin_lock(&fs_info->ordered_root_lock);
4036 }
4037 spin_unlock(&fs_info->ordered_root_lock);
4038}
4039
4040static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4041 struct btrfs_root *root)
4042{
4043 struct rb_node *node;
4044 struct btrfs_delayed_ref_root *delayed_refs;
4045 struct btrfs_delayed_ref_node *ref;
4046 int ret = 0;
4047
4048 delayed_refs = &trans->delayed_refs;
4049
4050 spin_lock(&delayed_refs->lock);
4051 if (atomic_read(&delayed_refs->num_entries) == 0) {
4052 spin_unlock(&delayed_refs->lock);
4053 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4054 return ret;
4055 }
4056
4057 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4058 struct btrfs_delayed_ref_head *head;
4059 bool pin_bytes = false;
4060
4061 head = rb_entry(node, struct btrfs_delayed_ref_head,
4062 href_node);
4063 if (!mutex_trylock(&head->mutex)) {
4064 atomic_inc(&head->node.refs);
4065 spin_unlock(&delayed_refs->lock);
4066
4067 mutex_lock(&head->mutex);
4068 mutex_unlock(&head->mutex);
4069 btrfs_put_delayed_ref(&head->node);
4070 spin_lock(&delayed_refs->lock);
4071 continue;
4072 }
4073 spin_lock(&head->lock);
4074 while ((node = rb_first(&head->ref_root)) != NULL) {
4075 ref = rb_entry(node, struct btrfs_delayed_ref_node,
4076 rb_node);
4077 ref->in_tree = 0;
4078 rb_erase(&ref->rb_node, &head->ref_root);
4079 atomic_dec(&delayed_refs->num_entries);
4080 btrfs_put_delayed_ref(ref);
4081 }
4082 if (head->must_insert_reserved)
4083 pin_bytes = true;
4084 btrfs_free_delayed_extent_op(head->extent_op);
4085 delayed_refs->num_heads--;
4086 if (head->processing == 0)
4087 delayed_refs->num_heads_ready--;
4088 atomic_dec(&delayed_refs->num_entries);
4089 head->node.in_tree = 0;
4090 rb_erase(&head->href_node, &delayed_refs->href_root);
4091 spin_unlock(&head->lock);
4092 spin_unlock(&delayed_refs->lock);
4093 mutex_unlock(&head->mutex);
4094
4095 if (pin_bytes)
4096 btrfs_pin_extent(root, head->node.bytenr,
4097 head->node.num_bytes, 1);
4098 btrfs_put_delayed_ref(&head->node);
4099 cond_resched();
4100 spin_lock(&delayed_refs->lock);
4101 }
4102
4103 spin_unlock(&delayed_refs->lock);
4104
4105 return ret;
4106}
4107
4108static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4109{
4110 struct btrfs_inode *btrfs_inode;
4111 struct list_head splice;
4112
4113 INIT_LIST_HEAD(&splice);
4114
4115 spin_lock(&root->delalloc_lock);
4116 list_splice_init(&root->delalloc_inodes, &splice);
4117
4118 while (!list_empty(&splice)) {
4119 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4120 delalloc_inodes);
4121
4122 list_del_init(&btrfs_inode->delalloc_inodes);
4123 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4124 &btrfs_inode->runtime_flags);
4125 spin_unlock(&root->delalloc_lock);
4126
4127 btrfs_invalidate_inodes(btrfs_inode->root);
4128
4129 spin_lock(&root->delalloc_lock);
4130 }
4131
4132 spin_unlock(&root->delalloc_lock);
4133}
4134
4135static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4136{
4137 struct btrfs_root *root;
4138 struct list_head splice;
4139
4140 INIT_LIST_HEAD(&splice);
4141
4142 spin_lock(&fs_info->delalloc_root_lock);
4143 list_splice_init(&fs_info->delalloc_roots, &splice);
4144 while (!list_empty(&splice)) {
4145 root = list_first_entry(&splice, struct btrfs_root,
4146 delalloc_root);
4147 list_del_init(&root->delalloc_root);
4148 root = btrfs_grab_fs_root(root);
4149 BUG_ON(!root);
4150 spin_unlock(&fs_info->delalloc_root_lock);
4151
4152 btrfs_destroy_delalloc_inodes(root);
4153 btrfs_put_fs_root(root);
4154
4155 spin_lock(&fs_info->delalloc_root_lock);
4156 }
4157 spin_unlock(&fs_info->delalloc_root_lock);
4158}
4159
4160static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4161 struct extent_io_tree *dirty_pages,
4162 int mark)
4163{
4164 int ret;
4165 struct extent_buffer *eb;
4166 u64 start = 0;
4167 u64 end;
4168
4169 while (1) {
4170 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4171 mark, NULL);
4172 if (ret)
4173 break;
4174
4175 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4176 while (start <= end) {
4177 eb = btrfs_find_tree_block(root->fs_info, start);
4178 start += root->nodesize;
4179 if (!eb)
4180 continue;
4181 wait_on_extent_buffer_writeback(eb);
4182
4183 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4184 &eb->bflags))
4185 clear_extent_buffer_dirty(eb);
4186 free_extent_buffer_stale(eb);
4187 }
4188 }
4189
4190 return ret;
4191}
4192
4193static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4194 struct extent_io_tree *pinned_extents)
4195{
4196 struct extent_io_tree *unpin;
4197 u64 start;
4198 u64 end;
4199 int ret;
4200 bool loop = true;
4201
4202 unpin = pinned_extents;
4203again:
4204 while (1) {
4205 ret = find_first_extent_bit(unpin, 0, &start, &end,
4206 EXTENT_DIRTY, NULL);
4207 if (ret)
4208 break;
4209
4210 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4211 btrfs_error_unpin_extent_range(root, start, end);
4212 cond_resched();
4213 }
4214
4215 if (loop) {
4216 if (unpin == &root->fs_info->freed_extents[0])
4217 unpin = &root->fs_info->freed_extents[1];
4218 else
4219 unpin = &root->fs_info->freed_extents[0];
4220 loop = false;
4221 goto again;
4222 }
4223
4224 return 0;
4225}
4226
4227static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans,
4228 struct btrfs_fs_info *fs_info)
4229{
4230 struct btrfs_ordered_extent *ordered;
4231
4232 spin_lock(&fs_info->trans_lock);
4233 while (!list_empty(&cur_trans->pending_ordered)) {
4234 ordered = list_first_entry(&cur_trans->pending_ordered,
4235 struct btrfs_ordered_extent,
4236 trans_list);
4237 list_del_init(&ordered->trans_list);
4238 spin_unlock(&fs_info->trans_lock);
4239
4240 btrfs_put_ordered_extent(ordered);
4241 spin_lock(&fs_info->trans_lock);
4242 }
4243 spin_unlock(&fs_info->trans_lock);
4244}
4245
4246void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4247 struct btrfs_root *root)
4248{
4249 btrfs_destroy_delayed_refs(cur_trans, root);
4250
4251 cur_trans->state = TRANS_STATE_COMMIT_START;
4252 wake_up(&root->fs_info->transaction_blocked_wait);
4253
4254 cur_trans->state = TRANS_STATE_UNBLOCKED;
4255 wake_up(&root->fs_info->transaction_wait);
4256
4257 btrfs_free_pending_ordered(cur_trans, root->fs_info);
4258 btrfs_destroy_delayed_inodes(root);
4259 btrfs_assert_delayed_root_empty(root);
4260
4261 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4262 EXTENT_DIRTY);
4263 btrfs_destroy_pinned_extent(root,
4264 root->fs_info->pinned_extents);
4265
4266 cur_trans->state =TRANS_STATE_COMPLETED;
4267 wake_up(&cur_trans->commit_wait);
4268
4269 /*
4270 memset(cur_trans, 0, sizeof(*cur_trans));
4271 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4272 */
4273}
4274
4275static int btrfs_cleanup_transaction(struct btrfs_root *root)
4276{
4277 struct btrfs_transaction *t;
4278
4279 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4280
4281 spin_lock(&root->fs_info->trans_lock);
4282 while (!list_empty(&root->fs_info->trans_list)) {
4283 t = list_first_entry(&root->fs_info->trans_list,
4284 struct btrfs_transaction, list);
4285 if (t->state >= TRANS_STATE_COMMIT_START) {
4286 atomic_inc(&t->use_count);
4287 spin_unlock(&root->fs_info->trans_lock);
4288 btrfs_wait_for_commit(root, t->transid);
4289 btrfs_put_transaction(t);
4290 spin_lock(&root->fs_info->trans_lock);
4291 continue;
4292 }
4293 if (t == root->fs_info->running_transaction) {
4294 t->state = TRANS_STATE_COMMIT_DOING;
4295 spin_unlock(&root->fs_info->trans_lock);
4296 /*
4297 * We wait for 0 num_writers since we don't hold a trans
4298 * handle open currently for this transaction.
4299 */
4300 wait_event(t->writer_wait,
4301 atomic_read(&t->num_writers) == 0);
4302 } else {
4303 spin_unlock(&root->fs_info->trans_lock);
4304 }
4305 btrfs_cleanup_one_transaction(t, root);
4306
4307 spin_lock(&root->fs_info->trans_lock);
4308 if (t == root->fs_info->running_transaction)
4309 root->fs_info->running_transaction = NULL;
4310 list_del_init(&t->list);
4311 spin_unlock(&root->fs_info->trans_lock);
4312
4313 btrfs_put_transaction(t);
4314 trace_btrfs_transaction_commit(root);
4315 spin_lock(&root->fs_info->trans_lock);
4316 }
4317 spin_unlock(&root->fs_info->trans_lock);
4318 btrfs_destroy_all_ordered_extents(root->fs_info);
4319 btrfs_destroy_delayed_inodes(root);
4320 btrfs_assert_delayed_root_empty(root);
4321 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4322 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4323 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4324
4325 return 0;
4326}
4327
4328static const struct extent_io_ops btree_extent_io_ops = {
4329 .readpage_end_io_hook = btree_readpage_end_io_hook,
4330 .readpage_io_failed_hook = btree_io_failed_hook,
4331 .submit_bio_hook = btree_submit_bio_hook,
4332 /* note we're sharing with inode.c for the merge bio hook */
4333 .merge_bio_hook = btrfs_merge_bio_hook,
4334};
Linux 6.x block layer and io_uring tree(s)