projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
btrfs: consistently use fs_info in close_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
2150int open_ctree(struct super_block *sb,
2151 struct btrfs_fs_devices *fs_devices,
2152 char *options)
2153{
2154 u32 sectorsize;
2155 u32 nodesize;
2156 u32 stripesize;
2157 u64 generation;
2158 u64 features;
2159 struct btrfs_key location;
2160 struct buffer_head *bh;
2161 struct btrfs_super_block *disk_super;
2162 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2163 struct btrfs_root *tree_root;
2164 struct btrfs_root *extent_root;
2165 struct btrfs_root *csum_root;
2166 struct btrfs_root *chunk_root;
2167 struct btrfs_root *dev_root;
2168 struct btrfs_root *quota_root;
2169 struct btrfs_root *uuid_root;
2170 struct btrfs_root *log_tree_root;
2171 int ret;
2172 int err = -EINVAL;
2173 int num_backups_tried = 0;
2174 int backup_index = 0;
2175 int max_active;
2176 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2177 bool create_uuid_tree;
2178 bool check_uuid_tree;
2179
2180 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2181 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2182 if (!tree_root || !chunk_root) {
2183 err = -ENOMEM;
2184 goto fail;
2185 }
2186
2187 ret = init_srcu_struct(&fs_info->subvol_srcu);
2188 if (ret) {
2189 err = ret;
2190 goto fail;
2191 }
2192
2193 ret = setup_bdi(fs_info, &fs_info->bdi);
2194 if (ret) {
2195 err = ret;
2196 goto fail_srcu;
2197 }
2198
2199 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2200 if (ret) {
2201 err = ret;
2202 goto fail_bdi;
2203 }
2204 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2205 (1 + ilog2(nr_cpu_ids));
2206
2207 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2208 if (ret) {
2209 err = ret;
2210 goto fail_dirty_metadata_bytes;
2211 }
2212
2213 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2214 if (ret) {
2215 err = ret;
2216 goto fail_delalloc_bytes;
2217 }
2218
2219 fs_info->btree_inode = new_inode(sb);
2220 if (!fs_info->btree_inode) {
2221 err = -ENOMEM;
2222 goto fail_bio_counter;
2223 }
2224
2225 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2226
2227 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2228 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2229 INIT_LIST_HEAD(&fs_info->trans_list);
2230 INIT_LIST_HEAD(&fs_info->dead_roots);
2231 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2232 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2233 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2234 spin_lock_init(&fs_info->delalloc_root_lock);
2235 spin_lock_init(&fs_info->trans_lock);
2236 spin_lock_init(&fs_info->fs_roots_radix_lock);
2237 spin_lock_init(&fs_info->delayed_iput_lock);
2238 spin_lock_init(&fs_info->defrag_inodes_lock);
2239 spin_lock_init(&fs_info->free_chunk_lock);
2240 spin_lock_init(&fs_info->tree_mod_seq_lock);
2241 spin_lock_init(&fs_info->super_lock);
2242 spin_lock_init(&fs_info->qgroup_op_lock);
2243 spin_lock_init(&fs_info->buffer_lock);
2244 spin_lock_init(&fs_info->unused_bgs_lock);
2245 mutex_init(&fs_info->unused_bg_unpin_mutex);
2246 rwlock_init(&fs_info->tree_mod_log_lock);
2247 mutex_init(&fs_info->reloc_mutex);
2248 mutex_init(&fs_info->delalloc_root_mutex);
2249 seqlock_init(&fs_info->profiles_lock);
2250
2251 init_completion(&fs_info->kobj_unregister);
2252 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2253 INIT_LIST_HEAD(&fs_info->space_info);
2254 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2255 INIT_LIST_HEAD(&fs_info->unused_bgs);
2256 btrfs_mapping_init(&fs_info->mapping_tree);
2257 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2258 BTRFS_BLOCK_RSV_GLOBAL);
2259 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2260 BTRFS_BLOCK_RSV_DELALLOC);
2261 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2262 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2263 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2264 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2265 BTRFS_BLOCK_RSV_DELOPS);
2266 atomic_set(&fs_info->nr_async_submits, 0);
2267 atomic_set(&fs_info->async_delalloc_pages, 0);
2268 atomic_set(&fs_info->async_submit_draining, 0);
2269 atomic_set(&fs_info->nr_async_bios, 0);
2270 atomic_set(&fs_info->defrag_running, 0);
2271 atomic_set(&fs_info->qgroup_op_seq, 0);
2272 atomic64_set(&fs_info->tree_mod_seq, 0);
2273 fs_info->sb = sb;
2274 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2275 fs_info->metadata_ratio = 0;
2276 fs_info->defrag_inodes = RB_ROOT;
2277 fs_info->free_chunk_space = 0;
2278 fs_info->tree_mod_log = RB_ROOT;
2279 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2280 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2281 /* readahead state */
2282 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2283 spin_lock_init(&fs_info->reada_lock);
2284
2285 fs_info->thread_pool_size = min_t(unsigned long,
2286 num_online_cpus() + 2, 8);
2287
2288 INIT_LIST_HEAD(&fs_info->ordered_roots);
2289 spin_lock_init(&fs_info->ordered_root_lock);
2290 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2291 GFP_NOFS);
2292 if (!fs_info->delayed_root) {
2293 err = -ENOMEM;
2294 goto fail_iput;
2295 }
2296 btrfs_init_delayed_root(fs_info->delayed_root);
2297
2298 mutex_init(&fs_info->scrub_lock);
2299 atomic_set(&fs_info->scrubs_running, 0);
2300 atomic_set(&fs_info->scrub_pause_req, 0);
2301 atomic_set(&fs_info->scrubs_paused, 0);
2302 atomic_set(&fs_info->scrub_cancel_req, 0);
2303 init_waitqueue_head(&fs_info->replace_wait);
2304 init_waitqueue_head(&fs_info->scrub_pause_wait);
2305 fs_info->scrub_workers_refcnt = 0;
2306#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2307 fs_info->check_integrity_print_mask = 0;
2308#endif
2309
2310 spin_lock_init(&fs_info->balance_lock);
2311 mutex_init(&fs_info->balance_mutex);
2312 atomic_set(&fs_info->balance_running, 0);
2313 atomic_set(&fs_info->balance_pause_req, 0);
2314 atomic_set(&fs_info->balance_cancel_req, 0);
2315 fs_info->balance_ctl = NULL;
2316 init_waitqueue_head(&fs_info->balance_wait_q);
2317 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2318
2319 sb->s_blocksize = 4096;
2320 sb->s_blocksize_bits = blksize_bits(4096);
2321 sb->s_bdi = &fs_info->bdi;
2322
2323 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2324 set_nlink(fs_info->btree_inode, 1);
2325 /*
2326 * we set the i_size on the btree inode to the max possible int.
2327 * the real end of the address space is determined by all of
2328 * the devices in the system
2329 */
2330 fs_info->btree_inode->i_size = OFFSET_MAX;
2331 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2332 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2333
2334 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2335 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2336 fs_info->btree_inode->i_mapping);
2337 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2338 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2339
2340 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2341
2342 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2343 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2344 sizeof(struct btrfs_key));
2345 set_bit(BTRFS_INODE_DUMMY,
2346 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2347 btrfs_insert_inode_hash(fs_info->btree_inode);
2348
2349 spin_lock_init(&fs_info->block_group_cache_lock);
2350 fs_info->block_group_cache_tree = RB_ROOT;
2351 fs_info->first_logical_byte = (u64)-1;
2352
2353 extent_io_tree_init(&fs_info->freed_extents[0],
2354 fs_info->btree_inode->i_mapping);
2355 extent_io_tree_init(&fs_info->freed_extents[1],
2356 fs_info->btree_inode->i_mapping);
2357 fs_info->pinned_extents = &fs_info->freed_extents[0];
2358 fs_info->do_barriers = 1;
2359
2360
2361 mutex_init(&fs_info->ordered_operations_mutex);
2362 mutex_init(&fs_info->ordered_extent_flush_mutex);
2363 mutex_init(&fs_info->tree_log_mutex);
2364 mutex_init(&fs_info->chunk_mutex);
2365 mutex_init(&fs_info->transaction_kthread_mutex);
2366 mutex_init(&fs_info->cleaner_mutex);
2367 mutex_init(&fs_info->volume_mutex);
2368 init_rwsem(&fs_info->commit_root_sem);
2369 init_rwsem(&fs_info->cleanup_work_sem);
2370 init_rwsem(&fs_info->subvol_sem);
2371 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2372 fs_info->dev_replace.lock_owner = 0;
2373 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2374 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2375 mutex_init(&fs_info->dev_replace.lock_management_lock);
2376 mutex_init(&fs_info->dev_replace.lock);
2377
2378 spin_lock_init(&fs_info->qgroup_lock);
2379 mutex_init(&fs_info->qgroup_ioctl_lock);
2380 fs_info->qgroup_tree = RB_ROOT;
2381 fs_info->qgroup_op_tree = RB_ROOT;
2382 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2383 fs_info->qgroup_seq = 1;
2384 fs_info->quota_enabled = 0;
2385 fs_info->pending_quota_state = 0;
2386 fs_info->qgroup_ulist = NULL;
2387 mutex_init(&fs_info->qgroup_rescan_lock);
2388
2389 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2390 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2391
2392 init_waitqueue_head(&fs_info->transaction_throttle);
2393 init_waitqueue_head(&fs_info->transaction_wait);
2394 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2395 init_waitqueue_head(&fs_info->async_submit_wait);
2396
2397 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2398
2399 ret = btrfs_alloc_stripe_hash_table(fs_info);
2400 if (ret) {
2401 err = ret;
2402 goto fail_alloc;
2403 }
2404
2405 __setup_root(4096, 4096, 4096, tree_root,
2406 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2407
2408 invalidate_bdev(fs_devices->latest_bdev);
2409
2410 /*
2411 * Read super block and check the signature bytes only
2412 */
2413 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2414 if (!bh) {
2415 err = -EINVAL;
2416 goto fail_alloc;
2417 }
2418
2419 /*
2420 * We want to check superblock checksum, the type is stored inside.
2421 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2422 */
2423 if (btrfs_check_super_csum(bh->b_data)) {
2424 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2425 err = -EINVAL;
2426 goto fail_alloc;
2427 }
2428
2429 /*
2430 * super_copy is zeroed at allocation time and we never touch the
2431 * following bytes up to INFO_SIZE, the checksum is calculated from
2432 * the whole block of INFO_SIZE
2433 */
2434 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2435 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2436 sizeof(*fs_info->super_for_commit));
2437 brelse(bh);
2438
2439 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2440
2441 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2442 if (ret) {
2443 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2444 err = -EINVAL;
2445 goto fail_alloc;
2446 }
2447
2448 disk_super = fs_info->super_copy;
2449 if (!btrfs_super_root(disk_super))
2450 goto fail_alloc;
2451
2452 /* check FS state, whether FS is broken. */
2453 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2454 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2455
2456 /*
2457 * run through our array of backup supers and setup
2458 * our ring pointer to the oldest one
2459 */
2460 generation = btrfs_super_generation(disk_super);
2461 find_oldest_super_backup(fs_info, generation);
2462
2463 /*
2464 * In the long term, we'll store the compression type in the super
2465 * block, and it'll be used for per file compression control.
2466 */
2467 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2468
2469 ret = btrfs_parse_options(tree_root, options);
2470 if (ret) {
2471 err = ret;
2472 goto fail_alloc;
2473 }
2474
2475 features = btrfs_super_incompat_flags(disk_super) &
2476 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2477 if (features) {
2478 printk(KERN_ERR "BTRFS: couldn't mount because of "
2479 "unsupported optional features (%Lx).\n",
2480 features);
2481 err = -EINVAL;
2482 goto fail_alloc;
2483 }
2484
2485 /*
2486 * Leafsize and nodesize were always equal, this is only a sanity check.
2487 */
2488 if (le32_to_cpu(disk_super->__unused_leafsize) !=
2489 btrfs_super_nodesize(disk_super)) {
2490 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2491 "blocksizes don't match. node %d leaf %d\n",
2492 btrfs_super_nodesize(disk_super),
2493 le32_to_cpu(disk_super->__unused_leafsize));
2494 err = -EINVAL;
2495 goto fail_alloc;
2496 }
2497 if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2498 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2499 "blocksize (%d) was too large\n",
2500 btrfs_super_nodesize(disk_super));
2501 err = -EINVAL;
2502 goto fail_alloc;
2503 }
2504
2505 features = btrfs_super_incompat_flags(disk_super);
2506 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2507 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2508 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2509
2510 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2511 printk(KERN_INFO "BTRFS: has skinny extents\n");
2512
2513 /*
2514 * flag our filesystem as having big metadata blocks if
2515 * they are bigger than the page size
2516 */
2517 if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2518 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2519 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2520 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2521 }
2522
2523 nodesize = btrfs_super_nodesize(disk_super);
2524 sectorsize = btrfs_super_sectorsize(disk_super);
2525 stripesize = btrfs_super_stripesize(disk_super);
2526 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2527 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2528
2529 /*
2530 * mixed block groups end up with duplicate but slightly offset
2531 * extent buffers for the same range. It leads to corruptions
2532 */
2533 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2534 (sectorsize != nodesize)) {
2535 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2536 "are not allowed for mixed block groups on %s\n",
2537 sb->s_id);
2538 goto fail_alloc;
2539 }
2540
2541 /*
2542 * Needn't use the lock because there is no other task which will
2543 * update the flag.
2544 */
2545 btrfs_set_super_incompat_flags(disk_super, features);
2546
2547 features = btrfs_super_compat_ro_flags(disk_super) &
2548 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2549 if (!(sb->s_flags & MS_RDONLY) && features) {
2550 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2551 "unsupported option features (%Lx).\n",
2552 features);
2553 err = -EINVAL;
2554 goto fail_alloc;
2555 }
2556
2557 max_active = fs_info->thread_pool_size;
2558
2559 fs_info->workers =
2560 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2561 max_active, 16);
2562
2563 fs_info->delalloc_workers =
2564 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2565
2566 fs_info->flush_workers =
2567 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2568
2569 fs_info->caching_workers =
2570 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2571
2572 /*
2573 * a higher idle thresh on the submit workers makes it much more
2574 * likely that bios will be send down in a sane order to the
2575 * devices
2576 */
2577 fs_info->submit_workers =
2578 btrfs_alloc_workqueue("submit", flags,
2579 min_t(u64, fs_devices->num_devices,
2580 max_active), 64);
2581
2582 fs_info->fixup_workers =
2583 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2584
2585 /*
2586 * endios are largely parallel and should have a very
2587 * low idle thresh
2588 */
2589 fs_info->endio_workers =
2590 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2591 fs_info->endio_meta_workers =
2592 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2593 fs_info->endio_meta_write_workers =
2594 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2595 fs_info->endio_raid56_workers =
2596 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2597 fs_info->endio_repair_workers =
2598 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2599 fs_info->rmw_workers =
2600 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2601 fs_info->endio_write_workers =
2602 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2603 fs_info->endio_freespace_worker =
2604 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2605 fs_info->delayed_workers =
2606 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2607 fs_info->readahead_workers =
2608 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2609 fs_info->qgroup_rescan_workers =
2610 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2611 fs_info->extent_workers =
2612 btrfs_alloc_workqueue("extent-refs", flags,
2613 min_t(u64, fs_devices->num_devices,
2614 max_active), 8);
2615
2616 if (!(fs_info->workers && fs_info->delalloc_workers &&
2617 fs_info->submit_workers && fs_info->flush_workers &&
2618 fs_info->endio_workers && fs_info->endio_meta_workers &&
2619 fs_info->endio_meta_write_workers &&
2620 fs_info->endio_repair_workers &&
2621 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2622 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2623 fs_info->caching_workers && fs_info->readahead_workers &&
2624 fs_info->fixup_workers && fs_info->delayed_workers &&
2625 fs_info->extent_workers &&
2626 fs_info->qgroup_rescan_workers)) {
2627 err = -ENOMEM;
2628 goto fail_sb_buffer;
2629 }
2630
2631 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2632 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2633 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2634
2635 tree_root->nodesize = nodesize;
2636 tree_root->sectorsize = sectorsize;
2637 tree_root->stripesize = stripesize;
2638
2639 sb->s_blocksize = sectorsize;
2640 sb->s_blocksize_bits = blksize_bits(sectorsize);
2641
2642 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2643 printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id);
2644 goto fail_sb_buffer;
2645 }
2646
2647 if (sectorsize != PAGE_SIZE) {
2648 printk(KERN_ERR "BTRFS: incompatible sector size (%lu) "
2649 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2650 goto fail_sb_buffer;
2651 }
2652
2653 mutex_lock(&fs_info->chunk_mutex);
2654 ret = btrfs_read_sys_array(tree_root);
2655 mutex_unlock(&fs_info->chunk_mutex);
2656 if (ret) {
2657 printk(KERN_ERR "BTRFS: failed to read the system "
2658 "array on %s\n", sb->s_id);
2659 goto fail_sb_buffer;
2660 }
2661
2662 generation = btrfs_super_chunk_root_generation(disk_super);
2663
2664 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2665 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2666
2667 chunk_root->node = read_tree_block(chunk_root,
2668 btrfs_super_chunk_root(disk_super),
2669 generation);
2670 if (!chunk_root->node ||
2671 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2672 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2673 sb->s_id);
2674 goto fail_tree_roots;
2675 }
2676 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2677 chunk_root->commit_root = btrfs_root_node(chunk_root);
2678
2679 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2680 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2681
2682 ret = btrfs_read_chunk_tree(chunk_root);
2683 if (ret) {
2684 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2685 sb->s_id);
2686 goto fail_tree_roots;
2687 }
2688
2689 /*
2690 * keep the device that is marked to be the target device for the
2691 * dev_replace procedure
2692 */
2693 btrfs_close_extra_devices(fs_devices, 0);
2694
2695 if (!fs_devices->latest_bdev) {
2696 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2697 sb->s_id);
2698 goto fail_tree_roots;
2699 }
2700
2701retry_root_backup:
2702 generation = btrfs_super_generation(disk_super);
2703
2704 tree_root->node = read_tree_block(tree_root,
2705 btrfs_super_root(disk_super),
2706 generation);
2707 if (!tree_root->node ||
2708 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2709 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2710 sb->s_id);
2711
2712 goto recovery_tree_root;
2713 }
2714
2715 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2716 tree_root->commit_root = btrfs_root_node(tree_root);
2717 btrfs_set_root_refs(&tree_root->root_item, 1);
2718
2719 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2720 location.type = BTRFS_ROOT_ITEM_KEY;
2721 location.offset = 0;
2722
2723 extent_root = btrfs_read_tree_root(tree_root, &location);
2724 if (IS_ERR(extent_root)) {
2725 ret = PTR_ERR(extent_root);
2726 goto recovery_tree_root;
2727 }
2728 set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2729 fs_info->extent_root = extent_root;
2730
2731 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2732 dev_root = btrfs_read_tree_root(tree_root, &location);
2733 if (IS_ERR(dev_root)) {
2734 ret = PTR_ERR(dev_root);
2735 goto recovery_tree_root;
2736 }
2737 set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2738 fs_info->dev_root = dev_root;
2739 btrfs_init_devices_late(fs_info);
2740
2741 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2742 csum_root = btrfs_read_tree_root(tree_root, &location);
2743 if (IS_ERR(csum_root)) {
2744 ret = PTR_ERR(csum_root);
2745 goto recovery_tree_root;
2746 }
2747 set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2748 fs_info->csum_root = csum_root;
2749
2750 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2751 quota_root = btrfs_read_tree_root(tree_root, &location);
2752 if (!IS_ERR(quota_root)) {
2753 set_bit(BTRFS_ROOT_TRACK_DIRTY, &quota_root->state);
2754 fs_info->quota_enabled = 1;
2755 fs_info->pending_quota_state = 1;
2756 fs_info->quota_root = quota_root;
2757 }
2758
2759 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2760 uuid_root = btrfs_read_tree_root(tree_root, &location);
2761 if (IS_ERR(uuid_root)) {
2762 ret = PTR_ERR(uuid_root);
2763 if (ret != -ENOENT)
2764 goto recovery_tree_root;
2765 create_uuid_tree = true;
2766 check_uuid_tree = false;
2767 } else {
2768 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2769 fs_info->uuid_root = uuid_root;
2770 create_uuid_tree = false;
2771 check_uuid_tree =
2772 generation != btrfs_super_uuid_tree_generation(disk_super);
2773 }
2774
2775 fs_info->generation = generation;
2776 fs_info->last_trans_committed = generation;
2777
2778 ret = btrfs_recover_balance(fs_info);
2779 if (ret) {
2780 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2781 goto fail_block_groups;
2782 }
2783
2784 ret = btrfs_init_dev_stats(fs_info);
2785 if (ret) {
2786 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2787 ret);
2788 goto fail_block_groups;
2789 }
2790
2791 ret = btrfs_init_dev_replace(fs_info);
2792 if (ret) {
2793 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2794 goto fail_block_groups;
2795 }
2796
2797 btrfs_close_extra_devices(fs_devices, 1);
2798
2799 ret = btrfs_sysfs_add_one(fs_info);
2800 if (ret) {
2801 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2802 goto fail_block_groups;
2803 }
2804
2805 ret = btrfs_init_space_info(fs_info);
2806 if (ret) {
2807 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2808 goto fail_sysfs;
2809 }
2810
2811 ret = btrfs_read_block_groups(extent_root);
2812 if (ret) {
2813 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2814 goto fail_sysfs;
2815 }
2816 fs_info->num_tolerated_disk_barrier_failures =
2817 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2818 if (fs_info->fs_devices->missing_devices >
2819 fs_info->num_tolerated_disk_barrier_failures &&
2820 !(sb->s_flags & MS_RDONLY)) {
2821 printk(KERN_WARNING "BTRFS: "
2822 "too many missing devices, writeable mount is not allowed\n");
2823 goto fail_sysfs;
2824 }
2825
2826 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2827 "btrfs-cleaner");
2828 if (IS_ERR(fs_info->cleaner_kthread))
2829 goto fail_sysfs;
2830
2831 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2832 tree_root,
2833 "btrfs-transaction");
2834 if (IS_ERR(fs_info->transaction_kthread))
2835 goto fail_cleaner;
2836
2837 if (!btrfs_test_opt(tree_root, SSD) &&
2838 !btrfs_test_opt(tree_root, NOSSD) &&
2839 !fs_info->fs_devices->rotating) {
2840 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2841 "mode\n");
2842 btrfs_set_opt(fs_info->mount_opt, SSD);
2843 }
2844
2845 /*
2846 * Mount does not set all options immediatelly, we can do it now and do
2847 * not have to wait for transaction commit
2848 */
2849 btrfs_apply_pending_changes(fs_info);
2850
2851#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2852 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2853 ret = btrfsic_mount(tree_root, fs_devices,
2854 btrfs_test_opt(tree_root,
2855 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2856 1 : 0,
2857 fs_info->check_integrity_print_mask);
2858 if (ret)
2859 printk(KERN_WARNING "BTRFS: failed to initialize"
2860 " integrity check module %s\n", sb->s_id);
2861 }
2862#endif
2863 ret = btrfs_read_qgroup_config(fs_info);
2864 if (ret)
2865 goto fail_trans_kthread;
2866
2867 /* do not make disk changes in broken FS */
2868 if (btrfs_super_log_root(disk_super) != 0) {
2869 u64 bytenr = btrfs_super_log_root(disk_super);
2870
2871 if (fs_devices->rw_devices == 0) {
2872 printk(KERN_WARNING "BTRFS: log replay required "
2873 "on RO media\n");
2874 err = -EIO;
2875 goto fail_qgroup;
2876 }
2877
2878 log_tree_root = btrfs_alloc_root(fs_info);
2879 if (!log_tree_root) {
2880 err = -ENOMEM;
2881 goto fail_qgroup;
2882 }
2883
2884 __setup_root(nodesize, sectorsize, stripesize,
2885 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2886
2887 log_tree_root->node = read_tree_block(tree_root, bytenr,
2888 generation + 1);
2889 if (!log_tree_root->node ||
2890 !extent_buffer_uptodate(log_tree_root->node)) {
2891 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2892 free_extent_buffer(log_tree_root->node);
2893 kfree(log_tree_root);
2894 goto fail_qgroup;
2895 }
2896 /* returns with log_tree_root freed on success */
2897 ret = btrfs_recover_log_trees(log_tree_root);
2898 if (ret) {
2899 btrfs_error(tree_root->fs_info, ret,
2900 "Failed to recover log tree");
2901 free_extent_buffer(log_tree_root->node);
2902 kfree(log_tree_root);
2903 goto fail_qgroup;
2904 }
2905
2906 if (sb->s_flags & MS_RDONLY) {
2907 ret = btrfs_commit_super(tree_root);
2908 if (ret)
2909 goto fail_qgroup;
2910 }
2911 }
2912
2913 ret = btrfs_find_orphan_roots(tree_root);
2914 if (ret)
2915 goto fail_qgroup;
2916
2917 if (!(sb->s_flags & MS_RDONLY)) {
2918 ret = btrfs_cleanup_fs_roots(fs_info);
2919 if (ret)
2920 goto fail_qgroup;
2921
2922 mutex_lock(&fs_info->cleaner_mutex);
2923 ret = btrfs_recover_relocation(tree_root);
2924 mutex_unlock(&fs_info->cleaner_mutex);
2925 if (ret < 0) {
2926 printk(KERN_WARNING
2927 "BTRFS: failed to recover relocation\n");
2928 err = -EINVAL;
2929 goto fail_qgroup;
2930 }
2931 }
2932
2933 location.objectid = BTRFS_FS_TREE_OBJECTID;
2934 location.type = BTRFS_ROOT_ITEM_KEY;
2935 location.offset = 0;
2936
2937 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2938 if (IS_ERR(fs_info->fs_root)) {
2939 err = PTR_ERR(fs_info->fs_root);
2940 goto fail_qgroup;
2941 }
2942
2943 if (sb->s_flags & MS_RDONLY)
2944 return 0;
2945
2946 down_read(&fs_info->cleanup_work_sem);
2947 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2948 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2949 up_read(&fs_info->cleanup_work_sem);
2950 close_ctree(tree_root);
2951 return ret;
2952 }
2953 up_read(&fs_info->cleanup_work_sem);
2954
2955 ret = btrfs_resume_balance_async(fs_info);
2956 if (ret) {
2957 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2958 close_ctree(tree_root);
2959 return ret;
2960 }
2961
2962 ret = btrfs_resume_dev_replace_async(fs_info);
2963 if (ret) {
2964 pr_warn("BTRFS: failed to resume dev_replace\n");
2965 close_ctree(tree_root);
2966 return ret;
2967 }
2968
2969 btrfs_qgroup_rescan_resume(fs_info);
2970
2971 if (create_uuid_tree) {
2972 pr_info("BTRFS: creating UUID tree\n");
2973 ret = btrfs_create_uuid_tree(fs_info);
2974 if (ret) {
2975 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2976 ret);
2977 close_ctree(tree_root);
2978 return ret;
2979 }
2980 } else if (check_uuid_tree ||
2981 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2982 pr_info("BTRFS: checking UUID tree\n");
2983 ret = btrfs_check_uuid_tree(fs_info);
2984 if (ret) {
2985 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2986 ret);
2987 close_ctree(tree_root);
2988 return ret;
2989 }
2990 } else {
2991 fs_info->update_uuid_tree_gen = 1;
2992 }
2993
2994 fs_info->open = 1;
2995
2996 return 0;
2997
2998fail_qgroup:
2999 btrfs_free_qgroup_config(fs_info);
3000fail_trans_kthread:
3001 kthread_stop(fs_info->transaction_kthread);
3002 btrfs_cleanup_transaction(fs_info->tree_root);
3003 btrfs_free_fs_roots(fs_info);
3004fail_cleaner:
3005 kthread_stop(fs_info->cleaner_kthread);
3006
3007 /*
3008 * make sure we're done with the btree inode before we stop our
3009 * kthreads
3010 */
3011 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3012
3013fail_sysfs:
3014 btrfs_sysfs_remove_one(fs_info);
3015
3016fail_block_groups:
3017 btrfs_put_block_group_cache(fs_info);
3018 btrfs_free_block_groups(fs_info);
3019
3020fail_tree_roots:
3021 free_root_pointers(fs_info, 1);
3022 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3023
3024fail_sb_buffer:
3025 btrfs_stop_all_workers(fs_info);
3026fail_alloc:
3027fail_iput:
3028 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3029
3030 iput(fs_info->btree_inode);
3031fail_bio_counter:
3032 percpu_counter_destroy(&fs_info->bio_counter);
3033fail_delalloc_bytes:
3034 percpu_counter_destroy(&fs_info->delalloc_bytes);
3035fail_dirty_metadata_bytes:
3036 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3037fail_bdi:
3038 bdi_destroy(&fs_info->bdi);
3039fail_srcu:
3040 cleanup_srcu_struct(&fs_info->subvol_srcu);
3041fail:
3042 btrfs_free_stripe_hash_table(fs_info);
3043 btrfs_close_devices(fs_info->fs_devices);
3044 return err;
3045
3046recovery_tree_root:
3047 if (!btrfs_test_opt(tree_root, RECOVERY))
3048 goto fail_tree_roots;
3049
3050 free_root_pointers(fs_info, 0);
3051
3052 /* don't use the log in recovery mode, it won't be valid */
3053 btrfs_set_super_log_root(disk_super, 0);
3054
3055 /* we can't trust the free space cache either */
3056 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3057
3058 ret = next_root_backup(fs_info, fs_info->super_copy,
3059 &num_backups_tried, &backup_index);
3060 if (ret == -1)
3061 goto fail_block_groups;
3062 goto retry_root_backup;
3063}
3064
3065static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3066{
3067 if (uptodate) {
3068 set_buffer_uptodate(bh);
3069 } else {
3070 struct btrfs_device *device = (struct btrfs_device *)
3071 bh->b_private;
3072
3073 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3074 "I/O error on %s\n",
3075 rcu_str_deref(device->name));
3076 /* note, we dont' set_buffer_write_io_error because we have
3077 * our own ways of dealing with the IO errors
3078 */
3079 clear_buffer_uptodate(bh);
3080 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3081 }
3082 unlock_buffer(bh);
3083 put_bh(bh);
3084}
3085
3086struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3087{
3088 struct buffer_head *bh;
3089 struct buffer_head *latest = NULL;
3090 struct btrfs_super_block *super;
3091 int i;
3092 u64 transid = 0;
3093 u64 bytenr;
3094
3095 /* we would like to check all the supers, but that would make
3096 * a btrfs mount succeed after a mkfs from a different FS.
3097 * So, we need to add a special mount option to scan for
3098 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3099 */
3100 for (i = 0; i < 1; i++) {
3101 bytenr = btrfs_sb_offset(i);
3102 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3103 i_size_read(bdev->bd_inode))
3104 break;
3105 bh = __bread(bdev, bytenr / 4096,
3106 BTRFS_SUPER_INFO_SIZE);
3107 if (!bh)
3108 continue;
3109
3110 super = (struct btrfs_super_block *)bh->b_data;
3111 if (btrfs_super_bytenr(super) != bytenr ||
3112 btrfs_super_magic(super) != BTRFS_MAGIC) {
3113 brelse(bh);
3114 continue;
3115 }
3116
3117 if (!latest || btrfs_super_generation(super) > transid) {
3118 brelse(latest);
3119 latest = bh;
3120 transid = btrfs_super_generation(super);
3121 } else {
3122 brelse(bh);
3123 }
3124 }
3125 return latest;
3126}
3127
3128/*
3129 * this should be called twice, once with wait == 0 and
3130 * once with wait == 1. When wait == 0 is done, all the buffer heads
3131 * we write are pinned.
3132 *
3133 * They are released when wait == 1 is done.
3134 * max_mirrors must be the same for both runs, and it indicates how
3135 * many supers on this one device should be written.
3136 *
3137 * max_mirrors == 0 means to write them all.
3138 */
3139static int write_dev_supers(struct btrfs_device *device,
3140 struct btrfs_super_block *sb,
3141 int do_barriers, int wait, int max_mirrors)
3142{
3143 struct buffer_head *bh;
3144 int i;
3145 int ret;
3146 int errors = 0;
3147 u32 crc;
3148 u64 bytenr;
3149
3150 if (max_mirrors == 0)
3151 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3152
3153 for (i = 0; i < max_mirrors; i++) {
3154 bytenr = btrfs_sb_offset(i);
3155 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3156 device->commit_total_bytes)
3157 break;
3158
3159 if (wait) {
3160 bh = __find_get_block(device->bdev, bytenr / 4096,
3161 BTRFS_SUPER_INFO_SIZE);
3162 if (!bh) {
3163 errors++;
3164 continue;
3165 }
3166 wait_on_buffer(bh);
3167 if (!buffer_uptodate(bh))
3168 errors++;
3169
3170 /* drop our reference */
3171 brelse(bh);
3172
3173 /* drop the reference from the wait == 0 run */
3174 brelse(bh);
3175 continue;
3176 } else {
3177 btrfs_set_super_bytenr(sb, bytenr);
3178
3179 crc = ~(u32)0;
3180 crc = btrfs_csum_data((char *)sb +
3181 BTRFS_CSUM_SIZE, crc,
3182 BTRFS_SUPER_INFO_SIZE -
3183 BTRFS_CSUM_SIZE);
3184 btrfs_csum_final(crc, sb->csum);
3185
3186 /*
3187 * one reference for us, and we leave it for the
3188 * caller
3189 */
3190 bh = __getblk(device->bdev, bytenr / 4096,
3191 BTRFS_SUPER_INFO_SIZE);
3192 if (!bh) {
3193 printk(KERN_ERR "BTRFS: couldn't get super "
3194 "buffer head for bytenr %Lu\n", bytenr);
3195 errors++;
3196 continue;
3197 }
3198
3199 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3200
3201 /* one reference for submit_bh */
3202 get_bh(bh);
3203
3204 set_buffer_uptodate(bh);
3205 lock_buffer(bh);
3206 bh->b_end_io = btrfs_end_buffer_write_sync;
3207 bh->b_private = device;
3208 }
3209
3210 /*
3211 * we fua the first super. The others we allow
3212 * to go down lazy.
3213 */
3214 if (i == 0)
3215 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3216 else
3217 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3218 if (ret)
3219 errors++;
3220 }
3221 return errors < i ? 0 : -1;
3222}
3223
3224/*
3225 * endio for the write_dev_flush, this will wake anyone waiting
3226 * for the barrier when it is done
3227 */
3228static void btrfs_end_empty_barrier(struct bio *bio, int err)
3229{
3230 if (err) {
3231 if (err == -EOPNOTSUPP)
3232 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3233 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3234 }
3235 if (bio->bi_private)
3236 complete(bio->bi_private);
3237 bio_put(bio);
3238}
3239
3240/*
3241 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3242 * sent down. With wait == 1, it waits for the previous flush.
3243 *
3244 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3245 * capable
3246 */
3247static int write_dev_flush(struct btrfs_device *device, int wait)
3248{
3249 struct bio *bio;
3250 int ret = 0;
3251
3252 if (device->nobarriers)
3253 return 0;
3254
3255 if (wait) {
3256 bio = device->flush_bio;
3257 if (!bio)
3258 return 0;
3259
3260 wait_for_completion(&device->flush_wait);
3261
3262 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3263 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3264 rcu_str_deref(device->name));
3265 device->nobarriers = 1;
3266 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3267 ret = -EIO;
3268 btrfs_dev_stat_inc_and_print(device,
3269 BTRFS_DEV_STAT_FLUSH_ERRS);
3270 }
3271
3272 /* drop the reference from the wait == 0 run */
3273 bio_put(bio);
3274 device->flush_bio = NULL;
3275
3276 return ret;
3277 }
3278
3279 /*
3280 * one reference for us, and we leave it for the
3281 * caller
3282 */
3283 device->flush_bio = NULL;
3284 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3285 if (!bio)
3286 return -ENOMEM;
3287
3288 bio->bi_end_io = btrfs_end_empty_barrier;
3289 bio->bi_bdev = device->bdev;
3290 init_completion(&device->flush_wait);
3291 bio->bi_private = &device->flush_wait;
3292 device->flush_bio = bio;
3293
3294 bio_get(bio);
3295 btrfsic_submit_bio(WRITE_FLUSH, bio);
3296
3297 return 0;
3298}
3299
3300/*
3301 * send an empty flush down to each device in parallel,
3302 * then wait for them
3303 */
3304static int barrier_all_devices(struct btrfs_fs_info *info)
3305{
3306 struct list_head *head;
3307 struct btrfs_device *dev;
3308 int errors_send = 0;
3309 int errors_wait = 0;
3310 int ret;
3311
3312 /* send down all the barriers */
3313 head = &info->fs_devices->devices;
3314 list_for_each_entry_rcu(dev, head, dev_list) {
3315 if (dev->missing)
3316 continue;
3317 if (!dev->bdev) {
3318 errors_send++;
3319 continue;
3320 }
3321 if (!dev->in_fs_metadata || !dev->writeable)
3322 continue;
3323
3324 ret = write_dev_flush(dev, 0);
3325 if (ret)
3326 errors_send++;
3327 }
3328
3329 /* wait for all the barriers */
3330 list_for_each_entry_rcu(dev, head, dev_list) {
3331 if (dev->missing)
3332 continue;
3333 if (!dev->bdev) {
3334 errors_wait++;
3335 continue;
3336 }
3337 if (!dev->in_fs_metadata || !dev->writeable)
3338 continue;
3339
3340 ret = write_dev_flush(dev, 1);
3341 if (ret)
3342 errors_wait++;
3343 }
3344 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3345 errors_wait > info->num_tolerated_disk_barrier_failures)
3346 return -EIO;
3347 return 0;
3348}
3349
3350int btrfs_calc_num_tolerated_disk_barrier_failures(
3351 struct btrfs_fs_info *fs_info)
3352{
3353 struct btrfs_ioctl_space_info space;
3354 struct btrfs_space_info *sinfo;
3355 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3356 BTRFS_BLOCK_GROUP_SYSTEM,
3357 BTRFS_BLOCK_GROUP_METADATA,
3358 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3359 int num_types = 4;
3360 int i;
3361 int c;
3362 int num_tolerated_disk_barrier_failures =
3363 (int)fs_info->fs_devices->num_devices;
3364
3365 for (i = 0; i < num_types; i++) {
3366 struct btrfs_space_info *tmp;
3367
3368 sinfo = NULL;
3369 rcu_read_lock();
3370 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3371 if (tmp->flags == types[i]) {
3372 sinfo = tmp;
3373 break;
3374 }
3375 }
3376 rcu_read_unlock();
3377
3378 if (!sinfo)
3379 continue;
3380
3381 down_read(&sinfo->groups_sem);
3382 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3383 if (!list_empty(&sinfo->block_groups[c])) {
3384 u64 flags;
3385
3386 btrfs_get_block_group_info(
3387 &sinfo->block_groups[c], &space);
3388 if (space.total_bytes == 0 ||
3389 space.used_bytes == 0)
3390 continue;
3391 flags = space.flags;
3392 /*
3393 * return
3394 * 0: if dup, single or RAID0 is configured for
3395 * any of metadata, system or data, else
3396 * 1: if RAID5 is configured, or if RAID1 or
3397 * RAID10 is configured and only two mirrors
3398 * are used, else
3399 * 2: if RAID6 is configured, else
3400 * num_mirrors - 1: if RAID1 or RAID10 is
3401 * configured and more than
3402 * 2 mirrors are used.
3403 */
3404 if (num_tolerated_disk_barrier_failures > 0 &&
3405 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3406 BTRFS_BLOCK_GROUP_RAID0)) ||
3407 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3408 == 0)))
3409 num_tolerated_disk_barrier_failures = 0;
3410 else if (num_tolerated_disk_barrier_failures > 1) {
3411 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3412 BTRFS_BLOCK_GROUP_RAID5 |
3413 BTRFS_BLOCK_GROUP_RAID10)) {
3414 num_tolerated_disk_barrier_failures = 1;
3415 } else if (flags &
3416 BTRFS_BLOCK_GROUP_RAID6) {
3417 num_tolerated_disk_barrier_failures = 2;
3418 }
3419 }
3420 }
3421 }
3422 up_read(&sinfo->groups_sem);
3423 }
3424
3425 return num_tolerated_disk_barrier_failures;
3426}
3427
3428static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3429{
3430 struct list_head *head;
3431 struct btrfs_device *dev;
3432 struct btrfs_super_block *sb;
3433 struct btrfs_dev_item *dev_item;
3434 int ret;
3435 int do_barriers;
3436 int max_errors;
3437 int total_errors = 0;
3438 u64 flags;
3439
3440 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3441 backup_super_roots(root->fs_info);
3442
3443 sb = root->fs_info->super_for_commit;
3444 dev_item = &sb->dev_item;
3445
3446 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3447 head = &root->fs_info->fs_devices->devices;
3448 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3449
3450 if (do_barriers) {
3451 ret = barrier_all_devices(root->fs_info);
3452 if (ret) {
3453 mutex_unlock(
3454 &root->fs_info->fs_devices->device_list_mutex);
3455 btrfs_error(root->fs_info, ret,
3456 "errors while submitting device barriers.");
3457 return ret;
3458 }
3459 }
3460
3461 list_for_each_entry_rcu(dev, head, dev_list) {
3462 if (!dev->bdev) {
3463 total_errors++;
3464 continue;
3465 }
3466 if (!dev->in_fs_metadata || !dev->writeable)
3467 continue;
3468
3469 btrfs_set_stack_device_generation(dev_item, 0);
3470 btrfs_set_stack_device_type(dev_item, dev->type);
3471 btrfs_set_stack_device_id(dev_item, dev->devid);
3472 btrfs_set_stack_device_total_bytes(dev_item,
3473 dev->commit_total_bytes);
3474 btrfs_set_stack_device_bytes_used(dev_item,
3475 dev->commit_bytes_used);
3476 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3477 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3478 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3479 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3480 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3481
3482 flags = btrfs_super_flags(sb);
3483 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3484
3485 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3486 if (ret)
3487 total_errors++;
3488 }
3489 if (total_errors > max_errors) {
3490 btrfs_err(root->fs_info, "%d errors while writing supers",
3491 total_errors);
3492 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3493
3494 /* FUA is masked off if unsupported and can't be the reason */
3495 btrfs_error(root->fs_info, -EIO,
3496 "%d errors while writing supers", total_errors);
3497 return -EIO;
3498 }
3499
3500 total_errors = 0;
3501 list_for_each_entry_rcu(dev, head, dev_list) {
3502 if (!dev->bdev)
3503 continue;
3504 if (!dev->in_fs_metadata || !dev->writeable)
3505 continue;
3506
3507 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3508 if (ret)
3509 total_errors++;
3510 }
3511 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3512 if (total_errors > max_errors) {
3513 btrfs_error(root->fs_info, -EIO,
3514 "%d errors while writing supers", total_errors);
3515 return -EIO;
3516 }
3517 return 0;
3518}
3519
3520int write_ctree_super(struct btrfs_trans_handle *trans,
3521 struct btrfs_root *root, int max_mirrors)
3522{
3523 return write_all_supers(root, max_mirrors);
3524}
3525
3526/* Drop a fs root from the radix tree and free it. */
3527void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3528 struct btrfs_root *root)
3529{
3530 spin_lock(&fs_info->fs_roots_radix_lock);
3531 radix_tree_delete(&fs_info->fs_roots_radix,
3532 (unsigned long)root->root_key.objectid);
3533 spin_unlock(&fs_info->fs_roots_radix_lock);
3534
3535 if (btrfs_root_refs(&root->root_item) == 0)
3536 synchronize_srcu(&fs_info->subvol_srcu);
3537
3538 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3539 btrfs_free_log(NULL, root);
3540
3541 if (root->free_ino_pinned)
3542 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3543 if (root->free_ino_ctl)
3544 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3545 free_fs_root(root);
3546}
3547
3548static void free_fs_root(struct btrfs_root *root)
3549{
3550 iput(root->ino_cache_inode);
3551 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3552 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3553 root->orphan_block_rsv = NULL;
3554 if (root->anon_dev)
3555 free_anon_bdev(root->anon_dev);
3556 if (root->subv_writers)
3557 btrfs_free_subvolume_writers(root->subv_writers);
3558 free_extent_buffer(root->node);
3559 free_extent_buffer(root->commit_root);
3560 kfree(root->free_ino_ctl);
3561 kfree(root->free_ino_pinned);
3562 kfree(root->name);
3563 btrfs_put_fs_root(root);
3564}
3565
3566void btrfs_free_fs_root(struct btrfs_root *root)
3567{
3568 free_fs_root(root);
3569}
3570
3571int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3572{
3573 u64 root_objectid = 0;
3574 struct btrfs_root *gang[8];
3575 int i = 0;
3576 int err = 0;
3577 unsigned int ret = 0;
3578 int index;
3579
3580 while (1) {
3581 index = srcu_read_lock(&fs_info->subvol_srcu);
3582 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3583 (void **)gang, root_objectid,
3584 ARRAY_SIZE(gang));
3585 if (!ret) {
3586 srcu_read_unlock(&fs_info->subvol_srcu, index);
3587 break;
3588 }
3589 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3590
3591 for (i = 0; i < ret; i++) {
3592 /* Avoid to grab roots in dead_roots */
3593 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3594 gang[i] = NULL;
3595 continue;
3596 }
3597 /* grab all the search result for later use */
3598 gang[i] = btrfs_grab_fs_root(gang[i]);
3599 }
3600 srcu_read_unlock(&fs_info->subvol_srcu, index);
3601
3602 for (i = 0; i < ret; i++) {
3603 if (!gang[i])
3604 continue;
3605 root_objectid = gang[i]->root_key.objectid;
3606 err = btrfs_orphan_cleanup(gang[i]);
3607 if (err)
3608 break;
3609 btrfs_put_fs_root(gang[i]);
3610 }
3611 root_objectid++;
3612 }
3613
3614 /* release the uncleaned roots due to error */
3615 for (; i < ret; i++) {
3616 if (gang[i])
3617 btrfs_put_fs_root(gang[i]);
3618 }
3619 return err;
3620}
3621
3622int btrfs_commit_super(struct btrfs_root *root)
3623{
3624 struct btrfs_trans_handle *trans;
3625
3626 mutex_lock(&root->fs_info->cleaner_mutex);
3627 btrfs_run_delayed_iputs(root);
3628 mutex_unlock(&root->fs_info->cleaner_mutex);
3629 wake_up_process(root->fs_info->cleaner_kthread);
3630
3631 /* wait until ongoing cleanup work done */
3632 down_write(&root->fs_info->cleanup_work_sem);
3633 up_write(&root->fs_info->cleanup_work_sem);
3634
3635 trans = btrfs_join_transaction(root);
3636 if (IS_ERR(trans))
3637 return PTR_ERR(trans);
3638 return btrfs_commit_transaction(trans, root);
3639}
3640
3641void close_ctree(struct btrfs_root *root)
3642{
3643 struct btrfs_fs_info *fs_info = root->fs_info;
3644 int ret;
3645
3646 fs_info->closing = 1;
3647 smp_mb();
3648
3649 /* wait for the uuid_scan task to finish */
3650 down(&fs_info->uuid_tree_rescan_sem);
3651 /* avoid complains from lockdep et al., set sem back to initial state */
3652 up(&fs_info->uuid_tree_rescan_sem);
3653
3654 /* pause restriper - we want to resume on mount */
3655 btrfs_pause_balance(fs_info);
3656
3657 btrfs_dev_replace_suspend_for_unmount(fs_info);
3658
3659 btrfs_scrub_cancel(fs_info);
3660
3661 /* wait for any defraggers to finish */
3662 wait_event(fs_info->transaction_wait,
3663 (atomic_read(&fs_info->defrag_running) == 0));
3664
3665 /* clear out the rbtree of defraggable inodes */
3666 btrfs_cleanup_defrag_inodes(fs_info);
3667
3668 cancel_work_sync(&fs_info->async_reclaim_work);
3669
3670 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3671 ret = btrfs_commit_super(root);
3672 if (ret)
3673 btrfs_err(fs_info, "commit super ret %d", ret);
3674 }
3675
3676 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3677 btrfs_error_commit_super(root);
3678
3679 kthread_stop(fs_info->transaction_kthread);
3680 kthread_stop(fs_info->cleaner_kthread);
3681
3682 fs_info->closing = 2;
3683 smp_mb();
3684
3685 btrfs_free_qgroup_config(fs_info);
3686
3687 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3688 btrfs_info(fs_info, "at unmount delalloc count %lld",
3689 percpu_counter_sum(&fs_info->delalloc_bytes));
3690 }
3691
3692 btrfs_sysfs_remove_one(fs_info);
3693
3694 btrfs_free_fs_roots(fs_info);
3695
3696 btrfs_put_block_group_cache(fs_info);
3697
3698 btrfs_free_block_groups(fs_info);
3699
3700 /*
3701 * we must make sure there is not any read request to
3702 * submit after we stopping all workers.
3703 */
3704 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3705 btrfs_stop_all_workers(fs_info);
3706
3707 fs_info->open = 0;
3708 free_root_pointers(fs_info, 1);
3709
3710 iput(fs_info->btree_inode);
3711
3712#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3713 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3714 btrfsic_unmount(root, fs_info->fs_devices);
3715#endif
3716
3717 btrfs_close_devices(fs_info->fs_devices);
3718 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3719
3720 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3721 percpu_counter_destroy(&fs_info->delalloc_bytes);
3722 percpu_counter_destroy(&fs_info->bio_counter);
3723 bdi_destroy(&fs_info->bdi);
3724 cleanup_srcu_struct(&fs_info->subvol_srcu);
3725
3726 btrfs_free_stripe_hash_table(fs_info);
3727
3728 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3729 root->orphan_block_rsv = NULL;
3730
3731 lock_chunks(root);
3732 while (!list_empty(&fs_info->pinned_chunks)) {
3733 struct extent_map *em;
3734
3735 em = list_first_entry(&fs_info->pinned_chunks,
3736 struct extent_map, list);
3737 list_del_init(&em->list);
3738 free_extent_map(em);
3739 }
3740 unlock_chunks(root);
3741}
3742
3743int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3744 int atomic)
3745{
3746 int ret;
3747 struct inode *btree_inode = buf->pages[0]->mapping->host;
3748
3749 ret = extent_buffer_uptodate(buf);
3750 if (!ret)
3751 return ret;
3752
3753 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3754 parent_transid, atomic);
3755 if (ret == -EAGAIN)
3756 return ret;
3757 return !ret;
3758}
3759
3760int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3761{
3762 return set_extent_buffer_uptodate(buf);
3763}
3764
3765void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3766{
3767 struct btrfs_root *root;
3768 u64 transid = btrfs_header_generation(buf);
3769 int was_dirty;
3770
3771#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3772 /*
3773 * This is a fast path so only do this check if we have sanity tests
3774 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3775 * outside of the sanity tests.
3776 */
3777 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3778 return;
3779#endif
3780 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3781 btrfs_assert_tree_locked(buf);
3782 if (transid != root->fs_info->generation)
3783 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3784 "found %llu running %llu\n",
3785 buf->start, transid, root->fs_info->generation);
3786 was_dirty = set_extent_buffer_dirty(buf);
3787 if (!was_dirty)
3788 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3789 buf->len,
3790 root->fs_info->dirty_metadata_batch);
3791#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3792 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3793 btrfs_print_leaf(root, buf);
3794 ASSERT(0);
3795 }
3796#endif
3797}
3798
3799static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3800 int flush_delayed)
3801{
3802 /*
3803 * looks as though older kernels can get into trouble with
3804 * this code, they end up stuck in balance_dirty_pages forever
3805 */
3806 int ret;
3807
3808 if (current->flags & PF_MEMALLOC)
3809 return;
3810
3811 if (flush_delayed)
3812 btrfs_balance_delayed_items(root);
3813
3814 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3815 BTRFS_DIRTY_METADATA_THRESH);
3816 if (ret > 0) {
3817 balance_dirty_pages_ratelimited(
3818 root->fs_info->btree_inode->i_mapping);
3819 }
3820 return;
3821}
3822
3823void btrfs_btree_balance_dirty(struct btrfs_root *root)
3824{
3825 __btrfs_btree_balance_dirty(root, 1);
3826}
3827
3828void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3829{
3830 __btrfs_btree_balance_dirty(root, 0);
3831}
3832
3833int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3834{
3835 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3836 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3837}
3838
3839static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3840 int read_only)
3841{
3842 struct btrfs_super_block *sb = fs_info->super_copy;
3843 int ret = 0;
3844
3845 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3846 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3847 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3848 ret = -EINVAL;
3849 }
3850 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3851 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3852 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3853 ret = -EINVAL;
3854 }
3855 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3856 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3857 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3858 ret = -EINVAL;
3859 }
3860
3861 /*
3862 * The common minimum, we don't know if we can trust the nodesize/sectorsize
3863 * items yet, they'll be verified later. Issue just a warning.
3864 */
3865 if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
3866 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
3867 btrfs_super_root(sb));
3868 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
3869 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
3870 btrfs_super_chunk_root(sb));
3871 if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
3872 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
3873 btrfs_super_log_root(sb));
3874
3875 /*
3876 * Check the lower bound, the alignment and other constraints are
3877 * checked later.
3878 */
3879 if (btrfs_super_nodesize(sb) < 4096) {
3880 printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n",
3881 btrfs_super_nodesize(sb));
3882 ret = -EINVAL;
3883 }
3884 if (btrfs_super_sectorsize(sb) < 4096) {
3885 printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n",
3886 btrfs_super_sectorsize(sb));
3887 ret = -EINVAL;
3888 }
3889
3890 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
3891 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
3892 fs_info->fsid, sb->dev_item.fsid);
3893 ret = -EINVAL;
3894 }
3895
3896 /*
3897 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3898 * done later
3899 */
3900 if (btrfs_super_num_devices(sb) > (1UL << 31))
3901 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
3902 btrfs_super_num_devices(sb));
3903 if (btrfs_super_num_devices(sb) == 0) {
3904 printk(KERN_ERR "BTRFS: number of devices is 0\n");
3905 ret = -EINVAL;
3906 }
3907
3908 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
3909 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
3910 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
3911 ret = -EINVAL;
3912 }
3913
3914 /*
3915 * Obvious sys_chunk_array corruptions, it must hold at least one key
3916 * and one chunk
3917 */
3918 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
3919 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
3920 btrfs_super_sys_array_size(sb),
3921 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
3922 ret = -EINVAL;
3923 }
3924 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
3925 + sizeof(struct btrfs_chunk)) {
3926 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
3927 btrfs_super_sys_array_size(sb),
3928 sizeof(struct btrfs_disk_key)
3929 + sizeof(struct btrfs_chunk));
3930 ret = -EINVAL;
3931 }
3932
3933 /*
3934 * The generation is a global counter, we'll trust it more than the others
3935 * but it's still possible that it's the one that's wrong.
3936 */
3937 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
3938 printk(KERN_WARNING
3939 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
3940 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
3941 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
3942 && btrfs_super_cache_generation(sb) != (u64)-1)
3943 printk(KERN_WARNING
3944 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
3945 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
3946
3947 return ret;
3948}
3949
3950static void btrfs_error_commit_super(struct btrfs_root *root)
3951{
3952 mutex_lock(&root->fs_info->cleaner_mutex);
3953 btrfs_run_delayed_iputs(root);
3954 mutex_unlock(&root->fs_info->cleaner_mutex);
3955
3956 down_write(&root->fs_info->cleanup_work_sem);
3957 up_write(&root->fs_info->cleanup_work_sem);
3958
3959 /* cleanup FS via transaction */
3960 btrfs_cleanup_transaction(root);
3961}
3962
3963static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3964{
3965 struct btrfs_ordered_extent *ordered;
3966
3967 spin_lock(&root->ordered_extent_lock);
3968 /*
3969 * This will just short circuit the ordered completion stuff which will
3970 * make sure the ordered extent gets properly cleaned up.
3971 */
3972 list_for_each_entry(ordered, &root->ordered_extents,
3973 root_extent_list)
3974 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3975 spin_unlock(&root->ordered_extent_lock);
3976}
3977
3978static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3979{
3980 struct btrfs_root *root;
3981 struct list_head splice;
3982
3983 INIT_LIST_HEAD(&splice);
3984
3985 spin_lock(&fs_info->ordered_root_lock);
3986 list_splice_init(&fs_info->ordered_roots, &splice);
3987 while (!list_empty(&splice)) {
3988 root = list_first_entry(&splice, struct btrfs_root,
3989 ordered_root);
3990 list_move_tail(&root->ordered_root,
3991 &fs_info->ordered_roots);
3992
3993 spin_unlock(&fs_info->ordered_root_lock);
3994 btrfs_destroy_ordered_extents(root);
3995
3996 cond_resched();
3997 spin_lock(&fs_info->ordered_root_lock);
3998 }
3999 spin_unlock(&fs_info->ordered_root_lock);
4000}
4001
4002static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4003 struct btrfs_root *root)
4004{
4005 struct rb_node *node;
4006 struct btrfs_delayed_ref_root *delayed_refs;
4007 struct btrfs_delayed_ref_node *ref;
4008 int ret = 0;
4009
4010 delayed_refs = &trans->delayed_refs;
4011
4012 spin_lock(&delayed_refs->lock);
4013 if (atomic_read(&delayed_refs->num_entries) == 0) {
4014 spin_unlock(&delayed_refs->lock);
4015 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4016 return ret;
4017 }
4018
4019 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4020 struct btrfs_delayed_ref_head *head;
4021 bool pin_bytes = false;
4022
4023 head = rb_entry(node, struct btrfs_delayed_ref_head,
4024 href_node);
4025 if (!mutex_trylock(&head->mutex)) {
4026 atomic_inc(&head->node.refs);
4027 spin_unlock(&delayed_refs->lock);
4028
4029 mutex_lock(&head->mutex);
4030 mutex_unlock(&head->mutex);
4031 btrfs_put_delayed_ref(&head->node);
4032 spin_lock(&delayed_refs->lock);
4033 continue;
4034 }
4035 spin_lock(&head->lock);
4036 while ((node = rb_first(&head->ref_root)) != NULL) {
4037 ref = rb_entry(node, struct btrfs_delayed_ref_node,
4038 rb_node);
4039 ref->in_tree = 0;
4040 rb_erase(&ref->rb_node, &head->ref_root);
4041 atomic_dec(&delayed_refs->num_entries);
4042 btrfs_put_delayed_ref(ref);
4043 }
4044 if (head->must_insert_reserved)
4045 pin_bytes = true;
4046 btrfs_free_delayed_extent_op(head->extent_op);
4047 delayed_refs->num_heads--;
4048 if (head->processing == 0)
4049 delayed_refs->num_heads_ready--;
4050 atomic_dec(&delayed_refs->num_entries);
4051 head->node.in_tree = 0;
4052 rb_erase(&head->href_node, &delayed_refs->href_root);
4053 spin_unlock(&head->lock);
4054 spin_unlock(&delayed_refs->lock);
4055 mutex_unlock(&head->mutex);
4056
4057 if (pin_bytes)
4058 btrfs_pin_extent(root, head->node.bytenr,
4059 head->node.num_bytes, 1);
4060 btrfs_put_delayed_ref(&head->node);
4061 cond_resched();
4062 spin_lock(&delayed_refs->lock);
4063 }
4064
4065 spin_unlock(&delayed_refs->lock);
4066
4067 return ret;
4068}
4069
4070static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4071{
4072 struct btrfs_inode *btrfs_inode;
4073 struct list_head splice;
4074
4075 INIT_LIST_HEAD(&splice);
4076
4077 spin_lock(&root->delalloc_lock);
4078 list_splice_init(&root->delalloc_inodes, &splice);
4079
4080 while (!list_empty(&splice)) {
4081 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4082 delalloc_inodes);
4083
4084 list_del_init(&btrfs_inode->delalloc_inodes);
4085 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4086 &btrfs_inode->runtime_flags);
4087 spin_unlock(&root->delalloc_lock);
4088
4089 btrfs_invalidate_inodes(btrfs_inode->root);
4090
4091 spin_lock(&root->delalloc_lock);
4092 }
4093
4094 spin_unlock(&root->delalloc_lock);
4095}
4096
4097static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4098{
4099 struct btrfs_root *root;
4100 struct list_head splice;
4101
4102 INIT_LIST_HEAD(&splice);
4103
4104 spin_lock(&fs_info->delalloc_root_lock);
4105 list_splice_init(&fs_info->delalloc_roots, &splice);
4106 while (!list_empty(&splice)) {
4107 root = list_first_entry(&splice, struct btrfs_root,
4108 delalloc_root);
4109 list_del_init(&root->delalloc_root);
4110 root = btrfs_grab_fs_root(root);
4111 BUG_ON(!root);
4112 spin_unlock(&fs_info->delalloc_root_lock);
4113
4114 btrfs_destroy_delalloc_inodes(root);
4115 btrfs_put_fs_root(root);
4116
4117 spin_lock(&fs_info->delalloc_root_lock);
4118 }
4119 spin_unlock(&fs_info->delalloc_root_lock);
4120}
4121
4122static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4123 struct extent_io_tree *dirty_pages,
4124 int mark)
4125{
4126 int ret;
4127 struct extent_buffer *eb;
4128 u64 start = 0;
4129 u64 end;
4130
4131 while (1) {
4132 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4133 mark, NULL);
4134 if (ret)
4135 break;
4136
4137 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4138 while (start <= end) {
4139 eb = btrfs_find_tree_block(root->fs_info, start);
4140 start += root->nodesize;
4141 if (!eb)
4142 continue;
4143 wait_on_extent_buffer_writeback(eb);
4144
4145 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4146 &eb->bflags))
4147 clear_extent_buffer_dirty(eb);
4148 free_extent_buffer_stale(eb);
4149 }
4150 }
4151
4152 return ret;
4153}
4154
4155static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4156 struct extent_io_tree *pinned_extents)
4157{
4158 struct extent_io_tree *unpin;
4159 u64 start;
4160 u64 end;
4161 int ret;
4162 bool loop = true;
4163
4164 unpin = pinned_extents;
4165again:
4166 while (1) {
4167 ret = find_first_extent_bit(unpin, 0, &start, &end,
4168 EXTENT_DIRTY, NULL);
4169 if (ret)
4170 break;
4171
4172 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4173 btrfs_error_unpin_extent_range(root, start, end);
4174 cond_resched();
4175 }
4176
4177 if (loop) {
4178 if (unpin == &root->fs_info->freed_extents[0])
4179 unpin = &root->fs_info->freed_extents[1];
4180 else
4181 unpin = &root->fs_info->freed_extents[0];
4182 loop = false;
4183 goto again;
4184 }
4185
4186 return 0;
4187}
4188
4189static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans,
4190 struct btrfs_fs_info *fs_info)
4191{
4192 struct btrfs_ordered_extent *ordered;
4193
4194 spin_lock(&fs_info->trans_lock);
4195 while (!list_empty(&cur_trans->pending_ordered)) {
4196 ordered = list_first_entry(&cur_trans->pending_ordered,
4197 struct btrfs_ordered_extent,
4198 trans_list);
4199 list_del_init(&ordered->trans_list);
4200 spin_unlock(&fs_info->trans_lock);
4201
4202 btrfs_put_ordered_extent(ordered);
4203 spin_lock(&fs_info->trans_lock);
4204 }
4205 spin_unlock(&fs_info->trans_lock);
4206}
4207
4208void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4209 struct btrfs_root *root)
4210{
4211 btrfs_destroy_delayed_refs(cur_trans, root);
4212
4213 cur_trans->state = TRANS_STATE_COMMIT_START;
4214 wake_up(&root->fs_info->transaction_blocked_wait);
4215
4216 cur_trans->state = TRANS_STATE_UNBLOCKED;
4217 wake_up(&root->fs_info->transaction_wait);
4218
4219 btrfs_free_pending_ordered(cur_trans, root->fs_info);
4220 btrfs_destroy_delayed_inodes(root);
4221 btrfs_assert_delayed_root_empty(root);
4222
4223 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4224 EXTENT_DIRTY);
4225 btrfs_destroy_pinned_extent(root,
4226 root->fs_info->pinned_extents);
4227
4228 cur_trans->state =TRANS_STATE_COMPLETED;
4229 wake_up(&cur_trans->commit_wait);
4230
4231 /*
4232 memset(cur_trans, 0, sizeof(*cur_trans));
4233 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4234 */
4235}
4236
4237static int btrfs_cleanup_transaction(struct btrfs_root *root)
4238{
4239 struct btrfs_transaction *t;
4240
4241 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4242
4243 spin_lock(&root->fs_info->trans_lock);
4244 while (!list_empty(&root->fs_info->trans_list)) {
4245 t = list_first_entry(&root->fs_info->trans_list,
4246 struct btrfs_transaction, list);
4247 if (t->state >= TRANS_STATE_COMMIT_START) {
4248 atomic_inc(&t->use_count);
4249 spin_unlock(&root->fs_info->trans_lock);
4250 btrfs_wait_for_commit(root, t->transid);
4251 btrfs_put_transaction(t);
4252 spin_lock(&root->fs_info->trans_lock);
4253 continue;
4254 }
4255 if (t == root->fs_info->running_transaction) {
4256 t->state = TRANS_STATE_COMMIT_DOING;
4257 spin_unlock(&root->fs_info->trans_lock);
4258 /*
4259 * We wait for 0 num_writers since we don't hold a trans
4260 * handle open currently for this transaction.
4261 */
4262 wait_event(t->writer_wait,
4263 atomic_read(&t->num_writers) == 0);
4264 } else {
4265 spin_unlock(&root->fs_info->trans_lock);
4266 }
4267 btrfs_cleanup_one_transaction(t, root);
4268
4269 spin_lock(&root->fs_info->trans_lock);
4270 if (t == root->fs_info->running_transaction)
4271 root->fs_info->running_transaction = NULL;
4272 list_del_init(&t->list);
4273 spin_unlock(&root->fs_info->trans_lock);
4274
4275 btrfs_put_transaction(t);
4276 trace_btrfs_transaction_commit(root);
4277 spin_lock(&root->fs_info->trans_lock);
4278 }
4279 spin_unlock(&root->fs_info->trans_lock);
4280 btrfs_destroy_all_ordered_extents(root->fs_info);
4281 btrfs_destroy_delayed_inodes(root);
4282 btrfs_assert_delayed_root_empty(root);
4283 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4284 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4285 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4286
4287 return 0;
4288}
4289
4290static const struct extent_io_ops btree_extent_io_ops = {
4291 .readpage_end_io_hook = btree_readpage_end_io_hook,
4292 .readpage_io_failed_hook = btree_io_failed_hook,
4293 .submit_bio_hook = btree_submit_bio_hook,
4294 /* note we're sharing with inode.c for the merge bio hook */
4295 .merge_bio_hook = btrfs_merge_bio_hook,
4296};
Linux 6.x block layer and io_uring tree(s)