projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
btrfs: add error handling for scrub_workers_get()
[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, 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
424 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
425 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
426 csum_type);
427 ret = 1;
428 }
429
430 return ret;
431}
432
433/*
434 * helper to read a given tree block, doing retries as required when
435 * the checksums don't match and we have alternate mirrors to try.
436 */
437static int btree_read_extent_buffer_pages(struct btrfs_root *root,
438 struct extent_buffer *eb,
439 u64 start, u64 parent_transid)
440{
441 struct extent_io_tree *io_tree;
442 int failed = 0;
443 int ret;
444 int num_copies = 0;
445 int mirror_num = 0;
446 int failed_mirror = 0;
447
448 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
449 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
450 while (1) {
451 ret = read_extent_buffer_pages(io_tree, eb, start,
452 WAIT_COMPLETE,
453 btree_get_extent, mirror_num);
454 if (!ret) {
455 if (!verify_parent_transid(io_tree, eb,
456 parent_transid, 0))
457 break;
458 else
459 ret = -EIO;
460 }
461
462 /*
463 * This buffer's crc is fine, but its contents are corrupted, so
464 * there is no reason to read the other copies, they won't be
465 * any less wrong.
466 */
467 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
468 break;
469
470 num_copies = btrfs_num_copies(root->fs_info,
471 eb->start, eb->len);
472 if (num_copies == 1)
473 break;
474
475 if (!failed_mirror) {
476 failed = 1;
477 failed_mirror = eb->read_mirror;
478 }
479
480 mirror_num++;
481 if (mirror_num == failed_mirror)
482 mirror_num++;
483
484 if (mirror_num > num_copies)
485 break;
486 }
487
488 if (failed && !ret && failed_mirror)
489 repair_eb_io_failure(root, eb, failed_mirror);
490
491 return ret;
492}
493
494/*
495 * checksum a dirty tree block before IO. This has extra checks to make sure
496 * we only fill in the checksum field in the first page of a multi-page block
497 */
498
499static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
500{
501 u64 start = page_offset(page);
502 u64 found_start;
503 struct extent_buffer *eb;
504
505 eb = (struct extent_buffer *)page->private;
506 if (page != eb->pages[0])
507 return 0;
508 found_start = btrfs_header_bytenr(eb);
509 if (WARN_ON(found_start != start || !PageUptodate(page)))
510 return 0;
511 csum_tree_block(fs_info, eb, 0);
512 return 0;
513}
514
515static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
516 struct extent_buffer *eb)
517{
518 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
519 u8 fsid[BTRFS_UUID_SIZE];
520 int ret = 1;
521
522 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
523 while (fs_devices) {
524 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
525 ret = 0;
526 break;
527 }
528 fs_devices = fs_devices->seed;
529 }
530 return ret;
531}
532
533#define CORRUPT(reason, eb, root, slot) \
534 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
535 "root=%llu, slot=%d", reason, \
536 btrfs_header_bytenr(eb), root->objectid, slot)
537
538static noinline int check_leaf(struct btrfs_root *root,
539 struct extent_buffer *leaf)
540{
541 struct btrfs_key key;
542 struct btrfs_key leaf_key;
543 u32 nritems = btrfs_header_nritems(leaf);
544 int slot;
545
546 if (nritems == 0)
547 return 0;
548
549 /* Check the 0 item */
550 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
551 BTRFS_LEAF_DATA_SIZE(root)) {
552 CORRUPT("invalid item offset size pair", leaf, root, 0);
553 return -EIO;
554 }
555
556 /*
557 * Check to make sure each items keys are in the correct order and their
558 * offsets make sense. We only have to loop through nritems-1 because
559 * we check the current slot against the next slot, which verifies the
560 * next slot's offset+size makes sense and that the current's slot
561 * offset is correct.
562 */
563 for (slot = 0; slot < nritems - 1; slot++) {
564 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
565 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
566
567 /* Make sure the keys are in the right order */
568 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
569 CORRUPT("bad key order", leaf, root, slot);
570 return -EIO;
571 }
572
573 /*
574 * Make sure the offset and ends are right, remember that the
575 * item data starts at the end of the leaf and grows towards the
576 * front.
577 */
578 if (btrfs_item_offset_nr(leaf, slot) !=
579 btrfs_item_end_nr(leaf, slot + 1)) {
580 CORRUPT("slot offset bad", leaf, root, slot);
581 return -EIO;
582 }
583
584 /*
585 * Check to make sure that we don't point outside of the leaf,
586 * just incase all the items are consistent to eachother, but
587 * all point outside of the leaf.
588 */
589 if (btrfs_item_end_nr(leaf, slot) >
590 BTRFS_LEAF_DATA_SIZE(root)) {
591 CORRUPT("slot end outside of leaf", leaf, root, slot);
592 return -EIO;
593 }
594 }
595
596 return 0;
597}
598
599static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
600 u64 phy_offset, struct page *page,
601 u64 start, u64 end, int mirror)
602{
603 u64 found_start;
604 int found_level;
605 struct extent_buffer *eb;
606 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
607 int ret = 0;
608 int reads_done;
609
610 if (!page->private)
611 goto out;
612
613 eb = (struct extent_buffer *)page->private;
614
615 /* the pending IO might have been the only thing that kept this buffer
616 * in memory. Make sure we have a ref for all this other checks
617 */
618 extent_buffer_get(eb);
619
620 reads_done = atomic_dec_and_test(&eb->io_pages);
621 if (!reads_done)
622 goto err;
623
624 eb->read_mirror = mirror;
625 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
626 ret = -EIO;
627 goto err;
628 }
629
630 found_start = btrfs_header_bytenr(eb);
631 if (found_start != eb->start) {
632 printk_ratelimited(KERN_ERR "BTRFS (device %s): bad tree block start "
633 "%llu %llu\n",
634 eb->fs_info->sb->s_id, found_start, eb->start);
635 ret = -EIO;
636 goto err;
637 }
638 if (check_tree_block_fsid(root->fs_info, eb)) {
639 printk_ratelimited(KERN_ERR "BTRFS (device %s): bad fsid on block %llu\n",
640 eb->fs_info->sb->s_id, eb->start);
641 ret = -EIO;
642 goto err;
643 }
644 found_level = btrfs_header_level(eb);
645 if (found_level >= BTRFS_MAX_LEVEL) {
646 btrfs_err(root->fs_info, "bad tree block level %d",
647 (int)btrfs_header_level(eb));
648 ret = -EIO;
649 goto err;
650 }
651
652 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
653 eb, found_level);
654
655 ret = csum_tree_block(root->fs_info, eb, 1);
656 if (ret) {
657 ret = -EIO;
658 goto err;
659 }
660
661 /*
662 * If this is a leaf block and it is corrupt, set the corrupt bit so
663 * that we don't try and read the other copies of this block, just
664 * return -EIO.
665 */
666 if (found_level == 0 && check_leaf(root, eb)) {
667 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
668 ret = -EIO;
669 }
670
671 if (!ret)
672 set_extent_buffer_uptodate(eb);
673err:
674 if (reads_done &&
675 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
676 btree_readahead_hook(root, eb, eb->start, ret);
677
678 if (ret) {
679 /*
680 * our io error hook is going to dec the io pages
681 * again, we have to make sure it has something
682 * to decrement
683 */
684 atomic_inc(&eb->io_pages);
685 clear_extent_buffer_uptodate(eb);
686 }
687 free_extent_buffer(eb);
688out:
689 return ret;
690}
691
692static int btree_io_failed_hook(struct page *page, int failed_mirror)
693{
694 struct extent_buffer *eb;
695 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
696
697 eb = (struct extent_buffer *)page->private;
698 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
699 eb->read_mirror = failed_mirror;
700 atomic_dec(&eb->io_pages);
701 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
702 btree_readahead_hook(root, eb, eb->start, -EIO);
703 return -EIO; /* we fixed nothing */
704}
705
706static void end_workqueue_bio(struct bio *bio, int err)
707{
708 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
709 struct btrfs_fs_info *fs_info;
710 struct btrfs_workqueue *wq;
711 btrfs_work_func_t func;
712
713 fs_info = end_io_wq->info;
714 end_io_wq->error = err;
715
716 if (bio->bi_rw & REQ_WRITE) {
717 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
718 wq = fs_info->endio_meta_write_workers;
719 func = btrfs_endio_meta_write_helper;
720 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
721 wq = fs_info->endio_freespace_worker;
722 func = btrfs_freespace_write_helper;
723 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
724 wq = fs_info->endio_raid56_workers;
725 func = btrfs_endio_raid56_helper;
726 } else {
727 wq = fs_info->endio_write_workers;
728 func = btrfs_endio_write_helper;
729 }
730 } else {
731 if (unlikely(end_io_wq->metadata ==
732 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
733 wq = fs_info->endio_repair_workers;
734 func = btrfs_endio_repair_helper;
735 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
736 wq = fs_info->endio_raid56_workers;
737 func = btrfs_endio_raid56_helper;
738 } else if (end_io_wq->metadata) {
739 wq = fs_info->endio_meta_workers;
740 func = btrfs_endio_meta_helper;
741 } else {
742 wq = fs_info->endio_workers;
743 func = btrfs_endio_helper;
744 }
745 }
746
747 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
748 btrfs_queue_work(wq, &end_io_wq->work);
749}
750
751int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
752 enum btrfs_wq_endio_type metadata)
753{
754 struct btrfs_end_io_wq *end_io_wq;
755
756 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
757 if (!end_io_wq)
758 return -ENOMEM;
759
760 end_io_wq->private = bio->bi_private;
761 end_io_wq->end_io = bio->bi_end_io;
762 end_io_wq->info = info;
763 end_io_wq->error = 0;
764 end_io_wq->bio = bio;
765 end_io_wq->metadata = metadata;
766
767 bio->bi_private = end_io_wq;
768 bio->bi_end_io = end_workqueue_bio;
769 return 0;
770}
771
772unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
773{
774 unsigned long limit = min_t(unsigned long,
775 info->thread_pool_size,
776 info->fs_devices->open_devices);
777 return 256 * limit;
778}
779
780static void run_one_async_start(struct btrfs_work *work)
781{
782 struct async_submit_bio *async;
783 int ret;
784
785 async = container_of(work, struct async_submit_bio, work);
786 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
787 async->mirror_num, async->bio_flags,
788 async->bio_offset);
789 if (ret)
790 async->error = ret;
791}
792
793static void run_one_async_done(struct btrfs_work *work)
794{
795 struct btrfs_fs_info *fs_info;
796 struct async_submit_bio *async;
797 int limit;
798
799 async = container_of(work, struct async_submit_bio, work);
800 fs_info = BTRFS_I(async->inode)->root->fs_info;
801
802 limit = btrfs_async_submit_limit(fs_info);
803 limit = limit * 2 / 3;
804
805 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
806 waitqueue_active(&fs_info->async_submit_wait))
807 wake_up(&fs_info->async_submit_wait);
808
809 /* If an error occured we just want to clean up the bio and move on */
810 if (async->error) {
811 bio_endio(async->bio, async->error);
812 return;
813 }
814
815 async->submit_bio_done(async->inode, async->rw, async->bio,
816 async->mirror_num, async->bio_flags,
817 async->bio_offset);
818}
819
820static void run_one_async_free(struct btrfs_work *work)
821{
822 struct async_submit_bio *async;
823
824 async = container_of(work, struct async_submit_bio, work);
825 kfree(async);
826}
827
828int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
829 int rw, struct bio *bio, int mirror_num,
830 unsigned long bio_flags,
831 u64 bio_offset,
832 extent_submit_bio_hook_t *submit_bio_start,
833 extent_submit_bio_hook_t *submit_bio_done)
834{
835 struct async_submit_bio *async;
836
837 async = kmalloc(sizeof(*async), GFP_NOFS);
838 if (!async)
839 return -ENOMEM;
840
841 async->inode = inode;
842 async->rw = rw;
843 async->bio = bio;
844 async->mirror_num = mirror_num;
845 async->submit_bio_start = submit_bio_start;
846 async->submit_bio_done = submit_bio_done;
847
848 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
849 run_one_async_done, run_one_async_free);
850
851 async->bio_flags = bio_flags;
852 async->bio_offset = bio_offset;
853
854 async->error = 0;
855
856 atomic_inc(&fs_info->nr_async_submits);
857
858 if (rw & REQ_SYNC)
859 btrfs_set_work_high_priority(&async->work);
860
861 btrfs_queue_work(fs_info->workers, &async->work);
862
863 while (atomic_read(&fs_info->async_submit_draining) &&
864 atomic_read(&fs_info->nr_async_submits)) {
865 wait_event(fs_info->async_submit_wait,
866 (atomic_read(&fs_info->nr_async_submits) == 0));
867 }
868
869 return 0;
870}
871
872static int btree_csum_one_bio(struct bio *bio)
873{
874 struct bio_vec *bvec;
875 struct btrfs_root *root;
876 int i, ret = 0;
877
878 bio_for_each_segment_all(bvec, bio, i) {
879 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
880 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
881 if (ret)
882 break;
883 }
884
885 return ret;
886}
887
888static int __btree_submit_bio_start(struct inode *inode, int rw,
889 struct bio *bio, int mirror_num,
890 unsigned long bio_flags,
891 u64 bio_offset)
892{
893 /*
894 * when we're called for a write, we're already in the async
895 * submission context. Just jump into btrfs_map_bio
896 */
897 return btree_csum_one_bio(bio);
898}
899
900static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
901 int mirror_num, unsigned long bio_flags,
902 u64 bio_offset)
903{
904 int ret;
905
906 /*
907 * when we're called for a write, we're already in the async
908 * submission context. Just jump into btrfs_map_bio
909 */
910 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
911 if (ret)
912 bio_endio(bio, ret);
913 return ret;
914}
915
916static int check_async_write(struct inode *inode, unsigned long bio_flags)
917{
918 if (bio_flags & EXTENT_BIO_TREE_LOG)
919 return 0;
920#ifdef CONFIG_X86
921 if (cpu_has_xmm4_2)
922 return 0;
923#endif
924 return 1;
925}
926
927static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
928 int mirror_num, unsigned long bio_flags,
929 u64 bio_offset)
930{
931 int async = check_async_write(inode, bio_flags);
932 int ret;
933
934 if (!(rw & REQ_WRITE)) {
935 /*
936 * called for a read, do the setup so that checksum validation
937 * can happen in the async kernel threads
938 */
939 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
940 bio, BTRFS_WQ_ENDIO_METADATA);
941 if (ret)
942 goto out_w_error;
943 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
944 mirror_num, 0);
945 } else if (!async) {
946 ret = btree_csum_one_bio(bio);
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 {
952 /*
953 * kthread helpers are used to submit writes so that
954 * checksumming can happen in parallel across all CPUs
955 */
956 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
957 inode, rw, bio, mirror_num, 0,
958 bio_offset,
959 __btree_submit_bio_start,
960 __btree_submit_bio_done);
961 }
962
963 if (ret) {
964out_w_error:
965 bio_endio(bio, ret);
966 }
967 return ret;
968}
969
970#ifdef CONFIG_MIGRATION
971static int btree_migratepage(struct address_space *mapping,
972 struct page *newpage, struct page *page,
973 enum migrate_mode mode)
974{
975 /*
976 * we can't safely write a btree page from here,
977 * we haven't done the locking hook
978 */
979 if (PageDirty(page))
980 return -EAGAIN;
981 /*
982 * Buffers may be managed in a filesystem specific way.
983 * We must have no buffers or drop them.
984 */
985 if (page_has_private(page) &&
986 !try_to_release_page(page, GFP_KERNEL))
987 return -EAGAIN;
988 return migrate_page(mapping, newpage, page, mode);
989}
990#endif
991
992
993static int btree_writepages(struct address_space *mapping,
994 struct writeback_control *wbc)
995{
996 struct btrfs_fs_info *fs_info;
997 int ret;
998
999 if (wbc->sync_mode == WB_SYNC_NONE) {
1000
1001 if (wbc->for_kupdate)
1002 return 0;
1003
1004 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1005 /* this is a bit racy, but that's ok */
1006 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1007 BTRFS_DIRTY_METADATA_THRESH);
1008 if (ret < 0)
1009 return 0;
1010 }
1011 return btree_write_cache_pages(mapping, wbc);
1012}
1013
1014static int btree_readpage(struct file *file, struct page *page)
1015{
1016 struct extent_io_tree *tree;
1017 tree = &BTRFS_I(page->mapping->host)->io_tree;
1018 return extent_read_full_page(tree, page, btree_get_extent, 0);
1019}
1020
1021static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1022{
1023 if (PageWriteback(page) || PageDirty(page))
1024 return 0;
1025
1026 return try_release_extent_buffer(page);
1027}
1028
1029static void btree_invalidatepage(struct page *page, unsigned int offset,
1030 unsigned int length)
1031{
1032 struct extent_io_tree *tree;
1033 tree = &BTRFS_I(page->mapping->host)->io_tree;
1034 extent_invalidatepage(tree, page, offset);
1035 btree_releasepage(page, GFP_NOFS);
1036 if (PagePrivate(page)) {
1037 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1038 "page private not zero on page %llu",
1039 (unsigned long long)page_offset(page));
1040 ClearPagePrivate(page);
1041 set_page_private(page, 0);
1042 page_cache_release(page);
1043 }
1044}
1045
1046static int btree_set_page_dirty(struct page *page)
1047{
1048#ifdef DEBUG
1049 struct extent_buffer *eb;
1050
1051 BUG_ON(!PagePrivate(page));
1052 eb = (struct extent_buffer *)page->private;
1053 BUG_ON(!eb);
1054 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1055 BUG_ON(!atomic_read(&eb->refs));
1056 btrfs_assert_tree_locked(eb);
1057#endif
1058 return __set_page_dirty_nobuffers(page);
1059}
1060
1061static const struct address_space_operations btree_aops = {
1062 .readpage = btree_readpage,
1063 .writepages = btree_writepages,
1064 .releasepage = btree_releasepage,
1065 .invalidatepage = btree_invalidatepage,
1066#ifdef CONFIG_MIGRATION
1067 .migratepage = btree_migratepage,
1068#endif
1069 .set_page_dirty = btree_set_page_dirty,
1070};
1071
1072void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1073{
1074 struct extent_buffer *buf = NULL;
1075 struct inode *btree_inode = root->fs_info->btree_inode;
1076
1077 buf = btrfs_find_create_tree_block(root, bytenr);
1078 if (!buf)
1079 return;
1080 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1081 buf, 0, WAIT_NONE, btree_get_extent, 0);
1082 free_extent_buffer(buf);
1083}
1084
1085int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1086 int mirror_num, struct extent_buffer **eb)
1087{
1088 struct extent_buffer *buf = NULL;
1089 struct inode *btree_inode = root->fs_info->btree_inode;
1090 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1091 int ret;
1092
1093 buf = btrfs_find_create_tree_block(root, bytenr);
1094 if (!buf)
1095 return 0;
1096
1097 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1098
1099 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1100 btree_get_extent, mirror_num);
1101 if (ret) {
1102 free_extent_buffer(buf);
1103 return ret;
1104 }
1105
1106 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1107 free_extent_buffer(buf);
1108 return -EIO;
1109 } else if (extent_buffer_uptodate(buf)) {
1110 *eb = buf;
1111 } else {
1112 free_extent_buffer(buf);
1113 }
1114 return 0;
1115}
1116
1117struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1118 u64 bytenr)
1119{
1120 return find_extent_buffer(fs_info, bytenr);
1121}
1122
1123struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1124 u64 bytenr)
1125{
1126 if (btrfs_test_is_dummy_root(root))
1127 return alloc_test_extent_buffer(root->fs_info, bytenr);
1128 return alloc_extent_buffer(root->fs_info, bytenr);
1129}
1130
1131
1132int btrfs_write_tree_block(struct extent_buffer *buf)
1133{
1134 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1135 buf->start + buf->len - 1);
1136}
1137
1138int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1139{
1140 return filemap_fdatawait_range(buf->pages[0]->mapping,
1141 buf->start, buf->start + buf->len - 1);
1142}
1143
1144struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1145 u64 parent_transid)
1146{
1147 struct extent_buffer *buf = NULL;
1148 int ret;
1149
1150 buf = btrfs_find_create_tree_block(root, bytenr);
1151 if (!buf)
1152 return ERR_PTR(-ENOMEM);
1153
1154 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1155 if (ret) {
1156 free_extent_buffer(buf);
1157 return ERR_PTR(ret);
1158 }
1159 return buf;
1160
1161}
1162
1163void clean_tree_block(struct btrfs_trans_handle *trans,
1164 struct btrfs_fs_info *fs_info,
1165 struct extent_buffer *buf)
1166{
1167 if (btrfs_header_generation(buf) ==
1168 fs_info->running_transaction->transid) {
1169 btrfs_assert_tree_locked(buf);
1170
1171 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1172 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1173 -buf->len,
1174 fs_info->dirty_metadata_batch);
1175 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1176 btrfs_set_lock_blocking(buf);
1177 clear_extent_buffer_dirty(buf);
1178 }
1179 }
1180}
1181
1182static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1183{
1184 struct btrfs_subvolume_writers *writers;
1185 int ret;
1186
1187 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1188 if (!writers)
1189 return ERR_PTR(-ENOMEM);
1190
1191 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1192 if (ret < 0) {
1193 kfree(writers);
1194 return ERR_PTR(ret);
1195 }
1196
1197 init_waitqueue_head(&writers->wait);
1198 return writers;
1199}
1200
1201static void
1202btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1203{
1204 percpu_counter_destroy(&writers->counter);
1205 kfree(writers);
1206}
1207
1208static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1209 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1210 u64 objectid)
1211{
1212 root->node = NULL;
1213 root->commit_root = NULL;
1214 root->sectorsize = sectorsize;
1215 root->nodesize = nodesize;
1216 root->stripesize = stripesize;
1217 root->state = 0;
1218 root->orphan_cleanup_state = 0;
1219
1220 root->objectid = objectid;
1221 root->last_trans = 0;
1222 root->highest_objectid = 0;
1223 root->nr_delalloc_inodes = 0;
1224 root->nr_ordered_extents = 0;
1225 root->name = NULL;
1226 root->inode_tree = RB_ROOT;
1227 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1228 root->block_rsv = NULL;
1229 root->orphan_block_rsv = NULL;
1230
1231 INIT_LIST_HEAD(&root->dirty_list);
1232 INIT_LIST_HEAD(&root->root_list);
1233 INIT_LIST_HEAD(&root->delalloc_inodes);
1234 INIT_LIST_HEAD(&root->delalloc_root);
1235 INIT_LIST_HEAD(&root->ordered_extents);
1236 INIT_LIST_HEAD(&root->ordered_root);
1237 INIT_LIST_HEAD(&root->logged_list[0]);
1238 INIT_LIST_HEAD(&root->logged_list[1]);
1239 spin_lock_init(&root->orphan_lock);
1240 spin_lock_init(&root->inode_lock);
1241 spin_lock_init(&root->delalloc_lock);
1242 spin_lock_init(&root->ordered_extent_lock);
1243 spin_lock_init(&root->accounting_lock);
1244 spin_lock_init(&root->log_extents_lock[0]);
1245 spin_lock_init(&root->log_extents_lock[1]);
1246 mutex_init(&root->objectid_mutex);
1247 mutex_init(&root->log_mutex);
1248 mutex_init(&root->ordered_extent_mutex);
1249 mutex_init(&root->delalloc_mutex);
1250 init_waitqueue_head(&root->log_writer_wait);
1251 init_waitqueue_head(&root->log_commit_wait[0]);
1252 init_waitqueue_head(&root->log_commit_wait[1]);
1253 INIT_LIST_HEAD(&root->log_ctxs[0]);
1254 INIT_LIST_HEAD(&root->log_ctxs[1]);
1255 atomic_set(&root->log_commit[0], 0);
1256 atomic_set(&root->log_commit[1], 0);
1257 atomic_set(&root->log_writers, 0);
1258 atomic_set(&root->log_batch, 0);
1259 atomic_set(&root->orphan_inodes, 0);
1260 atomic_set(&root->refs, 1);
1261 atomic_set(&root->will_be_snapshoted, 0);
1262 root->log_transid = 0;
1263 root->log_transid_committed = -1;
1264 root->last_log_commit = 0;
1265 if (fs_info)
1266 extent_io_tree_init(&root->dirty_log_pages,
1267 fs_info->btree_inode->i_mapping);
1268
1269 memset(&root->root_key, 0, sizeof(root->root_key));
1270 memset(&root->root_item, 0, sizeof(root->root_item));
1271 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1272 if (fs_info)
1273 root->defrag_trans_start = fs_info->generation;
1274 else
1275 root->defrag_trans_start = 0;
1276 root->root_key.objectid = objectid;
1277 root->anon_dev = 0;
1278
1279 spin_lock_init(&root->root_item_lock);
1280}
1281
1282static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1283{
1284 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1285 if (root)
1286 root->fs_info = fs_info;
1287 return root;
1288}
1289
1290#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1291/* Should only be used by the testing infrastructure */
1292struct btrfs_root *btrfs_alloc_dummy_root(void)
1293{
1294 struct btrfs_root *root;
1295
1296 root = btrfs_alloc_root(NULL);
1297 if (!root)
1298 return ERR_PTR(-ENOMEM);
1299 __setup_root(4096, 4096, 4096, root, NULL, 1);
1300 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1301 root->alloc_bytenr = 0;
1302
1303 return root;
1304}
1305#endif
1306
1307struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1308 struct btrfs_fs_info *fs_info,
1309 u64 objectid)
1310{
1311 struct extent_buffer *leaf;
1312 struct btrfs_root *tree_root = fs_info->tree_root;
1313 struct btrfs_root *root;
1314 struct btrfs_key key;
1315 int ret = 0;
1316 uuid_le uuid;
1317
1318 root = btrfs_alloc_root(fs_info);
1319 if (!root)
1320 return ERR_PTR(-ENOMEM);
1321
1322 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1323 tree_root->stripesize, root, fs_info, objectid);
1324 root->root_key.objectid = objectid;
1325 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1326 root->root_key.offset = 0;
1327
1328 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1329 if (IS_ERR(leaf)) {
1330 ret = PTR_ERR(leaf);
1331 leaf = NULL;
1332 goto fail;
1333 }
1334
1335 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1336 btrfs_set_header_bytenr(leaf, leaf->start);
1337 btrfs_set_header_generation(leaf, trans->transid);
1338 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1339 btrfs_set_header_owner(leaf, objectid);
1340 root->node = leaf;
1341
1342 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1343 BTRFS_FSID_SIZE);
1344 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1345 btrfs_header_chunk_tree_uuid(leaf),
1346 BTRFS_UUID_SIZE);
1347 btrfs_mark_buffer_dirty(leaf);
1348
1349 root->commit_root = btrfs_root_node(root);
1350 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1351
1352 root->root_item.flags = 0;
1353 root->root_item.byte_limit = 0;
1354 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1355 btrfs_set_root_generation(&root->root_item, trans->transid);
1356 btrfs_set_root_level(&root->root_item, 0);
1357 btrfs_set_root_refs(&root->root_item, 1);
1358 btrfs_set_root_used(&root->root_item, leaf->len);
1359 btrfs_set_root_last_snapshot(&root->root_item, 0);
1360 btrfs_set_root_dirid(&root->root_item, 0);
1361 uuid_le_gen(&uuid);
1362 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1363 root->root_item.drop_level = 0;
1364
1365 key.objectid = objectid;
1366 key.type = BTRFS_ROOT_ITEM_KEY;
1367 key.offset = 0;
1368 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1369 if (ret)
1370 goto fail;
1371
1372 btrfs_tree_unlock(leaf);
1373
1374 return root;
1375
1376fail:
1377 if (leaf) {
1378 btrfs_tree_unlock(leaf);
1379 free_extent_buffer(root->commit_root);
1380 free_extent_buffer(leaf);
1381 }
1382 kfree(root);
1383
1384 return ERR_PTR(ret);
1385}
1386
1387static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1388 struct btrfs_fs_info *fs_info)
1389{
1390 struct btrfs_root *root;
1391 struct btrfs_root *tree_root = fs_info->tree_root;
1392 struct extent_buffer *leaf;
1393
1394 root = btrfs_alloc_root(fs_info);
1395 if (!root)
1396 return ERR_PTR(-ENOMEM);
1397
1398 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1399 tree_root->stripesize, root, fs_info,
1400 BTRFS_TREE_LOG_OBJECTID);
1401
1402 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1403 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1404 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1405
1406 /*
1407 * DON'T set REF_COWS for log trees
1408 *
1409 * log trees do not get reference counted because they go away
1410 * before a real commit is actually done. They do store pointers
1411 * to file data extents, and those reference counts still get
1412 * updated (along with back refs to the log tree).
1413 */
1414
1415 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1416 NULL, 0, 0, 0);
1417 if (IS_ERR(leaf)) {
1418 kfree(root);
1419 return ERR_CAST(leaf);
1420 }
1421
1422 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1423 btrfs_set_header_bytenr(leaf, leaf->start);
1424 btrfs_set_header_generation(leaf, trans->transid);
1425 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1426 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1427 root->node = leaf;
1428
1429 write_extent_buffer(root->node, root->fs_info->fsid,
1430 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1431 btrfs_mark_buffer_dirty(root->node);
1432 btrfs_tree_unlock(root->node);
1433 return root;
1434}
1435
1436int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1437 struct btrfs_fs_info *fs_info)
1438{
1439 struct btrfs_root *log_root;
1440
1441 log_root = alloc_log_tree(trans, fs_info);
1442 if (IS_ERR(log_root))
1443 return PTR_ERR(log_root);
1444 WARN_ON(fs_info->log_root_tree);
1445 fs_info->log_root_tree = log_root;
1446 return 0;
1447}
1448
1449int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1450 struct btrfs_root *root)
1451{
1452 struct btrfs_root *log_root;
1453 struct btrfs_inode_item *inode_item;
1454
1455 log_root = alloc_log_tree(trans, root->fs_info);
1456 if (IS_ERR(log_root))
1457 return PTR_ERR(log_root);
1458
1459 log_root->last_trans = trans->transid;
1460 log_root->root_key.offset = root->root_key.objectid;
1461
1462 inode_item = &log_root->root_item.inode;
1463 btrfs_set_stack_inode_generation(inode_item, 1);
1464 btrfs_set_stack_inode_size(inode_item, 3);
1465 btrfs_set_stack_inode_nlink(inode_item, 1);
1466 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1467 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1468
1469 btrfs_set_root_node(&log_root->root_item, log_root->node);
1470
1471 WARN_ON(root->log_root);
1472 root->log_root = log_root;
1473 root->log_transid = 0;
1474 root->log_transid_committed = -1;
1475 root->last_log_commit = 0;
1476 return 0;
1477}
1478
1479static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1480 struct btrfs_key *key)
1481{
1482 struct btrfs_root *root;
1483 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1484 struct btrfs_path *path;
1485 u64 generation;
1486 int ret;
1487
1488 path = btrfs_alloc_path();
1489 if (!path)
1490 return ERR_PTR(-ENOMEM);
1491
1492 root = btrfs_alloc_root(fs_info);
1493 if (!root) {
1494 ret = -ENOMEM;
1495 goto alloc_fail;
1496 }
1497
1498 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1499 tree_root->stripesize, root, fs_info, key->objectid);
1500
1501 ret = btrfs_find_root(tree_root, key, path,
1502 &root->root_item, &root->root_key);
1503 if (ret) {
1504 if (ret > 0)
1505 ret = -ENOENT;
1506 goto find_fail;
1507 }
1508
1509 generation = btrfs_root_generation(&root->root_item);
1510 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1511 generation);
1512 if (IS_ERR(root->node)) {
1513 ret = PTR_ERR(root->node);
1514 goto find_fail;
1515 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1516 ret = -EIO;
1517 free_extent_buffer(root->node);
1518 goto find_fail;
1519 }
1520 root->commit_root = btrfs_root_node(root);
1521out:
1522 btrfs_free_path(path);
1523 return root;
1524
1525find_fail:
1526 kfree(root);
1527alloc_fail:
1528 root = ERR_PTR(ret);
1529 goto out;
1530}
1531
1532struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1533 struct btrfs_key *location)
1534{
1535 struct btrfs_root *root;
1536
1537 root = btrfs_read_tree_root(tree_root, location);
1538 if (IS_ERR(root))
1539 return root;
1540
1541 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1542 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1543 btrfs_check_and_init_root_item(&root->root_item);
1544 }
1545
1546 return root;
1547}
1548
1549int btrfs_init_fs_root(struct btrfs_root *root)
1550{
1551 int ret;
1552 struct btrfs_subvolume_writers *writers;
1553
1554 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1555 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1556 GFP_NOFS);
1557 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1558 ret = -ENOMEM;
1559 goto fail;
1560 }
1561
1562 writers = btrfs_alloc_subvolume_writers();
1563 if (IS_ERR(writers)) {
1564 ret = PTR_ERR(writers);
1565 goto fail;
1566 }
1567 root->subv_writers = writers;
1568
1569 btrfs_init_free_ino_ctl(root);
1570 spin_lock_init(&root->ino_cache_lock);
1571 init_waitqueue_head(&root->ino_cache_wait);
1572
1573 ret = get_anon_bdev(&root->anon_dev);
1574 if (ret)
1575 goto free_writers;
1576 return 0;
1577
1578free_writers:
1579 btrfs_free_subvolume_writers(root->subv_writers);
1580fail:
1581 kfree(root->free_ino_ctl);
1582 kfree(root->free_ino_pinned);
1583 return ret;
1584}
1585
1586static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1587 u64 root_id)
1588{
1589 struct btrfs_root *root;
1590
1591 spin_lock(&fs_info->fs_roots_radix_lock);
1592 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1593 (unsigned long)root_id);
1594 spin_unlock(&fs_info->fs_roots_radix_lock);
1595 return root;
1596}
1597
1598int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1599 struct btrfs_root *root)
1600{
1601 int ret;
1602
1603 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1604 if (ret)
1605 return ret;
1606
1607 spin_lock(&fs_info->fs_roots_radix_lock);
1608 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1609 (unsigned long)root->root_key.objectid,
1610 root);
1611 if (ret == 0)
1612 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1613 spin_unlock(&fs_info->fs_roots_radix_lock);
1614 radix_tree_preload_end();
1615
1616 return ret;
1617}
1618
1619struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1620 struct btrfs_key *location,
1621 bool check_ref)
1622{
1623 struct btrfs_root *root;
1624 struct btrfs_path *path;
1625 struct btrfs_key key;
1626 int ret;
1627
1628 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1629 return fs_info->tree_root;
1630 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1631 return fs_info->extent_root;
1632 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1633 return fs_info->chunk_root;
1634 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1635 return fs_info->dev_root;
1636 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1637 return fs_info->csum_root;
1638 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1639 return fs_info->quota_root ? fs_info->quota_root :
1640 ERR_PTR(-ENOENT);
1641 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1642 return fs_info->uuid_root ? fs_info->uuid_root :
1643 ERR_PTR(-ENOENT);
1644again:
1645 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1646 if (root) {
1647 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1648 return ERR_PTR(-ENOENT);
1649 return root;
1650 }
1651
1652 root = btrfs_read_fs_root(fs_info->tree_root, location);
1653 if (IS_ERR(root))
1654 return root;
1655
1656 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1657 ret = -ENOENT;
1658 goto fail;
1659 }
1660
1661 ret = btrfs_init_fs_root(root);
1662 if (ret)
1663 goto fail;
1664
1665 path = btrfs_alloc_path();
1666 if (!path) {
1667 ret = -ENOMEM;
1668 goto fail;
1669 }
1670 key.objectid = BTRFS_ORPHAN_OBJECTID;
1671 key.type = BTRFS_ORPHAN_ITEM_KEY;
1672 key.offset = location->objectid;
1673
1674 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1675 btrfs_free_path(path);
1676 if (ret < 0)
1677 goto fail;
1678 if (ret == 0)
1679 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1680
1681 ret = btrfs_insert_fs_root(fs_info, root);
1682 if (ret) {
1683 if (ret == -EEXIST) {
1684 free_fs_root(root);
1685 goto again;
1686 }
1687 goto fail;
1688 }
1689 return root;
1690fail:
1691 free_fs_root(root);
1692 return ERR_PTR(ret);
1693}
1694
1695static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1696{
1697 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1698 int ret = 0;
1699 struct btrfs_device *device;
1700 struct backing_dev_info *bdi;
1701
1702 rcu_read_lock();
1703 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1704 if (!device->bdev)
1705 continue;
1706 bdi = blk_get_backing_dev_info(device->bdev);
1707 if (bdi_congested(bdi, bdi_bits)) {
1708 ret = 1;
1709 break;
1710 }
1711 }
1712 rcu_read_unlock();
1713 return ret;
1714}
1715
1716static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1717{
1718 int err;
1719
1720 err = bdi_setup_and_register(bdi, "btrfs");
1721 if (err)
1722 return err;
1723
1724 bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE;
1725 bdi->congested_fn = btrfs_congested_fn;
1726 bdi->congested_data = info;
1727 return 0;
1728}
1729
1730/*
1731 * called by the kthread helper functions to finally call the bio end_io
1732 * functions. This is where read checksum verification actually happens
1733 */
1734static void end_workqueue_fn(struct btrfs_work *work)
1735{
1736 struct bio *bio;
1737 struct btrfs_end_io_wq *end_io_wq;
1738 int error;
1739
1740 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1741 bio = end_io_wq->bio;
1742
1743 error = end_io_wq->error;
1744 bio->bi_private = end_io_wq->private;
1745 bio->bi_end_io = end_io_wq->end_io;
1746 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1747 bio_endio_nodec(bio, error);
1748}
1749
1750static int cleaner_kthread(void *arg)
1751{
1752 struct btrfs_root *root = arg;
1753 int again;
1754
1755 do {
1756 again = 0;
1757
1758 /* Make the cleaner go to sleep early. */
1759 if (btrfs_need_cleaner_sleep(root))
1760 goto sleep;
1761
1762 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1763 goto sleep;
1764
1765 /*
1766 * Avoid the problem that we change the status of the fs
1767 * during the above check and trylock.
1768 */
1769 if (btrfs_need_cleaner_sleep(root)) {
1770 mutex_unlock(&root->fs_info->cleaner_mutex);
1771 goto sleep;
1772 }
1773
1774 btrfs_run_delayed_iputs(root);
1775 btrfs_delete_unused_bgs(root->fs_info);
1776 again = btrfs_clean_one_deleted_snapshot(root);
1777 mutex_unlock(&root->fs_info->cleaner_mutex);
1778
1779 /*
1780 * The defragger has dealt with the R/O remount and umount,
1781 * needn't do anything special here.
1782 */
1783 btrfs_run_defrag_inodes(root->fs_info);
1784sleep:
1785 if (!try_to_freeze() && !again) {
1786 set_current_state(TASK_INTERRUPTIBLE);
1787 if (!kthread_should_stop())
1788 schedule();
1789 __set_current_state(TASK_RUNNING);
1790 }
1791 } while (!kthread_should_stop());
1792 return 0;
1793}
1794
1795static int transaction_kthread(void *arg)
1796{
1797 struct btrfs_root *root = arg;
1798 struct btrfs_trans_handle *trans;
1799 struct btrfs_transaction *cur;
1800 u64 transid;
1801 unsigned long now;
1802 unsigned long delay;
1803 bool cannot_commit;
1804
1805 do {
1806 cannot_commit = false;
1807 delay = HZ * root->fs_info->commit_interval;
1808 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1809
1810 spin_lock(&root->fs_info->trans_lock);
1811 cur = root->fs_info->running_transaction;
1812 if (!cur) {
1813 spin_unlock(&root->fs_info->trans_lock);
1814 goto sleep;
1815 }
1816
1817 now = get_seconds();
1818 if (cur->state < TRANS_STATE_BLOCKED &&
1819 (now < cur->start_time ||
1820 now - cur->start_time < root->fs_info->commit_interval)) {
1821 spin_unlock(&root->fs_info->trans_lock);
1822 delay = HZ * 5;
1823 goto sleep;
1824 }
1825 transid = cur->transid;
1826 spin_unlock(&root->fs_info->trans_lock);
1827
1828 /* If the file system is aborted, this will always fail. */
1829 trans = btrfs_attach_transaction(root);
1830 if (IS_ERR(trans)) {
1831 if (PTR_ERR(trans) != -ENOENT)
1832 cannot_commit = true;
1833 goto sleep;
1834 }
1835 if (transid == trans->transid) {
1836 btrfs_commit_transaction(trans, root);
1837 } else {
1838 btrfs_end_transaction(trans, root);
1839 }
1840sleep:
1841 wake_up_process(root->fs_info->cleaner_kthread);
1842 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1843
1844 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1845 &root->fs_info->fs_state)))
1846 btrfs_cleanup_transaction(root);
1847 if (!try_to_freeze()) {
1848 set_current_state(TASK_INTERRUPTIBLE);
1849 if (!kthread_should_stop() &&
1850 (!btrfs_transaction_blocked(root->fs_info) ||
1851 cannot_commit))
1852 schedule_timeout(delay);
1853 __set_current_state(TASK_RUNNING);
1854 }
1855 } while (!kthread_should_stop());
1856 return 0;
1857}
1858
1859/*
1860 * this will find the highest generation in the array of
1861 * root backups. The index of the highest array is returned,
1862 * or -1 if we can't find anything.
1863 *
1864 * We check to make sure the array is valid by comparing the
1865 * generation of the latest root in the array with the generation
1866 * in the super block. If they don't match we pitch it.
1867 */
1868static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1869{
1870 u64 cur;
1871 int newest_index = -1;
1872 struct btrfs_root_backup *root_backup;
1873 int i;
1874
1875 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1876 root_backup = info->super_copy->super_roots + i;
1877 cur = btrfs_backup_tree_root_gen(root_backup);
1878 if (cur == newest_gen)
1879 newest_index = i;
1880 }
1881
1882 /* check to see if we actually wrapped around */
1883 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1884 root_backup = info->super_copy->super_roots;
1885 cur = btrfs_backup_tree_root_gen(root_backup);
1886 if (cur == newest_gen)
1887 newest_index = 0;
1888 }
1889 return newest_index;
1890}
1891
1892
1893/*
1894 * find the oldest backup so we know where to store new entries
1895 * in the backup array. This will set the backup_root_index
1896 * field in the fs_info struct
1897 */
1898static void find_oldest_super_backup(struct btrfs_fs_info *info,
1899 u64 newest_gen)
1900{
1901 int newest_index = -1;
1902
1903 newest_index = find_newest_super_backup(info, newest_gen);
1904 /* if there was garbage in there, just move along */
1905 if (newest_index == -1) {
1906 info->backup_root_index = 0;
1907 } else {
1908 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1909 }
1910}
1911
1912/*
1913 * copy all the root pointers into the super backup array.
1914 * this will bump the backup pointer by one when it is
1915 * done
1916 */
1917static void backup_super_roots(struct btrfs_fs_info *info)
1918{
1919 int next_backup;
1920 struct btrfs_root_backup *root_backup;
1921 int last_backup;
1922
1923 next_backup = info->backup_root_index;
1924 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1925 BTRFS_NUM_BACKUP_ROOTS;
1926
1927 /*
1928 * just overwrite the last backup if we're at the same generation
1929 * this happens only at umount
1930 */
1931 root_backup = info->super_for_commit->super_roots + last_backup;
1932 if (btrfs_backup_tree_root_gen(root_backup) ==
1933 btrfs_header_generation(info->tree_root->node))
1934 next_backup = last_backup;
1935
1936 root_backup = info->super_for_commit->super_roots + next_backup;
1937
1938 /*
1939 * make sure all of our padding and empty slots get zero filled
1940 * regardless of which ones we use today
1941 */
1942 memset(root_backup, 0, sizeof(*root_backup));
1943
1944 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1945
1946 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1947 btrfs_set_backup_tree_root_gen(root_backup,
1948 btrfs_header_generation(info->tree_root->node));
1949
1950 btrfs_set_backup_tree_root_level(root_backup,
1951 btrfs_header_level(info->tree_root->node));
1952
1953 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1954 btrfs_set_backup_chunk_root_gen(root_backup,
1955 btrfs_header_generation(info->chunk_root->node));
1956 btrfs_set_backup_chunk_root_level(root_backup,
1957 btrfs_header_level(info->chunk_root->node));
1958
1959 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1960 btrfs_set_backup_extent_root_gen(root_backup,
1961 btrfs_header_generation(info->extent_root->node));
1962 btrfs_set_backup_extent_root_level(root_backup,
1963 btrfs_header_level(info->extent_root->node));
1964
1965 /*
1966 * we might commit during log recovery, which happens before we set
1967 * the fs_root. Make sure it is valid before we fill it in.
1968 */
1969 if (info->fs_root && info->fs_root->node) {
1970 btrfs_set_backup_fs_root(root_backup,
1971 info->fs_root->node->start);
1972 btrfs_set_backup_fs_root_gen(root_backup,
1973 btrfs_header_generation(info->fs_root->node));
1974 btrfs_set_backup_fs_root_level(root_backup,
1975 btrfs_header_level(info->fs_root->node));
1976 }
1977
1978 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1979 btrfs_set_backup_dev_root_gen(root_backup,
1980 btrfs_header_generation(info->dev_root->node));
1981 btrfs_set_backup_dev_root_level(root_backup,
1982 btrfs_header_level(info->dev_root->node));
1983
1984 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1985 btrfs_set_backup_csum_root_gen(root_backup,
1986 btrfs_header_generation(info->csum_root->node));
1987 btrfs_set_backup_csum_root_level(root_backup,
1988 btrfs_header_level(info->csum_root->node));
1989
1990 btrfs_set_backup_total_bytes(root_backup,
1991 btrfs_super_total_bytes(info->super_copy));
1992 btrfs_set_backup_bytes_used(root_backup,
1993 btrfs_super_bytes_used(info->super_copy));
1994 btrfs_set_backup_num_devices(root_backup,
1995 btrfs_super_num_devices(info->super_copy));
1996
1997 /*
1998 * if we don't copy this out to the super_copy, it won't get remembered
1999 * for the next commit
2000 */
2001 memcpy(&info->super_copy->super_roots,
2002 &info->super_for_commit->super_roots,
2003 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2004}
2005
2006/*
2007 * this copies info out of the root backup array and back into
2008 * the in-memory super block. It is meant to help iterate through
2009 * the array, so you send it the number of backups you've already
2010 * tried and the last backup index you used.
2011 *
2012 * this returns -1 when it has tried all the backups
2013 */
2014static noinline int next_root_backup(struct btrfs_fs_info *info,
2015 struct btrfs_super_block *super,
2016 int *num_backups_tried, int *backup_index)
2017{
2018 struct btrfs_root_backup *root_backup;
2019 int newest = *backup_index;
2020
2021 if (*num_backups_tried == 0) {
2022 u64 gen = btrfs_super_generation(super);
2023
2024 newest = find_newest_super_backup(info, gen);
2025 if (newest == -1)
2026 return -1;
2027
2028 *backup_index = newest;
2029 *num_backups_tried = 1;
2030 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2031 /* we've tried all the backups, all done */
2032 return -1;
2033 } else {
2034 /* jump to the next oldest backup */
2035 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2036 BTRFS_NUM_BACKUP_ROOTS;
2037 *backup_index = newest;
2038 *num_backups_tried += 1;
2039 }
2040 root_backup = super->super_roots + newest;
2041
2042 btrfs_set_super_generation(super,
2043 btrfs_backup_tree_root_gen(root_backup));
2044 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2045 btrfs_set_super_root_level(super,
2046 btrfs_backup_tree_root_level(root_backup));
2047 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2048
2049 /*
2050 * fixme: the total bytes and num_devices need to match or we should
2051 * need a fsck
2052 */
2053 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2054 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2055 return 0;
2056}
2057
2058/* helper to cleanup workers */
2059static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2060{
2061 btrfs_destroy_workqueue(fs_info->fixup_workers);
2062 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2063 btrfs_destroy_workqueue(fs_info->workers);
2064 btrfs_destroy_workqueue(fs_info->endio_workers);
2065 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2066 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2067 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2068 btrfs_destroy_workqueue(fs_info->rmw_workers);
2069 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2070 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2071 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2072 btrfs_destroy_workqueue(fs_info->submit_workers);
2073 btrfs_destroy_workqueue(fs_info->delayed_workers);
2074 btrfs_destroy_workqueue(fs_info->caching_workers);
2075 btrfs_destroy_workqueue(fs_info->readahead_workers);
2076 btrfs_destroy_workqueue(fs_info->flush_workers);
2077 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2078 btrfs_destroy_workqueue(fs_info->extent_workers);
2079}
2080
2081static void free_root_extent_buffers(struct btrfs_root *root)
2082{
2083 if (root) {
2084 free_extent_buffer(root->node);
2085 free_extent_buffer(root->commit_root);
2086 root->node = NULL;
2087 root->commit_root = NULL;
2088 }
2089}
2090
2091/* helper to cleanup tree roots */
2092static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2093{
2094 free_root_extent_buffers(info->tree_root);
2095
2096 free_root_extent_buffers(info->dev_root);
2097 free_root_extent_buffers(info->extent_root);
2098 free_root_extent_buffers(info->csum_root);
2099 free_root_extent_buffers(info->quota_root);
2100 free_root_extent_buffers(info->uuid_root);
2101 if (chunk_root)
2102 free_root_extent_buffers(info->chunk_root);
2103}
2104
2105void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2106{
2107 int ret;
2108 struct btrfs_root *gang[8];
2109 int i;
2110
2111 while (!list_empty(&fs_info->dead_roots)) {
2112 gang[0] = list_entry(fs_info->dead_roots.next,
2113 struct btrfs_root, root_list);
2114 list_del(&gang[0]->root_list);
2115
2116 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2117 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2118 } else {
2119 free_extent_buffer(gang[0]->node);
2120 free_extent_buffer(gang[0]->commit_root);
2121 btrfs_put_fs_root(gang[0]);
2122 }
2123 }
2124
2125 while (1) {
2126 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2127 (void **)gang, 0,
2128 ARRAY_SIZE(gang));
2129 if (!ret)
2130 break;
2131 for (i = 0; i < ret; i++)
2132 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2133 }
2134
2135 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2136 btrfs_free_log_root_tree(NULL, fs_info);
2137 btrfs_destroy_pinned_extent(fs_info->tree_root,
2138 fs_info->pinned_extents);
2139 }
2140}
2141
2142static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2143{
2144 mutex_init(&fs_info->scrub_lock);
2145 atomic_set(&fs_info->scrubs_running, 0);
2146 atomic_set(&fs_info->scrub_pause_req, 0);
2147 atomic_set(&fs_info->scrubs_paused, 0);
2148 atomic_set(&fs_info->scrub_cancel_req, 0);
2149 init_waitqueue_head(&fs_info->scrub_pause_wait);
2150 fs_info->scrub_workers_refcnt = 0;
2151}
2152
2153static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2154{
2155 spin_lock_init(&fs_info->balance_lock);
2156 mutex_init(&fs_info->balance_mutex);
2157 atomic_set(&fs_info->balance_running, 0);
2158 atomic_set(&fs_info->balance_pause_req, 0);
2159 atomic_set(&fs_info->balance_cancel_req, 0);
2160 fs_info->balance_ctl = NULL;
2161 init_waitqueue_head(&fs_info->balance_wait_q);
2162}
2163
2164static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2165 struct btrfs_root *tree_root)
2166{
2167 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2168 set_nlink(fs_info->btree_inode, 1);
2169 /*
2170 * we set the i_size on the btree inode to the max possible int.
2171 * the real end of the address space is determined by all of
2172 * the devices in the system
2173 */
2174 fs_info->btree_inode->i_size = OFFSET_MAX;
2175 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2176
2177 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2178 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2179 fs_info->btree_inode->i_mapping);
2180 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2181 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2182
2183 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2184
2185 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2186 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2187 sizeof(struct btrfs_key));
2188 set_bit(BTRFS_INODE_DUMMY,
2189 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2190 btrfs_insert_inode_hash(fs_info->btree_inode);
2191}
2192
2193static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2194{
2195 fs_info->dev_replace.lock_owner = 0;
2196 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2197 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2198 mutex_init(&fs_info->dev_replace.lock_management_lock);
2199 mutex_init(&fs_info->dev_replace.lock);
2200 init_waitqueue_head(&fs_info->replace_wait);
2201}
2202
2203static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2204{
2205 spin_lock_init(&fs_info->qgroup_lock);
2206 mutex_init(&fs_info->qgroup_ioctl_lock);
2207 fs_info->qgroup_tree = RB_ROOT;
2208 fs_info->qgroup_op_tree = RB_ROOT;
2209 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2210 fs_info->qgroup_seq = 1;
2211 fs_info->quota_enabled = 0;
2212 fs_info->pending_quota_state = 0;
2213 fs_info->qgroup_ulist = NULL;
2214 mutex_init(&fs_info->qgroup_rescan_lock);
2215}
2216
2217static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2218 struct btrfs_fs_devices *fs_devices)
2219{
2220 int max_active = fs_info->thread_pool_size;
2221 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2222
2223 fs_info->workers =
2224 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2225 max_active, 16);
2226
2227 fs_info->delalloc_workers =
2228 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2229
2230 fs_info->flush_workers =
2231 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2232
2233 fs_info->caching_workers =
2234 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2235
2236 /*
2237 * a higher idle thresh on the submit workers makes it much more
2238 * likely that bios will be send down in a sane order to the
2239 * devices
2240 */
2241 fs_info->submit_workers =
2242 btrfs_alloc_workqueue("submit", flags,
2243 min_t(u64, fs_devices->num_devices,
2244 max_active), 64);
2245
2246 fs_info->fixup_workers =
2247 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2248
2249 /*
2250 * endios are largely parallel and should have a very
2251 * low idle thresh
2252 */
2253 fs_info->endio_workers =
2254 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2255 fs_info->endio_meta_workers =
2256 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2257 fs_info->endio_meta_write_workers =
2258 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2259 fs_info->endio_raid56_workers =
2260 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2261 fs_info->endio_repair_workers =
2262 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2263 fs_info->rmw_workers =
2264 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2265 fs_info->endio_write_workers =
2266 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2267 fs_info->endio_freespace_worker =
2268 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2269 fs_info->delayed_workers =
2270 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2271 fs_info->readahead_workers =
2272 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2273 fs_info->qgroup_rescan_workers =
2274 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2275 fs_info->extent_workers =
2276 btrfs_alloc_workqueue("extent-refs", flags,
2277 min_t(u64, fs_devices->num_devices,
2278 max_active), 8);
2279
2280 if (!(fs_info->workers && fs_info->delalloc_workers &&
2281 fs_info->submit_workers && fs_info->flush_workers &&
2282 fs_info->endio_workers && fs_info->endio_meta_workers &&
2283 fs_info->endio_meta_write_workers &&
2284 fs_info->endio_repair_workers &&
2285 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2286 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2287 fs_info->caching_workers && fs_info->readahead_workers &&
2288 fs_info->fixup_workers && fs_info->delayed_workers &&
2289 fs_info->extent_workers &&
2290 fs_info->qgroup_rescan_workers)) {
2291 return -ENOMEM;
2292 }
2293
2294 return 0;
2295}
2296
2297static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2298 struct btrfs_fs_devices *fs_devices)
2299{
2300 int ret;
2301 struct btrfs_root *tree_root = fs_info->tree_root;
2302 struct btrfs_root *log_tree_root;
2303 struct btrfs_super_block *disk_super = fs_info->super_copy;
2304 u64 bytenr = btrfs_super_log_root(disk_super);
2305
2306 if (fs_devices->rw_devices == 0) {
2307 printk(KERN_WARNING "BTRFS: log replay required "
2308 "on RO media\n");
2309 return -EIO;
2310 }
2311
2312 log_tree_root = btrfs_alloc_root(fs_info);
2313 if (!log_tree_root)
2314 return -ENOMEM;
2315
2316 __setup_root(tree_root->nodesize, tree_root->sectorsize,
2317 tree_root->stripesize, log_tree_root, fs_info,
2318 BTRFS_TREE_LOG_OBJECTID);
2319
2320 log_tree_root->node = read_tree_block(tree_root, bytenr,
2321 fs_info->generation + 1);
2322 if (IS_ERR(log_tree_root->node)) {
2323 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2324 ret = PTR_ERR(log_tree_root->node);
2325 kfree(log_tree_root);
2326 return ret;
2327 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2328 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2329 free_extent_buffer(log_tree_root->node);
2330 kfree(log_tree_root);
2331 return -EIO;
2332 }
2333 /* returns with log_tree_root freed on success */
2334 ret = btrfs_recover_log_trees(log_tree_root);
2335 if (ret) {
2336 btrfs_error(tree_root->fs_info, ret,
2337 "Failed to recover log tree");
2338 free_extent_buffer(log_tree_root->node);
2339 kfree(log_tree_root);
2340 return ret;
2341 }
2342
2343 if (fs_info->sb->s_flags & MS_RDONLY) {
2344 ret = btrfs_commit_super(tree_root);
2345 if (ret)
2346 return ret;
2347 }
2348
2349 return 0;
2350}
2351
2352static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2353 struct btrfs_root *tree_root)
2354{
2355 struct btrfs_root *root;
2356 struct btrfs_key location;
2357 int ret;
2358
2359 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2360 location.type = BTRFS_ROOT_ITEM_KEY;
2361 location.offset = 0;
2362
2363 root = btrfs_read_tree_root(tree_root, &location);
2364 if (IS_ERR(root))
2365 return PTR_ERR(root);
2366 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2367 fs_info->extent_root = root;
2368
2369 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2370 root = btrfs_read_tree_root(tree_root, &location);
2371 if (IS_ERR(root))
2372 return PTR_ERR(root);
2373 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2374 fs_info->dev_root = root;
2375 btrfs_init_devices_late(fs_info);
2376
2377 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2378 root = btrfs_read_tree_root(tree_root, &location);
2379 if (IS_ERR(root))
2380 return PTR_ERR(root);
2381 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2382 fs_info->csum_root = root;
2383
2384 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2385 root = btrfs_read_tree_root(tree_root, &location);
2386 if (!IS_ERR(root)) {
2387 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2388 fs_info->quota_enabled = 1;
2389 fs_info->pending_quota_state = 1;
2390 fs_info->quota_root = root;
2391 }
2392
2393 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2394 root = btrfs_read_tree_root(tree_root, &location);
2395 if (IS_ERR(root)) {
2396 ret = PTR_ERR(root);
2397 if (ret != -ENOENT)
2398 return ret;
2399 } else {
2400 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2401 fs_info->uuid_root = root;
2402 }
2403
2404 return 0;
2405}
2406
2407int open_ctree(struct super_block *sb,
2408 struct btrfs_fs_devices *fs_devices,
2409 char *options)
2410{
2411 u32 sectorsize;
2412 u32 nodesize;
2413 u32 stripesize;
2414 u64 generation;
2415 u64 features;
2416 struct btrfs_key location;
2417 struct buffer_head *bh;
2418 struct btrfs_super_block *disk_super;
2419 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2420 struct btrfs_root *tree_root;
2421 struct btrfs_root *chunk_root;
2422 int ret;
2423 int err = -EINVAL;
2424 int num_backups_tried = 0;
2425 int backup_index = 0;
2426 int max_active;
2427
2428 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2429 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2430 if (!tree_root || !chunk_root) {
2431 err = -ENOMEM;
2432 goto fail;
2433 }
2434
2435 ret = init_srcu_struct(&fs_info->subvol_srcu);
2436 if (ret) {
2437 err = ret;
2438 goto fail;
2439 }
2440
2441 ret = setup_bdi(fs_info, &fs_info->bdi);
2442 if (ret) {
2443 err = ret;
2444 goto fail_srcu;
2445 }
2446
2447 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2448 if (ret) {
2449 err = ret;
2450 goto fail_bdi;
2451 }
2452 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2453 (1 + ilog2(nr_cpu_ids));
2454
2455 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2456 if (ret) {
2457 err = ret;
2458 goto fail_dirty_metadata_bytes;
2459 }
2460
2461 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2462 if (ret) {
2463 err = ret;
2464 goto fail_delalloc_bytes;
2465 }
2466
2467 fs_info->btree_inode = new_inode(sb);
2468 if (!fs_info->btree_inode) {
2469 err = -ENOMEM;
2470 goto fail_bio_counter;
2471 }
2472
2473 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2474
2475 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2476 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2477 INIT_LIST_HEAD(&fs_info->trans_list);
2478 INIT_LIST_HEAD(&fs_info->dead_roots);
2479 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2480 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2481 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2482 spin_lock_init(&fs_info->delalloc_root_lock);
2483 spin_lock_init(&fs_info->trans_lock);
2484 spin_lock_init(&fs_info->fs_roots_radix_lock);
2485 spin_lock_init(&fs_info->delayed_iput_lock);
2486 spin_lock_init(&fs_info->defrag_inodes_lock);
2487 spin_lock_init(&fs_info->free_chunk_lock);
2488 spin_lock_init(&fs_info->tree_mod_seq_lock);
2489 spin_lock_init(&fs_info->super_lock);
2490 spin_lock_init(&fs_info->qgroup_op_lock);
2491 spin_lock_init(&fs_info->buffer_lock);
2492 spin_lock_init(&fs_info->unused_bgs_lock);
2493 rwlock_init(&fs_info->tree_mod_log_lock);
2494 mutex_init(&fs_info->unused_bg_unpin_mutex);
2495 mutex_init(&fs_info->reloc_mutex);
2496 mutex_init(&fs_info->delalloc_root_mutex);
2497 seqlock_init(&fs_info->profiles_lock);
2498 init_rwsem(&fs_info->delayed_iput_sem);
2499
2500 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2501 INIT_LIST_HEAD(&fs_info->space_info);
2502 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2503 INIT_LIST_HEAD(&fs_info->unused_bgs);
2504 btrfs_mapping_init(&fs_info->mapping_tree);
2505 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2506 BTRFS_BLOCK_RSV_GLOBAL);
2507 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2508 BTRFS_BLOCK_RSV_DELALLOC);
2509 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2510 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2511 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2512 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2513 BTRFS_BLOCK_RSV_DELOPS);
2514 atomic_set(&fs_info->nr_async_submits, 0);
2515 atomic_set(&fs_info->async_delalloc_pages, 0);
2516 atomic_set(&fs_info->async_submit_draining, 0);
2517 atomic_set(&fs_info->nr_async_bios, 0);
2518 atomic_set(&fs_info->defrag_running, 0);
2519 atomic_set(&fs_info->qgroup_op_seq, 0);
2520 atomic64_set(&fs_info->tree_mod_seq, 0);
2521 fs_info->sb = sb;
2522 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2523 fs_info->metadata_ratio = 0;
2524 fs_info->defrag_inodes = RB_ROOT;
2525 fs_info->free_chunk_space = 0;
2526 fs_info->tree_mod_log = RB_ROOT;
2527 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2528 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2529 /* readahead state */
2530 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2531 spin_lock_init(&fs_info->reada_lock);
2532
2533 fs_info->thread_pool_size = min_t(unsigned long,
2534 num_online_cpus() + 2, 8);
2535
2536 INIT_LIST_HEAD(&fs_info->ordered_roots);
2537 spin_lock_init(&fs_info->ordered_root_lock);
2538 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2539 GFP_NOFS);
2540 if (!fs_info->delayed_root) {
2541 err = -ENOMEM;
2542 goto fail_iput;
2543 }
2544 btrfs_init_delayed_root(fs_info->delayed_root);
2545
2546 btrfs_init_scrub(fs_info);
2547#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2548 fs_info->check_integrity_print_mask = 0;
2549#endif
2550 btrfs_init_balance(fs_info);
2551 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2552
2553 sb->s_blocksize = 4096;
2554 sb->s_blocksize_bits = blksize_bits(4096);
2555 sb->s_bdi = &fs_info->bdi;
2556
2557 btrfs_init_btree_inode(fs_info, tree_root);
2558
2559 spin_lock_init(&fs_info->block_group_cache_lock);
2560 fs_info->block_group_cache_tree = RB_ROOT;
2561 fs_info->first_logical_byte = (u64)-1;
2562
2563 extent_io_tree_init(&fs_info->freed_extents[0],
2564 fs_info->btree_inode->i_mapping);
2565 extent_io_tree_init(&fs_info->freed_extents[1],
2566 fs_info->btree_inode->i_mapping);
2567 fs_info->pinned_extents = &fs_info->freed_extents[0];
2568 fs_info->do_barriers = 1;
2569
2570
2571 mutex_init(&fs_info->ordered_operations_mutex);
2572 mutex_init(&fs_info->ordered_extent_flush_mutex);
2573 mutex_init(&fs_info->tree_log_mutex);
2574 mutex_init(&fs_info->chunk_mutex);
2575 mutex_init(&fs_info->transaction_kthread_mutex);
2576 mutex_init(&fs_info->cleaner_mutex);
2577 mutex_init(&fs_info->volume_mutex);
2578 mutex_init(&fs_info->ro_block_group_mutex);
2579 init_rwsem(&fs_info->commit_root_sem);
2580 init_rwsem(&fs_info->cleanup_work_sem);
2581 init_rwsem(&fs_info->subvol_sem);
2582 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2583
2584 btrfs_init_dev_replace_locks(fs_info);
2585 btrfs_init_qgroup(fs_info);
2586
2587 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2588 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2589
2590 init_waitqueue_head(&fs_info->transaction_throttle);
2591 init_waitqueue_head(&fs_info->transaction_wait);
2592 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2593 init_waitqueue_head(&fs_info->async_submit_wait);
2594
2595 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2596
2597 ret = btrfs_alloc_stripe_hash_table(fs_info);
2598 if (ret) {
2599 err = ret;
2600 goto fail_alloc;
2601 }
2602
2603 __setup_root(4096, 4096, 4096, tree_root,
2604 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2605
2606 invalidate_bdev(fs_devices->latest_bdev);
2607
2608 /*
2609 * Read super block and check the signature bytes only
2610 */
2611 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2612 if (!bh) {
2613 err = -EINVAL;
2614 goto fail_alloc;
2615 }
2616
2617 /*
2618 * We want to check superblock checksum, the type is stored inside.
2619 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2620 */
2621 if (btrfs_check_super_csum(bh->b_data)) {
2622 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2623 err = -EINVAL;
2624 goto fail_alloc;
2625 }
2626
2627 /*
2628 * super_copy is zeroed at allocation time and we never touch the
2629 * following bytes up to INFO_SIZE, the checksum is calculated from
2630 * the whole block of INFO_SIZE
2631 */
2632 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2633 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2634 sizeof(*fs_info->super_for_commit));
2635 brelse(bh);
2636
2637 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2638
2639 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2640 if (ret) {
2641 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2642 err = -EINVAL;
2643 goto fail_alloc;
2644 }
2645
2646 disk_super = fs_info->super_copy;
2647 if (!btrfs_super_root(disk_super))
2648 goto fail_alloc;
2649
2650 /* check FS state, whether FS is broken. */
2651 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2652 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2653
2654 /*
2655 * run through our array of backup supers and setup
2656 * our ring pointer to the oldest one
2657 */
2658 generation = btrfs_super_generation(disk_super);
2659 find_oldest_super_backup(fs_info, generation);
2660
2661 /*
2662 * In the long term, we'll store the compression type in the super
2663 * block, and it'll be used for per file compression control.
2664 */
2665 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2666
2667 ret = btrfs_parse_options(tree_root, options);
2668 if (ret) {
2669 err = ret;
2670 goto fail_alloc;
2671 }
2672
2673 features = btrfs_super_incompat_flags(disk_super) &
2674 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2675 if (features) {
2676 printk(KERN_ERR "BTRFS: couldn't mount because of "
2677 "unsupported optional features (%Lx).\n",
2678 features);
2679 err = -EINVAL;
2680 goto fail_alloc;
2681 }
2682
2683 /*
2684 * Leafsize and nodesize were always equal, this is only a sanity check.
2685 */
2686 if (le32_to_cpu(disk_super->__unused_leafsize) !=
2687 btrfs_super_nodesize(disk_super)) {
2688 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2689 "blocksizes don't match. node %d leaf %d\n",
2690 btrfs_super_nodesize(disk_super),
2691 le32_to_cpu(disk_super->__unused_leafsize));
2692 err = -EINVAL;
2693 goto fail_alloc;
2694 }
2695 if (btrfs_super_nodesize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2696 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2697 "blocksize (%d) was too large\n",
2698 btrfs_super_nodesize(disk_super));
2699 err = -EINVAL;
2700 goto fail_alloc;
2701 }
2702
2703 features = btrfs_super_incompat_flags(disk_super);
2704 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2705 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2706 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2707
2708 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2709 printk(KERN_INFO "BTRFS: has skinny extents\n");
2710
2711 /*
2712 * flag our filesystem as having big metadata blocks if
2713 * they are bigger than the page size
2714 */
2715 if (btrfs_super_nodesize(disk_super) > PAGE_CACHE_SIZE) {
2716 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2717 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2718 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2719 }
2720
2721 nodesize = btrfs_super_nodesize(disk_super);
2722 sectorsize = btrfs_super_sectorsize(disk_super);
2723 stripesize = btrfs_super_stripesize(disk_super);
2724 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2725 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2726
2727 /*
2728 * mixed block groups end up with duplicate but slightly offset
2729 * extent buffers for the same range. It leads to corruptions
2730 */
2731 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2732 (sectorsize != nodesize)) {
2733 printk(KERN_ERR "BTRFS: unequal leaf/node/sector sizes "
2734 "are not allowed for mixed block groups on %s\n",
2735 sb->s_id);
2736 goto fail_alloc;
2737 }
2738
2739 /*
2740 * Needn't use the lock because there is no other task which will
2741 * update the flag.
2742 */
2743 btrfs_set_super_incompat_flags(disk_super, features);
2744
2745 features = btrfs_super_compat_ro_flags(disk_super) &
2746 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2747 if (!(sb->s_flags & MS_RDONLY) && features) {
2748 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2749 "unsupported option features (%Lx).\n",
2750 features);
2751 err = -EINVAL;
2752 goto fail_alloc;
2753 }
2754
2755 max_active = fs_info->thread_pool_size;
2756
2757 ret = btrfs_init_workqueues(fs_info, fs_devices);
2758 if (ret) {
2759 err = ret;
2760 goto fail_sb_buffer;
2761 }
2762
2763 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2764 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2765 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2766
2767 tree_root->nodesize = nodesize;
2768 tree_root->sectorsize = sectorsize;
2769 tree_root->stripesize = stripesize;
2770
2771 sb->s_blocksize = sectorsize;
2772 sb->s_blocksize_bits = blksize_bits(sectorsize);
2773
2774 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2775 printk(KERN_ERR "BTRFS: valid FS not found on %s\n", sb->s_id);
2776 goto fail_sb_buffer;
2777 }
2778
2779 if (sectorsize != PAGE_SIZE) {
2780 printk(KERN_ERR "BTRFS: incompatible sector size (%lu) "
2781 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2782 goto fail_sb_buffer;
2783 }
2784
2785 mutex_lock(&fs_info->chunk_mutex);
2786 ret = btrfs_read_sys_array(tree_root);
2787 mutex_unlock(&fs_info->chunk_mutex);
2788 if (ret) {
2789 printk(KERN_ERR "BTRFS: failed to read the system "
2790 "array on %s\n", sb->s_id);
2791 goto fail_sb_buffer;
2792 }
2793
2794 generation = btrfs_super_chunk_root_generation(disk_super);
2795
2796 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2797 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2798
2799 chunk_root->node = read_tree_block(chunk_root,
2800 btrfs_super_chunk_root(disk_super),
2801 generation);
2802 if (IS_ERR(chunk_root->node) ||
2803 !extent_buffer_uptodate(chunk_root->node)) {
2804 printk(KERN_ERR "BTRFS: failed to read chunk root on %s\n",
2805 sb->s_id);
2806 goto fail_tree_roots;
2807 }
2808 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2809 chunk_root->commit_root = btrfs_root_node(chunk_root);
2810
2811 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2812 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2813
2814 ret = btrfs_read_chunk_tree(chunk_root);
2815 if (ret) {
2816 printk(KERN_ERR "BTRFS: failed to read chunk tree on %s\n",
2817 sb->s_id);
2818 goto fail_tree_roots;
2819 }
2820
2821 /*
2822 * keep the device that is marked to be the target device for the
2823 * dev_replace procedure
2824 */
2825 btrfs_close_extra_devices(fs_devices, 0);
2826
2827 if (!fs_devices->latest_bdev) {
2828 printk(KERN_ERR "BTRFS: failed to read devices on %s\n",
2829 sb->s_id);
2830 goto fail_tree_roots;
2831 }
2832
2833retry_root_backup:
2834 generation = btrfs_super_generation(disk_super);
2835
2836 tree_root->node = read_tree_block(tree_root,
2837 btrfs_super_root(disk_super),
2838 generation);
2839 if (IS_ERR(tree_root->node) ||
2840 !extent_buffer_uptodate(tree_root->node)) {
2841 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2842 sb->s_id);
2843
2844 goto recovery_tree_root;
2845 }
2846
2847 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2848 tree_root->commit_root = btrfs_root_node(tree_root);
2849 btrfs_set_root_refs(&tree_root->root_item, 1);
2850
2851 ret = btrfs_read_roots(fs_info, tree_root);
2852 if (ret)
2853 goto recovery_tree_root;
2854
2855 fs_info->generation = generation;
2856 fs_info->last_trans_committed = generation;
2857
2858 ret = btrfs_recover_balance(fs_info);
2859 if (ret) {
2860 printk(KERN_ERR "BTRFS: failed to recover balance\n");
2861 goto fail_block_groups;
2862 }
2863
2864 ret = btrfs_init_dev_stats(fs_info);
2865 if (ret) {
2866 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2867 ret);
2868 goto fail_block_groups;
2869 }
2870
2871 ret = btrfs_init_dev_replace(fs_info);
2872 if (ret) {
2873 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2874 goto fail_block_groups;
2875 }
2876
2877 btrfs_close_extra_devices(fs_devices, 1);
2878
2879 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2880 if (ret) {
2881 pr_err("BTRFS: failed to init sysfs fsid interface: %d\n", ret);
2882 goto fail_block_groups;
2883 }
2884
2885 ret = btrfs_sysfs_add_device(fs_devices);
2886 if (ret) {
2887 pr_err("BTRFS: failed to init sysfs device interface: %d\n", ret);
2888 goto fail_fsdev_sysfs;
2889 }
2890
2891 ret = btrfs_sysfs_add_one(fs_info);
2892 if (ret) {
2893 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2894 goto fail_fsdev_sysfs;
2895 }
2896
2897 ret = btrfs_init_space_info(fs_info);
2898 if (ret) {
2899 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2900 goto fail_sysfs;
2901 }
2902
2903 ret = btrfs_read_block_groups(fs_info->extent_root);
2904 if (ret) {
2905 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2906 goto fail_sysfs;
2907 }
2908 fs_info->num_tolerated_disk_barrier_failures =
2909 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2910 if (fs_info->fs_devices->missing_devices >
2911 fs_info->num_tolerated_disk_barrier_failures &&
2912 !(sb->s_flags & MS_RDONLY)) {
2913 printk(KERN_WARNING "BTRFS: "
2914 "too many missing devices, writeable mount is not allowed\n");
2915 goto fail_sysfs;
2916 }
2917
2918 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2919 "btrfs-cleaner");
2920 if (IS_ERR(fs_info->cleaner_kthread))
2921 goto fail_sysfs;
2922
2923 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2924 tree_root,
2925 "btrfs-transaction");
2926 if (IS_ERR(fs_info->transaction_kthread))
2927 goto fail_cleaner;
2928
2929 if (!btrfs_test_opt(tree_root, SSD) &&
2930 !btrfs_test_opt(tree_root, NOSSD) &&
2931 !fs_info->fs_devices->rotating) {
2932 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2933 "mode\n");
2934 btrfs_set_opt(fs_info->mount_opt, SSD);
2935 }
2936
2937 /*
2938 * Mount does not set all options immediatelly, we can do it now and do
2939 * not have to wait for transaction commit
2940 */
2941 btrfs_apply_pending_changes(fs_info);
2942
2943#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2944 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2945 ret = btrfsic_mount(tree_root, fs_devices,
2946 btrfs_test_opt(tree_root,
2947 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2948 1 : 0,
2949 fs_info->check_integrity_print_mask);
2950 if (ret)
2951 printk(KERN_WARNING "BTRFS: failed to initialize"
2952 " integrity check module %s\n", sb->s_id);
2953 }
2954#endif
2955 ret = btrfs_read_qgroup_config(fs_info);
2956 if (ret)
2957 goto fail_trans_kthread;
2958
2959 /* do not make disk changes in broken FS */
2960 if (btrfs_super_log_root(disk_super) != 0) {
2961 ret = btrfs_replay_log(fs_info, fs_devices);
2962 if (ret) {
2963 err = ret;
2964 goto fail_qgroup;
2965 }
2966 }
2967
2968 ret = btrfs_find_orphan_roots(tree_root);
2969 if (ret)
2970 goto fail_qgroup;
2971
2972 if (!(sb->s_flags & MS_RDONLY)) {
2973 ret = btrfs_cleanup_fs_roots(fs_info);
2974 if (ret)
2975 goto fail_qgroup;
2976
2977 mutex_lock(&fs_info->cleaner_mutex);
2978 ret = btrfs_recover_relocation(tree_root);
2979 mutex_unlock(&fs_info->cleaner_mutex);
2980 if (ret < 0) {
2981 printk(KERN_WARNING
2982 "BTRFS: failed to recover relocation\n");
2983 err = -EINVAL;
2984 goto fail_qgroup;
2985 }
2986 }
2987
2988 location.objectid = BTRFS_FS_TREE_OBJECTID;
2989 location.type = BTRFS_ROOT_ITEM_KEY;
2990 location.offset = 0;
2991
2992 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2993 if (IS_ERR(fs_info->fs_root)) {
2994 err = PTR_ERR(fs_info->fs_root);
2995 goto fail_qgroup;
2996 }
2997
2998 if (sb->s_flags & MS_RDONLY)
2999 return 0;
3000
3001 down_read(&fs_info->cleanup_work_sem);
3002 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3003 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3004 up_read(&fs_info->cleanup_work_sem);
3005 close_ctree(tree_root);
3006 return ret;
3007 }
3008 up_read(&fs_info->cleanup_work_sem);
3009
3010 ret = btrfs_resume_balance_async(fs_info);
3011 if (ret) {
3012 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
3013 close_ctree(tree_root);
3014 return ret;
3015 }
3016
3017 ret = btrfs_resume_dev_replace_async(fs_info);
3018 if (ret) {
3019 pr_warn("BTRFS: failed to resume dev_replace\n");
3020 close_ctree(tree_root);
3021 return ret;
3022 }
3023
3024 btrfs_qgroup_rescan_resume(fs_info);
3025
3026 if (!fs_info->uuid_root) {
3027 pr_info("BTRFS: creating UUID tree\n");
3028 ret = btrfs_create_uuid_tree(fs_info);
3029 if (ret) {
3030 pr_warn("BTRFS: failed to create the UUID tree %d\n",
3031 ret);
3032 close_ctree(tree_root);
3033 return ret;
3034 }
3035 } else if (btrfs_test_opt(tree_root, RESCAN_UUID_TREE) ||
3036 fs_info->generation !=
3037 btrfs_super_uuid_tree_generation(disk_super)) {
3038 pr_info("BTRFS: checking UUID tree\n");
3039 ret = btrfs_check_uuid_tree(fs_info);
3040 if (ret) {
3041 pr_warn("BTRFS: failed to check the UUID tree %d\n",
3042 ret);
3043 close_ctree(tree_root);
3044 return ret;
3045 }
3046 } else {
3047 fs_info->update_uuid_tree_gen = 1;
3048 }
3049
3050 fs_info->open = 1;
3051
3052 return 0;
3053
3054fail_qgroup:
3055 btrfs_free_qgroup_config(fs_info);
3056fail_trans_kthread:
3057 kthread_stop(fs_info->transaction_kthread);
3058 btrfs_cleanup_transaction(fs_info->tree_root);
3059 btrfs_free_fs_roots(fs_info);
3060fail_cleaner:
3061 kthread_stop(fs_info->cleaner_kthread);
3062
3063 /*
3064 * make sure we're done with the btree inode before we stop our
3065 * kthreads
3066 */
3067 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3068
3069fail_sysfs:
3070 btrfs_sysfs_remove_one(fs_info);
3071
3072fail_fsdev_sysfs:
3073 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3074
3075fail_block_groups:
3076 btrfs_put_block_group_cache(fs_info);
3077 btrfs_free_block_groups(fs_info);
3078
3079fail_tree_roots:
3080 free_root_pointers(fs_info, 1);
3081 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3082
3083fail_sb_buffer:
3084 btrfs_stop_all_workers(fs_info);
3085fail_alloc:
3086fail_iput:
3087 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3088
3089 iput(fs_info->btree_inode);
3090fail_bio_counter:
3091 percpu_counter_destroy(&fs_info->bio_counter);
3092fail_delalloc_bytes:
3093 percpu_counter_destroy(&fs_info->delalloc_bytes);
3094fail_dirty_metadata_bytes:
3095 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3096fail_bdi:
3097 bdi_destroy(&fs_info->bdi);
3098fail_srcu:
3099 cleanup_srcu_struct(&fs_info->subvol_srcu);
3100fail:
3101 btrfs_free_stripe_hash_table(fs_info);
3102 btrfs_close_devices(fs_info->fs_devices);
3103 return err;
3104
3105recovery_tree_root:
3106 if (!btrfs_test_opt(tree_root, RECOVERY))
3107 goto fail_tree_roots;
3108
3109 free_root_pointers(fs_info, 0);
3110
3111 /* don't use the log in recovery mode, it won't be valid */
3112 btrfs_set_super_log_root(disk_super, 0);
3113
3114 /* we can't trust the free space cache either */
3115 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3116
3117 ret = next_root_backup(fs_info, fs_info->super_copy,
3118 &num_backups_tried, &backup_index);
3119 if (ret == -1)
3120 goto fail_block_groups;
3121 goto retry_root_backup;
3122}
3123
3124static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3125{
3126 if (uptodate) {
3127 set_buffer_uptodate(bh);
3128 } else {
3129 struct btrfs_device *device = (struct btrfs_device *)
3130 bh->b_private;
3131
3132 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3133 "I/O error on %s\n",
3134 rcu_str_deref(device->name));
3135 /* note, we dont' set_buffer_write_io_error because we have
3136 * our own ways of dealing with the IO errors
3137 */
3138 clear_buffer_uptodate(bh);
3139 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3140 }
3141 unlock_buffer(bh);
3142 put_bh(bh);
3143}
3144
3145struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3146{
3147 struct buffer_head *bh;
3148 struct buffer_head *latest = NULL;
3149 struct btrfs_super_block *super;
3150 int i;
3151 u64 transid = 0;
3152 u64 bytenr;
3153
3154 /* we would like to check all the supers, but that would make
3155 * a btrfs mount succeed after a mkfs from a different FS.
3156 * So, we need to add a special mount option to scan for
3157 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3158 */
3159 for (i = 0; i < 1; i++) {
3160 bytenr = btrfs_sb_offset(i);
3161 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3162 i_size_read(bdev->bd_inode))
3163 break;
3164 bh = __bread(bdev, bytenr / 4096,
3165 BTRFS_SUPER_INFO_SIZE);
3166 if (!bh)
3167 continue;
3168
3169 super = (struct btrfs_super_block *)bh->b_data;
3170 if (btrfs_super_bytenr(super) != bytenr ||
3171 btrfs_super_magic(super) != BTRFS_MAGIC) {
3172 brelse(bh);
3173 continue;
3174 }
3175
3176 if (!latest || btrfs_super_generation(super) > transid) {
3177 brelse(latest);
3178 latest = bh;
3179 transid = btrfs_super_generation(super);
3180 } else {
3181 brelse(bh);
3182 }
3183 }
3184 return latest;
3185}
3186
3187/*
3188 * this should be called twice, once with wait == 0 and
3189 * once with wait == 1. When wait == 0 is done, all the buffer heads
3190 * we write are pinned.
3191 *
3192 * They are released when wait == 1 is done.
3193 * max_mirrors must be the same for both runs, and it indicates how
3194 * many supers on this one device should be written.
3195 *
3196 * max_mirrors == 0 means to write them all.
3197 */
3198static int write_dev_supers(struct btrfs_device *device,
3199 struct btrfs_super_block *sb,
3200 int do_barriers, int wait, int max_mirrors)
3201{
3202 struct buffer_head *bh;
3203 int i;
3204 int ret;
3205 int errors = 0;
3206 u32 crc;
3207 u64 bytenr;
3208
3209 if (max_mirrors == 0)
3210 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3211
3212 for (i = 0; i < max_mirrors; i++) {
3213 bytenr = btrfs_sb_offset(i);
3214 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3215 device->commit_total_bytes)
3216 break;
3217
3218 if (wait) {
3219 bh = __find_get_block(device->bdev, bytenr / 4096,
3220 BTRFS_SUPER_INFO_SIZE);
3221 if (!bh) {
3222 errors++;
3223 continue;
3224 }
3225 wait_on_buffer(bh);
3226 if (!buffer_uptodate(bh))
3227 errors++;
3228
3229 /* drop our reference */
3230 brelse(bh);
3231
3232 /* drop the reference from the wait == 0 run */
3233 brelse(bh);
3234 continue;
3235 } else {
3236 btrfs_set_super_bytenr(sb, bytenr);
3237
3238 crc = ~(u32)0;
3239 crc = btrfs_csum_data((char *)sb +
3240 BTRFS_CSUM_SIZE, crc,
3241 BTRFS_SUPER_INFO_SIZE -
3242 BTRFS_CSUM_SIZE);
3243 btrfs_csum_final(crc, sb->csum);
3244
3245 /*
3246 * one reference for us, and we leave it for the
3247 * caller
3248 */
3249 bh = __getblk(device->bdev, bytenr / 4096,
3250 BTRFS_SUPER_INFO_SIZE);
3251 if (!bh) {
3252 printk(KERN_ERR "BTRFS: couldn't get super "
3253 "buffer head for bytenr %Lu\n", bytenr);
3254 errors++;
3255 continue;
3256 }
3257
3258 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3259
3260 /* one reference for submit_bh */
3261 get_bh(bh);
3262
3263 set_buffer_uptodate(bh);
3264 lock_buffer(bh);
3265 bh->b_end_io = btrfs_end_buffer_write_sync;
3266 bh->b_private = device;
3267 }
3268
3269 /*
3270 * we fua the first super. The others we allow
3271 * to go down lazy.
3272 */
3273 if (i == 0)
3274 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3275 else
3276 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3277 if (ret)
3278 errors++;
3279 }
3280 return errors < i ? 0 : -1;
3281}
3282
3283/*
3284 * endio for the write_dev_flush, this will wake anyone waiting
3285 * for the barrier when it is done
3286 */
3287static void btrfs_end_empty_barrier(struct bio *bio, int err)
3288{
3289 if (err) {
3290 if (err == -EOPNOTSUPP)
3291 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3292 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3293 }
3294 if (bio->bi_private)
3295 complete(bio->bi_private);
3296 bio_put(bio);
3297}
3298
3299/*
3300 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3301 * sent down. With wait == 1, it waits for the previous flush.
3302 *
3303 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3304 * capable
3305 */
3306static int write_dev_flush(struct btrfs_device *device, int wait)
3307{
3308 struct bio *bio;
3309 int ret = 0;
3310
3311 if (device->nobarriers)
3312 return 0;
3313
3314 if (wait) {
3315 bio = device->flush_bio;
3316 if (!bio)
3317 return 0;
3318
3319 wait_for_completion(&device->flush_wait);
3320
3321 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3322 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3323 rcu_str_deref(device->name));
3324 device->nobarriers = 1;
3325 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3326 ret = -EIO;
3327 btrfs_dev_stat_inc_and_print(device,
3328 BTRFS_DEV_STAT_FLUSH_ERRS);
3329 }
3330
3331 /* drop the reference from the wait == 0 run */
3332 bio_put(bio);
3333 device->flush_bio = NULL;
3334
3335 return ret;
3336 }
3337
3338 /*
3339 * one reference for us, and we leave it for the
3340 * caller
3341 */
3342 device->flush_bio = NULL;
3343 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3344 if (!bio)
3345 return -ENOMEM;
3346
3347 bio->bi_end_io = btrfs_end_empty_barrier;
3348 bio->bi_bdev = device->bdev;
3349 init_completion(&device->flush_wait);
3350 bio->bi_private = &device->flush_wait;
3351 device->flush_bio = bio;
3352
3353 bio_get(bio);
3354 btrfsic_submit_bio(WRITE_FLUSH, bio);
3355
3356 return 0;
3357}
3358
3359/*
3360 * send an empty flush down to each device in parallel,
3361 * then wait for them
3362 */
3363static int barrier_all_devices(struct btrfs_fs_info *info)
3364{
3365 struct list_head *head;
3366 struct btrfs_device *dev;
3367 int errors_send = 0;
3368 int errors_wait = 0;
3369 int ret;
3370
3371 /* send down all the barriers */
3372 head = &info->fs_devices->devices;
3373 list_for_each_entry_rcu(dev, head, dev_list) {
3374 if (dev->missing)
3375 continue;
3376 if (!dev->bdev) {
3377 errors_send++;
3378 continue;
3379 }
3380 if (!dev->in_fs_metadata || !dev->writeable)
3381 continue;
3382
3383 ret = write_dev_flush(dev, 0);
3384 if (ret)
3385 errors_send++;
3386 }
3387
3388 /* wait for all the barriers */
3389 list_for_each_entry_rcu(dev, head, dev_list) {
3390 if (dev->missing)
3391 continue;
3392 if (!dev->bdev) {
3393 errors_wait++;
3394 continue;
3395 }
3396 if (!dev->in_fs_metadata || !dev->writeable)
3397 continue;
3398
3399 ret = write_dev_flush(dev, 1);
3400 if (ret)
3401 errors_wait++;
3402 }
3403 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3404 errors_wait > info->num_tolerated_disk_barrier_failures)
3405 return -EIO;
3406 return 0;
3407}
3408
3409int btrfs_calc_num_tolerated_disk_barrier_failures(
3410 struct btrfs_fs_info *fs_info)
3411{
3412 struct btrfs_ioctl_space_info space;
3413 struct btrfs_space_info *sinfo;
3414 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3415 BTRFS_BLOCK_GROUP_SYSTEM,
3416 BTRFS_BLOCK_GROUP_METADATA,
3417 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3418 int num_types = 4;
3419 int i;
3420 int c;
3421 int num_tolerated_disk_barrier_failures =
3422 (int)fs_info->fs_devices->num_devices;
3423
3424 for (i = 0; i < num_types; i++) {
3425 struct btrfs_space_info *tmp;
3426
3427 sinfo = NULL;
3428 rcu_read_lock();
3429 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3430 if (tmp->flags == types[i]) {
3431 sinfo = tmp;
3432 break;
3433 }
3434 }
3435 rcu_read_unlock();
3436
3437 if (!sinfo)
3438 continue;
3439
3440 down_read(&sinfo->groups_sem);
3441 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3442 if (!list_empty(&sinfo->block_groups[c])) {
3443 u64 flags;
3444
3445 btrfs_get_block_group_info(
3446 &sinfo->block_groups[c], &space);
3447 if (space.total_bytes == 0 ||
3448 space.used_bytes == 0)
3449 continue;
3450 flags = space.flags;
3451 /*
3452 * return
3453 * 0: if dup, single or RAID0 is configured for
3454 * any of metadata, system or data, else
3455 * 1: if RAID5 is configured, or if RAID1 or
3456 * RAID10 is configured and only two mirrors
3457 * are used, else
3458 * 2: if RAID6 is configured, else
3459 * num_mirrors - 1: if RAID1 or RAID10 is
3460 * configured and more than
3461 * 2 mirrors are used.
3462 */
3463 if (num_tolerated_disk_barrier_failures > 0 &&
3464 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3465 BTRFS_BLOCK_GROUP_RAID0)) ||
3466 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3467 == 0)))
3468 num_tolerated_disk_barrier_failures = 0;
3469 else if (num_tolerated_disk_barrier_failures > 1) {
3470 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3471 BTRFS_BLOCK_GROUP_RAID5 |
3472 BTRFS_BLOCK_GROUP_RAID10)) {
3473 num_tolerated_disk_barrier_failures = 1;
3474 } else if (flags &
3475 BTRFS_BLOCK_GROUP_RAID6) {
3476 num_tolerated_disk_barrier_failures = 2;
3477 }
3478 }
3479 }
3480 }
3481 up_read(&sinfo->groups_sem);
3482 }
3483
3484 return num_tolerated_disk_barrier_failures;
3485}
3486
3487static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3488{
3489 struct list_head *head;
3490 struct btrfs_device *dev;
3491 struct btrfs_super_block *sb;
3492 struct btrfs_dev_item *dev_item;
3493 int ret;
3494 int do_barriers;
3495 int max_errors;
3496 int total_errors = 0;
3497 u64 flags;
3498
3499 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3500 backup_super_roots(root->fs_info);
3501
3502 sb = root->fs_info->super_for_commit;
3503 dev_item = &sb->dev_item;
3504
3505 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3506 head = &root->fs_info->fs_devices->devices;
3507 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3508
3509 if (do_barriers) {
3510 ret = barrier_all_devices(root->fs_info);
3511 if (ret) {
3512 mutex_unlock(
3513 &root->fs_info->fs_devices->device_list_mutex);
3514 btrfs_error(root->fs_info, ret,
3515 "errors while submitting device barriers.");
3516 return ret;
3517 }
3518 }
3519
3520 list_for_each_entry_rcu(dev, head, dev_list) {
3521 if (!dev->bdev) {
3522 total_errors++;
3523 continue;
3524 }
3525 if (!dev->in_fs_metadata || !dev->writeable)
3526 continue;
3527
3528 btrfs_set_stack_device_generation(dev_item, 0);
3529 btrfs_set_stack_device_type(dev_item, dev->type);
3530 btrfs_set_stack_device_id(dev_item, dev->devid);
3531 btrfs_set_stack_device_total_bytes(dev_item,
3532 dev->commit_total_bytes);
3533 btrfs_set_stack_device_bytes_used(dev_item,
3534 dev->commit_bytes_used);
3535 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3536 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3537 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3538 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3539 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3540
3541 flags = btrfs_super_flags(sb);
3542 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3543
3544 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3545 if (ret)
3546 total_errors++;
3547 }
3548 if (total_errors > max_errors) {
3549 btrfs_err(root->fs_info, "%d errors while writing supers",
3550 total_errors);
3551 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3552
3553 /* FUA is masked off if unsupported and can't be the reason */
3554 btrfs_error(root->fs_info, -EIO,
3555 "%d errors while writing supers", total_errors);
3556 return -EIO;
3557 }
3558
3559 total_errors = 0;
3560 list_for_each_entry_rcu(dev, head, dev_list) {
3561 if (!dev->bdev)
3562 continue;
3563 if (!dev->in_fs_metadata || !dev->writeable)
3564 continue;
3565
3566 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3567 if (ret)
3568 total_errors++;
3569 }
3570 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3571 if (total_errors > max_errors) {
3572 btrfs_error(root->fs_info, -EIO,
3573 "%d errors while writing supers", total_errors);
3574 return -EIO;
3575 }
3576 return 0;
3577}
3578
3579int write_ctree_super(struct btrfs_trans_handle *trans,
3580 struct btrfs_root *root, int max_mirrors)
3581{
3582 return write_all_supers(root, max_mirrors);
3583}
3584
3585/* Drop a fs root from the radix tree and free it. */
3586void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3587 struct btrfs_root *root)
3588{
3589 spin_lock(&fs_info->fs_roots_radix_lock);
3590 radix_tree_delete(&fs_info->fs_roots_radix,
3591 (unsigned long)root->root_key.objectid);
3592 spin_unlock(&fs_info->fs_roots_radix_lock);
3593
3594 if (btrfs_root_refs(&root->root_item) == 0)
3595 synchronize_srcu(&fs_info->subvol_srcu);
3596
3597 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3598 btrfs_free_log(NULL, root);
3599
3600 if (root->free_ino_pinned)
3601 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3602 if (root->free_ino_ctl)
3603 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3604 free_fs_root(root);
3605}
3606
3607static void free_fs_root(struct btrfs_root *root)
3608{
3609 iput(root->ino_cache_inode);
3610 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3611 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3612 root->orphan_block_rsv = NULL;
3613 if (root->anon_dev)
3614 free_anon_bdev(root->anon_dev);
3615 if (root->subv_writers)
3616 btrfs_free_subvolume_writers(root->subv_writers);
3617 free_extent_buffer(root->node);
3618 free_extent_buffer(root->commit_root);
3619 kfree(root->free_ino_ctl);
3620 kfree(root->free_ino_pinned);
3621 kfree(root->name);
3622 btrfs_put_fs_root(root);
3623}
3624
3625void btrfs_free_fs_root(struct btrfs_root *root)
3626{
3627 free_fs_root(root);
3628}
3629
3630int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3631{
3632 u64 root_objectid = 0;
3633 struct btrfs_root *gang[8];
3634 int i = 0;
3635 int err = 0;
3636 unsigned int ret = 0;
3637 int index;
3638
3639 while (1) {
3640 index = srcu_read_lock(&fs_info->subvol_srcu);
3641 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3642 (void **)gang, root_objectid,
3643 ARRAY_SIZE(gang));
3644 if (!ret) {
3645 srcu_read_unlock(&fs_info->subvol_srcu, index);
3646 break;
3647 }
3648 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3649
3650 for (i = 0; i < ret; i++) {
3651 /* Avoid to grab roots in dead_roots */
3652 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3653 gang[i] = NULL;
3654 continue;
3655 }
3656 /* grab all the search result for later use */
3657 gang[i] = btrfs_grab_fs_root(gang[i]);
3658 }
3659 srcu_read_unlock(&fs_info->subvol_srcu, index);
3660
3661 for (i = 0; i < ret; i++) {
3662 if (!gang[i])
3663 continue;
3664 root_objectid = gang[i]->root_key.objectid;
3665 err = btrfs_orphan_cleanup(gang[i]);
3666 if (err)
3667 break;
3668 btrfs_put_fs_root(gang[i]);
3669 }
3670 root_objectid++;
3671 }
3672
3673 /* release the uncleaned roots due to error */
3674 for (; i < ret; i++) {
3675 if (gang[i])
3676 btrfs_put_fs_root(gang[i]);
3677 }
3678 return err;
3679}
3680
3681int btrfs_commit_super(struct btrfs_root *root)
3682{
3683 struct btrfs_trans_handle *trans;
3684
3685 mutex_lock(&root->fs_info->cleaner_mutex);
3686 btrfs_run_delayed_iputs(root);
3687 mutex_unlock(&root->fs_info->cleaner_mutex);
3688 wake_up_process(root->fs_info->cleaner_kthread);
3689
3690 /* wait until ongoing cleanup work done */
3691 down_write(&root->fs_info->cleanup_work_sem);
3692 up_write(&root->fs_info->cleanup_work_sem);
3693
3694 trans = btrfs_join_transaction(root);
3695 if (IS_ERR(trans))
3696 return PTR_ERR(trans);
3697 return btrfs_commit_transaction(trans, root);
3698}
3699
3700void close_ctree(struct btrfs_root *root)
3701{
3702 struct btrfs_fs_info *fs_info = root->fs_info;
3703 int ret;
3704
3705 fs_info->closing = 1;
3706 smp_mb();
3707
3708 /* wait for the uuid_scan task to finish */
3709 down(&fs_info->uuid_tree_rescan_sem);
3710 /* avoid complains from lockdep et al., set sem back to initial state */
3711 up(&fs_info->uuid_tree_rescan_sem);
3712
3713 /* pause restriper - we want to resume on mount */
3714 btrfs_pause_balance(fs_info);
3715
3716 btrfs_dev_replace_suspend_for_unmount(fs_info);
3717
3718 btrfs_scrub_cancel(fs_info);
3719
3720 /* wait for any defraggers to finish */
3721 wait_event(fs_info->transaction_wait,
3722 (atomic_read(&fs_info->defrag_running) == 0));
3723
3724 /* clear out the rbtree of defraggable inodes */
3725 btrfs_cleanup_defrag_inodes(fs_info);
3726
3727 cancel_work_sync(&fs_info->async_reclaim_work);
3728
3729 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3730 ret = btrfs_commit_super(root);
3731 if (ret)
3732 btrfs_err(fs_info, "commit super ret %d", ret);
3733 }
3734
3735 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3736 btrfs_error_commit_super(root);
3737
3738 kthread_stop(fs_info->transaction_kthread);
3739 kthread_stop(fs_info->cleaner_kthread);
3740
3741 fs_info->closing = 2;
3742 smp_mb();
3743
3744 btrfs_free_qgroup_config(fs_info);
3745
3746 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3747 btrfs_info(fs_info, "at unmount delalloc count %lld",
3748 percpu_counter_sum(&fs_info->delalloc_bytes));
3749 }
3750
3751 btrfs_sysfs_remove_one(fs_info);
3752 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3753
3754 btrfs_free_fs_roots(fs_info);
3755
3756 btrfs_put_block_group_cache(fs_info);
3757
3758 btrfs_free_block_groups(fs_info);
3759
3760 /*
3761 * we must make sure there is not any read request to
3762 * submit after we stopping all workers.
3763 */
3764 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3765 btrfs_stop_all_workers(fs_info);
3766
3767 fs_info->open = 0;
3768 free_root_pointers(fs_info, 1);
3769
3770 iput(fs_info->btree_inode);
3771
3772#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3773 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3774 btrfsic_unmount(root, fs_info->fs_devices);
3775#endif
3776
3777 btrfs_close_devices(fs_info->fs_devices);
3778 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3779
3780 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3781 percpu_counter_destroy(&fs_info->delalloc_bytes);
3782 percpu_counter_destroy(&fs_info->bio_counter);
3783 bdi_destroy(&fs_info->bdi);
3784 cleanup_srcu_struct(&fs_info->subvol_srcu);
3785
3786 btrfs_free_stripe_hash_table(fs_info);
3787
3788 __btrfs_free_block_rsv(root->orphan_block_rsv);
3789 root->orphan_block_rsv = NULL;
3790
3791 lock_chunks(root);
3792 while (!list_empty(&fs_info->pinned_chunks)) {
3793 struct extent_map *em;
3794
3795 em = list_first_entry(&fs_info->pinned_chunks,
3796 struct extent_map, list);
3797 list_del_init(&em->list);
3798 free_extent_map(em);
3799 }
3800 unlock_chunks(root);
3801}
3802
3803int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3804 int atomic)
3805{
3806 int ret;
3807 struct inode *btree_inode = buf->pages[0]->mapping->host;
3808
3809 ret = extent_buffer_uptodate(buf);
3810 if (!ret)
3811 return ret;
3812
3813 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3814 parent_transid, atomic);
3815 if (ret == -EAGAIN)
3816 return ret;
3817 return !ret;
3818}
3819
3820int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3821{
3822 return set_extent_buffer_uptodate(buf);
3823}
3824
3825void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3826{
3827 struct btrfs_root *root;
3828 u64 transid = btrfs_header_generation(buf);
3829 int was_dirty;
3830
3831#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3832 /*
3833 * This is a fast path so only do this check if we have sanity tests
3834 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3835 * outside of the sanity tests.
3836 */
3837 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3838 return;
3839#endif
3840 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3841 btrfs_assert_tree_locked(buf);
3842 if (transid != root->fs_info->generation)
3843 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3844 "found %llu running %llu\n",
3845 buf->start, transid, root->fs_info->generation);
3846 was_dirty = set_extent_buffer_dirty(buf);
3847 if (!was_dirty)
3848 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3849 buf->len,
3850 root->fs_info->dirty_metadata_batch);
3851#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3852 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3853 btrfs_print_leaf(root, buf);
3854 ASSERT(0);
3855 }
3856#endif
3857}
3858
3859static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3860 int flush_delayed)
3861{
3862 /*
3863 * looks as though older kernels can get into trouble with
3864 * this code, they end up stuck in balance_dirty_pages forever
3865 */
3866 int ret;
3867
3868 if (current->flags & PF_MEMALLOC)
3869 return;
3870
3871 if (flush_delayed)
3872 btrfs_balance_delayed_items(root);
3873
3874 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3875 BTRFS_DIRTY_METADATA_THRESH);
3876 if (ret > 0) {
3877 balance_dirty_pages_ratelimited(
3878 root->fs_info->btree_inode->i_mapping);
3879 }
3880 return;
3881}
3882
3883void btrfs_btree_balance_dirty(struct btrfs_root *root)
3884{
3885 __btrfs_btree_balance_dirty(root, 1);
3886}
3887
3888void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3889{
3890 __btrfs_btree_balance_dirty(root, 0);
3891}
3892
3893int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3894{
3895 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3896 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3897}
3898
3899static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3900 int read_only)
3901{
3902 struct btrfs_super_block *sb = fs_info->super_copy;
3903 int ret = 0;
3904
3905 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
3906 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
3907 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
3908 ret = -EINVAL;
3909 }
3910 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
3911 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
3912 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
3913 ret = -EINVAL;
3914 }
3915 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
3916 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
3917 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
3918 ret = -EINVAL;
3919 }
3920
3921 /*
3922 * The common minimum, we don't know if we can trust the nodesize/sectorsize
3923 * items yet, they'll be verified later. Issue just a warning.
3924 */
3925 if (!IS_ALIGNED(btrfs_super_root(sb), 4096))
3926 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
3927 btrfs_super_root(sb));
3928 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), 4096))
3929 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
3930 btrfs_super_chunk_root(sb));
3931 if (!IS_ALIGNED(btrfs_super_log_root(sb), 4096))
3932 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
3933 btrfs_super_log_root(sb));
3934
3935 /*
3936 * Check the lower bound, the alignment and other constraints are
3937 * checked later.
3938 */
3939 if (btrfs_super_nodesize(sb) < 4096) {
3940 printk(KERN_ERR "BTRFS: nodesize too small: %u < 4096\n",
3941 btrfs_super_nodesize(sb));
3942 ret = -EINVAL;
3943 }
3944 if (btrfs_super_sectorsize(sb) < 4096) {
3945 printk(KERN_ERR "BTRFS: sectorsize too small: %u < 4096\n",
3946 btrfs_super_sectorsize(sb));
3947 ret = -EINVAL;
3948 }
3949
3950 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
3951 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
3952 fs_info->fsid, sb->dev_item.fsid);
3953 ret = -EINVAL;
3954 }
3955
3956 /*
3957 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
3958 * done later
3959 */
3960 if (btrfs_super_num_devices(sb) > (1UL << 31))
3961 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
3962 btrfs_super_num_devices(sb));
3963 if (btrfs_super_num_devices(sb) == 0) {
3964 printk(KERN_ERR "BTRFS: number of devices is 0\n");
3965 ret = -EINVAL;
3966 }
3967
3968 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
3969 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
3970 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
3971 ret = -EINVAL;
3972 }
3973
3974 /*
3975 * Obvious sys_chunk_array corruptions, it must hold at least one key
3976 * and one chunk
3977 */
3978 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
3979 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
3980 btrfs_super_sys_array_size(sb),
3981 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
3982 ret = -EINVAL;
3983 }
3984 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
3985 + sizeof(struct btrfs_chunk)) {
3986 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
3987 btrfs_super_sys_array_size(sb),
3988 sizeof(struct btrfs_disk_key)
3989 + sizeof(struct btrfs_chunk));
3990 ret = -EINVAL;
3991 }
3992
3993 /*
3994 * The generation is a global counter, we'll trust it more than the others
3995 * but it's still possible that it's the one that's wrong.
3996 */
3997 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
3998 printk(KERN_WARNING
3999 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4000 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4001 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4002 && btrfs_super_cache_generation(sb) != (u64)-1)
4003 printk(KERN_WARNING
4004 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4005 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4006
4007 return ret;
4008}
4009
4010static void btrfs_error_commit_super(struct btrfs_root *root)
4011{
4012 mutex_lock(&root->fs_info->cleaner_mutex);
4013 btrfs_run_delayed_iputs(root);
4014 mutex_unlock(&root->fs_info->cleaner_mutex);
4015
4016 down_write(&root->fs_info->cleanup_work_sem);
4017 up_write(&root->fs_info->cleanup_work_sem);
4018
4019 /* cleanup FS via transaction */
4020 btrfs_cleanup_transaction(root);
4021}
4022
4023static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4024{
4025 struct btrfs_ordered_extent *ordered;
4026
4027 spin_lock(&root->ordered_extent_lock);
4028 /*
4029 * This will just short circuit the ordered completion stuff which will
4030 * make sure the ordered extent gets properly cleaned up.
4031 */
4032 list_for_each_entry(ordered, &root->ordered_extents,
4033 root_extent_list)
4034 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4035 spin_unlock(&root->ordered_extent_lock);
4036}
4037
4038static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4039{
4040 struct btrfs_root *root;
4041 struct list_head splice;
4042
4043 INIT_LIST_HEAD(&splice);
4044
4045 spin_lock(&fs_info->ordered_root_lock);
4046 list_splice_init(&fs_info->ordered_roots, &splice);
4047 while (!list_empty(&splice)) {
4048 root = list_first_entry(&splice, struct btrfs_root,
4049 ordered_root);
4050 list_move_tail(&root->ordered_root,
4051 &fs_info->ordered_roots);
4052
4053 spin_unlock(&fs_info->ordered_root_lock);
4054 btrfs_destroy_ordered_extents(root);
4055
4056 cond_resched();
4057 spin_lock(&fs_info->ordered_root_lock);
4058 }
4059 spin_unlock(&fs_info->ordered_root_lock);
4060}
4061
4062static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4063 struct btrfs_root *root)
4064{
4065 struct rb_node *node;
4066 struct btrfs_delayed_ref_root *delayed_refs;
4067 struct btrfs_delayed_ref_node *ref;
4068 int ret = 0;
4069
4070 delayed_refs = &trans->delayed_refs;
4071
4072 spin_lock(&delayed_refs->lock);
4073 if (atomic_read(&delayed_refs->num_entries) == 0) {
4074 spin_unlock(&delayed_refs->lock);
4075 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4076 return ret;
4077 }
4078
4079 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4080 struct btrfs_delayed_ref_head *head;
4081 struct btrfs_delayed_ref_node *tmp;
4082 bool pin_bytes = false;
4083
4084 head = rb_entry(node, struct btrfs_delayed_ref_head,
4085 href_node);
4086 if (!mutex_trylock(&head->mutex)) {
4087 atomic_inc(&head->node.refs);
4088 spin_unlock(&delayed_refs->lock);
4089
4090 mutex_lock(&head->mutex);
4091 mutex_unlock(&head->mutex);
4092 btrfs_put_delayed_ref(&head->node);
4093 spin_lock(&delayed_refs->lock);
4094 continue;
4095 }
4096 spin_lock(&head->lock);
4097 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4098 list) {
4099 ref->in_tree = 0;
4100 list_del(&ref->list);
4101 atomic_dec(&delayed_refs->num_entries);
4102 btrfs_put_delayed_ref(ref);
4103 }
4104 if (head->must_insert_reserved)
4105 pin_bytes = true;
4106 btrfs_free_delayed_extent_op(head->extent_op);
4107 delayed_refs->num_heads--;
4108 if (head->processing == 0)
4109 delayed_refs->num_heads_ready--;
4110 atomic_dec(&delayed_refs->num_entries);
4111 head->node.in_tree = 0;
4112 rb_erase(&head->href_node, &delayed_refs->href_root);
4113 spin_unlock(&head->lock);
4114 spin_unlock(&delayed_refs->lock);
4115 mutex_unlock(&head->mutex);
4116
4117 if (pin_bytes)
4118 btrfs_pin_extent(root, head->node.bytenr,
4119 head->node.num_bytes, 1);
4120 btrfs_put_delayed_ref(&head->node);
4121 cond_resched();
4122 spin_lock(&delayed_refs->lock);
4123 }
4124
4125 spin_unlock(&delayed_refs->lock);
4126
4127 return ret;
4128}
4129
4130static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4131{
4132 struct btrfs_inode *btrfs_inode;
4133 struct list_head splice;
4134
4135 INIT_LIST_HEAD(&splice);
4136
4137 spin_lock(&root->delalloc_lock);
4138 list_splice_init(&root->delalloc_inodes, &splice);
4139
4140 while (!list_empty(&splice)) {
4141 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4142 delalloc_inodes);
4143
4144 list_del_init(&btrfs_inode->delalloc_inodes);
4145 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4146 &btrfs_inode->runtime_flags);
4147 spin_unlock(&root->delalloc_lock);
4148
4149 btrfs_invalidate_inodes(btrfs_inode->root);
4150
4151 spin_lock(&root->delalloc_lock);
4152 }
4153
4154 spin_unlock(&root->delalloc_lock);
4155}
4156
4157static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4158{
4159 struct btrfs_root *root;
4160 struct list_head splice;
4161
4162 INIT_LIST_HEAD(&splice);
4163
4164 spin_lock(&fs_info->delalloc_root_lock);
4165 list_splice_init(&fs_info->delalloc_roots, &splice);
4166 while (!list_empty(&splice)) {
4167 root = list_first_entry(&splice, struct btrfs_root,
4168 delalloc_root);
4169 list_del_init(&root->delalloc_root);
4170 root = btrfs_grab_fs_root(root);
4171 BUG_ON(!root);
4172 spin_unlock(&fs_info->delalloc_root_lock);
4173
4174 btrfs_destroy_delalloc_inodes(root);
4175 btrfs_put_fs_root(root);
4176
4177 spin_lock(&fs_info->delalloc_root_lock);
4178 }
4179 spin_unlock(&fs_info->delalloc_root_lock);
4180}
4181
4182static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4183 struct extent_io_tree *dirty_pages,
4184 int mark)
4185{
4186 int ret;
4187 struct extent_buffer *eb;
4188 u64 start = 0;
4189 u64 end;
4190
4191 while (1) {
4192 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4193 mark, NULL);
4194 if (ret)
4195 break;
4196
4197 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4198 while (start <= end) {
4199 eb = btrfs_find_tree_block(root->fs_info, start);
4200 start += root->nodesize;
4201 if (!eb)
4202 continue;
4203 wait_on_extent_buffer_writeback(eb);
4204
4205 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4206 &eb->bflags))
4207 clear_extent_buffer_dirty(eb);
4208 free_extent_buffer_stale(eb);
4209 }
4210 }
4211
4212 return ret;
4213}
4214
4215static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4216 struct extent_io_tree *pinned_extents)
4217{
4218 struct extent_io_tree *unpin;
4219 u64 start;
4220 u64 end;
4221 int ret;
4222 bool loop = true;
4223
4224 unpin = pinned_extents;
4225again:
4226 while (1) {
4227 ret = find_first_extent_bit(unpin, 0, &start, &end,
4228 EXTENT_DIRTY, NULL);
4229 if (ret)
4230 break;
4231
4232 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4233 btrfs_error_unpin_extent_range(root, start, end);
4234 cond_resched();
4235 }
4236
4237 if (loop) {
4238 if (unpin == &root->fs_info->freed_extents[0])
4239 unpin = &root->fs_info->freed_extents[1];
4240 else
4241 unpin = &root->fs_info->freed_extents[0];
4242 loop = false;
4243 goto again;
4244 }
4245
4246 return 0;
4247}
4248
4249static void btrfs_free_pending_ordered(struct btrfs_transaction *cur_trans,
4250 struct btrfs_fs_info *fs_info)
4251{
4252 struct btrfs_ordered_extent *ordered;
4253
4254 spin_lock(&fs_info->trans_lock);
4255 while (!list_empty(&cur_trans->pending_ordered)) {
4256 ordered = list_first_entry(&cur_trans->pending_ordered,
4257 struct btrfs_ordered_extent,
4258 trans_list);
4259 list_del_init(&ordered->trans_list);
4260 spin_unlock(&fs_info->trans_lock);
4261
4262 btrfs_put_ordered_extent(ordered);
4263 spin_lock(&fs_info->trans_lock);
4264 }
4265 spin_unlock(&fs_info->trans_lock);
4266}
4267
4268void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4269 struct btrfs_root *root)
4270{
4271 btrfs_destroy_delayed_refs(cur_trans, root);
4272
4273 cur_trans->state = TRANS_STATE_COMMIT_START;
4274 wake_up(&root->fs_info->transaction_blocked_wait);
4275
4276 cur_trans->state = TRANS_STATE_UNBLOCKED;
4277 wake_up(&root->fs_info->transaction_wait);
4278
4279 btrfs_free_pending_ordered(cur_trans, root->fs_info);
4280 btrfs_destroy_delayed_inodes(root);
4281 btrfs_assert_delayed_root_empty(root);
4282
4283 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4284 EXTENT_DIRTY);
4285 btrfs_destroy_pinned_extent(root,
4286 root->fs_info->pinned_extents);
4287
4288 cur_trans->state =TRANS_STATE_COMPLETED;
4289 wake_up(&cur_trans->commit_wait);
4290
4291 /*
4292 memset(cur_trans, 0, sizeof(*cur_trans));
4293 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4294 */
4295}
4296
4297static int btrfs_cleanup_transaction(struct btrfs_root *root)
4298{
4299 struct btrfs_transaction *t;
4300
4301 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4302
4303 spin_lock(&root->fs_info->trans_lock);
4304 while (!list_empty(&root->fs_info->trans_list)) {
4305 t = list_first_entry(&root->fs_info->trans_list,
4306 struct btrfs_transaction, list);
4307 if (t->state >= TRANS_STATE_COMMIT_START) {
4308 atomic_inc(&t->use_count);
4309 spin_unlock(&root->fs_info->trans_lock);
4310 btrfs_wait_for_commit(root, t->transid);
4311 btrfs_put_transaction(t);
4312 spin_lock(&root->fs_info->trans_lock);
4313 continue;
4314 }
4315 if (t == root->fs_info->running_transaction) {
4316 t->state = TRANS_STATE_COMMIT_DOING;
4317 spin_unlock(&root->fs_info->trans_lock);
4318 /*
4319 * We wait for 0 num_writers since we don't hold a trans
4320 * handle open currently for this transaction.
4321 */
4322 wait_event(t->writer_wait,
4323 atomic_read(&t->num_writers) == 0);
4324 } else {
4325 spin_unlock(&root->fs_info->trans_lock);
4326 }
4327 btrfs_cleanup_one_transaction(t, root);
4328
4329 spin_lock(&root->fs_info->trans_lock);
4330 if (t == root->fs_info->running_transaction)
4331 root->fs_info->running_transaction = NULL;
4332 list_del_init(&t->list);
4333 spin_unlock(&root->fs_info->trans_lock);
4334
4335 btrfs_put_transaction(t);
4336 trace_btrfs_transaction_commit(root);
4337 spin_lock(&root->fs_info->trans_lock);
4338 }
4339 spin_unlock(&root->fs_info->trans_lock);
4340 btrfs_destroy_all_ordered_extents(root->fs_info);
4341 btrfs_destroy_delayed_inodes(root);
4342 btrfs_assert_delayed_root_empty(root);
4343 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4344 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4345 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4346
4347 return 0;
4348}
4349
4350static const struct extent_io_ops btree_extent_io_ops = {
4351 .readpage_end_io_hook = btree_readpage_end_io_hook,
4352 .readpage_io_failed_hook = btree_io_failed_hook,
4353 .submit_bio_hook = btree_submit_bio_hook,
4354 /* note we're sharing with inode.c for the merge bio hook */
4355 .merge_bio_hook = btrfs_merge_bio_hook,
4356};
Linux 6.x block layer and io_uring tree(s)