projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
btrfs: properly track when rescan worker is running
[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/slab.h>
29#include <linux/migrate.h>
30#include <linux/ratelimit.h>
31#include <linux/uuid.h>
32#include <linux/semaphore.h>
33#include <asm/unaligned.h>
34#include "ctree.h"
35#include "disk-io.h"
36#include "hash.h"
37#include "transaction.h"
38#include "btrfs_inode.h"
39#include "volumes.h"
40#include "print-tree.h"
41#include "locking.h"
42#include "tree-log.h"
43#include "free-space-cache.h"
44#include "free-space-tree.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#include "compression.h"
53
54#ifdef CONFIG_X86
55#include <asm/cpufeature.h>
56#endif
57
58#define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
59 BTRFS_HEADER_FLAG_RELOC |\
60 BTRFS_SUPER_FLAG_ERROR |\
61 BTRFS_SUPER_FLAG_SEEDING |\
62 BTRFS_SUPER_FLAG_METADUMP)
63
64static const struct extent_io_ops btree_extent_io_ops;
65static void end_workqueue_fn(struct btrfs_work *work);
66static void free_fs_root(struct btrfs_root *root);
67static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
68 int read_only);
69static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
70static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
71 struct btrfs_root *root);
72static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
73static int btrfs_destroy_marked_extents(struct btrfs_root *root,
74 struct extent_io_tree *dirty_pages,
75 int mark);
76static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
77 struct extent_io_tree *pinned_extents);
78static int btrfs_cleanup_transaction(struct btrfs_root *root);
79static void btrfs_error_commit_super(struct btrfs_root *root);
80
81/*
82 * btrfs_end_io_wq structs are used to do processing in task context when an IO
83 * is complete. This is used during reads to verify checksums, and it is used
84 * by writes to insert metadata for new file extents after IO is complete.
85 */
86struct btrfs_end_io_wq {
87 struct bio *bio;
88 bio_end_io_t *end_io;
89 void *private;
90 struct btrfs_fs_info *info;
91 int error;
92 enum btrfs_wq_endio_type metadata;
93 struct list_head list;
94 struct btrfs_work work;
95};
96
97static struct kmem_cache *btrfs_end_io_wq_cache;
98
99int __init btrfs_end_io_wq_init(void)
100{
101 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
102 sizeof(struct btrfs_end_io_wq),
103 0,
104 SLAB_MEM_SPREAD,
105 NULL);
106 if (!btrfs_end_io_wq_cache)
107 return -ENOMEM;
108 return 0;
109}
110
111void btrfs_end_io_wq_exit(void)
112{
113 kmem_cache_destroy(btrfs_end_io_wq_cache);
114}
115
116/*
117 * async submit bios are used to offload expensive checksumming
118 * onto the worker threads. They checksum file and metadata bios
119 * just before they are sent down the IO stack.
120 */
121struct async_submit_bio {
122 struct inode *inode;
123 struct bio *bio;
124 struct list_head list;
125 extent_submit_bio_hook_t *submit_bio_start;
126 extent_submit_bio_hook_t *submit_bio_done;
127 int rw;
128 int mirror_num;
129 unsigned long bio_flags;
130 /*
131 * bio_offset is optional, can be used if the pages in the bio
132 * can't tell us where in the file the bio should go
133 */
134 u64 bio_offset;
135 struct btrfs_work work;
136 int error;
137};
138
139/*
140 * Lockdep class keys for extent_buffer->lock's in this root. For a given
141 * eb, the lockdep key is determined by the btrfs_root it belongs to and
142 * the level the eb occupies in the tree.
143 *
144 * Different roots are used for different purposes and may nest inside each
145 * other and they require separate keysets. As lockdep keys should be
146 * static, assign keysets according to the purpose of the root as indicated
147 * by btrfs_root->objectid. This ensures that all special purpose roots
148 * have separate keysets.
149 *
150 * Lock-nesting across peer nodes is always done with the immediate parent
151 * node locked thus preventing deadlock. As lockdep doesn't know this, use
152 * subclass to avoid triggering lockdep warning in such cases.
153 *
154 * The key is set by the readpage_end_io_hook after the buffer has passed
155 * csum validation but before the pages are unlocked. It is also set by
156 * btrfs_init_new_buffer on freshly allocated blocks.
157 *
158 * We also add a check to make sure the highest level of the tree is the
159 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
160 * needs update as well.
161 */
162#ifdef CONFIG_DEBUG_LOCK_ALLOC
163# if BTRFS_MAX_LEVEL != 8
164# error
165# endif
166
167static struct btrfs_lockdep_keyset {
168 u64 id; /* root objectid */
169 const char *name_stem; /* lock name stem */
170 char names[BTRFS_MAX_LEVEL + 1][20];
171 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
172} btrfs_lockdep_keysets[] = {
173 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
174 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
175 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
176 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
177 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
178 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
179 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
180 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
181 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
182 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
183 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
184 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
185 { .id = 0, .name_stem = "tree" },
186};
187
188void __init btrfs_init_lockdep(void)
189{
190 int i, j;
191
192 /* initialize lockdep class names */
193 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
194 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
195
196 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
197 snprintf(ks->names[j], sizeof(ks->names[j]),
198 "btrfs-%s-%02d", ks->name_stem, j);
199 }
200}
201
202void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
203 int level)
204{
205 struct btrfs_lockdep_keyset *ks;
206
207 BUG_ON(level >= ARRAY_SIZE(ks->keys));
208
209 /* find the matching keyset, id 0 is the default entry */
210 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
211 if (ks->id == objectid)
212 break;
213
214 lockdep_set_class_and_name(&eb->lock,
215 &ks->keys[level], ks->names[level]);
216}
217
218#endif
219
220/*
221 * extents on the btree inode are pretty simple, there's one extent
222 * that covers the entire device
223 */
224static struct extent_map *btree_get_extent(struct inode *inode,
225 struct page *page, size_t pg_offset, u64 start, u64 len,
226 int create)
227{
228 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
229 struct extent_map *em;
230 int ret;
231
232 read_lock(&em_tree->lock);
233 em = lookup_extent_mapping(em_tree, start, len);
234 if (em) {
235 em->bdev =
236 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
237 read_unlock(&em_tree->lock);
238 goto out;
239 }
240 read_unlock(&em_tree->lock);
241
242 em = alloc_extent_map();
243 if (!em) {
244 em = ERR_PTR(-ENOMEM);
245 goto out;
246 }
247 em->start = 0;
248 em->len = (u64)-1;
249 em->block_len = (u64)-1;
250 em->block_start = 0;
251 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
252
253 write_lock(&em_tree->lock);
254 ret = add_extent_mapping(em_tree, em, 0);
255 if (ret == -EEXIST) {
256 free_extent_map(em);
257 em = lookup_extent_mapping(em_tree, start, len);
258 if (!em)
259 em = ERR_PTR(-EIO);
260 } else if (ret) {
261 free_extent_map(em);
262 em = ERR_PTR(ret);
263 }
264 write_unlock(&em_tree->lock);
265
266out:
267 return em;
268}
269
270u32 btrfs_csum_data(char *data, u32 seed, size_t len)
271{
272 return btrfs_crc32c(seed, data, len);
273}
274
275void btrfs_csum_final(u32 crc, char *result)
276{
277 put_unaligned_le32(~crc, result);
278}
279
280/*
281 * compute the csum for a btree block, and either verify it or write it
282 * into the csum field of the block.
283 */
284static int csum_tree_block(struct btrfs_fs_info *fs_info,
285 struct extent_buffer *buf,
286 int verify)
287{
288 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
289 char *result = NULL;
290 unsigned long len;
291 unsigned long cur_len;
292 unsigned long offset = BTRFS_CSUM_SIZE;
293 char *kaddr;
294 unsigned long map_start;
295 unsigned long map_len;
296 int err;
297 u32 crc = ~(u32)0;
298 unsigned long inline_result;
299
300 len = buf->len - offset;
301 while (len > 0) {
302 err = map_private_extent_buffer(buf, offset, 32,
303 &kaddr, &map_start, &map_len);
304 if (err)
305 return err;
306 cur_len = min(len, map_len - (offset - map_start));
307 crc = btrfs_csum_data(kaddr + offset - map_start,
308 crc, cur_len);
309 len -= cur_len;
310 offset += cur_len;
311 }
312 if (csum_size > sizeof(inline_result)) {
313 result = kzalloc(csum_size, GFP_NOFS);
314 if (!result)
315 return -ENOMEM;
316 } else {
317 result = (char *)&inline_result;
318 }
319
320 btrfs_csum_final(crc, result);
321
322 if (verify) {
323 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
324 u32 val;
325 u32 found = 0;
326 memcpy(&found, result, csum_size);
327
328 read_extent_buffer(buf, &val, 0, csum_size);
329 btrfs_warn_rl(fs_info,
330 "%s checksum verify failed on %llu wanted %X found %X "
331 "level %d",
332 fs_info->sb->s_id, buf->start,
333 val, found, btrfs_header_level(buf));
334 if (result != (char *)&inline_result)
335 kfree(result);
336 return -EUCLEAN;
337 }
338 } else {
339 write_extent_buffer(buf, result, 0, csum_size);
340 }
341 if (result != (char *)&inline_result)
342 kfree(result);
343 return 0;
344}
345
346/*
347 * we can't consider a given block up to date unless the transid of the
348 * block matches the transid in the parent node's pointer. This is how we
349 * detect blocks that either didn't get written at all or got written
350 * in the wrong place.
351 */
352static int verify_parent_transid(struct extent_io_tree *io_tree,
353 struct extent_buffer *eb, u64 parent_transid,
354 int atomic)
355{
356 struct extent_state *cached_state = NULL;
357 int ret;
358 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
359
360 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
361 return 0;
362
363 if (atomic)
364 return -EAGAIN;
365
366 if (need_lock) {
367 btrfs_tree_read_lock(eb);
368 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
369 }
370
371 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
372 &cached_state);
373 if (extent_buffer_uptodate(eb) &&
374 btrfs_header_generation(eb) == parent_transid) {
375 ret = 0;
376 goto out;
377 }
378 btrfs_err_rl(eb->fs_info,
379 "parent transid verify failed on %llu wanted %llu found %llu",
380 eb->start,
381 parent_transid, btrfs_header_generation(eb));
382 ret = 1;
383
384 /*
385 * Things reading via commit roots that don't have normal protection,
386 * like send, can have a really old block in cache that may point at a
387 * block that has been freed and re-allocated. So don't clear uptodate
388 * if we find an eb that is under IO (dirty/writeback) because we could
389 * end up reading in the stale data and then writing it back out and
390 * making everybody very sad.
391 */
392 if (!extent_buffer_under_io(eb))
393 clear_extent_buffer_uptodate(eb);
394out:
395 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
396 &cached_state, GFP_NOFS);
397 if (need_lock)
398 btrfs_tree_read_unlock_blocking(eb);
399 return ret;
400}
401
402/*
403 * Return 0 if the superblock checksum type matches the checksum value of that
404 * algorithm. Pass the raw disk superblock data.
405 */
406static int btrfs_check_super_csum(char *raw_disk_sb)
407{
408 struct btrfs_super_block *disk_sb =
409 (struct btrfs_super_block *)raw_disk_sb;
410 u16 csum_type = btrfs_super_csum_type(disk_sb);
411 int ret = 0;
412
413 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
414 u32 crc = ~(u32)0;
415 const int csum_size = sizeof(crc);
416 char result[csum_size];
417
418 /*
419 * The super_block structure does not span the whole
420 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
421 * is filled with zeros and is included in the checksum.
422 */
423 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
424 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
425 btrfs_csum_final(crc, result);
426
427 if (memcmp(raw_disk_sb, result, csum_size))
428 ret = 1;
429 }
430
431 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
432 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
433 csum_type);
434 ret = 1;
435 }
436
437 return ret;
438}
439
440/*
441 * helper to read a given tree block, doing retries as required when
442 * the checksums don't match and we have alternate mirrors to try.
443 */
444static int btree_read_extent_buffer_pages(struct btrfs_root *root,
445 struct extent_buffer *eb,
446 u64 start, u64 parent_transid)
447{
448 struct extent_io_tree *io_tree;
449 int failed = 0;
450 int ret;
451 int num_copies = 0;
452 int mirror_num = 0;
453 int failed_mirror = 0;
454
455 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
456 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
457 while (1) {
458 ret = read_extent_buffer_pages(io_tree, eb, start,
459 WAIT_COMPLETE,
460 btree_get_extent, mirror_num);
461 if (!ret) {
462 if (!verify_parent_transid(io_tree, eb,
463 parent_transid, 0))
464 break;
465 else
466 ret = -EIO;
467 }
468
469 /*
470 * This buffer's crc is fine, but its contents are corrupted, so
471 * there is no reason to read the other copies, they won't be
472 * any less wrong.
473 */
474 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
475 break;
476
477 num_copies = btrfs_num_copies(root->fs_info,
478 eb->start, eb->len);
479 if (num_copies == 1)
480 break;
481
482 if (!failed_mirror) {
483 failed = 1;
484 failed_mirror = eb->read_mirror;
485 }
486
487 mirror_num++;
488 if (mirror_num == failed_mirror)
489 mirror_num++;
490
491 if (mirror_num > num_copies)
492 break;
493 }
494
495 if (failed && !ret && failed_mirror)
496 repair_eb_io_failure(root, eb, failed_mirror);
497
498 return ret;
499}
500
501/*
502 * checksum a dirty tree block before IO. This has extra checks to make sure
503 * we only fill in the checksum field in the first page of a multi-page block
504 */
505
506static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
507{
508 u64 start = page_offset(page);
509 u64 found_start;
510 struct extent_buffer *eb;
511
512 eb = (struct extent_buffer *)page->private;
513 if (page != eb->pages[0])
514 return 0;
515
516 found_start = btrfs_header_bytenr(eb);
517 /*
518 * Please do not consolidate these warnings into a single if.
519 * It is useful to know what went wrong.
520 */
521 if (WARN_ON(found_start != start))
522 return -EUCLEAN;
523 if (WARN_ON(!PageUptodate(page)))
524 return -EUCLEAN;
525
526 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
527 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
528
529 return csum_tree_block(fs_info, eb, 0);
530}
531
532static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
533 struct extent_buffer *eb)
534{
535 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
536 u8 fsid[BTRFS_UUID_SIZE];
537 int ret = 1;
538
539 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
540 while (fs_devices) {
541 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
542 ret = 0;
543 break;
544 }
545 fs_devices = fs_devices->seed;
546 }
547 return ret;
548}
549
550#define CORRUPT(reason, eb, root, slot) \
551 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
552 "root=%llu, slot=%d", reason, \
553 btrfs_header_bytenr(eb), root->objectid, slot)
554
555static noinline int check_leaf(struct btrfs_root *root,
556 struct extent_buffer *leaf)
557{
558 struct btrfs_key key;
559 struct btrfs_key leaf_key;
560 u32 nritems = btrfs_header_nritems(leaf);
561 int slot;
562
563 if (nritems == 0)
564 return 0;
565
566 /* Check the 0 item */
567 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
568 BTRFS_LEAF_DATA_SIZE(root)) {
569 CORRUPT("invalid item offset size pair", leaf, root, 0);
570 return -EIO;
571 }
572
573 /*
574 * Check to make sure each items keys are in the correct order and their
575 * offsets make sense. We only have to loop through nritems-1 because
576 * we check the current slot against the next slot, which verifies the
577 * next slot's offset+size makes sense and that the current's slot
578 * offset is correct.
579 */
580 for (slot = 0; slot < nritems - 1; slot++) {
581 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
582 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
583
584 /* Make sure the keys are in the right order */
585 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
586 CORRUPT("bad key order", leaf, root, slot);
587 return -EIO;
588 }
589
590 /*
591 * Make sure the offset and ends are right, remember that the
592 * item data starts at the end of the leaf and grows towards the
593 * front.
594 */
595 if (btrfs_item_offset_nr(leaf, slot) !=
596 btrfs_item_end_nr(leaf, slot + 1)) {
597 CORRUPT("slot offset bad", leaf, root, slot);
598 return -EIO;
599 }
600
601 /*
602 * Check to make sure that we don't point outside of the leaf,
603 * just in case all the items are consistent to each other, but
604 * all point outside of the leaf.
605 */
606 if (btrfs_item_end_nr(leaf, slot) >
607 BTRFS_LEAF_DATA_SIZE(root)) {
608 CORRUPT("slot end outside of leaf", leaf, root, slot);
609 return -EIO;
610 }
611 }
612
613 return 0;
614}
615
616static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
617 u64 phy_offset, struct page *page,
618 u64 start, u64 end, int mirror)
619{
620 u64 found_start;
621 int found_level;
622 struct extent_buffer *eb;
623 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
624 struct btrfs_fs_info *fs_info = root->fs_info;
625 int ret = 0;
626 int reads_done;
627
628 if (!page->private)
629 goto out;
630
631 eb = (struct extent_buffer *)page->private;
632
633 /* the pending IO might have been the only thing that kept this buffer
634 * in memory. Make sure we have a ref for all this other checks
635 */
636 extent_buffer_get(eb);
637
638 reads_done = atomic_dec_and_test(&eb->io_pages);
639 if (!reads_done)
640 goto err;
641
642 eb->read_mirror = mirror;
643 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
644 ret = -EIO;
645 goto err;
646 }
647
648 found_start = btrfs_header_bytenr(eb);
649 if (found_start != eb->start) {
650 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
651 found_start, eb->start);
652 ret = -EIO;
653 goto err;
654 }
655 if (check_tree_block_fsid(fs_info, eb)) {
656 btrfs_err_rl(fs_info, "bad fsid on block %llu",
657 eb->start);
658 ret = -EIO;
659 goto err;
660 }
661 found_level = btrfs_header_level(eb);
662 if (found_level >= BTRFS_MAX_LEVEL) {
663 btrfs_err(fs_info, "bad tree block level %d",
664 (int)btrfs_header_level(eb));
665 ret = -EIO;
666 goto err;
667 }
668
669 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
670 eb, found_level);
671
672 ret = csum_tree_block(fs_info, eb, 1);
673 if (ret)
674 goto err;
675
676 /*
677 * If this is a leaf block and it is corrupt, set the corrupt bit so
678 * that we don't try and read the other copies of this block, just
679 * return -EIO.
680 */
681 if (found_level == 0 && check_leaf(root, eb)) {
682 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
683 ret = -EIO;
684 }
685
686 if (!ret)
687 set_extent_buffer_uptodate(eb);
688err:
689 if (reads_done &&
690 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
691 btree_readahead_hook(fs_info, eb, eb->start, ret);
692
693 if (ret) {
694 /*
695 * our io error hook is going to dec the io pages
696 * again, we have to make sure it has something
697 * to decrement
698 */
699 atomic_inc(&eb->io_pages);
700 clear_extent_buffer_uptodate(eb);
701 }
702 free_extent_buffer(eb);
703out:
704 return ret;
705}
706
707static int btree_io_failed_hook(struct page *page, int failed_mirror)
708{
709 struct extent_buffer *eb;
710
711 eb = (struct extent_buffer *)page->private;
712 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
713 eb->read_mirror = failed_mirror;
714 atomic_dec(&eb->io_pages);
715 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
716 btree_readahead_hook(eb->fs_info, eb, eb->start, -EIO);
717 return -EIO; /* we fixed nothing */
718}
719
720static void end_workqueue_bio(struct bio *bio)
721{
722 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
723 struct btrfs_fs_info *fs_info;
724 struct btrfs_workqueue *wq;
725 btrfs_work_func_t func;
726
727 fs_info = end_io_wq->info;
728 end_io_wq->error = bio->bi_error;
729
730 if (bio->bi_rw & REQ_WRITE) {
731 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
732 wq = fs_info->endio_meta_write_workers;
733 func = btrfs_endio_meta_write_helper;
734 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
735 wq = fs_info->endio_freespace_worker;
736 func = btrfs_freespace_write_helper;
737 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
738 wq = fs_info->endio_raid56_workers;
739 func = btrfs_endio_raid56_helper;
740 } else {
741 wq = fs_info->endio_write_workers;
742 func = btrfs_endio_write_helper;
743 }
744 } else {
745 if (unlikely(end_io_wq->metadata ==
746 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
747 wq = fs_info->endio_repair_workers;
748 func = btrfs_endio_repair_helper;
749 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
750 wq = fs_info->endio_raid56_workers;
751 func = btrfs_endio_raid56_helper;
752 } else if (end_io_wq->metadata) {
753 wq = fs_info->endio_meta_workers;
754 func = btrfs_endio_meta_helper;
755 } else {
756 wq = fs_info->endio_workers;
757 func = btrfs_endio_helper;
758 }
759 }
760
761 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
762 btrfs_queue_work(wq, &end_io_wq->work);
763}
764
765int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
766 enum btrfs_wq_endio_type metadata)
767{
768 struct btrfs_end_io_wq *end_io_wq;
769
770 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
771 if (!end_io_wq)
772 return -ENOMEM;
773
774 end_io_wq->private = bio->bi_private;
775 end_io_wq->end_io = bio->bi_end_io;
776 end_io_wq->info = info;
777 end_io_wq->error = 0;
778 end_io_wq->bio = bio;
779 end_io_wq->metadata = metadata;
780
781 bio->bi_private = end_io_wq;
782 bio->bi_end_io = end_workqueue_bio;
783 return 0;
784}
785
786unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
787{
788 unsigned long limit = min_t(unsigned long,
789 info->thread_pool_size,
790 info->fs_devices->open_devices);
791 return 256 * limit;
792}
793
794static void run_one_async_start(struct btrfs_work *work)
795{
796 struct async_submit_bio *async;
797 int ret;
798
799 async = container_of(work, struct async_submit_bio, work);
800 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
801 async->mirror_num, async->bio_flags,
802 async->bio_offset);
803 if (ret)
804 async->error = ret;
805}
806
807static void run_one_async_done(struct btrfs_work *work)
808{
809 struct btrfs_fs_info *fs_info;
810 struct async_submit_bio *async;
811 int limit;
812
813 async = container_of(work, struct async_submit_bio, work);
814 fs_info = BTRFS_I(async->inode)->root->fs_info;
815
816 limit = btrfs_async_submit_limit(fs_info);
817 limit = limit * 2 / 3;
818
819 /*
820 * atomic_dec_return implies a barrier for waitqueue_active
821 */
822 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
823 waitqueue_active(&fs_info->async_submit_wait))
824 wake_up(&fs_info->async_submit_wait);
825
826 /* If an error occurred we just want to clean up the bio and move on */
827 if (async->error) {
828 async->bio->bi_error = async->error;
829 bio_endio(async->bio);
830 return;
831 }
832
833 async->submit_bio_done(async->inode, async->rw, async->bio,
834 async->mirror_num, async->bio_flags,
835 async->bio_offset);
836}
837
838static void run_one_async_free(struct btrfs_work *work)
839{
840 struct async_submit_bio *async;
841
842 async = container_of(work, struct async_submit_bio, work);
843 kfree(async);
844}
845
846int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
847 int rw, struct bio *bio, int mirror_num,
848 unsigned long bio_flags,
849 u64 bio_offset,
850 extent_submit_bio_hook_t *submit_bio_start,
851 extent_submit_bio_hook_t *submit_bio_done)
852{
853 struct async_submit_bio *async;
854
855 async = kmalloc(sizeof(*async), GFP_NOFS);
856 if (!async)
857 return -ENOMEM;
858
859 async->inode = inode;
860 async->rw = rw;
861 async->bio = bio;
862 async->mirror_num = mirror_num;
863 async->submit_bio_start = submit_bio_start;
864 async->submit_bio_done = submit_bio_done;
865
866 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
867 run_one_async_done, run_one_async_free);
868
869 async->bio_flags = bio_flags;
870 async->bio_offset = bio_offset;
871
872 async->error = 0;
873
874 atomic_inc(&fs_info->nr_async_submits);
875
876 if (rw & REQ_SYNC)
877 btrfs_set_work_high_priority(&async->work);
878
879 btrfs_queue_work(fs_info->workers, &async->work);
880
881 while (atomic_read(&fs_info->async_submit_draining) &&
882 atomic_read(&fs_info->nr_async_submits)) {
883 wait_event(fs_info->async_submit_wait,
884 (atomic_read(&fs_info->nr_async_submits) == 0));
885 }
886
887 return 0;
888}
889
890static int btree_csum_one_bio(struct bio *bio)
891{
892 struct bio_vec *bvec;
893 struct btrfs_root *root;
894 int i, ret = 0;
895
896 bio_for_each_segment_all(bvec, bio, i) {
897 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
898 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
899 if (ret)
900 break;
901 }
902
903 return ret;
904}
905
906static int __btree_submit_bio_start(struct inode *inode, int rw,
907 struct bio *bio, int mirror_num,
908 unsigned long bio_flags,
909 u64 bio_offset)
910{
911 /*
912 * when we're called for a write, we're already in the async
913 * submission context. Just jump into btrfs_map_bio
914 */
915 return btree_csum_one_bio(bio);
916}
917
918static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
919 int mirror_num, unsigned long bio_flags,
920 u64 bio_offset)
921{
922 int ret;
923
924 /*
925 * when we're called for a write, we're already in the async
926 * submission context. Just jump into btrfs_map_bio
927 */
928 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
929 if (ret) {
930 bio->bi_error = ret;
931 bio_endio(bio);
932 }
933 return ret;
934}
935
936static int check_async_write(struct inode *inode, unsigned long bio_flags)
937{
938 if (bio_flags & EXTENT_BIO_TREE_LOG)
939 return 0;
940#ifdef CONFIG_X86
941 if (static_cpu_has(X86_FEATURE_XMM4_2))
942 return 0;
943#endif
944 return 1;
945}
946
947static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
948 int mirror_num, unsigned long bio_flags,
949 u64 bio_offset)
950{
951 int async = check_async_write(inode, bio_flags);
952 int ret;
953
954 if (!(rw & REQ_WRITE)) {
955 /*
956 * called for a read, do the setup so that checksum validation
957 * can happen in the async kernel threads
958 */
959 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
960 bio, BTRFS_WQ_ENDIO_METADATA);
961 if (ret)
962 goto out_w_error;
963 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
964 mirror_num, 0);
965 } else if (!async) {
966 ret = btree_csum_one_bio(bio);
967 if (ret)
968 goto out_w_error;
969 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
970 mirror_num, 0);
971 } else {
972 /*
973 * kthread helpers are used to submit writes so that
974 * checksumming can happen in parallel across all CPUs
975 */
976 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
977 inode, rw, bio, mirror_num, 0,
978 bio_offset,
979 __btree_submit_bio_start,
980 __btree_submit_bio_done);
981 }
982
983 if (ret)
984 goto out_w_error;
985 return 0;
986
987out_w_error:
988 bio->bi_error = ret;
989 bio_endio(bio);
990 return ret;
991}
992
993#ifdef CONFIG_MIGRATION
994static int btree_migratepage(struct address_space *mapping,
995 struct page *newpage, struct page *page,
996 enum migrate_mode mode)
997{
998 /*
999 * we can't safely write a btree page from here,
1000 * we haven't done the locking hook
1001 */
1002 if (PageDirty(page))
1003 return -EAGAIN;
1004 /*
1005 * Buffers may be managed in a filesystem specific way.
1006 * We must have no buffers or drop them.
1007 */
1008 if (page_has_private(page) &&
1009 !try_to_release_page(page, GFP_KERNEL))
1010 return -EAGAIN;
1011 return migrate_page(mapping, newpage, page, mode);
1012}
1013#endif
1014
1015
1016static int btree_writepages(struct address_space *mapping,
1017 struct writeback_control *wbc)
1018{
1019 struct btrfs_fs_info *fs_info;
1020 int ret;
1021
1022 if (wbc->sync_mode == WB_SYNC_NONE) {
1023
1024 if (wbc->for_kupdate)
1025 return 0;
1026
1027 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1028 /* this is a bit racy, but that's ok */
1029 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1030 BTRFS_DIRTY_METADATA_THRESH);
1031 if (ret < 0)
1032 return 0;
1033 }
1034 return btree_write_cache_pages(mapping, wbc);
1035}
1036
1037static int btree_readpage(struct file *file, struct page *page)
1038{
1039 struct extent_io_tree *tree;
1040 tree = &BTRFS_I(page->mapping->host)->io_tree;
1041 return extent_read_full_page(tree, page, btree_get_extent, 0);
1042}
1043
1044static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1045{
1046 if (PageWriteback(page) || PageDirty(page))
1047 return 0;
1048
1049 return try_release_extent_buffer(page);
1050}
1051
1052static void btree_invalidatepage(struct page *page, unsigned int offset,
1053 unsigned int length)
1054{
1055 struct extent_io_tree *tree;
1056 tree = &BTRFS_I(page->mapping->host)->io_tree;
1057 extent_invalidatepage(tree, page, offset);
1058 btree_releasepage(page, GFP_NOFS);
1059 if (PagePrivate(page)) {
1060 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1061 "page private not zero on page %llu",
1062 (unsigned long long)page_offset(page));
1063 ClearPagePrivate(page);
1064 set_page_private(page, 0);
1065 put_page(page);
1066 }
1067}
1068
1069static int btree_set_page_dirty(struct page *page)
1070{
1071#ifdef DEBUG
1072 struct extent_buffer *eb;
1073
1074 BUG_ON(!PagePrivate(page));
1075 eb = (struct extent_buffer *)page->private;
1076 BUG_ON(!eb);
1077 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1078 BUG_ON(!atomic_read(&eb->refs));
1079 btrfs_assert_tree_locked(eb);
1080#endif
1081 return __set_page_dirty_nobuffers(page);
1082}
1083
1084static const struct address_space_operations btree_aops = {
1085 .readpage = btree_readpage,
1086 .writepages = btree_writepages,
1087 .releasepage = btree_releasepage,
1088 .invalidatepage = btree_invalidatepage,
1089#ifdef CONFIG_MIGRATION
1090 .migratepage = btree_migratepage,
1091#endif
1092 .set_page_dirty = btree_set_page_dirty,
1093};
1094
1095void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1096{
1097 struct extent_buffer *buf = NULL;
1098 struct inode *btree_inode = root->fs_info->btree_inode;
1099
1100 buf = btrfs_find_create_tree_block(root, bytenr);
1101 if (IS_ERR(buf))
1102 return;
1103 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1104 buf, 0, WAIT_NONE, btree_get_extent, 0);
1105 free_extent_buffer(buf);
1106}
1107
1108int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1109 int mirror_num, struct extent_buffer **eb)
1110{
1111 struct extent_buffer *buf = NULL;
1112 struct inode *btree_inode = root->fs_info->btree_inode;
1113 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1114 int ret;
1115
1116 buf = btrfs_find_create_tree_block(root, bytenr);
1117 if (IS_ERR(buf))
1118 return 0;
1119
1120 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1121
1122 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1123 btree_get_extent, mirror_num);
1124 if (ret) {
1125 free_extent_buffer(buf);
1126 return ret;
1127 }
1128
1129 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1130 free_extent_buffer(buf);
1131 return -EIO;
1132 } else if (extent_buffer_uptodate(buf)) {
1133 *eb = buf;
1134 } else {
1135 free_extent_buffer(buf);
1136 }
1137 return 0;
1138}
1139
1140struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1141 u64 bytenr)
1142{
1143 return find_extent_buffer(fs_info, bytenr);
1144}
1145
1146struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1147 u64 bytenr)
1148{
1149 if (btrfs_is_testing(root->fs_info))
1150 return alloc_test_extent_buffer(root->fs_info, bytenr,
1151 root->nodesize);
1152 return alloc_extent_buffer(root->fs_info, bytenr);
1153}
1154
1155
1156int btrfs_write_tree_block(struct extent_buffer *buf)
1157{
1158 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1159 buf->start + buf->len - 1);
1160}
1161
1162int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1163{
1164 return filemap_fdatawait_range(buf->pages[0]->mapping,
1165 buf->start, buf->start + buf->len - 1);
1166}
1167
1168struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1169 u64 parent_transid)
1170{
1171 struct extent_buffer *buf = NULL;
1172 int ret;
1173
1174 buf = btrfs_find_create_tree_block(root, bytenr);
1175 if (IS_ERR(buf))
1176 return buf;
1177
1178 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1179 if (ret) {
1180 free_extent_buffer(buf);
1181 return ERR_PTR(ret);
1182 }
1183 return buf;
1184
1185}
1186
1187void clean_tree_block(struct btrfs_trans_handle *trans,
1188 struct btrfs_fs_info *fs_info,
1189 struct extent_buffer *buf)
1190{
1191 if (btrfs_header_generation(buf) ==
1192 fs_info->running_transaction->transid) {
1193 btrfs_assert_tree_locked(buf);
1194
1195 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1196 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1197 -buf->len,
1198 fs_info->dirty_metadata_batch);
1199 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1200 btrfs_set_lock_blocking(buf);
1201 clear_extent_buffer_dirty(buf);
1202 }
1203 }
1204}
1205
1206static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1207{
1208 struct btrfs_subvolume_writers *writers;
1209 int ret;
1210
1211 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1212 if (!writers)
1213 return ERR_PTR(-ENOMEM);
1214
1215 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1216 if (ret < 0) {
1217 kfree(writers);
1218 return ERR_PTR(ret);
1219 }
1220
1221 init_waitqueue_head(&writers->wait);
1222 return writers;
1223}
1224
1225static void
1226btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1227{
1228 percpu_counter_destroy(&writers->counter);
1229 kfree(writers);
1230}
1231
1232static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1233 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1234 u64 objectid)
1235{
1236 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1237 root->node = NULL;
1238 root->commit_root = NULL;
1239 root->sectorsize = sectorsize;
1240 root->nodesize = nodesize;
1241 root->stripesize = stripesize;
1242 root->state = 0;
1243 root->orphan_cleanup_state = 0;
1244
1245 root->objectid = objectid;
1246 root->last_trans = 0;
1247 root->highest_objectid = 0;
1248 root->nr_delalloc_inodes = 0;
1249 root->nr_ordered_extents = 0;
1250 root->name = NULL;
1251 root->inode_tree = RB_ROOT;
1252 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1253 root->block_rsv = NULL;
1254 root->orphan_block_rsv = NULL;
1255
1256 INIT_LIST_HEAD(&root->dirty_list);
1257 INIT_LIST_HEAD(&root->root_list);
1258 INIT_LIST_HEAD(&root->delalloc_inodes);
1259 INIT_LIST_HEAD(&root->delalloc_root);
1260 INIT_LIST_HEAD(&root->ordered_extents);
1261 INIT_LIST_HEAD(&root->ordered_root);
1262 INIT_LIST_HEAD(&root->logged_list[0]);
1263 INIT_LIST_HEAD(&root->logged_list[1]);
1264 spin_lock_init(&root->orphan_lock);
1265 spin_lock_init(&root->inode_lock);
1266 spin_lock_init(&root->delalloc_lock);
1267 spin_lock_init(&root->ordered_extent_lock);
1268 spin_lock_init(&root->accounting_lock);
1269 spin_lock_init(&root->log_extents_lock[0]);
1270 spin_lock_init(&root->log_extents_lock[1]);
1271 mutex_init(&root->objectid_mutex);
1272 mutex_init(&root->log_mutex);
1273 mutex_init(&root->ordered_extent_mutex);
1274 mutex_init(&root->delalloc_mutex);
1275 init_waitqueue_head(&root->log_writer_wait);
1276 init_waitqueue_head(&root->log_commit_wait[0]);
1277 init_waitqueue_head(&root->log_commit_wait[1]);
1278 INIT_LIST_HEAD(&root->log_ctxs[0]);
1279 INIT_LIST_HEAD(&root->log_ctxs[1]);
1280 atomic_set(&root->log_commit[0], 0);
1281 atomic_set(&root->log_commit[1], 0);
1282 atomic_set(&root->log_writers, 0);
1283 atomic_set(&root->log_batch, 0);
1284 atomic_set(&root->orphan_inodes, 0);
1285 atomic_set(&root->refs, 1);
1286 atomic_set(&root->will_be_snapshoted, 0);
1287 atomic_set(&root->qgroup_meta_rsv, 0);
1288 root->log_transid = 0;
1289 root->log_transid_committed = -1;
1290 root->last_log_commit = 0;
1291 if (!dummy)
1292 extent_io_tree_init(&root->dirty_log_pages,
1293 fs_info->btree_inode->i_mapping);
1294
1295 memset(&root->root_key, 0, sizeof(root->root_key));
1296 memset(&root->root_item, 0, sizeof(root->root_item));
1297 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1298 if (!dummy)
1299 root->defrag_trans_start = fs_info->generation;
1300 else
1301 root->defrag_trans_start = 0;
1302 root->root_key.objectid = objectid;
1303 root->anon_dev = 0;
1304
1305 spin_lock_init(&root->root_item_lock);
1306}
1307
1308static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1309 gfp_t flags)
1310{
1311 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1312 if (root)
1313 root->fs_info = fs_info;
1314 return root;
1315}
1316
1317#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1318/* Should only be used by the testing infrastructure */
1319struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info,
1320 u32 sectorsize, u32 nodesize)
1321{
1322 struct btrfs_root *root;
1323
1324 if (!fs_info)
1325 return ERR_PTR(-EINVAL);
1326
1327 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1328 if (!root)
1329 return ERR_PTR(-ENOMEM);
1330 /* We don't use the stripesize in selftest, set it as sectorsize */
1331 __setup_root(nodesize, sectorsize, sectorsize, root, fs_info,
1332 BTRFS_ROOT_TREE_OBJECTID);
1333 root->alloc_bytenr = 0;
1334
1335 return root;
1336}
1337#endif
1338
1339struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1340 struct btrfs_fs_info *fs_info,
1341 u64 objectid)
1342{
1343 struct extent_buffer *leaf;
1344 struct btrfs_root *tree_root = fs_info->tree_root;
1345 struct btrfs_root *root;
1346 struct btrfs_key key;
1347 int ret = 0;
1348 uuid_le uuid;
1349
1350 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1351 if (!root)
1352 return ERR_PTR(-ENOMEM);
1353
1354 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1355 tree_root->stripesize, root, fs_info, objectid);
1356 root->root_key.objectid = objectid;
1357 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1358 root->root_key.offset = 0;
1359
1360 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1361 if (IS_ERR(leaf)) {
1362 ret = PTR_ERR(leaf);
1363 leaf = NULL;
1364 goto fail;
1365 }
1366
1367 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1368 btrfs_set_header_bytenr(leaf, leaf->start);
1369 btrfs_set_header_generation(leaf, trans->transid);
1370 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1371 btrfs_set_header_owner(leaf, objectid);
1372 root->node = leaf;
1373
1374 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1375 BTRFS_FSID_SIZE);
1376 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1377 btrfs_header_chunk_tree_uuid(leaf),
1378 BTRFS_UUID_SIZE);
1379 btrfs_mark_buffer_dirty(leaf);
1380
1381 root->commit_root = btrfs_root_node(root);
1382 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1383
1384 root->root_item.flags = 0;
1385 root->root_item.byte_limit = 0;
1386 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1387 btrfs_set_root_generation(&root->root_item, trans->transid);
1388 btrfs_set_root_level(&root->root_item, 0);
1389 btrfs_set_root_refs(&root->root_item, 1);
1390 btrfs_set_root_used(&root->root_item, leaf->len);
1391 btrfs_set_root_last_snapshot(&root->root_item, 0);
1392 btrfs_set_root_dirid(&root->root_item, 0);
1393 uuid_le_gen(&uuid);
1394 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1395 root->root_item.drop_level = 0;
1396
1397 key.objectid = objectid;
1398 key.type = BTRFS_ROOT_ITEM_KEY;
1399 key.offset = 0;
1400 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1401 if (ret)
1402 goto fail;
1403
1404 btrfs_tree_unlock(leaf);
1405
1406 return root;
1407
1408fail:
1409 if (leaf) {
1410 btrfs_tree_unlock(leaf);
1411 free_extent_buffer(root->commit_root);
1412 free_extent_buffer(leaf);
1413 }
1414 kfree(root);
1415
1416 return ERR_PTR(ret);
1417}
1418
1419static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1420 struct btrfs_fs_info *fs_info)
1421{
1422 struct btrfs_root *root;
1423 struct btrfs_root *tree_root = fs_info->tree_root;
1424 struct extent_buffer *leaf;
1425
1426 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1427 if (!root)
1428 return ERR_PTR(-ENOMEM);
1429
1430 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1431 tree_root->stripesize, root, fs_info,
1432 BTRFS_TREE_LOG_OBJECTID);
1433
1434 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1435 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1436 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1437
1438 /*
1439 * DON'T set REF_COWS for log trees
1440 *
1441 * log trees do not get reference counted because they go away
1442 * before a real commit is actually done. They do store pointers
1443 * to file data extents, and those reference counts still get
1444 * updated (along with back refs to the log tree).
1445 */
1446
1447 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1448 NULL, 0, 0, 0);
1449 if (IS_ERR(leaf)) {
1450 kfree(root);
1451 return ERR_CAST(leaf);
1452 }
1453
1454 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1455 btrfs_set_header_bytenr(leaf, leaf->start);
1456 btrfs_set_header_generation(leaf, trans->transid);
1457 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1458 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1459 root->node = leaf;
1460
1461 write_extent_buffer(root->node, root->fs_info->fsid,
1462 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1463 btrfs_mark_buffer_dirty(root->node);
1464 btrfs_tree_unlock(root->node);
1465 return root;
1466}
1467
1468int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1469 struct btrfs_fs_info *fs_info)
1470{
1471 struct btrfs_root *log_root;
1472
1473 log_root = alloc_log_tree(trans, fs_info);
1474 if (IS_ERR(log_root))
1475 return PTR_ERR(log_root);
1476 WARN_ON(fs_info->log_root_tree);
1477 fs_info->log_root_tree = log_root;
1478 return 0;
1479}
1480
1481int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1482 struct btrfs_root *root)
1483{
1484 struct btrfs_root *log_root;
1485 struct btrfs_inode_item *inode_item;
1486
1487 log_root = alloc_log_tree(trans, root->fs_info);
1488 if (IS_ERR(log_root))
1489 return PTR_ERR(log_root);
1490
1491 log_root->last_trans = trans->transid;
1492 log_root->root_key.offset = root->root_key.objectid;
1493
1494 inode_item = &log_root->root_item.inode;
1495 btrfs_set_stack_inode_generation(inode_item, 1);
1496 btrfs_set_stack_inode_size(inode_item, 3);
1497 btrfs_set_stack_inode_nlink(inode_item, 1);
1498 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1499 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1500
1501 btrfs_set_root_node(&log_root->root_item, log_root->node);
1502
1503 WARN_ON(root->log_root);
1504 root->log_root = log_root;
1505 root->log_transid = 0;
1506 root->log_transid_committed = -1;
1507 root->last_log_commit = 0;
1508 return 0;
1509}
1510
1511static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1512 struct btrfs_key *key)
1513{
1514 struct btrfs_root *root;
1515 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1516 struct btrfs_path *path;
1517 u64 generation;
1518 int ret;
1519
1520 path = btrfs_alloc_path();
1521 if (!path)
1522 return ERR_PTR(-ENOMEM);
1523
1524 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1525 if (!root) {
1526 ret = -ENOMEM;
1527 goto alloc_fail;
1528 }
1529
1530 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1531 tree_root->stripesize, root, fs_info, key->objectid);
1532
1533 ret = btrfs_find_root(tree_root, key, path,
1534 &root->root_item, &root->root_key);
1535 if (ret) {
1536 if (ret > 0)
1537 ret = -ENOENT;
1538 goto find_fail;
1539 }
1540
1541 generation = btrfs_root_generation(&root->root_item);
1542 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1543 generation);
1544 if (IS_ERR(root->node)) {
1545 ret = PTR_ERR(root->node);
1546 goto find_fail;
1547 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1548 ret = -EIO;
1549 free_extent_buffer(root->node);
1550 goto find_fail;
1551 }
1552 root->commit_root = btrfs_root_node(root);
1553out:
1554 btrfs_free_path(path);
1555 return root;
1556
1557find_fail:
1558 kfree(root);
1559alloc_fail:
1560 root = ERR_PTR(ret);
1561 goto out;
1562}
1563
1564struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1565 struct btrfs_key *location)
1566{
1567 struct btrfs_root *root;
1568
1569 root = btrfs_read_tree_root(tree_root, location);
1570 if (IS_ERR(root))
1571 return root;
1572
1573 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1574 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1575 btrfs_check_and_init_root_item(&root->root_item);
1576 }
1577
1578 return root;
1579}
1580
1581int btrfs_init_fs_root(struct btrfs_root *root)
1582{
1583 int ret;
1584 struct btrfs_subvolume_writers *writers;
1585
1586 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1587 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1588 GFP_NOFS);
1589 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1590 ret = -ENOMEM;
1591 goto fail;
1592 }
1593
1594 writers = btrfs_alloc_subvolume_writers();
1595 if (IS_ERR(writers)) {
1596 ret = PTR_ERR(writers);
1597 goto fail;
1598 }
1599 root->subv_writers = writers;
1600
1601 btrfs_init_free_ino_ctl(root);
1602 spin_lock_init(&root->ino_cache_lock);
1603 init_waitqueue_head(&root->ino_cache_wait);
1604
1605 ret = get_anon_bdev(&root->anon_dev);
1606 if (ret)
1607 goto fail;
1608
1609 mutex_lock(&root->objectid_mutex);
1610 ret = btrfs_find_highest_objectid(root,
1611 &root->highest_objectid);
1612 if (ret) {
1613 mutex_unlock(&root->objectid_mutex);
1614 goto fail;
1615 }
1616
1617 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1618
1619 mutex_unlock(&root->objectid_mutex);
1620
1621 return 0;
1622fail:
1623 /* the caller is responsible to call free_fs_root */
1624 return ret;
1625}
1626
1627static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1628 u64 root_id)
1629{
1630 struct btrfs_root *root;
1631
1632 spin_lock(&fs_info->fs_roots_radix_lock);
1633 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1634 (unsigned long)root_id);
1635 spin_unlock(&fs_info->fs_roots_radix_lock);
1636 return root;
1637}
1638
1639int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1640 struct btrfs_root *root)
1641{
1642 int ret;
1643
1644 ret = radix_tree_preload(GFP_NOFS);
1645 if (ret)
1646 return ret;
1647
1648 spin_lock(&fs_info->fs_roots_radix_lock);
1649 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1650 (unsigned long)root->root_key.objectid,
1651 root);
1652 if (ret == 0)
1653 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1654 spin_unlock(&fs_info->fs_roots_radix_lock);
1655 radix_tree_preload_end();
1656
1657 return ret;
1658}
1659
1660struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1661 struct btrfs_key *location,
1662 bool check_ref)
1663{
1664 struct btrfs_root *root;
1665 struct btrfs_path *path;
1666 struct btrfs_key key;
1667 int ret;
1668
1669 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1670 return fs_info->tree_root;
1671 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1672 return fs_info->extent_root;
1673 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1674 return fs_info->chunk_root;
1675 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1676 return fs_info->dev_root;
1677 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1678 return fs_info->csum_root;
1679 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1680 return fs_info->quota_root ? fs_info->quota_root :
1681 ERR_PTR(-ENOENT);
1682 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1683 return fs_info->uuid_root ? fs_info->uuid_root :
1684 ERR_PTR(-ENOENT);
1685 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1686 return fs_info->free_space_root ? fs_info->free_space_root :
1687 ERR_PTR(-ENOENT);
1688again:
1689 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1690 if (root) {
1691 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1692 return ERR_PTR(-ENOENT);
1693 return root;
1694 }
1695
1696 root = btrfs_read_fs_root(fs_info->tree_root, location);
1697 if (IS_ERR(root))
1698 return root;
1699
1700 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1701 ret = -ENOENT;
1702 goto fail;
1703 }
1704
1705 ret = btrfs_init_fs_root(root);
1706 if (ret)
1707 goto fail;
1708
1709 path = btrfs_alloc_path();
1710 if (!path) {
1711 ret = -ENOMEM;
1712 goto fail;
1713 }
1714 key.objectid = BTRFS_ORPHAN_OBJECTID;
1715 key.type = BTRFS_ORPHAN_ITEM_KEY;
1716 key.offset = location->objectid;
1717
1718 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1719 btrfs_free_path(path);
1720 if (ret < 0)
1721 goto fail;
1722 if (ret == 0)
1723 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1724
1725 ret = btrfs_insert_fs_root(fs_info, root);
1726 if (ret) {
1727 if (ret == -EEXIST) {
1728 free_fs_root(root);
1729 goto again;
1730 }
1731 goto fail;
1732 }
1733 return root;
1734fail:
1735 free_fs_root(root);
1736 return ERR_PTR(ret);
1737}
1738
1739static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1740{
1741 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1742 int ret = 0;
1743 struct btrfs_device *device;
1744 struct backing_dev_info *bdi;
1745
1746 rcu_read_lock();
1747 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1748 if (!device->bdev)
1749 continue;
1750 bdi = blk_get_backing_dev_info(device->bdev);
1751 if (bdi_congested(bdi, bdi_bits)) {
1752 ret = 1;
1753 break;
1754 }
1755 }
1756 rcu_read_unlock();
1757 return ret;
1758}
1759
1760static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1761{
1762 int err;
1763
1764 err = bdi_setup_and_register(bdi, "btrfs");
1765 if (err)
1766 return err;
1767
1768 bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1769 bdi->congested_fn = btrfs_congested_fn;
1770 bdi->congested_data = info;
1771 bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1772 return 0;
1773}
1774
1775/*
1776 * called by the kthread helper functions to finally call the bio end_io
1777 * functions. This is where read checksum verification actually happens
1778 */
1779static void end_workqueue_fn(struct btrfs_work *work)
1780{
1781 struct bio *bio;
1782 struct btrfs_end_io_wq *end_io_wq;
1783
1784 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1785 bio = end_io_wq->bio;
1786
1787 bio->bi_error = end_io_wq->error;
1788 bio->bi_private = end_io_wq->private;
1789 bio->bi_end_io = end_io_wq->end_io;
1790 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1791 bio_endio(bio);
1792}
1793
1794static int cleaner_kthread(void *arg)
1795{
1796 struct btrfs_root *root = arg;
1797 int again;
1798 struct btrfs_trans_handle *trans;
1799
1800 do {
1801 again = 0;
1802
1803 /* Make the cleaner go to sleep early. */
1804 if (btrfs_need_cleaner_sleep(root))
1805 goto sleep;
1806
1807 /*
1808 * Do not do anything if we might cause open_ctree() to block
1809 * before we have finished mounting the filesystem.
1810 */
1811 if (!root->fs_info->open)
1812 goto sleep;
1813
1814 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1815 goto sleep;
1816
1817 /*
1818 * Avoid the problem that we change the status of the fs
1819 * during the above check and trylock.
1820 */
1821 if (btrfs_need_cleaner_sleep(root)) {
1822 mutex_unlock(&root->fs_info->cleaner_mutex);
1823 goto sleep;
1824 }
1825
1826 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1827 btrfs_run_delayed_iputs(root);
1828 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1829
1830 again = btrfs_clean_one_deleted_snapshot(root);
1831 mutex_unlock(&root->fs_info->cleaner_mutex);
1832
1833 /*
1834 * The defragger has dealt with the R/O remount and umount,
1835 * needn't do anything special here.
1836 */
1837 btrfs_run_defrag_inodes(root->fs_info);
1838
1839 /*
1840 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1841 * with relocation (btrfs_relocate_chunk) and relocation
1842 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1843 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1844 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1845 * unused block groups.
1846 */
1847 btrfs_delete_unused_bgs(root->fs_info);
1848sleep:
1849 if (!again) {
1850 set_current_state(TASK_INTERRUPTIBLE);
1851 if (!kthread_should_stop())
1852 schedule();
1853 __set_current_state(TASK_RUNNING);
1854 }
1855 } while (!kthread_should_stop());
1856
1857 /*
1858 * Transaction kthread is stopped before us and wakes us up.
1859 * However we might have started a new transaction and COWed some
1860 * tree blocks when deleting unused block groups for example. So
1861 * make sure we commit the transaction we started to have a clean
1862 * shutdown when evicting the btree inode - if it has dirty pages
1863 * when we do the final iput() on it, eviction will trigger a
1864 * writeback for it which will fail with null pointer dereferences
1865 * since work queues and other resources were already released and
1866 * destroyed by the time the iput/eviction/writeback is made.
1867 */
1868 trans = btrfs_attach_transaction(root);
1869 if (IS_ERR(trans)) {
1870 if (PTR_ERR(trans) != -ENOENT)
1871 btrfs_err(root->fs_info,
1872 "cleaner transaction attach returned %ld",
1873 PTR_ERR(trans));
1874 } else {
1875 int ret;
1876
1877 ret = btrfs_commit_transaction(trans, root);
1878 if (ret)
1879 btrfs_err(root->fs_info,
1880 "cleaner open transaction commit returned %d",
1881 ret);
1882 }
1883
1884 return 0;
1885}
1886
1887static int transaction_kthread(void *arg)
1888{
1889 struct btrfs_root *root = arg;
1890 struct btrfs_trans_handle *trans;
1891 struct btrfs_transaction *cur;
1892 u64 transid;
1893 unsigned long now;
1894 unsigned long delay;
1895 bool cannot_commit;
1896
1897 do {
1898 cannot_commit = false;
1899 delay = HZ * root->fs_info->commit_interval;
1900 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1901
1902 spin_lock(&root->fs_info->trans_lock);
1903 cur = root->fs_info->running_transaction;
1904 if (!cur) {
1905 spin_unlock(&root->fs_info->trans_lock);
1906 goto sleep;
1907 }
1908
1909 now = get_seconds();
1910 if (cur->state < TRANS_STATE_BLOCKED &&
1911 (now < cur->start_time ||
1912 now - cur->start_time < root->fs_info->commit_interval)) {
1913 spin_unlock(&root->fs_info->trans_lock);
1914 delay = HZ * 5;
1915 goto sleep;
1916 }
1917 transid = cur->transid;
1918 spin_unlock(&root->fs_info->trans_lock);
1919
1920 /* If the file system is aborted, this will always fail. */
1921 trans = btrfs_attach_transaction(root);
1922 if (IS_ERR(trans)) {
1923 if (PTR_ERR(trans) != -ENOENT)
1924 cannot_commit = true;
1925 goto sleep;
1926 }
1927 if (transid == trans->transid) {
1928 btrfs_commit_transaction(trans, root);
1929 } else {
1930 btrfs_end_transaction(trans, root);
1931 }
1932sleep:
1933 wake_up_process(root->fs_info->cleaner_kthread);
1934 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1935
1936 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1937 &root->fs_info->fs_state)))
1938 btrfs_cleanup_transaction(root);
1939 set_current_state(TASK_INTERRUPTIBLE);
1940 if (!kthread_should_stop() &&
1941 (!btrfs_transaction_blocked(root->fs_info) ||
1942 cannot_commit))
1943 schedule_timeout(delay);
1944 __set_current_state(TASK_RUNNING);
1945 } while (!kthread_should_stop());
1946 return 0;
1947}
1948
1949/*
1950 * this will find the highest generation in the array of
1951 * root backups. The index of the highest array is returned,
1952 * or -1 if we can't find anything.
1953 *
1954 * We check to make sure the array is valid by comparing the
1955 * generation of the latest root in the array with the generation
1956 * in the super block. If they don't match we pitch it.
1957 */
1958static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1959{
1960 u64 cur;
1961 int newest_index = -1;
1962 struct btrfs_root_backup *root_backup;
1963 int i;
1964
1965 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1966 root_backup = info->super_copy->super_roots + i;
1967 cur = btrfs_backup_tree_root_gen(root_backup);
1968 if (cur == newest_gen)
1969 newest_index = i;
1970 }
1971
1972 /* check to see if we actually wrapped around */
1973 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1974 root_backup = info->super_copy->super_roots;
1975 cur = btrfs_backup_tree_root_gen(root_backup);
1976 if (cur == newest_gen)
1977 newest_index = 0;
1978 }
1979 return newest_index;
1980}
1981
1982
1983/*
1984 * find the oldest backup so we know where to store new entries
1985 * in the backup array. This will set the backup_root_index
1986 * field in the fs_info struct
1987 */
1988static void find_oldest_super_backup(struct btrfs_fs_info *info,
1989 u64 newest_gen)
1990{
1991 int newest_index = -1;
1992
1993 newest_index = find_newest_super_backup(info, newest_gen);
1994 /* if there was garbage in there, just move along */
1995 if (newest_index == -1) {
1996 info->backup_root_index = 0;
1997 } else {
1998 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1999 }
2000}
2001
2002/*
2003 * copy all the root pointers into the super backup array.
2004 * this will bump the backup pointer by one when it is
2005 * done
2006 */
2007static void backup_super_roots(struct btrfs_fs_info *info)
2008{
2009 int next_backup;
2010 struct btrfs_root_backup *root_backup;
2011 int last_backup;
2012
2013 next_backup = info->backup_root_index;
2014 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2015 BTRFS_NUM_BACKUP_ROOTS;
2016
2017 /*
2018 * just overwrite the last backup if we're at the same generation
2019 * this happens only at umount
2020 */
2021 root_backup = info->super_for_commit->super_roots + last_backup;
2022 if (btrfs_backup_tree_root_gen(root_backup) ==
2023 btrfs_header_generation(info->tree_root->node))
2024 next_backup = last_backup;
2025
2026 root_backup = info->super_for_commit->super_roots + next_backup;
2027
2028 /*
2029 * make sure all of our padding and empty slots get zero filled
2030 * regardless of which ones we use today
2031 */
2032 memset(root_backup, 0, sizeof(*root_backup));
2033
2034 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2035
2036 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2037 btrfs_set_backup_tree_root_gen(root_backup,
2038 btrfs_header_generation(info->tree_root->node));
2039
2040 btrfs_set_backup_tree_root_level(root_backup,
2041 btrfs_header_level(info->tree_root->node));
2042
2043 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2044 btrfs_set_backup_chunk_root_gen(root_backup,
2045 btrfs_header_generation(info->chunk_root->node));
2046 btrfs_set_backup_chunk_root_level(root_backup,
2047 btrfs_header_level(info->chunk_root->node));
2048
2049 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2050 btrfs_set_backup_extent_root_gen(root_backup,
2051 btrfs_header_generation(info->extent_root->node));
2052 btrfs_set_backup_extent_root_level(root_backup,
2053 btrfs_header_level(info->extent_root->node));
2054
2055 /*
2056 * we might commit during log recovery, which happens before we set
2057 * the fs_root. Make sure it is valid before we fill it in.
2058 */
2059 if (info->fs_root && info->fs_root->node) {
2060 btrfs_set_backup_fs_root(root_backup,
2061 info->fs_root->node->start);
2062 btrfs_set_backup_fs_root_gen(root_backup,
2063 btrfs_header_generation(info->fs_root->node));
2064 btrfs_set_backup_fs_root_level(root_backup,
2065 btrfs_header_level(info->fs_root->node));
2066 }
2067
2068 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2069 btrfs_set_backup_dev_root_gen(root_backup,
2070 btrfs_header_generation(info->dev_root->node));
2071 btrfs_set_backup_dev_root_level(root_backup,
2072 btrfs_header_level(info->dev_root->node));
2073
2074 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2075 btrfs_set_backup_csum_root_gen(root_backup,
2076 btrfs_header_generation(info->csum_root->node));
2077 btrfs_set_backup_csum_root_level(root_backup,
2078 btrfs_header_level(info->csum_root->node));
2079
2080 btrfs_set_backup_total_bytes(root_backup,
2081 btrfs_super_total_bytes(info->super_copy));
2082 btrfs_set_backup_bytes_used(root_backup,
2083 btrfs_super_bytes_used(info->super_copy));
2084 btrfs_set_backup_num_devices(root_backup,
2085 btrfs_super_num_devices(info->super_copy));
2086
2087 /*
2088 * if we don't copy this out to the super_copy, it won't get remembered
2089 * for the next commit
2090 */
2091 memcpy(&info->super_copy->super_roots,
2092 &info->super_for_commit->super_roots,
2093 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2094}
2095
2096/*
2097 * this copies info out of the root backup array and back into
2098 * the in-memory super block. It is meant to help iterate through
2099 * the array, so you send it the number of backups you've already
2100 * tried and the last backup index you used.
2101 *
2102 * this returns -1 when it has tried all the backups
2103 */
2104static noinline int next_root_backup(struct btrfs_fs_info *info,
2105 struct btrfs_super_block *super,
2106 int *num_backups_tried, int *backup_index)
2107{
2108 struct btrfs_root_backup *root_backup;
2109 int newest = *backup_index;
2110
2111 if (*num_backups_tried == 0) {
2112 u64 gen = btrfs_super_generation(super);
2113
2114 newest = find_newest_super_backup(info, gen);
2115 if (newest == -1)
2116 return -1;
2117
2118 *backup_index = newest;
2119 *num_backups_tried = 1;
2120 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2121 /* we've tried all the backups, all done */
2122 return -1;
2123 } else {
2124 /* jump to the next oldest backup */
2125 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2126 BTRFS_NUM_BACKUP_ROOTS;
2127 *backup_index = newest;
2128 *num_backups_tried += 1;
2129 }
2130 root_backup = super->super_roots + newest;
2131
2132 btrfs_set_super_generation(super,
2133 btrfs_backup_tree_root_gen(root_backup));
2134 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2135 btrfs_set_super_root_level(super,
2136 btrfs_backup_tree_root_level(root_backup));
2137 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2138
2139 /*
2140 * fixme: the total bytes and num_devices need to match or we should
2141 * need a fsck
2142 */
2143 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2144 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2145 return 0;
2146}
2147
2148/* helper to cleanup workers */
2149static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2150{
2151 btrfs_destroy_workqueue(fs_info->fixup_workers);
2152 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2153 btrfs_destroy_workqueue(fs_info->workers);
2154 btrfs_destroy_workqueue(fs_info->endio_workers);
2155 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2156 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2157 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2158 btrfs_destroy_workqueue(fs_info->rmw_workers);
2159 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2160 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2161 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2162 btrfs_destroy_workqueue(fs_info->submit_workers);
2163 btrfs_destroy_workqueue(fs_info->delayed_workers);
2164 btrfs_destroy_workqueue(fs_info->caching_workers);
2165 btrfs_destroy_workqueue(fs_info->readahead_workers);
2166 btrfs_destroy_workqueue(fs_info->flush_workers);
2167 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2168 btrfs_destroy_workqueue(fs_info->extent_workers);
2169}
2170
2171static void free_root_extent_buffers(struct btrfs_root *root)
2172{
2173 if (root) {
2174 free_extent_buffer(root->node);
2175 free_extent_buffer(root->commit_root);
2176 root->node = NULL;
2177 root->commit_root = NULL;
2178 }
2179}
2180
2181/* helper to cleanup tree roots */
2182static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2183{
2184 free_root_extent_buffers(info->tree_root);
2185
2186 free_root_extent_buffers(info->dev_root);
2187 free_root_extent_buffers(info->extent_root);
2188 free_root_extent_buffers(info->csum_root);
2189 free_root_extent_buffers(info->quota_root);
2190 free_root_extent_buffers(info->uuid_root);
2191 if (chunk_root)
2192 free_root_extent_buffers(info->chunk_root);
2193 free_root_extent_buffers(info->free_space_root);
2194}
2195
2196void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2197{
2198 int ret;
2199 struct btrfs_root *gang[8];
2200 int i;
2201
2202 while (!list_empty(&fs_info->dead_roots)) {
2203 gang[0] = list_entry(fs_info->dead_roots.next,
2204 struct btrfs_root, root_list);
2205 list_del(&gang[0]->root_list);
2206
2207 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2208 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2209 } else {
2210 free_extent_buffer(gang[0]->node);
2211 free_extent_buffer(gang[0]->commit_root);
2212 btrfs_put_fs_root(gang[0]);
2213 }
2214 }
2215
2216 while (1) {
2217 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2218 (void **)gang, 0,
2219 ARRAY_SIZE(gang));
2220 if (!ret)
2221 break;
2222 for (i = 0; i < ret; i++)
2223 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2224 }
2225
2226 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2227 btrfs_free_log_root_tree(NULL, fs_info);
2228 btrfs_destroy_pinned_extent(fs_info->tree_root,
2229 fs_info->pinned_extents);
2230 }
2231}
2232
2233static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2234{
2235 mutex_init(&fs_info->scrub_lock);
2236 atomic_set(&fs_info->scrubs_running, 0);
2237 atomic_set(&fs_info->scrub_pause_req, 0);
2238 atomic_set(&fs_info->scrubs_paused, 0);
2239 atomic_set(&fs_info->scrub_cancel_req, 0);
2240 init_waitqueue_head(&fs_info->scrub_pause_wait);
2241 fs_info->scrub_workers_refcnt = 0;
2242}
2243
2244static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2245{
2246 spin_lock_init(&fs_info->balance_lock);
2247 mutex_init(&fs_info->balance_mutex);
2248 atomic_set(&fs_info->balance_running, 0);
2249 atomic_set(&fs_info->balance_pause_req, 0);
2250 atomic_set(&fs_info->balance_cancel_req, 0);
2251 fs_info->balance_ctl = NULL;
2252 init_waitqueue_head(&fs_info->balance_wait_q);
2253}
2254
2255static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2256 struct btrfs_root *tree_root)
2257{
2258 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2259 set_nlink(fs_info->btree_inode, 1);
2260 /*
2261 * we set the i_size on the btree inode to the max possible int.
2262 * the real end of the address space is determined by all of
2263 * the devices in the system
2264 */
2265 fs_info->btree_inode->i_size = OFFSET_MAX;
2266 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2267
2268 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2269 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2270 fs_info->btree_inode->i_mapping);
2271 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2272 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2273
2274 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2275
2276 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2277 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2278 sizeof(struct btrfs_key));
2279 set_bit(BTRFS_INODE_DUMMY,
2280 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2281 btrfs_insert_inode_hash(fs_info->btree_inode);
2282}
2283
2284static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2285{
2286 fs_info->dev_replace.lock_owner = 0;
2287 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2288 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2289 rwlock_init(&fs_info->dev_replace.lock);
2290 atomic_set(&fs_info->dev_replace.read_locks, 0);
2291 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2292 init_waitqueue_head(&fs_info->replace_wait);
2293 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2294}
2295
2296static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2297{
2298 spin_lock_init(&fs_info->qgroup_lock);
2299 mutex_init(&fs_info->qgroup_ioctl_lock);
2300 fs_info->qgroup_tree = RB_ROOT;
2301 fs_info->qgroup_op_tree = RB_ROOT;
2302 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2303 fs_info->qgroup_seq = 1;
2304 fs_info->quota_enabled = 0;
2305 fs_info->pending_quota_state = 0;
2306 fs_info->qgroup_ulist = NULL;
2307 fs_info->qgroup_rescan_running = false;
2308 mutex_init(&fs_info->qgroup_rescan_lock);
2309}
2310
2311static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2312 struct btrfs_fs_devices *fs_devices)
2313{
2314 int max_active = fs_info->thread_pool_size;
2315 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2316
2317 fs_info->workers =
2318 btrfs_alloc_workqueue(fs_info, "worker",
2319 flags | WQ_HIGHPRI, max_active, 16);
2320
2321 fs_info->delalloc_workers =
2322 btrfs_alloc_workqueue(fs_info, "delalloc",
2323 flags, max_active, 2);
2324
2325 fs_info->flush_workers =
2326 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2327 flags, max_active, 0);
2328
2329 fs_info->caching_workers =
2330 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2331
2332 /*
2333 * a higher idle thresh on the submit workers makes it much more
2334 * likely that bios will be send down in a sane order to the
2335 * devices
2336 */
2337 fs_info->submit_workers =
2338 btrfs_alloc_workqueue(fs_info, "submit", flags,
2339 min_t(u64, fs_devices->num_devices,
2340 max_active), 64);
2341
2342 fs_info->fixup_workers =
2343 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2344
2345 /*
2346 * endios are largely parallel and should have a very
2347 * low idle thresh
2348 */
2349 fs_info->endio_workers =
2350 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2351 fs_info->endio_meta_workers =
2352 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2353 max_active, 4);
2354 fs_info->endio_meta_write_workers =
2355 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2356 max_active, 2);
2357 fs_info->endio_raid56_workers =
2358 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2359 max_active, 4);
2360 fs_info->endio_repair_workers =
2361 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2362 fs_info->rmw_workers =
2363 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2364 fs_info->endio_write_workers =
2365 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2366 max_active, 2);
2367 fs_info->endio_freespace_worker =
2368 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2369 max_active, 0);
2370 fs_info->delayed_workers =
2371 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2372 max_active, 0);
2373 fs_info->readahead_workers =
2374 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2375 max_active, 2);
2376 fs_info->qgroup_rescan_workers =
2377 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2378 fs_info->extent_workers =
2379 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2380 min_t(u64, fs_devices->num_devices,
2381 max_active), 8);
2382
2383 if (!(fs_info->workers && fs_info->delalloc_workers &&
2384 fs_info->submit_workers && fs_info->flush_workers &&
2385 fs_info->endio_workers && fs_info->endio_meta_workers &&
2386 fs_info->endio_meta_write_workers &&
2387 fs_info->endio_repair_workers &&
2388 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2389 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2390 fs_info->caching_workers && fs_info->readahead_workers &&
2391 fs_info->fixup_workers && fs_info->delayed_workers &&
2392 fs_info->extent_workers &&
2393 fs_info->qgroup_rescan_workers)) {
2394 return -ENOMEM;
2395 }
2396
2397 return 0;
2398}
2399
2400static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2401 struct btrfs_fs_devices *fs_devices)
2402{
2403 int ret;
2404 struct btrfs_root *tree_root = fs_info->tree_root;
2405 struct btrfs_root *log_tree_root;
2406 struct btrfs_super_block *disk_super = fs_info->super_copy;
2407 u64 bytenr = btrfs_super_log_root(disk_super);
2408
2409 if (fs_devices->rw_devices == 0) {
2410 btrfs_warn(fs_info, "log replay required on RO media");
2411 return -EIO;
2412 }
2413
2414 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2415 if (!log_tree_root)
2416 return -ENOMEM;
2417
2418 __setup_root(tree_root->nodesize, tree_root->sectorsize,
2419 tree_root->stripesize, log_tree_root, fs_info,
2420 BTRFS_TREE_LOG_OBJECTID);
2421
2422 log_tree_root->node = read_tree_block(tree_root, bytenr,
2423 fs_info->generation + 1);
2424 if (IS_ERR(log_tree_root->node)) {
2425 btrfs_warn(fs_info, "failed to read log tree");
2426 ret = PTR_ERR(log_tree_root->node);
2427 kfree(log_tree_root);
2428 return ret;
2429 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2430 btrfs_err(fs_info, "failed to read log tree");
2431 free_extent_buffer(log_tree_root->node);
2432 kfree(log_tree_root);
2433 return -EIO;
2434 }
2435 /* returns with log_tree_root freed on success */
2436 ret = btrfs_recover_log_trees(log_tree_root);
2437 if (ret) {
2438 btrfs_handle_fs_error(tree_root->fs_info, ret,
2439 "Failed to recover log tree");
2440 free_extent_buffer(log_tree_root->node);
2441 kfree(log_tree_root);
2442 return ret;
2443 }
2444
2445 if (fs_info->sb->s_flags & MS_RDONLY) {
2446 ret = btrfs_commit_super(tree_root);
2447 if (ret)
2448 return ret;
2449 }
2450
2451 return 0;
2452}
2453
2454static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2455 struct btrfs_root *tree_root)
2456{
2457 struct btrfs_root *root;
2458 struct btrfs_key location;
2459 int ret;
2460
2461 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2462 location.type = BTRFS_ROOT_ITEM_KEY;
2463 location.offset = 0;
2464
2465 root = btrfs_read_tree_root(tree_root, &location);
2466 if (IS_ERR(root))
2467 return PTR_ERR(root);
2468 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2469 fs_info->extent_root = root;
2470
2471 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2472 root = btrfs_read_tree_root(tree_root, &location);
2473 if (IS_ERR(root))
2474 return PTR_ERR(root);
2475 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2476 fs_info->dev_root = root;
2477 btrfs_init_devices_late(fs_info);
2478
2479 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2480 root = btrfs_read_tree_root(tree_root, &location);
2481 if (IS_ERR(root))
2482 return PTR_ERR(root);
2483 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2484 fs_info->csum_root = root;
2485
2486 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2487 root = btrfs_read_tree_root(tree_root, &location);
2488 if (!IS_ERR(root)) {
2489 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2490 fs_info->quota_enabled = 1;
2491 fs_info->pending_quota_state = 1;
2492 fs_info->quota_root = root;
2493 }
2494
2495 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2496 root = btrfs_read_tree_root(tree_root, &location);
2497 if (IS_ERR(root)) {
2498 ret = PTR_ERR(root);
2499 if (ret != -ENOENT)
2500 return ret;
2501 } else {
2502 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2503 fs_info->uuid_root = root;
2504 }
2505
2506 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2507 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2508 root = btrfs_read_tree_root(tree_root, &location);
2509 if (IS_ERR(root))
2510 return PTR_ERR(root);
2511 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2512 fs_info->free_space_root = root;
2513 }
2514
2515 return 0;
2516}
2517
2518int open_ctree(struct super_block *sb,
2519 struct btrfs_fs_devices *fs_devices,
2520 char *options)
2521{
2522 u32 sectorsize;
2523 u32 nodesize;
2524 u32 stripesize;
2525 u64 generation;
2526 u64 features;
2527 struct btrfs_key location;
2528 struct buffer_head *bh;
2529 struct btrfs_super_block *disk_super;
2530 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2531 struct btrfs_root *tree_root;
2532 struct btrfs_root *chunk_root;
2533 int ret;
2534 int err = -EINVAL;
2535 int num_backups_tried = 0;
2536 int backup_index = 0;
2537 int max_active;
2538
2539 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2540 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2541 if (!tree_root || !chunk_root) {
2542 err = -ENOMEM;
2543 goto fail;
2544 }
2545
2546 ret = init_srcu_struct(&fs_info->subvol_srcu);
2547 if (ret) {
2548 err = ret;
2549 goto fail;
2550 }
2551
2552 ret = setup_bdi(fs_info, &fs_info->bdi);
2553 if (ret) {
2554 err = ret;
2555 goto fail_srcu;
2556 }
2557
2558 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2559 if (ret) {
2560 err = ret;
2561 goto fail_bdi;
2562 }
2563 fs_info->dirty_metadata_batch = PAGE_SIZE *
2564 (1 + ilog2(nr_cpu_ids));
2565
2566 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2567 if (ret) {
2568 err = ret;
2569 goto fail_dirty_metadata_bytes;
2570 }
2571
2572 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2573 if (ret) {
2574 err = ret;
2575 goto fail_delalloc_bytes;
2576 }
2577
2578 fs_info->btree_inode = new_inode(sb);
2579 if (!fs_info->btree_inode) {
2580 err = -ENOMEM;
2581 goto fail_bio_counter;
2582 }
2583
2584 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2585
2586 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2587 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2588 INIT_LIST_HEAD(&fs_info->trans_list);
2589 INIT_LIST_HEAD(&fs_info->dead_roots);
2590 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2591 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2592 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2593 spin_lock_init(&fs_info->delalloc_root_lock);
2594 spin_lock_init(&fs_info->trans_lock);
2595 spin_lock_init(&fs_info->fs_roots_radix_lock);
2596 spin_lock_init(&fs_info->delayed_iput_lock);
2597 spin_lock_init(&fs_info->defrag_inodes_lock);
2598 spin_lock_init(&fs_info->free_chunk_lock);
2599 spin_lock_init(&fs_info->tree_mod_seq_lock);
2600 spin_lock_init(&fs_info->super_lock);
2601 spin_lock_init(&fs_info->qgroup_op_lock);
2602 spin_lock_init(&fs_info->buffer_lock);
2603 spin_lock_init(&fs_info->unused_bgs_lock);
2604 rwlock_init(&fs_info->tree_mod_log_lock);
2605 mutex_init(&fs_info->unused_bg_unpin_mutex);
2606 mutex_init(&fs_info->delete_unused_bgs_mutex);
2607 mutex_init(&fs_info->reloc_mutex);
2608 mutex_init(&fs_info->delalloc_root_mutex);
2609 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2610 seqlock_init(&fs_info->profiles_lock);
2611
2612 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2613 INIT_LIST_HEAD(&fs_info->space_info);
2614 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2615 INIT_LIST_HEAD(&fs_info->unused_bgs);
2616 btrfs_mapping_init(&fs_info->mapping_tree);
2617 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2618 BTRFS_BLOCK_RSV_GLOBAL);
2619 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2620 BTRFS_BLOCK_RSV_DELALLOC);
2621 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2622 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2623 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2624 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2625 BTRFS_BLOCK_RSV_DELOPS);
2626 atomic_set(&fs_info->nr_async_submits, 0);
2627 atomic_set(&fs_info->async_delalloc_pages, 0);
2628 atomic_set(&fs_info->async_submit_draining, 0);
2629 atomic_set(&fs_info->nr_async_bios, 0);
2630 atomic_set(&fs_info->defrag_running, 0);
2631 atomic_set(&fs_info->qgroup_op_seq, 0);
2632 atomic_set(&fs_info->reada_works_cnt, 0);
2633 atomic64_set(&fs_info->tree_mod_seq, 0);
2634 fs_info->sb = sb;
2635 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2636 fs_info->metadata_ratio = 0;
2637 fs_info->defrag_inodes = RB_ROOT;
2638 fs_info->free_chunk_space = 0;
2639 fs_info->tree_mod_log = RB_ROOT;
2640 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2641 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2642 /* readahead state */
2643 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2644 spin_lock_init(&fs_info->reada_lock);
2645
2646 fs_info->thread_pool_size = min_t(unsigned long,
2647 num_online_cpus() + 2, 8);
2648
2649 INIT_LIST_HEAD(&fs_info->ordered_roots);
2650 spin_lock_init(&fs_info->ordered_root_lock);
2651 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2652 GFP_KERNEL);
2653 if (!fs_info->delayed_root) {
2654 err = -ENOMEM;
2655 goto fail_iput;
2656 }
2657 btrfs_init_delayed_root(fs_info->delayed_root);
2658
2659 btrfs_init_scrub(fs_info);
2660#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2661 fs_info->check_integrity_print_mask = 0;
2662#endif
2663 btrfs_init_balance(fs_info);
2664 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2665
2666 sb->s_blocksize = 4096;
2667 sb->s_blocksize_bits = blksize_bits(4096);
2668 sb->s_bdi = &fs_info->bdi;
2669
2670 btrfs_init_btree_inode(fs_info, tree_root);
2671
2672 spin_lock_init(&fs_info->block_group_cache_lock);
2673 fs_info->block_group_cache_tree = RB_ROOT;
2674 fs_info->first_logical_byte = (u64)-1;
2675
2676 extent_io_tree_init(&fs_info->freed_extents[0],
2677 fs_info->btree_inode->i_mapping);
2678 extent_io_tree_init(&fs_info->freed_extents[1],
2679 fs_info->btree_inode->i_mapping);
2680 fs_info->pinned_extents = &fs_info->freed_extents[0];
2681 fs_info->do_barriers = 1;
2682
2683
2684 mutex_init(&fs_info->ordered_operations_mutex);
2685 mutex_init(&fs_info->tree_log_mutex);
2686 mutex_init(&fs_info->chunk_mutex);
2687 mutex_init(&fs_info->transaction_kthread_mutex);
2688 mutex_init(&fs_info->cleaner_mutex);
2689 mutex_init(&fs_info->volume_mutex);
2690 mutex_init(&fs_info->ro_block_group_mutex);
2691 init_rwsem(&fs_info->commit_root_sem);
2692 init_rwsem(&fs_info->cleanup_work_sem);
2693 init_rwsem(&fs_info->subvol_sem);
2694 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2695
2696 btrfs_init_dev_replace_locks(fs_info);
2697 btrfs_init_qgroup(fs_info);
2698
2699 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2700 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2701
2702 init_waitqueue_head(&fs_info->transaction_throttle);
2703 init_waitqueue_head(&fs_info->transaction_wait);
2704 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2705 init_waitqueue_head(&fs_info->async_submit_wait);
2706
2707 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2708
2709 ret = btrfs_alloc_stripe_hash_table(fs_info);
2710 if (ret) {
2711 err = ret;
2712 goto fail_alloc;
2713 }
2714
2715 __setup_root(4096, 4096, 4096, tree_root,
2716 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2717
2718 invalidate_bdev(fs_devices->latest_bdev);
2719
2720 /*
2721 * Read super block and check the signature bytes only
2722 */
2723 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2724 if (IS_ERR(bh)) {
2725 err = PTR_ERR(bh);
2726 goto fail_alloc;
2727 }
2728
2729 /*
2730 * We want to check superblock checksum, the type is stored inside.
2731 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2732 */
2733 if (btrfs_check_super_csum(bh->b_data)) {
2734 btrfs_err(fs_info, "superblock checksum mismatch");
2735 err = -EINVAL;
2736 brelse(bh);
2737 goto fail_alloc;
2738 }
2739
2740 /*
2741 * super_copy is zeroed at allocation time and we never touch the
2742 * following bytes up to INFO_SIZE, the checksum is calculated from
2743 * the whole block of INFO_SIZE
2744 */
2745 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2746 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2747 sizeof(*fs_info->super_for_commit));
2748 brelse(bh);
2749
2750 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2751
2752 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2753 if (ret) {
2754 btrfs_err(fs_info, "superblock contains fatal errors");
2755 err = -EINVAL;
2756 goto fail_alloc;
2757 }
2758
2759 disk_super = fs_info->super_copy;
2760 if (!btrfs_super_root(disk_super))
2761 goto fail_alloc;
2762
2763 /* check FS state, whether FS is broken. */
2764 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2765 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2766
2767 /*
2768 * run through our array of backup supers and setup
2769 * our ring pointer to the oldest one
2770 */
2771 generation = btrfs_super_generation(disk_super);
2772 find_oldest_super_backup(fs_info, generation);
2773
2774 /*
2775 * In the long term, we'll store the compression type in the super
2776 * block, and it'll be used for per file compression control.
2777 */
2778 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2779
2780 ret = btrfs_parse_options(tree_root, options, sb->s_flags);
2781 if (ret) {
2782 err = ret;
2783 goto fail_alloc;
2784 }
2785
2786 features = btrfs_super_incompat_flags(disk_super) &
2787 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2788 if (features) {
2789 btrfs_err(fs_info,
2790 "cannot mount because of unsupported optional features (%llx)",
2791 features);
2792 err = -EINVAL;
2793 goto fail_alloc;
2794 }
2795
2796 features = btrfs_super_incompat_flags(disk_super);
2797 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2798 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2799 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2800
2801 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2802 btrfs_info(fs_info, "has skinny extents");
2803
2804 /*
2805 * flag our filesystem as having big metadata blocks if
2806 * they are bigger than the page size
2807 */
2808 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2809 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2810 btrfs_info(fs_info,
2811 "flagging fs with big metadata feature");
2812 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2813 }
2814
2815 nodesize = btrfs_super_nodesize(disk_super);
2816 sectorsize = btrfs_super_sectorsize(disk_super);
2817 stripesize = sectorsize;
2818 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2819 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2820
2821 /*
2822 * mixed block groups end up with duplicate but slightly offset
2823 * extent buffers for the same range. It leads to corruptions
2824 */
2825 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2826 (sectorsize != nodesize)) {
2827 btrfs_err(fs_info,
2828"unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2829 nodesize, sectorsize);
2830 goto fail_alloc;
2831 }
2832
2833 /*
2834 * Needn't use the lock because there is no other task which will
2835 * update the flag.
2836 */
2837 btrfs_set_super_incompat_flags(disk_super, features);
2838
2839 features = btrfs_super_compat_ro_flags(disk_super) &
2840 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2841 if (!(sb->s_flags & MS_RDONLY) && features) {
2842 btrfs_err(fs_info,
2843 "cannot mount read-write because of unsupported optional features (%llx)",
2844 features);
2845 err = -EINVAL;
2846 goto fail_alloc;
2847 }
2848
2849 max_active = fs_info->thread_pool_size;
2850
2851 ret = btrfs_init_workqueues(fs_info, fs_devices);
2852 if (ret) {
2853 err = ret;
2854 goto fail_sb_buffer;
2855 }
2856
2857 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2858 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2859 SZ_4M / PAGE_SIZE);
2860
2861 tree_root->nodesize = nodesize;
2862 tree_root->sectorsize = sectorsize;
2863 tree_root->stripesize = stripesize;
2864
2865 sb->s_blocksize = sectorsize;
2866 sb->s_blocksize_bits = blksize_bits(sectorsize);
2867
2868 mutex_lock(&fs_info->chunk_mutex);
2869 ret = btrfs_read_sys_array(tree_root);
2870 mutex_unlock(&fs_info->chunk_mutex);
2871 if (ret) {
2872 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2873 goto fail_sb_buffer;
2874 }
2875
2876 generation = btrfs_super_chunk_root_generation(disk_super);
2877
2878 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2879 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2880
2881 chunk_root->node = read_tree_block(chunk_root,
2882 btrfs_super_chunk_root(disk_super),
2883 generation);
2884 if (IS_ERR(chunk_root->node) ||
2885 !extent_buffer_uptodate(chunk_root->node)) {
2886 btrfs_err(fs_info, "failed to read chunk root");
2887 if (!IS_ERR(chunk_root->node))
2888 free_extent_buffer(chunk_root->node);
2889 chunk_root->node = NULL;
2890 goto fail_tree_roots;
2891 }
2892 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2893 chunk_root->commit_root = btrfs_root_node(chunk_root);
2894
2895 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2896 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2897
2898 ret = btrfs_read_chunk_tree(chunk_root);
2899 if (ret) {
2900 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2901 goto fail_tree_roots;
2902 }
2903
2904 /*
2905 * keep the device that is marked to be the target device for the
2906 * dev_replace procedure
2907 */
2908 btrfs_close_extra_devices(fs_devices, 0);
2909
2910 if (!fs_devices->latest_bdev) {
2911 btrfs_err(fs_info, "failed to read devices");
2912 goto fail_tree_roots;
2913 }
2914
2915retry_root_backup:
2916 generation = btrfs_super_generation(disk_super);
2917
2918 tree_root->node = read_tree_block(tree_root,
2919 btrfs_super_root(disk_super),
2920 generation);
2921 if (IS_ERR(tree_root->node) ||
2922 !extent_buffer_uptodate(tree_root->node)) {
2923 btrfs_warn(fs_info, "failed to read tree root");
2924 if (!IS_ERR(tree_root->node))
2925 free_extent_buffer(tree_root->node);
2926 tree_root->node = NULL;
2927 goto recovery_tree_root;
2928 }
2929
2930 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2931 tree_root->commit_root = btrfs_root_node(tree_root);
2932 btrfs_set_root_refs(&tree_root->root_item, 1);
2933
2934 mutex_lock(&tree_root->objectid_mutex);
2935 ret = btrfs_find_highest_objectid(tree_root,
2936 &tree_root->highest_objectid);
2937 if (ret) {
2938 mutex_unlock(&tree_root->objectid_mutex);
2939 goto recovery_tree_root;
2940 }
2941
2942 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2943
2944 mutex_unlock(&tree_root->objectid_mutex);
2945
2946 ret = btrfs_read_roots(fs_info, tree_root);
2947 if (ret)
2948 goto recovery_tree_root;
2949
2950 fs_info->generation = generation;
2951 fs_info->last_trans_committed = generation;
2952
2953 ret = btrfs_recover_balance(fs_info);
2954 if (ret) {
2955 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2956 goto fail_block_groups;
2957 }
2958
2959 ret = btrfs_init_dev_stats(fs_info);
2960 if (ret) {
2961 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2962 goto fail_block_groups;
2963 }
2964
2965 ret = btrfs_init_dev_replace(fs_info);
2966 if (ret) {
2967 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2968 goto fail_block_groups;
2969 }
2970
2971 btrfs_close_extra_devices(fs_devices, 1);
2972
2973 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2974 if (ret) {
2975 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2976 ret);
2977 goto fail_block_groups;
2978 }
2979
2980 ret = btrfs_sysfs_add_device(fs_devices);
2981 if (ret) {
2982 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2983 ret);
2984 goto fail_fsdev_sysfs;
2985 }
2986
2987 ret = btrfs_sysfs_add_mounted(fs_info);
2988 if (ret) {
2989 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2990 goto fail_fsdev_sysfs;
2991 }
2992
2993 ret = btrfs_init_space_info(fs_info);
2994 if (ret) {
2995 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
2996 goto fail_sysfs;
2997 }
2998
2999 ret = btrfs_read_block_groups(fs_info->extent_root);
3000 if (ret) {
3001 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3002 goto fail_sysfs;
3003 }
3004 fs_info->num_tolerated_disk_barrier_failures =
3005 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3006 if (fs_info->fs_devices->missing_devices >
3007 fs_info->num_tolerated_disk_barrier_failures &&
3008 !(sb->s_flags & MS_RDONLY)) {
3009 btrfs_warn(fs_info,
3010"missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3011 fs_info->fs_devices->missing_devices,
3012 fs_info->num_tolerated_disk_barrier_failures);
3013 goto fail_sysfs;
3014 }
3015
3016 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3017 "btrfs-cleaner");
3018 if (IS_ERR(fs_info->cleaner_kthread))
3019 goto fail_sysfs;
3020
3021 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3022 tree_root,
3023 "btrfs-transaction");
3024 if (IS_ERR(fs_info->transaction_kthread))
3025 goto fail_cleaner;
3026
3027 if (!btrfs_test_opt(tree_root->fs_info, SSD) &&
3028 !btrfs_test_opt(tree_root->fs_info, NOSSD) &&
3029 !fs_info->fs_devices->rotating) {
3030 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3031 btrfs_set_opt(fs_info->mount_opt, SSD);
3032 }
3033
3034 /*
3035 * Mount does not set all options immediately, we can do it now and do
3036 * not have to wait for transaction commit
3037 */
3038 btrfs_apply_pending_changes(fs_info);
3039
3040#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3041 if (btrfs_test_opt(tree_root->fs_info, CHECK_INTEGRITY)) {
3042 ret = btrfsic_mount(tree_root, fs_devices,
3043 btrfs_test_opt(tree_root->fs_info,
3044 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3045 1 : 0,
3046 fs_info->check_integrity_print_mask);
3047 if (ret)
3048 btrfs_warn(fs_info,
3049 "failed to initialize integrity check module: %d",
3050 ret);
3051 }
3052#endif
3053 ret = btrfs_read_qgroup_config(fs_info);
3054 if (ret)
3055 goto fail_trans_kthread;
3056
3057 /* do not make disk changes in broken FS or nologreplay is given */
3058 if (btrfs_super_log_root(disk_super) != 0 &&
3059 !btrfs_test_opt(tree_root->fs_info, NOLOGREPLAY)) {
3060 ret = btrfs_replay_log(fs_info, fs_devices);
3061 if (ret) {
3062 err = ret;
3063 goto fail_qgroup;
3064 }
3065 }
3066
3067 ret = btrfs_find_orphan_roots(tree_root);
3068 if (ret)
3069 goto fail_qgroup;
3070
3071 if (!(sb->s_flags & MS_RDONLY)) {
3072 ret = btrfs_cleanup_fs_roots(fs_info);
3073 if (ret)
3074 goto fail_qgroup;
3075
3076 mutex_lock(&fs_info->cleaner_mutex);
3077 ret = btrfs_recover_relocation(tree_root);
3078 mutex_unlock(&fs_info->cleaner_mutex);
3079 if (ret < 0) {
3080 btrfs_warn(fs_info, "failed to recover relocation: %d",
3081 ret);
3082 err = -EINVAL;
3083 goto fail_qgroup;
3084 }
3085 }
3086
3087 location.objectid = BTRFS_FS_TREE_OBJECTID;
3088 location.type = BTRFS_ROOT_ITEM_KEY;
3089 location.offset = 0;
3090
3091 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3092 if (IS_ERR(fs_info->fs_root)) {
3093 err = PTR_ERR(fs_info->fs_root);
3094 goto fail_qgroup;
3095 }
3096
3097 if (sb->s_flags & MS_RDONLY)
3098 return 0;
3099
3100 if (btrfs_test_opt(tree_root->fs_info, FREE_SPACE_TREE) &&
3101 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3102 btrfs_info(fs_info, "creating free space tree");
3103 ret = btrfs_create_free_space_tree(fs_info);
3104 if (ret) {
3105 btrfs_warn(fs_info,
3106 "failed to create free space tree: %d", ret);
3107 close_ctree(tree_root);
3108 return ret;
3109 }
3110 }
3111
3112 down_read(&fs_info->cleanup_work_sem);
3113 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3114 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3115 up_read(&fs_info->cleanup_work_sem);
3116 close_ctree(tree_root);
3117 return ret;
3118 }
3119 up_read(&fs_info->cleanup_work_sem);
3120
3121 ret = btrfs_resume_balance_async(fs_info);
3122 if (ret) {
3123 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3124 close_ctree(tree_root);
3125 return ret;
3126 }
3127
3128 ret = btrfs_resume_dev_replace_async(fs_info);
3129 if (ret) {
3130 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3131 close_ctree(tree_root);
3132 return ret;
3133 }
3134
3135 btrfs_qgroup_rescan_resume(fs_info);
3136
3137 if (btrfs_test_opt(tree_root->fs_info, CLEAR_CACHE) &&
3138 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3139 btrfs_info(fs_info, "clearing free space tree");
3140 ret = btrfs_clear_free_space_tree(fs_info);
3141 if (ret) {
3142 btrfs_warn(fs_info,
3143 "failed to clear free space tree: %d", ret);
3144 close_ctree(tree_root);
3145 return ret;
3146 }
3147 }
3148
3149 if (!fs_info->uuid_root) {
3150 btrfs_info(fs_info, "creating UUID tree");
3151 ret = btrfs_create_uuid_tree(fs_info);
3152 if (ret) {
3153 btrfs_warn(fs_info,
3154 "failed to create the UUID tree: %d", ret);
3155 close_ctree(tree_root);
3156 return ret;
3157 }
3158 } else if (btrfs_test_opt(tree_root->fs_info, RESCAN_UUID_TREE) ||
3159 fs_info->generation !=
3160 btrfs_super_uuid_tree_generation(disk_super)) {
3161 btrfs_info(fs_info, "checking UUID tree");
3162 ret = btrfs_check_uuid_tree(fs_info);
3163 if (ret) {
3164 btrfs_warn(fs_info,
3165 "failed to check the UUID tree: %d", ret);
3166 close_ctree(tree_root);
3167 return ret;
3168 }
3169 } else {
3170 fs_info->update_uuid_tree_gen = 1;
3171 }
3172
3173 fs_info->open = 1;
3174
3175 /*
3176 * backuproot only affect mount behavior, and if open_ctree succeeded,
3177 * no need to keep the flag
3178 */
3179 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3180
3181 return 0;
3182
3183fail_qgroup:
3184 btrfs_free_qgroup_config(fs_info);
3185fail_trans_kthread:
3186 kthread_stop(fs_info->transaction_kthread);
3187 btrfs_cleanup_transaction(fs_info->tree_root);
3188 btrfs_free_fs_roots(fs_info);
3189fail_cleaner:
3190 kthread_stop(fs_info->cleaner_kthread);
3191
3192 /*
3193 * make sure we're done with the btree inode before we stop our
3194 * kthreads
3195 */
3196 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3197
3198fail_sysfs:
3199 btrfs_sysfs_remove_mounted(fs_info);
3200
3201fail_fsdev_sysfs:
3202 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3203
3204fail_block_groups:
3205 btrfs_put_block_group_cache(fs_info);
3206 btrfs_free_block_groups(fs_info);
3207
3208fail_tree_roots:
3209 free_root_pointers(fs_info, 1);
3210 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3211
3212fail_sb_buffer:
3213 btrfs_stop_all_workers(fs_info);
3214fail_alloc:
3215fail_iput:
3216 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3217
3218 iput(fs_info->btree_inode);
3219fail_bio_counter:
3220 percpu_counter_destroy(&fs_info->bio_counter);
3221fail_delalloc_bytes:
3222 percpu_counter_destroy(&fs_info->delalloc_bytes);
3223fail_dirty_metadata_bytes:
3224 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3225fail_bdi:
3226 bdi_destroy(&fs_info->bdi);
3227fail_srcu:
3228 cleanup_srcu_struct(&fs_info->subvol_srcu);
3229fail:
3230 btrfs_free_stripe_hash_table(fs_info);
3231 btrfs_close_devices(fs_info->fs_devices);
3232 return err;
3233
3234recovery_tree_root:
3235 if (!btrfs_test_opt(tree_root->fs_info, USEBACKUPROOT))
3236 goto fail_tree_roots;
3237
3238 free_root_pointers(fs_info, 0);
3239
3240 /* don't use the log in recovery mode, it won't be valid */
3241 btrfs_set_super_log_root(disk_super, 0);
3242
3243 /* we can't trust the free space cache either */
3244 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3245
3246 ret = next_root_backup(fs_info, fs_info->super_copy,
3247 &num_backups_tried, &backup_index);
3248 if (ret == -1)
3249 goto fail_block_groups;
3250 goto retry_root_backup;
3251}
3252
3253static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3254{
3255 if (uptodate) {
3256 set_buffer_uptodate(bh);
3257 } else {
3258 struct btrfs_device *device = (struct btrfs_device *)
3259 bh->b_private;
3260
3261 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3262 "lost page write due to IO error on %s",
3263 rcu_str_deref(device->name));
3264 /* note, we don't set_buffer_write_io_error because we have
3265 * our own ways of dealing with the IO errors
3266 */
3267 clear_buffer_uptodate(bh);
3268 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3269 }
3270 unlock_buffer(bh);
3271 put_bh(bh);
3272}
3273
3274int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3275 struct buffer_head **bh_ret)
3276{
3277 struct buffer_head *bh;
3278 struct btrfs_super_block *super;
3279 u64 bytenr;
3280
3281 bytenr = btrfs_sb_offset(copy_num);
3282 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3283 return -EINVAL;
3284
3285 bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3286 /*
3287 * If we fail to read from the underlying devices, as of now
3288 * the best option we have is to mark it EIO.
3289 */
3290 if (!bh)
3291 return -EIO;
3292
3293 super = (struct btrfs_super_block *)bh->b_data;
3294 if (btrfs_super_bytenr(super) != bytenr ||
3295 btrfs_super_magic(super) != BTRFS_MAGIC) {
3296 brelse(bh);
3297 return -EINVAL;
3298 }
3299
3300 *bh_ret = bh;
3301 return 0;
3302}
3303
3304
3305struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3306{
3307 struct buffer_head *bh;
3308 struct buffer_head *latest = NULL;
3309 struct btrfs_super_block *super;
3310 int i;
3311 u64 transid = 0;
3312 int ret = -EINVAL;
3313
3314 /* we would like to check all the supers, but that would make
3315 * a btrfs mount succeed after a mkfs from a different FS.
3316 * So, we need to add a special mount option to scan for
3317 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3318 */
3319 for (i = 0; i < 1; i++) {
3320 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3321 if (ret)
3322 continue;
3323
3324 super = (struct btrfs_super_block *)bh->b_data;
3325
3326 if (!latest || btrfs_super_generation(super) > transid) {
3327 brelse(latest);
3328 latest = bh;
3329 transid = btrfs_super_generation(super);
3330 } else {
3331 brelse(bh);
3332 }
3333 }
3334
3335 if (!latest)
3336 return ERR_PTR(ret);
3337
3338 return latest;
3339}
3340
3341/*
3342 * this should be called twice, once with wait == 0 and
3343 * once with wait == 1. When wait == 0 is done, all the buffer heads
3344 * we write are pinned.
3345 *
3346 * They are released when wait == 1 is done.
3347 * max_mirrors must be the same for both runs, and it indicates how
3348 * many supers on this one device should be written.
3349 *
3350 * max_mirrors == 0 means to write them all.
3351 */
3352static int write_dev_supers(struct btrfs_device *device,
3353 struct btrfs_super_block *sb,
3354 int do_barriers, int wait, int max_mirrors)
3355{
3356 struct buffer_head *bh;
3357 int i;
3358 int ret;
3359 int errors = 0;
3360 u32 crc;
3361 u64 bytenr;
3362
3363 if (max_mirrors == 0)
3364 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3365
3366 for (i = 0; i < max_mirrors; i++) {
3367 bytenr = btrfs_sb_offset(i);
3368 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3369 device->commit_total_bytes)
3370 break;
3371
3372 if (wait) {
3373 bh = __find_get_block(device->bdev, bytenr / 4096,
3374 BTRFS_SUPER_INFO_SIZE);
3375 if (!bh) {
3376 errors++;
3377 continue;
3378 }
3379 wait_on_buffer(bh);
3380 if (!buffer_uptodate(bh))
3381 errors++;
3382
3383 /* drop our reference */
3384 brelse(bh);
3385
3386 /* drop the reference from the wait == 0 run */
3387 brelse(bh);
3388 continue;
3389 } else {
3390 btrfs_set_super_bytenr(sb, bytenr);
3391
3392 crc = ~(u32)0;
3393 crc = btrfs_csum_data((char *)sb +
3394 BTRFS_CSUM_SIZE, crc,
3395 BTRFS_SUPER_INFO_SIZE -
3396 BTRFS_CSUM_SIZE);
3397 btrfs_csum_final(crc, sb->csum);
3398
3399 /*
3400 * one reference for us, and we leave it for the
3401 * caller
3402 */
3403 bh = __getblk(device->bdev, bytenr / 4096,
3404 BTRFS_SUPER_INFO_SIZE);
3405 if (!bh) {
3406 btrfs_err(device->dev_root->fs_info,
3407 "couldn't get super buffer head for bytenr %llu",
3408 bytenr);
3409 errors++;
3410 continue;
3411 }
3412
3413 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3414
3415 /* one reference for submit_bh */
3416 get_bh(bh);
3417
3418 set_buffer_uptodate(bh);
3419 lock_buffer(bh);
3420 bh->b_end_io = btrfs_end_buffer_write_sync;
3421 bh->b_private = device;
3422 }
3423
3424 /*
3425 * we fua the first super. The others we allow
3426 * to go down lazy.
3427 */
3428 if (i == 0)
3429 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3430 else
3431 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3432 if (ret)
3433 errors++;
3434 }
3435 return errors < i ? 0 : -1;
3436}
3437
3438/*
3439 * endio for the write_dev_flush, this will wake anyone waiting
3440 * for the barrier when it is done
3441 */
3442static void btrfs_end_empty_barrier(struct bio *bio)
3443{
3444 if (bio->bi_private)
3445 complete(bio->bi_private);
3446 bio_put(bio);
3447}
3448
3449/*
3450 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3451 * sent down. With wait == 1, it waits for the previous flush.
3452 *
3453 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3454 * capable
3455 */
3456static int write_dev_flush(struct btrfs_device *device, int wait)
3457{
3458 struct bio *bio;
3459 int ret = 0;
3460
3461 if (device->nobarriers)
3462 return 0;
3463
3464 if (wait) {
3465 bio = device->flush_bio;
3466 if (!bio)
3467 return 0;
3468
3469 wait_for_completion(&device->flush_wait);
3470
3471 if (bio->bi_error) {
3472 ret = bio->bi_error;
3473 btrfs_dev_stat_inc_and_print(device,
3474 BTRFS_DEV_STAT_FLUSH_ERRS);
3475 }
3476
3477 /* drop the reference from the wait == 0 run */
3478 bio_put(bio);
3479 device->flush_bio = NULL;
3480
3481 return ret;
3482 }
3483
3484 /*
3485 * one reference for us, and we leave it for the
3486 * caller
3487 */
3488 device->flush_bio = NULL;
3489 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3490 if (!bio)
3491 return -ENOMEM;
3492
3493 bio->bi_end_io = btrfs_end_empty_barrier;
3494 bio->bi_bdev = device->bdev;
3495 init_completion(&device->flush_wait);
3496 bio->bi_private = &device->flush_wait;
3497 device->flush_bio = bio;
3498
3499 bio_get(bio);
3500 btrfsic_submit_bio(WRITE_FLUSH, bio);
3501
3502 return 0;
3503}
3504
3505/*
3506 * send an empty flush down to each device in parallel,
3507 * then wait for them
3508 */
3509static int barrier_all_devices(struct btrfs_fs_info *info)
3510{
3511 struct list_head *head;
3512 struct btrfs_device *dev;
3513 int errors_send = 0;
3514 int errors_wait = 0;
3515 int ret;
3516
3517 /* send down all the barriers */
3518 head = &info->fs_devices->devices;
3519 list_for_each_entry_rcu(dev, head, dev_list) {
3520 if (dev->missing)
3521 continue;
3522 if (!dev->bdev) {
3523 errors_send++;
3524 continue;
3525 }
3526 if (!dev->in_fs_metadata || !dev->writeable)
3527 continue;
3528
3529 ret = write_dev_flush(dev, 0);
3530 if (ret)
3531 errors_send++;
3532 }
3533
3534 /* wait for all the barriers */
3535 list_for_each_entry_rcu(dev, head, dev_list) {
3536 if (dev->missing)
3537 continue;
3538 if (!dev->bdev) {
3539 errors_wait++;
3540 continue;
3541 }
3542 if (!dev->in_fs_metadata || !dev->writeable)
3543 continue;
3544
3545 ret = write_dev_flush(dev, 1);
3546 if (ret)
3547 errors_wait++;
3548 }
3549 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3550 errors_wait > info->num_tolerated_disk_barrier_failures)
3551 return -EIO;
3552 return 0;
3553}
3554
3555int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3556{
3557 int raid_type;
3558 int min_tolerated = INT_MAX;
3559
3560 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3561 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3562 min_tolerated = min(min_tolerated,
3563 btrfs_raid_array[BTRFS_RAID_SINGLE].
3564 tolerated_failures);
3565
3566 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3567 if (raid_type == BTRFS_RAID_SINGLE)
3568 continue;
3569 if (!(flags & btrfs_raid_group[raid_type]))
3570 continue;
3571 min_tolerated = min(min_tolerated,
3572 btrfs_raid_array[raid_type].
3573 tolerated_failures);
3574 }
3575
3576 if (min_tolerated == INT_MAX) {
3577 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3578 min_tolerated = 0;
3579 }
3580
3581 return min_tolerated;
3582}
3583
3584int btrfs_calc_num_tolerated_disk_barrier_failures(
3585 struct btrfs_fs_info *fs_info)
3586{
3587 struct btrfs_ioctl_space_info space;
3588 struct btrfs_space_info *sinfo;
3589 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3590 BTRFS_BLOCK_GROUP_SYSTEM,
3591 BTRFS_BLOCK_GROUP_METADATA,
3592 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3593 int i;
3594 int c;
3595 int num_tolerated_disk_barrier_failures =
3596 (int)fs_info->fs_devices->num_devices;
3597
3598 for (i = 0; i < ARRAY_SIZE(types); i++) {
3599 struct btrfs_space_info *tmp;
3600
3601 sinfo = NULL;
3602 rcu_read_lock();
3603 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3604 if (tmp->flags == types[i]) {
3605 sinfo = tmp;
3606 break;
3607 }
3608 }
3609 rcu_read_unlock();
3610
3611 if (!sinfo)
3612 continue;
3613
3614 down_read(&sinfo->groups_sem);
3615 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3616 u64 flags;
3617
3618 if (list_empty(&sinfo->block_groups[c]))
3619 continue;
3620
3621 btrfs_get_block_group_info(&sinfo->block_groups[c],
3622 &space);
3623 if (space.total_bytes == 0 || space.used_bytes == 0)
3624 continue;
3625 flags = space.flags;
3626
3627 num_tolerated_disk_barrier_failures = min(
3628 num_tolerated_disk_barrier_failures,
3629 btrfs_get_num_tolerated_disk_barrier_failures(
3630 flags));
3631 }
3632 up_read(&sinfo->groups_sem);
3633 }
3634
3635 return num_tolerated_disk_barrier_failures;
3636}
3637
3638static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3639{
3640 struct list_head *head;
3641 struct btrfs_device *dev;
3642 struct btrfs_super_block *sb;
3643 struct btrfs_dev_item *dev_item;
3644 int ret;
3645 int do_barriers;
3646 int max_errors;
3647 int total_errors = 0;
3648 u64 flags;
3649
3650 do_barriers = !btrfs_test_opt(root->fs_info, NOBARRIER);
3651 backup_super_roots(root->fs_info);
3652
3653 sb = root->fs_info->super_for_commit;
3654 dev_item = &sb->dev_item;
3655
3656 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3657 head = &root->fs_info->fs_devices->devices;
3658 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3659
3660 if (do_barriers) {
3661 ret = barrier_all_devices(root->fs_info);
3662 if (ret) {
3663 mutex_unlock(
3664 &root->fs_info->fs_devices->device_list_mutex);
3665 btrfs_handle_fs_error(root->fs_info, ret,
3666 "errors while submitting device barriers.");
3667 return ret;
3668 }
3669 }
3670
3671 list_for_each_entry_rcu(dev, head, dev_list) {
3672 if (!dev->bdev) {
3673 total_errors++;
3674 continue;
3675 }
3676 if (!dev->in_fs_metadata || !dev->writeable)
3677 continue;
3678
3679 btrfs_set_stack_device_generation(dev_item, 0);
3680 btrfs_set_stack_device_type(dev_item, dev->type);
3681 btrfs_set_stack_device_id(dev_item, dev->devid);
3682 btrfs_set_stack_device_total_bytes(dev_item,
3683 dev->commit_total_bytes);
3684 btrfs_set_stack_device_bytes_used(dev_item,
3685 dev->commit_bytes_used);
3686 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3687 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3688 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3689 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3690 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3691
3692 flags = btrfs_super_flags(sb);
3693 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3694
3695 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3696 if (ret)
3697 total_errors++;
3698 }
3699 if (total_errors > max_errors) {
3700 btrfs_err(root->fs_info, "%d errors while writing supers",
3701 total_errors);
3702 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3703
3704 /* FUA is masked off if unsupported and can't be the reason */
3705 btrfs_handle_fs_error(root->fs_info, -EIO,
3706 "%d errors while writing supers", total_errors);
3707 return -EIO;
3708 }
3709
3710 total_errors = 0;
3711 list_for_each_entry_rcu(dev, head, dev_list) {
3712 if (!dev->bdev)
3713 continue;
3714 if (!dev->in_fs_metadata || !dev->writeable)
3715 continue;
3716
3717 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3718 if (ret)
3719 total_errors++;
3720 }
3721 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3722 if (total_errors > max_errors) {
3723 btrfs_handle_fs_error(root->fs_info, -EIO,
3724 "%d errors while writing supers", total_errors);
3725 return -EIO;
3726 }
3727 return 0;
3728}
3729
3730int write_ctree_super(struct btrfs_trans_handle *trans,
3731 struct btrfs_root *root, int max_mirrors)
3732{
3733 return write_all_supers(root, max_mirrors);
3734}
3735
3736/* Drop a fs root from the radix tree and free it. */
3737void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3738 struct btrfs_root *root)
3739{
3740 spin_lock(&fs_info->fs_roots_radix_lock);
3741 radix_tree_delete(&fs_info->fs_roots_radix,
3742 (unsigned long)root->root_key.objectid);
3743 spin_unlock(&fs_info->fs_roots_radix_lock);
3744
3745 if (btrfs_root_refs(&root->root_item) == 0)
3746 synchronize_srcu(&fs_info->subvol_srcu);
3747
3748 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3749 btrfs_free_log(NULL, root);
3750 if (root->reloc_root) {
3751 free_extent_buffer(root->reloc_root->node);
3752 free_extent_buffer(root->reloc_root->commit_root);
3753 btrfs_put_fs_root(root->reloc_root);
3754 root->reloc_root = NULL;
3755 }
3756 }
3757
3758 if (root->free_ino_pinned)
3759 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3760 if (root->free_ino_ctl)
3761 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3762 free_fs_root(root);
3763}
3764
3765static void free_fs_root(struct btrfs_root *root)
3766{
3767 iput(root->ino_cache_inode);
3768 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3769 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3770 root->orphan_block_rsv = NULL;
3771 if (root->anon_dev)
3772 free_anon_bdev(root->anon_dev);
3773 if (root->subv_writers)
3774 btrfs_free_subvolume_writers(root->subv_writers);
3775 free_extent_buffer(root->node);
3776 free_extent_buffer(root->commit_root);
3777 kfree(root->free_ino_ctl);
3778 kfree(root->free_ino_pinned);
3779 kfree(root->name);
3780 btrfs_put_fs_root(root);
3781}
3782
3783void btrfs_free_fs_root(struct btrfs_root *root)
3784{
3785 free_fs_root(root);
3786}
3787
3788int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3789{
3790 u64 root_objectid = 0;
3791 struct btrfs_root *gang[8];
3792 int i = 0;
3793 int err = 0;
3794 unsigned int ret = 0;
3795 int index;
3796
3797 while (1) {
3798 index = srcu_read_lock(&fs_info->subvol_srcu);
3799 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3800 (void **)gang, root_objectid,
3801 ARRAY_SIZE(gang));
3802 if (!ret) {
3803 srcu_read_unlock(&fs_info->subvol_srcu, index);
3804 break;
3805 }
3806 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3807
3808 for (i = 0; i < ret; i++) {
3809 /* Avoid to grab roots in dead_roots */
3810 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3811 gang[i] = NULL;
3812 continue;
3813 }
3814 /* grab all the search result for later use */
3815 gang[i] = btrfs_grab_fs_root(gang[i]);
3816 }
3817 srcu_read_unlock(&fs_info->subvol_srcu, index);
3818
3819 for (i = 0; i < ret; i++) {
3820 if (!gang[i])
3821 continue;
3822 root_objectid = gang[i]->root_key.objectid;
3823 err = btrfs_orphan_cleanup(gang[i]);
3824 if (err)
3825 break;
3826 btrfs_put_fs_root(gang[i]);
3827 }
3828 root_objectid++;
3829 }
3830
3831 /* release the uncleaned roots due to error */
3832 for (; i < ret; i++) {
3833 if (gang[i])
3834 btrfs_put_fs_root(gang[i]);
3835 }
3836 return err;
3837}
3838
3839int btrfs_commit_super(struct btrfs_root *root)
3840{
3841 struct btrfs_trans_handle *trans;
3842
3843 mutex_lock(&root->fs_info->cleaner_mutex);
3844 btrfs_run_delayed_iputs(root);
3845 mutex_unlock(&root->fs_info->cleaner_mutex);
3846 wake_up_process(root->fs_info->cleaner_kthread);
3847
3848 /* wait until ongoing cleanup work done */
3849 down_write(&root->fs_info->cleanup_work_sem);
3850 up_write(&root->fs_info->cleanup_work_sem);
3851
3852 trans = btrfs_join_transaction(root);
3853 if (IS_ERR(trans))
3854 return PTR_ERR(trans);
3855 return btrfs_commit_transaction(trans, root);
3856}
3857
3858void close_ctree(struct btrfs_root *root)
3859{
3860 struct btrfs_fs_info *fs_info = root->fs_info;
3861 int ret;
3862
3863 fs_info->closing = 1;
3864 smp_mb();
3865
3866 /* wait for the qgroup rescan worker to stop */
3867 btrfs_qgroup_wait_for_completion(fs_info);
3868
3869 /* wait for the uuid_scan task to finish */
3870 down(&fs_info->uuid_tree_rescan_sem);
3871 /* avoid complains from lockdep et al., set sem back to initial state */
3872 up(&fs_info->uuid_tree_rescan_sem);
3873
3874 /* pause restriper - we want to resume on mount */
3875 btrfs_pause_balance(fs_info);
3876
3877 btrfs_dev_replace_suspend_for_unmount(fs_info);
3878
3879 btrfs_scrub_cancel(fs_info);
3880
3881 /* wait for any defraggers to finish */
3882 wait_event(fs_info->transaction_wait,
3883 (atomic_read(&fs_info->defrag_running) == 0));
3884
3885 /* clear out the rbtree of defraggable inodes */
3886 btrfs_cleanup_defrag_inodes(fs_info);
3887
3888 cancel_work_sync(&fs_info->async_reclaim_work);
3889
3890 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3891 /*
3892 * If the cleaner thread is stopped and there are
3893 * block groups queued for removal, the deletion will be
3894 * skipped when we quit the cleaner thread.
3895 */
3896 btrfs_delete_unused_bgs(root->fs_info);
3897
3898 ret = btrfs_commit_super(root);
3899 if (ret)
3900 btrfs_err(fs_info, "commit super ret %d", ret);
3901 }
3902
3903 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3904 btrfs_error_commit_super(root);
3905
3906 kthread_stop(fs_info->transaction_kthread);
3907 kthread_stop(fs_info->cleaner_kthread);
3908
3909 fs_info->closing = 2;
3910 smp_mb();
3911
3912 btrfs_free_qgroup_config(fs_info);
3913
3914 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3915 btrfs_info(fs_info, "at unmount delalloc count %lld",
3916 percpu_counter_sum(&fs_info->delalloc_bytes));
3917 }
3918
3919 btrfs_sysfs_remove_mounted(fs_info);
3920 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3921
3922 btrfs_free_fs_roots(fs_info);
3923
3924 btrfs_put_block_group_cache(fs_info);
3925
3926 btrfs_free_block_groups(fs_info);
3927
3928 /*
3929 * we must make sure there is not any read request to
3930 * submit after we stopping all workers.
3931 */
3932 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3933 btrfs_stop_all_workers(fs_info);
3934
3935 fs_info->open = 0;
3936 free_root_pointers(fs_info, 1);
3937
3938 iput(fs_info->btree_inode);
3939
3940#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3941 if (btrfs_test_opt(root->fs_info, CHECK_INTEGRITY))
3942 btrfsic_unmount(root, fs_info->fs_devices);
3943#endif
3944
3945 btrfs_close_devices(fs_info->fs_devices);
3946 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3947
3948 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3949 percpu_counter_destroy(&fs_info->delalloc_bytes);
3950 percpu_counter_destroy(&fs_info->bio_counter);
3951 bdi_destroy(&fs_info->bdi);
3952 cleanup_srcu_struct(&fs_info->subvol_srcu);
3953
3954 btrfs_free_stripe_hash_table(fs_info);
3955
3956 __btrfs_free_block_rsv(root->orphan_block_rsv);
3957 root->orphan_block_rsv = NULL;
3958
3959 lock_chunks(root);
3960 while (!list_empty(&fs_info->pinned_chunks)) {
3961 struct extent_map *em;
3962
3963 em = list_first_entry(&fs_info->pinned_chunks,
3964 struct extent_map, list);
3965 list_del_init(&em->list);
3966 free_extent_map(em);
3967 }
3968 unlock_chunks(root);
3969}
3970
3971int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3972 int atomic)
3973{
3974 int ret;
3975 struct inode *btree_inode = buf->pages[0]->mapping->host;
3976
3977 ret = extent_buffer_uptodate(buf);
3978 if (!ret)
3979 return ret;
3980
3981 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3982 parent_transid, atomic);
3983 if (ret == -EAGAIN)
3984 return ret;
3985 return !ret;
3986}
3987
3988void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3989{
3990 struct btrfs_root *root;
3991 u64 transid = btrfs_header_generation(buf);
3992 int was_dirty;
3993
3994#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3995 /*
3996 * This is a fast path so only do this check if we have sanity tests
3997 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3998 * outside of the sanity tests.
3999 */
4000 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4001 return;
4002#endif
4003 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4004 btrfs_assert_tree_locked(buf);
4005 if (transid != root->fs_info->generation)
4006 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
4007 "found %llu running %llu\n",
4008 buf->start, transid, root->fs_info->generation);
4009 was_dirty = set_extent_buffer_dirty(buf);
4010 if (!was_dirty)
4011 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
4012 buf->len,
4013 root->fs_info->dirty_metadata_batch);
4014#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4015 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4016 btrfs_print_leaf(root, buf);
4017 ASSERT(0);
4018 }
4019#endif
4020}
4021
4022static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
4023 int flush_delayed)
4024{
4025 /*
4026 * looks as though older kernels can get into trouble with
4027 * this code, they end up stuck in balance_dirty_pages forever
4028 */
4029 int ret;
4030
4031 if (current->flags & PF_MEMALLOC)
4032 return;
4033
4034 if (flush_delayed)
4035 btrfs_balance_delayed_items(root);
4036
4037 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
4038 BTRFS_DIRTY_METADATA_THRESH);
4039 if (ret > 0) {
4040 balance_dirty_pages_ratelimited(
4041 root->fs_info->btree_inode->i_mapping);
4042 }
4043}
4044
4045void btrfs_btree_balance_dirty(struct btrfs_root *root)
4046{
4047 __btrfs_btree_balance_dirty(root, 1);
4048}
4049
4050void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4051{
4052 __btrfs_btree_balance_dirty(root, 0);
4053}
4054
4055int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4056{
4057 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4058 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4059}
4060
4061static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4062 int read_only)
4063{
4064 struct btrfs_super_block *sb = fs_info->super_copy;
4065 u64 nodesize = btrfs_super_nodesize(sb);
4066 u64 sectorsize = btrfs_super_sectorsize(sb);
4067 int ret = 0;
4068
4069 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4070 printk(KERN_ERR "BTRFS: no valid FS found\n");
4071 ret = -EINVAL;
4072 }
4073 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4074 printk(KERN_WARNING "BTRFS: unrecognized super flag: %llu\n",
4075 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4076 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4077 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4078 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4079 ret = -EINVAL;
4080 }
4081 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4082 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4083 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4084 ret = -EINVAL;
4085 }
4086 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4087 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4088 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4089 ret = -EINVAL;
4090 }
4091
4092 /*
4093 * Check sectorsize and nodesize first, other check will need it.
4094 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4095 */
4096 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4097 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4098 printk(KERN_ERR "BTRFS: invalid sectorsize %llu\n", sectorsize);
4099 ret = -EINVAL;
4100 }
4101 /* Only PAGE SIZE is supported yet */
4102 if (sectorsize != PAGE_SIZE) {
4103 printk(KERN_ERR "BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4104 sectorsize, PAGE_SIZE);
4105 ret = -EINVAL;
4106 }
4107 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4108 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4109 printk(KERN_ERR "BTRFS: invalid nodesize %llu\n", nodesize);
4110 ret = -EINVAL;
4111 }
4112 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4113 printk(KERN_ERR "BTRFS: invalid leafsize %u, should be %llu\n",
4114 le32_to_cpu(sb->__unused_leafsize),
4115 nodesize);
4116 ret = -EINVAL;
4117 }
4118
4119 /* Root alignment check */
4120 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4121 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4122 btrfs_super_root(sb));
4123 ret = -EINVAL;
4124 }
4125 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4126 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4127 btrfs_super_chunk_root(sb));
4128 ret = -EINVAL;
4129 }
4130 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4131 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4132 btrfs_super_log_root(sb));
4133 ret = -EINVAL;
4134 }
4135
4136 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4137 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4138 fs_info->fsid, sb->dev_item.fsid);
4139 ret = -EINVAL;
4140 }
4141
4142 /*
4143 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4144 * done later
4145 */
4146 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4147 btrfs_err(fs_info, "bytes_used is too small %llu",
4148 btrfs_super_bytes_used(sb));
4149 ret = -EINVAL;
4150 }
4151 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4152 btrfs_err(fs_info, "invalid stripesize %u",
4153 btrfs_super_stripesize(sb));
4154 ret = -EINVAL;
4155 }
4156 if (btrfs_super_num_devices(sb) > (1UL << 31))
4157 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4158 btrfs_super_num_devices(sb));
4159 if (btrfs_super_num_devices(sb) == 0) {
4160 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4161 ret = -EINVAL;
4162 }
4163
4164 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4165 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4166 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4167 ret = -EINVAL;
4168 }
4169
4170 /*
4171 * Obvious sys_chunk_array corruptions, it must hold at least one key
4172 * and one chunk
4173 */
4174 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4175 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4176 btrfs_super_sys_array_size(sb),
4177 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4178 ret = -EINVAL;
4179 }
4180 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4181 + sizeof(struct btrfs_chunk)) {
4182 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4183 btrfs_super_sys_array_size(sb),
4184 sizeof(struct btrfs_disk_key)
4185 + sizeof(struct btrfs_chunk));
4186 ret = -EINVAL;
4187 }
4188
4189 /*
4190 * The generation is a global counter, we'll trust it more than the others
4191 * but it's still possible that it's the one that's wrong.
4192 */
4193 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4194 printk(KERN_WARNING
4195 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4196 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4197 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4198 && btrfs_super_cache_generation(sb) != (u64)-1)
4199 printk(KERN_WARNING
4200 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4201 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4202
4203 return ret;
4204}
4205
4206static void btrfs_error_commit_super(struct btrfs_root *root)
4207{
4208 mutex_lock(&root->fs_info->cleaner_mutex);
4209 btrfs_run_delayed_iputs(root);
4210 mutex_unlock(&root->fs_info->cleaner_mutex);
4211
4212 down_write(&root->fs_info->cleanup_work_sem);
4213 up_write(&root->fs_info->cleanup_work_sem);
4214
4215 /* cleanup FS via transaction */
4216 btrfs_cleanup_transaction(root);
4217}
4218
4219static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4220{
4221 struct btrfs_ordered_extent *ordered;
4222
4223 spin_lock(&root->ordered_extent_lock);
4224 /*
4225 * This will just short circuit the ordered completion stuff which will
4226 * make sure the ordered extent gets properly cleaned up.
4227 */
4228 list_for_each_entry(ordered, &root->ordered_extents,
4229 root_extent_list)
4230 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4231 spin_unlock(&root->ordered_extent_lock);
4232}
4233
4234static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4235{
4236 struct btrfs_root *root;
4237 struct list_head splice;
4238
4239 INIT_LIST_HEAD(&splice);
4240
4241 spin_lock(&fs_info->ordered_root_lock);
4242 list_splice_init(&fs_info->ordered_roots, &splice);
4243 while (!list_empty(&splice)) {
4244 root = list_first_entry(&splice, struct btrfs_root,
4245 ordered_root);
4246 list_move_tail(&root->ordered_root,
4247 &fs_info->ordered_roots);
4248
4249 spin_unlock(&fs_info->ordered_root_lock);
4250 btrfs_destroy_ordered_extents(root);
4251
4252 cond_resched();
4253 spin_lock(&fs_info->ordered_root_lock);
4254 }
4255 spin_unlock(&fs_info->ordered_root_lock);
4256}
4257
4258static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4259 struct btrfs_root *root)
4260{
4261 struct rb_node *node;
4262 struct btrfs_delayed_ref_root *delayed_refs;
4263 struct btrfs_delayed_ref_node *ref;
4264 int ret = 0;
4265
4266 delayed_refs = &trans->delayed_refs;
4267
4268 spin_lock(&delayed_refs->lock);
4269 if (atomic_read(&delayed_refs->num_entries) == 0) {
4270 spin_unlock(&delayed_refs->lock);
4271 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4272 return ret;
4273 }
4274
4275 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4276 struct btrfs_delayed_ref_head *head;
4277 struct btrfs_delayed_ref_node *tmp;
4278 bool pin_bytes = false;
4279
4280 head = rb_entry(node, struct btrfs_delayed_ref_head,
4281 href_node);
4282 if (!mutex_trylock(&head->mutex)) {
4283 atomic_inc(&head->node.refs);
4284 spin_unlock(&delayed_refs->lock);
4285
4286 mutex_lock(&head->mutex);
4287 mutex_unlock(&head->mutex);
4288 btrfs_put_delayed_ref(&head->node);
4289 spin_lock(&delayed_refs->lock);
4290 continue;
4291 }
4292 spin_lock(&head->lock);
4293 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4294 list) {
4295 ref->in_tree = 0;
4296 list_del(&ref->list);
4297 atomic_dec(&delayed_refs->num_entries);
4298 btrfs_put_delayed_ref(ref);
4299 }
4300 if (head->must_insert_reserved)
4301 pin_bytes = true;
4302 btrfs_free_delayed_extent_op(head->extent_op);
4303 delayed_refs->num_heads--;
4304 if (head->processing == 0)
4305 delayed_refs->num_heads_ready--;
4306 atomic_dec(&delayed_refs->num_entries);
4307 head->node.in_tree = 0;
4308 rb_erase(&head->href_node, &delayed_refs->href_root);
4309 spin_unlock(&head->lock);
4310 spin_unlock(&delayed_refs->lock);
4311 mutex_unlock(&head->mutex);
4312
4313 if (pin_bytes)
4314 btrfs_pin_extent(root, head->node.bytenr,
4315 head->node.num_bytes, 1);
4316 btrfs_put_delayed_ref(&head->node);
4317 cond_resched();
4318 spin_lock(&delayed_refs->lock);
4319 }
4320
4321 spin_unlock(&delayed_refs->lock);
4322
4323 return ret;
4324}
4325
4326static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4327{
4328 struct btrfs_inode *btrfs_inode;
4329 struct list_head splice;
4330
4331 INIT_LIST_HEAD(&splice);
4332
4333 spin_lock(&root->delalloc_lock);
4334 list_splice_init(&root->delalloc_inodes, &splice);
4335
4336 while (!list_empty(&splice)) {
4337 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4338 delalloc_inodes);
4339
4340 list_del_init(&btrfs_inode->delalloc_inodes);
4341 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4342 &btrfs_inode->runtime_flags);
4343 spin_unlock(&root->delalloc_lock);
4344
4345 btrfs_invalidate_inodes(btrfs_inode->root);
4346
4347 spin_lock(&root->delalloc_lock);
4348 }
4349
4350 spin_unlock(&root->delalloc_lock);
4351}
4352
4353static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4354{
4355 struct btrfs_root *root;
4356 struct list_head splice;
4357
4358 INIT_LIST_HEAD(&splice);
4359
4360 spin_lock(&fs_info->delalloc_root_lock);
4361 list_splice_init(&fs_info->delalloc_roots, &splice);
4362 while (!list_empty(&splice)) {
4363 root = list_first_entry(&splice, struct btrfs_root,
4364 delalloc_root);
4365 list_del_init(&root->delalloc_root);
4366 root = btrfs_grab_fs_root(root);
4367 BUG_ON(!root);
4368 spin_unlock(&fs_info->delalloc_root_lock);
4369
4370 btrfs_destroy_delalloc_inodes(root);
4371 btrfs_put_fs_root(root);
4372
4373 spin_lock(&fs_info->delalloc_root_lock);
4374 }
4375 spin_unlock(&fs_info->delalloc_root_lock);
4376}
4377
4378static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4379 struct extent_io_tree *dirty_pages,
4380 int mark)
4381{
4382 int ret;
4383 struct extent_buffer *eb;
4384 u64 start = 0;
4385 u64 end;
4386
4387 while (1) {
4388 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4389 mark, NULL);
4390 if (ret)
4391 break;
4392
4393 clear_extent_bits(dirty_pages, start, end, mark);
4394 while (start <= end) {
4395 eb = btrfs_find_tree_block(root->fs_info, start);
4396 start += root->nodesize;
4397 if (!eb)
4398 continue;
4399 wait_on_extent_buffer_writeback(eb);
4400
4401 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4402 &eb->bflags))
4403 clear_extent_buffer_dirty(eb);
4404 free_extent_buffer_stale(eb);
4405 }
4406 }
4407
4408 return ret;
4409}
4410
4411static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4412 struct extent_io_tree *pinned_extents)
4413{
4414 struct extent_io_tree *unpin;
4415 u64 start;
4416 u64 end;
4417 int ret;
4418 bool loop = true;
4419
4420 unpin = pinned_extents;
4421again:
4422 while (1) {
4423 ret = find_first_extent_bit(unpin, 0, &start, &end,
4424 EXTENT_DIRTY, NULL);
4425 if (ret)
4426 break;
4427
4428 clear_extent_dirty(unpin, start, end);
4429 btrfs_error_unpin_extent_range(root, start, end);
4430 cond_resched();
4431 }
4432
4433 if (loop) {
4434 if (unpin == &root->fs_info->freed_extents[0])
4435 unpin = &root->fs_info->freed_extents[1];
4436 else
4437 unpin = &root->fs_info->freed_extents[0];
4438 loop = false;
4439 goto again;
4440 }
4441
4442 return 0;
4443}
4444
4445void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4446 struct btrfs_root *root)
4447{
4448 btrfs_destroy_delayed_refs(cur_trans, root);
4449
4450 cur_trans->state = TRANS_STATE_COMMIT_START;
4451 wake_up(&root->fs_info->transaction_blocked_wait);
4452
4453 cur_trans->state = TRANS_STATE_UNBLOCKED;
4454 wake_up(&root->fs_info->transaction_wait);
4455
4456 btrfs_destroy_delayed_inodes(root);
4457 btrfs_assert_delayed_root_empty(root);
4458
4459 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4460 EXTENT_DIRTY);
4461 btrfs_destroy_pinned_extent(root,
4462 root->fs_info->pinned_extents);
4463
4464 cur_trans->state =TRANS_STATE_COMPLETED;
4465 wake_up(&cur_trans->commit_wait);
4466
4467 /*
4468 memset(cur_trans, 0, sizeof(*cur_trans));
4469 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4470 */
4471}
4472
4473static int btrfs_cleanup_transaction(struct btrfs_root *root)
4474{
4475 struct btrfs_transaction *t;
4476
4477 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4478
4479 spin_lock(&root->fs_info->trans_lock);
4480 while (!list_empty(&root->fs_info->trans_list)) {
4481 t = list_first_entry(&root->fs_info->trans_list,
4482 struct btrfs_transaction, list);
4483 if (t->state >= TRANS_STATE_COMMIT_START) {
4484 atomic_inc(&t->use_count);
4485 spin_unlock(&root->fs_info->trans_lock);
4486 btrfs_wait_for_commit(root, t->transid);
4487 btrfs_put_transaction(t);
4488 spin_lock(&root->fs_info->trans_lock);
4489 continue;
4490 }
4491 if (t == root->fs_info->running_transaction) {
4492 t->state = TRANS_STATE_COMMIT_DOING;
4493 spin_unlock(&root->fs_info->trans_lock);
4494 /*
4495 * We wait for 0 num_writers since we don't hold a trans
4496 * handle open currently for this transaction.
4497 */
4498 wait_event(t->writer_wait,
4499 atomic_read(&t->num_writers) == 0);
4500 } else {
4501 spin_unlock(&root->fs_info->trans_lock);
4502 }
4503 btrfs_cleanup_one_transaction(t, root);
4504
4505 spin_lock(&root->fs_info->trans_lock);
4506 if (t == root->fs_info->running_transaction)
4507 root->fs_info->running_transaction = NULL;
4508 list_del_init(&t->list);
4509 spin_unlock(&root->fs_info->trans_lock);
4510
4511 btrfs_put_transaction(t);
4512 trace_btrfs_transaction_commit(root);
4513 spin_lock(&root->fs_info->trans_lock);
4514 }
4515 spin_unlock(&root->fs_info->trans_lock);
4516 btrfs_destroy_all_ordered_extents(root->fs_info);
4517 btrfs_destroy_delayed_inodes(root);
4518 btrfs_assert_delayed_root_empty(root);
4519 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4520 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4521 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4522
4523 return 0;
4524}
4525
4526static const struct extent_io_ops btree_extent_io_ops = {
4527 .readpage_end_io_hook = btree_readpage_end_io_hook,
4528 .readpage_io_failed_hook = btree_io_failed_hook,
4529 .submit_bio_hook = btree_submit_bio_hook,
4530 /* note we're sharing with inode.c for the merge bio hook */
4531 .merge_bio_hook = btrfs_merge_bio_hook,
4532};
Linux 6.x block layer and io_uring tree(s)