projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: don't write any data into a readonly device when scrub
[linux-2.6-block.git] / fs / btrfs / disk-io.c
... / ...
CommitLineData
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/fs.h>
20#include <linux/blkdev.h>
21#include <linux/scatterlist.h>
22#include <linux/swap.h>
23#include <linux/radix-tree.h>
24#include <linux/writeback.h>
25#include <linux/buffer_head.h>
26#include <linux/workqueue.h>
27#include <linux/kthread.h>
28#include <linux/freezer.h>
29#include <linux/slab.h>
30#include <linux/migrate.h>
31#include <linux/ratelimit.h>
32#include <linux/uuid.h>
33#include <linux/semaphore.h>
34#include <asm/unaligned.h>
35#include "ctree.h"
36#include "disk-io.h"
37#include "hash.h"
38#include "transaction.h"
39#include "btrfs_inode.h"
40#include "volumes.h"
41#include "print-tree.h"
42#include "async-thread.h"
43#include "locking.h"
44#include "tree-log.h"
45#include "free-space-cache.h"
46#include "inode-map.h"
47#include "check-integrity.h"
48#include "rcu-string.h"
49#include "dev-replace.h"
50#include "raid56.h"
51#include "sysfs.h"
52#include "qgroup.h"
53
54#ifdef CONFIG_X86
55#include <asm/cpufeature.h>
56#endif
57
58static struct extent_io_ops btree_extent_io_ops;
59static void end_workqueue_fn(struct btrfs_work *work);
60static void free_fs_root(struct btrfs_root *root);
61static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
62 int read_only);
63static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
64static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
65 struct btrfs_root *root);
66static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
67static int btrfs_destroy_marked_extents(struct btrfs_root *root,
68 struct extent_io_tree *dirty_pages,
69 int mark);
70static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
71 struct extent_io_tree *pinned_extents);
72static int btrfs_cleanup_transaction(struct btrfs_root *root);
73static void btrfs_error_commit_super(struct btrfs_root *root);
74
75/*
76 * end_io_wq structs are used to do processing in task context when an IO is
77 * complete. This is used during reads to verify checksums, and it is used
78 * by writes to insert metadata for new file extents after IO is complete.
79 */
80struct end_io_wq {
81 struct bio *bio;
82 bio_end_io_t *end_io;
83 void *private;
84 struct btrfs_fs_info *info;
85 int error;
86 int metadata;
87 struct list_head list;
88 struct btrfs_work work;
89};
90
91/*
92 * async submit bios are used to offload expensive checksumming
93 * onto the worker threads. They checksum file and metadata bios
94 * just before they are sent down the IO stack.
95 */
96struct async_submit_bio {
97 struct inode *inode;
98 struct bio *bio;
99 struct list_head list;
100 extent_submit_bio_hook_t *submit_bio_start;
101 extent_submit_bio_hook_t *submit_bio_done;
102 int rw;
103 int mirror_num;
104 unsigned long bio_flags;
105 /*
106 * bio_offset is optional, can be used if the pages in the bio
107 * can't tell us where in the file the bio should go
108 */
109 u64 bio_offset;
110 struct btrfs_work work;
111 int error;
112};
113
114/*
115 * Lockdep class keys for extent_buffer->lock's in this root. For a given
116 * eb, the lockdep key is determined by the btrfs_root it belongs to and
117 * the level the eb occupies in the tree.
118 *
119 * Different roots are used for different purposes and may nest inside each
120 * other and they require separate keysets. As lockdep keys should be
121 * static, assign keysets according to the purpose of the root as indicated
122 * by btrfs_root->objectid. This ensures that all special purpose roots
123 * have separate keysets.
124 *
125 * Lock-nesting across peer nodes is always done with the immediate parent
126 * node locked thus preventing deadlock. As lockdep doesn't know this, use
127 * subclass to avoid triggering lockdep warning in such cases.
128 *
129 * The key is set by the readpage_end_io_hook after the buffer has passed
130 * csum validation but before the pages are unlocked. It is also set by
131 * btrfs_init_new_buffer on freshly allocated blocks.
132 *
133 * We also add a check to make sure the highest level of the tree is the
134 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
135 * needs update as well.
136 */
137#ifdef CONFIG_DEBUG_LOCK_ALLOC
138# if BTRFS_MAX_LEVEL != 8
139# error
140# endif
141
142static struct btrfs_lockdep_keyset {
143 u64 id; /* root objectid */
144 const char *name_stem; /* lock name stem */
145 char names[BTRFS_MAX_LEVEL + 1][20];
146 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
147} btrfs_lockdep_keysets[] = {
148 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
149 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
150 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
151 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
152 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
153 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
154 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
155 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
156 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
157 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
158 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
159 { .id = 0, .name_stem = "tree" },
160};
161
162void __init btrfs_init_lockdep(void)
163{
164 int i, j;
165
166 /* initialize lockdep class names */
167 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
168 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
169
170 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
171 snprintf(ks->names[j], sizeof(ks->names[j]),
172 "btrfs-%s-%02d", ks->name_stem, j);
173 }
174}
175
176void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
177 int level)
178{
179 struct btrfs_lockdep_keyset *ks;
180
181 BUG_ON(level >= ARRAY_SIZE(ks->keys));
182
183 /* find the matching keyset, id 0 is the default entry */
184 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
185 if (ks->id == objectid)
186 break;
187
188 lockdep_set_class_and_name(&eb->lock,
189 &ks->keys[level], ks->names[level]);
190}
191
192#endif
193
194/*
195 * extents on the btree inode are pretty simple, there's one extent
196 * that covers the entire device
197 */
198static struct extent_map *btree_get_extent(struct inode *inode,
199 struct page *page, size_t pg_offset, u64 start, u64 len,
200 int create)
201{
202 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
203 struct extent_map *em;
204 int ret;
205
206 read_lock(&em_tree->lock);
207 em = lookup_extent_mapping(em_tree, start, len);
208 if (em) {
209 em->bdev =
210 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
211 read_unlock(&em_tree->lock);
212 goto out;
213 }
214 read_unlock(&em_tree->lock);
215
216 em = alloc_extent_map();
217 if (!em) {
218 em = ERR_PTR(-ENOMEM);
219 goto out;
220 }
221 em->start = 0;
222 em->len = (u64)-1;
223 em->block_len = (u64)-1;
224 em->block_start = 0;
225 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
226
227 write_lock(&em_tree->lock);
228 ret = add_extent_mapping(em_tree, em, 0);
229 if (ret == -EEXIST) {
230 free_extent_map(em);
231 em = lookup_extent_mapping(em_tree, start, len);
232 if (!em)
233 em = ERR_PTR(-EIO);
234 } else if (ret) {
235 free_extent_map(em);
236 em = ERR_PTR(ret);
237 }
238 write_unlock(&em_tree->lock);
239
240out:
241 return em;
242}
243
244u32 btrfs_csum_data(char *data, u32 seed, size_t len)
245{
246 return btrfs_crc32c(seed, data, len);
247}
248
249void btrfs_csum_final(u32 crc, char *result)
250{
251 put_unaligned_le32(~crc, result);
252}
253
254/*
255 * compute the csum for a btree block, and either verify it or write it
256 * into the csum field of the block.
257 */
258static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
259 int verify)
260{
261 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
262 char *result = NULL;
263 unsigned long len;
264 unsigned long cur_len;
265 unsigned long offset = BTRFS_CSUM_SIZE;
266 char *kaddr;
267 unsigned long map_start;
268 unsigned long map_len;
269 int err;
270 u32 crc = ~(u32)0;
271 unsigned long inline_result;
272
273 len = buf->len - offset;
274 while (len > 0) {
275 err = map_private_extent_buffer(buf, offset, 32,
276 &kaddr, &map_start, &map_len);
277 if (err)
278 return 1;
279 cur_len = min(len, map_len - (offset - map_start));
280 crc = btrfs_csum_data(kaddr + offset - map_start,
281 crc, cur_len);
282 len -= cur_len;
283 offset += cur_len;
284 }
285 if (csum_size > sizeof(inline_result)) {
286 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
287 if (!result)
288 return 1;
289 } else {
290 result = (char *)&inline_result;
291 }
292
293 btrfs_csum_final(crc, result);
294
295 if (verify) {
296 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
297 u32 val;
298 u32 found = 0;
299 memcpy(&found, result, csum_size);
300
301 read_extent_buffer(buf, &val, 0, csum_size);
302 printk_ratelimited(KERN_INFO
303 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
304 "level %d\n",
305 root->fs_info->sb->s_id, buf->start,
306 val, found, btrfs_header_level(buf));
307 if (result != (char *)&inline_result)
308 kfree(result);
309 return 1;
310 }
311 } else {
312 write_extent_buffer(buf, result, 0, csum_size);
313 }
314 if (result != (char *)&inline_result)
315 kfree(result);
316 return 0;
317}
318
319/*
320 * we can't consider a given block up to date unless the transid of the
321 * block matches the transid in the parent node's pointer. This is how we
322 * detect blocks that either didn't get written at all or got written
323 * in the wrong place.
324 */
325static int verify_parent_transid(struct extent_io_tree *io_tree,
326 struct extent_buffer *eb, u64 parent_transid,
327 int atomic)
328{
329 struct extent_state *cached_state = NULL;
330 int ret;
331 bool need_lock = (current->journal_info ==
332 (void *)BTRFS_SEND_TRANS_STUB);
333
334 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
335 return 0;
336
337 if (atomic)
338 return -EAGAIN;
339
340 if (need_lock) {
341 btrfs_tree_read_lock(eb);
342 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
343 }
344
345 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
346 0, &cached_state);
347 if (extent_buffer_uptodate(eb) &&
348 btrfs_header_generation(eb) == parent_transid) {
349 ret = 0;
350 goto out;
351 }
352 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
353 "found %llu\n",
354 eb->start, parent_transid, btrfs_header_generation(eb));
355 ret = 1;
356
357 /*
358 * Things reading via commit roots that don't have normal protection,
359 * like send, can have a really old block in cache that may point at a
360 * block that has been free'd and re-allocated. So don't clear uptodate
361 * if we find an eb that is under IO (dirty/writeback) because we could
362 * end up reading in the stale data and then writing it back out and
363 * making everybody very sad.
364 */
365 if (!extent_buffer_under_io(eb))
366 clear_extent_buffer_uptodate(eb);
367out:
368 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
369 &cached_state, GFP_NOFS);
370 if (need_lock)
371 btrfs_tree_read_unlock_blocking(eb);
372 return ret;
373}
374
375/*
376 * Return 0 if the superblock checksum type matches the checksum value of that
377 * algorithm. Pass the raw disk superblock data.
378 */
379static int btrfs_check_super_csum(char *raw_disk_sb)
380{
381 struct btrfs_super_block *disk_sb =
382 (struct btrfs_super_block *)raw_disk_sb;
383 u16 csum_type = btrfs_super_csum_type(disk_sb);
384 int ret = 0;
385
386 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
387 u32 crc = ~(u32)0;
388 const int csum_size = sizeof(crc);
389 char result[csum_size];
390
391 /*
392 * The super_block structure does not span the whole
393 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
394 * is filled with zeros and is included in the checkum.
395 */
396 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
397 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
398 btrfs_csum_final(crc, result);
399
400 if (memcmp(raw_disk_sb, result, csum_size))
401 ret = 1;
402
403 if (ret && btrfs_super_generation(disk_sb) < 10) {
404 printk(KERN_WARNING
405 "BTRFS: super block crcs don't match, older mkfs detected\n");
406 ret = 0;
407 }
408 }
409
410 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
411 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
412 csum_type);
413 ret = 1;
414 }
415
416 return ret;
417}
418
419/*
420 * helper to read a given tree block, doing retries as required when
421 * the checksums don't match and we have alternate mirrors to try.
422 */
423static int btree_read_extent_buffer_pages(struct btrfs_root *root,
424 struct extent_buffer *eb,
425 u64 start, u64 parent_transid)
426{
427 struct extent_io_tree *io_tree;
428 int failed = 0;
429 int ret;
430 int num_copies = 0;
431 int mirror_num = 0;
432 int failed_mirror = 0;
433
434 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
435 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
436 while (1) {
437 ret = read_extent_buffer_pages(io_tree, eb, start,
438 WAIT_COMPLETE,
439 btree_get_extent, mirror_num);
440 if (!ret) {
441 if (!verify_parent_transid(io_tree, eb,
442 parent_transid, 0))
443 break;
444 else
445 ret = -EIO;
446 }
447
448 /*
449 * This buffer's crc is fine, but its contents are corrupted, so
450 * there is no reason to read the other copies, they won't be
451 * any less wrong.
452 */
453 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
454 break;
455
456 num_copies = btrfs_num_copies(root->fs_info,
457 eb->start, eb->len);
458 if (num_copies == 1)
459 break;
460
461 if (!failed_mirror) {
462 failed = 1;
463 failed_mirror = eb->read_mirror;
464 }
465
466 mirror_num++;
467 if (mirror_num == failed_mirror)
468 mirror_num++;
469
470 if (mirror_num > num_copies)
471 break;
472 }
473
474 if (failed && !ret && failed_mirror)
475 repair_eb_io_failure(root, eb, failed_mirror);
476
477 return ret;
478}
479
480/*
481 * checksum a dirty tree block before IO. This has extra checks to make sure
482 * we only fill in the checksum field in the first page of a multi-page block
483 */
484
485static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
486{
487 u64 start = page_offset(page);
488 u64 found_start;
489 struct extent_buffer *eb;
490
491 eb = (struct extent_buffer *)page->private;
492 if (page != eb->pages[0])
493 return 0;
494 found_start = btrfs_header_bytenr(eb);
495 if (WARN_ON(found_start != start || !PageUptodate(page)))
496 return 0;
497 csum_tree_block(root, eb, 0);
498 return 0;
499}
500
501static int check_tree_block_fsid(struct btrfs_root *root,
502 struct extent_buffer *eb)
503{
504 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
505 u8 fsid[BTRFS_UUID_SIZE];
506 int ret = 1;
507
508 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
509 while (fs_devices) {
510 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
511 ret = 0;
512 break;
513 }
514 fs_devices = fs_devices->seed;
515 }
516 return ret;
517}
518
519#define CORRUPT(reason, eb, root, slot) \
520 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
521 "root=%llu, slot=%d", reason, \
522 btrfs_header_bytenr(eb), root->objectid, slot)
523
524static noinline int check_leaf(struct btrfs_root *root,
525 struct extent_buffer *leaf)
526{
527 struct btrfs_key key;
528 struct btrfs_key leaf_key;
529 u32 nritems = btrfs_header_nritems(leaf);
530 int slot;
531
532 if (nritems == 0)
533 return 0;
534
535 /* Check the 0 item */
536 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
537 BTRFS_LEAF_DATA_SIZE(root)) {
538 CORRUPT("invalid item offset size pair", leaf, root, 0);
539 return -EIO;
540 }
541
542 /*
543 * Check to make sure each items keys are in the correct order and their
544 * offsets make sense. We only have to loop through nritems-1 because
545 * we check the current slot against the next slot, which verifies the
546 * next slot's offset+size makes sense and that the current's slot
547 * offset is correct.
548 */
549 for (slot = 0; slot < nritems - 1; slot++) {
550 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
551 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
552
553 /* Make sure the keys are in the right order */
554 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
555 CORRUPT("bad key order", leaf, root, slot);
556 return -EIO;
557 }
558
559 /*
560 * Make sure the offset and ends are right, remember that the
561 * item data starts at the end of the leaf and grows towards the
562 * front.
563 */
564 if (btrfs_item_offset_nr(leaf, slot) !=
565 btrfs_item_end_nr(leaf, slot + 1)) {
566 CORRUPT("slot offset bad", leaf, root, slot);
567 return -EIO;
568 }
569
570 /*
571 * Check to make sure that we don't point outside of the leaf,
572 * just incase all the items are consistent to eachother, but
573 * all point outside of the leaf.
574 */
575 if (btrfs_item_end_nr(leaf, slot) >
576 BTRFS_LEAF_DATA_SIZE(root)) {
577 CORRUPT("slot end outside of leaf", leaf, root, slot);
578 return -EIO;
579 }
580 }
581
582 return 0;
583}
584
585static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
586 u64 phy_offset, struct page *page,
587 u64 start, u64 end, int mirror)
588{
589 u64 found_start;
590 int found_level;
591 struct extent_buffer *eb;
592 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
593 int ret = 0;
594 int reads_done;
595
596 if (!page->private)
597 goto out;
598
599 eb = (struct extent_buffer *)page->private;
600
601 /* the pending IO might have been the only thing that kept this buffer
602 * in memory. Make sure we have a ref for all this other checks
603 */
604 extent_buffer_get(eb);
605
606 reads_done = atomic_dec_and_test(&eb->io_pages);
607 if (!reads_done)
608 goto err;
609
610 eb->read_mirror = mirror;
611 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
612 ret = -EIO;
613 goto err;
614 }
615
616 found_start = btrfs_header_bytenr(eb);
617 if (found_start != eb->start) {
618 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
619 "%llu %llu\n",
620 found_start, eb->start);
621 ret = -EIO;
622 goto err;
623 }
624 if (check_tree_block_fsid(root, eb)) {
625 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
626 eb->start);
627 ret = -EIO;
628 goto err;
629 }
630 found_level = btrfs_header_level(eb);
631 if (found_level >= BTRFS_MAX_LEVEL) {
632 btrfs_info(root->fs_info, "bad tree block level %d",
633 (int)btrfs_header_level(eb));
634 ret = -EIO;
635 goto err;
636 }
637
638 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
639 eb, found_level);
640
641 ret = csum_tree_block(root, eb, 1);
642 if (ret) {
643 ret = -EIO;
644 goto err;
645 }
646
647 /*
648 * If this is a leaf block and it is corrupt, set the corrupt bit so
649 * that we don't try and read the other copies of this block, just
650 * return -EIO.
651 */
652 if (found_level == 0 && check_leaf(root, eb)) {
653 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
654 ret = -EIO;
655 }
656
657 if (!ret)
658 set_extent_buffer_uptodate(eb);
659err:
660 if (reads_done &&
661 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
662 btree_readahead_hook(root, eb, eb->start, ret);
663
664 if (ret) {
665 /*
666 * our io error hook is going to dec the io pages
667 * again, we have to make sure it has something
668 * to decrement
669 */
670 atomic_inc(&eb->io_pages);
671 clear_extent_buffer_uptodate(eb);
672 }
673 free_extent_buffer(eb);
674out:
675 return ret;
676}
677
678static int btree_io_failed_hook(struct page *page, int failed_mirror)
679{
680 struct extent_buffer *eb;
681 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
682
683 eb = (struct extent_buffer *)page->private;
684 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
685 eb->read_mirror = failed_mirror;
686 atomic_dec(&eb->io_pages);
687 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
688 btree_readahead_hook(root, eb, eb->start, -EIO);
689 return -EIO; /* we fixed nothing */
690}
691
692static void end_workqueue_bio(struct bio *bio, int err)
693{
694 struct end_io_wq *end_io_wq = bio->bi_private;
695 struct btrfs_fs_info *fs_info;
696
697 fs_info = end_io_wq->info;
698 end_io_wq->error = err;
699 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
700
701 if (bio->bi_rw & REQ_WRITE) {
702 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
703 btrfs_queue_work(fs_info->endio_meta_write_workers,
704 &end_io_wq->work);
705 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
706 btrfs_queue_work(fs_info->endio_freespace_worker,
707 &end_io_wq->work);
708 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
709 btrfs_queue_work(fs_info->endio_raid56_workers,
710 &end_io_wq->work);
711 else
712 btrfs_queue_work(fs_info->endio_write_workers,
713 &end_io_wq->work);
714 } else {
715 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
716 btrfs_queue_work(fs_info->endio_raid56_workers,
717 &end_io_wq->work);
718 else if (end_io_wq->metadata)
719 btrfs_queue_work(fs_info->endio_meta_workers,
720 &end_io_wq->work);
721 else
722 btrfs_queue_work(fs_info->endio_workers,
723 &end_io_wq->work);
724 }
725}
726
727/*
728 * For the metadata arg you want
729 *
730 * 0 - if data
731 * 1 - if normal metadta
732 * 2 - if writing to the free space cache area
733 * 3 - raid parity work
734 */
735int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
736 int metadata)
737{
738 struct end_io_wq *end_io_wq;
739 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
740 if (!end_io_wq)
741 return -ENOMEM;
742
743 end_io_wq->private = bio->bi_private;
744 end_io_wq->end_io = bio->bi_end_io;
745 end_io_wq->info = info;
746 end_io_wq->error = 0;
747 end_io_wq->bio = bio;
748 end_io_wq->metadata = metadata;
749
750 bio->bi_private = end_io_wq;
751 bio->bi_end_io = end_workqueue_bio;
752 return 0;
753}
754
755unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
756{
757 unsigned long limit = min_t(unsigned long,
758 info->thread_pool_size,
759 info->fs_devices->open_devices);
760 return 256 * limit;
761}
762
763static void run_one_async_start(struct btrfs_work *work)
764{
765 struct async_submit_bio *async;
766 int ret;
767
768 async = container_of(work, struct async_submit_bio, work);
769 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
770 async->mirror_num, async->bio_flags,
771 async->bio_offset);
772 if (ret)
773 async->error = ret;
774}
775
776static void run_one_async_done(struct btrfs_work *work)
777{
778 struct btrfs_fs_info *fs_info;
779 struct async_submit_bio *async;
780 int limit;
781
782 async = container_of(work, struct async_submit_bio, work);
783 fs_info = BTRFS_I(async->inode)->root->fs_info;
784
785 limit = btrfs_async_submit_limit(fs_info);
786 limit = limit * 2 / 3;
787
788 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
789 waitqueue_active(&fs_info->async_submit_wait))
790 wake_up(&fs_info->async_submit_wait);
791
792 /* If an error occured we just want to clean up the bio and move on */
793 if (async->error) {
794 bio_endio(async->bio, async->error);
795 return;
796 }
797
798 async->submit_bio_done(async->inode, async->rw, async->bio,
799 async->mirror_num, async->bio_flags,
800 async->bio_offset);
801}
802
803static void run_one_async_free(struct btrfs_work *work)
804{
805 struct async_submit_bio *async;
806
807 async = container_of(work, struct async_submit_bio, work);
808 kfree(async);
809}
810
811int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
812 int rw, struct bio *bio, int mirror_num,
813 unsigned long bio_flags,
814 u64 bio_offset,
815 extent_submit_bio_hook_t *submit_bio_start,
816 extent_submit_bio_hook_t *submit_bio_done)
817{
818 struct async_submit_bio *async;
819
820 async = kmalloc(sizeof(*async), GFP_NOFS);
821 if (!async)
822 return -ENOMEM;
823
824 async->inode = inode;
825 async->rw = rw;
826 async->bio = bio;
827 async->mirror_num = mirror_num;
828 async->submit_bio_start = submit_bio_start;
829 async->submit_bio_done = submit_bio_done;
830
831 btrfs_init_work(&async->work, run_one_async_start,
832 run_one_async_done, run_one_async_free);
833
834 async->bio_flags = bio_flags;
835 async->bio_offset = bio_offset;
836
837 async->error = 0;
838
839 atomic_inc(&fs_info->nr_async_submits);
840
841 if (rw & REQ_SYNC)
842 btrfs_set_work_high_priority(&async->work);
843
844 btrfs_queue_work(fs_info->workers, &async->work);
845
846 while (atomic_read(&fs_info->async_submit_draining) &&
847 atomic_read(&fs_info->nr_async_submits)) {
848 wait_event(fs_info->async_submit_wait,
849 (atomic_read(&fs_info->nr_async_submits) == 0));
850 }
851
852 return 0;
853}
854
855static int btree_csum_one_bio(struct bio *bio)
856{
857 struct bio_vec *bvec;
858 struct btrfs_root *root;
859 int i, ret = 0;
860
861 bio_for_each_segment_all(bvec, bio, i) {
862 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
863 ret = csum_dirty_buffer(root, bvec->bv_page);
864 if (ret)
865 break;
866 }
867
868 return ret;
869}
870
871static int __btree_submit_bio_start(struct inode *inode, int rw,
872 struct bio *bio, int mirror_num,
873 unsigned long bio_flags,
874 u64 bio_offset)
875{
876 /*
877 * when we're called for a write, we're already in the async
878 * submission context. Just jump into btrfs_map_bio
879 */
880 return btree_csum_one_bio(bio);
881}
882
883static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
884 int mirror_num, unsigned long bio_flags,
885 u64 bio_offset)
886{
887 int ret;
888
889 /*
890 * when we're called for a write, we're already in the async
891 * submission context. Just jump into btrfs_map_bio
892 */
893 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
894 if (ret)
895 bio_endio(bio, ret);
896 return ret;
897}
898
899static int check_async_write(struct inode *inode, unsigned long bio_flags)
900{
901 if (bio_flags & EXTENT_BIO_TREE_LOG)
902 return 0;
903#ifdef CONFIG_X86
904 if (cpu_has_xmm4_2)
905 return 0;
906#endif
907 return 1;
908}
909
910static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
911 int mirror_num, unsigned long bio_flags,
912 u64 bio_offset)
913{
914 int async = check_async_write(inode, bio_flags);
915 int ret;
916
917 if (!(rw & REQ_WRITE)) {
918 /*
919 * called for a read, do the setup so that checksum validation
920 * can happen in the async kernel threads
921 */
922 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
923 bio, 1);
924 if (ret)
925 goto out_w_error;
926 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
927 mirror_num, 0);
928 } else if (!async) {
929 ret = btree_csum_one_bio(bio);
930 if (ret)
931 goto out_w_error;
932 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
933 mirror_num, 0);
934 } else {
935 /*
936 * kthread helpers are used to submit writes so that
937 * checksumming can happen in parallel across all CPUs
938 */
939 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
940 inode, rw, bio, mirror_num, 0,
941 bio_offset,
942 __btree_submit_bio_start,
943 __btree_submit_bio_done);
944 }
945
946 if (ret) {
947out_w_error:
948 bio_endio(bio, ret);
949 }
950 return ret;
951}
952
953#ifdef CONFIG_MIGRATION
954static int btree_migratepage(struct address_space *mapping,
955 struct page *newpage, struct page *page,
956 enum migrate_mode mode)
957{
958 /*
959 * we can't safely write a btree page from here,
960 * we haven't done the locking hook
961 */
962 if (PageDirty(page))
963 return -EAGAIN;
964 /*
965 * Buffers may be managed in a filesystem specific way.
966 * We must have no buffers or drop them.
967 */
968 if (page_has_private(page) &&
969 !try_to_release_page(page, GFP_KERNEL))
970 return -EAGAIN;
971 return migrate_page(mapping, newpage, page, mode);
972}
973#endif
974
975
976static int btree_writepages(struct address_space *mapping,
977 struct writeback_control *wbc)
978{
979 struct btrfs_fs_info *fs_info;
980 int ret;
981
982 if (wbc->sync_mode == WB_SYNC_NONE) {
983
984 if (wbc->for_kupdate)
985 return 0;
986
987 fs_info = BTRFS_I(mapping->host)->root->fs_info;
988 /* this is a bit racy, but that's ok */
989 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
990 BTRFS_DIRTY_METADATA_THRESH);
991 if (ret < 0)
992 return 0;
993 }
994 return btree_write_cache_pages(mapping, wbc);
995}
996
997static int btree_readpage(struct file *file, struct page *page)
998{
999 struct extent_io_tree *tree;
1000 tree = &BTRFS_I(page->mapping->host)->io_tree;
1001 return extent_read_full_page(tree, page, btree_get_extent, 0);
1002}
1003
1004static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1005{
1006 if (PageWriteback(page) || PageDirty(page))
1007 return 0;
1008
1009 return try_release_extent_buffer(page);
1010}
1011
1012static void btree_invalidatepage(struct page *page, unsigned int offset,
1013 unsigned int length)
1014{
1015 struct extent_io_tree *tree;
1016 tree = &BTRFS_I(page->mapping->host)->io_tree;
1017 extent_invalidatepage(tree, page, offset);
1018 btree_releasepage(page, GFP_NOFS);
1019 if (PagePrivate(page)) {
1020 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1021 "page private not zero on page %llu",
1022 (unsigned long long)page_offset(page));
1023 ClearPagePrivate(page);
1024 set_page_private(page, 0);
1025 page_cache_release(page);
1026 }
1027}
1028
1029static int btree_set_page_dirty(struct page *page)
1030{
1031#ifdef DEBUG
1032 struct extent_buffer *eb;
1033
1034 BUG_ON(!PagePrivate(page));
1035 eb = (struct extent_buffer *)page->private;
1036 BUG_ON(!eb);
1037 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 BUG_ON(!atomic_read(&eb->refs));
1039 btrfs_assert_tree_locked(eb);
1040#endif
1041 return __set_page_dirty_nobuffers(page);
1042}
1043
1044static const struct address_space_operations btree_aops = {
1045 .readpage = btree_readpage,
1046 .writepages = btree_writepages,
1047 .releasepage = btree_releasepage,
1048 .invalidatepage = btree_invalidatepage,
1049#ifdef CONFIG_MIGRATION
1050 .migratepage = btree_migratepage,
1051#endif
1052 .set_page_dirty = btree_set_page_dirty,
1053};
1054
1055int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1056 u64 parent_transid)
1057{
1058 struct extent_buffer *buf = NULL;
1059 struct inode *btree_inode = root->fs_info->btree_inode;
1060 int ret = 0;
1061
1062 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1063 if (!buf)
1064 return 0;
1065 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1066 buf, 0, WAIT_NONE, btree_get_extent, 0);
1067 free_extent_buffer(buf);
1068 return ret;
1069}
1070
1071int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1072 int mirror_num, struct extent_buffer **eb)
1073{
1074 struct extent_buffer *buf = NULL;
1075 struct inode *btree_inode = root->fs_info->btree_inode;
1076 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1077 int ret;
1078
1079 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1080 if (!buf)
1081 return 0;
1082
1083 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1084
1085 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1086 btree_get_extent, mirror_num);
1087 if (ret) {
1088 free_extent_buffer(buf);
1089 return ret;
1090 }
1091
1092 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1093 free_extent_buffer(buf);
1094 return -EIO;
1095 } else if (extent_buffer_uptodate(buf)) {
1096 *eb = buf;
1097 } else {
1098 free_extent_buffer(buf);
1099 }
1100 return 0;
1101}
1102
1103struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1104 u64 bytenr, u32 blocksize)
1105{
1106 return find_extent_buffer(root->fs_info, bytenr);
1107}
1108
1109struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1110 u64 bytenr, u32 blocksize)
1111{
1112#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1113 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state)))
1114 return alloc_test_extent_buffer(root->fs_info, bytenr,
1115 blocksize);
1116#endif
1117 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1118}
1119
1120
1121int btrfs_write_tree_block(struct extent_buffer *buf)
1122{
1123 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1124 buf->start + buf->len - 1);
1125}
1126
1127int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1128{
1129 return filemap_fdatawait_range(buf->pages[0]->mapping,
1130 buf->start, buf->start + buf->len - 1);
1131}
1132
1133struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1134 u32 blocksize, u64 parent_transid)
1135{
1136 struct extent_buffer *buf = NULL;
1137 int ret;
1138
1139 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1140 if (!buf)
1141 return NULL;
1142
1143 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1144 if (ret) {
1145 free_extent_buffer(buf);
1146 return NULL;
1147 }
1148 return buf;
1149
1150}
1151
1152void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1153 struct extent_buffer *buf)
1154{
1155 struct btrfs_fs_info *fs_info = root->fs_info;
1156
1157 if (btrfs_header_generation(buf) ==
1158 fs_info->running_transaction->transid) {
1159 btrfs_assert_tree_locked(buf);
1160
1161 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1162 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1163 -buf->len,
1164 fs_info->dirty_metadata_batch);
1165 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1166 btrfs_set_lock_blocking(buf);
1167 clear_extent_buffer_dirty(buf);
1168 }
1169 }
1170}
1171
1172static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1173{
1174 struct btrfs_subvolume_writers *writers;
1175 int ret;
1176
1177 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1178 if (!writers)
1179 return ERR_PTR(-ENOMEM);
1180
1181 ret = percpu_counter_init(&writers->counter, 0);
1182 if (ret < 0) {
1183 kfree(writers);
1184 return ERR_PTR(ret);
1185 }
1186
1187 init_waitqueue_head(&writers->wait);
1188 return writers;
1189}
1190
1191static void
1192btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1193{
1194 percpu_counter_destroy(&writers->counter);
1195 kfree(writers);
1196}
1197
1198static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1199 u32 stripesize, struct btrfs_root *root,
1200 struct btrfs_fs_info *fs_info,
1201 u64 objectid)
1202{
1203 root->node = NULL;
1204 root->commit_root = NULL;
1205 root->sectorsize = sectorsize;
1206 root->nodesize = nodesize;
1207 root->leafsize = leafsize;
1208 root->stripesize = stripesize;
1209 root->state = 0;
1210 root->orphan_cleanup_state = 0;
1211
1212 root->objectid = objectid;
1213 root->last_trans = 0;
1214 root->highest_objectid = 0;
1215 root->nr_delalloc_inodes = 0;
1216 root->nr_ordered_extents = 0;
1217 root->name = NULL;
1218 root->inode_tree = RB_ROOT;
1219 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1220 root->block_rsv = NULL;
1221 root->orphan_block_rsv = NULL;
1222
1223 INIT_LIST_HEAD(&root->dirty_list);
1224 INIT_LIST_HEAD(&root->root_list);
1225 INIT_LIST_HEAD(&root->delalloc_inodes);
1226 INIT_LIST_HEAD(&root->delalloc_root);
1227 INIT_LIST_HEAD(&root->ordered_extents);
1228 INIT_LIST_HEAD(&root->ordered_root);
1229 INIT_LIST_HEAD(&root->logged_list[0]);
1230 INIT_LIST_HEAD(&root->logged_list[1]);
1231 spin_lock_init(&root->orphan_lock);
1232 spin_lock_init(&root->inode_lock);
1233 spin_lock_init(&root->delalloc_lock);
1234 spin_lock_init(&root->ordered_extent_lock);
1235 spin_lock_init(&root->accounting_lock);
1236 spin_lock_init(&root->log_extents_lock[0]);
1237 spin_lock_init(&root->log_extents_lock[1]);
1238 mutex_init(&root->objectid_mutex);
1239 mutex_init(&root->log_mutex);
1240 mutex_init(&root->ordered_extent_mutex);
1241 mutex_init(&root->delalloc_mutex);
1242 init_waitqueue_head(&root->log_writer_wait);
1243 init_waitqueue_head(&root->log_commit_wait[0]);
1244 init_waitqueue_head(&root->log_commit_wait[1]);
1245 INIT_LIST_HEAD(&root->log_ctxs[0]);
1246 INIT_LIST_HEAD(&root->log_ctxs[1]);
1247 atomic_set(&root->log_commit[0], 0);
1248 atomic_set(&root->log_commit[1], 0);
1249 atomic_set(&root->log_writers, 0);
1250 atomic_set(&root->log_batch, 0);
1251 atomic_set(&root->orphan_inodes, 0);
1252 atomic_set(&root->refs, 1);
1253 atomic_set(&root->will_be_snapshoted, 0);
1254 root->log_transid = 0;
1255 root->log_transid_committed = -1;
1256 root->last_log_commit = 0;
1257 if (fs_info)
1258 extent_io_tree_init(&root->dirty_log_pages,
1259 fs_info->btree_inode->i_mapping);
1260
1261 memset(&root->root_key, 0, sizeof(root->root_key));
1262 memset(&root->root_item, 0, sizeof(root->root_item));
1263 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1264 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1265 if (fs_info)
1266 root->defrag_trans_start = fs_info->generation;
1267 else
1268 root->defrag_trans_start = 0;
1269 init_completion(&root->kobj_unregister);
1270 root->root_key.objectid = objectid;
1271 root->anon_dev = 0;
1272
1273 spin_lock_init(&root->root_item_lock);
1274}
1275
1276static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1277{
1278 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1279 if (root)
1280 root->fs_info = fs_info;
1281 return root;
1282}
1283
1284#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1285/* Should only be used by the testing infrastructure */
1286struct btrfs_root *btrfs_alloc_dummy_root(void)
1287{
1288 struct btrfs_root *root;
1289
1290 root = btrfs_alloc_root(NULL);
1291 if (!root)
1292 return ERR_PTR(-ENOMEM);
1293 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1294 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1295 root->alloc_bytenr = 0;
1296
1297 return root;
1298}
1299#endif
1300
1301struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1302 struct btrfs_fs_info *fs_info,
1303 u64 objectid)
1304{
1305 struct extent_buffer *leaf;
1306 struct btrfs_root *tree_root = fs_info->tree_root;
1307 struct btrfs_root *root;
1308 struct btrfs_key key;
1309 int ret = 0;
1310 uuid_le uuid;
1311
1312 root = btrfs_alloc_root(fs_info);
1313 if (!root)
1314 return ERR_PTR(-ENOMEM);
1315
1316 __setup_root(tree_root->nodesize, tree_root->leafsize,
1317 tree_root->sectorsize, tree_root->stripesize,
1318 root, fs_info, objectid);
1319 root->root_key.objectid = objectid;
1320 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1321 root->root_key.offset = 0;
1322
1323 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1324 0, objectid, NULL, 0, 0, 0);
1325 if (IS_ERR(leaf)) {
1326 ret = PTR_ERR(leaf);
1327 leaf = NULL;
1328 goto fail;
1329 }
1330
1331 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1332 btrfs_set_header_bytenr(leaf, leaf->start);
1333 btrfs_set_header_generation(leaf, trans->transid);
1334 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1335 btrfs_set_header_owner(leaf, objectid);
1336 root->node = leaf;
1337
1338 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1339 BTRFS_FSID_SIZE);
1340 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1341 btrfs_header_chunk_tree_uuid(leaf),
1342 BTRFS_UUID_SIZE);
1343 btrfs_mark_buffer_dirty(leaf);
1344
1345 root->commit_root = btrfs_root_node(root);
1346 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1347
1348 root->root_item.flags = 0;
1349 root->root_item.byte_limit = 0;
1350 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1351 btrfs_set_root_generation(&root->root_item, trans->transid);
1352 btrfs_set_root_level(&root->root_item, 0);
1353 btrfs_set_root_refs(&root->root_item, 1);
1354 btrfs_set_root_used(&root->root_item, leaf->len);
1355 btrfs_set_root_last_snapshot(&root->root_item, 0);
1356 btrfs_set_root_dirid(&root->root_item, 0);
1357 uuid_le_gen(&uuid);
1358 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1359 root->root_item.drop_level = 0;
1360
1361 key.objectid = objectid;
1362 key.type = BTRFS_ROOT_ITEM_KEY;
1363 key.offset = 0;
1364 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1365 if (ret)
1366 goto fail;
1367
1368 btrfs_tree_unlock(leaf);
1369
1370 return root;
1371
1372fail:
1373 if (leaf) {
1374 btrfs_tree_unlock(leaf);
1375 free_extent_buffer(root->commit_root);
1376 free_extent_buffer(leaf);
1377 }
1378 kfree(root);
1379
1380 return ERR_PTR(ret);
1381}
1382
1383static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1384 struct btrfs_fs_info *fs_info)
1385{
1386 struct btrfs_root *root;
1387 struct btrfs_root *tree_root = fs_info->tree_root;
1388 struct extent_buffer *leaf;
1389
1390 root = btrfs_alloc_root(fs_info);
1391 if (!root)
1392 return ERR_PTR(-ENOMEM);
1393
1394 __setup_root(tree_root->nodesize, tree_root->leafsize,
1395 tree_root->sectorsize, tree_root->stripesize,
1396 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1397
1398 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1399 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1400 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1401
1402 /*
1403 * DON'T set REF_COWS for log trees
1404 *
1405 * log trees do not get reference counted because they go away
1406 * before a real commit is actually done. They do store pointers
1407 * to file data extents, and those reference counts still get
1408 * updated (along with back refs to the log tree).
1409 */
1410
1411 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1412 BTRFS_TREE_LOG_OBJECTID, NULL,
1413 0, 0, 0);
1414 if (IS_ERR(leaf)) {
1415 kfree(root);
1416 return ERR_CAST(leaf);
1417 }
1418
1419 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1420 btrfs_set_header_bytenr(leaf, leaf->start);
1421 btrfs_set_header_generation(leaf, trans->transid);
1422 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1423 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1424 root->node = leaf;
1425
1426 write_extent_buffer(root->node, root->fs_info->fsid,
1427 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1428 btrfs_mark_buffer_dirty(root->node);
1429 btrfs_tree_unlock(root->node);
1430 return root;
1431}
1432
1433int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1434 struct btrfs_fs_info *fs_info)
1435{
1436 struct btrfs_root *log_root;
1437
1438 log_root = alloc_log_tree(trans, fs_info);
1439 if (IS_ERR(log_root))
1440 return PTR_ERR(log_root);
1441 WARN_ON(fs_info->log_root_tree);
1442 fs_info->log_root_tree = log_root;
1443 return 0;
1444}
1445
1446int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1447 struct btrfs_root *root)
1448{
1449 struct btrfs_root *log_root;
1450 struct btrfs_inode_item *inode_item;
1451
1452 log_root = alloc_log_tree(trans, root->fs_info);
1453 if (IS_ERR(log_root))
1454 return PTR_ERR(log_root);
1455
1456 log_root->last_trans = trans->transid;
1457 log_root->root_key.offset = root->root_key.objectid;
1458
1459 inode_item = &log_root->root_item.inode;
1460 btrfs_set_stack_inode_generation(inode_item, 1);
1461 btrfs_set_stack_inode_size(inode_item, 3);
1462 btrfs_set_stack_inode_nlink(inode_item, 1);
1463 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1464 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1465
1466 btrfs_set_root_node(&log_root->root_item, log_root->node);
1467
1468 WARN_ON(root->log_root);
1469 root->log_root = log_root;
1470 root->log_transid = 0;
1471 root->log_transid_committed = -1;
1472 root->last_log_commit = 0;
1473 return 0;
1474}
1475
1476static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1477 struct btrfs_key *key)
1478{
1479 struct btrfs_root *root;
1480 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1481 struct btrfs_path *path;
1482 u64 generation;
1483 u32 blocksize;
1484 int ret;
1485
1486 path = btrfs_alloc_path();
1487 if (!path)
1488 return ERR_PTR(-ENOMEM);
1489
1490 root = btrfs_alloc_root(fs_info);
1491 if (!root) {
1492 ret = -ENOMEM;
1493 goto alloc_fail;
1494 }
1495
1496 __setup_root(tree_root->nodesize, tree_root->leafsize,
1497 tree_root->sectorsize, tree_root->stripesize,
1498 root, fs_info, key->objectid);
1499
1500 ret = btrfs_find_root(tree_root, key, path,
1501 &root->root_item, &root->root_key);
1502 if (ret) {
1503 if (ret > 0)
1504 ret = -ENOENT;
1505 goto find_fail;
1506 }
1507
1508 generation = btrfs_root_generation(&root->root_item);
1509 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1510 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1511 blocksize, generation);
1512 if (!root->node) {
1513 ret = -ENOMEM;
1514 goto find_fail;
1515 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1516 ret = -EIO;
1517 goto read_fail;
1518 }
1519 root->commit_root = btrfs_root_node(root);
1520out:
1521 btrfs_free_path(path);
1522 return root;
1523
1524read_fail:
1525 free_extent_buffer(root->node);
1526find_fail:
1527 kfree(root);
1528alloc_fail:
1529 root = ERR_PTR(ret);
1530 goto out;
1531}
1532
1533struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1534 struct btrfs_key *location)
1535{
1536 struct btrfs_root *root;
1537
1538 root = btrfs_read_tree_root(tree_root, location);
1539 if (IS_ERR(root))
1540 return root;
1541
1542 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1543 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1544 btrfs_check_and_init_root_item(&root->root_item);
1545 }
1546
1547 return root;
1548}
1549
1550int btrfs_init_fs_root(struct btrfs_root *root)
1551{
1552 int ret;
1553 struct btrfs_subvolume_writers *writers;
1554
1555 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1556 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1557 GFP_NOFS);
1558 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1559 ret = -ENOMEM;
1560 goto fail;
1561 }
1562
1563 writers = btrfs_alloc_subvolume_writers();
1564 if (IS_ERR(writers)) {
1565 ret = PTR_ERR(writers);
1566 goto fail;
1567 }
1568 root->subv_writers = writers;
1569
1570 btrfs_init_free_ino_ctl(root);
1571 spin_lock_init(&root->cache_lock);
1572 init_waitqueue_head(&root->cache_wait);
1573
1574 ret = get_anon_bdev(&root->anon_dev);
1575 if (ret)
1576 goto free_writers;
1577 return 0;
1578
1579free_writers:
1580 btrfs_free_subvolume_writers(root->subv_writers);
1581fail:
1582 kfree(root->free_ino_ctl);
1583 kfree(root->free_ino_pinned);
1584 return ret;
1585}
1586
1587static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1588 u64 root_id)
1589{
1590 struct btrfs_root *root;
1591
1592 spin_lock(&fs_info->fs_roots_radix_lock);
1593 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1594 (unsigned long)root_id);
1595 spin_unlock(&fs_info->fs_roots_radix_lock);
1596 return root;
1597}
1598
1599int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1600 struct btrfs_root *root)
1601{
1602 int ret;
1603
1604 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1605 if (ret)
1606 return ret;
1607
1608 spin_lock(&fs_info->fs_roots_radix_lock);
1609 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1610 (unsigned long)root->root_key.objectid,
1611 root);
1612 if (ret == 0)
1613 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1614 spin_unlock(&fs_info->fs_roots_radix_lock);
1615 radix_tree_preload_end();
1616
1617 return ret;
1618}
1619
1620struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1621 struct btrfs_key *location,
1622 bool check_ref)
1623{
1624 struct btrfs_root *root;
1625 int ret;
1626
1627 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1628 return fs_info->tree_root;
1629 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1630 return fs_info->extent_root;
1631 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1632 return fs_info->chunk_root;
1633 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1634 return fs_info->dev_root;
1635 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1636 return fs_info->csum_root;
1637 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1638 return fs_info->quota_root ? fs_info->quota_root :
1639 ERR_PTR(-ENOENT);
1640 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1641 return fs_info->uuid_root ? fs_info->uuid_root :
1642 ERR_PTR(-ENOENT);
1643again:
1644 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1645 if (root) {
1646 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1647 return ERR_PTR(-ENOENT);
1648 return root;
1649 }
1650
1651 root = btrfs_read_fs_root(fs_info->tree_root, location);
1652 if (IS_ERR(root))
1653 return root;
1654
1655 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1656 ret = -ENOENT;
1657 goto fail;
1658 }
1659
1660 ret = btrfs_init_fs_root(root);
1661 if (ret)
1662 goto fail;
1663
1664 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1665 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1666 if (ret < 0)
1667 goto fail;
1668 if (ret == 0)
1669 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1670
1671 ret = btrfs_insert_fs_root(fs_info, root);
1672 if (ret) {
1673 if (ret == -EEXIST) {
1674 free_fs_root(root);
1675 goto again;
1676 }
1677 goto fail;
1678 }
1679 return root;
1680fail:
1681 free_fs_root(root);
1682 return ERR_PTR(ret);
1683}
1684
1685static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1686{
1687 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1688 int ret = 0;
1689 struct btrfs_device *device;
1690 struct backing_dev_info *bdi;
1691
1692 rcu_read_lock();
1693 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1694 if (!device->bdev)
1695 continue;
1696 bdi = blk_get_backing_dev_info(device->bdev);
1697 if (bdi && bdi_congested(bdi, bdi_bits)) {
1698 ret = 1;
1699 break;
1700 }
1701 }
1702 rcu_read_unlock();
1703 return ret;
1704}
1705
1706/*
1707 * If this fails, caller must call bdi_destroy() to get rid of the
1708 * bdi again.
1709 */
1710static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1711{
1712 int err;
1713
1714 bdi->capabilities = BDI_CAP_MAP_COPY;
1715 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1716 if (err)
1717 return err;
1718
1719 bdi->ra_pages = default_backing_dev_info.ra_pages;
1720 bdi->congested_fn = btrfs_congested_fn;
1721 bdi->congested_data = info;
1722 return 0;
1723}
1724
1725/*
1726 * called by the kthread helper functions to finally call the bio end_io
1727 * functions. This is where read checksum verification actually happens
1728 */
1729static void end_workqueue_fn(struct btrfs_work *work)
1730{
1731 struct bio *bio;
1732 struct end_io_wq *end_io_wq;
1733 int error;
1734
1735 end_io_wq = container_of(work, struct end_io_wq, work);
1736 bio = end_io_wq->bio;
1737
1738 error = end_io_wq->error;
1739 bio->bi_private = end_io_wq->private;
1740 bio->bi_end_io = end_io_wq->end_io;
1741 kfree(end_io_wq);
1742 bio_endio_nodec(bio, error);
1743}
1744
1745static int cleaner_kthread(void *arg)
1746{
1747 struct btrfs_root *root = arg;
1748 int again;
1749
1750 do {
1751 again = 0;
1752
1753 /* Make the cleaner go to sleep early. */
1754 if (btrfs_need_cleaner_sleep(root))
1755 goto sleep;
1756
1757 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1758 goto sleep;
1759
1760 /*
1761 * Avoid the problem that we change the status of the fs
1762 * during the above check and trylock.
1763 */
1764 if (btrfs_need_cleaner_sleep(root)) {
1765 mutex_unlock(&root->fs_info->cleaner_mutex);
1766 goto sleep;
1767 }
1768
1769 btrfs_run_delayed_iputs(root);
1770 again = btrfs_clean_one_deleted_snapshot(root);
1771 mutex_unlock(&root->fs_info->cleaner_mutex);
1772
1773 /*
1774 * The defragger has dealt with the R/O remount and umount,
1775 * needn't do anything special here.
1776 */
1777 btrfs_run_defrag_inodes(root->fs_info);
1778sleep:
1779 if (!try_to_freeze() && !again) {
1780 set_current_state(TASK_INTERRUPTIBLE);
1781 if (!kthread_should_stop())
1782 schedule();
1783 __set_current_state(TASK_RUNNING);
1784 }
1785 } while (!kthread_should_stop());
1786 return 0;
1787}
1788
1789static int transaction_kthread(void *arg)
1790{
1791 struct btrfs_root *root = arg;
1792 struct btrfs_trans_handle *trans;
1793 struct btrfs_transaction *cur;
1794 u64 transid;
1795 unsigned long now;
1796 unsigned long delay;
1797 bool cannot_commit;
1798
1799 do {
1800 cannot_commit = false;
1801 delay = HZ * root->fs_info->commit_interval;
1802 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1803
1804 spin_lock(&root->fs_info->trans_lock);
1805 cur = root->fs_info->running_transaction;
1806 if (!cur) {
1807 spin_unlock(&root->fs_info->trans_lock);
1808 goto sleep;
1809 }
1810
1811 now = get_seconds();
1812 if (cur->state < TRANS_STATE_BLOCKED &&
1813 (now < cur->start_time ||
1814 now - cur->start_time < root->fs_info->commit_interval)) {
1815 spin_unlock(&root->fs_info->trans_lock);
1816 delay = HZ * 5;
1817 goto sleep;
1818 }
1819 transid = cur->transid;
1820 spin_unlock(&root->fs_info->trans_lock);
1821
1822 /* If the file system is aborted, this will always fail. */
1823 trans = btrfs_attach_transaction(root);
1824 if (IS_ERR(trans)) {
1825 if (PTR_ERR(trans) != -ENOENT)
1826 cannot_commit = true;
1827 goto sleep;
1828 }
1829 if (transid == trans->transid) {
1830 btrfs_commit_transaction(trans, root);
1831 } else {
1832 btrfs_end_transaction(trans, root);
1833 }
1834sleep:
1835 wake_up_process(root->fs_info->cleaner_kthread);
1836 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1837
1838 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1839 &root->fs_info->fs_state)))
1840 btrfs_cleanup_transaction(root);
1841 if (!try_to_freeze()) {
1842 set_current_state(TASK_INTERRUPTIBLE);
1843 if (!kthread_should_stop() &&
1844 (!btrfs_transaction_blocked(root->fs_info) ||
1845 cannot_commit))
1846 schedule_timeout(delay);
1847 __set_current_state(TASK_RUNNING);
1848 }
1849 } while (!kthread_should_stop());
1850 return 0;
1851}
1852
1853/*
1854 * this will find the highest generation in the array of
1855 * root backups. The index of the highest array is returned,
1856 * or -1 if we can't find anything.
1857 *
1858 * We check to make sure the array is valid by comparing the
1859 * generation of the latest root in the array with the generation
1860 * in the super block. If they don't match we pitch it.
1861 */
1862static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1863{
1864 u64 cur;
1865 int newest_index = -1;
1866 struct btrfs_root_backup *root_backup;
1867 int i;
1868
1869 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1870 root_backup = info->super_copy->super_roots + i;
1871 cur = btrfs_backup_tree_root_gen(root_backup);
1872 if (cur == newest_gen)
1873 newest_index = i;
1874 }
1875
1876 /* check to see if we actually wrapped around */
1877 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1878 root_backup = info->super_copy->super_roots;
1879 cur = btrfs_backup_tree_root_gen(root_backup);
1880 if (cur == newest_gen)
1881 newest_index = 0;
1882 }
1883 return newest_index;
1884}
1885
1886
1887/*
1888 * find the oldest backup so we know where to store new entries
1889 * in the backup array. This will set the backup_root_index
1890 * field in the fs_info struct
1891 */
1892static void find_oldest_super_backup(struct btrfs_fs_info *info,
1893 u64 newest_gen)
1894{
1895 int newest_index = -1;
1896
1897 newest_index = find_newest_super_backup(info, newest_gen);
1898 /* if there was garbage in there, just move along */
1899 if (newest_index == -1) {
1900 info->backup_root_index = 0;
1901 } else {
1902 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1903 }
1904}
1905
1906/*
1907 * copy all the root pointers into the super backup array.
1908 * this will bump the backup pointer by one when it is
1909 * done
1910 */
1911static void backup_super_roots(struct btrfs_fs_info *info)
1912{
1913 int next_backup;
1914 struct btrfs_root_backup *root_backup;
1915 int last_backup;
1916
1917 next_backup = info->backup_root_index;
1918 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1919 BTRFS_NUM_BACKUP_ROOTS;
1920
1921 /*
1922 * just overwrite the last backup if we're at the same generation
1923 * this happens only at umount
1924 */
1925 root_backup = info->super_for_commit->super_roots + last_backup;
1926 if (btrfs_backup_tree_root_gen(root_backup) ==
1927 btrfs_header_generation(info->tree_root->node))
1928 next_backup = last_backup;
1929
1930 root_backup = info->super_for_commit->super_roots + next_backup;
1931
1932 /*
1933 * make sure all of our padding and empty slots get zero filled
1934 * regardless of which ones we use today
1935 */
1936 memset(root_backup, 0, sizeof(*root_backup));
1937
1938 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1939
1940 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1941 btrfs_set_backup_tree_root_gen(root_backup,
1942 btrfs_header_generation(info->tree_root->node));
1943
1944 btrfs_set_backup_tree_root_level(root_backup,
1945 btrfs_header_level(info->tree_root->node));
1946
1947 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1948 btrfs_set_backup_chunk_root_gen(root_backup,
1949 btrfs_header_generation(info->chunk_root->node));
1950 btrfs_set_backup_chunk_root_level(root_backup,
1951 btrfs_header_level(info->chunk_root->node));
1952
1953 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1954 btrfs_set_backup_extent_root_gen(root_backup,
1955 btrfs_header_generation(info->extent_root->node));
1956 btrfs_set_backup_extent_root_level(root_backup,
1957 btrfs_header_level(info->extent_root->node));
1958
1959 /*
1960 * we might commit during log recovery, which happens before we set
1961 * the fs_root. Make sure it is valid before we fill it in.
1962 */
1963 if (info->fs_root && info->fs_root->node) {
1964 btrfs_set_backup_fs_root(root_backup,
1965 info->fs_root->node->start);
1966 btrfs_set_backup_fs_root_gen(root_backup,
1967 btrfs_header_generation(info->fs_root->node));
1968 btrfs_set_backup_fs_root_level(root_backup,
1969 btrfs_header_level(info->fs_root->node));
1970 }
1971
1972 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1973 btrfs_set_backup_dev_root_gen(root_backup,
1974 btrfs_header_generation(info->dev_root->node));
1975 btrfs_set_backup_dev_root_level(root_backup,
1976 btrfs_header_level(info->dev_root->node));
1977
1978 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1979 btrfs_set_backup_csum_root_gen(root_backup,
1980 btrfs_header_generation(info->csum_root->node));
1981 btrfs_set_backup_csum_root_level(root_backup,
1982 btrfs_header_level(info->csum_root->node));
1983
1984 btrfs_set_backup_total_bytes(root_backup,
1985 btrfs_super_total_bytes(info->super_copy));
1986 btrfs_set_backup_bytes_used(root_backup,
1987 btrfs_super_bytes_used(info->super_copy));
1988 btrfs_set_backup_num_devices(root_backup,
1989 btrfs_super_num_devices(info->super_copy));
1990
1991 /*
1992 * if we don't copy this out to the super_copy, it won't get remembered
1993 * for the next commit
1994 */
1995 memcpy(&info->super_copy->super_roots,
1996 &info->super_for_commit->super_roots,
1997 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1998}
1999
2000/*
2001 * this copies info out of the root backup array and back into
2002 * the in-memory super block. It is meant to help iterate through
2003 * the array, so you send it the number of backups you've already
2004 * tried and the last backup index you used.
2005 *
2006 * this returns -1 when it has tried all the backups
2007 */
2008static noinline int next_root_backup(struct btrfs_fs_info *info,
2009 struct btrfs_super_block *super,
2010 int *num_backups_tried, int *backup_index)
2011{
2012 struct btrfs_root_backup *root_backup;
2013 int newest = *backup_index;
2014
2015 if (*num_backups_tried == 0) {
2016 u64 gen = btrfs_super_generation(super);
2017
2018 newest = find_newest_super_backup(info, gen);
2019 if (newest == -1)
2020 return -1;
2021
2022 *backup_index = newest;
2023 *num_backups_tried = 1;
2024 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2025 /* we've tried all the backups, all done */
2026 return -1;
2027 } else {
2028 /* jump to the next oldest backup */
2029 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2030 BTRFS_NUM_BACKUP_ROOTS;
2031 *backup_index = newest;
2032 *num_backups_tried += 1;
2033 }
2034 root_backup = super->super_roots + newest;
2035
2036 btrfs_set_super_generation(super,
2037 btrfs_backup_tree_root_gen(root_backup));
2038 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2039 btrfs_set_super_root_level(super,
2040 btrfs_backup_tree_root_level(root_backup));
2041 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2042
2043 /*
2044 * fixme: the total bytes and num_devices need to match or we should
2045 * need a fsck
2046 */
2047 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2048 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2049 return 0;
2050}
2051
2052/* helper to cleanup workers */
2053static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2054{
2055 btrfs_destroy_workqueue(fs_info->fixup_workers);
2056 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2057 btrfs_destroy_workqueue(fs_info->workers);
2058 btrfs_destroy_workqueue(fs_info->endio_workers);
2059 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2060 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2061 btrfs_destroy_workqueue(fs_info->rmw_workers);
2062 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2063 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2064 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2065 btrfs_destroy_workqueue(fs_info->submit_workers);
2066 btrfs_destroy_workqueue(fs_info->delayed_workers);
2067 btrfs_destroy_workqueue(fs_info->caching_workers);
2068 btrfs_destroy_workqueue(fs_info->readahead_workers);
2069 btrfs_destroy_workqueue(fs_info->flush_workers);
2070 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2071 btrfs_destroy_workqueue(fs_info->extent_workers);
2072}
2073
2074static void free_root_extent_buffers(struct btrfs_root *root)
2075{
2076 if (root) {
2077 free_extent_buffer(root->node);
2078 free_extent_buffer(root->commit_root);
2079 root->node = NULL;
2080 root->commit_root = NULL;
2081 }
2082}
2083
2084/* helper to cleanup tree roots */
2085static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2086{
2087 free_root_extent_buffers(info->tree_root);
2088
2089 free_root_extent_buffers(info->dev_root);
2090 free_root_extent_buffers(info->extent_root);
2091 free_root_extent_buffers(info->csum_root);
2092 free_root_extent_buffers(info->quota_root);
2093 free_root_extent_buffers(info->uuid_root);
2094 if (chunk_root)
2095 free_root_extent_buffers(info->chunk_root);
2096}
2097
2098void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2099{
2100 int ret;
2101 struct btrfs_root *gang[8];
2102 int i;
2103
2104 while (!list_empty(&fs_info->dead_roots)) {
2105 gang[0] = list_entry(fs_info->dead_roots.next,
2106 struct btrfs_root, root_list);
2107 list_del(&gang[0]->root_list);
2108
2109 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2110 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2111 } else {
2112 free_extent_buffer(gang[0]->node);
2113 free_extent_buffer(gang[0]->commit_root);
2114 btrfs_put_fs_root(gang[0]);
2115 }
2116 }
2117
2118 while (1) {
2119 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2120 (void **)gang, 0,
2121 ARRAY_SIZE(gang));
2122 if (!ret)
2123 break;
2124 for (i = 0; i < ret; i++)
2125 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2126 }
2127
2128 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2129 btrfs_free_log_root_tree(NULL, fs_info);
2130 btrfs_destroy_pinned_extent(fs_info->tree_root,
2131 fs_info->pinned_extents);
2132 }
2133}
2134
2135int open_ctree(struct super_block *sb,
2136 struct btrfs_fs_devices *fs_devices,
2137 char *options)
2138{
2139 u32 sectorsize;
2140 u32 nodesize;
2141 u32 leafsize;
2142 u32 blocksize;
2143 u32 stripesize;
2144 u64 generation;
2145 u64 features;
2146 struct btrfs_key location;
2147 struct buffer_head *bh;
2148 struct btrfs_super_block *disk_super;
2149 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2150 struct btrfs_root *tree_root;
2151 struct btrfs_root *extent_root;
2152 struct btrfs_root *csum_root;
2153 struct btrfs_root *chunk_root;
2154 struct btrfs_root *dev_root;
2155 struct btrfs_root *quota_root;
2156 struct btrfs_root *uuid_root;
2157 struct btrfs_root *log_tree_root;
2158 int ret;
2159 int err = -EINVAL;
2160 int num_backups_tried = 0;
2161 int backup_index = 0;
2162 int max_active;
2163 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2164 bool create_uuid_tree;
2165 bool check_uuid_tree;
2166
2167 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2168 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2169 if (!tree_root || !chunk_root) {
2170 err = -ENOMEM;
2171 goto fail;
2172 }
2173
2174 ret = init_srcu_struct(&fs_info->subvol_srcu);
2175 if (ret) {
2176 err = ret;
2177 goto fail;
2178 }
2179
2180 ret = setup_bdi(fs_info, &fs_info->bdi);
2181 if (ret) {
2182 err = ret;
2183 goto fail_srcu;
2184 }
2185
2186 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2187 if (ret) {
2188 err = ret;
2189 goto fail_bdi;
2190 }
2191 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2192 (1 + ilog2(nr_cpu_ids));
2193
2194 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2195 if (ret) {
2196 err = ret;
2197 goto fail_dirty_metadata_bytes;
2198 }
2199
2200 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2201 if (ret) {
2202 err = ret;
2203 goto fail_delalloc_bytes;
2204 }
2205
2206 fs_info->btree_inode = new_inode(sb);
2207 if (!fs_info->btree_inode) {
2208 err = -ENOMEM;
2209 goto fail_bio_counter;
2210 }
2211
2212 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2213
2214 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2215 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2216 INIT_LIST_HEAD(&fs_info->trans_list);
2217 INIT_LIST_HEAD(&fs_info->dead_roots);
2218 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2219 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2220 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2221 spin_lock_init(&fs_info->delalloc_root_lock);
2222 spin_lock_init(&fs_info->trans_lock);
2223 spin_lock_init(&fs_info->fs_roots_radix_lock);
2224 spin_lock_init(&fs_info->delayed_iput_lock);
2225 spin_lock_init(&fs_info->defrag_inodes_lock);
2226 spin_lock_init(&fs_info->free_chunk_lock);
2227 spin_lock_init(&fs_info->tree_mod_seq_lock);
2228 spin_lock_init(&fs_info->super_lock);
2229 spin_lock_init(&fs_info->qgroup_op_lock);
2230 spin_lock_init(&fs_info->buffer_lock);
2231 rwlock_init(&fs_info->tree_mod_log_lock);
2232 mutex_init(&fs_info->reloc_mutex);
2233 mutex_init(&fs_info->delalloc_root_mutex);
2234 seqlock_init(&fs_info->profiles_lock);
2235
2236 init_completion(&fs_info->kobj_unregister);
2237 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2238 INIT_LIST_HEAD(&fs_info->space_info);
2239 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2240 btrfs_mapping_init(&fs_info->mapping_tree);
2241 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2242 BTRFS_BLOCK_RSV_GLOBAL);
2243 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2244 BTRFS_BLOCK_RSV_DELALLOC);
2245 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2246 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2247 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2248 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2249 BTRFS_BLOCK_RSV_DELOPS);
2250 atomic_set(&fs_info->nr_async_submits, 0);
2251 atomic_set(&fs_info->async_delalloc_pages, 0);
2252 atomic_set(&fs_info->async_submit_draining, 0);
2253 atomic_set(&fs_info->nr_async_bios, 0);
2254 atomic_set(&fs_info->defrag_running, 0);
2255 atomic_set(&fs_info->qgroup_op_seq, 0);
2256 atomic64_set(&fs_info->tree_mod_seq, 0);
2257 fs_info->sb = sb;
2258 fs_info->max_inline = 8192 * 1024;
2259 fs_info->metadata_ratio = 0;
2260 fs_info->defrag_inodes = RB_ROOT;
2261 fs_info->free_chunk_space = 0;
2262 fs_info->tree_mod_log = RB_ROOT;
2263 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2264 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2265 /* readahead state */
2266 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2267 spin_lock_init(&fs_info->reada_lock);
2268
2269 fs_info->thread_pool_size = min_t(unsigned long,
2270 num_online_cpus() + 2, 8);
2271
2272 INIT_LIST_HEAD(&fs_info->ordered_roots);
2273 spin_lock_init(&fs_info->ordered_root_lock);
2274 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2275 GFP_NOFS);
2276 if (!fs_info->delayed_root) {
2277 err = -ENOMEM;
2278 goto fail_iput;
2279 }
2280 btrfs_init_delayed_root(fs_info->delayed_root);
2281
2282 mutex_init(&fs_info->scrub_lock);
2283 atomic_set(&fs_info->scrubs_running, 0);
2284 atomic_set(&fs_info->scrub_pause_req, 0);
2285 atomic_set(&fs_info->scrubs_paused, 0);
2286 atomic_set(&fs_info->scrub_cancel_req, 0);
2287 init_waitqueue_head(&fs_info->replace_wait);
2288 init_waitqueue_head(&fs_info->scrub_pause_wait);
2289 fs_info->scrub_workers_refcnt = 0;
2290#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2291 fs_info->check_integrity_print_mask = 0;
2292#endif
2293
2294 spin_lock_init(&fs_info->balance_lock);
2295 mutex_init(&fs_info->balance_mutex);
2296 atomic_set(&fs_info->balance_running, 0);
2297 atomic_set(&fs_info->balance_pause_req, 0);
2298 atomic_set(&fs_info->balance_cancel_req, 0);
2299 fs_info->balance_ctl = NULL;
2300 init_waitqueue_head(&fs_info->balance_wait_q);
2301 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2302
2303 sb->s_blocksize = 4096;
2304 sb->s_blocksize_bits = blksize_bits(4096);
2305 sb->s_bdi = &fs_info->bdi;
2306
2307 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2308 set_nlink(fs_info->btree_inode, 1);
2309 /*
2310 * we set the i_size on the btree inode to the max possible int.
2311 * the real end of the address space is determined by all of
2312 * the devices in the system
2313 */
2314 fs_info->btree_inode->i_size = OFFSET_MAX;
2315 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2316 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2317
2318 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2319 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2320 fs_info->btree_inode->i_mapping);
2321 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2322 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2323
2324 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2325
2326 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2327 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2328 sizeof(struct btrfs_key));
2329 set_bit(BTRFS_INODE_DUMMY,
2330 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2331 btrfs_insert_inode_hash(fs_info->btree_inode);
2332
2333 spin_lock_init(&fs_info->block_group_cache_lock);
2334 fs_info->block_group_cache_tree = RB_ROOT;
2335 fs_info->first_logical_byte = (u64)-1;
2336
2337 extent_io_tree_init(&fs_info->freed_extents[0],
2338 fs_info->btree_inode->i_mapping);
2339 extent_io_tree_init(&fs_info->freed_extents[1],
2340 fs_info->btree_inode->i_mapping);
2341 fs_info->pinned_extents = &fs_info->freed_extents[0];
2342 fs_info->do_barriers = 1;
2343
2344
2345 mutex_init(&fs_info->ordered_operations_mutex);
2346 mutex_init(&fs_info->ordered_extent_flush_mutex);
2347 mutex_init(&fs_info->tree_log_mutex);
2348 mutex_init(&fs_info->chunk_mutex);
2349 mutex_init(&fs_info->transaction_kthread_mutex);
2350 mutex_init(&fs_info->cleaner_mutex);
2351 mutex_init(&fs_info->volume_mutex);
2352 init_rwsem(&fs_info->commit_root_sem);
2353 init_rwsem(&fs_info->cleanup_work_sem);
2354 init_rwsem(&fs_info->subvol_sem);
2355 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2356 fs_info->dev_replace.lock_owner = 0;
2357 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2358 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2359 mutex_init(&fs_info->dev_replace.lock_management_lock);
2360 mutex_init(&fs_info->dev_replace.lock);
2361
2362 spin_lock_init(&fs_info->qgroup_lock);
2363 mutex_init(&fs_info->qgroup_ioctl_lock);
2364 fs_info->qgroup_tree = RB_ROOT;
2365 fs_info->qgroup_op_tree = RB_ROOT;
2366 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2367 fs_info->qgroup_seq = 1;
2368 fs_info->quota_enabled = 0;
2369 fs_info->pending_quota_state = 0;
2370 fs_info->qgroup_ulist = NULL;
2371 mutex_init(&fs_info->qgroup_rescan_lock);
2372
2373 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2374 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2375
2376 init_waitqueue_head(&fs_info->transaction_throttle);
2377 init_waitqueue_head(&fs_info->transaction_wait);
2378 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2379 init_waitqueue_head(&fs_info->async_submit_wait);
2380
2381 ret = btrfs_alloc_stripe_hash_table(fs_info);
2382 if (ret) {
2383 err = ret;
2384 goto fail_alloc;
2385 }
2386
2387 __setup_root(4096, 4096, 4096, 4096, tree_root,
2388 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2389
2390 invalidate_bdev(fs_devices->latest_bdev);
2391
2392 /*
2393 * Read super block and check the signature bytes only
2394 */
2395 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2396 if (!bh) {
2397 err = -EINVAL;
2398 goto fail_alloc;
2399 }
2400
2401 /*
2402 * We want to check superblock checksum, the type is stored inside.
2403 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2404 */
2405 if (btrfs_check_super_csum(bh->b_data)) {
2406 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2407 err = -EINVAL;
2408 goto fail_alloc;
2409 }
2410
2411 /*
2412 * super_copy is zeroed at allocation time and we never touch the
2413 * following bytes up to INFO_SIZE, the checksum is calculated from
2414 * the whole block of INFO_SIZE
2415 */
2416 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2417 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2418 sizeof(*fs_info->super_for_commit));
2419 brelse(bh);
2420
2421 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2422
2423 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2424 if (ret) {
2425 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2426 err = -EINVAL;
2427 goto fail_alloc;
2428 }
2429
2430 disk_super = fs_info->super_copy;
2431 if (!btrfs_super_root(disk_super))
2432 goto fail_alloc;
2433
2434 /* check FS state, whether FS is broken. */
2435 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2436 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2437
2438 /*
2439 * run through our array of backup supers and setup
2440 * our ring pointer to the oldest one
2441 */
2442 generation = btrfs_super_generation(disk_super);
2443 find_oldest_super_backup(fs_info, generation);
2444
2445 /*
2446 * In the long term, we'll store the compression type in the super
2447 * block, and it'll be used for per file compression control.
2448 */
2449 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2450
2451 ret = btrfs_parse_options(tree_root, options);
2452 if (ret) {
2453 err = ret;
2454 goto fail_alloc;
2455 }
2456
2457 features = btrfs_super_incompat_flags(disk_super) &
2458 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2459 if (features) {
2460 printk(KERN_ERR "BTRFS: couldn't mount because of "
2461 "unsupported optional features (%Lx).\n",
2462 features);
2463 err = -EINVAL;
2464 goto fail_alloc;
2465 }
2466
2467 if (btrfs_super_leafsize(disk_super) !=
2468 btrfs_super_nodesize(disk_super)) {
2469 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2470 "blocksizes don't match. node %d leaf %d\n",
2471 btrfs_super_nodesize(disk_super),
2472 btrfs_super_leafsize(disk_super));
2473 err = -EINVAL;
2474 goto fail_alloc;
2475 }
2476 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2477 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2478 "blocksize (%d) was too large\n",
2479 btrfs_super_leafsize(disk_super));
2480 err = -EINVAL;
2481 goto fail_alloc;
2482 }
2483
2484 features = btrfs_super_incompat_flags(disk_super);
2485 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2486 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2487 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2488
2489 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2490 printk(KERN_ERR "BTRFS: has skinny extents\n");
2491
2492 /*
2493 * flag our filesystem as having big metadata blocks if
2494 * they are bigger than the page size
2495 */
2496 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2497 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2498 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2499 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2500 }
2501
2502 nodesize = btrfs_super_nodesize(disk_super);
2503 leafsize = btrfs_super_leafsize(disk_super);
2504 sectorsize = btrfs_super_sectorsize(disk_super);
2505 stripesize = btrfs_super_stripesize(disk_super);
2506 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2507 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2508
2509 /*
2510 * mixed block groups end up with duplicate but slightly offset
2511 * extent buffers for the same range. It leads to corruptions
2512 */
2513 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2514 (sectorsize != leafsize)) {
2515 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2516 "are not allowed for mixed block groups on %s\n",
2517 sb->s_id);
2518 goto fail_alloc;
2519 }
2520
2521 /*
2522 * Needn't use the lock because there is no other task which will
2523 * update the flag.
2524 */
2525 btrfs_set_super_incompat_flags(disk_super, features);
2526
2527 features = btrfs_super_compat_ro_flags(disk_super) &
2528 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2529 if (!(sb->s_flags & MS_RDONLY) && features) {
2530 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2531 "unsupported option features (%Lx).\n",
2532 features);
2533 err = -EINVAL;
2534 goto fail_alloc;
2535 }
2536
2537 max_active = fs_info->thread_pool_size;
2538
2539 fs_info->workers =
2540 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2541 max_active, 16);
2542
2543 fs_info->delalloc_workers =
2544 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2545
2546 fs_info->flush_workers =
2547 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2548
2549 fs_info->caching_workers =
2550 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2551
2552 /*
2553 * a higher idle thresh on the submit workers makes it much more
2554 * likely that bios will be send down in a sane order to the
2555 * devices
2556 */
2557 fs_info->submit_workers =
2558 btrfs_alloc_workqueue("submit", flags,
2559 min_t(u64, fs_devices->num_devices,
2560 max_active), 64);
2561
2562 fs_info->fixup_workers =
2563 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2564
2565 /*
2566 * endios are largely parallel and should have a very
2567 * low idle thresh
2568 */
2569 fs_info->endio_workers =
2570 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2571 fs_info->endio_meta_workers =
2572 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2573 fs_info->endio_meta_write_workers =
2574 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2575 fs_info->endio_raid56_workers =
2576 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2577 fs_info->rmw_workers =
2578 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2579 fs_info->endio_write_workers =
2580 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2581 fs_info->endio_freespace_worker =
2582 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2583 fs_info->delayed_workers =
2584 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2585 fs_info->readahead_workers =
2586 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2587 fs_info->qgroup_rescan_workers =
2588 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2589 fs_info->extent_workers =
2590 btrfs_alloc_workqueue("extent-refs", flags,
2591 min_t(u64, fs_devices->num_devices,
2592 max_active), 8);
2593
2594 if (!(fs_info->workers && fs_info->delalloc_workers &&
2595 fs_info->submit_workers && fs_info->flush_workers &&
2596 fs_info->endio_workers && fs_info->endio_meta_workers &&
2597 fs_info->endio_meta_write_workers &&
2598 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2599 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2600 fs_info->caching_workers && fs_info->readahead_workers &&
2601 fs_info->fixup_workers && fs_info->delayed_workers &&
2602 fs_info->fixup_workers && fs_info->extent_workers &&
2603 fs_info->qgroup_rescan_workers)) {
2604 err = -ENOMEM;
2605 goto fail_sb_buffer;
2606 }
2607
2608 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2609 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2610 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2611
2612 tree_root->nodesize = nodesize;
2613 tree_root->leafsize = leafsize;
2614 tree_root->sectorsize = sectorsize;
2615 tree_root->stripesize = stripesize;
2616
2617 sb->s_blocksize = sectorsize;
2618 sb->s_blocksize_bits = blksize_bits(sectorsize);
2619
2620 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2621 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2622 goto fail_sb_buffer;
2623 }
2624
2625 if (sectorsize != PAGE_SIZE) {
2626 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2627 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2628 goto fail_sb_buffer;
2629 }
2630
2631 mutex_lock(&fs_info->chunk_mutex);
2632 ret = btrfs_read_sys_array(tree_root);
2633 mutex_unlock(&fs_info->chunk_mutex);
2634 if (ret) {
2635 printk(KERN_WARNING "BTRFS: failed to read the system "
2636 "array on %s\n", sb->s_id);
2637 goto fail_sb_buffer;
2638 }
2639
2640 blocksize = btrfs_level_size(tree_root,
2641 btrfs_super_chunk_root_level(disk_super));
2642 generation = btrfs_super_chunk_root_generation(disk_super);
2643
2644 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2645 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2646
2647 chunk_root->node = read_tree_block(chunk_root,
2648 btrfs_super_chunk_root(disk_super),
2649 blocksize, generation);
2650 if (!chunk_root->node ||
2651 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2652 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2653 sb->s_id);
2654 goto fail_tree_roots;
2655 }
2656 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2657 chunk_root->commit_root = btrfs_root_node(chunk_root);
2658
2659 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2660 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2661
2662 ret = btrfs_read_chunk_tree(chunk_root);
2663 if (ret) {
2664 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2665 sb->s_id);
2666 goto fail_tree_roots;
2667 }
2668
2669 /*
2670 * keep the device that is marked to be the target device for the
2671 * dev_replace procedure
2672 */
2673 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2674
2675 if (!fs_devices->latest_bdev) {
2676 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2677 sb->s_id);
2678 goto fail_tree_roots;
2679 }
2680
2681retry_root_backup:
2682 blocksize = btrfs_level_size(tree_root,
2683 btrfs_super_root_level(disk_super));
2684 generation = btrfs_super_generation(disk_super);
2685
2686 tree_root->node = read_tree_block(tree_root,
2687 btrfs_super_root(disk_super),
2688 blocksize, generation);
2689 if (!tree_root->node ||
2690 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2691 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2692 sb->s_id);
2693
2694 goto recovery_tree_root;
2695 }
2696
2697 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2698 tree_root->commit_root = btrfs_root_node(tree_root);
2699 btrfs_set_root_refs(&tree_root->root_item, 1);
2700
2701 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2702 location.type = BTRFS_ROOT_ITEM_KEY;
2703 location.offset = 0;
2704
2705 extent_root = btrfs_read_tree_root(tree_root, &location);
2706 if (IS_ERR(extent_root)) {
2707 ret = PTR_ERR(extent_root);
2708 goto recovery_tree_root;
2709 }
2710 set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2711 fs_info->extent_root = extent_root;
2712
2713 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2714 dev_root = btrfs_read_tree_root(tree_root, &location);
2715 if (IS_ERR(dev_root)) {
2716 ret = PTR_ERR(dev_root);
2717 goto recovery_tree_root;
2718 }
2719 set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2720 fs_info->dev_root = dev_root;
2721 btrfs_init_devices_late(fs_info);
2722
2723 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2724 csum_root = btrfs_read_tree_root(tree_root, &location);
2725 if (IS_ERR(csum_root)) {
2726 ret = PTR_ERR(csum_root);
2727 goto recovery_tree_root;
2728 }
2729 set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2730 fs_info->csum_root = csum_root;
2731
2732 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2733 quota_root = btrfs_read_tree_root(tree_root, &location);
2734 if (!IS_ERR(quota_root)) {
2735 set_bit(BTRFS_ROOT_TRACK_DIRTY, &quota_root->state);
2736 fs_info->quota_enabled = 1;
2737 fs_info->pending_quota_state = 1;
2738 fs_info->quota_root = quota_root;
2739 }
2740
2741 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2742 uuid_root = btrfs_read_tree_root(tree_root, &location);
2743 if (IS_ERR(uuid_root)) {
2744 ret = PTR_ERR(uuid_root);
2745 if (ret != -ENOENT)
2746 goto recovery_tree_root;
2747 create_uuid_tree = true;
2748 check_uuid_tree = false;
2749 } else {
2750 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2751 fs_info->uuid_root = uuid_root;
2752 create_uuid_tree = false;
2753 check_uuid_tree =
2754 generation != btrfs_super_uuid_tree_generation(disk_super);
2755 }
2756
2757 fs_info->generation = generation;
2758 fs_info->last_trans_committed = generation;
2759
2760 ret = btrfs_recover_balance(fs_info);
2761 if (ret) {
2762 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2763 goto fail_block_groups;
2764 }
2765
2766 ret = btrfs_init_dev_stats(fs_info);
2767 if (ret) {
2768 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2769 ret);
2770 goto fail_block_groups;
2771 }
2772
2773 ret = btrfs_init_dev_replace(fs_info);
2774 if (ret) {
2775 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2776 goto fail_block_groups;
2777 }
2778
2779 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2780
2781 ret = btrfs_sysfs_add_one(fs_info);
2782 if (ret) {
2783 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2784 goto fail_block_groups;
2785 }
2786
2787 ret = btrfs_init_space_info(fs_info);
2788 if (ret) {
2789 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2790 goto fail_sysfs;
2791 }
2792
2793 ret = btrfs_read_block_groups(extent_root);
2794 if (ret) {
2795 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2796 goto fail_sysfs;
2797 }
2798 fs_info->num_tolerated_disk_barrier_failures =
2799 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2800 if (fs_info->fs_devices->missing_devices >
2801 fs_info->num_tolerated_disk_barrier_failures &&
2802 !(sb->s_flags & MS_RDONLY)) {
2803 printk(KERN_WARNING "BTRFS: "
2804 "too many missing devices, writeable mount is not allowed\n");
2805 goto fail_sysfs;
2806 }
2807
2808 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2809 "btrfs-cleaner");
2810 if (IS_ERR(fs_info->cleaner_kthread))
2811 goto fail_sysfs;
2812
2813 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2814 tree_root,
2815 "btrfs-transaction");
2816 if (IS_ERR(fs_info->transaction_kthread))
2817 goto fail_cleaner;
2818
2819 if (!btrfs_test_opt(tree_root, SSD) &&
2820 !btrfs_test_opt(tree_root, NOSSD) &&
2821 !fs_info->fs_devices->rotating) {
2822 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2823 "mode\n");
2824 btrfs_set_opt(fs_info->mount_opt, SSD);
2825 }
2826
2827 /* Set the real inode map cache flag */
2828 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2829 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2830
2831#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2832 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2833 ret = btrfsic_mount(tree_root, fs_devices,
2834 btrfs_test_opt(tree_root,
2835 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2836 1 : 0,
2837 fs_info->check_integrity_print_mask);
2838 if (ret)
2839 printk(KERN_WARNING "BTRFS: failed to initialize"
2840 " integrity check module %s\n", sb->s_id);
2841 }
2842#endif
2843 ret = btrfs_read_qgroup_config(fs_info);
2844 if (ret)
2845 goto fail_trans_kthread;
2846
2847 /* do not make disk changes in broken FS */
2848 if (btrfs_super_log_root(disk_super) != 0) {
2849 u64 bytenr = btrfs_super_log_root(disk_super);
2850
2851 if (fs_devices->rw_devices == 0) {
2852 printk(KERN_WARNING "BTRFS: log replay required "
2853 "on RO media\n");
2854 err = -EIO;
2855 goto fail_qgroup;
2856 }
2857 blocksize =
2858 btrfs_level_size(tree_root,
2859 btrfs_super_log_root_level(disk_super));
2860
2861 log_tree_root = btrfs_alloc_root(fs_info);
2862 if (!log_tree_root) {
2863 err = -ENOMEM;
2864 goto fail_qgroup;
2865 }
2866
2867 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2868 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2869
2870 log_tree_root->node = read_tree_block(tree_root, bytenr,
2871 blocksize,
2872 generation + 1);
2873 if (!log_tree_root->node ||
2874 !extent_buffer_uptodate(log_tree_root->node)) {
2875 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2876 free_extent_buffer(log_tree_root->node);
2877 kfree(log_tree_root);
2878 goto fail_qgroup;
2879 }
2880 /* returns with log_tree_root freed on success */
2881 ret = btrfs_recover_log_trees(log_tree_root);
2882 if (ret) {
2883 btrfs_error(tree_root->fs_info, ret,
2884 "Failed to recover log tree");
2885 free_extent_buffer(log_tree_root->node);
2886 kfree(log_tree_root);
2887 goto fail_qgroup;
2888 }
2889
2890 if (sb->s_flags & MS_RDONLY) {
2891 ret = btrfs_commit_super(tree_root);
2892 if (ret)
2893 goto fail_qgroup;
2894 }
2895 }
2896
2897 ret = btrfs_find_orphan_roots(tree_root);
2898 if (ret)
2899 goto fail_qgroup;
2900
2901 if (!(sb->s_flags & MS_RDONLY)) {
2902 ret = btrfs_cleanup_fs_roots(fs_info);
2903 if (ret)
2904 goto fail_qgroup;
2905
2906 mutex_lock(&fs_info->cleaner_mutex);
2907 ret = btrfs_recover_relocation(tree_root);
2908 mutex_unlock(&fs_info->cleaner_mutex);
2909 if (ret < 0) {
2910 printk(KERN_WARNING
2911 "BTRFS: failed to recover relocation\n");
2912 err = -EINVAL;
2913 goto fail_qgroup;
2914 }
2915 }
2916
2917 location.objectid = BTRFS_FS_TREE_OBJECTID;
2918 location.type = BTRFS_ROOT_ITEM_KEY;
2919 location.offset = 0;
2920
2921 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2922 if (IS_ERR(fs_info->fs_root)) {
2923 err = PTR_ERR(fs_info->fs_root);
2924 goto fail_qgroup;
2925 }
2926
2927 if (sb->s_flags & MS_RDONLY)
2928 return 0;
2929
2930 down_read(&fs_info->cleanup_work_sem);
2931 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2932 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2933 up_read(&fs_info->cleanup_work_sem);
2934 close_ctree(tree_root);
2935 return ret;
2936 }
2937 up_read(&fs_info->cleanup_work_sem);
2938
2939 ret = btrfs_resume_balance_async(fs_info);
2940 if (ret) {
2941 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2942 close_ctree(tree_root);
2943 return ret;
2944 }
2945
2946 ret = btrfs_resume_dev_replace_async(fs_info);
2947 if (ret) {
2948 pr_warn("BTRFS: failed to resume dev_replace\n");
2949 close_ctree(tree_root);
2950 return ret;
2951 }
2952
2953 btrfs_qgroup_rescan_resume(fs_info);
2954
2955 if (create_uuid_tree) {
2956 pr_info("BTRFS: creating UUID tree\n");
2957 ret = btrfs_create_uuid_tree(fs_info);
2958 if (ret) {
2959 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2960 ret);
2961 close_ctree(tree_root);
2962 return ret;
2963 }
2964 } else if (check_uuid_tree ||
2965 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2966 pr_info("BTRFS: checking UUID tree\n");
2967 ret = btrfs_check_uuid_tree(fs_info);
2968 if (ret) {
2969 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2970 ret);
2971 close_ctree(tree_root);
2972 return ret;
2973 }
2974 } else {
2975 fs_info->update_uuid_tree_gen = 1;
2976 }
2977
2978 return 0;
2979
2980fail_qgroup:
2981 btrfs_free_qgroup_config(fs_info);
2982fail_trans_kthread:
2983 kthread_stop(fs_info->transaction_kthread);
2984 btrfs_cleanup_transaction(fs_info->tree_root);
2985 btrfs_free_fs_roots(fs_info);
2986fail_cleaner:
2987 kthread_stop(fs_info->cleaner_kthread);
2988
2989 /*
2990 * make sure we're done with the btree inode before we stop our
2991 * kthreads
2992 */
2993 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2994
2995fail_sysfs:
2996 btrfs_sysfs_remove_one(fs_info);
2997
2998fail_block_groups:
2999 btrfs_put_block_group_cache(fs_info);
3000 btrfs_free_block_groups(fs_info);
3001
3002fail_tree_roots:
3003 free_root_pointers(fs_info, 1);
3004 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3005
3006fail_sb_buffer:
3007 btrfs_stop_all_workers(fs_info);
3008fail_alloc:
3009fail_iput:
3010 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3011
3012 iput(fs_info->btree_inode);
3013fail_bio_counter:
3014 percpu_counter_destroy(&fs_info->bio_counter);
3015fail_delalloc_bytes:
3016 percpu_counter_destroy(&fs_info->delalloc_bytes);
3017fail_dirty_metadata_bytes:
3018 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3019fail_bdi:
3020 bdi_destroy(&fs_info->bdi);
3021fail_srcu:
3022 cleanup_srcu_struct(&fs_info->subvol_srcu);
3023fail:
3024 btrfs_free_stripe_hash_table(fs_info);
3025 btrfs_close_devices(fs_info->fs_devices);
3026 return err;
3027
3028recovery_tree_root:
3029 if (!btrfs_test_opt(tree_root, RECOVERY))
3030 goto fail_tree_roots;
3031
3032 free_root_pointers(fs_info, 0);
3033
3034 /* don't use the log in recovery mode, it won't be valid */
3035 btrfs_set_super_log_root(disk_super, 0);
3036
3037 /* we can't trust the free space cache either */
3038 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3039
3040 ret = next_root_backup(fs_info, fs_info->super_copy,
3041 &num_backups_tried, &backup_index);
3042 if (ret == -1)
3043 goto fail_block_groups;
3044 goto retry_root_backup;
3045}
3046
3047static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3048{
3049 if (uptodate) {
3050 set_buffer_uptodate(bh);
3051 } else {
3052 struct btrfs_device *device = (struct btrfs_device *)
3053 bh->b_private;
3054
3055 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3056 "I/O error on %s\n",
3057 rcu_str_deref(device->name));
3058 /* note, we dont' set_buffer_write_io_error because we have
3059 * our own ways of dealing with the IO errors
3060 */
3061 clear_buffer_uptodate(bh);
3062 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3063 }
3064 unlock_buffer(bh);
3065 put_bh(bh);
3066}
3067
3068struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3069{
3070 struct buffer_head *bh;
3071 struct buffer_head *latest = NULL;
3072 struct btrfs_super_block *super;
3073 int i;
3074 u64 transid = 0;
3075 u64 bytenr;
3076
3077 /* we would like to check all the supers, but that would make
3078 * a btrfs mount succeed after a mkfs from a different FS.
3079 * So, we need to add a special mount option to scan for
3080 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3081 */
3082 for (i = 0; i < 1; i++) {
3083 bytenr = btrfs_sb_offset(i);
3084 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3085 i_size_read(bdev->bd_inode))
3086 break;
3087 bh = __bread(bdev, bytenr / 4096,
3088 BTRFS_SUPER_INFO_SIZE);
3089 if (!bh)
3090 continue;
3091
3092 super = (struct btrfs_super_block *)bh->b_data;
3093 if (btrfs_super_bytenr(super) != bytenr ||
3094 btrfs_super_magic(super) != BTRFS_MAGIC) {
3095 brelse(bh);
3096 continue;
3097 }
3098
3099 if (!latest || btrfs_super_generation(super) > transid) {
3100 brelse(latest);
3101 latest = bh;
3102 transid = btrfs_super_generation(super);
3103 } else {
3104 brelse(bh);
3105 }
3106 }
3107 return latest;
3108}
3109
3110/*
3111 * this should be called twice, once with wait == 0 and
3112 * once with wait == 1. When wait == 0 is done, all the buffer heads
3113 * we write are pinned.
3114 *
3115 * They are released when wait == 1 is done.
3116 * max_mirrors must be the same for both runs, and it indicates how
3117 * many supers on this one device should be written.
3118 *
3119 * max_mirrors == 0 means to write them all.
3120 */
3121static int write_dev_supers(struct btrfs_device *device,
3122 struct btrfs_super_block *sb,
3123 int do_barriers, int wait, int max_mirrors)
3124{
3125 struct buffer_head *bh;
3126 int i;
3127 int ret;
3128 int errors = 0;
3129 u32 crc;
3130 u64 bytenr;
3131
3132 if (max_mirrors == 0)
3133 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3134
3135 for (i = 0; i < max_mirrors; i++) {
3136 bytenr = btrfs_sb_offset(i);
3137 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3138 break;
3139
3140 if (wait) {
3141 bh = __find_get_block(device->bdev, bytenr / 4096,
3142 BTRFS_SUPER_INFO_SIZE);
3143 if (!bh) {
3144 errors++;
3145 continue;
3146 }
3147 wait_on_buffer(bh);
3148 if (!buffer_uptodate(bh))
3149 errors++;
3150
3151 /* drop our reference */
3152 brelse(bh);
3153
3154 /* drop the reference from the wait == 0 run */
3155 brelse(bh);
3156 continue;
3157 } else {
3158 btrfs_set_super_bytenr(sb, bytenr);
3159
3160 crc = ~(u32)0;
3161 crc = btrfs_csum_data((char *)sb +
3162 BTRFS_CSUM_SIZE, crc,
3163 BTRFS_SUPER_INFO_SIZE -
3164 BTRFS_CSUM_SIZE);
3165 btrfs_csum_final(crc, sb->csum);
3166
3167 /*
3168 * one reference for us, and we leave it for the
3169 * caller
3170 */
3171 bh = __getblk(device->bdev, bytenr / 4096,
3172 BTRFS_SUPER_INFO_SIZE);
3173 if (!bh) {
3174 printk(KERN_ERR "BTRFS: couldn't get super "
3175 "buffer head for bytenr %Lu\n", bytenr);
3176 errors++;
3177 continue;
3178 }
3179
3180 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3181
3182 /* one reference for submit_bh */
3183 get_bh(bh);
3184
3185 set_buffer_uptodate(bh);
3186 lock_buffer(bh);
3187 bh->b_end_io = btrfs_end_buffer_write_sync;
3188 bh->b_private = device;
3189 }
3190
3191 /*
3192 * we fua the first super. The others we allow
3193 * to go down lazy.
3194 */
3195 if (i == 0)
3196 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3197 else
3198 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3199 if (ret)
3200 errors++;
3201 }
3202 return errors < i ? 0 : -1;
3203}
3204
3205/*
3206 * endio for the write_dev_flush, this will wake anyone waiting
3207 * for the barrier when it is done
3208 */
3209static void btrfs_end_empty_barrier(struct bio *bio, int err)
3210{
3211 if (err) {
3212 if (err == -EOPNOTSUPP)
3213 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3214 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3215 }
3216 if (bio->bi_private)
3217 complete(bio->bi_private);
3218 bio_put(bio);
3219}
3220
3221/*
3222 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3223 * sent down. With wait == 1, it waits for the previous flush.
3224 *
3225 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3226 * capable
3227 */
3228static int write_dev_flush(struct btrfs_device *device, int wait)
3229{
3230 struct bio *bio;
3231 int ret = 0;
3232
3233 if (device->nobarriers)
3234 return 0;
3235
3236 if (wait) {
3237 bio = device->flush_bio;
3238 if (!bio)
3239 return 0;
3240
3241 wait_for_completion(&device->flush_wait);
3242
3243 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3244 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3245 rcu_str_deref(device->name));
3246 device->nobarriers = 1;
3247 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3248 ret = -EIO;
3249 btrfs_dev_stat_inc_and_print(device,
3250 BTRFS_DEV_STAT_FLUSH_ERRS);
3251 }
3252
3253 /* drop the reference from the wait == 0 run */
3254 bio_put(bio);
3255 device->flush_bio = NULL;
3256
3257 return ret;
3258 }
3259
3260 /*
3261 * one reference for us, and we leave it for the
3262 * caller
3263 */
3264 device->flush_bio = NULL;
3265 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3266 if (!bio)
3267 return -ENOMEM;
3268
3269 bio->bi_end_io = btrfs_end_empty_barrier;
3270 bio->bi_bdev = device->bdev;
3271 init_completion(&device->flush_wait);
3272 bio->bi_private = &device->flush_wait;
3273 device->flush_bio = bio;
3274
3275 bio_get(bio);
3276 btrfsic_submit_bio(WRITE_FLUSH, bio);
3277
3278 return 0;
3279}
3280
3281/*
3282 * send an empty flush down to each device in parallel,
3283 * then wait for them
3284 */
3285static int barrier_all_devices(struct btrfs_fs_info *info)
3286{
3287 struct list_head *head;
3288 struct btrfs_device *dev;
3289 int errors_send = 0;
3290 int errors_wait = 0;
3291 int ret;
3292
3293 /* send down all the barriers */
3294 head = &info->fs_devices->devices;
3295 list_for_each_entry_rcu(dev, head, dev_list) {
3296 if (dev->missing)
3297 continue;
3298 if (!dev->bdev) {
3299 errors_send++;
3300 continue;
3301 }
3302 if (!dev->in_fs_metadata || !dev->writeable)
3303 continue;
3304
3305 ret = write_dev_flush(dev, 0);
3306 if (ret)
3307 errors_send++;
3308 }
3309
3310 /* wait for all the barriers */
3311 list_for_each_entry_rcu(dev, head, dev_list) {
3312 if (dev->missing)
3313 continue;
3314 if (!dev->bdev) {
3315 errors_wait++;
3316 continue;
3317 }
3318 if (!dev->in_fs_metadata || !dev->writeable)
3319 continue;
3320
3321 ret = write_dev_flush(dev, 1);
3322 if (ret)
3323 errors_wait++;
3324 }
3325 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3326 errors_wait > info->num_tolerated_disk_barrier_failures)
3327 return -EIO;
3328 return 0;
3329}
3330
3331int btrfs_calc_num_tolerated_disk_barrier_failures(
3332 struct btrfs_fs_info *fs_info)
3333{
3334 struct btrfs_ioctl_space_info space;
3335 struct btrfs_space_info *sinfo;
3336 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3337 BTRFS_BLOCK_GROUP_SYSTEM,
3338 BTRFS_BLOCK_GROUP_METADATA,
3339 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3340 int num_types = 4;
3341 int i;
3342 int c;
3343 int num_tolerated_disk_barrier_failures =
3344 (int)fs_info->fs_devices->num_devices;
3345
3346 for (i = 0; i < num_types; i++) {
3347 struct btrfs_space_info *tmp;
3348
3349 sinfo = NULL;
3350 rcu_read_lock();
3351 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3352 if (tmp->flags == types[i]) {
3353 sinfo = tmp;
3354 break;
3355 }
3356 }
3357 rcu_read_unlock();
3358
3359 if (!sinfo)
3360 continue;
3361
3362 down_read(&sinfo->groups_sem);
3363 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3364 if (!list_empty(&sinfo->block_groups[c])) {
3365 u64 flags;
3366
3367 btrfs_get_block_group_info(
3368 &sinfo->block_groups[c], &space);
3369 if (space.total_bytes == 0 ||
3370 space.used_bytes == 0)
3371 continue;
3372 flags = space.flags;
3373 /*
3374 * return
3375 * 0: if dup, single or RAID0 is configured for
3376 * any of metadata, system or data, else
3377 * 1: if RAID5 is configured, or if RAID1 or
3378 * RAID10 is configured and only two mirrors
3379 * are used, else
3380 * 2: if RAID6 is configured, else
3381 * num_mirrors - 1: if RAID1 or RAID10 is
3382 * configured and more than
3383 * 2 mirrors are used.
3384 */
3385 if (num_tolerated_disk_barrier_failures > 0 &&
3386 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3387 BTRFS_BLOCK_GROUP_RAID0)) ||
3388 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3389 == 0)))
3390 num_tolerated_disk_barrier_failures = 0;
3391 else if (num_tolerated_disk_barrier_failures > 1) {
3392 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3393 BTRFS_BLOCK_GROUP_RAID5 |
3394 BTRFS_BLOCK_GROUP_RAID10)) {
3395 num_tolerated_disk_barrier_failures = 1;
3396 } else if (flags &
3397 BTRFS_BLOCK_GROUP_RAID6) {
3398 num_tolerated_disk_barrier_failures = 2;
3399 }
3400 }
3401 }
3402 }
3403 up_read(&sinfo->groups_sem);
3404 }
3405
3406 return num_tolerated_disk_barrier_failures;
3407}
3408
3409static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3410{
3411 struct list_head *head;
3412 struct btrfs_device *dev;
3413 struct btrfs_super_block *sb;
3414 struct btrfs_dev_item *dev_item;
3415 int ret;
3416 int do_barriers;
3417 int max_errors;
3418 int total_errors = 0;
3419 u64 flags;
3420
3421 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3422 backup_super_roots(root->fs_info);
3423
3424 sb = root->fs_info->super_for_commit;
3425 dev_item = &sb->dev_item;
3426
3427 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3428 head = &root->fs_info->fs_devices->devices;
3429 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3430
3431 if (do_barriers) {
3432 ret = barrier_all_devices(root->fs_info);
3433 if (ret) {
3434 mutex_unlock(
3435 &root->fs_info->fs_devices->device_list_mutex);
3436 btrfs_error(root->fs_info, ret,
3437 "errors while submitting device barriers.");
3438 return ret;
3439 }
3440 }
3441
3442 list_for_each_entry_rcu(dev, head, dev_list) {
3443 if (!dev->bdev) {
3444 total_errors++;
3445 continue;
3446 }
3447 if (!dev->in_fs_metadata || !dev->writeable)
3448 continue;
3449
3450 btrfs_set_stack_device_generation(dev_item, 0);
3451 btrfs_set_stack_device_type(dev_item, dev->type);
3452 btrfs_set_stack_device_id(dev_item, dev->devid);
3453 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3454 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3455 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3456 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3457 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3458 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3459 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3460
3461 flags = btrfs_super_flags(sb);
3462 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3463
3464 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3465 if (ret)
3466 total_errors++;
3467 }
3468 if (total_errors > max_errors) {
3469 btrfs_err(root->fs_info, "%d errors while writing supers",
3470 total_errors);
3471 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3472
3473 /* FUA is masked off if unsupported and can't be the reason */
3474 btrfs_error(root->fs_info, -EIO,
3475 "%d errors while writing supers", total_errors);
3476 return -EIO;
3477 }
3478
3479 total_errors = 0;
3480 list_for_each_entry_rcu(dev, head, dev_list) {
3481 if (!dev->bdev)
3482 continue;
3483 if (!dev->in_fs_metadata || !dev->writeable)
3484 continue;
3485
3486 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3487 if (ret)
3488 total_errors++;
3489 }
3490 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3491 if (total_errors > max_errors) {
3492 btrfs_error(root->fs_info, -EIO,
3493 "%d errors while writing supers", total_errors);
3494 return -EIO;
3495 }
3496 return 0;
3497}
3498
3499int write_ctree_super(struct btrfs_trans_handle *trans,
3500 struct btrfs_root *root, int max_mirrors)
3501{
3502 return write_all_supers(root, max_mirrors);
3503}
3504
3505/* Drop a fs root from the radix tree and free it. */
3506void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3507 struct btrfs_root *root)
3508{
3509 spin_lock(&fs_info->fs_roots_radix_lock);
3510 radix_tree_delete(&fs_info->fs_roots_radix,
3511 (unsigned long)root->root_key.objectid);
3512 spin_unlock(&fs_info->fs_roots_radix_lock);
3513
3514 if (btrfs_root_refs(&root->root_item) == 0)
3515 synchronize_srcu(&fs_info->subvol_srcu);
3516
3517 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3518 btrfs_free_log(NULL, root);
3519
3520 if (root->free_ino_pinned)
3521 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3522 if (root->free_ino_ctl)
3523 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3524 free_fs_root(root);
3525}
3526
3527static void free_fs_root(struct btrfs_root *root)
3528{
3529 iput(root->cache_inode);
3530 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3531 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3532 root->orphan_block_rsv = NULL;
3533 if (root->anon_dev)
3534 free_anon_bdev(root->anon_dev);
3535 if (root->subv_writers)
3536 btrfs_free_subvolume_writers(root->subv_writers);
3537 free_extent_buffer(root->node);
3538 free_extent_buffer(root->commit_root);
3539 kfree(root->free_ino_ctl);
3540 kfree(root->free_ino_pinned);
3541 kfree(root->name);
3542 btrfs_put_fs_root(root);
3543}
3544
3545void btrfs_free_fs_root(struct btrfs_root *root)
3546{
3547 free_fs_root(root);
3548}
3549
3550int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3551{
3552 u64 root_objectid = 0;
3553 struct btrfs_root *gang[8];
3554 int i = 0;
3555 int err = 0;
3556 unsigned int ret = 0;
3557 int index;
3558
3559 while (1) {
3560 index = srcu_read_lock(&fs_info->subvol_srcu);
3561 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3562 (void **)gang, root_objectid,
3563 ARRAY_SIZE(gang));
3564 if (!ret) {
3565 srcu_read_unlock(&fs_info->subvol_srcu, index);
3566 break;
3567 }
3568 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3569
3570 for (i = 0; i < ret; i++) {
3571 /* Avoid to grab roots in dead_roots */
3572 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3573 gang[i] = NULL;
3574 continue;
3575 }
3576 /* grab all the search result for later use */
3577 gang[i] = btrfs_grab_fs_root(gang[i]);
3578 }
3579 srcu_read_unlock(&fs_info->subvol_srcu, index);
3580
3581 for (i = 0; i < ret; i++) {
3582 if (!gang[i])
3583 continue;
3584 root_objectid = gang[i]->root_key.objectid;
3585 err = btrfs_orphan_cleanup(gang[i]);
3586 if (err)
3587 break;
3588 btrfs_put_fs_root(gang[i]);
3589 }
3590 root_objectid++;
3591 }
3592
3593 /* release the uncleaned roots due to error */
3594 for (; i < ret; i++) {
3595 if (gang[i])
3596 btrfs_put_fs_root(gang[i]);
3597 }
3598 return err;
3599}
3600
3601int btrfs_commit_super(struct btrfs_root *root)
3602{
3603 struct btrfs_trans_handle *trans;
3604
3605 mutex_lock(&root->fs_info->cleaner_mutex);
3606 btrfs_run_delayed_iputs(root);
3607 mutex_unlock(&root->fs_info->cleaner_mutex);
3608 wake_up_process(root->fs_info->cleaner_kthread);
3609
3610 /* wait until ongoing cleanup work done */
3611 down_write(&root->fs_info->cleanup_work_sem);
3612 up_write(&root->fs_info->cleanup_work_sem);
3613
3614 trans = btrfs_join_transaction(root);
3615 if (IS_ERR(trans))
3616 return PTR_ERR(trans);
3617 return btrfs_commit_transaction(trans, root);
3618}
3619
3620int close_ctree(struct btrfs_root *root)
3621{
3622 struct btrfs_fs_info *fs_info = root->fs_info;
3623 int ret;
3624
3625 fs_info->closing = 1;
3626 smp_mb();
3627
3628 /* wait for the uuid_scan task to finish */
3629 down(&fs_info->uuid_tree_rescan_sem);
3630 /* avoid complains from lockdep et al., set sem back to initial state */
3631 up(&fs_info->uuid_tree_rescan_sem);
3632
3633 /* pause restriper - we want to resume on mount */
3634 btrfs_pause_balance(fs_info);
3635
3636 btrfs_dev_replace_suspend_for_unmount(fs_info);
3637
3638 btrfs_scrub_cancel(fs_info);
3639
3640 /* wait for any defraggers to finish */
3641 wait_event(fs_info->transaction_wait,
3642 (atomic_read(&fs_info->defrag_running) == 0));
3643
3644 /* clear out the rbtree of defraggable inodes */
3645 btrfs_cleanup_defrag_inodes(fs_info);
3646
3647 cancel_work_sync(&fs_info->async_reclaim_work);
3648
3649 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3650 ret = btrfs_commit_super(root);
3651 if (ret)
3652 btrfs_err(root->fs_info, "commit super ret %d", ret);
3653 }
3654
3655 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3656 btrfs_error_commit_super(root);
3657
3658 kthread_stop(fs_info->transaction_kthread);
3659 kthread_stop(fs_info->cleaner_kthread);
3660
3661 fs_info->closing = 2;
3662 smp_mb();
3663
3664 btrfs_free_qgroup_config(root->fs_info);
3665
3666 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3667 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3668 percpu_counter_sum(&fs_info->delalloc_bytes));
3669 }
3670
3671 btrfs_sysfs_remove_one(fs_info);
3672
3673 btrfs_free_fs_roots(fs_info);
3674
3675 btrfs_put_block_group_cache(fs_info);
3676
3677 btrfs_free_block_groups(fs_info);
3678
3679 /*
3680 * we must make sure there is not any read request to
3681 * submit after we stopping all workers.
3682 */
3683 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3684 btrfs_stop_all_workers(fs_info);
3685
3686 free_root_pointers(fs_info, 1);
3687
3688 iput(fs_info->btree_inode);
3689
3690#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3691 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3692 btrfsic_unmount(root, fs_info->fs_devices);
3693#endif
3694
3695 btrfs_close_devices(fs_info->fs_devices);
3696 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3697
3698 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3699 percpu_counter_destroy(&fs_info->delalloc_bytes);
3700 percpu_counter_destroy(&fs_info->bio_counter);
3701 bdi_destroy(&fs_info->bdi);
3702 cleanup_srcu_struct(&fs_info->subvol_srcu);
3703
3704 btrfs_free_stripe_hash_table(fs_info);
3705
3706 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3707 root->orphan_block_rsv = NULL;
3708
3709 return 0;
3710}
3711
3712int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3713 int atomic)
3714{
3715 int ret;
3716 struct inode *btree_inode = buf->pages[0]->mapping->host;
3717
3718 ret = extent_buffer_uptodate(buf);
3719 if (!ret)
3720 return ret;
3721
3722 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3723 parent_transid, atomic);
3724 if (ret == -EAGAIN)
3725 return ret;
3726 return !ret;
3727}
3728
3729int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3730{
3731 return set_extent_buffer_uptodate(buf);
3732}
3733
3734void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3735{
3736 struct btrfs_root *root;
3737 u64 transid = btrfs_header_generation(buf);
3738 int was_dirty;
3739
3740#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3741 /*
3742 * This is a fast path so only do this check if we have sanity tests
3743 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3744 * outside of the sanity tests.
3745 */
3746 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3747 return;
3748#endif
3749 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3750 btrfs_assert_tree_locked(buf);
3751 if (transid != root->fs_info->generation)
3752 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3753 "found %llu running %llu\n",
3754 buf->start, transid, root->fs_info->generation);
3755 was_dirty = set_extent_buffer_dirty(buf);
3756 if (!was_dirty)
3757 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3758 buf->len,
3759 root->fs_info->dirty_metadata_batch);
3760#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3761 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3762 btrfs_print_leaf(root, buf);
3763 ASSERT(0);
3764 }
3765#endif
3766}
3767
3768static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3769 int flush_delayed)
3770{
3771 /*
3772 * looks as though older kernels can get into trouble with
3773 * this code, they end up stuck in balance_dirty_pages forever
3774 */
3775 int ret;
3776
3777 if (current->flags & PF_MEMALLOC)
3778 return;
3779
3780 if (flush_delayed)
3781 btrfs_balance_delayed_items(root);
3782
3783 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3784 BTRFS_DIRTY_METADATA_THRESH);
3785 if (ret > 0) {
3786 balance_dirty_pages_ratelimited(
3787 root->fs_info->btree_inode->i_mapping);
3788 }
3789 return;
3790}
3791
3792void btrfs_btree_balance_dirty(struct btrfs_root *root)
3793{
3794 __btrfs_btree_balance_dirty(root, 1);
3795}
3796
3797void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3798{
3799 __btrfs_btree_balance_dirty(root, 0);
3800}
3801
3802int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3803{
3804 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3805 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3806}
3807
3808static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3809 int read_only)
3810{
3811 /*
3812 * Placeholder for checks
3813 */
3814 return 0;
3815}
3816
3817static void btrfs_error_commit_super(struct btrfs_root *root)
3818{
3819 mutex_lock(&root->fs_info->cleaner_mutex);
3820 btrfs_run_delayed_iputs(root);
3821 mutex_unlock(&root->fs_info->cleaner_mutex);
3822
3823 down_write(&root->fs_info->cleanup_work_sem);
3824 up_write(&root->fs_info->cleanup_work_sem);
3825
3826 /* cleanup FS via transaction */
3827 btrfs_cleanup_transaction(root);
3828}
3829
3830static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3831{
3832 struct btrfs_ordered_extent *ordered;
3833
3834 spin_lock(&root->ordered_extent_lock);
3835 /*
3836 * This will just short circuit the ordered completion stuff which will
3837 * make sure the ordered extent gets properly cleaned up.
3838 */
3839 list_for_each_entry(ordered, &root->ordered_extents,
3840 root_extent_list)
3841 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3842 spin_unlock(&root->ordered_extent_lock);
3843}
3844
3845static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3846{
3847 struct btrfs_root *root;
3848 struct list_head splice;
3849
3850 INIT_LIST_HEAD(&splice);
3851
3852 spin_lock(&fs_info->ordered_root_lock);
3853 list_splice_init(&fs_info->ordered_roots, &splice);
3854 while (!list_empty(&splice)) {
3855 root = list_first_entry(&splice, struct btrfs_root,
3856 ordered_root);
3857 list_move_tail(&root->ordered_root,
3858 &fs_info->ordered_roots);
3859
3860 spin_unlock(&fs_info->ordered_root_lock);
3861 btrfs_destroy_ordered_extents(root);
3862
3863 cond_resched();
3864 spin_lock(&fs_info->ordered_root_lock);
3865 }
3866 spin_unlock(&fs_info->ordered_root_lock);
3867}
3868
3869static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3870 struct btrfs_root *root)
3871{
3872 struct rb_node *node;
3873 struct btrfs_delayed_ref_root *delayed_refs;
3874 struct btrfs_delayed_ref_node *ref;
3875 int ret = 0;
3876
3877 delayed_refs = &trans->delayed_refs;
3878
3879 spin_lock(&delayed_refs->lock);
3880 if (atomic_read(&delayed_refs->num_entries) == 0) {
3881 spin_unlock(&delayed_refs->lock);
3882 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3883 return ret;
3884 }
3885
3886 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3887 struct btrfs_delayed_ref_head *head;
3888 bool pin_bytes = false;
3889
3890 head = rb_entry(node, struct btrfs_delayed_ref_head,
3891 href_node);
3892 if (!mutex_trylock(&head->mutex)) {
3893 atomic_inc(&head->node.refs);
3894 spin_unlock(&delayed_refs->lock);
3895
3896 mutex_lock(&head->mutex);
3897 mutex_unlock(&head->mutex);
3898 btrfs_put_delayed_ref(&head->node);
3899 spin_lock(&delayed_refs->lock);
3900 continue;
3901 }
3902 spin_lock(&head->lock);
3903 while ((node = rb_first(&head->ref_root)) != NULL) {
3904 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3905 rb_node);
3906 ref->in_tree = 0;
3907 rb_erase(&ref->rb_node, &head->ref_root);
3908 atomic_dec(&delayed_refs->num_entries);
3909 btrfs_put_delayed_ref(ref);
3910 }
3911 if (head->must_insert_reserved)
3912 pin_bytes = true;
3913 btrfs_free_delayed_extent_op(head->extent_op);
3914 delayed_refs->num_heads--;
3915 if (head->processing == 0)
3916 delayed_refs->num_heads_ready--;
3917 atomic_dec(&delayed_refs->num_entries);
3918 head->node.in_tree = 0;
3919 rb_erase(&head->href_node, &delayed_refs->href_root);
3920 spin_unlock(&head->lock);
3921 spin_unlock(&delayed_refs->lock);
3922 mutex_unlock(&head->mutex);
3923
3924 if (pin_bytes)
3925 btrfs_pin_extent(root, head->node.bytenr,
3926 head->node.num_bytes, 1);
3927 btrfs_put_delayed_ref(&head->node);
3928 cond_resched();
3929 spin_lock(&delayed_refs->lock);
3930 }
3931
3932 spin_unlock(&delayed_refs->lock);
3933
3934 return ret;
3935}
3936
3937static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3938{
3939 struct btrfs_inode *btrfs_inode;
3940 struct list_head splice;
3941
3942 INIT_LIST_HEAD(&splice);
3943
3944 spin_lock(&root->delalloc_lock);
3945 list_splice_init(&root->delalloc_inodes, &splice);
3946
3947 while (!list_empty(&splice)) {
3948 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3949 delalloc_inodes);
3950
3951 list_del_init(&btrfs_inode->delalloc_inodes);
3952 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3953 &btrfs_inode->runtime_flags);
3954 spin_unlock(&root->delalloc_lock);
3955
3956 btrfs_invalidate_inodes(btrfs_inode->root);
3957
3958 spin_lock(&root->delalloc_lock);
3959 }
3960
3961 spin_unlock(&root->delalloc_lock);
3962}
3963
3964static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3965{
3966 struct btrfs_root *root;
3967 struct list_head splice;
3968
3969 INIT_LIST_HEAD(&splice);
3970
3971 spin_lock(&fs_info->delalloc_root_lock);
3972 list_splice_init(&fs_info->delalloc_roots, &splice);
3973 while (!list_empty(&splice)) {
3974 root = list_first_entry(&splice, struct btrfs_root,
3975 delalloc_root);
3976 list_del_init(&root->delalloc_root);
3977 root = btrfs_grab_fs_root(root);
3978 BUG_ON(!root);
3979 spin_unlock(&fs_info->delalloc_root_lock);
3980
3981 btrfs_destroy_delalloc_inodes(root);
3982 btrfs_put_fs_root(root);
3983
3984 spin_lock(&fs_info->delalloc_root_lock);
3985 }
3986 spin_unlock(&fs_info->delalloc_root_lock);
3987}
3988
3989static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3990 struct extent_io_tree *dirty_pages,
3991 int mark)
3992{
3993 int ret;
3994 struct extent_buffer *eb;
3995 u64 start = 0;
3996 u64 end;
3997
3998 while (1) {
3999 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4000 mark, NULL);
4001 if (ret)
4002 break;
4003
4004 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4005 while (start <= end) {
4006 eb = btrfs_find_tree_block(root, start,
4007 root->leafsize);
4008 start += root->leafsize;
4009 if (!eb)
4010 continue;
4011 wait_on_extent_buffer_writeback(eb);
4012
4013 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4014 &eb->bflags))
4015 clear_extent_buffer_dirty(eb);
4016 free_extent_buffer_stale(eb);
4017 }
4018 }
4019
4020 return ret;
4021}
4022
4023static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4024 struct extent_io_tree *pinned_extents)
4025{
4026 struct extent_io_tree *unpin;
4027 u64 start;
4028 u64 end;
4029 int ret;
4030 bool loop = true;
4031
4032 unpin = pinned_extents;
4033again:
4034 while (1) {
4035 ret = find_first_extent_bit(unpin, 0, &start, &end,
4036 EXTENT_DIRTY, NULL);
4037 if (ret)
4038 break;
4039
4040 /* opt_discard */
4041 if (btrfs_test_opt(root, DISCARD))
4042 ret = btrfs_error_discard_extent(root, start,
4043 end + 1 - start,
4044 NULL);
4045
4046 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4047 btrfs_error_unpin_extent_range(root, start, end);
4048 cond_resched();
4049 }
4050
4051 if (loop) {
4052 if (unpin == &root->fs_info->freed_extents[0])
4053 unpin = &root->fs_info->freed_extents[1];
4054 else
4055 unpin = &root->fs_info->freed_extents[0];
4056 loop = false;
4057 goto again;
4058 }
4059
4060 return 0;
4061}
4062
4063void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4064 struct btrfs_root *root)
4065{
4066 btrfs_destroy_delayed_refs(cur_trans, root);
4067
4068 cur_trans->state = TRANS_STATE_COMMIT_START;
4069 wake_up(&root->fs_info->transaction_blocked_wait);
4070
4071 cur_trans->state = TRANS_STATE_UNBLOCKED;
4072 wake_up(&root->fs_info->transaction_wait);
4073
4074 btrfs_destroy_delayed_inodes(root);
4075 btrfs_assert_delayed_root_empty(root);
4076
4077 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4078 EXTENT_DIRTY);
4079 btrfs_destroy_pinned_extent(root,
4080 root->fs_info->pinned_extents);
4081
4082 cur_trans->state =TRANS_STATE_COMPLETED;
4083 wake_up(&cur_trans->commit_wait);
4084
4085 /*
4086 memset(cur_trans, 0, sizeof(*cur_trans));
4087 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4088 */
4089}
4090
4091static int btrfs_cleanup_transaction(struct btrfs_root *root)
4092{
4093 struct btrfs_transaction *t;
4094
4095 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4096
4097 spin_lock(&root->fs_info->trans_lock);
4098 while (!list_empty(&root->fs_info->trans_list)) {
4099 t = list_first_entry(&root->fs_info->trans_list,
4100 struct btrfs_transaction, list);
4101 if (t->state >= TRANS_STATE_COMMIT_START) {
4102 atomic_inc(&t->use_count);
4103 spin_unlock(&root->fs_info->trans_lock);
4104 btrfs_wait_for_commit(root, t->transid);
4105 btrfs_put_transaction(t);
4106 spin_lock(&root->fs_info->trans_lock);
4107 continue;
4108 }
4109 if (t == root->fs_info->running_transaction) {
4110 t->state = TRANS_STATE_COMMIT_DOING;
4111 spin_unlock(&root->fs_info->trans_lock);
4112 /*
4113 * We wait for 0 num_writers since we don't hold a trans
4114 * handle open currently for this transaction.
4115 */
4116 wait_event(t->writer_wait,
4117 atomic_read(&t->num_writers) == 0);
4118 } else {
4119 spin_unlock(&root->fs_info->trans_lock);
4120 }
4121 btrfs_cleanup_one_transaction(t, root);
4122
4123 spin_lock(&root->fs_info->trans_lock);
4124 if (t == root->fs_info->running_transaction)
4125 root->fs_info->running_transaction = NULL;
4126 list_del_init(&t->list);
4127 spin_unlock(&root->fs_info->trans_lock);
4128
4129 btrfs_put_transaction(t);
4130 trace_btrfs_transaction_commit(root);
4131 spin_lock(&root->fs_info->trans_lock);
4132 }
4133 spin_unlock(&root->fs_info->trans_lock);
4134 btrfs_destroy_all_ordered_extents(root->fs_info);
4135 btrfs_destroy_delayed_inodes(root);
4136 btrfs_assert_delayed_root_empty(root);
4137 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4138 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4139 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4140
4141 return 0;
4142}
4143
4144static struct extent_io_ops btree_extent_io_ops = {
4145 .readpage_end_io_hook = btree_readpage_end_io_hook,
4146 .readpage_io_failed_hook = btree_io_failed_hook,
4147 .submit_bio_hook = btree_submit_bio_hook,
4148 /* note we're sharing with inode.c for the merge bio hook */
4149 .merge_bio_hook = btrfs_merge_bio_hook,
4150};
Linux 6.x block layer and io_uring tree(s)