projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: introduce uuid-tree-gen field
[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/crc32c.h>
30#include <linux/slab.h>
31#include <linux/migrate.h>
32#include <linux/ratelimit.h>
33#include <linux/uuid.h>
34#include <linux/semaphore.h>
35#include <asm/unaligned.h>
36#include "compat.h"
37#include "ctree.h"
38#include "disk-io.h"
39#include "transaction.h"
40#include "btrfs_inode.h"
41#include "volumes.h"
42#include "print-tree.h"
43#include "async-thread.h"
44#include "locking.h"
45#include "tree-log.h"
46#include "free-space-cache.h"
47#include "inode-map.h"
48#include "check-integrity.h"
49#include "rcu-string.h"
50#include "dev-replace.h"
51#include "raid56.h"
52
53#ifdef CONFIG_X86
54#include <asm/cpufeature.h>
55#endif
56
57static struct extent_io_ops btree_extent_io_ops;
58static void end_workqueue_fn(struct btrfs_work *work);
59static void free_fs_root(struct btrfs_root *root);
60static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
61 int read_only);
62static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
63 struct btrfs_root *root);
64static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
65static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
66 struct btrfs_root *root);
67static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t);
68static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
69static int btrfs_destroy_marked_extents(struct btrfs_root *root,
70 struct extent_io_tree *dirty_pages,
71 int mark);
72static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
73 struct extent_io_tree *pinned_extents);
74static int btrfs_cleanup_transaction(struct btrfs_root *root);
75static void btrfs_error_commit_super(struct btrfs_root *root);
76
77/*
78 * end_io_wq structs are used to do processing in task context when an IO is
79 * complete. This is used during reads to verify checksums, and it is used
80 * by writes to insert metadata for new file extents after IO is complete.
81 */
82struct end_io_wq {
83 struct bio *bio;
84 bio_end_io_t *end_io;
85 void *private;
86 struct btrfs_fs_info *info;
87 int error;
88 int metadata;
89 struct list_head list;
90 struct btrfs_work work;
91};
92
93/*
94 * async submit bios are used to offload expensive checksumming
95 * onto the worker threads. They checksum file and metadata bios
96 * just before they are sent down the IO stack.
97 */
98struct async_submit_bio {
99 struct inode *inode;
100 struct bio *bio;
101 struct list_head list;
102 extent_submit_bio_hook_t *submit_bio_start;
103 extent_submit_bio_hook_t *submit_bio_done;
104 int rw;
105 int mirror_num;
106 unsigned long bio_flags;
107 /*
108 * bio_offset is optional, can be used if the pages in the bio
109 * can't tell us where in the file the bio should go
110 */
111 u64 bio_offset;
112 struct btrfs_work work;
113 int error;
114};
115
116/*
117 * Lockdep class keys for extent_buffer->lock's in this root. For a given
118 * eb, the lockdep key is determined by the btrfs_root it belongs to and
119 * the level the eb occupies in the tree.
120 *
121 * Different roots are used for different purposes and may nest inside each
122 * other and they require separate keysets. As lockdep keys should be
123 * static, assign keysets according to the purpose of the root as indicated
124 * by btrfs_root->objectid. This ensures that all special purpose roots
125 * have separate keysets.
126 *
127 * Lock-nesting across peer nodes is always done with the immediate parent
128 * node locked thus preventing deadlock. As lockdep doesn't know this, use
129 * subclass to avoid triggering lockdep warning in such cases.
130 *
131 * The key is set by the readpage_end_io_hook after the buffer has passed
132 * csum validation but before the pages are unlocked. It is also set by
133 * btrfs_init_new_buffer on freshly allocated blocks.
134 *
135 * We also add a check to make sure the highest level of the tree is the
136 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
137 * needs update as well.
138 */
139#ifdef CONFIG_DEBUG_LOCK_ALLOC
140# if BTRFS_MAX_LEVEL != 8
141# error
142# endif
143
144static struct btrfs_lockdep_keyset {
145 u64 id; /* root objectid */
146 const char *name_stem; /* lock name stem */
147 char names[BTRFS_MAX_LEVEL + 1][20];
148 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
149} btrfs_lockdep_keysets[] = {
150 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
151 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
152 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
153 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
154 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
155 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
156 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
157 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
158 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
159 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
160 { .id = 0, .name_stem = "tree" },
161};
162
163void __init btrfs_init_lockdep(void)
164{
165 int i, j;
166
167 /* initialize lockdep class names */
168 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
169 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
170
171 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
172 snprintf(ks->names[j], sizeof(ks->names[j]),
173 "btrfs-%s-%02d", ks->name_stem, j);
174 }
175}
176
177void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
178 int level)
179{
180 struct btrfs_lockdep_keyset *ks;
181
182 BUG_ON(level >= ARRAY_SIZE(ks->keys));
183
184 /* find the matching keyset, id 0 is the default entry */
185 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
186 if (ks->id == objectid)
187 break;
188
189 lockdep_set_class_and_name(&eb->lock,
190 &ks->keys[level], ks->names[level]);
191}
192
193#endif
194
195/*
196 * extents on the btree inode are pretty simple, there's one extent
197 * that covers the entire device
198 */
199static struct extent_map *btree_get_extent(struct inode *inode,
200 struct page *page, size_t pg_offset, u64 start, u64 len,
201 int create)
202{
203 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
204 struct extent_map *em;
205 int ret;
206
207 read_lock(&em_tree->lock);
208 em = lookup_extent_mapping(em_tree, start, len);
209 if (em) {
210 em->bdev =
211 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
212 read_unlock(&em_tree->lock);
213 goto out;
214 }
215 read_unlock(&em_tree->lock);
216
217 em = alloc_extent_map();
218 if (!em) {
219 em = ERR_PTR(-ENOMEM);
220 goto out;
221 }
222 em->start = 0;
223 em->len = (u64)-1;
224 em->block_len = (u64)-1;
225 em->block_start = 0;
226 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
227
228 write_lock(&em_tree->lock);
229 ret = add_extent_mapping(em_tree, em, 0);
230 if (ret == -EEXIST) {
231 free_extent_map(em);
232 em = lookup_extent_mapping(em_tree, start, len);
233 if (!em)
234 em = ERR_PTR(-EIO);
235 } else if (ret) {
236 free_extent_map(em);
237 em = ERR_PTR(ret);
238 }
239 write_unlock(&em_tree->lock);
240
241out:
242 return em;
243}
244
245u32 btrfs_csum_data(char *data, u32 seed, size_t len)
246{
247 return crc32c(seed, data, len);
248}
249
250void btrfs_csum_final(u32 crc, char *result)
251{
252 put_unaligned_le32(~crc, result);
253}
254
255/*
256 * compute the csum for a btree block, and either verify it or write it
257 * into the csum field of the block.
258 */
259static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
260 int verify)
261{
262 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
263 char *result = NULL;
264 unsigned long len;
265 unsigned long cur_len;
266 unsigned long offset = BTRFS_CSUM_SIZE;
267 char *kaddr;
268 unsigned long map_start;
269 unsigned long map_len;
270 int err;
271 u32 crc = ~(u32)0;
272 unsigned long inline_result;
273
274 len = buf->len - offset;
275 while (len > 0) {
276 err = map_private_extent_buffer(buf, offset, 32,
277 &kaddr, &map_start, &map_len);
278 if (err)
279 return 1;
280 cur_len = min(len, map_len - (offset - map_start));
281 crc = btrfs_csum_data(kaddr + offset - map_start,
282 crc, cur_len);
283 len -= cur_len;
284 offset += cur_len;
285 }
286 if (csum_size > sizeof(inline_result)) {
287 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
288 if (!result)
289 return 1;
290 } else {
291 result = (char *)&inline_result;
292 }
293
294 btrfs_csum_final(crc, result);
295
296 if (verify) {
297 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
298 u32 val;
299 u32 found = 0;
300 memcpy(&found, result, csum_size);
301
302 read_extent_buffer(buf, &val, 0, csum_size);
303 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
304 "failed on %llu wanted %X found %X "
305 "level %d\n",
306 root->fs_info->sb->s_id,
307 (unsigned long long)buf->start, val, found,
308 btrfs_header_level(buf));
309 if (result != (char *)&inline_result)
310 kfree(result);
311 return 1;
312 }
313 } else {
314 write_extent_buffer(buf, result, 0, csum_size);
315 }
316 if (result != (char *)&inline_result)
317 kfree(result);
318 return 0;
319}
320
321/*
322 * we can't consider a given block up to date unless the transid of the
323 * block matches the transid in the parent node's pointer. This is how we
324 * detect blocks that either didn't get written at all or got written
325 * in the wrong place.
326 */
327static int verify_parent_transid(struct extent_io_tree *io_tree,
328 struct extent_buffer *eb, u64 parent_transid,
329 int atomic)
330{
331 struct extent_state *cached_state = NULL;
332 int ret;
333
334 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
335 return 0;
336
337 if (atomic)
338 return -EAGAIN;
339
340 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
341 0, &cached_state);
342 if (extent_buffer_uptodate(eb) &&
343 btrfs_header_generation(eb) == parent_transid) {
344 ret = 0;
345 goto out;
346 }
347 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
348 "found %llu\n",
349 (unsigned long long)eb->start,
350 (unsigned long long)parent_transid,
351 (unsigned long long)btrfs_header_generation(eb));
352 ret = 1;
353 clear_extent_buffer_uptodate(eb);
354out:
355 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
356 &cached_state, GFP_NOFS);
357 return ret;
358}
359
360/*
361 * Return 0 if the superblock checksum type matches the checksum value of that
362 * algorithm. Pass the raw disk superblock data.
363 */
364static int btrfs_check_super_csum(char *raw_disk_sb)
365{
366 struct btrfs_super_block *disk_sb =
367 (struct btrfs_super_block *)raw_disk_sb;
368 u16 csum_type = btrfs_super_csum_type(disk_sb);
369 int ret = 0;
370
371 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
372 u32 crc = ~(u32)0;
373 const int csum_size = sizeof(crc);
374 char result[csum_size];
375
376 /*
377 * The super_block structure does not span the whole
378 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
379 * is filled with zeros and is included in the checkum.
380 */
381 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
382 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
383 btrfs_csum_final(crc, result);
384
385 if (memcmp(raw_disk_sb, result, csum_size))
386 ret = 1;
387
388 if (ret && btrfs_super_generation(disk_sb) < 10) {
389 printk(KERN_WARNING "btrfs: super block crcs don't match, older mkfs detected\n");
390 ret = 0;
391 }
392 }
393
394 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
395 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
396 csum_type);
397 ret = 1;
398 }
399
400 return ret;
401}
402
403/*
404 * helper to read a given tree block, doing retries as required when
405 * the checksums don't match and we have alternate mirrors to try.
406 */
407static int btree_read_extent_buffer_pages(struct btrfs_root *root,
408 struct extent_buffer *eb,
409 u64 start, u64 parent_transid)
410{
411 struct extent_io_tree *io_tree;
412 int failed = 0;
413 int ret;
414 int num_copies = 0;
415 int mirror_num = 0;
416 int failed_mirror = 0;
417
418 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
419 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
420 while (1) {
421 ret = read_extent_buffer_pages(io_tree, eb, start,
422 WAIT_COMPLETE,
423 btree_get_extent, mirror_num);
424 if (!ret) {
425 if (!verify_parent_transid(io_tree, eb,
426 parent_transid, 0))
427 break;
428 else
429 ret = -EIO;
430 }
431
432 /*
433 * This buffer's crc is fine, but its contents are corrupted, so
434 * there is no reason to read the other copies, they won't be
435 * any less wrong.
436 */
437 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
438 break;
439
440 num_copies = btrfs_num_copies(root->fs_info,
441 eb->start, eb->len);
442 if (num_copies == 1)
443 break;
444
445 if (!failed_mirror) {
446 failed = 1;
447 failed_mirror = eb->read_mirror;
448 }
449
450 mirror_num++;
451 if (mirror_num == failed_mirror)
452 mirror_num++;
453
454 if (mirror_num > num_copies)
455 break;
456 }
457
458 if (failed && !ret && failed_mirror)
459 repair_eb_io_failure(root, eb, failed_mirror);
460
461 return ret;
462}
463
464/*
465 * checksum a dirty tree block before IO. This has extra checks to make sure
466 * we only fill in the checksum field in the first page of a multi-page block
467 */
468
469static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
470{
471 struct extent_io_tree *tree;
472 u64 start = page_offset(page);
473 u64 found_start;
474 struct extent_buffer *eb;
475
476 tree = &BTRFS_I(page->mapping->host)->io_tree;
477
478 eb = (struct extent_buffer *)page->private;
479 if (page != eb->pages[0])
480 return 0;
481 found_start = btrfs_header_bytenr(eb);
482 if (found_start != start) {
483 WARN_ON(1);
484 return 0;
485 }
486 if (!PageUptodate(page)) {
487 WARN_ON(1);
488 return 0;
489 }
490 csum_tree_block(root, eb, 0);
491 return 0;
492}
493
494static int check_tree_block_fsid(struct btrfs_root *root,
495 struct extent_buffer *eb)
496{
497 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
498 u8 fsid[BTRFS_UUID_SIZE];
499 int ret = 1;
500
501 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
502 BTRFS_FSID_SIZE);
503 while (fs_devices) {
504 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
505 ret = 0;
506 break;
507 }
508 fs_devices = fs_devices->seed;
509 }
510 return ret;
511}
512
513#define CORRUPT(reason, eb, root, slot) \
514 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
515 "root=%llu, slot=%d\n", reason, \
516 (unsigned long long)btrfs_header_bytenr(eb), \
517 (unsigned long long)root->objectid, slot)
518
519static noinline int check_leaf(struct btrfs_root *root,
520 struct extent_buffer *leaf)
521{
522 struct btrfs_key key;
523 struct btrfs_key leaf_key;
524 u32 nritems = btrfs_header_nritems(leaf);
525 int slot;
526
527 if (nritems == 0)
528 return 0;
529
530 /* Check the 0 item */
531 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
532 BTRFS_LEAF_DATA_SIZE(root)) {
533 CORRUPT("invalid item offset size pair", leaf, root, 0);
534 return -EIO;
535 }
536
537 /*
538 * Check to make sure each items keys are in the correct order and their
539 * offsets make sense. We only have to loop through nritems-1 because
540 * we check the current slot against the next slot, which verifies the
541 * next slot's offset+size makes sense and that the current's slot
542 * offset is correct.
543 */
544 for (slot = 0; slot < nritems - 1; slot++) {
545 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
546 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
547
548 /* Make sure the keys are in the right order */
549 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
550 CORRUPT("bad key order", leaf, root, slot);
551 return -EIO;
552 }
553
554 /*
555 * Make sure the offset and ends are right, remember that the
556 * item data starts at the end of the leaf and grows towards the
557 * front.
558 */
559 if (btrfs_item_offset_nr(leaf, slot) !=
560 btrfs_item_end_nr(leaf, slot + 1)) {
561 CORRUPT("slot offset bad", leaf, root, slot);
562 return -EIO;
563 }
564
565 /*
566 * Check to make sure that we don't point outside of the leaf,
567 * just incase all the items are consistent to eachother, but
568 * all point outside of the leaf.
569 */
570 if (btrfs_item_end_nr(leaf, slot) >
571 BTRFS_LEAF_DATA_SIZE(root)) {
572 CORRUPT("slot end outside of leaf", leaf, root, slot);
573 return -EIO;
574 }
575 }
576
577 return 0;
578}
579
580static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
581 u64 phy_offset, struct page *page,
582 u64 start, u64 end, int mirror)
583{
584 struct extent_io_tree *tree;
585 u64 found_start;
586 int found_level;
587 struct extent_buffer *eb;
588 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
589 int ret = 0;
590 int reads_done;
591
592 if (!page->private)
593 goto out;
594
595 tree = &BTRFS_I(page->mapping->host)->io_tree;
596 eb = (struct extent_buffer *)page->private;
597
598 /* the pending IO might have been the only thing that kept this buffer
599 * in memory. Make sure we have a ref for all this other checks
600 */
601 extent_buffer_get(eb);
602
603 reads_done = atomic_dec_and_test(&eb->io_pages);
604 if (!reads_done)
605 goto err;
606
607 eb->read_mirror = mirror;
608 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
609 ret = -EIO;
610 goto err;
611 }
612
613 found_start = btrfs_header_bytenr(eb);
614 if (found_start != eb->start) {
615 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
616 "%llu %llu\n",
617 (unsigned long long)found_start,
618 (unsigned long long)eb->start);
619 ret = -EIO;
620 goto err;
621 }
622 if (check_tree_block_fsid(root, eb)) {
623 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
624 (unsigned long long)eb->start);
625 ret = -EIO;
626 goto err;
627 }
628 found_level = btrfs_header_level(eb);
629 if (found_level >= BTRFS_MAX_LEVEL) {
630 btrfs_info(root->fs_info, "bad tree block level %d\n",
631 (int)btrfs_header_level(eb));
632 ret = -EIO;
633 goto err;
634 }
635
636 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
637 eb, found_level);
638
639 ret = csum_tree_block(root, eb, 1);
640 if (ret) {
641 ret = -EIO;
642 goto err;
643 }
644
645 /*
646 * If this is a leaf block and it is corrupt, set the corrupt bit so
647 * that we don't try and read the other copies of this block, just
648 * return -EIO.
649 */
650 if (found_level == 0 && check_leaf(root, eb)) {
651 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
652 ret = -EIO;
653 }
654
655 if (!ret)
656 set_extent_buffer_uptodate(eb);
657err:
658 if (reads_done &&
659 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
660 btree_readahead_hook(root, eb, eb->start, ret);
661
662 if (ret) {
663 /*
664 * our io error hook is going to dec the io pages
665 * again, we have to make sure it has something
666 * to decrement
667 */
668 atomic_inc(&eb->io_pages);
669 clear_extent_buffer_uptodate(eb);
670 }
671 free_extent_buffer(eb);
672out:
673 return ret;
674}
675
676static int btree_io_failed_hook(struct page *page, int failed_mirror)
677{
678 struct extent_buffer *eb;
679 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
680
681 eb = (struct extent_buffer *)page->private;
682 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
683 eb->read_mirror = failed_mirror;
684 atomic_dec(&eb->io_pages);
685 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
686 btree_readahead_hook(root, eb, eb->start, -EIO);
687 return -EIO; /* we fixed nothing */
688}
689
690static void end_workqueue_bio(struct bio *bio, int err)
691{
692 struct end_io_wq *end_io_wq = bio->bi_private;
693 struct btrfs_fs_info *fs_info;
694
695 fs_info = end_io_wq->info;
696 end_io_wq->error = err;
697 end_io_wq->work.func = end_workqueue_fn;
698 end_io_wq->work.flags = 0;
699
700 if (bio->bi_rw & REQ_WRITE) {
701 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
702 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
703 &end_io_wq->work);
704 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
705 btrfs_queue_worker(&fs_info->endio_freespace_worker,
706 &end_io_wq->work);
707 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
708 btrfs_queue_worker(&fs_info->endio_raid56_workers,
709 &end_io_wq->work);
710 else
711 btrfs_queue_worker(&fs_info->endio_write_workers,
712 &end_io_wq->work);
713 } else {
714 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
715 btrfs_queue_worker(&fs_info->endio_raid56_workers,
716 &end_io_wq->work);
717 else if (end_io_wq->metadata)
718 btrfs_queue_worker(&fs_info->endio_meta_workers,
719 &end_io_wq->work);
720 else
721 btrfs_queue_worker(&fs_info->endio_workers,
722 &end_io_wq->work);
723 }
724}
725
726/*
727 * For the metadata arg you want
728 *
729 * 0 - if data
730 * 1 - if normal metadta
731 * 2 - if writing to the free space cache area
732 * 3 - raid parity work
733 */
734int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
735 int metadata)
736{
737 struct end_io_wq *end_io_wq;
738 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
739 if (!end_io_wq)
740 return -ENOMEM;
741
742 end_io_wq->private = bio->bi_private;
743 end_io_wq->end_io = bio->bi_end_io;
744 end_io_wq->info = info;
745 end_io_wq->error = 0;
746 end_io_wq->bio = bio;
747 end_io_wq->metadata = metadata;
748
749 bio->bi_private = end_io_wq;
750 bio->bi_end_io = end_workqueue_bio;
751 return 0;
752}
753
754unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
755{
756 unsigned long limit = min_t(unsigned long,
757 info->workers.max_workers,
758 info->fs_devices->open_devices);
759 return 256 * limit;
760}
761
762static void run_one_async_start(struct btrfs_work *work)
763{
764 struct async_submit_bio *async;
765 int ret;
766
767 async = container_of(work, struct async_submit_bio, work);
768 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
769 async->mirror_num, async->bio_flags,
770 async->bio_offset);
771 if (ret)
772 async->error = ret;
773}
774
775static void run_one_async_done(struct btrfs_work *work)
776{
777 struct btrfs_fs_info *fs_info;
778 struct async_submit_bio *async;
779 int limit;
780
781 async = container_of(work, struct async_submit_bio, work);
782 fs_info = BTRFS_I(async->inode)->root->fs_info;
783
784 limit = btrfs_async_submit_limit(fs_info);
785 limit = limit * 2 / 3;
786
787 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
788 waitqueue_active(&fs_info->async_submit_wait))
789 wake_up(&fs_info->async_submit_wait);
790
791 /* If an error occured we just want to clean up the bio and move on */
792 if (async->error) {
793 bio_endio(async->bio, async->error);
794 return;
795 }
796
797 async->submit_bio_done(async->inode, async->rw, async->bio,
798 async->mirror_num, async->bio_flags,
799 async->bio_offset);
800}
801
802static void run_one_async_free(struct btrfs_work *work)
803{
804 struct async_submit_bio *async;
805
806 async = container_of(work, struct async_submit_bio, work);
807 kfree(async);
808}
809
810int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
811 int rw, struct bio *bio, int mirror_num,
812 unsigned long bio_flags,
813 u64 bio_offset,
814 extent_submit_bio_hook_t *submit_bio_start,
815 extent_submit_bio_hook_t *submit_bio_done)
816{
817 struct async_submit_bio *async;
818
819 async = kmalloc(sizeof(*async), GFP_NOFS);
820 if (!async)
821 return -ENOMEM;
822
823 async->inode = inode;
824 async->rw = rw;
825 async->bio = bio;
826 async->mirror_num = mirror_num;
827 async->submit_bio_start = submit_bio_start;
828 async->submit_bio_done = submit_bio_done;
829
830 async->work.func = run_one_async_start;
831 async->work.ordered_func = run_one_async_done;
832 async->work.ordered_free = run_one_async_free;
833
834 async->work.flags = 0;
835 async->bio_flags = bio_flags;
836 async->bio_offset = bio_offset;
837
838 async->error = 0;
839
840 atomic_inc(&fs_info->nr_async_submits);
841
842 if (rw & REQ_SYNC)
843 btrfs_set_work_high_prio(&async->work);
844
845 btrfs_queue_worker(&fs_info->workers, &async->work);
846
847 while (atomic_read(&fs_info->async_submit_draining) &&
848 atomic_read(&fs_info->nr_async_submits)) {
849 wait_event(fs_info->async_submit_wait,
850 (atomic_read(&fs_info->nr_async_submits) == 0));
851 }
852
853 return 0;
854}
855
856static int btree_csum_one_bio(struct bio *bio)
857{
858 struct bio_vec *bvec = bio->bi_io_vec;
859 int bio_index = 0;
860 struct btrfs_root *root;
861 int ret = 0;
862
863 WARN_ON(bio->bi_vcnt <= 0);
864 while (bio_index < bio->bi_vcnt) {
865 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
866 ret = csum_dirty_buffer(root, bvec->bv_page);
867 if (ret)
868 break;
869 bio_index++;
870 bvec++;
871 }
872 return ret;
873}
874
875static int __btree_submit_bio_start(struct inode *inode, int rw,
876 struct bio *bio, int mirror_num,
877 unsigned long bio_flags,
878 u64 bio_offset)
879{
880 /*
881 * when we're called for a write, we're already in the async
882 * submission context. Just jump into btrfs_map_bio
883 */
884 return btree_csum_one_bio(bio);
885}
886
887static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
888 int mirror_num, unsigned long bio_flags,
889 u64 bio_offset)
890{
891 int ret;
892
893 /*
894 * when we're called for a write, we're already in the async
895 * submission context. Just jump into btrfs_map_bio
896 */
897 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
898 if (ret)
899 bio_endio(bio, ret);
900 return ret;
901}
902
903static int check_async_write(struct inode *inode, unsigned long bio_flags)
904{
905 if (bio_flags & EXTENT_BIO_TREE_LOG)
906 return 0;
907#ifdef CONFIG_X86
908 if (cpu_has_xmm4_2)
909 return 0;
910#endif
911 return 1;
912}
913
914static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
915 int mirror_num, unsigned long bio_flags,
916 u64 bio_offset)
917{
918 int async = check_async_write(inode, bio_flags);
919 int ret;
920
921 if (!(rw & REQ_WRITE)) {
922 /*
923 * called for a read, do the setup so that checksum validation
924 * can happen in the async kernel threads
925 */
926 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
927 bio, 1);
928 if (ret)
929 goto out_w_error;
930 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
931 mirror_num, 0);
932 } else if (!async) {
933 ret = btree_csum_one_bio(bio);
934 if (ret)
935 goto out_w_error;
936 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
937 mirror_num, 0);
938 } else {
939 /*
940 * kthread helpers are used to submit writes so that
941 * checksumming can happen in parallel across all CPUs
942 */
943 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
944 inode, rw, bio, mirror_num, 0,
945 bio_offset,
946 __btree_submit_bio_start,
947 __btree_submit_bio_done);
948 }
949
950 if (ret) {
951out_w_error:
952 bio_endio(bio, ret);
953 }
954 return ret;
955}
956
957#ifdef CONFIG_MIGRATION
958static int btree_migratepage(struct address_space *mapping,
959 struct page *newpage, struct page *page,
960 enum migrate_mode mode)
961{
962 /*
963 * we can't safely write a btree page from here,
964 * we haven't done the locking hook
965 */
966 if (PageDirty(page))
967 return -EAGAIN;
968 /*
969 * Buffers may be managed in a filesystem specific way.
970 * We must have no buffers or drop them.
971 */
972 if (page_has_private(page) &&
973 !try_to_release_page(page, GFP_KERNEL))
974 return -EAGAIN;
975 return migrate_page(mapping, newpage, page, mode);
976}
977#endif
978
979
980static int btree_writepages(struct address_space *mapping,
981 struct writeback_control *wbc)
982{
983 struct extent_io_tree *tree;
984 struct btrfs_fs_info *fs_info;
985 int ret;
986
987 tree = &BTRFS_I(mapping->host)->io_tree;
988 if (wbc->sync_mode == WB_SYNC_NONE) {
989
990 if (wbc->for_kupdate)
991 return 0;
992
993 fs_info = BTRFS_I(mapping->host)->root->fs_info;
994 /* this is a bit racy, but that's ok */
995 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
996 BTRFS_DIRTY_METADATA_THRESH);
997 if (ret < 0)
998 return 0;
999 }
1000 return btree_write_cache_pages(mapping, wbc);
1001}
1002
1003static int btree_readpage(struct file *file, struct page *page)
1004{
1005 struct extent_io_tree *tree;
1006 tree = &BTRFS_I(page->mapping->host)->io_tree;
1007 return extent_read_full_page(tree, page, btree_get_extent, 0);
1008}
1009
1010static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1011{
1012 if (PageWriteback(page) || PageDirty(page))
1013 return 0;
1014
1015 return try_release_extent_buffer(page);
1016}
1017
1018static void btree_invalidatepage(struct page *page, unsigned int offset,
1019 unsigned int length)
1020{
1021 struct extent_io_tree *tree;
1022 tree = &BTRFS_I(page->mapping->host)->io_tree;
1023 extent_invalidatepage(tree, page, offset);
1024 btree_releasepage(page, GFP_NOFS);
1025 if (PagePrivate(page)) {
1026 printk(KERN_WARNING "btrfs warning page private not zero "
1027 "on page %llu\n", (unsigned long long)page_offset(page));
1028 ClearPagePrivate(page);
1029 set_page_private(page, 0);
1030 page_cache_release(page);
1031 }
1032}
1033
1034static int btree_set_page_dirty(struct page *page)
1035{
1036#ifdef DEBUG
1037 struct extent_buffer *eb;
1038
1039 BUG_ON(!PagePrivate(page));
1040 eb = (struct extent_buffer *)page->private;
1041 BUG_ON(!eb);
1042 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1043 BUG_ON(!atomic_read(&eb->refs));
1044 btrfs_assert_tree_locked(eb);
1045#endif
1046 return __set_page_dirty_nobuffers(page);
1047}
1048
1049static const struct address_space_operations btree_aops = {
1050 .readpage = btree_readpage,
1051 .writepages = btree_writepages,
1052 .releasepage = btree_releasepage,
1053 .invalidatepage = btree_invalidatepage,
1054#ifdef CONFIG_MIGRATION
1055 .migratepage = btree_migratepage,
1056#endif
1057 .set_page_dirty = btree_set_page_dirty,
1058};
1059
1060int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1061 u64 parent_transid)
1062{
1063 struct extent_buffer *buf = NULL;
1064 struct inode *btree_inode = root->fs_info->btree_inode;
1065 int ret = 0;
1066
1067 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1068 if (!buf)
1069 return 0;
1070 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1071 buf, 0, WAIT_NONE, btree_get_extent, 0);
1072 free_extent_buffer(buf);
1073 return ret;
1074}
1075
1076int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1077 int mirror_num, struct extent_buffer **eb)
1078{
1079 struct extent_buffer *buf = NULL;
1080 struct inode *btree_inode = root->fs_info->btree_inode;
1081 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1082 int ret;
1083
1084 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1085 if (!buf)
1086 return 0;
1087
1088 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1089
1090 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1091 btree_get_extent, mirror_num);
1092 if (ret) {
1093 free_extent_buffer(buf);
1094 return ret;
1095 }
1096
1097 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1098 free_extent_buffer(buf);
1099 return -EIO;
1100 } else if (extent_buffer_uptodate(buf)) {
1101 *eb = buf;
1102 } else {
1103 free_extent_buffer(buf);
1104 }
1105 return 0;
1106}
1107
1108struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1109 u64 bytenr, u32 blocksize)
1110{
1111 struct inode *btree_inode = root->fs_info->btree_inode;
1112 struct extent_buffer *eb;
1113 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1114 bytenr, blocksize);
1115 return eb;
1116}
1117
1118struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1119 u64 bytenr, u32 blocksize)
1120{
1121 struct inode *btree_inode = root->fs_info->btree_inode;
1122 struct extent_buffer *eb;
1123
1124 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1125 bytenr, blocksize);
1126 return eb;
1127}
1128
1129
1130int btrfs_write_tree_block(struct extent_buffer *buf)
1131{
1132 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1133 buf->start + buf->len - 1);
1134}
1135
1136int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1137{
1138 return filemap_fdatawait_range(buf->pages[0]->mapping,
1139 buf->start, buf->start + buf->len - 1);
1140}
1141
1142struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1143 u32 blocksize, u64 parent_transid)
1144{
1145 struct extent_buffer *buf = NULL;
1146 int ret;
1147
1148 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1149 if (!buf)
1150 return NULL;
1151
1152 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1153 if (ret) {
1154 free_extent_buffer(buf);
1155 return NULL;
1156 }
1157 return buf;
1158
1159}
1160
1161void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1162 struct extent_buffer *buf)
1163{
1164 struct btrfs_fs_info *fs_info = root->fs_info;
1165
1166 if (btrfs_header_generation(buf) ==
1167 fs_info->running_transaction->transid) {
1168 btrfs_assert_tree_locked(buf);
1169
1170 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1171 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1172 -buf->len,
1173 fs_info->dirty_metadata_batch);
1174 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1175 btrfs_set_lock_blocking(buf);
1176 clear_extent_buffer_dirty(buf);
1177 }
1178 }
1179}
1180
1181static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1182 u32 stripesize, struct btrfs_root *root,
1183 struct btrfs_fs_info *fs_info,
1184 u64 objectid)
1185{
1186 root->node = NULL;
1187 root->commit_root = NULL;
1188 root->sectorsize = sectorsize;
1189 root->nodesize = nodesize;
1190 root->leafsize = leafsize;
1191 root->stripesize = stripesize;
1192 root->ref_cows = 0;
1193 root->track_dirty = 0;
1194 root->in_radix = 0;
1195 root->orphan_item_inserted = 0;
1196 root->orphan_cleanup_state = 0;
1197
1198 root->objectid = objectid;
1199 root->last_trans = 0;
1200 root->highest_objectid = 0;
1201 root->nr_delalloc_inodes = 0;
1202 root->nr_ordered_extents = 0;
1203 root->name = NULL;
1204 root->inode_tree = RB_ROOT;
1205 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1206 root->block_rsv = NULL;
1207 root->orphan_block_rsv = NULL;
1208
1209 INIT_LIST_HEAD(&root->dirty_list);
1210 INIT_LIST_HEAD(&root->root_list);
1211 INIT_LIST_HEAD(&root->delalloc_inodes);
1212 INIT_LIST_HEAD(&root->delalloc_root);
1213 INIT_LIST_HEAD(&root->ordered_extents);
1214 INIT_LIST_HEAD(&root->ordered_root);
1215 INIT_LIST_HEAD(&root->logged_list[0]);
1216 INIT_LIST_HEAD(&root->logged_list[1]);
1217 spin_lock_init(&root->orphan_lock);
1218 spin_lock_init(&root->inode_lock);
1219 spin_lock_init(&root->delalloc_lock);
1220 spin_lock_init(&root->ordered_extent_lock);
1221 spin_lock_init(&root->accounting_lock);
1222 spin_lock_init(&root->log_extents_lock[0]);
1223 spin_lock_init(&root->log_extents_lock[1]);
1224 mutex_init(&root->objectid_mutex);
1225 mutex_init(&root->log_mutex);
1226 init_waitqueue_head(&root->log_writer_wait);
1227 init_waitqueue_head(&root->log_commit_wait[0]);
1228 init_waitqueue_head(&root->log_commit_wait[1]);
1229 atomic_set(&root->log_commit[0], 0);
1230 atomic_set(&root->log_commit[1], 0);
1231 atomic_set(&root->log_writers, 0);
1232 atomic_set(&root->log_batch, 0);
1233 atomic_set(&root->orphan_inodes, 0);
1234 atomic_set(&root->refs, 1);
1235 root->log_transid = 0;
1236 root->last_log_commit = 0;
1237 extent_io_tree_init(&root->dirty_log_pages,
1238 fs_info->btree_inode->i_mapping);
1239
1240 memset(&root->root_key, 0, sizeof(root->root_key));
1241 memset(&root->root_item, 0, sizeof(root->root_item));
1242 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1243 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1244 root->defrag_trans_start = fs_info->generation;
1245 init_completion(&root->kobj_unregister);
1246 root->defrag_running = 0;
1247 root->root_key.objectid = objectid;
1248 root->anon_dev = 0;
1249
1250 spin_lock_init(&root->root_item_lock);
1251}
1252
1253static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1254{
1255 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1256 if (root)
1257 root->fs_info = fs_info;
1258 return root;
1259}
1260
1261struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1262 struct btrfs_fs_info *fs_info,
1263 u64 objectid)
1264{
1265 struct extent_buffer *leaf;
1266 struct btrfs_root *tree_root = fs_info->tree_root;
1267 struct btrfs_root *root;
1268 struct btrfs_key key;
1269 int ret = 0;
1270 u64 bytenr;
1271 uuid_le uuid;
1272
1273 root = btrfs_alloc_root(fs_info);
1274 if (!root)
1275 return ERR_PTR(-ENOMEM);
1276
1277 __setup_root(tree_root->nodesize, tree_root->leafsize,
1278 tree_root->sectorsize, tree_root->stripesize,
1279 root, fs_info, objectid);
1280 root->root_key.objectid = objectid;
1281 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1282 root->root_key.offset = 0;
1283
1284 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1285 0, objectid, NULL, 0, 0, 0);
1286 if (IS_ERR(leaf)) {
1287 ret = PTR_ERR(leaf);
1288 leaf = NULL;
1289 goto fail;
1290 }
1291
1292 bytenr = leaf->start;
1293 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1294 btrfs_set_header_bytenr(leaf, leaf->start);
1295 btrfs_set_header_generation(leaf, trans->transid);
1296 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1297 btrfs_set_header_owner(leaf, objectid);
1298 root->node = leaf;
1299
1300 write_extent_buffer(leaf, fs_info->fsid,
1301 (unsigned long)btrfs_header_fsid(leaf),
1302 BTRFS_FSID_SIZE);
1303 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1304 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1305 BTRFS_UUID_SIZE);
1306 btrfs_mark_buffer_dirty(leaf);
1307
1308 root->commit_root = btrfs_root_node(root);
1309 root->track_dirty = 1;
1310
1311
1312 root->root_item.flags = 0;
1313 root->root_item.byte_limit = 0;
1314 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1315 btrfs_set_root_generation(&root->root_item, trans->transid);
1316 btrfs_set_root_level(&root->root_item, 0);
1317 btrfs_set_root_refs(&root->root_item, 1);
1318 btrfs_set_root_used(&root->root_item, leaf->len);
1319 btrfs_set_root_last_snapshot(&root->root_item, 0);
1320 btrfs_set_root_dirid(&root->root_item, 0);
1321 uuid_le_gen(&uuid);
1322 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1323 root->root_item.drop_level = 0;
1324
1325 key.objectid = objectid;
1326 key.type = BTRFS_ROOT_ITEM_KEY;
1327 key.offset = 0;
1328 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1329 if (ret)
1330 goto fail;
1331
1332 btrfs_tree_unlock(leaf);
1333
1334 return root;
1335
1336fail:
1337 if (leaf) {
1338 btrfs_tree_unlock(leaf);
1339 free_extent_buffer(leaf);
1340 }
1341 kfree(root);
1342
1343 return ERR_PTR(ret);
1344}
1345
1346static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1347 struct btrfs_fs_info *fs_info)
1348{
1349 struct btrfs_root *root;
1350 struct btrfs_root *tree_root = fs_info->tree_root;
1351 struct extent_buffer *leaf;
1352
1353 root = btrfs_alloc_root(fs_info);
1354 if (!root)
1355 return ERR_PTR(-ENOMEM);
1356
1357 __setup_root(tree_root->nodesize, tree_root->leafsize,
1358 tree_root->sectorsize, tree_root->stripesize,
1359 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1360
1361 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1362 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1363 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1364 /*
1365 * log trees do not get reference counted because they go away
1366 * before a real commit is actually done. They do store pointers
1367 * to file data extents, and those reference counts still get
1368 * updated (along with back refs to the log tree).
1369 */
1370 root->ref_cows = 0;
1371
1372 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1373 BTRFS_TREE_LOG_OBJECTID, NULL,
1374 0, 0, 0);
1375 if (IS_ERR(leaf)) {
1376 kfree(root);
1377 return ERR_CAST(leaf);
1378 }
1379
1380 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1381 btrfs_set_header_bytenr(leaf, leaf->start);
1382 btrfs_set_header_generation(leaf, trans->transid);
1383 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1384 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1385 root->node = leaf;
1386
1387 write_extent_buffer(root->node, root->fs_info->fsid,
1388 (unsigned long)btrfs_header_fsid(root->node),
1389 BTRFS_FSID_SIZE);
1390 btrfs_mark_buffer_dirty(root->node);
1391 btrfs_tree_unlock(root->node);
1392 return root;
1393}
1394
1395int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1396 struct btrfs_fs_info *fs_info)
1397{
1398 struct btrfs_root *log_root;
1399
1400 log_root = alloc_log_tree(trans, fs_info);
1401 if (IS_ERR(log_root))
1402 return PTR_ERR(log_root);
1403 WARN_ON(fs_info->log_root_tree);
1404 fs_info->log_root_tree = log_root;
1405 return 0;
1406}
1407
1408int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1409 struct btrfs_root *root)
1410{
1411 struct btrfs_root *log_root;
1412 struct btrfs_inode_item *inode_item;
1413
1414 log_root = alloc_log_tree(trans, root->fs_info);
1415 if (IS_ERR(log_root))
1416 return PTR_ERR(log_root);
1417
1418 log_root->last_trans = trans->transid;
1419 log_root->root_key.offset = root->root_key.objectid;
1420
1421 inode_item = &log_root->root_item.inode;
1422 btrfs_set_stack_inode_generation(inode_item, 1);
1423 btrfs_set_stack_inode_size(inode_item, 3);
1424 btrfs_set_stack_inode_nlink(inode_item, 1);
1425 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1426 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1427
1428 btrfs_set_root_node(&log_root->root_item, log_root->node);
1429
1430 WARN_ON(root->log_root);
1431 root->log_root = log_root;
1432 root->log_transid = 0;
1433 root->last_log_commit = 0;
1434 return 0;
1435}
1436
1437static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1438 struct btrfs_key *key)
1439{
1440 struct btrfs_root *root;
1441 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1442 struct btrfs_path *path;
1443 u64 generation;
1444 u32 blocksize;
1445 int ret;
1446
1447 path = btrfs_alloc_path();
1448 if (!path)
1449 return ERR_PTR(-ENOMEM);
1450
1451 root = btrfs_alloc_root(fs_info);
1452 if (!root) {
1453 ret = -ENOMEM;
1454 goto alloc_fail;
1455 }
1456
1457 __setup_root(tree_root->nodesize, tree_root->leafsize,
1458 tree_root->sectorsize, tree_root->stripesize,
1459 root, fs_info, key->objectid);
1460
1461 ret = btrfs_find_root(tree_root, key, path,
1462 &root->root_item, &root->root_key);
1463 if (ret) {
1464 if (ret > 0)
1465 ret = -ENOENT;
1466 goto find_fail;
1467 }
1468
1469 generation = btrfs_root_generation(&root->root_item);
1470 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1471 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1472 blocksize, generation);
1473 if (!root->node) {
1474 ret = -ENOMEM;
1475 goto find_fail;
1476 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1477 ret = -EIO;
1478 goto read_fail;
1479 }
1480 root->commit_root = btrfs_root_node(root);
1481out:
1482 btrfs_free_path(path);
1483 return root;
1484
1485read_fail:
1486 free_extent_buffer(root->node);
1487find_fail:
1488 kfree(root);
1489alloc_fail:
1490 root = ERR_PTR(ret);
1491 goto out;
1492}
1493
1494struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1495 struct btrfs_key *location)
1496{
1497 struct btrfs_root *root;
1498
1499 root = btrfs_read_tree_root(tree_root, location);
1500 if (IS_ERR(root))
1501 return root;
1502
1503 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1504 root->ref_cows = 1;
1505 btrfs_check_and_init_root_item(&root->root_item);
1506 }
1507
1508 return root;
1509}
1510
1511int btrfs_init_fs_root(struct btrfs_root *root)
1512{
1513 int ret;
1514
1515 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1516 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1517 GFP_NOFS);
1518 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1519 ret = -ENOMEM;
1520 goto fail;
1521 }
1522
1523 btrfs_init_free_ino_ctl(root);
1524 mutex_init(&root->fs_commit_mutex);
1525 spin_lock_init(&root->cache_lock);
1526 init_waitqueue_head(&root->cache_wait);
1527
1528 ret = get_anon_bdev(&root->anon_dev);
1529 if (ret)
1530 goto fail;
1531 return 0;
1532fail:
1533 kfree(root->free_ino_ctl);
1534 kfree(root->free_ino_pinned);
1535 return ret;
1536}
1537
1538static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1539 u64 root_id)
1540{
1541 struct btrfs_root *root;
1542
1543 spin_lock(&fs_info->fs_roots_radix_lock);
1544 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1545 (unsigned long)root_id);
1546 spin_unlock(&fs_info->fs_roots_radix_lock);
1547 return root;
1548}
1549
1550int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1551 struct btrfs_root *root)
1552{
1553 int ret;
1554
1555 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1556 if (ret)
1557 return ret;
1558
1559 spin_lock(&fs_info->fs_roots_radix_lock);
1560 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1561 (unsigned long)root->root_key.objectid,
1562 root);
1563 if (ret == 0)
1564 root->in_radix = 1;
1565 spin_unlock(&fs_info->fs_roots_radix_lock);
1566 radix_tree_preload_end();
1567
1568 return ret;
1569}
1570
1571struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1572 struct btrfs_key *location)
1573{
1574 struct btrfs_root *root;
1575 int ret;
1576
1577 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1578 return fs_info->tree_root;
1579 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1580 return fs_info->extent_root;
1581 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1582 return fs_info->chunk_root;
1583 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1584 return fs_info->dev_root;
1585 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1586 return fs_info->csum_root;
1587 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1588 return fs_info->quota_root ? fs_info->quota_root :
1589 ERR_PTR(-ENOENT);
1590 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1591 return fs_info->uuid_root ? fs_info->uuid_root :
1592 ERR_PTR(-ENOENT);
1593again:
1594 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1595 if (root)
1596 return root;
1597
1598 root = btrfs_read_fs_root(fs_info->tree_root, location);
1599 if (IS_ERR(root))
1600 return root;
1601
1602 if (btrfs_root_refs(&root->root_item) == 0) {
1603 ret = -ENOENT;
1604 goto fail;
1605 }
1606
1607 ret = btrfs_init_fs_root(root);
1608 if (ret)
1609 goto fail;
1610
1611 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1612 if (ret < 0)
1613 goto fail;
1614 if (ret == 0)
1615 root->orphan_item_inserted = 1;
1616
1617 ret = btrfs_insert_fs_root(fs_info, root);
1618 if (ret) {
1619 if (ret == -EEXIST) {
1620 free_fs_root(root);
1621 goto again;
1622 }
1623 goto fail;
1624 }
1625 return root;
1626fail:
1627 free_fs_root(root);
1628 return ERR_PTR(ret);
1629}
1630
1631static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1632{
1633 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1634 int ret = 0;
1635 struct btrfs_device *device;
1636 struct backing_dev_info *bdi;
1637
1638 rcu_read_lock();
1639 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1640 if (!device->bdev)
1641 continue;
1642 bdi = blk_get_backing_dev_info(device->bdev);
1643 if (bdi && bdi_congested(bdi, bdi_bits)) {
1644 ret = 1;
1645 break;
1646 }
1647 }
1648 rcu_read_unlock();
1649 return ret;
1650}
1651
1652/*
1653 * If this fails, caller must call bdi_destroy() to get rid of the
1654 * bdi again.
1655 */
1656static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1657{
1658 int err;
1659
1660 bdi->capabilities = BDI_CAP_MAP_COPY;
1661 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1662 if (err)
1663 return err;
1664
1665 bdi->ra_pages = default_backing_dev_info.ra_pages;
1666 bdi->congested_fn = btrfs_congested_fn;
1667 bdi->congested_data = info;
1668 return 0;
1669}
1670
1671/*
1672 * called by the kthread helper functions to finally call the bio end_io
1673 * functions. This is where read checksum verification actually happens
1674 */
1675static void end_workqueue_fn(struct btrfs_work *work)
1676{
1677 struct bio *bio;
1678 struct end_io_wq *end_io_wq;
1679 struct btrfs_fs_info *fs_info;
1680 int error;
1681
1682 end_io_wq = container_of(work, struct end_io_wq, work);
1683 bio = end_io_wq->bio;
1684 fs_info = end_io_wq->info;
1685
1686 error = end_io_wq->error;
1687 bio->bi_private = end_io_wq->private;
1688 bio->bi_end_io = end_io_wq->end_io;
1689 kfree(end_io_wq);
1690 bio_endio(bio, error);
1691}
1692
1693static int cleaner_kthread(void *arg)
1694{
1695 struct btrfs_root *root = arg;
1696 int again;
1697
1698 do {
1699 again = 0;
1700
1701 /* Make the cleaner go to sleep early. */
1702 if (btrfs_need_cleaner_sleep(root))
1703 goto sleep;
1704
1705 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1706 goto sleep;
1707
1708 /*
1709 * Avoid the problem that we change the status of the fs
1710 * during the above check and trylock.
1711 */
1712 if (btrfs_need_cleaner_sleep(root)) {
1713 mutex_unlock(&root->fs_info->cleaner_mutex);
1714 goto sleep;
1715 }
1716
1717 btrfs_run_delayed_iputs(root);
1718 again = btrfs_clean_one_deleted_snapshot(root);
1719 mutex_unlock(&root->fs_info->cleaner_mutex);
1720
1721 /*
1722 * The defragger has dealt with the R/O remount and umount,
1723 * needn't do anything special here.
1724 */
1725 btrfs_run_defrag_inodes(root->fs_info);
1726sleep:
1727 if (!try_to_freeze() && !again) {
1728 set_current_state(TASK_INTERRUPTIBLE);
1729 if (!kthread_should_stop())
1730 schedule();
1731 __set_current_state(TASK_RUNNING);
1732 }
1733 } while (!kthread_should_stop());
1734 return 0;
1735}
1736
1737static int transaction_kthread(void *arg)
1738{
1739 struct btrfs_root *root = arg;
1740 struct btrfs_trans_handle *trans;
1741 struct btrfs_transaction *cur;
1742 u64 transid;
1743 unsigned long now;
1744 unsigned long delay;
1745 bool cannot_commit;
1746
1747 do {
1748 cannot_commit = false;
1749 delay = HZ * root->fs_info->commit_interval;
1750 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1751
1752 spin_lock(&root->fs_info->trans_lock);
1753 cur = root->fs_info->running_transaction;
1754 if (!cur) {
1755 spin_unlock(&root->fs_info->trans_lock);
1756 goto sleep;
1757 }
1758
1759 now = get_seconds();
1760 if (cur->state < TRANS_STATE_BLOCKED &&
1761 (now < cur->start_time ||
1762 now - cur->start_time < root->fs_info->commit_interval)) {
1763 spin_unlock(&root->fs_info->trans_lock);
1764 delay = HZ * 5;
1765 goto sleep;
1766 }
1767 transid = cur->transid;
1768 spin_unlock(&root->fs_info->trans_lock);
1769
1770 /* If the file system is aborted, this will always fail. */
1771 trans = btrfs_attach_transaction(root);
1772 if (IS_ERR(trans)) {
1773 if (PTR_ERR(trans) != -ENOENT)
1774 cannot_commit = true;
1775 goto sleep;
1776 }
1777 if (transid == trans->transid) {
1778 btrfs_commit_transaction(trans, root);
1779 } else {
1780 btrfs_end_transaction(trans, root);
1781 }
1782sleep:
1783 wake_up_process(root->fs_info->cleaner_kthread);
1784 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1785
1786 if (!try_to_freeze()) {
1787 set_current_state(TASK_INTERRUPTIBLE);
1788 if (!kthread_should_stop() &&
1789 (!btrfs_transaction_blocked(root->fs_info) ||
1790 cannot_commit))
1791 schedule_timeout(delay);
1792 __set_current_state(TASK_RUNNING);
1793 }
1794 } while (!kthread_should_stop());
1795 return 0;
1796}
1797
1798/*
1799 * this will find the highest generation in the array of
1800 * root backups. The index of the highest array is returned,
1801 * or -1 if we can't find anything.
1802 *
1803 * We check to make sure the array is valid by comparing the
1804 * generation of the latest root in the array with the generation
1805 * in the super block. If they don't match we pitch it.
1806 */
1807static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1808{
1809 u64 cur;
1810 int newest_index = -1;
1811 struct btrfs_root_backup *root_backup;
1812 int i;
1813
1814 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1815 root_backup = info->super_copy->super_roots + i;
1816 cur = btrfs_backup_tree_root_gen(root_backup);
1817 if (cur == newest_gen)
1818 newest_index = i;
1819 }
1820
1821 /* check to see if we actually wrapped around */
1822 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1823 root_backup = info->super_copy->super_roots;
1824 cur = btrfs_backup_tree_root_gen(root_backup);
1825 if (cur == newest_gen)
1826 newest_index = 0;
1827 }
1828 return newest_index;
1829}
1830
1831
1832/*
1833 * find the oldest backup so we know where to store new entries
1834 * in the backup array. This will set the backup_root_index
1835 * field in the fs_info struct
1836 */
1837static void find_oldest_super_backup(struct btrfs_fs_info *info,
1838 u64 newest_gen)
1839{
1840 int newest_index = -1;
1841
1842 newest_index = find_newest_super_backup(info, newest_gen);
1843 /* if there was garbage in there, just move along */
1844 if (newest_index == -1) {
1845 info->backup_root_index = 0;
1846 } else {
1847 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1848 }
1849}
1850
1851/*
1852 * copy all the root pointers into the super backup array.
1853 * this will bump the backup pointer by one when it is
1854 * done
1855 */
1856static void backup_super_roots(struct btrfs_fs_info *info)
1857{
1858 int next_backup;
1859 struct btrfs_root_backup *root_backup;
1860 int last_backup;
1861
1862 next_backup = info->backup_root_index;
1863 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1864 BTRFS_NUM_BACKUP_ROOTS;
1865
1866 /*
1867 * just overwrite the last backup if we're at the same generation
1868 * this happens only at umount
1869 */
1870 root_backup = info->super_for_commit->super_roots + last_backup;
1871 if (btrfs_backup_tree_root_gen(root_backup) ==
1872 btrfs_header_generation(info->tree_root->node))
1873 next_backup = last_backup;
1874
1875 root_backup = info->super_for_commit->super_roots + next_backup;
1876
1877 /*
1878 * make sure all of our padding and empty slots get zero filled
1879 * regardless of which ones we use today
1880 */
1881 memset(root_backup, 0, sizeof(*root_backup));
1882
1883 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1884
1885 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1886 btrfs_set_backup_tree_root_gen(root_backup,
1887 btrfs_header_generation(info->tree_root->node));
1888
1889 btrfs_set_backup_tree_root_level(root_backup,
1890 btrfs_header_level(info->tree_root->node));
1891
1892 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1893 btrfs_set_backup_chunk_root_gen(root_backup,
1894 btrfs_header_generation(info->chunk_root->node));
1895 btrfs_set_backup_chunk_root_level(root_backup,
1896 btrfs_header_level(info->chunk_root->node));
1897
1898 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1899 btrfs_set_backup_extent_root_gen(root_backup,
1900 btrfs_header_generation(info->extent_root->node));
1901 btrfs_set_backup_extent_root_level(root_backup,
1902 btrfs_header_level(info->extent_root->node));
1903
1904 /*
1905 * we might commit during log recovery, which happens before we set
1906 * the fs_root. Make sure it is valid before we fill it in.
1907 */
1908 if (info->fs_root && info->fs_root->node) {
1909 btrfs_set_backup_fs_root(root_backup,
1910 info->fs_root->node->start);
1911 btrfs_set_backup_fs_root_gen(root_backup,
1912 btrfs_header_generation(info->fs_root->node));
1913 btrfs_set_backup_fs_root_level(root_backup,
1914 btrfs_header_level(info->fs_root->node));
1915 }
1916
1917 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1918 btrfs_set_backup_dev_root_gen(root_backup,
1919 btrfs_header_generation(info->dev_root->node));
1920 btrfs_set_backup_dev_root_level(root_backup,
1921 btrfs_header_level(info->dev_root->node));
1922
1923 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1924 btrfs_set_backup_csum_root_gen(root_backup,
1925 btrfs_header_generation(info->csum_root->node));
1926 btrfs_set_backup_csum_root_level(root_backup,
1927 btrfs_header_level(info->csum_root->node));
1928
1929 btrfs_set_backup_total_bytes(root_backup,
1930 btrfs_super_total_bytes(info->super_copy));
1931 btrfs_set_backup_bytes_used(root_backup,
1932 btrfs_super_bytes_used(info->super_copy));
1933 btrfs_set_backup_num_devices(root_backup,
1934 btrfs_super_num_devices(info->super_copy));
1935
1936 /*
1937 * if we don't copy this out to the super_copy, it won't get remembered
1938 * for the next commit
1939 */
1940 memcpy(&info->super_copy->super_roots,
1941 &info->super_for_commit->super_roots,
1942 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1943}
1944
1945/*
1946 * this copies info out of the root backup array and back into
1947 * the in-memory super block. It is meant to help iterate through
1948 * the array, so you send it the number of backups you've already
1949 * tried and the last backup index you used.
1950 *
1951 * this returns -1 when it has tried all the backups
1952 */
1953static noinline int next_root_backup(struct btrfs_fs_info *info,
1954 struct btrfs_super_block *super,
1955 int *num_backups_tried, int *backup_index)
1956{
1957 struct btrfs_root_backup *root_backup;
1958 int newest = *backup_index;
1959
1960 if (*num_backups_tried == 0) {
1961 u64 gen = btrfs_super_generation(super);
1962
1963 newest = find_newest_super_backup(info, gen);
1964 if (newest == -1)
1965 return -1;
1966
1967 *backup_index = newest;
1968 *num_backups_tried = 1;
1969 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1970 /* we've tried all the backups, all done */
1971 return -1;
1972 } else {
1973 /* jump to the next oldest backup */
1974 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1975 BTRFS_NUM_BACKUP_ROOTS;
1976 *backup_index = newest;
1977 *num_backups_tried += 1;
1978 }
1979 root_backup = super->super_roots + newest;
1980
1981 btrfs_set_super_generation(super,
1982 btrfs_backup_tree_root_gen(root_backup));
1983 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1984 btrfs_set_super_root_level(super,
1985 btrfs_backup_tree_root_level(root_backup));
1986 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1987
1988 /*
1989 * fixme: the total bytes and num_devices need to match or we should
1990 * need a fsck
1991 */
1992 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1993 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1994 return 0;
1995}
1996
1997/* helper to cleanup workers */
1998static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
1999{
2000 btrfs_stop_workers(&fs_info->generic_worker);
2001 btrfs_stop_workers(&fs_info->fixup_workers);
2002 btrfs_stop_workers(&fs_info->delalloc_workers);
2003 btrfs_stop_workers(&fs_info->workers);
2004 btrfs_stop_workers(&fs_info->endio_workers);
2005 btrfs_stop_workers(&fs_info->endio_meta_workers);
2006 btrfs_stop_workers(&fs_info->endio_raid56_workers);
2007 btrfs_stop_workers(&fs_info->rmw_workers);
2008 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2009 btrfs_stop_workers(&fs_info->endio_write_workers);
2010 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2011 btrfs_stop_workers(&fs_info->submit_workers);
2012 btrfs_stop_workers(&fs_info->delayed_workers);
2013 btrfs_stop_workers(&fs_info->caching_workers);
2014 btrfs_stop_workers(&fs_info->readahead_workers);
2015 btrfs_stop_workers(&fs_info->flush_workers);
2016 btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2017}
2018
2019/* helper to cleanup tree roots */
2020static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2021{
2022 free_extent_buffer(info->tree_root->node);
2023 free_extent_buffer(info->tree_root->commit_root);
2024 info->tree_root->node = NULL;
2025 info->tree_root->commit_root = NULL;
2026
2027 if (info->dev_root) {
2028 free_extent_buffer(info->dev_root->node);
2029 free_extent_buffer(info->dev_root->commit_root);
2030 info->dev_root->node = NULL;
2031 info->dev_root->commit_root = NULL;
2032 }
2033 if (info->extent_root) {
2034 free_extent_buffer(info->extent_root->node);
2035 free_extent_buffer(info->extent_root->commit_root);
2036 info->extent_root->node = NULL;
2037 info->extent_root->commit_root = NULL;
2038 }
2039 if (info->csum_root) {
2040 free_extent_buffer(info->csum_root->node);
2041 free_extent_buffer(info->csum_root->commit_root);
2042 info->csum_root->node = NULL;
2043 info->csum_root->commit_root = NULL;
2044 }
2045 if (info->quota_root) {
2046 free_extent_buffer(info->quota_root->node);
2047 free_extent_buffer(info->quota_root->commit_root);
2048 info->quota_root->node = NULL;
2049 info->quota_root->commit_root = NULL;
2050 }
2051 if (info->uuid_root) {
2052 free_extent_buffer(info->uuid_root->node);
2053 free_extent_buffer(info->uuid_root->commit_root);
2054 info->uuid_root->node = NULL;
2055 info->uuid_root->commit_root = NULL;
2056 }
2057 if (chunk_root) {
2058 free_extent_buffer(info->chunk_root->node);
2059 free_extent_buffer(info->chunk_root->commit_root);
2060 info->chunk_root->node = NULL;
2061 info->chunk_root->commit_root = NULL;
2062 }
2063}
2064
2065static void del_fs_roots(struct btrfs_fs_info *fs_info)
2066{
2067 int ret;
2068 struct btrfs_root *gang[8];
2069 int i;
2070
2071 while (!list_empty(&fs_info->dead_roots)) {
2072 gang[0] = list_entry(fs_info->dead_roots.next,
2073 struct btrfs_root, root_list);
2074 list_del(&gang[0]->root_list);
2075
2076 if (gang[0]->in_radix) {
2077 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2078 } else {
2079 free_extent_buffer(gang[0]->node);
2080 free_extent_buffer(gang[0]->commit_root);
2081 btrfs_put_fs_root(gang[0]);
2082 }
2083 }
2084
2085 while (1) {
2086 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2087 (void **)gang, 0,
2088 ARRAY_SIZE(gang));
2089 if (!ret)
2090 break;
2091 for (i = 0; i < ret; i++)
2092 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2093 }
2094}
2095
2096int open_ctree(struct super_block *sb,
2097 struct btrfs_fs_devices *fs_devices,
2098 char *options)
2099{
2100 u32 sectorsize;
2101 u32 nodesize;
2102 u32 leafsize;
2103 u32 blocksize;
2104 u32 stripesize;
2105 u64 generation;
2106 u64 features;
2107 struct btrfs_key location;
2108 struct buffer_head *bh;
2109 struct btrfs_super_block *disk_super;
2110 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2111 struct btrfs_root *tree_root;
2112 struct btrfs_root *extent_root;
2113 struct btrfs_root *csum_root;
2114 struct btrfs_root *chunk_root;
2115 struct btrfs_root *dev_root;
2116 struct btrfs_root *quota_root;
2117 struct btrfs_root *uuid_root;
2118 struct btrfs_root *log_tree_root;
2119 int ret;
2120 int err = -EINVAL;
2121 int num_backups_tried = 0;
2122 int backup_index = 0;
2123 bool create_uuid_tree = false;
2124
2125 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2126 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2127 if (!tree_root || !chunk_root) {
2128 err = -ENOMEM;
2129 goto fail;
2130 }
2131
2132 ret = init_srcu_struct(&fs_info->subvol_srcu);
2133 if (ret) {
2134 err = ret;
2135 goto fail;
2136 }
2137
2138 ret = setup_bdi(fs_info, &fs_info->bdi);
2139 if (ret) {
2140 err = ret;
2141 goto fail_srcu;
2142 }
2143
2144 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2145 if (ret) {
2146 err = ret;
2147 goto fail_bdi;
2148 }
2149 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2150 (1 + ilog2(nr_cpu_ids));
2151
2152 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2153 if (ret) {
2154 err = ret;
2155 goto fail_dirty_metadata_bytes;
2156 }
2157
2158 fs_info->btree_inode = new_inode(sb);
2159 if (!fs_info->btree_inode) {
2160 err = -ENOMEM;
2161 goto fail_delalloc_bytes;
2162 }
2163
2164 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2165
2166 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2167 INIT_LIST_HEAD(&fs_info->trans_list);
2168 INIT_LIST_HEAD(&fs_info->dead_roots);
2169 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2170 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2171 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2172 spin_lock_init(&fs_info->delalloc_root_lock);
2173 spin_lock_init(&fs_info->trans_lock);
2174 spin_lock_init(&fs_info->fs_roots_radix_lock);
2175 spin_lock_init(&fs_info->delayed_iput_lock);
2176 spin_lock_init(&fs_info->defrag_inodes_lock);
2177 spin_lock_init(&fs_info->free_chunk_lock);
2178 spin_lock_init(&fs_info->tree_mod_seq_lock);
2179 spin_lock_init(&fs_info->super_lock);
2180 rwlock_init(&fs_info->tree_mod_log_lock);
2181 mutex_init(&fs_info->reloc_mutex);
2182 seqlock_init(&fs_info->profiles_lock);
2183
2184 init_completion(&fs_info->kobj_unregister);
2185 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2186 INIT_LIST_HEAD(&fs_info->space_info);
2187 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2188 btrfs_mapping_init(&fs_info->mapping_tree);
2189 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2190 BTRFS_BLOCK_RSV_GLOBAL);
2191 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2192 BTRFS_BLOCK_RSV_DELALLOC);
2193 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2194 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2195 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2196 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2197 BTRFS_BLOCK_RSV_DELOPS);
2198 atomic_set(&fs_info->nr_async_submits, 0);
2199 atomic_set(&fs_info->async_delalloc_pages, 0);
2200 atomic_set(&fs_info->async_submit_draining, 0);
2201 atomic_set(&fs_info->nr_async_bios, 0);
2202 atomic_set(&fs_info->defrag_running, 0);
2203 atomic64_set(&fs_info->tree_mod_seq, 0);
2204 fs_info->sb = sb;
2205 fs_info->max_inline = 8192 * 1024;
2206 fs_info->metadata_ratio = 0;
2207 fs_info->defrag_inodes = RB_ROOT;
2208 fs_info->free_chunk_space = 0;
2209 fs_info->tree_mod_log = RB_ROOT;
2210 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2211
2212 /* readahead state */
2213 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2214 spin_lock_init(&fs_info->reada_lock);
2215
2216 fs_info->thread_pool_size = min_t(unsigned long,
2217 num_online_cpus() + 2, 8);
2218
2219 INIT_LIST_HEAD(&fs_info->ordered_roots);
2220 spin_lock_init(&fs_info->ordered_root_lock);
2221 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2222 GFP_NOFS);
2223 if (!fs_info->delayed_root) {
2224 err = -ENOMEM;
2225 goto fail_iput;
2226 }
2227 btrfs_init_delayed_root(fs_info->delayed_root);
2228
2229 mutex_init(&fs_info->scrub_lock);
2230 atomic_set(&fs_info->scrubs_running, 0);
2231 atomic_set(&fs_info->scrub_pause_req, 0);
2232 atomic_set(&fs_info->scrubs_paused, 0);
2233 atomic_set(&fs_info->scrub_cancel_req, 0);
2234 init_waitqueue_head(&fs_info->scrub_pause_wait);
2235 init_rwsem(&fs_info->scrub_super_lock);
2236 fs_info->scrub_workers_refcnt = 0;
2237#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2238 fs_info->check_integrity_print_mask = 0;
2239#endif
2240
2241 spin_lock_init(&fs_info->balance_lock);
2242 mutex_init(&fs_info->balance_mutex);
2243 atomic_set(&fs_info->balance_running, 0);
2244 atomic_set(&fs_info->balance_pause_req, 0);
2245 atomic_set(&fs_info->balance_cancel_req, 0);
2246 fs_info->balance_ctl = NULL;
2247 init_waitqueue_head(&fs_info->balance_wait_q);
2248
2249 sb->s_blocksize = 4096;
2250 sb->s_blocksize_bits = blksize_bits(4096);
2251 sb->s_bdi = &fs_info->bdi;
2252
2253 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2254 set_nlink(fs_info->btree_inode, 1);
2255 /*
2256 * we set the i_size on the btree inode to the max possible int.
2257 * the real end of the address space is determined by all of
2258 * the devices in the system
2259 */
2260 fs_info->btree_inode->i_size = OFFSET_MAX;
2261 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2262 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2263
2264 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2265 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2266 fs_info->btree_inode->i_mapping);
2267 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2268 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2269
2270 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2271
2272 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2273 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2274 sizeof(struct btrfs_key));
2275 set_bit(BTRFS_INODE_DUMMY,
2276 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2277 insert_inode_hash(fs_info->btree_inode);
2278
2279 spin_lock_init(&fs_info->block_group_cache_lock);
2280 fs_info->block_group_cache_tree = RB_ROOT;
2281 fs_info->first_logical_byte = (u64)-1;
2282
2283 extent_io_tree_init(&fs_info->freed_extents[0],
2284 fs_info->btree_inode->i_mapping);
2285 extent_io_tree_init(&fs_info->freed_extents[1],
2286 fs_info->btree_inode->i_mapping);
2287 fs_info->pinned_extents = &fs_info->freed_extents[0];
2288 fs_info->do_barriers = 1;
2289
2290
2291 mutex_init(&fs_info->ordered_operations_mutex);
2292 mutex_init(&fs_info->ordered_extent_flush_mutex);
2293 mutex_init(&fs_info->tree_log_mutex);
2294 mutex_init(&fs_info->chunk_mutex);
2295 mutex_init(&fs_info->transaction_kthread_mutex);
2296 mutex_init(&fs_info->cleaner_mutex);
2297 mutex_init(&fs_info->volume_mutex);
2298 init_rwsem(&fs_info->extent_commit_sem);
2299 init_rwsem(&fs_info->cleanup_work_sem);
2300 init_rwsem(&fs_info->subvol_sem);
2301 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2302 fs_info->dev_replace.lock_owner = 0;
2303 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2304 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2305 mutex_init(&fs_info->dev_replace.lock_management_lock);
2306 mutex_init(&fs_info->dev_replace.lock);
2307
2308 spin_lock_init(&fs_info->qgroup_lock);
2309 mutex_init(&fs_info->qgroup_ioctl_lock);
2310 fs_info->qgroup_tree = RB_ROOT;
2311 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2312 fs_info->qgroup_seq = 1;
2313 fs_info->quota_enabled = 0;
2314 fs_info->pending_quota_state = 0;
2315 fs_info->qgroup_ulist = NULL;
2316 mutex_init(&fs_info->qgroup_rescan_lock);
2317
2318 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2319 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2320
2321 init_waitqueue_head(&fs_info->transaction_throttle);
2322 init_waitqueue_head(&fs_info->transaction_wait);
2323 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2324 init_waitqueue_head(&fs_info->async_submit_wait);
2325
2326 ret = btrfs_alloc_stripe_hash_table(fs_info);
2327 if (ret) {
2328 err = ret;
2329 goto fail_alloc;
2330 }
2331
2332 __setup_root(4096, 4096, 4096, 4096, tree_root,
2333 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2334
2335 invalidate_bdev(fs_devices->latest_bdev);
2336
2337 /*
2338 * Read super block and check the signature bytes only
2339 */
2340 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2341 if (!bh) {
2342 err = -EINVAL;
2343 goto fail_alloc;
2344 }
2345
2346 /*
2347 * We want to check superblock checksum, the type is stored inside.
2348 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2349 */
2350 if (btrfs_check_super_csum(bh->b_data)) {
2351 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2352 err = -EINVAL;
2353 goto fail_alloc;
2354 }
2355
2356 /*
2357 * super_copy is zeroed at allocation time and we never touch the
2358 * following bytes up to INFO_SIZE, the checksum is calculated from
2359 * the whole block of INFO_SIZE
2360 */
2361 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2362 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2363 sizeof(*fs_info->super_for_commit));
2364 brelse(bh);
2365
2366 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2367
2368 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2369 if (ret) {
2370 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2371 err = -EINVAL;
2372 goto fail_alloc;
2373 }
2374
2375 disk_super = fs_info->super_copy;
2376 if (!btrfs_super_root(disk_super))
2377 goto fail_alloc;
2378
2379 /* check FS state, whether FS is broken. */
2380 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2381 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2382
2383 /*
2384 * run through our array of backup supers and setup
2385 * our ring pointer to the oldest one
2386 */
2387 generation = btrfs_super_generation(disk_super);
2388 find_oldest_super_backup(fs_info, generation);
2389
2390 /*
2391 * In the long term, we'll store the compression type in the super
2392 * block, and it'll be used for per file compression control.
2393 */
2394 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2395
2396 ret = btrfs_parse_options(tree_root, options);
2397 if (ret) {
2398 err = ret;
2399 goto fail_alloc;
2400 }
2401
2402 features = btrfs_super_incompat_flags(disk_super) &
2403 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2404 if (features) {
2405 printk(KERN_ERR "BTRFS: couldn't mount because of "
2406 "unsupported optional features (%Lx).\n",
2407 (unsigned long long)features);
2408 err = -EINVAL;
2409 goto fail_alloc;
2410 }
2411
2412 if (btrfs_super_leafsize(disk_super) !=
2413 btrfs_super_nodesize(disk_super)) {
2414 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2415 "blocksizes don't match. node %d leaf %d\n",
2416 btrfs_super_nodesize(disk_super),
2417 btrfs_super_leafsize(disk_super));
2418 err = -EINVAL;
2419 goto fail_alloc;
2420 }
2421 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2422 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2423 "blocksize (%d) was too large\n",
2424 btrfs_super_leafsize(disk_super));
2425 err = -EINVAL;
2426 goto fail_alloc;
2427 }
2428
2429 features = btrfs_super_incompat_flags(disk_super);
2430 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2431 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2432 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2433
2434 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2435 printk(KERN_ERR "btrfs: has skinny extents\n");
2436
2437 /*
2438 * flag our filesystem as having big metadata blocks if
2439 * they are bigger than the page size
2440 */
2441 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2442 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2443 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2444 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2445 }
2446
2447 nodesize = btrfs_super_nodesize(disk_super);
2448 leafsize = btrfs_super_leafsize(disk_super);
2449 sectorsize = btrfs_super_sectorsize(disk_super);
2450 stripesize = btrfs_super_stripesize(disk_super);
2451 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2452 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2453
2454 /*
2455 * mixed block groups end up with duplicate but slightly offset
2456 * extent buffers for the same range. It leads to corruptions
2457 */
2458 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2459 (sectorsize != leafsize)) {
2460 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2461 "are not allowed for mixed block groups on %s\n",
2462 sb->s_id);
2463 goto fail_alloc;
2464 }
2465
2466 /*
2467 * Needn't use the lock because there is no other task which will
2468 * update the flag.
2469 */
2470 btrfs_set_super_incompat_flags(disk_super, features);
2471
2472 features = btrfs_super_compat_ro_flags(disk_super) &
2473 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2474 if (!(sb->s_flags & MS_RDONLY) && features) {
2475 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2476 "unsupported option features (%Lx).\n",
2477 (unsigned long long)features);
2478 err = -EINVAL;
2479 goto fail_alloc;
2480 }
2481
2482 btrfs_init_workers(&fs_info->generic_worker,
2483 "genwork", 1, NULL);
2484
2485 btrfs_init_workers(&fs_info->workers, "worker",
2486 fs_info->thread_pool_size,
2487 &fs_info->generic_worker);
2488
2489 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2490 fs_info->thread_pool_size,
2491 &fs_info->generic_worker);
2492
2493 btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2494 fs_info->thread_pool_size,
2495 &fs_info->generic_worker);
2496
2497 btrfs_init_workers(&fs_info->submit_workers, "submit",
2498 min_t(u64, fs_devices->num_devices,
2499 fs_info->thread_pool_size),
2500 &fs_info->generic_worker);
2501
2502 btrfs_init_workers(&fs_info->caching_workers, "cache",
2503 2, &fs_info->generic_worker);
2504
2505 /* a higher idle thresh on the submit workers makes it much more
2506 * likely that bios will be send down in a sane order to the
2507 * devices
2508 */
2509 fs_info->submit_workers.idle_thresh = 64;
2510
2511 fs_info->workers.idle_thresh = 16;
2512 fs_info->workers.ordered = 1;
2513
2514 fs_info->delalloc_workers.idle_thresh = 2;
2515 fs_info->delalloc_workers.ordered = 1;
2516
2517 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2518 &fs_info->generic_worker);
2519 btrfs_init_workers(&fs_info->endio_workers, "endio",
2520 fs_info->thread_pool_size,
2521 &fs_info->generic_worker);
2522 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2523 fs_info->thread_pool_size,
2524 &fs_info->generic_worker);
2525 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2526 "endio-meta-write", fs_info->thread_pool_size,
2527 &fs_info->generic_worker);
2528 btrfs_init_workers(&fs_info->endio_raid56_workers,
2529 "endio-raid56", fs_info->thread_pool_size,
2530 &fs_info->generic_worker);
2531 btrfs_init_workers(&fs_info->rmw_workers,
2532 "rmw", fs_info->thread_pool_size,
2533 &fs_info->generic_worker);
2534 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2535 fs_info->thread_pool_size,
2536 &fs_info->generic_worker);
2537 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2538 1, &fs_info->generic_worker);
2539 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2540 fs_info->thread_pool_size,
2541 &fs_info->generic_worker);
2542 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2543 fs_info->thread_pool_size,
2544 &fs_info->generic_worker);
2545 btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2546 &fs_info->generic_worker);
2547
2548 /*
2549 * endios are largely parallel and should have a very
2550 * low idle thresh
2551 */
2552 fs_info->endio_workers.idle_thresh = 4;
2553 fs_info->endio_meta_workers.idle_thresh = 4;
2554 fs_info->endio_raid56_workers.idle_thresh = 4;
2555 fs_info->rmw_workers.idle_thresh = 2;
2556
2557 fs_info->endio_write_workers.idle_thresh = 2;
2558 fs_info->endio_meta_write_workers.idle_thresh = 2;
2559 fs_info->readahead_workers.idle_thresh = 2;
2560
2561 /*
2562 * btrfs_start_workers can really only fail because of ENOMEM so just
2563 * return -ENOMEM if any of these fail.
2564 */
2565 ret = btrfs_start_workers(&fs_info->workers);
2566 ret |= btrfs_start_workers(&fs_info->generic_worker);
2567 ret |= btrfs_start_workers(&fs_info->submit_workers);
2568 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2569 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2570 ret |= btrfs_start_workers(&fs_info->endio_workers);
2571 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2572 ret |= btrfs_start_workers(&fs_info->rmw_workers);
2573 ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2574 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2575 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2576 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2577 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2578 ret |= btrfs_start_workers(&fs_info->caching_workers);
2579 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2580 ret |= btrfs_start_workers(&fs_info->flush_workers);
2581 ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2582 if (ret) {
2583 err = -ENOMEM;
2584 goto fail_sb_buffer;
2585 }
2586
2587 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2588 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2589 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2590
2591 tree_root->nodesize = nodesize;
2592 tree_root->leafsize = leafsize;
2593 tree_root->sectorsize = sectorsize;
2594 tree_root->stripesize = stripesize;
2595
2596 sb->s_blocksize = sectorsize;
2597 sb->s_blocksize_bits = blksize_bits(sectorsize);
2598
2599 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2600 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2601 goto fail_sb_buffer;
2602 }
2603
2604 if (sectorsize != PAGE_SIZE) {
2605 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2606 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2607 goto fail_sb_buffer;
2608 }
2609
2610 mutex_lock(&fs_info->chunk_mutex);
2611 ret = btrfs_read_sys_array(tree_root);
2612 mutex_unlock(&fs_info->chunk_mutex);
2613 if (ret) {
2614 printk(KERN_WARNING "btrfs: failed to read the system "
2615 "array on %s\n", sb->s_id);
2616 goto fail_sb_buffer;
2617 }
2618
2619 blocksize = btrfs_level_size(tree_root,
2620 btrfs_super_chunk_root_level(disk_super));
2621 generation = btrfs_super_chunk_root_generation(disk_super);
2622
2623 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2624 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2625
2626 chunk_root->node = read_tree_block(chunk_root,
2627 btrfs_super_chunk_root(disk_super),
2628 blocksize, generation);
2629 if (!chunk_root->node ||
2630 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2631 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2632 sb->s_id);
2633 goto fail_tree_roots;
2634 }
2635 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2636 chunk_root->commit_root = btrfs_root_node(chunk_root);
2637
2638 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2639 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2640 BTRFS_UUID_SIZE);
2641
2642 ret = btrfs_read_chunk_tree(chunk_root);
2643 if (ret) {
2644 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2645 sb->s_id);
2646 goto fail_tree_roots;
2647 }
2648
2649 /*
2650 * keep the device that is marked to be the target device for the
2651 * dev_replace procedure
2652 */
2653 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2654
2655 if (!fs_devices->latest_bdev) {
2656 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2657 sb->s_id);
2658 goto fail_tree_roots;
2659 }
2660
2661retry_root_backup:
2662 blocksize = btrfs_level_size(tree_root,
2663 btrfs_super_root_level(disk_super));
2664 generation = btrfs_super_generation(disk_super);
2665
2666 tree_root->node = read_tree_block(tree_root,
2667 btrfs_super_root(disk_super),
2668 blocksize, generation);
2669 if (!tree_root->node ||
2670 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2671 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2672 sb->s_id);
2673
2674 goto recovery_tree_root;
2675 }
2676
2677 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2678 tree_root->commit_root = btrfs_root_node(tree_root);
2679
2680 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2681 location.type = BTRFS_ROOT_ITEM_KEY;
2682 location.offset = 0;
2683
2684 extent_root = btrfs_read_tree_root(tree_root, &location);
2685 if (IS_ERR(extent_root)) {
2686 ret = PTR_ERR(extent_root);
2687 goto recovery_tree_root;
2688 }
2689 extent_root->track_dirty = 1;
2690 fs_info->extent_root = extent_root;
2691
2692 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2693 dev_root = btrfs_read_tree_root(tree_root, &location);
2694 if (IS_ERR(dev_root)) {
2695 ret = PTR_ERR(dev_root);
2696 goto recovery_tree_root;
2697 }
2698 dev_root->track_dirty = 1;
2699 fs_info->dev_root = dev_root;
2700 btrfs_init_devices_late(fs_info);
2701
2702 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2703 csum_root = btrfs_read_tree_root(tree_root, &location);
2704 if (IS_ERR(csum_root)) {
2705 ret = PTR_ERR(csum_root);
2706 goto recovery_tree_root;
2707 }
2708 csum_root->track_dirty = 1;
2709 fs_info->csum_root = csum_root;
2710
2711 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2712 quota_root = btrfs_read_tree_root(tree_root, &location);
2713 if (!IS_ERR(quota_root)) {
2714 quota_root->track_dirty = 1;
2715 fs_info->quota_enabled = 1;
2716 fs_info->pending_quota_state = 1;
2717 fs_info->quota_root = quota_root;
2718 }
2719
2720 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2721 uuid_root = btrfs_read_tree_root(tree_root, &location);
2722 if (IS_ERR(uuid_root)) {
2723 ret = PTR_ERR(uuid_root);
2724 if (ret != -ENOENT)
2725 goto recovery_tree_root;
2726 create_uuid_tree = true;
2727 } else {
2728 uuid_root->track_dirty = 1;
2729 fs_info->uuid_root = uuid_root;
2730 }
2731
2732 fs_info->generation = generation;
2733 fs_info->last_trans_committed = generation;
2734
2735 ret = btrfs_recover_balance(fs_info);
2736 if (ret) {
2737 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2738 goto fail_block_groups;
2739 }
2740
2741 ret = btrfs_init_dev_stats(fs_info);
2742 if (ret) {
2743 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2744 ret);
2745 goto fail_block_groups;
2746 }
2747
2748 ret = btrfs_init_dev_replace(fs_info);
2749 if (ret) {
2750 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2751 goto fail_block_groups;
2752 }
2753
2754 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2755
2756 ret = btrfs_init_space_info(fs_info);
2757 if (ret) {
2758 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2759 goto fail_block_groups;
2760 }
2761
2762 ret = btrfs_read_block_groups(extent_root);
2763 if (ret) {
2764 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2765 goto fail_block_groups;
2766 }
2767 fs_info->num_tolerated_disk_barrier_failures =
2768 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2769 if (fs_info->fs_devices->missing_devices >
2770 fs_info->num_tolerated_disk_barrier_failures &&
2771 !(sb->s_flags & MS_RDONLY)) {
2772 printk(KERN_WARNING
2773 "Btrfs: too many missing devices, writeable mount is not allowed\n");
2774 goto fail_block_groups;
2775 }
2776
2777 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2778 "btrfs-cleaner");
2779 if (IS_ERR(fs_info->cleaner_kthread))
2780 goto fail_block_groups;
2781
2782 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2783 tree_root,
2784 "btrfs-transaction");
2785 if (IS_ERR(fs_info->transaction_kthread))
2786 goto fail_cleaner;
2787
2788 if (!btrfs_test_opt(tree_root, SSD) &&
2789 !btrfs_test_opt(tree_root, NOSSD) &&
2790 !fs_info->fs_devices->rotating) {
2791 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2792 "mode\n");
2793 btrfs_set_opt(fs_info->mount_opt, SSD);
2794 }
2795
2796#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2797 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2798 ret = btrfsic_mount(tree_root, fs_devices,
2799 btrfs_test_opt(tree_root,
2800 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2801 1 : 0,
2802 fs_info->check_integrity_print_mask);
2803 if (ret)
2804 printk(KERN_WARNING "btrfs: failed to initialize"
2805 " integrity check module %s\n", sb->s_id);
2806 }
2807#endif
2808 ret = btrfs_read_qgroup_config(fs_info);
2809 if (ret)
2810 goto fail_trans_kthread;
2811
2812 /* do not make disk changes in broken FS */
2813 if (btrfs_super_log_root(disk_super) != 0) {
2814 u64 bytenr = btrfs_super_log_root(disk_super);
2815
2816 if (fs_devices->rw_devices == 0) {
2817 printk(KERN_WARNING "Btrfs log replay required "
2818 "on RO media\n");
2819 err = -EIO;
2820 goto fail_qgroup;
2821 }
2822 blocksize =
2823 btrfs_level_size(tree_root,
2824 btrfs_super_log_root_level(disk_super));
2825
2826 log_tree_root = btrfs_alloc_root(fs_info);
2827 if (!log_tree_root) {
2828 err = -ENOMEM;
2829 goto fail_qgroup;
2830 }
2831
2832 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2833 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2834
2835 log_tree_root->node = read_tree_block(tree_root, bytenr,
2836 blocksize,
2837 generation + 1);
2838 if (!log_tree_root->node ||
2839 !extent_buffer_uptodate(log_tree_root->node)) {
2840 printk(KERN_ERR "btrfs: failed to read log tree\n");
2841 free_extent_buffer(log_tree_root->node);
2842 kfree(log_tree_root);
2843 goto fail_trans_kthread;
2844 }
2845 /* returns with log_tree_root freed on success */
2846 ret = btrfs_recover_log_trees(log_tree_root);
2847 if (ret) {
2848 btrfs_error(tree_root->fs_info, ret,
2849 "Failed to recover log tree");
2850 free_extent_buffer(log_tree_root->node);
2851 kfree(log_tree_root);
2852 goto fail_trans_kthread;
2853 }
2854
2855 if (sb->s_flags & MS_RDONLY) {
2856 ret = btrfs_commit_super(tree_root);
2857 if (ret)
2858 goto fail_trans_kthread;
2859 }
2860 }
2861
2862 ret = btrfs_find_orphan_roots(tree_root);
2863 if (ret)
2864 goto fail_trans_kthread;
2865
2866 if (!(sb->s_flags & MS_RDONLY)) {
2867 ret = btrfs_cleanup_fs_roots(fs_info);
2868 if (ret)
2869 goto fail_trans_kthread;
2870
2871 ret = btrfs_recover_relocation(tree_root);
2872 if (ret < 0) {
2873 printk(KERN_WARNING
2874 "btrfs: failed to recover relocation\n");
2875 err = -EINVAL;
2876 goto fail_qgroup;
2877 }
2878 }
2879
2880 location.objectid = BTRFS_FS_TREE_OBJECTID;
2881 location.type = BTRFS_ROOT_ITEM_KEY;
2882 location.offset = 0;
2883
2884 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2885 if (IS_ERR(fs_info->fs_root)) {
2886 err = PTR_ERR(fs_info->fs_root);
2887 goto fail_qgroup;
2888 }
2889
2890 if (sb->s_flags & MS_RDONLY)
2891 return 0;
2892
2893 down_read(&fs_info->cleanup_work_sem);
2894 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2895 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2896 up_read(&fs_info->cleanup_work_sem);
2897 close_ctree(tree_root);
2898 return ret;
2899 }
2900 up_read(&fs_info->cleanup_work_sem);
2901
2902 ret = btrfs_resume_balance_async(fs_info);
2903 if (ret) {
2904 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2905 close_ctree(tree_root);
2906 return ret;
2907 }
2908
2909 ret = btrfs_resume_dev_replace_async(fs_info);
2910 if (ret) {
2911 pr_warn("btrfs: failed to resume dev_replace\n");
2912 close_ctree(tree_root);
2913 return ret;
2914 }
2915
2916 btrfs_qgroup_rescan_resume(fs_info);
2917
2918 if (create_uuid_tree) {
2919 pr_info("btrfs: creating UUID tree\n");
2920 ret = btrfs_create_uuid_tree(fs_info);
2921 if (ret) {
2922 pr_warn("btrfs: failed to create the UUID tree %d\n",
2923 ret);
2924 close_ctree(tree_root);
2925 return ret;
2926 }
2927 }
2928
2929 return 0;
2930
2931fail_qgroup:
2932 btrfs_free_qgroup_config(fs_info);
2933fail_trans_kthread:
2934 kthread_stop(fs_info->transaction_kthread);
2935 btrfs_cleanup_transaction(fs_info->tree_root);
2936 del_fs_roots(fs_info);
2937fail_cleaner:
2938 kthread_stop(fs_info->cleaner_kthread);
2939
2940 /*
2941 * make sure we're done with the btree inode before we stop our
2942 * kthreads
2943 */
2944 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2945
2946fail_block_groups:
2947 btrfs_put_block_group_cache(fs_info);
2948 btrfs_free_block_groups(fs_info);
2949
2950fail_tree_roots:
2951 free_root_pointers(fs_info, 1);
2952 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2953
2954fail_sb_buffer:
2955 btrfs_stop_all_workers(fs_info);
2956fail_alloc:
2957fail_iput:
2958 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2959
2960 iput(fs_info->btree_inode);
2961fail_delalloc_bytes:
2962 percpu_counter_destroy(&fs_info->delalloc_bytes);
2963fail_dirty_metadata_bytes:
2964 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2965fail_bdi:
2966 bdi_destroy(&fs_info->bdi);
2967fail_srcu:
2968 cleanup_srcu_struct(&fs_info->subvol_srcu);
2969fail:
2970 btrfs_free_stripe_hash_table(fs_info);
2971 btrfs_close_devices(fs_info->fs_devices);
2972 return err;
2973
2974recovery_tree_root:
2975 if (!btrfs_test_opt(tree_root, RECOVERY))
2976 goto fail_tree_roots;
2977
2978 free_root_pointers(fs_info, 0);
2979
2980 /* don't use the log in recovery mode, it won't be valid */
2981 btrfs_set_super_log_root(disk_super, 0);
2982
2983 /* we can't trust the free space cache either */
2984 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2985
2986 ret = next_root_backup(fs_info, fs_info->super_copy,
2987 &num_backups_tried, &backup_index);
2988 if (ret == -1)
2989 goto fail_block_groups;
2990 goto retry_root_backup;
2991}
2992
2993static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2994{
2995 if (uptodate) {
2996 set_buffer_uptodate(bh);
2997 } else {
2998 struct btrfs_device *device = (struct btrfs_device *)
2999 bh->b_private;
3000
3001 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3002 "I/O error on %s\n",
3003 rcu_str_deref(device->name));
3004 /* note, we dont' set_buffer_write_io_error because we have
3005 * our own ways of dealing with the IO errors
3006 */
3007 clear_buffer_uptodate(bh);
3008 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3009 }
3010 unlock_buffer(bh);
3011 put_bh(bh);
3012}
3013
3014struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3015{
3016 struct buffer_head *bh;
3017 struct buffer_head *latest = NULL;
3018 struct btrfs_super_block *super;
3019 int i;
3020 u64 transid = 0;
3021 u64 bytenr;
3022
3023 /* we would like to check all the supers, but that would make
3024 * a btrfs mount succeed after a mkfs from a different FS.
3025 * So, we need to add a special mount option to scan for
3026 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3027 */
3028 for (i = 0; i < 1; i++) {
3029 bytenr = btrfs_sb_offset(i);
3030 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3031 i_size_read(bdev->bd_inode))
3032 break;
3033 bh = __bread(bdev, bytenr / 4096,
3034 BTRFS_SUPER_INFO_SIZE);
3035 if (!bh)
3036 continue;
3037
3038 super = (struct btrfs_super_block *)bh->b_data;
3039 if (btrfs_super_bytenr(super) != bytenr ||
3040 btrfs_super_magic(super) != BTRFS_MAGIC) {
3041 brelse(bh);
3042 continue;
3043 }
3044
3045 if (!latest || btrfs_super_generation(super) > transid) {
3046 brelse(latest);
3047 latest = bh;
3048 transid = btrfs_super_generation(super);
3049 } else {
3050 brelse(bh);
3051 }
3052 }
3053 return latest;
3054}
3055
3056/*
3057 * this should be called twice, once with wait == 0 and
3058 * once with wait == 1. When wait == 0 is done, all the buffer heads
3059 * we write are pinned.
3060 *
3061 * They are released when wait == 1 is done.
3062 * max_mirrors must be the same for both runs, and it indicates how
3063 * many supers on this one device should be written.
3064 *
3065 * max_mirrors == 0 means to write them all.
3066 */
3067static int write_dev_supers(struct btrfs_device *device,
3068 struct btrfs_super_block *sb,
3069 int do_barriers, int wait, int max_mirrors)
3070{
3071 struct buffer_head *bh;
3072 int i;
3073 int ret;
3074 int errors = 0;
3075 u32 crc;
3076 u64 bytenr;
3077
3078 if (max_mirrors == 0)
3079 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3080
3081 for (i = 0; i < max_mirrors; i++) {
3082 bytenr = btrfs_sb_offset(i);
3083 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3084 break;
3085
3086 if (wait) {
3087 bh = __find_get_block(device->bdev, bytenr / 4096,
3088 BTRFS_SUPER_INFO_SIZE);
3089 if (!bh) {
3090 errors++;
3091 continue;
3092 }
3093 wait_on_buffer(bh);
3094 if (!buffer_uptodate(bh))
3095 errors++;
3096
3097 /* drop our reference */
3098 brelse(bh);
3099
3100 /* drop the reference from the wait == 0 run */
3101 brelse(bh);
3102 continue;
3103 } else {
3104 btrfs_set_super_bytenr(sb, bytenr);
3105
3106 crc = ~(u32)0;
3107 crc = btrfs_csum_data((char *)sb +
3108 BTRFS_CSUM_SIZE, crc,
3109 BTRFS_SUPER_INFO_SIZE -
3110 BTRFS_CSUM_SIZE);
3111 btrfs_csum_final(crc, sb->csum);
3112
3113 /*
3114 * one reference for us, and we leave it for the
3115 * caller
3116 */
3117 bh = __getblk(device->bdev, bytenr / 4096,
3118 BTRFS_SUPER_INFO_SIZE);
3119 if (!bh) {
3120 printk(KERN_ERR "btrfs: couldn't get super "
3121 "buffer head for bytenr %Lu\n", bytenr);
3122 errors++;
3123 continue;
3124 }
3125
3126 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3127
3128 /* one reference for submit_bh */
3129 get_bh(bh);
3130
3131 set_buffer_uptodate(bh);
3132 lock_buffer(bh);
3133 bh->b_end_io = btrfs_end_buffer_write_sync;
3134 bh->b_private = device;
3135 }
3136
3137 /*
3138 * we fua the first super. The others we allow
3139 * to go down lazy.
3140 */
3141 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3142 if (ret)
3143 errors++;
3144 }
3145 return errors < i ? 0 : -1;
3146}
3147
3148/*
3149 * endio for the write_dev_flush, this will wake anyone waiting
3150 * for the barrier when it is done
3151 */
3152static void btrfs_end_empty_barrier(struct bio *bio, int err)
3153{
3154 if (err) {
3155 if (err == -EOPNOTSUPP)
3156 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3157 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3158 }
3159 if (bio->bi_private)
3160 complete(bio->bi_private);
3161 bio_put(bio);
3162}
3163
3164/*
3165 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3166 * sent down. With wait == 1, it waits for the previous flush.
3167 *
3168 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3169 * capable
3170 */
3171static int write_dev_flush(struct btrfs_device *device, int wait)
3172{
3173 struct bio *bio;
3174 int ret = 0;
3175
3176 if (device->nobarriers)
3177 return 0;
3178
3179 if (wait) {
3180 bio = device->flush_bio;
3181 if (!bio)
3182 return 0;
3183
3184 wait_for_completion(&device->flush_wait);
3185
3186 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3187 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3188 rcu_str_deref(device->name));
3189 device->nobarriers = 1;
3190 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3191 ret = -EIO;
3192 btrfs_dev_stat_inc_and_print(device,
3193 BTRFS_DEV_STAT_FLUSH_ERRS);
3194 }
3195
3196 /* drop the reference from the wait == 0 run */
3197 bio_put(bio);
3198 device->flush_bio = NULL;
3199
3200 return ret;
3201 }
3202
3203 /*
3204 * one reference for us, and we leave it for the
3205 * caller
3206 */
3207 device->flush_bio = NULL;
3208 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3209 if (!bio)
3210 return -ENOMEM;
3211
3212 bio->bi_end_io = btrfs_end_empty_barrier;
3213 bio->bi_bdev = device->bdev;
3214 init_completion(&device->flush_wait);
3215 bio->bi_private = &device->flush_wait;
3216 device->flush_bio = bio;
3217
3218 bio_get(bio);
3219 btrfsic_submit_bio(WRITE_FLUSH, bio);
3220
3221 return 0;
3222}
3223
3224/*
3225 * send an empty flush down to each device in parallel,
3226 * then wait for them
3227 */
3228static int barrier_all_devices(struct btrfs_fs_info *info)
3229{
3230 struct list_head *head;
3231 struct btrfs_device *dev;
3232 int errors_send = 0;
3233 int errors_wait = 0;
3234 int ret;
3235
3236 /* send down all the barriers */
3237 head = &info->fs_devices->devices;
3238 list_for_each_entry_rcu(dev, head, dev_list) {
3239 if (!dev->bdev) {
3240 errors_send++;
3241 continue;
3242 }
3243 if (!dev->in_fs_metadata || !dev->writeable)
3244 continue;
3245
3246 ret = write_dev_flush(dev, 0);
3247 if (ret)
3248 errors_send++;
3249 }
3250
3251 /* wait for all the barriers */
3252 list_for_each_entry_rcu(dev, head, dev_list) {
3253 if (!dev->bdev) {
3254 errors_wait++;
3255 continue;
3256 }
3257 if (!dev->in_fs_metadata || !dev->writeable)
3258 continue;
3259
3260 ret = write_dev_flush(dev, 1);
3261 if (ret)
3262 errors_wait++;
3263 }
3264 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3265 errors_wait > info->num_tolerated_disk_barrier_failures)
3266 return -EIO;
3267 return 0;
3268}
3269
3270int btrfs_calc_num_tolerated_disk_barrier_failures(
3271 struct btrfs_fs_info *fs_info)
3272{
3273 struct btrfs_ioctl_space_info space;
3274 struct btrfs_space_info *sinfo;
3275 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3276 BTRFS_BLOCK_GROUP_SYSTEM,
3277 BTRFS_BLOCK_GROUP_METADATA,
3278 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3279 int num_types = 4;
3280 int i;
3281 int c;
3282 int num_tolerated_disk_barrier_failures =
3283 (int)fs_info->fs_devices->num_devices;
3284
3285 for (i = 0; i < num_types; i++) {
3286 struct btrfs_space_info *tmp;
3287
3288 sinfo = NULL;
3289 rcu_read_lock();
3290 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3291 if (tmp->flags == types[i]) {
3292 sinfo = tmp;
3293 break;
3294 }
3295 }
3296 rcu_read_unlock();
3297
3298 if (!sinfo)
3299 continue;
3300
3301 down_read(&sinfo->groups_sem);
3302 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3303 if (!list_empty(&sinfo->block_groups[c])) {
3304 u64 flags;
3305
3306 btrfs_get_block_group_info(
3307 &sinfo->block_groups[c], &space);
3308 if (space.total_bytes == 0 ||
3309 space.used_bytes == 0)
3310 continue;
3311 flags = space.flags;
3312 /*
3313 * return
3314 * 0: if dup, single or RAID0 is configured for
3315 * any of metadata, system or data, else
3316 * 1: if RAID5 is configured, or if RAID1 or
3317 * RAID10 is configured and only two mirrors
3318 * are used, else
3319 * 2: if RAID6 is configured, else
3320 * num_mirrors - 1: if RAID1 or RAID10 is
3321 * configured and more than
3322 * 2 mirrors are used.
3323 */
3324 if (num_tolerated_disk_barrier_failures > 0 &&
3325 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3326 BTRFS_BLOCK_GROUP_RAID0)) ||
3327 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3328 == 0)))
3329 num_tolerated_disk_barrier_failures = 0;
3330 else if (num_tolerated_disk_barrier_failures > 1) {
3331 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3332 BTRFS_BLOCK_GROUP_RAID5 |
3333 BTRFS_BLOCK_GROUP_RAID10)) {
3334 num_tolerated_disk_barrier_failures = 1;
3335 } else if (flags &
3336 BTRFS_BLOCK_GROUP_RAID6) {
3337 num_tolerated_disk_barrier_failures = 2;
3338 }
3339 }
3340 }
3341 }
3342 up_read(&sinfo->groups_sem);
3343 }
3344
3345 return num_tolerated_disk_barrier_failures;
3346}
3347
3348static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3349{
3350 struct list_head *head;
3351 struct btrfs_device *dev;
3352 struct btrfs_super_block *sb;
3353 struct btrfs_dev_item *dev_item;
3354 int ret;
3355 int do_barriers;
3356 int max_errors;
3357 int total_errors = 0;
3358 u64 flags;
3359
3360 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3361 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3362 backup_super_roots(root->fs_info);
3363
3364 sb = root->fs_info->super_for_commit;
3365 dev_item = &sb->dev_item;
3366
3367 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3368 head = &root->fs_info->fs_devices->devices;
3369
3370 if (do_barriers) {
3371 ret = barrier_all_devices(root->fs_info);
3372 if (ret) {
3373 mutex_unlock(
3374 &root->fs_info->fs_devices->device_list_mutex);
3375 btrfs_error(root->fs_info, ret,
3376 "errors while submitting device barriers.");
3377 return ret;
3378 }
3379 }
3380
3381 list_for_each_entry_rcu(dev, head, dev_list) {
3382 if (!dev->bdev) {
3383 total_errors++;
3384 continue;
3385 }
3386 if (!dev->in_fs_metadata || !dev->writeable)
3387 continue;
3388
3389 btrfs_set_stack_device_generation(dev_item, 0);
3390 btrfs_set_stack_device_type(dev_item, dev->type);
3391 btrfs_set_stack_device_id(dev_item, dev->devid);
3392 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3393 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3394 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3395 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3396 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3397 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3398 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3399
3400 flags = btrfs_super_flags(sb);
3401 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3402
3403 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3404 if (ret)
3405 total_errors++;
3406 }
3407 if (total_errors > max_errors) {
3408 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3409 total_errors);
3410
3411 /* This shouldn't happen. FUA is masked off if unsupported */
3412 BUG();
3413 }
3414
3415 total_errors = 0;
3416 list_for_each_entry_rcu(dev, head, dev_list) {
3417 if (!dev->bdev)
3418 continue;
3419 if (!dev->in_fs_metadata || !dev->writeable)
3420 continue;
3421
3422 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3423 if (ret)
3424 total_errors++;
3425 }
3426 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3427 if (total_errors > max_errors) {
3428 btrfs_error(root->fs_info, -EIO,
3429 "%d errors while writing supers", total_errors);
3430 return -EIO;
3431 }
3432 return 0;
3433}
3434
3435int write_ctree_super(struct btrfs_trans_handle *trans,
3436 struct btrfs_root *root, int max_mirrors)
3437{
3438 int ret;
3439
3440 ret = write_all_supers(root, max_mirrors);
3441 return ret;
3442}
3443
3444/* Drop a fs root from the radix tree and free it. */
3445void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3446 struct btrfs_root *root)
3447{
3448 spin_lock(&fs_info->fs_roots_radix_lock);
3449 radix_tree_delete(&fs_info->fs_roots_radix,
3450 (unsigned long)root->root_key.objectid);
3451 spin_unlock(&fs_info->fs_roots_radix_lock);
3452
3453 if (btrfs_root_refs(&root->root_item) == 0)
3454 synchronize_srcu(&fs_info->subvol_srcu);
3455
3456 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3457 btrfs_free_log(NULL, root);
3458 btrfs_free_log_root_tree(NULL, fs_info);
3459 }
3460
3461 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3462 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3463 free_fs_root(root);
3464}
3465
3466static void free_fs_root(struct btrfs_root *root)
3467{
3468 iput(root->cache_inode);
3469 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3470 if (root->anon_dev)
3471 free_anon_bdev(root->anon_dev);
3472 free_extent_buffer(root->node);
3473 free_extent_buffer(root->commit_root);
3474 kfree(root->free_ino_ctl);
3475 kfree(root->free_ino_pinned);
3476 kfree(root->name);
3477 btrfs_put_fs_root(root);
3478}
3479
3480void btrfs_free_fs_root(struct btrfs_root *root)
3481{
3482 free_fs_root(root);
3483}
3484
3485int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3486{
3487 u64 root_objectid = 0;
3488 struct btrfs_root *gang[8];
3489 int i;
3490 int ret;
3491
3492 while (1) {
3493 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3494 (void **)gang, root_objectid,
3495 ARRAY_SIZE(gang));
3496 if (!ret)
3497 break;
3498
3499 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3500 for (i = 0; i < ret; i++) {
3501 int err;
3502
3503 root_objectid = gang[i]->root_key.objectid;
3504 err = btrfs_orphan_cleanup(gang[i]);
3505 if (err)
3506 return err;
3507 }
3508 root_objectid++;
3509 }
3510 return 0;
3511}
3512
3513int btrfs_commit_super(struct btrfs_root *root)
3514{
3515 struct btrfs_trans_handle *trans;
3516 int ret;
3517
3518 mutex_lock(&root->fs_info->cleaner_mutex);
3519 btrfs_run_delayed_iputs(root);
3520 mutex_unlock(&root->fs_info->cleaner_mutex);
3521 wake_up_process(root->fs_info->cleaner_kthread);
3522
3523 /* wait until ongoing cleanup work done */
3524 down_write(&root->fs_info->cleanup_work_sem);
3525 up_write(&root->fs_info->cleanup_work_sem);
3526
3527 trans = btrfs_join_transaction(root);
3528 if (IS_ERR(trans))
3529 return PTR_ERR(trans);
3530 ret = btrfs_commit_transaction(trans, root);
3531 if (ret)
3532 return ret;
3533 /* run commit again to drop the original snapshot */
3534 trans = btrfs_join_transaction(root);
3535 if (IS_ERR(trans))
3536 return PTR_ERR(trans);
3537 ret = btrfs_commit_transaction(trans, root);
3538 if (ret)
3539 return ret;
3540 ret = btrfs_write_and_wait_transaction(NULL, root);
3541 if (ret) {
3542 btrfs_error(root->fs_info, ret,
3543 "Failed to sync btree inode to disk.");
3544 return ret;
3545 }
3546
3547 ret = write_ctree_super(NULL, root, 0);
3548 return ret;
3549}
3550
3551int close_ctree(struct btrfs_root *root)
3552{
3553 struct btrfs_fs_info *fs_info = root->fs_info;
3554 int ret;
3555
3556 fs_info->closing = 1;
3557 smp_mb();
3558
3559 /* wait for the uuid_scan task to finish */
3560 down(&fs_info->uuid_tree_rescan_sem);
3561 /* avoid complains from lockdep et al., set sem back to initial state */
3562 up(&fs_info->uuid_tree_rescan_sem);
3563
3564 /* pause restriper - we want to resume on mount */
3565 btrfs_pause_balance(fs_info);
3566
3567 btrfs_dev_replace_suspend_for_unmount(fs_info);
3568
3569 btrfs_scrub_cancel(fs_info);
3570
3571 /* wait for any defraggers to finish */
3572 wait_event(fs_info->transaction_wait,
3573 (atomic_read(&fs_info->defrag_running) == 0));
3574
3575 /* clear out the rbtree of defraggable inodes */
3576 btrfs_cleanup_defrag_inodes(fs_info);
3577
3578 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3579 ret = btrfs_commit_super(root);
3580 if (ret)
3581 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3582 }
3583
3584 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3585 btrfs_error_commit_super(root);
3586
3587 btrfs_put_block_group_cache(fs_info);
3588
3589 kthread_stop(fs_info->transaction_kthread);
3590 kthread_stop(fs_info->cleaner_kthread);
3591
3592 fs_info->closing = 2;
3593 smp_mb();
3594
3595 btrfs_free_qgroup_config(root->fs_info);
3596
3597 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3598 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3599 percpu_counter_sum(&fs_info->delalloc_bytes));
3600 }
3601
3602 btrfs_free_block_groups(fs_info);
3603
3604 btrfs_stop_all_workers(fs_info);
3605
3606 del_fs_roots(fs_info);
3607
3608 free_root_pointers(fs_info, 1);
3609
3610 iput(fs_info->btree_inode);
3611
3612#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3613 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3614 btrfsic_unmount(root, fs_info->fs_devices);
3615#endif
3616
3617 btrfs_close_devices(fs_info->fs_devices);
3618 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3619
3620 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3621 percpu_counter_destroy(&fs_info->delalloc_bytes);
3622 bdi_destroy(&fs_info->bdi);
3623 cleanup_srcu_struct(&fs_info->subvol_srcu);
3624
3625 btrfs_free_stripe_hash_table(fs_info);
3626
3627 return 0;
3628}
3629
3630int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3631 int atomic)
3632{
3633 int ret;
3634 struct inode *btree_inode = buf->pages[0]->mapping->host;
3635
3636 ret = extent_buffer_uptodate(buf);
3637 if (!ret)
3638 return ret;
3639
3640 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3641 parent_transid, atomic);
3642 if (ret == -EAGAIN)
3643 return ret;
3644 return !ret;
3645}
3646
3647int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3648{
3649 return set_extent_buffer_uptodate(buf);
3650}
3651
3652void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3653{
3654 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3655 u64 transid = btrfs_header_generation(buf);
3656 int was_dirty;
3657
3658 btrfs_assert_tree_locked(buf);
3659 if (transid != root->fs_info->generation)
3660 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3661 "found %llu running %llu\n",
3662 (unsigned long long)buf->start,
3663 (unsigned long long)transid,
3664 (unsigned long long)root->fs_info->generation);
3665 was_dirty = set_extent_buffer_dirty(buf);
3666 if (!was_dirty)
3667 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3668 buf->len,
3669 root->fs_info->dirty_metadata_batch);
3670}
3671
3672static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3673 int flush_delayed)
3674{
3675 /*
3676 * looks as though older kernels can get into trouble with
3677 * this code, they end up stuck in balance_dirty_pages forever
3678 */
3679 int ret;
3680
3681 if (current->flags & PF_MEMALLOC)
3682 return;
3683
3684 if (flush_delayed)
3685 btrfs_balance_delayed_items(root);
3686
3687 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3688 BTRFS_DIRTY_METADATA_THRESH);
3689 if (ret > 0) {
3690 balance_dirty_pages_ratelimited(
3691 root->fs_info->btree_inode->i_mapping);
3692 }
3693 return;
3694}
3695
3696void btrfs_btree_balance_dirty(struct btrfs_root *root)
3697{
3698 __btrfs_btree_balance_dirty(root, 1);
3699}
3700
3701void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3702{
3703 __btrfs_btree_balance_dirty(root, 0);
3704}
3705
3706int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3707{
3708 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3709 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3710}
3711
3712static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3713 int read_only)
3714{
3715 /*
3716 * Placeholder for checks
3717 */
3718 return 0;
3719}
3720
3721static void btrfs_error_commit_super(struct btrfs_root *root)
3722{
3723 mutex_lock(&root->fs_info->cleaner_mutex);
3724 btrfs_run_delayed_iputs(root);
3725 mutex_unlock(&root->fs_info->cleaner_mutex);
3726
3727 down_write(&root->fs_info->cleanup_work_sem);
3728 up_write(&root->fs_info->cleanup_work_sem);
3729
3730 /* cleanup FS via transaction */
3731 btrfs_cleanup_transaction(root);
3732}
3733
3734static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3735 struct btrfs_root *root)
3736{
3737 struct btrfs_inode *btrfs_inode;
3738 struct list_head splice;
3739
3740 INIT_LIST_HEAD(&splice);
3741
3742 mutex_lock(&root->fs_info->ordered_operations_mutex);
3743 spin_lock(&root->fs_info->ordered_root_lock);
3744
3745 list_splice_init(&t->ordered_operations, &splice);
3746 while (!list_empty(&splice)) {
3747 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3748 ordered_operations);
3749
3750 list_del_init(&btrfs_inode->ordered_operations);
3751 spin_unlock(&root->fs_info->ordered_root_lock);
3752
3753 btrfs_invalidate_inodes(btrfs_inode->root);
3754
3755 spin_lock(&root->fs_info->ordered_root_lock);
3756 }
3757
3758 spin_unlock(&root->fs_info->ordered_root_lock);
3759 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3760}
3761
3762static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3763{
3764 struct btrfs_ordered_extent *ordered;
3765
3766 spin_lock(&root->ordered_extent_lock);
3767 /*
3768 * This will just short circuit the ordered completion stuff which will
3769 * make sure the ordered extent gets properly cleaned up.
3770 */
3771 list_for_each_entry(ordered, &root->ordered_extents,
3772 root_extent_list)
3773 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3774 spin_unlock(&root->ordered_extent_lock);
3775}
3776
3777static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3778{
3779 struct btrfs_root *root;
3780 struct list_head splice;
3781
3782 INIT_LIST_HEAD(&splice);
3783
3784 spin_lock(&fs_info->ordered_root_lock);
3785 list_splice_init(&fs_info->ordered_roots, &splice);
3786 while (!list_empty(&splice)) {
3787 root = list_first_entry(&splice, struct btrfs_root,
3788 ordered_root);
3789 list_del_init(&root->ordered_root);
3790
3791 btrfs_destroy_ordered_extents(root);
3792
3793 cond_resched_lock(&fs_info->ordered_root_lock);
3794 }
3795 spin_unlock(&fs_info->ordered_root_lock);
3796}
3797
3798static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3799 struct btrfs_root *root)
3800{
3801 struct rb_node *node;
3802 struct btrfs_delayed_ref_root *delayed_refs;
3803 struct btrfs_delayed_ref_node *ref;
3804 int ret = 0;
3805
3806 delayed_refs = &trans->delayed_refs;
3807
3808 spin_lock(&delayed_refs->lock);
3809 if (delayed_refs->num_entries == 0) {
3810 spin_unlock(&delayed_refs->lock);
3811 printk(KERN_INFO "delayed_refs has NO entry\n");
3812 return ret;
3813 }
3814
3815 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3816 struct btrfs_delayed_ref_head *head = NULL;
3817 bool pin_bytes = false;
3818
3819 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3820 atomic_set(&ref->refs, 1);
3821 if (btrfs_delayed_ref_is_head(ref)) {
3822
3823 head = btrfs_delayed_node_to_head(ref);
3824 if (!mutex_trylock(&head->mutex)) {
3825 atomic_inc(&ref->refs);
3826 spin_unlock(&delayed_refs->lock);
3827
3828 /* Need to wait for the delayed ref to run */
3829 mutex_lock(&head->mutex);
3830 mutex_unlock(&head->mutex);
3831 btrfs_put_delayed_ref(ref);
3832
3833 spin_lock(&delayed_refs->lock);
3834 continue;
3835 }
3836
3837 if (head->must_insert_reserved)
3838 pin_bytes = true;
3839 btrfs_free_delayed_extent_op(head->extent_op);
3840 delayed_refs->num_heads--;
3841 if (list_empty(&head->cluster))
3842 delayed_refs->num_heads_ready--;
3843 list_del_init(&head->cluster);
3844 }
3845
3846 ref->in_tree = 0;
3847 rb_erase(&ref->rb_node, &delayed_refs->root);
3848 delayed_refs->num_entries--;
3849 spin_unlock(&delayed_refs->lock);
3850 if (head) {
3851 if (pin_bytes)
3852 btrfs_pin_extent(root, ref->bytenr,
3853 ref->num_bytes, 1);
3854 mutex_unlock(&head->mutex);
3855 }
3856 btrfs_put_delayed_ref(ref);
3857
3858 cond_resched();
3859 spin_lock(&delayed_refs->lock);
3860 }
3861
3862 spin_unlock(&delayed_refs->lock);
3863
3864 return ret;
3865}
3866
3867static void btrfs_evict_pending_snapshots(struct btrfs_transaction *t)
3868{
3869 struct btrfs_pending_snapshot *snapshot;
3870 struct list_head splice;
3871
3872 INIT_LIST_HEAD(&splice);
3873
3874 list_splice_init(&t->pending_snapshots, &splice);
3875
3876 while (!list_empty(&splice)) {
3877 snapshot = list_entry(splice.next,
3878 struct btrfs_pending_snapshot,
3879 list);
3880 snapshot->error = -ECANCELED;
3881 list_del_init(&snapshot->list);
3882 }
3883}
3884
3885static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3886{
3887 struct btrfs_inode *btrfs_inode;
3888 struct list_head splice;
3889
3890 INIT_LIST_HEAD(&splice);
3891
3892 spin_lock(&root->delalloc_lock);
3893 list_splice_init(&root->delalloc_inodes, &splice);
3894
3895 while (!list_empty(&splice)) {
3896 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3897 delalloc_inodes);
3898
3899 list_del_init(&btrfs_inode->delalloc_inodes);
3900 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3901 &btrfs_inode->runtime_flags);
3902 spin_unlock(&root->delalloc_lock);
3903
3904 btrfs_invalidate_inodes(btrfs_inode->root);
3905
3906 spin_lock(&root->delalloc_lock);
3907 }
3908
3909 spin_unlock(&root->delalloc_lock);
3910}
3911
3912static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3913{
3914 struct btrfs_root *root;
3915 struct list_head splice;
3916
3917 INIT_LIST_HEAD(&splice);
3918
3919 spin_lock(&fs_info->delalloc_root_lock);
3920 list_splice_init(&fs_info->delalloc_roots, &splice);
3921 while (!list_empty(&splice)) {
3922 root = list_first_entry(&splice, struct btrfs_root,
3923 delalloc_root);
3924 list_del_init(&root->delalloc_root);
3925 root = btrfs_grab_fs_root(root);
3926 BUG_ON(!root);
3927 spin_unlock(&fs_info->delalloc_root_lock);
3928
3929 btrfs_destroy_delalloc_inodes(root);
3930 btrfs_put_fs_root(root);
3931
3932 spin_lock(&fs_info->delalloc_root_lock);
3933 }
3934 spin_unlock(&fs_info->delalloc_root_lock);
3935}
3936
3937static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3938 struct extent_io_tree *dirty_pages,
3939 int mark)
3940{
3941 int ret;
3942 struct extent_buffer *eb;
3943 u64 start = 0;
3944 u64 end;
3945
3946 while (1) {
3947 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3948 mark, NULL);
3949 if (ret)
3950 break;
3951
3952 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3953 while (start <= end) {
3954 eb = btrfs_find_tree_block(root, start,
3955 root->leafsize);
3956 start += root->leafsize;
3957 if (!eb)
3958 continue;
3959 wait_on_extent_buffer_writeback(eb);
3960
3961 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3962 &eb->bflags))
3963 clear_extent_buffer_dirty(eb);
3964 free_extent_buffer_stale(eb);
3965 }
3966 }
3967
3968 return ret;
3969}
3970
3971static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3972 struct extent_io_tree *pinned_extents)
3973{
3974 struct extent_io_tree *unpin;
3975 u64 start;
3976 u64 end;
3977 int ret;
3978 bool loop = true;
3979
3980 unpin = pinned_extents;
3981again:
3982 while (1) {
3983 ret = find_first_extent_bit(unpin, 0, &start, &end,
3984 EXTENT_DIRTY, NULL);
3985 if (ret)
3986 break;
3987
3988 /* opt_discard */
3989 if (btrfs_test_opt(root, DISCARD))
3990 ret = btrfs_error_discard_extent(root, start,
3991 end + 1 - start,
3992 NULL);
3993
3994 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3995 btrfs_error_unpin_extent_range(root, start, end);
3996 cond_resched();
3997 }
3998
3999 if (loop) {
4000 if (unpin == &root->fs_info->freed_extents[0])
4001 unpin = &root->fs_info->freed_extents[1];
4002 else
4003 unpin = &root->fs_info->freed_extents[0];
4004 loop = false;
4005 goto again;
4006 }
4007
4008 return 0;
4009}
4010
4011void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4012 struct btrfs_root *root)
4013{
4014 btrfs_destroy_delayed_refs(cur_trans, root);
4015 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
4016 cur_trans->dirty_pages.dirty_bytes);
4017
4018 cur_trans->state = TRANS_STATE_COMMIT_START;
4019 wake_up(&root->fs_info->transaction_blocked_wait);
4020
4021 btrfs_evict_pending_snapshots(cur_trans);
4022
4023 cur_trans->state = TRANS_STATE_UNBLOCKED;
4024 wake_up(&root->fs_info->transaction_wait);
4025
4026 btrfs_destroy_delayed_inodes(root);
4027 btrfs_assert_delayed_root_empty(root);
4028
4029 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4030 EXTENT_DIRTY);
4031 btrfs_destroy_pinned_extent(root,
4032 root->fs_info->pinned_extents);
4033
4034 cur_trans->state =TRANS_STATE_COMPLETED;
4035 wake_up(&cur_trans->commit_wait);
4036
4037 /*
4038 memset(cur_trans, 0, sizeof(*cur_trans));
4039 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4040 */
4041}
4042
4043static int btrfs_cleanup_transaction(struct btrfs_root *root)
4044{
4045 struct btrfs_transaction *t;
4046 LIST_HEAD(list);
4047
4048 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4049
4050 spin_lock(&root->fs_info->trans_lock);
4051 list_splice_init(&root->fs_info->trans_list, &list);
4052 root->fs_info->running_transaction = NULL;
4053 spin_unlock(&root->fs_info->trans_lock);
4054
4055 while (!list_empty(&list)) {
4056 t = list_entry(list.next, struct btrfs_transaction, list);
4057
4058 btrfs_destroy_ordered_operations(t, root);
4059
4060 btrfs_destroy_all_ordered_extents(root->fs_info);
4061
4062 btrfs_destroy_delayed_refs(t, root);
4063
4064 /*
4065 * FIXME: cleanup wait for commit
4066 * We needn't acquire the lock here, because we are during
4067 * the umount, there is no other task which will change it.
4068 */
4069 t->state = TRANS_STATE_COMMIT_START;
4070 smp_mb();
4071 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
4072 wake_up(&root->fs_info->transaction_blocked_wait);
4073
4074 btrfs_evict_pending_snapshots(t);
4075
4076 t->state = TRANS_STATE_UNBLOCKED;
4077 smp_mb();
4078 if (waitqueue_active(&root->fs_info->transaction_wait))
4079 wake_up(&root->fs_info->transaction_wait);
4080
4081 btrfs_destroy_delayed_inodes(root);
4082 btrfs_assert_delayed_root_empty(root);
4083
4084 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4085
4086 btrfs_destroy_marked_extents(root, &t->dirty_pages,
4087 EXTENT_DIRTY);
4088
4089 btrfs_destroy_pinned_extent(root,
4090 root->fs_info->pinned_extents);
4091
4092 t->state = TRANS_STATE_COMPLETED;
4093 smp_mb();
4094 if (waitqueue_active(&t->commit_wait))
4095 wake_up(&t->commit_wait);
4096
4097 atomic_set(&t->use_count, 0);
4098 list_del_init(&t->list);
4099 memset(t, 0, sizeof(*t));
4100 kmem_cache_free(btrfs_transaction_cachep, t);
4101 }
4102
4103 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4104
4105 return 0;
4106}
4107
4108static struct extent_io_ops btree_extent_io_ops = {
4109 .readpage_end_io_hook = btree_readpage_end_io_hook,
4110 .readpage_io_failed_hook = btree_io_failed_hook,
4111 .submit_bio_hook = btree_submit_bio_hook,
4112 /* note we're sharing with inode.c for the merge bio hook */
4113 .merge_bio_hook = btrfs_merge_bio_hook,
4114};
Linux 6.x block layer and io_uring tree(s)