Loading...
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/fs.h>
20#include <linux/blkdev.h>
21#include <linux/scatterlist.h>
22#include <linux/swap.h>
23#include <linux/radix-tree.h>
24#include <linux/writeback.h>
25#include <linux/buffer_head.h>
26#include <linux/workqueue.h>
27#include <linux/kthread.h>
28#include <linux/freezer.h>
29#include <linux/slab.h>
30#include <linux/migrate.h>
31#include <linux/ratelimit.h>
32#include <linux/uuid.h>
33#include <linux/semaphore.h>
34#include <asm/unaligned.h>
35#include "ctree.h"
36#include "disk-io.h"
37#include "hash.h"
38#include "transaction.h"
39#include "btrfs_inode.h"
40#include "volumes.h"
41#include "print-tree.h"
42#include "async-thread.h"
43#include "locking.h"
44#include "tree-log.h"
45#include "free-space-cache.h"
46#include "inode-map.h"
47#include "check-integrity.h"
48#include "rcu-string.h"
49#include "dev-replace.h"
50#include "raid56.h"
51#include "sysfs.h"
52
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_destroy_delalloc_inodes(struct btrfs_root *root);
68static int btrfs_destroy_marked_extents(struct btrfs_root *root,
69 struct extent_io_tree *dirty_pages,
70 int mark);
71static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
72 struct extent_io_tree *pinned_extents);
73static int btrfs_cleanup_transaction(struct btrfs_root *root);
74static void btrfs_error_commit_super(struct btrfs_root *root);
75
76/*
77 * end_io_wq structs are used to do processing in task context when an IO is
78 * complete. This is used during reads to verify checksums, and it is used
79 * by writes to insert metadata for new file extents after IO is complete.
80 */
81struct end_io_wq {
82 struct bio *bio;
83 bio_end_io_t *end_io;
84 void *private;
85 struct btrfs_fs_info *info;
86 int error;
87 int metadata;
88 struct list_head list;
89 struct btrfs_work work;
90};
91
92/*
93 * async submit bios are used to offload expensive checksumming
94 * onto the worker threads. They checksum file and metadata bios
95 * just before they are sent down the IO stack.
96 */
97struct async_submit_bio {
98 struct inode *inode;
99 struct bio *bio;
100 struct list_head list;
101 extent_submit_bio_hook_t *submit_bio_start;
102 extent_submit_bio_hook_t *submit_bio_done;
103 int rw;
104 int mirror_num;
105 unsigned long bio_flags;
106 /*
107 * bio_offset is optional, can be used if the pages in the bio
108 * can't tell us where in the file the bio should go
109 */
110 u64 bio_offset;
111 struct btrfs_work work;
112 int error;
113};
114
115/*
116 * Lockdep class keys for extent_buffer->lock's in this root. For a given
117 * eb, the lockdep key is determined by the btrfs_root it belongs to and
118 * the level the eb occupies in the tree.
119 *
120 * Different roots are used for different purposes and may nest inside each
121 * other and they require separate keysets. As lockdep keys should be
122 * static, assign keysets according to the purpose of the root as indicated
123 * by btrfs_root->objectid. This ensures that all special purpose roots
124 * have separate keysets.
125 *
126 * Lock-nesting across peer nodes is always done with the immediate parent
127 * node locked thus preventing deadlock. As lockdep doesn't know this, use
128 * subclass to avoid triggering lockdep warning in such cases.
129 *
130 * The key is set by the readpage_end_io_hook after the buffer has passed
131 * csum validation but before the pages are unlocked. It is also set by
132 * btrfs_init_new_buffer on freshly allocated blocks.
133 *
134 * We also add a check to make sure the highest level of the tree is the
135 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
136 * needs update as well.
137 */
138#ifdef CONFIG_DEBUG_LOCK_ALLOC
139# if BTRFS_MAX_LEVEL != 8
140# error
141# endif
142
143static struct btrfs_lockdep_keyset {
144 u64 id; /* root objectid */
145 const char *name_stem; /* lock name stem */
146 char names[BTRFS_MAX_LEVEL + 1][20];
147 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
148} btrfs_lockdep_keysets[] = {
149 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
150 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
151 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
152 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
153 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
154 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
155 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
156 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
157 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
158 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
159 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
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 btrfs_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
304 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
305 "level %d\n",
306 root->fs_info->sb->s_id, buf->start,
307 val, found, btrfs_header_level(buf));
308 if (result != (char *)&inline_result)
309 kfree(result);
310 return 1;
311 }
312 } else {
313 write_extent_buffer(buf, result, 0, csum_size);
314 }
315 if (result != (char *)&inline_result)
316 kfree(result);
317 return 0;
318}
319
320/*
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
325 */
326static int verify_parent_transid(struct extent_io_tree *io_tree,
327 struct extent_buffer *eb, u64 parent_transid,
328 int atomic)
329{
330 struct extent_state *cached_state = NULL;
331 int ret;
332 bool need_lock = (current->journal_info ==
333 (void *)BTRFS_SEND_TRANS_STUB);
334
335 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
336 return 0;
337
338 if (atomic)
339 return -EAGAIN;
340
341 if (need_lock) {
342 btrfs_tree_read_lock(eb);
343 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
344 }
345
346 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
347 0, &cached_state);
348 if (extent_buffer_uptodate(eb) &&
349 btrfs_header_generation(eb) == parent_transid) {
350 ret = 0;
351 goto out;
352 }
353 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
354 "found %llu\n",
355 eb->start, parent_transid, btrfs_header_generation(eb));
356 ret = 1;
357
358 /*
359 * Things reading via commit roots that don't have normal protection,
360 * like send, can have a really old block in cache that may point at a
361 * block that has been free'd and re-allocated. So don't clear uptodate
362 * if we find an eb that is under IO (dirty/writeback) because we could
363 * end up reading in the stale data and then writing it back out and
364 * making everybody very sad.
365 */
366 if (!extent_buffer_under_io(eb))
367 clear_extent_buffer_uptodate(eb);
368out:
369 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
370 &cached_state, GFP_NOFS);
371 btrfs_tree_read_unlock_blocking(eb);
372 return ret;
373}
374
375/*
376 * Return 0 if the superblock checksum type matches the checksum value of that
377 * algorithm. Pass the raw disk superblock data.
378 */
379static int btrfs_check_super_csum(char *raw_disk_sb)
380{
381 struct btrfs_super_block *disk_sb =
382 (struct btrfs_super_block *)raw_disk_sb;
383 u16 csum_type = btrfs_super_csum_type(disk_sb);
384 int ret = 0;
385
386 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
387 u32 crc = ~(u32)0;
388 const int csum_size = sizeof(crc);
389 char result[csum_size];
390
391 /*
392 * The super_block structure does not span the whole
393 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
394 * is filled with zeros and is included in the checkum.
395 */
396 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
397 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
398 btrfs_csum_final(crc, result);
399
400 if (memcmp(raw_disk_sb, result, csum_size))
401 ret = 1;
402
403 if (ret && btrfs_super_generation(disk_sb) < 10) {
404 printk(KERN_WARNING
405 "BTRFS: super block crcs don't match, older mkfs detected\n");
406 ret = 0;
407 }
408 }
409
410 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
411 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
412 csum_type);
413 ret = 1;
414 }
415
416 return ret;
417}
418
419/*
420 * helper to read a given tree block, doing retries as required when
421 * the checksums don't match and we have alternate mirrors to try.
422 */
423static int btree_read_extent_buffer_pages(struct btrfs_root *root,
424 struct extent_buffer *eb,
425 u64 start, u64 parent_transid)
426{
427 struct extent_io_tree *io_tree;
428 int failed = 0;
429 int ret;
430 int num_copies = 0;
431 int mirror_num = 0;
432 int failed_mirror = 0;
433
434 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
435 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
436 while (1) {
437 ret = read_extent_buffer_pages(io_tree, eb, start,
438 WAIT_COMPLETE,
439 btree_get_extent, mirror_num);
440 if (!ret) {
441 if (!verify_parent_transid(io_tree, eb,
442 parent_transid, 0))
443 break;
444 else
445 ret = -EIO;
446 }
447
448 /*
449 * This buffer's crc is fine, but its contents are corrupted, so
450 * there is no reason to read the other copies, they won't be
451 * any less wrong.
452 */
453 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
454 break;
455
456 num_copies = btrfs_num_copies(root->fs_info,
457 eb->start, eb->len);
458 if (num_copies == 1)
459 break;
460
461 if (!failed_mirror) {
462 failed = 1;
463 failed_mirror = eb->read_mirror;
464 }
465
466 mirror_num++;
467 if (mirror_num == failed_mirror)
468 mirror_num++;
469
470 if (mirror_num > num_copies)
471 break;
472 }
473
474 if (failed && !ret && failed_mirror)
475 repair_eb_io_failure(root, eb, failed_mirror);
476
477 return ret;
478}
479
480/*
481 * checksum a dirty tree block before IO. This has extra checks to make sure
482 * we only fill in the checksum field in the first page of a multi-page block
483 */
484
485static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
486{
487 u64 start = page_offset(page);
488 u64 found_start;
489 struct extent_buffer *eb;
490
491 eb = (struct extent_buffer *)page->private;
492 if (page != eb->pages[0])
493 return 0;
494 found_start = btrfs_header_bytenr(eb);
495 if (WARN_ON(found_start != start || !PageUptodate(page)))
496 return 0;
497 csum_tree_block(root, eb, 0);
498 return 0;
499}
500
501static int check_tree_block_fsid(struct btrfs_root *root,
502 struct extent_buffer *eb)
503{
504 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
505 u8 fsid[BTRFS_UUID_SIZE];
506 int ret = 1;
507
508 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
509 while (fs_devices) {
510 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
511 ret = 0;
512 break;
513 }
514 fs_devices = fs_devices->seed;
515 }
516 return ret;
517}
518
519#define CORRUPT(reason, eb, root, slot) \
520 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
521 "root=%llu, slot=%d", reason, \
522 btrfs_header_bytenr(eb), root->objectid, slot)
523
524static noinline int check_leaf(struct btrfs_root *root,
525 struct extent_buffer *leaf)
526{
527 struct btrfs_key key;
528 struct btrfs_key leaf_key;
529 u32 nritems = btrfs_header_nritems(leaf);
530 int slot;
531
532 if (nritems == 0)
533 return 0;
534
535 /* Check the 0 item */
536 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
537 BTRFS_LEAF_DATA_SIZE(root)) {
538 CORRUPT("invalid item offset size pair", leaf, root, 0);
539 return -EIO;
540 }
541
542 /*
543 * Check to make sure each items keys are in the correct order and their
544 * offsets make sense. We only have to loop through nritems-1 because
545 * we check the current slot against the next slot, which verifies the
546 * next slot's offset+size makes sense and that the current's slot
547 * offset is correct.
548 */
549 for (slot = 0; slot < nritems - 1; slot++) {
550 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
551 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
552
553 /* Make sure the keys are in the right order */
554 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
555 CORRUPT("bad key order", leaf, root, slot);
556 return -EIO;
557 }
558
559 /*
560 * Make sure the offset and ends are right, remember that the
561 * item data starts at the end of the leaf and grows towards the
562 * front.
563 */
564 if (btrfs_item_offset_nr(leaf, slot) !=
565 btrfs_item_end_nr(leaf, slot + 1)) {
566 CORRUPT("slot offset bad", leaf, root, slot);
567 return -EIO;
568 }
569
570 /*
571 * Check to make sure that we don't point outside of the leaf,
572 * just incase all the items are consistent to eachother, but
573 * all point outside of the leaf.
574 */
575 if (btrfs_item_end_nr(leaf, slot) >
576 BTRFS_LEAF_DATA_SIZE(root)) {
577 CORRUPT("slot end outside of leaf", leaf, root, slot);
578 return -EIO;
579 }
580 }
581
582 return 0;
583}
584
585static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
586 u64 phy_offset, struct page *page,
587 u64 start, u64 end, int mirror)
588{
589 u64 found_start;
590 int found_level;
591 struct extent_buffer *eb;
592 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
593 int ret = 0;
594 int reads_done;
595
596 if (!page->private)
597 goto out;
598
599 eb = (struct extent_buffer *)page->private;
600
601 /* the pending IO might have been the only thing that kept this buffer
602 * in memory. Make sure we have a ref for all this other checks
603 */
604 extent_buffer_get(eb);
605
606 reads_done = atomic_dec_and_test(&eb->io_pages);
607 if (!reads_done)
608 goto err;
609
610 eb->read_mirror = mirror;
611 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
612 ret = -EIO;
613 goto err;
614 }
615
616 found_start = btrfs_header_bytenr(eb);
617 if (found_start != eb->start) {
618 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
619 "%llu %llu\n",
620 found_start, eb->start);
621 ret = -EIO;
622 goto err;
623 }
624 if (check_tree_block_fsid(root, eb)) {
625 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
626 eb->start);
627 ret = -EIO;
628 goto err;
629 }
630 found_level = btrfs_header_level(eb);
631 if (found_level >= BTRFS_MAX_LEVEL) {
632 btrfs_info(root->fs_info, "bad tree block level %d",
633 (int)btrfs_header_level(eb));
634 ret = -EIO;
635 goto err;
636 }
637
638 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
639 eb, found_level);
640
641 ret = csum_tree_block(root, eb, 1);
642 if (ret) {
643 ret = -EIO;
644 goto err;
645 }
646
647 /*
648 * If this is a leaf block and it is corrupt, set the corrupt bit so
649 * that we don't try and read the other copies of this block, just
650 * return -EIO.
651 */
652 if (found_level == 0 && check_leaf(root, eb)) {
653 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
654 ret = -EIO;
655 }
656
657 if (!ret)
658 set_extent_buffer_uptodate(eb);
659err:
660 if (reads_done &&
661 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
662 btree_readahead_hook(root, eb, eb->start, ret);
663
664 if (ret) {
665 /*
666 * our io error hook is going to dec the io pages
667 * again, we have to make sure it has something
668 * to decrement
669 */
670 atomic_inc(&eb->io_pages);
671 clear_extent_buffer_uptodate(eb);
672 }
673 free_extent_buffer(eb);
674out:
675 return ret;
676}
677
678static int btree_io_failed_hook(struct page *page, int failed_mirror)
679{
680 struct extent_buffer *eb;
681 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
682
683 eb = (struct extent_buffer *)page->private;
684 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
685 eb->read_mirror = failed_mirror;
686 atomic_dec(&eb->io_pages);
687 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
688 btree_readahead_hook(root, eb, eb->start, -EIO);
689 return -EIO; /* we fixed nothing */
690}
691
692static void end_workqueue_bio(struct bio *bio, int err)
693{
694 struct end_io_wq *end_io_wq = bio->bi_private;
695 struct btrfs_fs_info *fs_info;
696
697 fs_info = end_io_wq->info;
698 end_io_wq->error = err;
699 btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
700
701 if (bio->bi_rw & REQ_WRITE) {
702 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
703 btrfs_queue_work(fs_info->endio_meta_write_workers,
704 &end_io_wq->work);
705 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
706 btrfs_queue_work(fs_info->endio_freespace_worker,
707 &end_io_wq->work);
708 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
709 btrfs_queue_work(fs_info->endio_raid56_workers,
710 &end_io_wq->work);
711 else
712 btrfs_queue_work(fs_info->endio_write_workers,
713 &end_io_wq->work);
714 } else {
715 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
716 btrfs_queue_work(fs_info->endio_raid56_workers,
717 &end_io_wq->work);
718 else if (end_io_wq->metadata)
719 btrfs_queue_work(fs_info->endio_meta_workers,
720 &end_io_wq->work);
721 else
722 btrfs_queue_work(fs_info->endio_workers,
723 &end_io_wq->work);
724 }
725}
726
727/*
728 * For the metadata arg you want
729 *
730 * 0 - if data
731 * 1 - if normal metadta
732 * 2 - if writing to the free space cache area
733 * 3 - raid parity work
734 */
735int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
736 int metadata)
737{
738 struct end_io_wq *end_io_wq;
739 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
740 if (!end_io_wq)
741 return -ENOMEM;
742
743 end_io_wq->private = bio->bi_private;
744 end_io_wq->end_io = bio->bi_end_io;
745 end_io_wq->info = info;
746 end_io_wq->error = 0;
747 end_io_wq->bio = bio;
748 end_io_wq->metadata = metadata;
749
750 bio->bi_private = end_io_wq;
751 bio->bi_end_io = end_workqueue_bio;
752 return 0;
753}
754
755unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
756{
757 unsigned long limit = min_t(unsigned long,
758 info->thread_pool_size,
759 info->fs_devices->open_devices);
760 return 256 * limit;
761}
762
763static void run_one_async_start(struct btrfs_work *work)
764{
765 struct async_submit_bio *async;
766 int ret;
767
768 async = container_of(work, struct async_submit_bio, work);
769 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
770 async->mirror_num, async->bio_flags,
771 async->bio_offset);
772 if (ret)
773 async->error = ret;
774}
775
776static void run_one_async_done(struct btrfs_work *work)
777{
778 struct btrfs_fs_info *fs_info;
779 struct async_submit_bio *async;
780 int limit;
781
782 async = container_of(work, struct async_submit_bio, work);
783 fs_info = BTRFS_I(async->inode)->root->fs_info;
784
785 limit = btrfs_async_submit_limit(fs_info);
786 limit = limit * 2 / 3;
787
788 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
789 waitqueue_active(&fs_info->async_submit_wait))
790 wake_up(&fs_info->async_submit_wait);
791
792 /* If an error occured we just want to clean up the bio and move on */
793 if (async->error) {
794 bio_endio(async->bio, async->error);
795 return;
796 }
797
798 async->submit_bio_done(async->inode, async->rw, async->bio,
799 async->mirror_num, async->bio_flags,
800 async->bio_offset);
801}
802
803static void run_one_async_free(struct btrfs_work *work)
804{
805 struct async_submit_bio *async;
806
807 async = container_of(work, struct async_submit_bio, work);
808 kfree(async);
809}
810
811int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
812 int rw, struct bio *bio, int mirror_num,
813 unsigned long bio_flags,
814 u64 bio_offset,
815 extent_submit_bio_hook_t *submit_bio_start,
816 extent_submit_bio_hook_t *submit_bio_done)
817{
818 struct async_submit_bio *async;
819
820 async = kmalloc(sizeof(*async), GFP_NOFS);
821 if (!async)
822 return -ENOMEM;
823
824 async->inode = inode;
825 async->rw = rw;
826 async->bio = bio;
827 async->mirror_num = mirror_num;
828 async->submit_bio_start = submit_bio_start;
829 async->submit_bio_done = submit_bio_done;
830
831 btrfs_init_work(&async->work, run_one_async_start,
832 run_one_async_done, run_one_async_free);
833
834 async->bio_flags = bio_flags;
835 async->bio_offset = bio_offset;
836
837 async->error = 0;
838
839 atomic_inc(&fs_info->nr_async_submits);
840
841 if (rw & REQ_SYNC)
842 btrfs_set_work_high_priority(&async->work);
843
844 btrfs_queue_work(fs_info->workers, &async->work);
845
846 while (atomic_read(&fs_info->async_submit_draining) &&
847 atomic_read(&fs_info->nr_async_submits)) {
848 wait_event(fs_info->async_submit_wait,
849 (atomic_read(&fs_info->nr_async_submits) == 0));
850 }
851
852 return 0;
853}
854
855static int btree_csum_one_bio(struct bio *bio)
856{
857 struct bio_vec *bvec;
858 struct btrfs_root *root;
859 int i, ret = 0;
860
861 bio_for_each_segment_all(bvec, bio, i) {
862 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
863 ret = csum_dirty_buffer(root, bvec->bv_page);
864 if (ret)
865 break;
866 }
867
868 return ret;
869}
870
871static int __btree_submit_bio_start(struct inode *inode, int rw,
872 struct bio *bio, int mirror_num,
873 unsigned long bio_flags,
874 u64 bio_offset)
875{
876 /*
877 * when we're called for a write, we're already in the async
878 * submission context. Just jump into btrfs_map_bio
879 */
880 return btree_csum_one_bio(bio);
881}
882
883static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
884 int mirror_num, unsigned long bio_flags,
885 u64 bio_offset)
886{
887 int ret;
888
889 /*
890 * when we're called for a write, we're already in the async
891 * submission context. Just jump into btrfs_map_bio
892 */
893 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
894 if (ret)
895 bio_endio(bio, ret);
896 return ret;
897}
898
899static int check_async_write(struct inode *inode, unsigned long bio_flags)
900{
901 if (bio_flags & EXTENT_BIO_TREE_LOG)
902 return 0;
903#ifdef CONFIG_X86
904 if (cpu_has_xmm4_2)
905 return 0;
906#endif
907 return 1;
908}
909
910static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
911 int mirror_num, unsigned long bio_flags,
912 u64 bio_offset)
913{
914 int async = check_async_write(inode, bio_flags);
915 int ret;
916
917 if (!(rw & REQ_WRITE)) {
918 /*
919 * called for a read, do the setup so that checksum validation
920 * can happen in the async kernel threads
921 */
922 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
923 bio, 1);
924 if (ret)
925 goto out_w_error;
926 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
927 mirror_num, 0);
928 } else if (!async) {
929 ret = btree_csum_one_bio(bio);
930 if (ret)
931 goto out_w_error;
932 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
933 mirror_num, 0);
934 } else {
935 /*
936 * kthread helpers are used to submit writes so that
937 * checksumming can happen in parallel across all CPUs
938 */
939 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
940 inode, rw, bio, mirror_num, 0,
941 bio_offset,
942 __btree_submit_bio_start,
943 __btree_submit_bio_done);
944 }
945
946 if (ret) {
947out_w_error:
948 bio_endio(bio, ret);
949 }
950 return ret;
951}
952
953#ifdef CONFIG_MIGRATION
954static int btree_migratepage(struct address_space *mapping,
955 struct page *newpage, struct page *page,
956 enum migrate_mode mode)
957{
958 /*
959 * we can't safely write a btree page from here,
960 * we haven't done the locking hook
961 */
962 if (PageDirty(page))
963 return -EAGAIN;
964 /*
965 * Buffers may be managed in a filesystem specific way.
966 * We must have no buffers or drop them.
967 */
968 if (page_has_private(page) &&
969 !try_to_release_page(page, GFP_KERNEL))
970 return -EAGAIN;
971 return migrate_page(mapping, newpage, page, mode);
972}
973#endif
974
975
976static int btree_writepages(struct address_space *mapping,
977 struct writeback_control *wbc)
978{
979 struct btrfs_fs_info *fs_info;
980 int ret;
981
982 if (wbc->sync_mode == WB_SYNC_NONE) {
983
984 if (wbc->for_kupdate)
985 return 0;
986
987 fs_info = BTRFS_I(mapping->host)->root->fs_info;
988 /* this is a bit racy, but that's ok */
989 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
990 BTRFS_DIRTY_METADATA_THRESH);
991 if (ret < 0)
992 return 0;
993 }
994 return btree_write_cache_pages(mapping, wbc);
995}
996
997static int btree_readpage(struct file *file, struct page *page)
998{
999 struct extent_io_tree *tree;
1000 tree = &BTRFS_I(page->mapping->host)->io_tree;
1001 return extent_read_full_page(tree, page, btree_get_extent, 0);
1002}
1003
1004static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1005{
1006 if (PageWriteback(page) || PageDirty(page))
1007 return 0;
1008
1009 return try_release_extent_buffer(page);
1010}
1011
1012static void btree_invalidatepage(struct page *page, unsigned int offset,
1013 unsigned int length)
1014{
1015 struct extent_io_tree *tree;
1016 tree = &BTRFS_I(page->mapping->host)->io_tree;
1017 extent_invalidatepage(tree, page, offset);
1018 btree_releasepage(page, GFP_NOFS);
1019 if (PagePrivate(page)) {
1020 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1021 "page private not zero on page %llu",
1022 (unsigned long long)page_offset(page));
1023 ClearPagePrivate(page);
1024 set_page_private(page, 0);
1025 page_cache_release(page);
1026 }
1027}
1028
1029static int btree_set_page_dirty(struct page *page)
1030{
1031#ifdef DEBUG
1032 struct extent_buffer *eb;
1033
1034 BUG_ON(!PagePrivate(page));
1035 eb = (struct extent_buffer *)page->private;
1036 BUG_ON(!eb);
1037 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1038 BUG_ON(!atomic_read(&eb->refs));
1039 btrfs_assert_tree_locked(eb);
1040#endif
1041 return __set_page_dirty_nobuffers(page);
1042}
1043
1044static const struct address_space_operations btree_aops = {
1045 .readpage = btree_readpage,
1046 .writepages = btree_writepages,
1047 .releasepage = btree_releasepage,
1048 .invalidatepage = btree_invalidatepage,
1049#ifdef CONFIG_MIGRATION
1050 .migratepage = btree_migratepage,
1051#endif
1052 .set_page_dirty = btree_set_page_dirty,
1053};
1054
1055int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1056 u64 parent_transid)
1057{
1058 struct extent_buffer *buf = NULL;
1059 struct inode *btree_inode = root->fs_info->btree_inode;
1060 int ret = 0;
1061
1062 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1063 if (!buf)
1064 return 0;
1065 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1066 buf, 0, WAIT_NONE, btree_get_extent, 0);
1067 free_extent_buffer(buf);
1068 return ret;
1069}
1070
1071int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1072 int mirror_num, struct extent_buffer **eb)
1073{
1074 struct extent_buffer *buf = NULL;
1075 struct inode *btree_inode = root->fs_info->btree_inode;
1076 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1077 int ret;
1078
1079 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1080 if (!buf)
1081 return 0;
1082
1083 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1084
1085 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1086 btree_get_extent, mirror_num);
1087 if (ret) {
1088 free_extent_buffer(buf);
1089 return ret;
1090 }
1091
1092 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1093 free_extent_buffer(buf);
1094 return -EIO;
1095 } else if (extent_buffer_uptodate(buf)) {
1096 *eb = buf;
1097 } else {
1098 free_extent_buffer(buf);
1099 }
1100 return 0;
1101}
1102
1103struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1104 u64 bytenr, u32 blocksize)
1105{
1106 return find_extent_buffer(root->fs_info, bytenr);
1107}
1108
1109struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1110 u64 bytenr, u32 blocksize)
1111{
1112 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1113}
1114
1115
1116int btrfs_write_tree_block(struct extent_buffer *buf)
1117{
1118 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1119 buf->start + buf->len - 1);
1120}
1121
1122int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1123{
1124 return filemap_fdatawait_range(buf->pages[0]->mapping,
1125 buf->start, buf->start + buf->len - 1);
1126}
1127
1128struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1129 u32 blocksize, u64 parent_transid)
1130{
1131 struct extent_buffer *buf = NULL;
1132 int ret;
1133
1134 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1135 if (!buf)
1136 return NULL;
1137
1138 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1139 if (ret) {
1140 free_extent_buffer(buf);
1141 return NULL;
1142 }
1143 return buf;
1144
1145}
1146
1147void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1148 struct extent_buffer *buf)
1149{
1150 struct btrfs_fs_info *fs_info = root->fs_info;
1151
1152 if (btrfs_header_generation(buf) ==
1153 fs_info->running_transaction->transid) {
1154 btrfs_assert_tree_locked(buf);
1155
1156 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1157 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1158 -buf->len,
1159 fs_info->dirty_metadata_batch);
1160 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1161 btrfs_set_lock_blocking(buf);
1162 clear_extent_buffer_dirty(buf);
1163 }
1164 }
1165}
1166
1167static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1168{
1169 struct btrfs_subvolume_writers *writers;
1170 int ret;
1171
1172 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1173 if (!writers)
1174 return ERR_PTR(-ENOMEM);
1175
1176 ret = percpu_counter_init(&writers->counter, 0);
1177 if (ret < 0) {
1178 kfree(writers);
1179 return ERR_PTR(ret);
1180 }
1181
1182 init_waitqueue_head(&writers->wait);
1183 return writers;
1184}
1185
1186static void
1187btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1188{
1189 percpu_counter_destroy(&writers->counter);
1190 kfree(writers);
1191}
1192
1193static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1194 u32 stripesize, struct btrfs_root *root,
1195 struct btrfs_fs_info *fs_info,
1196 u64 objectid)
1197{
1198 root->node = NULL;
1199 root->commit_root = NULL;
1200 root->sectorsize = sectorsize;
1201 root->nodesize = nodesize;
1202 root->leafsize = leafsize;
1203 root->stripesize = stripesize;
1204 root->ref_cows = 0;
1205 root->track_dirty = 0;
1206 root->in_radix = 0;
1207 root->orphan_item_inserted = 0;
1208 root->orphan_cleanup_state = 0;
1209
1210 root->objectid = objectid;
1211 root->last_trans = 0;
1212 root->highest_objectid = 0;
1213 root->nr_delalloc_inodes = 0;
1214 root->nr_ordered_extents = 0;
1215 root->name = NULL;
1216 root->inode_tree = RB_ROOT;
1217 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1218 root->block_rsv = NULL;
1219 root->orphan_block_rsv = NULL;
1220
1221 INIT_LIST_HEAD(&root->dirty_list);
1222 INIT_LIST_HEAD(&root->root_list);
1223 INIT_LIST_HEAD(&root->delalloc_inodes);
1224 INIT_LIST_HEAD(&root->delalloc_root);
1225 INIT_LIST_HEAD(&root->ordered_extents);
1226 INIT_LIST_HEAD(&root->ordered_root);
1227 INIT_LIST_HEAD(&root->logged_list[0]);
1228 INIT_LIST_HEAD(&root->logged_list[1]);
1229 spin_lock_init(&root->orphan_lock);
1230 spin_lock_init(&root->inode_lock);
1231 spin_lock_init(&root->delalloc_lock);
1232 spin_lock_init(&root->ordered_extent_lock);
1233 spin_lock_init(&root->accounting_lock);
1234 spin_lock_init(&root->log_extents_lock[0]);
1235 spin_lock_init(&root->log_extents_lock[1]);
1236 mutex_init(&root->objectid_mutex);
1237 mutex_init(&root->log_mutex);
1238 mutex_init(&root->ordered_extent_mutex);
1239 mutex_init(&root->delalloc_mutex);
1240 init_waitqueue_head(&root->log_writer_wait);
1241 init_waitqueue_head(&root->log_commit_wait[0]);
1242 init_waitqueue_head(&root->log_commit_wait[1]);
1243 INIT_LIST_HEAD(&root->log_ctxs[0]);
1244 INIT_LIST_HEAD(&root->log_ctxs[1]);
1245 atomic_set(&root->log_commit[0], 0);
1246 atomic_set(&root->log_commit[1], 0);
1247 atomic_set(&root->log_writers, 0);
1248 atomic_set(&root->log_batch, 0);
1249 atomic_set(&root->orphan_inodes, 0);
1250 atomic_set(&root->refs, 1);
1251 atomic_set(&root->will_be_snapshoted, 0);
1252 root->log_transid = 0;
1253 root->log_transid_committed = -1;
1254 root->last_log_commit = 0;
1255 if (fs_info)
1256 extent_io_tree_init(&root->dirty_log_pages,
1257 fs_info->btree_inode->i_mapping);
1258
1259 memset(&root->root_key, 0, sizeof(root->root_key));
1260 memset(&root->root_item, 0, sizeof(root->root_item));
1261 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1262 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1263 if (fs_info)
1264 root->defrag_trans_start = fs_info->generation;
1265 else
1266 root->defrag_trans_start = 0;
1267 init_completion(&root->kobj_unregister);
1268 root->defrag_running = 0;
1269 root->root_key.objectid = objectid;
1270 root->anon_dev = 0;
1271
1272 spin_lock_init(&root->root_item_lock);
1273}
1274
1275static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1276{
1277 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1278 if (root)
1279 root->fs_info = fs_info;
1280 return root;
1281}
1282
1283#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1284/* Should only be used by the testing infrastructure */
1285struct btrfs_root *btrfs_alloc_dummy_root(void)
1286{
1287 struct btrfs_root *root;
1288
1289 root = btrfs_alloc_root(NULL);
1290 if (!root)
1291 return ERR_PTR(-ENOMEM);
1292 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1293 root->dummy_root = 1;
1294
1295 return root;
1296}
1297#endif
1298
1299struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1300 struct btrfs_fs_info *fs_info,
1301 u64 objectid)
1302{
1303 struct extent_buffer *leaf;
1304 struct btrfs_root *tree_root = fs_info->tree_root;
1305 struct btrfs_root *root;
1306 struct btrfs_key key;
1307 int ret = 0;
1308 uuid_le uuid;
1309
1310 root = btrfs_alloc_root(fs_info);
1311 if (!root)
1312 return ERR_PTR(-ENOMEM);
1313
1314 __setup_root(tree_root->nodesize, tree_root->leafsize,
1315 tree_root->sectorsize, tree_root->stripesize,
1316 root, fs_info, objectid);
1317 root->root_key.objectid = objectid;
1318 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1319 root->root_key.offset = 0;
1320
1321 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1322 0, objectid, NULL, 0, 0, 0);
1323 if (IS_ERR(leaf)) {
1324 ret = PTR_ERR(leaf);
1325 leaf = NULL;
1326 goto fail;
1327 }
1328
1329 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1330 btrfs_set_header_bytenr(leaf, leaf->start);
1331 btrfs_set_header_generation(leaf, trans->transid);
1332 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1333 btrfs_set_header_owner(leaf, objectid);
1334 root->node = leaf;
1335
1336 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1337 BTRFS_FSID_SIZE);
1338 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1339 btrfs_header_chunk_tree_uuid(leaf),
1340 BTRFS_UUID_SIZE);
1341 btrfs_mark_buffer_dirty(leaf);
1342
1343 root->commit_root = btrfs_root_node(root);
1344 root->track_dirty = 1;
1345
1346
1347 root->root_item.flags = 0;
1348 root->root_item.byte_limit = 0;
1349 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1350 btrfs_set_root_generation(&root->root_item, trans->transid);
1351 btrfs_set_root_level(&root->root_item, 0);
1352 btrfs_set_root_refs(&root->root_item, 1);
1353 btrfs_set_root_used(&root->root_item, leaf->len);
1354 btrfs_set_root_last_snapshot(&root->root_item, 0);
1355 btrfs_set_root_dirid(&root->root_item, 0);
1356 uuid_le_gen(&uuid);
1357 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1358 root->root_item.drop_level = 0;
1359
1360 key.objectid = objectid;
1361 key.type = BTRFS_ROOT_ITEM_KEY;
1362 key.offset = 0;
1363 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1364 if (ret)
1365 goto fail;
1366
1367 btrfs_tree_unlock(leaf);
1368
1369 return root;
1370
1371fail:
1372 if (leaf) {
1373 btrfs_tree_unlock(leaf);
1374 free_extent_buffer(leaf);
1375 }
1376 kfree(root);
1377
1378 return ERR_PTR(ret);
1379}
1380
1381static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1382 struct btrfs_fs_info *fs_info)
1383{
1384 struct btrfs_root *root;
1385 struct btrfs_root *tree_root = fs_info->tree_root;
1386 struct extent_buffer *leaf;
1387
1388 root = btrfs_alloc_root(fs_info);
1389 if (!root)
1390 return ERR_PTR(-ENOMEM);
1391
1392 __setup_root(tree_root->nodesize, tree_root->leafsize,
1393 tree_root->sectorsize, tree_root->stripesize,
1394 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1395
1396 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1397 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1398 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1399 /*
1400 * log trees do not get reference counted because they go away
1401 * before a real commit is actually done. They do store pointers
1402 * to file data extents, and those reference counts still get
1403 * updated (along with back refs to the log tree).
1404 */
1405 root->ref_cows = 0;
1406
1407 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1408 BTRFS_TREE_LOG_OBJECTID, NULL,
1409 0, 0, 0);
1410 if (IS_ERR(leaf)) {
1411 kfree(root);
1412 return ERR_CAST(leaf);
1413 }
1414
1415 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1416 btrfs_set_header_bytenr(leaf, leaf->start);
1417 btrfs_set_header_generation(leaf, trans->transid);
1418 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1419 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1420 root->node = leaf;
1421
1422 write_extent_buffer(root->node, root->fs_info->fsid,
1423 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1424 btrfs_mark_buffer_dirty(root->node);
1425 btrfs_tree_unlock(root->node);
1426 return root;
1427}
1428
1429int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1430 struct btrfs_fs_info *fs_info)
1431{
1432 struct btrfs_root *log_root;
1433
1434 log_root = alloc_log_tree(trans, fs_info);
1435 if (IS_ERR(log_root))
1436 return PTR_ERR(log_root);
1437 WARN_ON(fs_info->log_root_tree);
1438 fs_info->log_root_tree = log_root;
1439 return 0;
1440}
1441
1442int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1443 struct btrfs_root *root)
1444{
1445 struct btrfs_root *log_root;
1446 struct btrfs_inode_item *inode_item;
1447
1448 log_root = alloc_log_tree(trans, root->fs_info);
1449 if (IS_ERR(log_root))
1450 return PTR_ERR(log_root);
1451
1452 log_root->last_trans = trans->transid;
1453 log_root->root_key.offset = root->root_key.objectid;
1454
1455 inode_item = &log_root->root_item.inode;
1456 btrfs_set_stack_inode_generation(inode_item, 1);
1457 btrfs_set_stack_inode_size(inode_item, 3);
1458 btrfs_set_stack_inode_nlink(inode_item, 1);
1459 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1460 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1461
1462 btrfs_set_root_node(&log_root->root_item, log_root->node);
1463
1464 WARN_ON(root->log_root);
1465 root->log_root = log_root;
1466 root->log_transid = 0;
1467 root->log_transid_committed = -1;
1468 root->last_log_commit = 0;
1469 return 0;
1470}
1471
1472static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1473 struct btrfs_key *key)
1474{
1475 struct btrfs_root *root;
1476 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1477 struct btrfs_path *path;
1478 u64 generation;
1479 u32 blocksize;
1480 int ret;
1481
1482 path = btrfs_alloc_path();
1483 if (!path)
1484 return ERR_PTR(-ENOMEM);
1485
1486 root = btrfs_alloc_root(fs_info);
1487 if (!root) {
1488 ret = -ENOMEM;
1489 goto alloc_fail;
1490 }
1491
1492 __setup_root(tree_root->nodesize, tree_root->leafsize,
1493 tree_root->sectorsize, tree_root->stripesize,
1494 root, fs_info, key->objectid);
1495
1496 ret = btrfs_find_root(tree_root, key, path,
1497 &root->root_item, &root->root_key);
1498 if (ret) {
1499 if (ret > 0)
1500 ret = -ENOENT;
1501 goto find_fail;
1502 }
1503
1504 generation = btrfs_root_generation(&root->root_item);
1505 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1506 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1507 blocksize, generation);
1508 if (!root->node) {
1509 ret = -ENOMEM;
1510 goto find_fail;
1511 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1512 ret = -EIO;
1513 goto read_fail;
1514 }
1515 root->commit_root = btrfs_root_node(root);
1516out:
1517 btrfs_free_path(path);
1518 return root;
1519
1520read_fail:
1521 free_extent_buffer(root->node);
1522find_fail:
1523 kfree(root);
1524alloc_fail:
1525 root = ERR_PTR(ret);
1526 goto out;
1527}
1528
1529struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1530 struct btrfs_key *location)
1531{
1532 struct btrfs_root *root;
1533
1534 root = btrfs_read_tree_root(tree_root, location);
1535 if (IS_ERR(root))
1536 return root;
1537
1538 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1539 root->ref_cows = 1;
1540 btrfs_check_and_init_root_item(&root->root_item);
1541 }
1542
1543 return root;
1544}
1545
1546int btrfs_init_fs_root(struct btrfs_root *root)
1547{
1548 int ret;
1549 struct btrfs_subvolume_writers *writers;
1550
1551 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1552 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1553 GFP_NOFS);
1554 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1555 ret = -ENOMEM;
1556 goto fail;
1557 }
1558
1559 writers = btrfs_alloc_subvolume_writers();
1560 if (IS_ERR(writers)) {
1561 ret = PTR_ERR(writers);
1562 goto fail;
1563 }
1564 root->subv_writers = writers;
1565
1566 btrfs_init_free_ino_ctl(root);
1567 spin_lock_init(&root->cache_lock);
1568 init_waitqueue_head(&root->cache_wait);
1569
1570 ret = get_anon_bdev(&root->anon_dev);
1571 if (ret)
1572 goto free_writers;
1573 return 0;
1574
1575free_writers:
1576 btrfs_free_subvolume_writers(root->subv_writers);
1577fail:
1578 kfree(root->free_ino_ctl);
1579 kfree(root->free_ino_pinned);
1580 return ret;
1581}
1582
1583static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1584 u64 root_id)
1585{
1586 struct btrfs_root *root;
1587
1588 spin_lock(&fs_info->fs_roots_radix_lock);
1589 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1590 (unsigned long)root_id);
1591 spin_unlock(&fs_info->fs_roots_radix_lock);
1592 return root;
1593}
1594
1595int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1596 struct btrfs_root *root)
1597{
1598 int ret;
1599
1600 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1601 if (ret)
1602 return ret;
1603
1604 spin_lock(&fs_info->fs_roots_radix_lock);
1605 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1606 (unsigned long)root->root_key.objectid,
1607 root);
1608 if (ret == 0)
1609 root->in_radix = 1;
1610 spin_unlock(&fs_info->fs_roots_radix_lock);
1611 radix_tree_preload_end();
1612
1613 return ret;
1614}
1615
1616struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1617 struct btrfs_key *location,
1618 bool check_ref)
1619{
1620 struct btrfs_root *root;
1621 int ret;
1622
1623 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1624 return fs_info->tree_root;
1625 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1626 return fs_info->extent_root;
1627 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1628 return fs_info->chunk_root;
1629 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1630 return fs_info->dev_root;
1631 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1632 return fs_info->csum_root;
1633 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1634 return fs_info->quota_root ? fs_info->quota_root :
1635 ERR_PTR(-ENOENT);
1636 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1637 return fs_info->uuid_root ? fs_info->uuid_root :
1638 ERR_PTR(-ENOENT);
1639again:
1640 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1641 if (root) {
1642 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1643 return ERR_PTR(-ENOENT);
1644 return root;
1645 }
1646
1647 root = btrfs_read_fs_root(fs_info->tree_root, location);
1648 if (IS_ERR(root))
1649 return root;
1650
1651 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1652 ret = -ENOENT;
1653 goto fail;
1654 }
1655
1656 ret = btrfs_init_fs_root(root);
1657 if (ret)
1658 goto fail;
1659
1660 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1661 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1662 if (ret < 0)
1663 goto fail;
1664 if (ret == 0)
1665 root->orphan_item_inserted = 1;
1666
1667 ret = btrfs_insert_fs_root(fs_info, root);
1668 if (ret) {
1669 if (ret == -EEXIST) {
1670 free_fs_root(root);
1671 goto again;
1672 }
1673 goto fail;
1674 }
1675 return root;
1676fail:
1677 free_fs_root(root);
1678 return ERR_PTR(ret);
1679}
1680
1681static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1682{
1683 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1684 int ret = 0;
1685 struct btrfs_device *device;
1686 struct backing_dev_info *bdi;
1687
1688 rcu_read_lock();
1689 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1690 if (!device->bdev)
1691 continue;
1692 bdi = blk_get_backing_dev_info(device->bdev);
1693 if (bdi && bdi_congested(bdi, bdi_bits)) {
1694 ret = 1;
1695 break;
1696 }
1697 }
1698 rcu_read_unlock();
1699 return ret;
1700}
1701
1702/*
1703 * If this fails, caller must call bdi_destroy() to get rid of the
1704 * bdi again.
1705 */
1706static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1707{
1708 int err;
1709
1710 bdi->capabilities = BDI_CAP_MAP_COPY;
1711 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1712 if (err)
1713 return err;
1714
1715 bdi->ra_pages = default_backing_dev_info.ra_pages;
1716 bdi->congested_fn = btrfs_congested_fn;
1717 bdi->congested_data = info;
1718 return 0;
1719}
1720
1721/*
1722 * called by the kthread helper functions to finally call the bio end_io
1723 * functions. This is where read checksum verification actually happens
1724 */
1725static void end_workqueue_fn(struct btrfs_work *work)
1726{
1727 struct bio *bio;
1728 struct end_io_wq *end_io_wq;
1729 int error;
1730
1731 end_io_wq = container_of(work, struct end_io_wq, work);
1732 bio = end_io_wq->bio;
1733
1734 error = end_io_wq->error;
1735 bio->bi_private = end_io_wq->private;
1736 bio->bi_end_io = end_io_wq->end_io;
1737 kfree(end_io_wq);
1738 bio_endio_nodec(bio, error);
1739}
1740
1741static int cleaner_kthread(void *arg)
1742{
1743 struct btrfs_root *root = arg;
1744 int again;
1745
1746 do {
1747 again = 0;
1748
1749 /* Make the cleaner go to sleep early. */
1750 if (btrfs_need_cleaner_sleep(root))
1751 goto sleep;
1752
1753 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1754 goto sleep;
1755
1756 /*
1757 * Avoid the problem that we change the status of the fs
1758 * during the above check and trylock.
1759 */
1760 if (btrfs_need_cleaner_sleep(root)) {
1761 mutex_unlock(&root->fs_info->cleaner_mutex);
1762 goto sleep;
1763 }
1764
1765 btrfs_run_delayed_iputs(root);
1766 again = btrfs_clean_one_deleted_snapshot(root);
1767 mutex_unlock(&root->fs_info->cleaner_mutex);
1768
1769 /*
1770 * The defragger has dealt with the R/O remount and umount,
1771 * needn't do anything special here.
1772 */
1773 btrfs_run_defrag_inodes(root->fs_info);
1774sleep:
1775 if (!try_to_freeze() && !again) {
1776 set_current_state(TASK_INTERRUPTIBLE);
1777 if (!kthread_should_stop())
1778 schedule();
1779 __set_current_state(TASK_RUNNING);
1780 }
1781 } while (!kthread_should_stop());
1782 return 0;
1783}
1784
1785static int transaction_kthread(void *arg)
1786{
1787 struct btrfs_root *root = arg;
1788 struct btrfs_trans_handle *trans;
1789 struct btrfs_transaction *cur;
1790 u64 transid;
1791 unsigned long now;
1792 unsigned long delay;
1793 bool cannot_commit;
1794
1795 do {
1796 cannot_commit = false;
1797 delay = HZ * root->fs_info->commit_interval;
1798 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1799
1800 spin_lock(&root->fs_info->trans_lock);
1801 cur = root->fs_info->running_transaction;
1802 if (!cur) {
1803 spin_unlock(&root->fs_info->trans_lock);
1804 goto sleep;
1805 }
1806
1807 now = get_seconds();
1808 if (cur->state < TRANS_STATE_BLOCKED &&
1809 (now < cur->start_time ||
1810 now - cur->start_time < root->fs_info->commit_interval)) {
1811 spin_unlock(&root->fs_info->trans_lock);
1812 delay = HZ * 5;
1813 goto sleep;
1814 }
1815 transid = cur->transid;
1816 spin_unlock(&root->fs_info->trans_lock);
1817
1818 /* If the file system is aborted, this will always fail. */
1819 trans = btrfs_attach_transaction(root);
1820 if (IS_ERR(trans)) {
1821 if (PTR_ERR(trans) != -ENOENT)
1822 cannot_commit = true;
1823 goto sleep;
1824 }
1825 if (transid == trans->transid) {
1826 btrfs_commit_transaction(trans, root);
1827 } else {
1828 btrfs_end_transaction(trans, root);
1829 }
1830sleep:
1831 wake_up_process(root->fs_info->cleaner_kthread);
1832 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1833
1834 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1835 &root->fs_info->fs_state)))
1836 btrfs_cleanup_transaction(root);
1837 if (!try_to_freeze()) {
1838 set_current_state(TASK_INTERRUPTIBLE);
1839 if (!kthread_should_stop() &&
1840 (!btrfs_transaction_blocked(root->fs_info) ||
1841 cannot_commit))
1842 schedule_timeout(delay);
1843 __set_current_state(TASK_RUNNING);
1844 }
1845 } while (!kthread_should_stop());
1846 return 0;
1847}
1848
1849/*
1850 * this will find the highest generation in the array of
1851 * root backups. The index of the highest array is returned,
1852 * or -1 if we can't find anything.
1853 *
1854 * We check to make sure the array is valid by comparing the
1855 * generation of the latest root in the array with the generation
1856 * in the super block. If they don't match we pitch it.
1857 */
1858static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1859{
1860 u64 cur;
1861 int newest_index = -1;
1862 struct btrfs_root_backup *root_backup;
1863 int i;
1864
1865 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1866 root_backup = info->super_copy->super_roots + i;
1867 cur = btrfs_backup_tree_root_gen(root_backup);
1868 if (cur == newest_gen)
1869 newest_index = i;
1870 }
1871
1872 /* check to see if we actually wrapped around */
1873 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1874 root_backup = info->super_copy->super_roots;
1875 cur = btrfs_backup_tree_root_gen(root_backup);
1876 if (cur == newest_gen)
1877 newest_index = 0;
1878 }
1879 return newest_index;
1880}
1881
1882
1883/*
1884 * find the oldest backup so we know where to store new entries
1885 * in the backup array. This will set the backup_root_index
1886 * field in the fs_info struct
1887 */
1888static void find_oldest_super_backup(struct btrfs_fs_info *info,
1889 u64 newest_gen)
1890{
1891 int newest_index = -1;
1892
1893 newest_index = find_newest_super_backup(info, newest_gen);
1894 /* if there was garbage in there, just move along */
1895 if (newest_index == -1) {
1896 info->backup_root_index = 0;
1897 } else {
1898 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1899 }
1900}
1901
1902/*
1903 * copy all the root pointers into the super backup array.
1904 * this will bump the backup pointer by one when it is
1905 * done
1906 */
1907static void backup_super_roots(struct btrfs_fs_info *info)
1908{
1909 int next_backup;
1910 struct btrfs_root_backup *root_backup;
1911 int last_backup;
1912
1913 next_backup = info->backup_root_index;
1914 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1915 BTRFS_NUM_BACKUP_ROOTS;
1916
1917 /*
1918 * just overwrite the last backup if we're at the same generation
1919 * this happens only at umount
1920 */
1921 root_backup = info->super_for_commit->super_roots + last_backup;
1922 if (btrfs_backup_tree_root_gen(root_backup) ==
1923 btrfs_header_generation(info->tree_root->node))
1924 next_backup = last_backup;
1925
1926 root_backup = info->super_for_commit->super_roots + next_backup;
1927
1928 /*
1929 * make sure all of our padding and empty slots get zero filled
1930 * regardless of which ones we use today
1931 */
1932 memset(root_backup, 0, sizeof(*root_backup));
1933
1934 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1935
1936 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1937 btrfs_set_backup_tree_root_gen(root_backup,
1938 btrfs_header_generation(info->tree_root->node));
1939
1940 btrfs_set_backup_tree_root_level(root_backup,
1941 btrfs_header_level(info->tree_root->node));
1942
1943 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1944 btrfs_set_backup_chunk_root_gen(root_backup,
1945 btrfs_header_generation(info->chunk_root->node));
1946 btrfs_set_backup_chunk_root_level(root_backup,
1947 btrfs_header_level(info->chunk_root->node));
1948
1949 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1950 btrfs_set_backup_extent_root_gen(root_backup,
1951 btrfs_header_generation(info->extent_root->node));
1952 btrfs_set_backup_extent_root_level(root_backup,
1953 btrfs_header_level(info->extent_root->node));
1954
1955 /*
1956 * we might commit during log recovery, which happens before we set
1957 * the fs_root. Make sure it is valid before we fill it in.
1958 */
1959 if (info->fs_root && info->fs_root->node) {
1960 btrfs_set_backup_fs_root(root_backup,
1961 info->fs_root->node->start);
1962 btrfs_set_backup_fs_root_gen(root_backup,
1963 btrfs_header_generation(info->fs_root->node));
1964 btrfs_set_backup_fs_root_level(root_backup,
1965 btrfs_header_level(info->fs_root->node));
1966 }
1967
1968 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1969 btrfs_set_backup_dev_root_gen(root_backup,
1970 btrfs_header_generation(info->dev_root->node));
1971 btrfs_set_backup_dev_root_level(root_backup,
1972 btrfs_header_level(info->dev_root->node));
1973
1974 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1975 btrfs_set_backup_csum_root_gen(root_backup,
1976 btrfs_header_generation(info->csum_root->node));
1977 btrfs_set_backup_csum_root_level(root_backup,
1978 btrfs_header_level(info->csum_root->node));
1979
1980 btrfs_set_backup_total_bytes(root_backup,
1981 btrfs_super_total_bytes(info->super_copy));
1982 btrfs_set_backup_bytes_used(root_backup,
1983 btrfs_super_bytes_used(info->super_copy));
1984 btrfs_set_backup_num_devices(root_backup,
1985 btrfs_super_num_devices(info->super_copy));
1986
1987 /*
1988 * if we don't copy this out to the super_copy, it won't get remembered
1989 * for the next commit
1990 */
1991 memcpy(&info->super_copy->super_roots,
1992 &info->super_for_commit->super_roots,
1993 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1994}
1995
1996/*
1997 * this copies info out of the root backup array and back into
1998 * the in-memory super block. It is meant to help iterate through
1999 * the array, so you send it the number of backups you've already
2000 * tried and the last backup index you used.
2001 *
2002 * this returns -1 when it has tried all the backups
2003 */
2004static noinline int next_root_backup(struct btrfs_fs_info *info,
2005 struct btrfs_super_block *super,
2006 int *num_backups_tried, int *backup_index)
2007{
2008 struct btrfs_root_backup *root_backup;
2009 int newest = *backup_index;
2010
2011 if (*num_backups_tried == 0) {
2012 u64 gen = btrfs_super_generation(super);
2013
2014 newest = find_newest_super_backup(info, gen);
2015 if (newest == -1)
2016 return -1;
2017
2018 *backup_index = newest;
2019 *num_backups_tried = 1;
2020 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2021 /* we've tried all the backups, all done */
2022 return -1;
2023 } else {
2024 /* jump to the next oldest backup */
2025 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2026 BTRFS_NUM_BACKUP_ROOTS;
2027 *backup_index = newest;
2028 *num_backups_tried += 1;
2029 }
2030 root_backup = super->super_roots + newest;
2031
2032 btrfs_set_super_generation(super,
2033 btrfs_backup_tree_root_gen(root_backup));
2034 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2035 btrfs_set_super_root_level(super,
2036 btrfs_backup_tree_root_level(root_backup));
2037 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2038
2039 /*
2040 * fixme: the total bytes and num_devices need to match or we should
2041 * need a fsck
2042 */
2043 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2044 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2045 return 0;
2046}
2047
2048/* helper to cleanup workers */
2049static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2050{
2051 btrfs_destroy_workqueue(fs_info->fixup_workers);
2052 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2053 btrfs_destroy_workqueue(fs_info->workers);
2054 btrfs_destroy_workqueue(fs_info->endio_workers);
2055 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2056 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2057 btrfs_destroy_workqueue(fs_info->rmw_workers);
2058 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2059 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2060 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2061 btrfs_destroy_workqueue(fs_info->submit_workers);
2062 btrfs_destroy_workqueue(fs_info->delayed_workers);
2063 btrfs_destroy_workqueue(fs_info->caching_workers);
2064 btrfs_destroy_workqueue(fs_info->readahead_workers);
2065 btrfs_destroy_workqueue(fs_info->flush_workers);
2066 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2067}
2068
2069static void free_root_extent_buffers(struct btrfs_root *root)
2070{
2071 if (root) {
2072 free_extent_buffer(root->node);
2073 free_extent_buffer(root->commit_root);
2074 root->node = NULL;
2075 root->commit_root = NULL;
2076 }
2077}
2078
2079/* helper to cleanup tree roots */
2080static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2081{
2082 free_root_extent_buffers(info->tree_root);
2083
2084 free_root_extent_buffers(info->dev_root);
2085 free_root_extent_buffers(info->extent_root);
2086 free_root_extent_buffers(info->csum_root);
2087 free_root_extent_buffers(info->quota_root);
2088 free_root_extent_buffers(info->uuid_root);
2089 if (chunk_root)
2090 free_root_extent_buffers(info->chunk_root);
2091}
2092
2093static void del_fs_roots(struct btrfs_fs_info *fs_info)
2094{
2095 int ret;
2096 struct btrfs_root *gang[8];
2097 int i;
2098
2099 while (!list_empty(&fs_info->dead_roots)) {
2100 gang[0] = list_entry(fs_info->dead_roots.next,
2101 struct btrfs_root, root_list);
2102 list_del(&gang[0]->root_list);
2103
2104 if (gang[0]->in_radix) {
2105 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2106 } else {
2107 free_extent_buffer(gang[0]->node);
2108 free_extent_buffer(gang[0]->commit_root);
2109 btrfs_put_fs_root(gang[0]);
2110 }
2111 }
2112
2113 while (1) {
2114 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2115 (void **)gang, 0,
2116 ARRAY_SIZE(gang));
2117 if (!ret)
2118 break;
2119 for (i = 0; i < ret; i++)
2120 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2121 }
2122
2123 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2124 btrfs_free_log_root_tree(NULL, fs_info);
2125 btrfs_destroy_pinned_extent(fs_info->tree_root,
2126 fs_info->pinned_extents);
2127 }
2128}
2129
2130int open_ctree(struct super_block *sb,
2131 struct btrfs_fs_devices *fs_devices,
2132 char *options)
2133{
2134 u32 sectorsize;
2135 u32 nodesize;
2136 u32 leafsize;
2137 u32 blocksize;
2138 u32 stripesize;
2139 u64 generation;
2140 u64 features;
2141 struct btrfs_key location;
2142 struct buffer_head *bh;
2143 struct btrfs_super_block *disk_super;
2144 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2145 struct btrfs_root *tree_root;
2146 struct btrfs_root *extent_root;
2147 struct btrfs_root *csum_root;
2148 struct btrfs_root *chunk_root;
2149 struct btrfs_root *dev_root;
2150 struct btrfs_root *quota_root;
2151 struct btrfs_root *uuid_root;
2152 struct btrfs_root *log_tree_root;
2153 int ret;
2154 int err = -EINVAL;
2155 int num_backups_tried = 0;
2156 int backup_index = 0;
2157 int max_active;
2158 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2159 bool create_uuid_tree;
2160 bool check_uuid_tree;
2161
2162 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2163 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2164 if (!tree_root || !chunk_root) {
2165 err = -ENOMEM;
2166 goto fail;
2167 }
2168
2169 ret = init_srcu_struct(&fs_info->subvol_srcu);
2170 if (ret) {
2171 err = ret;
2172 goto fail;
2173 }
2174
2175 ret = setup_bdi(fs_info, &fs_info->bdi);
2176 if (ret) {
2177 err = ret;
2178 goto fail_srcu;
2179 }
2180
2181 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2182 if (ret) {
2183 err = ret;
2184 goto fail_bdi;
2185 }
2186 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2187 (1 + ilog2(nr_cpu_ids));
2188
2189 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2190 if (ret) {
2191 err = ret;
2192 goto fail_dirty_metadata_bytes;
2193 }
2194
2195 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2196 if (ret) {
2197 err = ret;
2198 goto fail_delalloc_bytes;
2199 }
2200
2201 fs_info->btree_inode = new_inode(sb);
2202 if (!fs_info->btree_inode) {
2203 err = -ENOMEM;
2204 goto fail_bio_counter;
2205 }
2206
2207 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2208
2209 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2210 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2211 INIT_LIST_HEAD(&fs_info->trans_list);
2212 INIT_LIST_HEAD(&fs_info->dead_roots);
2213 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2214 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2215 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2216 spin_lock_init(&fs_info->delalloc_root_lock);
2217 spin_lock_init(&fs_info->trans_lock);
2218 spin_lock_init(&fs_info->fs_roots_radix_lock);
2219 spin_lock_init(&fs_info->delayed_iput_lock);
2220 spin_lock_init(&fs_info->defrag_inodes_lock);
2221 spin_lock_init(&fs_info->free_chunk_lock);
2222 spin_lock_init(&fs_info->tree_mod_seq_lock);
2223 spin_lock_init(&fs_info->super_lock);
2224 spin_lock_init(&fs_info->buffer_lock);
2225 rwlock_init(&fs_info->tree_mod_log_lock);
2226 mutex_init(&fs_info->reloc_mutex);
2227 mutex_init(&fs_info->delalloc_root_mutex);
2228 seqlock_init(&fs_info->profiles_lock);
2229
2230 init_completion(&fs_info->kobj_unregister);
2231 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2232 INIT_LIST_HEAD(&fs_info->space_info);
2233 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2234 btrfs_mapping_init(&fs_info->mapping_tree);
2235 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2236 BTRFS_BLOCK_RSV_GLOBAL);
2237 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2238 BTRFS_BLOCK_RSV_DELALLOC);
2239 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2240 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2241 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2242 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2243 BTRFS_BLOCK_RSV_DELOPS);
2244 atomic_set(&fs_info->nr_async_submits, 0);
2245 atomic_set(&fs_info->async_delalloc_pages, 0);
2246 atomic_set(&fs_info->async_submit_draining, 0);
2247 atomic_set(&fs_info->nr_async_bios, 0);
2248 atomic_set(&fs_info->defrag_running, 0);
2249 atomic64_set(&fs_info->tree_mod_seq, 0);
2250 fs_info->sb = sb;
2251 fs_info->max_inline = 8192 * 1024;
2252 fs_info->metadata_ratio = 0;
2253 fs_info->defrag_inodes = RB_ROOT;
2254 fs_info->free_chunk_space = 0;
2255 fs_info->tree_mod_log = RB_ROOT;
2256 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2257 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2258 /* readahead state */
2259 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2260 spin_lock_init(&fs_info->reada_lock);
2261
2262 fs_info->thread_pool_size = min_t(unsigned long,
2263 num_online_cpus() + 2, 8);
2264
2265 INIT_LIST_HEAD(&fs_info->ordered_roots);
2266 spin_lock_init(&fs_info->ordered_root_lock);
2267 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2268 GFP_NOFS);
2269 if (!fs_info->delayed_root) {
2270 err = -ENOMEM;
2271 goto fail_iput;
2272 }
2273 btrfs_init_delayed_root(fs_info->delayed_root);
2274
2275 mutex_init(&fs_info->scrub_lock);
2276 atomic_set(&fs_info->scrubs_running, 0);
2277 atomic_set(&fs_info->scrub_pause_req, 0);
2278 atomic_set(&fs_info->scrubs_paused, 0);
2279 atomic_set(&fs_info->scrub_cancel_req, 0);
2280 init_waitqueue_head(&fs_info->replace_wait);
2281 init_waitqueue_head(&fs_info->scrub_pause_wait);
2282 fs_info->scrub_workers_refcnt = 0;
2283#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2284 fs_info->check_integrity_print_mask = 0;
2285#endif
2286
2287 spin_lock_init(&fs_info->balance_lock);
2288 mutex_init(&fs_info->balance_mutex);
2289 atomic_set(&fs_info->balance_running, 0);
2290 atomic_set(&fs_info->balance_pause_req, 0);
2291 atomic_set(&fs_info->balance_cancel_req, 0);
2292 fs_info->balance_ctl = NULL;
2293 init_waitqueue_head(&fs_info->balance_wait_q);
2294
2295 sb->s_blocksize = 4096;
2296 sb->s_blocksize_bits = blksize_bits(4096);
2297 sb->s_bdi = &fs_info->bdi;
2298
2299 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2300 set_nlink(fs_info->btree_inode, 1);
2301 /*
2302 * we set the i_size on the btree inode to the max possible int.
2303 * the real end of the address space is determined by all of
2304 * the devices in the system
2305 */
2306 fs_info->btree_inode->i_size = OFFSET_MAX;
2307 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2308 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2309
2310 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2311 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2312 fs_info->btree_inode->i_mapping);
2313 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2314 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2315
2316 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2317
2318 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2319 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2320 sizeof(struct btrfs_key));
2321 set_bit(BTRFS_INODE_DUMMY,
2322 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2323 btrfs_insert_inode_hash(fs_info->btree_inode);
2324
2325 spin_lock_init(&fs_info->block_group_cache_lock);
2326 fs_info->block_group_cache_tree = RB_ROOT;
2327 fs_info->first_logical_byte = (u64)-1;
2328
2329 extent_io_tree_init(&fs_info->freed_extents[0],
2330 fs_info->btree_inode->i_mapping);
2331 extent_io_tree_init(&fs_info->freed_extents[1],
2332 fs_info->btree_inode->i_mapping);
2333 fs_info->pinned_extents = &fs_info->freed_extents[0];
2334 fs_info->do_barriers = 1;
2335
2336
2337 mutex_init(&fs_info->ordered_operations_mutex);
2338 mutex_init(&fs_info->ordered_extent_flush_mutex);
2339 mutex_init(&fs_info->tree_log_mutex);
2340 mutex_init(&fs_info->chunk_mutex);
2341 mutex_init(&fs_info->transaction_kthread_mutex);
2342 mutex_init(&fs_info->cleaner_mutex);
2343 mutex_init(&fs_info->volume_mutex);
2344 init_rwsem(&fs_info->commit_root_sem);
2345 init_rwsem(&fs_info->cleanup_work_sem);
2346 init_rwsem(&fs_info->subvol_sem);
2347 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2348 fs_info->dev_replace.lock_owner = 0;
2349 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2350 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2351 mutex_init(&fs_info->dev_replace.lock_management_lock);
2352 mutex_init(&fs_info->dev_replace.lock);
2353
2354 spin_lock_init(&fs_info->qgroup_lock);
2355 mutex_init(&fs_info->qgroup_ioctl_lock);
2356 fs_info->qgroup_tree = RB_ROOT;
2357 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2358 fs_info->qgroup_seq = 1;
2359 fs_info->quota_enabled = 0;
2360 fs_info->pending_quota_state = 0;
2361 fs_info->qgroup_ulist = NULL;
2362 mutex_init(&fs_info->qgroup_rescan_lock);
2363
2364 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2365 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2366
2367 init_waitqueue_head(&fs_info->transaction_throttle);
2368 init_waitqueue_head(&fs_info->transaction_wait);
2369 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2370 init_waitqueue_head(&fs_info->async_submit_wait);
2371
2372 ret = btrfs_alloc_stripe_hash_table(fs_info);
2373 if (ret) {
2374 err = ret;
2375 goto fail_alloc;
2376 }
2377
2378 __setup_root(4096, 4096, 4096, 4096, tree_root,
2379 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2380
2381 invalidate_bdev(fs_devices->latest_bdev);
2382
2383 /*
2384 * Read super block and check the signature bytes only
2385 */
2386 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2387 if (!bh) {
2388 err = -EINVAL;
2389 goto fail_alloc;
2390 }
2391
2392 /*
2393 * We want to check superblock checksum, the type is stored inside.
2394 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2395 */
2396 if (btrfs_check_super_csum(bh->b_data)) {
2397 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2398 err = -EINVAL;
2399 goto fail_alloc;
2400 }
2401
2402 /*
2403 * super_copy is zeroed at allocation time and we never touch the
2404 * following bytes up to INFO_SIZE, the checksum is calculated from
2405 * the whole block of INFO_SIZE
2406 */
2407 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2408 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2409 sizeof(*fs_info->super_for_commit));
2410 brelse(bh);
2411
2412 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2413
2414 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2415 if (ret) {
2416 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2417 err = -EINVAL;
2418 goto fail_alloc;
2419 }
2420
2421 disk_super = fs_info->super_copy;
2422 if (!btrfs_super_root(disk_super))
2423 goto fail_alloc;
2424
2425 /* check FS state, whether FS is broken. */
2426 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2427 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2428
2429 /*
2430 * run through our array of backup supers and setup
2431 * our ring pointer to the oldest one
2432 */
2433 generation = btrfs_super_generation(disk_super);
2434 find_oldest_super_backup(fs_info, generation);
2435
2436 /*
2437 * In the long term, we'll store the compression type in the super
2438 * block, and it'll be used for per file compression control.
2439 */
2440 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2441
2442 ret = btrfs_parse_options(tree_root, options);
2443 if (ret) {
2444 err = ret;
2445 goto fail_alloc;
2446 }
2447
2448 features = btrfs_super_incompat_flags(disk_super) &
2449 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2450 if (features) {
2451 printk(KERN_ERR "BTRFS: couldn't mount because of "
2452 "unsupported optional features (%Lx).\n",
2453 features);
2454 err = -EINVAL;
2455 goto fail_alloc;
2456 }
2457
2458 if (btrfs_super_leafsize(disk_super) !=
2459 btrfs_super_nodesize(disk_super)) {
2460 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2461 "blocksizes don't match. node %d leaf %d\n",
2462 btrfs_super_nodesize(disk_super),
2463 btrfs_super_leafsize(disk_super));
2464 err = -EINVAL;
2465 goto fail_alloc;
2466 }
2467 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2468 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2469 "blocksize (%d) was too large\n",
2470 btrfs_super_leafsize(disk_super));
2471 err = -EINVAL;
2472 goto fail_alloc;
2473 }
2474
2475 features = btrfs_super_incompat_flags(disk_super);
2476 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2477 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2478 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2479
2480 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2481 printk(KERN_ERR "BTRFS: has skinny extents\n");
2482
2483 /*
2484 * flag our filesystem as having big metadata blocks if
2485 * they are bigger than the page size
2486 */
2487 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2488 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2489 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2490 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2491 }
2492
2493 nodesize = btrfs_super_nodesize(disk_super);
2494 leafsize = btrfs_super_leafsize(disk_super);
2495 sectorsize = btrfs_super_sectorsize(disk_super);
2496 stripesize = btrfs_super_stripesize(disk_super);
2497 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2498 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2499
2500 /*
2501 * mixed block groups end up with duplicate but slightly offset
2502 * extent buffers for the same range. It leads to corruptions
2503 */
2504 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2505 (sectorsize != leafsize)) {
2506 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2507 "are not allowed for mixed block groups on %s\n",
2508 sb->s_id);
2509 goto fail_alloc;
2510 }
2511
2512 /*
2513 * Needn't use the lock because there is no other task which will
2514 * update the flag.
2515 */
2516 btrfs_set_super_incompat_flags(disk_super, features);
2517
2518 features = btrfs_super_compat_ro_flags(disk_super) &
2519 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2520 if (!(sb->s_flags & MS_RDONLY) && features) {
2521 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2522 "unsupported option features (%Lx).\n",
2523 features);
2524 err = -EINVAL;
2525 goto fail_alloc;
2526 }
2527
2528 max_active = fs_info->thread_pool_size;
2529
2530 fs_info->workers =
2531 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2532 max_active, 16);
2533
2534 fs_info->delalloc_workers =
2535 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2536
2537 fs_info->flush_workers =
2538 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2539
2540 fs_info->caching_workers =
2541 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2542
2543 /*
2544 * a higher idle thresh on the submit workers makes it much more
2545 * likely that bios will be send down in a sane order to the
2546 * devices
2547 */
2548 fs_info->submit_workers =
2549 btrfs_alloc_workqueue("submit", flags,
2550 min_t(u64, fs_devices->num_devices,
2551 max_active), 64);
2552
2553 fs_info->fixup_workers =
2554 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2555
2556 /*
2557 * endios are largely parallel and should have a very
2558 * low idle thresh
2559 */
2560 fs_info->endio_workers =
2561 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2562 fs_info->endio_meta_workers =
2563 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2564 fs_info->endio_meta_write_workers =
2565 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2566 fs_info->endio_raid56_workers =
2567 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2568 fs_info->rmw_workers =
2569 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2570 fs_info->endio_write_workers =
2571 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2572 fs_info->endio_freespace_worker =
2573 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2574 fs_info->delayed_workers =
2575 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2576 fs_info->readahead_workers =
2577 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2578 fs_info->qgroup_rescan_workers =
2579 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2580
2581 if (!(fs_info->workers && fs_info->delalloc_workers &&
2582 fs_info->submit_workers && fs_info->flush_workers &&
2583 fs_info->endio_workers && fs_info->endio_meta_workers &&
2584 fs_info->endio_meta_write_workers &&
2585 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2586 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2587 fs_info->caching_workers && fs_info->readahead_workers &&
2588 fs_info->fixup_workers && fs_info->delayed_workers &&
2589 fs_info->qgroup_rescan_workers)) {
2590 err = -ENOMEM;
2591 goto fail_sb_buffer;
2592 }
2593
2594 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2595 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2596 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2597
2598 tree_root->nodesize = nodesize;
2599 tree_root->leafsize = leafsize;
2600 tree_root->sectorsize = sectorsize;
2601 tree_root->stripesize = stripesize;
2602
2603 sb->s_blocksize = sectorsize;
2604 sb->s_blocksize_bits = blksize_bits(sectorsize);
2605
2606 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2607 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2608 goto fail_sb_buffer;
2609 }
2610
2611 if (sectorsize != PAGE_SIZE) {
2612 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2613 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2614 goto fail_sb_buffer;
2615 }
2616
2617 mutex_lock(&fs_info->chunk_mutex);
2618 ret = btrfs_read_sys_array(tree_root);
2619 mutex_unlock(&fs_info->chunk_mutex);
2620 if (ret) {
2621 printk(KERN_WARNING "BTRFS: failed to read the system "
2622 "array on %s\n", sb->s_id);
2623 goto fail_sb_buffer;
2624 }
2625
2626 blocksize = btrfs_level_size(tree_root,
2627 btrfs_super_chunk_root_level(disk_super));
2628 generation = btrfs_super_chunk_root_generation(disk_super);
2629
2630 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2631 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2632
2633 chunk_root->node = read_tree_block(chunk_root,
2634 btrfs_super_chunk_root(disk_super),
2635 blocksize, generation);
2636 if (!chunk_root->node ||
2637 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2638 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2639 sb->s_id);
2640 goto fail_tree_roots;
2641 }
2642 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2643 chunk_root->commit_root = btrfs_root_node(chunk_root);
2644
2645 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2646 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2647
2648 ret = btrfs_read_chunk_tree(chunk_root);
2649 if (ret) {
2650 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2651 sb->s_id);
2652 goto fail_tree_roots;
2653 }
2654
2655 /*
2656 * keep the device that is marked to be the target device for the
2657 * dev_replace procedure
2658 */
2659 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2660
2661 if (!fs_devices->latest_bdev) {
2662 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2663 sb->s_id);
2664 goto fail_tree_roots;
2665 }
2666
2667retry_root_backup:
2668 blocksize = btrfs_level_size(tree_root,
2669 btrfs_super_root_level(disk_super));
2670 generation = btrfs_super_generation(disk_super);
2671
2672 tree_root->node = read_tree_block(tree_root,
2673 btrfs_super_root(disk_super),
2674 blocksize, generation);
2675 if (!tree_root->node ||
2676 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2677 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2678 sb->s_id);
2679
2680 goto recovery_tree_root;
2681 }
2682
2683 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2684 tree_root->commit_root = btrfs_root_node(tree_root);
2685 btrfs_set_root_refs(&tree_root->root_item, 1);
2686
2687 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2688 location.type = BTRFS_ROOT_ITEM_KEY;
2689 location.offset = 0;
2690
2691 extent_root = btrfs_read_tree_root(tree_root, &location);
2692 if (IS_ERR(extent_root)) {
2693 ret = PTR_ERR(extent_root);
2694 goto recovery_tree_root;
2695 }
2696 extent_root->track_dirty = 1;
2697 fs_info->extent_root = extent_root;
2698
2699 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2700 dev_root = btrfs_read_tree_root(tree_root, &location);
2701 if (IS_ERR(dev_root)) {
2702 ret = PTR_ERR(dev_root);
2703 goto recovery_tree_root;
2704 }
2705 dev_root->track_dirty = 1;
2706 fs_info->dev_root = dev_root;
2707 btrfs_init_devices_late(fs_info);
2708
2709 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2710 csum_root = btrfs_read_tree_root(tree_root, &location);
2711 if (IS_ERR(csum_root)) {
2712 ret = PTR_ERR(csum_root);
2713 goto recovery_tree_root;
2714 }
2715 csum_root->track_dirty = 1;
2716 fs_info->csum_root = csum_root;
2717
2718 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2719 quota_root = btrfs_read_tree_root(tree_root, &location);
2720 if (!IS_ERR(quota_root)) {
2721 quota_root->track_dirty = 1;
2722 fs_info->quota_enabled = 1;
2723 fs_info->pending_quota_state = 1;
2724 fs_info->quota_root = quota_root;
2725 }
2726
2727 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2728 uuid_root = btrfs_read_tree_root(tree_root, &location);
2729 if (IS_ERR(uuid_root)) {
2730 ret = PTR_ERR(uuid_root);
2731 if (ret != -ENOENT)
2732 goto recovery_tree_root;
2733 create_uuid_tree = true;
2734 check_uuid_tree = false;
2735 } else {
2736 uuid_root->track_dirty = 1;
2737 fs_info->uuid_root = uuid_root;
2738 create_uuid_tree = false;
2739 check_uuid_tree =
2740 generation != btrfs_super_uuid_tree_generation(disk_super);
2741 }
2742
2743 fs_info->generation = generation;
2744 fs_info->last_trans_committed = generation;
2745
2746 ret = btrfs_recover_balance(fs_info);
2747 if (ret) {
2748 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2749 goto fail_block_groups;
2750 }
2751
2752 ret = btrfs_init_dev_stats(fs_info);
2753 if (ret) {
2754 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2755 ret);
2756 goto fail_block_groups;
2757 }
2758
2759 ret = btrfs_init_dev_replace(fs_info);
2760 if (ret) {
2761 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2762 goto fail_block_groups;
2763 }
2764
2765 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2766
2767 ret = btrfs_sysfs_add_one(fs_info);
2768 if (ret) {
2769 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2770 goto fail_block_groups;
2771 }
2772
2773 ret = btrfs_init_space_info(fs_info);
2774 if (ret) {
2775 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2776 goto fail_sysfs;
2777 }
2778
2779 ret = btrfs_read_block_groups(extent_root);
2780 if (ret) {
2781 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2782 goto fail_sysfs;
2783 }
2784 fs_info->num_tolerated_disk_barrier_failures =
2785 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2786 if (fs_info->fs_devices->missing_devices >
2787 fs_info->num_tolerated_disk_barrier_failures &&
2788 !(sb->s_flags & MS_RDONLY)) {
2789 printk(KERN_WARNING "BTRFS: "
2790 "too many missing devices, writeable mount is not allowed\n");
2791 goto fail_sysfs;
2792 }
2793
2794 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2795 "btrfs-cleaner");
2796 if (IS_ERR(fs_info->cleaner_kthread))
2797 goto fail_sysfs;
2798
2799 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2800 tree_root,
2801 "btrfs-transaction");
2802 if (IS_ERR(fs_info->transaction_kthread))
2803 goto fail_cleaner;
2804
2805 if (!btrfs_test_opt(tree_root, SSD) &&
2806 !btrfs_test_opt(tree_root, NOSSD) &&
2807 !fs_info->fs_devices->rotating) {
2808 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2809 "mode\n");
2810 btrfs_set_opt(fs_info->mount_opt, SSD);
2811 }
2812
2813 /* Set the real inode map cache flag */
2814 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2815 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2816
2817#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2818 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2819 ret = btrfsic_mount(tree_root, fs_devices,
2820 btrfs_test_opt(tree_root,
2821 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2822 1 : 0,
2823 fs_info->check_integrity_print_mask);
2824 if (ret)
2825 printk(KERN_WARNING "BTRFS: failed to initialize"
2826 " integrity check module %s\n", sb->s_id);
2827 }
2828#endif
2829 ret = btrfs_read_qgroup_config(fs_info);
2830 if (ret)
2831 goto fail_trans_kthread;
2832
2833 /* do not make disk changes in broken FS */
2834 if (btrfs_super_log_root(disk_super) != 0) {
2835 u64 bytenr = btrfs_super_log_root(disk_super);
2836
2837 if (fs_devices->rw_devices == 0) {
2838 printk(KERN_WARNING "BTRFS: log replay required "
2839 "on RO media\n");
2840 err = -EIO;
2841 goto fail_qgroup;
2842 }
2843 blocksize =
2844 btrfs_level_size(tree_root,
2845 btrfs_super_log_root_level(disk_super));
2846
2847 log_tree_root = btrfs_alloc_root(fs_info);
2848 if (!log_tree_root) {
2849 err = -ENOMEM;
2850 goto fail_qgroup;
2851 }
2852
2853 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2854 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2855
2856 log_tree_root->node = read_tree_block(tree_root, bytenr,
2857 blocksize,
2858 generation + 1);
2859 if (!log_tree_root->node ||
2860 !extent_buffer_uptodate(log_tree_root->node)) {
2861 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2862 free_extent_buffer(log_tree_root->node);
2863 kfree(log_tree_root);
2864 goto fail_qgroup;
2865 }
2866 /* returns with log_tree_root freed on success */
2867 ret = btrfs_recover_log_trees(log_tree_root);
2868 if (ret) {
2869 btrfs_error(tree_root->fs_info, ret,
2870 "Failed to recover log tree");
2871 free_extent_buffer(log_tree_root->node);
2872 kfree(log_tree_root);
2873 goto fail_qgroup;
2874 }
2875
2876 if (sb->s_flags & MS_RDONLY) {
2877 ret = btrfs_commit_super(tree_root);
2878 if (ret)
2879 goto fail_qgroup;
2880 }
2881 }
2882
2883 ret = btrfs_find_orphan_roots(tree_root);
2884 if (ret)
2885 goto fail_qgroup;
2886
2887 if (!(sb->s_flags & MS_RDONLY)) {
2888 ret = btrfs_cleanup_fs_roots(fs_info);
2889 if (ret)
2890 goto fail_qgroup;
2891
2892 ret = btrfs_recover_relocation(tree_root);
2893 if (ret < 0) {
2894 printk(KERN_WARNING
2895 "BTRFS: failed to recover relocation\n");
2896 err = -EINVAL;
2897 goto fail_qgroup;
2898 }
2899 }
2900
2901 location.objectid = BTRFS_FS_TREE_OBJECTID;
2902 location.type = BTRFS_ROOT_ITEM_KEY;
2903 location.offset = 0;
2904
2905 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2906 if (IS_ERR(fs_info->fs_root)) {
2907 err = PTR_ERR(fs_info->fs_root);
2908 goto fail_qgroup;
2909 }
2910
2911 if (sb->s_flags & MS_RDONLY)
2912 return 0;
2913
2914 down_read(&fs_info->cleanup_work_sem);
2915 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2916 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2917 up_read(&fs_info->cleanup_work_sem);
2918 close_ctree(tree_root);
2919 return ret;
2920 }
2921 up_read(&fs_info->cleanup_work_sem);
2922
2923 ret = btrfs_resume_balance_async(fs_info);
2924 if (ret) {
2925 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2926 close_ctree(tree_root);
2927 return ret;
2928 }
2929
2930 ret = btrfs_resume_dev_replace_async(fs_info);
2931 if (ret) {
2932 pr_warn("BTRFS: failed to resume dev_replace\n");
2933 close_ctree(tree_root);
2934 return ret;
2935 }
2936
2937 btrfs_qgroup_rescan_resume(fs_info);
2938
2939 if (create_uuid_tree) {
2940 pr_info("BTRFS: creating UUID tree\n");
2941 ret = btrfs_create_uuid_tree(fs_info);
2942 if (ret) {
2943 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2944 ret);
2945 close_ctree(tree_root);
2946 return ret;
2947 }
2948 } else if (check_uuid_tree ||
2949 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2950 pr_info("BTRFS: checking UUID tree\n");
2951 ret = btrfs_check_uuid_tree(fs_info);
2952 if (ret) {
2953 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2954 ret);
2955 close_ctree(tree_root);
2956 return ret;
2957 }
2958 } else {
2959 fs_info->update_uuid_tree_gen = 1;
2960 }
2961
2962 return 0;
2963
2964fail_qgroup:
2965 btrfs_free_qgroup_config(fs_info);
2966fail_trans_kthread:
2967 kthread_stop(fs_info->transaction_kthread);
2968 btrfs_cleanup_transaction(fs_info->tree_root);
2969 del_fs_roots(fs_info);
2970fail_cleaner:
2971 kthread_stop(fs_info->cleaner_kthread);
2972
2973 /*
2974 * make sure we're done with the btree inode before we stop our
2975 * kthreads
2976 */
2977 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2978
2979fail_sysfs:
2980 btrfs_sysfs_remove_one(fs_info);
2981
2982fail_block_groups:
2983 btrfs_put_block_group_cache(fs_info);
2984 btrfs_free_block_groups(fs_info);
2985
2986fail_tree_roots:
2987 free_root_pointers(fs_info, 1);
2988 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2989
2990fail_sb_buffer:
2991 btrfs_stop_all_workers(fs_info);
2992fail_alloc:
2993fail_iput:
2994 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2995
2996 iput(fs_info->btree_inode);
2997fail_bio_counter:
2998 percpu_counter_destroy(&fs_info->bio_counter);
2999fail_delalloc_bytes:
3000 percpu_counter_destroy(&fs_info->delalloc_bytes);
3001fail_dirty_metadata_bytes:
3002 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3003fail_bdi:
3004 bdi_destroy(&fs_info->bdi);
3005fail_srcu:
3006 cleanup_srcu_struct(&fs_info->subvol_srcu);
3007fail:
3008 btrfs_free_stripe_hash_table(fs_info);
3009 btrfs_close_devices(fs_info->fs_devices);
3010 return err;
3011
3012recovery_tree_root:
3013 if (!btrfs_test_opt(tree_root, RECOVERY))
3014 goto fail_tree_roots;
3015
3016 free_root_pointers(fs_info, 0);
3017
3018 /* don't use the log in recovery mode, it won't be valid */
3019 btrfs_set_super_log_root(disk_super, 0);
3020
3021 /* we can't trust the free space cache either */
3022 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3023
3024 ret = next_root_backup(fs_info, fs_info->super_copy,
3025 &num_backups_tried, &backup_index);
3026 if (ret == -1)
3027 goto fail_block_groups;
3028 goto retry_root_backup;
3029}
3030
3031static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3032{
3033 if (uptodate) {
3034 set_buffer_uptodate(bh);
3035 } else {
3036 struct btrfs_device *device = (struct btrfs_device *)
3037 bh->b_private;
3038
3039 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3040 "I/O error on %s\n",
3041 rcu_str_deref(device->name));
3042 /* note, we dont' set_buffer_write_io_error because we have
3043 * our own ways of dealing with the IO errors
3044 */
3045 clear_buffer_uptodate(bh);
3046 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3047 }
3048 unlock_buffer(bh);
3049 put_bh(bh);
3050}
3051
3052struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3053{
3054 struct buffer_head *bh;
3055 struct buffer_head *latest = NULL;
3056 struct btrfs_super_block *super;
3057 int i;
3058 u64 transid = 0;
3059 u64 bytenr;
3060
3061 /* we would like to check all the supers, but that would make
3062 * a btrfs mount succeed after a mkfs from a different FS.
3063 * So, we need to add a special mount option to scan for
3064 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3065 */
3066 for (i = 0; i < 1; i++) {
3067 bytenr = btrfs_sb_offset(i);
3068 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3069 i_size_read(bdev->bd_inode))
3070 break;
3071 bh = __bread(bdev, bytenr / 4096,
3072 BTRFS_SUPER_INFO_SIZE);
3073 if (!bh)
3074 continue;
3075
3076 super = (struct btrfs_super_block *)bh->b_data;
3077 if (btrfs_super_bytenr(super) != bytenr ||
3078 btrfs_super_magic(super) != BTRFS_MAGIC) {
3079 brelse(bh);
3080 continue;
3081 }
3082
3083 if (!latest || btrfs_super_generation(super) > transid) {
3084 brelse(latest);
3085 latest = bh;
3086 transid = btrfs_super_generation(super);
3087 } else {
3088 brelse(bh);
3089 }
3090 }
3091 return latest;
3092}
3093
3094/*
3095 * this should be called twice, once with wait == 0 and
3096 * once with wait == 1. When wait == 0 is done, all the buffer heads
3097 * we write are pinned.
3098 *
3099 * They are released when wait == 1 is done.
3100 * max_mirrors must be the same for both runs, and it indicates how
3101 * many supers on this one device should be written.
3102 *
3103 * max_mirrors == 0 means to write them all.
3104 */
3105static int write_dev_supers(struct btrfs_device *device,
3106 struct btrfs_super_block *sb,
3107 int do_barriers, int wait, int max_mirrors)
3108{
3109 struct buffer_head *bh;
3110 int i;
3111 int ret;
3112 int errors = 0;
3113 u32 crc;
3114 u64 bytenr;
3115
3116 if (max_mirrors == 0)
3117 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3118
3119 for (i = 0; i < max_mirrors; i++) {
3120 bytenr = btrfs_sb_offset(i);
3121 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3122 break;
3123
3124 if (wait) {
3125 bh = __find_get_block(device->bdev, bytenr / 4096,
3126 BTRFS_SUPER_INFO_SIZE);
3127 if (!bh) {
3128 errors++;
3129 continue;
3130 }
3131 wait_on_buffer(bh);
3132 if (!buffer_uptodate(bh))
3133 errors++;
3134
3135 /* drop our reference */
3136 brelse(bh);
3137
3138 /* drop the reference from the wait == 0 run */
3139 brelse(bh);
3140 continue;
3141 } else {
3142 btrfs_set_super_bytenr(sb, bytenr);
3143
3144 crc = ~(u32)0;
3145 crc = btrfs_csum_data((char *)sb +
3146 BTRFS_CSUM_SIZE, crc,
3147 BTRFS_SUPER_INFO_SIZE -
3148 BTRFS_CSUM_SIZE);
3149 btrfs_csum_final(crc, sb->csum);
3150
3151 /*
3152 * one reference for us, and we leave it for the
3153 * caller
3154 */
3155 bh = __getblk(device->bdev, bytenr / 4096,
3156 BTRFS_SUPER_INFO_SIZE);
3157 if (!bh) {
3158 printk(KERN_ERR "BTRFS: couldn't get super "
3159 "buffer head for bytenr %Lu\n", bytenr);
3160 errors++;
3161 continue;
3162 }
3163
3164 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3165
3166 /* one reference for submit_bh */
3167 get_bh(bh);
3168
3169 set_buffer_uptodate(bh);
3170 lock_buffer(bh);
3171 bh->b_end_io = btrfs_end_buffer_write_sync;
3172 bh->b_private = device;
3173 }
3174
3175 /*
3176 * we fua the first super. The others we allow
3177 * to go down lazy.
3178 */
3179 if (i == 0)
3180 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3181 else
3182 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3183 if (ret)
3184 errors++;
3185 }
3186 return errors < i ? 0 : -1;
3187}
3188
3189/*
3190 * endio for the write_dev_flush, this will wake anyone waiting
3191 * for the barrier when it is done
3192 */
3193static void btrfs_end_empty_barrier(struct bio *bio, int err)
3194{
3195 if (err) {
3196 if (err == -EOPNOTSUPP)
3197 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3198 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3199 }
3200 if (bio->bi_private)
3201 complete(bio->bi_private);
3202 bio_put(bio);
3203}
3204
3205/*
3206 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3207 * sent down. With wait == 1, it waits for the previous flush.
3208 *
3209 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3210 * capable
3211 */
3212static int write_dev_flush(struct btrfs_device *device, int wait)
3213{
3214 struct bio *bio;
3215 int ret = 0;
3216
3217 if (device->nobarriers)
3218 return 0;
3219
3220 if (wait) {
3221 bio = device->flush_bio;
3222 if (!bio)
3223 return 0;
3224
3225 wait_for_completion(&device->flush_wait);
3226
3227 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3228 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3229 rcu_str_deref(device->name));
3230 device->nobarriers = 1;
3231 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3232 ret = -EIO;
3233 btrfs_dev_stat_inc_and_print(device,
3234 BTRFS_DEV_STAT_FLUSH_ERRS);
3235 }
3236
3237 /* drop the reference from the wait == 0 run */
3238 bio_put(bio);
3239 device->flush_bio = NULL;
3240
3241 return ret;
3242 }
3243
3244 /*
3245 * one reference for us, and we leave it for the
3246 * caller
3247 */
3248 device->flush_bio = NULL;
3249 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3250 if (!bio)
3251 return -ENOMEM;
3252
3253 bio->bi_end_io = btrfs_end_empty_barrier;
3254 bio->bi_bdev = device->bdev;
3255 init_completion(&device->flush_wait);
3256 bio->bi_private = &device->flush_wait;
3257 device->flush_bio = bio;
3258
3259 bio_get(bio);
3260 btrfsic_submit_bio(WRITE_FLUSH, bio);
3261
3262 return 0;
3263}
3264
3265/*
3266 * send an empty flush down to each device in parallel,
3267 * then wait for them
3268 */
3269static int barrier_all_devices(struct btrfs_fs_info *info)
3270{
3271 struct list_head *head;
3272 struct btrfs_device *dev;
3273 int errors_send = 0;
3274 int errors_wait = 0;
3275 int ret;
3276
3277 /* send down all the barriers */
3278 head = &info->fs_devices->devices;
3279 list_for_each_entry_rcu(dev, head, dev_list) {
3280 if (dev->missing)
3281 continue;
3282 if (!dev->bdev) {
3283 errors_send++;
3284 continue;
3285 }
3286 if (!dev->in_fs_metadata || !dev->writeable)
3287 continue;
3288
3289 ret = write_dev_flush(dev, 0);
3290 if (ret)
3291 errors_send++;
3292 }
3293
3294 /* wait for all the barriers */
3295 list_for_each_entry_rcu(dev, head, dev_list) {
3296 if (dev->missing)
3297 continue;
3298 if (!dev->bdev) {
3299 errors_wait++;
3300 continue;
3301 }
3302 if (!dev->in_fs_metadata || !dev->writeable)
3303 continue;
3304
3305 ret = write_dev_flush(dev, 1);
3306 if (ret)
3307 errors_wait++;
3308 }
3309 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3310 errors_wait > info->num_tolerated_disk_barrier_failures)
3311 return -EIO;
3312 return 0;
3313}
3314
3315int btrfs_calc_num_tolerated_disk_barrier_failures(
3316 struct btrfs_fs_info *fs_info)
3317{
3318 struct btrfs_ioctl_space_info space;
3319 struct btrfs_space_info *sinfo;
3320 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3321 BTRFS_BLOCK_GROUP_SYSTEM,
3322 BTRFS_BLOCK_GROUP_METADATA,
3323 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3324 int num_types = 4;
3325 int i;
3326 int c;
3327 int num_tolerated_disk_barrier_failures =
3328 (int)fs_info->fs_devices->num_devices;
3329
3330 for (i = 0; i < num_types; i++) {
3331 struct btrfs_space_info *tmp;
3332
3333 sinfo = NULL;
3334 rcu_read_lock();
3335 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3336 if (tmp->flags == types[i]) {
3337 sinfo = tmp;
3338 break;
3339 }
3340 }
3341 rcu_read_unlock();
3342
3343 if (!sinfo)
3344 continue;
3345
3346 down_read(&sinfo->groups_sem);
3347 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3348 if (!list_empty(&sinfo->block_groups[c])) {
3349 u64 flags;
3350
3351 btrfs_get_block_group_info(
3352 &sinfo->block_groups[c], &space);
3353 if (space.total_bytes == 0 ||
3354 space.used_bytes == 0)
3355 continue;
3356 flags = space.flags;
3357 /*
3358 * return
3359 * 0: if dup, single or RAID0 is configured for
3360 * any of metadata, system or data, else
3361 * 1: if RAID5 is configured, or if RAID1 or
3362 * RAID10 is configured and only two mirrors
3363 * are used, else
3364 * 2: if RAID6 is configured, else
3365 * num_mirrors - 1: if RAID1 or RAID10 is
3366 * configured and more than
3367 * 2 mirrors are used.
3368 */
3369 if (num_tolerated_disk_barrier_failures > 0 &&
3370 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3371 BTRFS_BLOCK_GROUP_RAID0)) ||
3372 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3373 == 0)))
3374 num_tolerated_disk_barrier_failures = 0;
3375 else if (num_tolerated_disk_barrier_failures > 1) {
3376 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3377 BTRFS_BLOCK_GROUP_RAID5 |
3378 BTRFS_BLOCK_GROUP_RAID10)) {
3379 num_tolerated_disk_barrier_failures = 1;
3380 } else if (flags &
3381 BTRFS_BLOCK_GROUP_RAID6) {
3382 num_tolerated_disk_barrier_failures = 2;
3383 }
3384 }
3385 }
3386 }
3387 up_read(&sinfo->groups_sem);
3388 }
3389
3390 return num_tolerated_disk_barrier_failures;
3391}
3392
3393static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3394{
3395 struct list_head *head;
3396 struct btrfs_device *dev;
3397 struct btrfs_super_block *sb;
3398 struct btrfs_dev_item *dev_item;
3399 int ret;
3400 int do_barriers;
3401 int max_errors;
3402 int total_errors = 0;
3403 u64 flags;
3404
3405 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3406 backup_super_roots(root->fs_info);
3407
3408 sb = root->fs_info->super_for_commit;
3409 dev_item = &sb->dev_item;
3410
3411 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3412 head = &root->fs_info->fs_devices->devices;
3413 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3414
3415 if (do_barriers) {
3416 ret = barrier_all_devices(root->fs_info);
3417 if (ret) {
3418 mutex_unlock(
3419 &root->fs_info->fs_devices->device_list_mutex);
3420 btrfs_error(root->fs_info, ret,
3421 "errors while submitting device barriers.");
3422 return ret;
3423 }
3424 }
3425
3426 list_for_each_entry_rcu(dev, head, dev_list) {
3427 if (!dev->bdev) {
3428 total_errors++;
3429 continue;
3430 }
3431 if (!dev->in_fs_metadata || !dev->writeable)
3432 continue;
3433
3434 btrfs_set_stack_device_generation(dev_item, 0);
3435 btrfs_set_stack_device_type(dev_item, dev->type);
3436 btrfs_set_stack_device_id(dev_item, dev->devid);
3437 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3438 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3439 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3440 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3441 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3442 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3443 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3444
3445 flags = btrfs_super_flags(sb);
3446 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3447
3448 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3449 if (ret)
3450 total_errors++;
3451 }
3452 if (total_errors > max_errors) {
3453 btrfs_err(root->fs_info, "%d errors while writing supers",
3454 total_errors);
3455 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3456
3457 /* FUA is masked off if unsupported and can't be the reason */
3458 btrfs_error(root->fs_info, -EIO,
3459 "%d errors while writing supers", total_errors);
3460 return -EIO;
3461 }
3462
3463 total_errors = 0;
3464 list_for_each_entry_rcu(dev, head, dev_list) {
3465 if (!dev->bdev)
3466 continue;
3467 if (!dev->in_fs_metadata || !dev->writeable)
3468 continue;
3469
3470 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3471 if (ret)
3472 total_errors++;
3473 }
3474 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3475 if (total_errors > max_errors) {
3476 btrfs_error(root->fs_info, -EIO,
3477 "%d errors while writing supers", total_errors);
3478 return -EIO;
3479 }
3480 return 0;
3481}
3482
3483int write_ctree_super(struct btrfs_trans_handle *trans,
3484 struct btrfs_root *root, int max_mirrors)
3485{
3486 return write_all_supers(root, max_mirrors);
3487}
3488
3489/* Drop a fs root from the radix tree and free it. */
3490void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3491 struct btrfs_root *root)
3492{
3493 spin_lock(&fs_info->fs_roots_radix_lock);
3494 radix_tree_delete(&fs_info->fs_roots_radix,
3495 (unsigned long)root->root_key.objectid);
3496 spin_unlock(&fs_info->fs_roots_radix_lock);
3497
3498 if (btrfs_root_refs(&root->root_item) == 0)
3499 synchronize_srcu(&fs_info->subvol_srcu);
3500
3501 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3502 btrfs_free_log(NULL, root);
3503
3504 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3505 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3506 free_fs_root(root);
3507}
3508
3509static void free_fs_root(struct btrfs_root *root)
3510{
3511 iput(root->cache_inode);
3512 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3513 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3514 root->orphan_block_rsv = NULL;
3515 if (root->anon_dev)
3516 free_anon_bdev(root->anon_dev);
3517 if (root->subv_writers)
3518 btrfs_free_subvolume_writers(root->subv_writers);
3519 free_extent_buffer(root->node);
3520 free_extent_buffer(root->commit_root);
3521 kfree(root->free_ino_ctl);
3522 kfree(root->free_ino_pinned);
3523 kfree(root->name);
3524 btrfs_put_fs_root(root);
3525}
3526
3527void btrfs_free_fs_root(struct btrfs_root *root)
3528{
3529 free_fs_root(root);
3530}
3531
3532int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3533{
3534 u64 root_objectid = 0;
3535 struct btrfs_root *gang[8];
3536 int i;
3537 int ret;
3538
3539 while (1) {
3540 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3541 (void **)gang, root_objectid,
3542 ARRAY_SIZE(gang));
3543 if (!ret)
3544 break;
3545
3546 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3547 for (i = 0; i < ret; i++) {
3548 int err;
3549
3550 root_objectid = gang[i]->root_key.objectid;
3551 err = btrfs_orphan_cleanup(gang[i]);
3552 if (err)
3553 return err;
3554 }
3555 root_objectid++;
3556 }
3557 return 0;
3558}
3559
3560int btrfs_commit_super(struct btrfs_root *root)
3561{
3562 struct btrfs_trans_handle *trans;
3563
3564 mutex_lock(&root->fs_info->cleaner_mutex);
3565 btrfs_run_delayed_iputs(root);
3566 mutex_unlock(&root->fs_info->cleaner_mutex);
3567 wake_up_process(root->fs_info->cleaner_kthread);
3568
3569 /* wait until ongoing cleanup work done */
3570 down_write(&root->fs_info->cleanup_work_sem);
3571 up_write(&root->fs_info->cleanup_work_sem);
3572
3573 trans = btrfs_join_transaction(root);
3574 if (IS_ERR(trans))
3575 return PTR_ERR(trans);
3576 return btrfs_commit_transaction(trans, root);
3577}
3578
3579int close_ctree(struct btrfs_root *root)
3580{
3581 struct btrfs_fs_info *fs_info = root->fs_info;
3582 int ret;
3583
3584 fs_info->closing = 1;
3585 smp_mb();
3586
3587 /* wait for the uuid_scan task to finish */
3588 down(&fs_info->uuid_tree_rescan_sem);
3589 /* avoid complains from lockdep et al., set sem back to initial state */
3590 up(&fs_info->uuid_tree_rescan_sem);
3591
3592 /* pause restriper - we want to resume on mount */
3593 btrfs_pause_balance(fs_info);
3594
3595 btrfs_dev_replace_suspend_for_unmount(fs_info);
3596
3597 btrfs_scrub_cancel(fs_info);
3598
3599 /* wait for any defraggers to finish */
3600 wait_event(fs_info->transaction_wait,
3601 (atomic_read(&fs_info->defrag_running) == 0));
3602
3603 /* clear out the rbtree of defraggable inodes */
3604 btrfs_cleanup_defrag_inodes(fs_info);
3605
3606 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3607 ret = btrfs_commit_super(root);
3608 if (ret)
3609 btrfs_err(root->fs_info, "commit super ret %d", ret);
3610 }
3611
3612 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3613 btrfs_error_commit_super(root);
3614
3615 kthread_stop(fs_info->transaction_kthread);
3616 kthread_stop(fs_info->cleaner_kthread);
3617
3618 fs_info->closing = 2;
3619 smp_mb();
3620
3621 btrfs_free_qgroup_config(root->fs_info);
3622
3623 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3624 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3625 percpu_counter_sum(&fs_info->delalloc_bytes));
3626 }
3627
3628 btrfs_sysfs_remove_one(fs_info);
3629
3630 del_fs_roots(fs_info);
3631
3632 btrfs_put_block_group_cache(fs_info);
3633
3634 btrfs_free_block_groups(fs_info);
3635
3636 btrfs_stop_all_workers(fs_info);
3637
3638 free_root_pointers(fs_info, 1);
3639
3640 iput(fs_info->btree_inode);
3641
3642#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3643 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3644 btrfsic_unmount(root, fs_info->fs_devices);
3645#endif
3646
3647 btrfs_close_devices(fs_info->fs_devices);
3648 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3649
3650 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3651 percpu_counter_destroy(&fs_info->delalloc_bytes);
3652 percpu_counter_destroy(&fs_info->bio_counter);
3653 bdi_destroy(&fs_info->bdi);
3654 cleanup_srcu_struct(&fs_info->subvol_srcu);
3655
3656 btrfs_free_stripe_hash_table(fs_info);
3657
3658 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3659 root->orphan_block_rsv = NULL;
3660
3661 return 0;
3662}
3663
3664int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3665 int atomic)
3666{
3667 int ret;
3668 struct inode *btree_inode = buf->pages[0]->mapping->host;
3669
3670 ret = extent_buffer_uptodate(buf);
3671 if (!ret)
3672 return ret;
3673
3674 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3675 parent_transid, atomic);
3676 if (ret == -EAGAIN)
3677 return ret;
3678 return !ret;
3679}
3680
3681int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3682{
3683 return set_extent_buffer_uptodate(buf);
3684}
3685
3686void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3687{
3688 struct btrfs_root *root;
3689 u64 transid = btrfs_header_generation(buf);
3690 int was_dirty;
3691
3692#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3693 /*
3694 * This is a fast path so only do this check if we have sanity tests
3695 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3696 * outside of the sanity tests.
3697 */
3698 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3699 return;
3700#endif
3701 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3702 btrfs_assert_tree_locked(buf);
3703 if (transid != root->fs_info->generation)
3704 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3705 "found %llu running %llu\n",
3706 buf->start, transid, root->fs_info->generation);
3707 was_dirty = set_extent_buffer_dirty(buf);
3708 if (!was_dirty)
3709 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3710 buf->len,
3711 root->fs_info->dirty_metadata_batch);
3712}
3713
3714static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3715 int flush_delayed)
3716{
3717 /*
3718 * looks as though older kernels can get into trouble with
3719 * this code, they end up stuck in balance_dirty_pages forever
3720 */
3721 int ret;
3722
3723 if (current->flags & PF_MEMALLOC)
3724 return;
3725
3726 if (flush_delayed)
3727 btrfs_balance_delayed_items(root);
3728
3729 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3730 BTRFS_DIRTY_METADATA_THRESH);
3731 if (ret > 0) {
3732 balance_dirty_pages_ratelimited(
3733 root->fs_info->btree_inode->i_mapping);
3734 }
3735 return;
3736}
3737
3738void btrfs_btree_balance_dirty(struct btrfs_root *root)
3739{
3740 __btrfs_btree_balance_dirty(root, 1);
3741}
3742
3743void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3744{
3745 __btrfs_btree_balance_dirty(root, 0);
3746}
3747
3748int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3749{
3750 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3751 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3752}
3753
3754static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3755 int read_only)
3756{
3757 /*
3758 * Placeholder for checks
3759 */
3760 return 0;
3761}
3762
3763static void btrfs_error_commit_super(struct btrfs_root *root)
3764{
3765 mutex_lock(&root->fs_info->cleaner_mutex);
3766 btrfs_run_delayed_iputs(root);
3767 mutex_unlock(&root->fs_info->cleaner_mutex);
3768
3769 down_write(&root->fs_info->cleanup_work_sem);
3770 up_write(&root->fs_info->cleanup_work_sem);
3771
3772 /* cleanup FS via transaction */
3773 btrfs_cleanup_transaction(root);
3774}
3775
3776static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3777 struct btrfs_root *root)
3778{
3779 struct btrfs_inode *btrfs_inode;
3780 struct list_head splice;
3781
3782 INIT_LIST_HEAD(&splice);
3783
3784 mutex_lock(&root->fs_info->ordered_operations_mutex);
3785 spin_lock(&root->fs_info->ordered_root_lock);
3786
3787 list_splice_init(&t->ordered_operations, &splice);
3788 while (!list_empty(&splice)) {
3789 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3790 ordered_operations);
3791
3792 list_del_init(&btrfs_inode->ordered_operations);
3793 spin_unlock(&root->fs_info->ordered_root_lock);
3794
3795 btrfs_invalidate_inodes(btrfs_inode->root);
3796
3797 spin_lock(&root->fs_info->ordered_root_lock);
3798 }
3799
3800 spin_unlock(&root->fs_info->ordered_root_lock);
3801 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3802}
3803
3804static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3805{
3806 struct btrfs_ordered_extent *ordered;
3807
3808 spin_lock(&root->ordered_extent_lock);
3809 /*
3810 * This will just short circuit the ordered completion stuff which will
3811 * make sure the ordered extent gets properly cleaned up.
3812 */
3813 list_for_each_entry(ordered, &root->ordered_extents,
3814 root_extent_list)
3815 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3816 spin_unlock(&root->ordered_extent_lock);
3817}
3818
3819static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3820{
3821 struct btrfs_root *root;
3822 struct list_head splice;
3823
3824 INIT_LIST_HEAD(&splice);
3825
3826 spin_lock(&fs_info->ordered_root_lock);
3827 list_splice_init(&fs_info->ordered_roots, &splice);
3828 while (!list_empty(&splice)) {
3829 root = list_first_entry(&splice, struct btrfs_root,
3830 ordered_root);
3831 list_move_tail(&root->ordered_root,
3832 &fs_info->ordered_roots);
3833
3834 spin_unlock(&fs_info->ordered_root_lock);
3835 btrfs_destroy_ordered_extents(root);
3836
3837 cond_resched();
3838 spin_lock(&fs_info->ordered_root_lock);
3839 }
3840 spin_unlock(&fs_info->ordered_root_lock);
3841}
3842
3843static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3844 struct btrfs_root *root)
3845{
3846 struct rb_node *node;
3847 struct btrfs_delayed_ref_root *delayed_refs;
3848 struct btrfs_delayed_ref_node *ref;
3849 int ret = 0;
3850
3851 delayed_refs = &trans->delayed_refs;
3852
3853 spin_lock(&delayed_refs->lock);
3854 if (atomic_read(&delayed_refs->num_entries) == 0) {
3855 spin_unlock(&delayed_refs->lock);
3856 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3857 return ret;
3858 }
3859
3860 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3861 struct btrfs_delayed_ref_head *head;
3862 bool pin_bytes = false;
3863
3864 head = rb_entry(node, struct btrfs_delayed_ref_head,
3865 href_node);
3866 if (!mutex_trylock(&head->mutex)) {
3867 atomic_inc(&head->node.refs);
3868 spin_unlock(&delayed_refs->lock);
3869
3870 mutex_lock(&head->mutex);
3871 mutex_unlock(&head->mutex);
3872 btrfs_put_delayed_ref(&head->node);
3873 spin_lock(&delayed_refs->lock);
3874 continue;
3875 }
3876 spin_lock(&head->lock);
3877 while ((node = rb_first(&head->ref_root)) != NULL) {
3878 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3879 rb_node);
3880 ref->in_tree = 0;
3881 rb_erase(&ref->rb_node, &head->ref_root);
3882 atomic_dec(&delayed_refs->num_entries);
3883 btrfs_put_delayed_ref(ref);
3884 }
3885 if (head->must_insert_reserved)
3886 pin_bytes = true;
3887 btrfs_free_delayed_extent_op(head->extent_op);
3888 delayed_refs->num_heads--;
3889 if (head->processing == 0)
3890 delayed_refs->num_heads_ready--;
3891 atomic_dec(&delayed_refs->num_entries);
3892 head->node.in_tree = 0;
3893 rb_erase(&head->href_node, &delayed_refs->href_root);
3894 spin_unlock(&head->lock);
3895 spin_unlock(&delayed_refs->lock);
3896 mutex_unlock(&head->mutex);
3897
3898 if (pin_bytes)
3899 btrfs_pin_extent(root, head->node.bytenr,
3900 head->node.num_bytes, 1);
3901 btrfs_put_delayed_ref(&head->node);
3902 cond_resched();
3903 spin_lock(&delayed_refs->lock);
3904 }
3905
3906 spin_unlock(&delayed_refs->lock);
3907
3908 return ret;
3909}
3910
3911static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3912{
3913 struct btrfs_inode *btrfs_inode;
3914 struct list_head splice;
3915
3916 INIT_LIST_HEAD(&splice);
3917
3918 spin_lock(&root->delalloc_lock);
3919 list_splice_init(&root->delalloc_inodes, &splice);
3920
3921 while (!list_empty(&splice)) {
3922 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3923 delalloc_inodes);
3924
3925 list_del_init(&btrfs_inode->delalloc_inodes);
3926 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3927 &btrfs_inode->runtime_flags);
3928 spin_unlock(&root->delalloc_lock);
3929
3930 btrfs_invalidate_inodes(btrfs_inode->root);
3931
3932 spin_lock(&root->delalloc_lock);
3933 }
3934
3935 spin_unlock(&root->delalloc_lock);
3936}
3937
3938static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3939{
3940 struct btrfs_root *root;
3941 struct list_head splice;
3942
3943 INIT_LIST_HEAD(&splice);
3944
3945 spin_lock(&fs_info->delalloc_root_lock);
3946 list_splice_init(&fs_info->delalloc_roots, &splice);
3947 while (!list_empty(&splice)) {
3948 root = list_first_entry(&splice, struct btrfs_root,
3949 delalloc_root);
3950 list_del_init(&root->delalloc_root);
3951 root = btrfs_grab_fs_root(root);
3952 BUG_ON(!root);
3953 spin_unlock(&fs_info->delalloc_root_lock);
3954
3955 btrfs_destroy_delalloc_inodes(root);
3956 btrfs_put_fs_root(root);
3957
3958 spin_lock(&fs_info->delalloc_root_lock);
3959 }
3960 spin_unlock(&fs_info->delalloc_root_lock);
3961}
3962
3963static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3964 struct extent_io_tree *dirty_pages,
3965 int mark)
3966{
3967 int ret;
3968 struct extent_buffer *eb;
3969 u64 start = 0;
3970 u64 end;
3971
3972 while (1) {
3973 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3974 mark, NULL);
3975 if (ret)
3976 break;
3977
3978 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3979 while (start <= end) {
3980 eb = btrfs_find_tree_block(root, start,
3981 root->leafsize);
3982 start += root->leafsize;
3983 if (!eb)
3984 continue;
3985 wait_on_extent_buffer_writeback(eb);
3986
3987 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3988 &eb->bflags))
3989 clear_extent_buffer_dirty(eb);
3990 free_extent_buffer_stale(eb);
3991 }
3992 }
3993
3994 return ret;
3995}
3996
3997static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3998 struct extent_io_tree *pinned_extents)
3999{
4000 struct extent_io_tree *unpin;
4001 u64 start;
4002 u64 end;
4003 int ret;
4004 bool loop = true;
4005
4006 unpin = pinned_extents;
4007again:
4008 while (1) {
4009 ret = find_first_extent_bit(unpin, 0, &start, &end,
4010 EXTENT_DIRTY, NULL);
4011 if (ret)
4012 break;
4013
4014 /* opt_discard */
4015 if (btrfs_test_opt(root, DISCARD))
4016 ret = btrfs_error_discard_extent(root, start,
4017 end + 1 - start,
4018 NULL);
4019
4020 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4021 btrfs_error_unpin_extent_range(root, start, end);
4022 cond_resched();
4023 }
4024
4025 if (loop) {
4026 if (unpin == &root->fs_info->freed_extents[0])
4027 unpin = &root->fs_info->freed_extents[1];
4028 else
4029 unpin = &root->fs_info->freed_extents[0];
4030 loop = false;
4031 goto again;
4032 }
4033
4034 return 0;
4035}
4036
4037void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4038 struct btrfs_root *root)
4039{
4040 btrfs_destroy_ordered_operations(cur_trans, root);
4041
4042 btrfs_destroy_delayed_refs(cur_trans, root);
4043
4044 cur_trans->state = TRANS_STATE_COMMIT_START;
4045 wake_up(&root->fs_info->transaction_blocked_wait);
4046
4047 cur_trans->state = TRANS_STATE_UNBLOCKED;
4048 wake_up(&root->fs_info->transaction_wait);
4049
4050 btrfs_destroy_delayed_inodes(root);
4051 btrfs_assert_delayed_root_empty(root);
4052
4053 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4054 EXTENT_DIRTY);
4055 btrfs_destroy_pinned_extent(root,
4056 root->fs_info->pinned_extents);
4057
4058 cur_trans->state =TRANS_STATE_COMPLETED;
4059 wake_up(&cur_trans->commit_wait);
4060
4061 /*
4062 memset(cur_trans, 0, sizeof(*cur_trans));
4063 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4064 */
4065}
4066
4067static int btrfs_cleanup_transaction(struct btrfs_root *root)
4068{
4069 struct btrfs_transaction *t;
4070
4071 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4072
4073 spin_lock(&root->fs_info->trans_lock);
4074 while (!list_empty(&root->fs_info->trans_list)) {
4075 t = list_first_entry(&root->fs_info->trans_list,
4076 struct btrfs_transaction, list);
4077 if (t->state >= TRANS_STATE_COMMIT_START) {
4078 atomic_inc(&t->use_count);
4079 spin_unlock(&root->fs_info->trans_lock);
4080 btrfs_wait_for_commit(root, t->transid);
4081 btrfs_put_transaction(t);
4082 spin_lock(&root->fs_info->trans_lock);
4083 continue;
4084 }
4085 if (t == root->fs_info->running_transaction) {
4086 t->state = TRANS_STATE_COMMIT_DOING;
4087 spin_unlock(&root->fs_info->trans_lock);
4088 /*
4089 * We wait for 0 num_writers since we don't hold a trans
4090 * handle open currently for this transaction.
4091 */
4092 wait_event(t->writer_wait,
4093 atomic_read(&t->num_writers) == 0);
4094 } else {
4095 spin_unlock(&root->fs_info->trans_lock);
4096 }
4097 btrfs_cleanup_one_transaction(t, root);
4098
4099 spin_lock(&root->fs_info->trans_lock);
4100 if (t == root->fs_info->running_transaction)
4101 root->fs_info->running_transaction = NULL;
4102 list_del_init(&t->list);
4103 spin_unlock(&root->fs_info->trans_lock);
4104
4105 btrfs_put_transaction(t);
4106 trace_btrfs_transaction_commit(root);
4107 spin_lock(&root->fs_info->trans_lock);
4108 }
4109 spin_unlock(&root->fs_info->trans_lock);
4110 btrfs_destroy_all_ordered_extents(root->fs_info);
4111 btrfs_destroy_delayed_inodes(root);
4112 btrfs_assert_delayed_root_empty(root);
4113 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4114 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4115 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4116
4117 return 0;
4118}
4119
4120static struct extent_io_ops btree_extent_io_ops = {
4121 .readpage_end_io_hook = btree_readpage_end_io_hook,
4122 .readpage_io_failed_hook = btree_io_failed_hook,
4123 .submit_bio_hook = btree_submit_bio_hook,
4124 /* note we're sharing with inode.c for the merge bio hook */
4125 .merge_bio_hook = btrfs_merge_bio_hook,
4126};
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 <asm/unaligned.h>
34#include "compat.h"
35#include "ctree.h"
36#include "disk-io.h"
37#include "transaction.h"
38#include "btrfs_inode.h"
39#include "volumes.h"
40#include "print-tree.h"
41#include "async-thread.h"
42#include "locking.h"
43#include "tree-log.h"
44#include "free-space-cache.h"
45#include "inode-map.h"
46#include "check-integrity.h"
47#include "rcu-string.h"
48
49static struct extent_io_ops btree_extent_io_ops;
50static void end_workqueue_fn(struct btrfs_work *work);
51static void free_fs_root(struct btrfs_root *root);
52static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
53 int read_only);
54static void btrfs_destroy_ordered_operations(struct btrfs_root *root);
55static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 struct btrfs_root *root);
58static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
59static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
60static int btrfs_destroy_marked_extents(struct btrfs_root *root,
61 struct extent_io_tree *dirty_pages,
62 int mark);
63static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
64 struct extent_io_tree *pinned_extents);
65
66/*
67 * end_io_wq structs are used to do processing in task context when an IO is
68 * complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
70 */
71struct end_io_wq {
72 struct bio *bio;
73 bio_end_io_t *end_io;
74 void *private;
75 struct btrfs_fs_info *info;
76 int error;
77 int metadata;
78 struct list_head list;
79 struct btrfs_work work;
80};
81
82/*
83 * async submit bios are used to offload expensive checksumming
84 * onto the worker threads. They checksum file and metadata bios
85 * just before they are sent down the IO stack.
86 */
87struct async_submit_bio {
88 struct inode *inode;
89 struct bio *bio;
90 struct list_head list;
91 extent_submit_bio_hook_t *submit_bio_start;
92 extent_submit_bio_hook_t *submit_bio_done;
93 int rw;
94 int mirror_num;
95 unsigned long bio_flags;
96 /*
97 * bio_offset is optional, can be used if the pages in the bio
98 * can't tell us where in the file the bio should go
99 */
100 u64 bio_offset;
101 struct btrfs_work work;
102 int error;
103};
104
105/*
106 * Lockdep class keys for extent_buffer->lock's in this root. For a given
107 * eb, the lockdep key is determined by the btrfs_root it belongs to and
108 * the level the eb occupies in the tree.
109 *
110 * Different roots are used for different purposes and may nest inside each
111 * other and they require separate keysets. As lockdep keys should be
112 * static, assign keysets according to the purpose of the root as indicated
113 * by btrfs_root->objectid. This ensures that all special purpose roots
114 * have separate keysets.
115 *
116 * Lock-nesting across peer nodes is always done with the immediate parent
117 * node locked thus preventing deadlock. As lockdep doesn't know this, use
118 * subclass to avoid triggering lockdep warning in such cases.
119 *
120 * The key is set by the readpage_end_io_hook after the buffer has passed
121 * csum validation but before the pages are unlocked. It is also set by
122 * btrfs_init_new_buffer on freshly allocated blocks.
123 *
124 * We also add a check to make sure the highest level of the tree is the
125 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
126 * needs update as well.
127 */
128#ifdef CONFIG_DEBUG_LOCK_ALLOC
129# if BTRFS_MAX_LEVEL != 8
130# error
131# endif
132
133static struct btrfs_lockdep_keyset {
134 u64 id; /* root objectid */
135 const char *name_stem; /* lock name stem */
136 char names[BTRFS_MAX_LEVEL + 1][20];
137 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
138} btrfs_lockdep_keysets[] = {
139 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
140 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
141 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
142 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
143 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
144 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
145 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
146 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
147 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
148 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
149 { .id = 0, .name_stem = "tree" },
150};
151
152void __init btrfs_init_lockdep(void)
153{
154 int i, j;
155
156 /* initialize lockdep class names */
157 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
158 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
159
160 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
161 snprintf(ks->names[j], sizeof(ks->names[j]),
162 "btrfs-%s-%02d", ks->name_stem, j);
163 }
164}
165
166void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
167 int level)
168{
169 struct btrfs_lockdep_keyset *ks;
170
171 BUG_ON(level >= ARRAY_SIZE(ks->keys));
172
173 /* find the matching keyset, id 0 is the default entry */
174 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
175 if (ks->id == objectid)
176 break;
177
178 lockdep_set_class_and_name(&eb->lock,
179 &ks->keys[level], ks->names[level]);
180}
181
182#endif
183
184/*
185 * extents on the btree inode are pretty simple, there's one extent
186 * that covers the entire device
187 */
188static struct extent_map *btree_get_extent(struct inode *inode,
189 struct page *page, size_t pg_offset, u64 start, u64 len,
190 int create)
191{
192 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
193 struct extent_map *em;
194 int ret;
195
196 read_lock(&em_tree->lock);
197 em = lookup_extent_mapping(em_tree, start, len);
198 if (em) {
199 em->bdev =
200 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
201 read_unlock(&em_tree->lock);
202 goto out;
203 }
204 read_unlock(&em_tree->lock);
205
206 em = alloc_extent_map();
207 if (!em) {
208 em = ERR_PTR(-ENOMEM);
209 goto out;
210 }
211 em->start = 0;
212 em->len = (u64)-1;
213 em->block_len = (u64)-1;
214 em->block_start = 0;
215 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
216
217 write_lock(&em_tree->lock);
218 ret = add_extent_mapping(em_tree, em);
219 if (ret == -EEXIST) {
220 u64 failed_start = em->start;
221 u64 failed_len = em->len;
222
223 free_extent_map(em);
224 em = lookup_extent_mapping(em_tree, start, len);
225 if (em) {
226 ret = 0;
227 } else {
228 em = lookup_extent_mapping(em_tree, failed_start,
229 failed_len);
230 ret = -EIO;
231 }
232 } else if (ret) {
233 free_extent_map(em);
234 em = NULL;
235 }
236 write_unlock(&em_tree->lock);
237
238 if (ret)
239 em = ERR_PTR(ret);
240out:
241 return em;
242}
243
244u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
245{
246 return crc32c(seed, data, len);
247}
248
249void btrfs_csum_final(u32 crc, char *result)
250{
251 put_unaligned_le32(~crc, result);
252}
253
254/*
255 * compute the csum for a btree block, and either verify it or write it
256 * into the csum field of the block.
257 */
258static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
259 int verify)
260{
261 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
262 char *result = NULL;
263 unsigned long len;
264 unsigned long cur_len;
265 unsigned long offset = BTRFS_CSUM_SIZE;
266 char *kaddr;
267 unsigned long map_start;
268 unsigned long map_len;
269 int err;
270 u32 crc = ~(u32)0;
271 unsigned long inline_result;
272
273 len = buf->len - offset;
274 while (len > 0) {
275 err = map_private_extent_buffer(buf, offset, 32,
276 &kaddr, &map_start, &map_len);
277 if (err)
278 return 1;
279 cur_len = min(len, map_len - (offset - map_start));
280 crc = btrfs_csum_data(root, kaddr + offset - map_start,
281 crc, cur_len);
282 len -= cur_len;
283 offset += cur_len;
284 }
285 if (csum_size > sizeof(inline_result)) {
286 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
287 if (!result)
288 return 1;
289 } else {
290 result = (char *)&inline_result;
291 }
292
293 btrfs_csum_final(crc, result);
294
295 if (verify) {
296 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
297 u32 val;
298 u32 found = 0;
299 memcpy(&found, result, csum_size);
300
301 read_extent_buffer(buf, &val, 0, csum_size);
302 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
303 "failed on %llu wanted %X found %X "
304 "level %d\n",
305 root->fs_info->sb->s_id,
306 (unsigned long long)buf->start, val, found,
307 btrfs_header_level(buf));
308 if (result != (char *)&inline_result)
309 kfree(result);
310 return 1;
311 }
312 } else {
313 write_extent_buffer(buf, result, 0, csum_size);
314 }
315 if (result != (char *)&inline_result)
316 kfree(result);
317 return 0;
318}
319
320/*
321 * we can't consider a given block up to date unless the transid of the
322 * block matches the transid in the parent node's pointer. This is how we
323 * detect blocks that either didn't get written at all or got written
324 * in the wrong place.
325 */
326static int verify_parent_transid(struct extent_io_tree *io_tree,
327 struct extent_buffer *eb, u64 parent_transid,
328 int atomic)
329{
330 struct extent_state *cached_state = NULL;
331 int ret;
332
333 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334 return 0;
335
336 if (atomic)
337 return -EAGAIN;
338
339 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
340 0, &cached_state);
341 if (extent_buffer_uptodate(eb) &&
342 btrfs_header_generation(eb) == parent_transid) {
343 ret = 0;
344 goto out;
345 }
346 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
347 "found %llu\n",
348 (unsigned long long)eb->start,
349 (unsigned long long)parent_transid,
350 (unsigned long long)btrfs_header_generation(eb));
351 ret = 1;
352 clear_extent_buffer_uptodate(eb);
353out:
354 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
355 &cached_state, GFP_NOFS);
356 return ret;
357}
358
359/*
360 * helper to read a given tree block, doing retries as required when
361 * the checksums don't match and we have alternate mirrors to try.
362 */
363static int btree_read_extent_buffer_pages(struct btrfs_root *root,
364 struct extent_buffer *eb,
365 u64 start, u64 parent_transid)
366{
367 struct extent_io_tree *io_tree;
368 int failed = 0;
369 int ret;
370 int num_copies = 0;
371 int mirror_num = 0;
372 int failed_mirror = 0;
373
374 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
375 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
376 while (1) {
377 ret = read_extent_buffer_pages(io_tree, eb, start,
378 WAIT_COMPLETE,
379 btree_get_extent, mirror_num);
380 if (!ret && !verify_parent_transid(io_tree, eb,
381 parent_transid, 0))
382 break;
383
384 /*
385 * This buffer's crc is fine, but its contents are corrupted, so
386 * there is no reason to read the other copies, they won't be
387 * any less wrong.
388 */
389 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
390 break;
391
392 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
393 eb->start, eb->len);
394 if (num_copies == 1)
395 break;
396
397 if (!failed_mirror) {
398 failed = 1;
399 failed_mirror = eb->read_mirror;
400 }
401
402 mirror_num++;
403 if (mirror_num == failed_mirror)
404 mirror_num++;
405
406 if (mirror_num > num_copies)
407 break;
408 }
409
410 if (failed && !ret)
411 repair_eb_io_failure(root, eb, failed_mirror);
412
413 return ret;
414}
415
416/*
417 * checksum a dirty tree block before IO. This has extra checks to make sure
418 * we only fill in the checksum field in the first page of a multi-page block
419 */
420
421static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
422{
423 struct extent_io_tree *tree;
424 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
425 u64 found_start;
426 struct extent_buffer *eb;
427
428 tree = &BTRFS_I(page->mapping->host)->io_tree;
429
430 eb = (struct extent_buffer *)page->private;
431 if (page != eb->pages[0])
432 return 0;
433 found_start = btrfs_header_bytenr(eb);
434 if (found_start != start) {
435 WARN_ON(1);
436 return 0;
437 }
438 if (eb->pages[0] != page) {
439 WARN_ON(1);
440 return 0;
441 }
442 if (!PageUptodate(page)) {
443 WARN_ON(1);
444 return 0;
445 }
446 csum_tree_block(root, eb, 0);
447 return 0;
448}
449
450static int check_tree_block_fsid(struct btrfs_root *root,
451 struct extent_buffer *eb)
452{
453 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
454 u8 fsid[BTRFS_UUID_SIZE];
455 int ret = 1;
456
457 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
458 BTRFS_FSID_SIZE);
459 while (fs_devices) {
460 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
461 ret = 0;
462 break;
463 }
464 fs_devices = fs_devices->seed;
465 }
466 return ret;
467}
468
469#define CORRUPT(reason, eb, root, slot) \
470 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
471 "root=%llu, slot=%d\n", reason, \
472 (unsigned long long)btrfs_header_bytenr(eb), \
473 (unsigned long long)root->objectid, slot)
474
475static noinline int check_leaf(struct btrfs_root *root,
476 struct extent_buffer *leaf)
477{
478 struct btrfs_key key;
479 struct btrfs_key leaf_key;
480 u32 nritems = btrfs_header_nritems(leaf);
481 int slot;
482
483 if (nritems == 0)
484 return 0;
485
486 /* Check the 0 item */
487 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
488 BTRFS_LEAF_DATA_SIZE(root)) {
489 CORRUPT("invalid item offset size pair", leaf, root, 0);
490 return -EIO;
491 }
492
493 /*
494 * Check to make sure each items keys are in the correct order and their
495 * offsets make sense. We only have to loop through nritems-1 because
496 * we check the current slot against the next slot, which verifies the
497 * next slot's offset+size makes sense and that the current's slot
498 * offset is correct.
499 */
500 for (slot = 0; slot < nritems - 1; slot++) {
501 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
502 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
503
504 /* Make sure the keys are in the right order */
505 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
506 CORRUPT("bad key order", leaf, root, slot);
507 return -EIO;
508 }
509
510 /*
511 * Make sure the offset and ends are right, remember that the
512 * item data starts at the end of the leaf and grows towards the
513 * front.
514 */
515 if (btrfs_item_offset_nr(leaf, slot) !=
516 btrfs_item_end_nr(leaf, slot + 1)) {
517 CORRUPT("slot offset bad", leaf, root, slot);
518 return -EIO;
519 }
520
521 /*
522 * Check to make sure that we don't point outside of the leaf,
523 * just incase all the items are consistent to eachother, but
524 * all point outside of the leaf.
525 */
526 if (btrfs_item_end_nr(leaf, slot) >
527 BTRFS_LEAF_DATA_SIZE(root)) {
528 CORRUPT("slot end outside of leaf", leaf, root, slot);
529 return -EIO;
530 }
531 }
532
533 return 0;
534}
535
536struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
537 struct page *page, int max_walk)
538{
539 struct extent_buffer *eb;
540 u64 start = page_offset(page);
541 u64 target = start;
542 u64 min_start;
543
544 if (start < max_walk)
545 min_start = 0;
546 else
547 min_start = start - max_walk;
548
549 while (start >= min_start) {
550 eb = find_extent_buffer(tree, start, 0);
551 if (eb) {
552 /*
553 * we found an extent buffer and it contains our page
554 * horray!
555 */
556 if (eb->start <= target &&
557 eb->start + eb->len > target)
558 return eb;
559
560 /* we found an extent buffer that wasn't for us */
561 free_extent_buffer(eb);
562 return NULL;
563 }
564 if (start == 0)
565 break;
566 start -= PAGE_CACHE_SIZE;
567 }
568 return NULL;
569}
570
571static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
572 struct extent_state *state, int mirror)
573{
574 struct extent_io_tree *tree;
575 u64 found_start;
576 int found_level;
577 struct extent_buffer *eb;
578 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
579 int ret = 0;
580 int reads_done;
581
582 if (!page->private)
583 goto out;
584
585 tree = &BTRFS_I(page->mapping->host)->io_tree;
586 eb = (struct extent_buffer *)page->private;
587
588 /* the pending IO might have been the only thing that kept this buffer
589 * in memory. Make sure we have a ref for all this other checks
590 */
591 extent_buffer_get(eb);
592
593 reads_done = atomic_dec_and_test(&eb->io_pages);
594 if (!reads_done)
595 goto err;
596
597 eb->read_mirror = mirror;
598 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
599 ret = -EIO;
600 goto err;
601 }
602
603 found_start = btrfs_header_bytenr(eb);
604 if (found_start != eb->start) {
605 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
606 "%llu %llu\n",
607 (unsigned long long)found_start,
608 (unsigned long long)eb->start);
609 ret = -EIO;
610 goto err;
611 }
612 if (check_tree_block_fsid(root, eb)) {
613 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
614 (unsigned long long)eb->start);
615 ret = -EIO;
616 goto err;
617 }
618 found_level = btrfs_header_level(eb);
619
620 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
621 eb, found_level);
622
623 ret = csum_tree_block(root, eb, 1);
624 if (ret) {
625 ret = -EIO;
626 goto err;
627 }
628
629 /*
630 * If this is a leaf block and it is corrupt, set the corrupt bit so
631 * that we don't try and read the other copies of this block, just
632 * return -EIO.
633 */
634 if (found_level == 0 && check_leaf(root, eb)) {
635 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
636 ret = -EIO;
637 }
638
639 if (!ret)
640 set_extent_buffer_uptodate(eb);
641err:
642 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
643 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
644 btree_readahead_hook(root, eb, eb->start, ret);
645 }
646
647 if (ret)
648 clear_extent_buffer_uptodate(eb);
649 free_extent_buffer(eb);
650out:
651 return ret;
652}
653
654static int btree_io_failed_hook(struct page *page, int failed_mirror)
655{
656 struct extent_buffer *eb;
657 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
658
659 eb = (struct extent_buffer *)page->private;
660 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
661 eb->read_mirror = failed_mirror;
662 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
663 btree_readahead_hook(root, eb, eb->start, -EIO);
664 return -EIO; /* we fixed nothing */
665}
666
667static void end_workqueue_bio(struct bio *bio, int err)
668{
669 struct end_io_wq *end_io_wq = bio->bi_private;
670 struct btrfs_fs_info *fs_info;
671
672 fs_info = end_io_wq->info;
673 end_io_wq->error = err;
674 end_io_wq->work.func = end_workqueue_fn;
675 end_io_wq->work.flags = 0;
676
677 if (bio->bi_rw & REQ_WRITE) {
678 if (end_io_wq->metadata == 1)
679 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
680 &end_io_wq->work);
681 else if (end_io_wq->metadata == 2)
682 btrfs_queue_worker(&fs_info->endio_freespace_worker,
683 &end_io_wq->work);
684 else
685 btrfs_queue_worker(&fs_info->endio_write_workers,
686 &end_io_wq->work);
687 } else {
688 if (end_io_wq->metadata)
689 btrfs_queue_worker(&fs_info->endio_meta_workers,
690 &end_io_wq->work);
691 else
692 btrfs_queue_worker(&fs_info->endio_workers,
693 &end_io_wq->work);
694 }
695}
696
697/*
698 * For the metadata arg you want
699 *
700 * 0 - if data
701 * 1 - if normal metadta
702 * 2 - if writing to the free space cache area
703 */
704int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
705 int metadata)
706{
707 struct end_io_wq *end_io_wq;
708 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
709 if (!end_io_wq)
710 return -ENOMEM;
711
712 end_io_wq->private = bio->bi_private;
713 end_io_wq->end_io = bio->bi_end_io;
714 end_io_wq->info = info;
715 end_io_wq->error = 0;
716 end_io_wq->bio = bio;
717 end_io_wq->metadata = metadata;
718
719 bio->bi_private = end_io_wq;
720 bio->bi_end_io = end_workqueue_bio;
721 return 0;
722}
723
724unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
725{
726 unsigned long limit = min_t(unsigned long,
727 info->workers.max_workers,
728 info->fs_devices->open_devices);
729 return 256 * limit;
730}
731
732static void run_one_async_start(struct btrfs_work *work)
733{
734 struct async_submit_bio *async;
735 int ret;
736
737 async = container_of(work, struct async_submit_bio, work);
738 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
739 async->mirror_num, async->bio_flags,
740 async->bio_offset);
741 if (ret)
742 async->error = ret;
743}
744
745static void run_one_async_done(struct btrfs_work *work)
746{
747 struct btrfs_fs_info *fs_info;
748 struct async_submit_bio *async;
749 int limit;
750
751 async = container_of(work, struct async_submit_bio, work);
752 fs_info = BTRFS_I(async->inode)->root->fs_info;
753
754 limit = btrfs_async_submit_limit(fs_info);
755 limit = limit * 2 / 3;
756
757 atomic_dec(&fs_info->nr_async_submits);
758
759 if (atomic_read(&fs_info->nr_async_submits) < limit &&
760 waitqueue_active(&fs_info->async_submit_wait))
761 wake_up(&fs_info->async_submit_wait);
762
763 /* If an error occured we just want to clean up the bio and move on */
764 if (async->error) {
765 bio_endio(async->bio, async->error);
766 return;
767 }
768
769 async->submit_bio_done(async->inode, async->rw, async->bio,
770 async->mirror_num, async->bio_flags,
771 async->bio_offset);
772}
773
774static void run_one_async_free(struct btrfs_work *work)
775{
776 struct async_submit_bio *async;
777
778 async = container_of(work, struct async_submit_bio, work);
779 kfree(async);
780}
781
782int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
783 int rw, struct bio *bio, int mirror_num,
784 unsigned long bio_flags,
785 u64 bio_offset,
786 extent_submit_bio_hook_t *submit_bio_start,
787 extent_submit_bio_hook_t *submit_bio_done)
788{
789 struct async_submit_bio *async;
790
791 async = kmalloc(sizeof(*async), GFP_NOFS);
792 if (!async)
793 return -ENOMEM;
794
795 async->inode = inode;
796 async->rw = rw;
797 async->bio = bio;
798 async->mirror_num = mirror_num;
799 async->submit_bio_start = submit_bio_start;
800 async->submit_bio_done = submit_bio_done;
801
802 async->work.func = run_one_async_start;
803 async->work.ordered_func = run_one_async_done;
804 async->work.ordered_free = run_one_async_free;
805
806 async->work.flags = 0;
807 async->bio_flags = bio_flags;
808 async->bio_offset = bio_offset;
809
810 async->error = 0;
811
812 atomic_inc(&fs_info->nr_async_submits);
813
814 if (rw & REQ_SYNC)
815 btrfs_set_work_high_prio(&async->work);
816
817 btrfs_queue_worker(&fs_info->workers, &async->work);
818
819 while (atomic_read(&fs_info->async_submit_draining) &&
820 atomic_read(&fs_info->nr_async_submits)) {
821 wait_event(fs_info->async_submit_wait,
822 (atomic_read(&fs_info->nr_async_submits) == 0));
823 }
824
825 return 0;
826}
827
828static int btree_csum_one_bio(struct bio *bio)
829{
830 struct bio_vec *bvec = bio->bi_io_vec;
831 int bio_index = 0;
832 struct btrfs_root *root;
833 int ret = 0;
834
835 WARN_ON(bio->bi_vcnt <= 0);
836 while (bio_index < bio->bi_vcnt) {
837 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
838 ret = csum_dirty_buffer(root, bvec->bv_page);
839 if (ret)
840 break;
841 bio_index++;
842 bvec++;
843 }
844 return ret;
845}
846
847static int __btree_submit_bio_start(struct inode *inode, int rw,
848 struct bio *bio, int mirror_num,
849 unsigned long bio_flags,
850 u64 bio_offset)
851{
852 /*
853 * when we're called for a write, we're already in the async
854 * submission context. Just jump into btrfs_map_bio
855 */
856 return btree_csum_one_bio(bio);
857}
858
859static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
860 int mirror_num, unsigned long bio_flags,
861 u64 bio_offset)
862{
863 /*
864 * when we're called for a write, we're already in the async
865 * submission context. Just jump into btrfs_map_bio
866 */
867 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
868}
869
870static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
871 int mirror_num, unsigned long bio_flags,
872 u64 bio_offset)
873{
874 int ret;
875
876 if (!(rw & REQ_WRITE)) {
877
878 /*
879 * called for a read, do the setup so that checksum validation
880 * can happen in the async kernel threads
881 */
882 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
883 bio, 1);
884 if (ret)
885 return ret;
886 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
887 mirror_num, 0);
888 }
889
890 /*
891 * kthread helpers are used to submit writes so that checksumming
892 * can happen in parallel across all CPUs
893 */
894 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
895 inode, rw, bio, mirror_num, 0,
896 bio_offset,
897 __btree_submit_bio_start,
898 __btree_submit_bio_done);
899}
900
901#ifdef CONFIG_MIGRATION
902static int btree_migratepage(struct address_space *mapping,
903 struct page *newpage, struct page *page,
904 enum migrate_mode mode)
905{
906 /*
907 * we can't safely write a btree page from here,
908 * we haven't done the locking hook
909 */
910 if (PageDirty(page))
911 return -EAGAIN;
912 /*
913 * Buffers may be managed in a filesystem specific way.
914 * We must have no buffers or drop them.
915 */
916 if (page_has_private(page) &&
917 !try_to_release_page(page, GFP_KERNEL))
918 return -EAGAIN;
919 return migrate_page(mapping, newpage, page, mode);
920}
921#endif
922
923
924static int btree_writepages(struct address_space *mapping,
925 struct writeback_control *wbc)
926{
927 struct extent_io_tree *tree;
928 tree = &BTRFS_I(mapping->host)->io_tree;
929 if (wbc->sync_mode == WB_SYNC_NONE) {
930 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
931 u64 num_dirty;
932 unsigned long thresh = 32 * 1024 * 1024;
933
934 if (wbc->for_kupdate)
935 return 0;
936
937 /* this is a bit racy, but that's ok */
938 num_dirty = root->fs_info->dirty_metadata_bytes;
939 if (num_dirty < thresh)
940 return 0;
941 }
942 return btree_write_cache_pages(mapping, wbc);
943}
944
945static int btree_readpage(struct file *file, struct page *page)
946{
947 struct extent_io_tree *tree;
948 tree = &BTRFS_I(page->mapping->host)->io_tree;
949 return extent_read_full_page(tree, page, btree_get_extent, 0);
950}
951
952static int btree_releasepage(struct page *page, gfp_t gfp_flags)
953{
954 if (PageWriteback(page) || PageDirty(page))
955 return 0;
956 /*
957 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
958 * slab allocation from alloc_extent_state down the callchain where
959 * it'd hit a BUG_ON as those flags are not allowed.
960 */
961 gfp_flags &= ~GFP_SLAB_BUG_MASK;
962
963 return try_release_extent_buffer(page, gfp_flags);
964}
965
966static void btree_invalidatepage(struct page *page, unsigned long offset)
967{
968 struct extent_io_tree *tree;
969 tree = &BTRFS_I(page->mapping->host)->io_tree;
970 extent_invalidatepage(tree, page, offset);
971 btree_releasepage(page, GFP_NOFS);
972 if (PagePrivate(page)) {
973 printk(KERN_WARNING "btrfs warning page private not zero "
974 "on page %llu\n", (unsigned long long)page_offset(page));
975 ClearPagePrivate(page);
976 set_page_private(page, 0);
977 page_cache_release(page);
978 }
979}
980
981static int btree_set_page_dirty(struct page *page)
982{
983 struct extent_buffer *eb;
984
985 BUG_ON(!PagePrivate(page));
986 eb = (struct extent_buffer *)page->private;
987 BUG_ON(!eb);
988 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
989 BUG_ON(!atomic_read(&eb->refs));
990 btrfs_assert_tree_locked(eb);
991 return __set_page_dirty_nobuffers(page);
992}
993
994static const struct address_space_operations btree_aops = {
995 .readpage = btree_readpage,
996 .writepages = btree_writepages,
997 .releasepage = btree_releasepage,
998 .invalidatepage = btree_invalidatepage,
999#ifdef CONFIG_MIGRATION
1000 .migratepage = btree_migratepage,
1001#endif
1002 .set_page_dirty = btree_set_page_dirty,
1003};
1004
1005int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1006 u64 parent_transid)
1007{
1008 struct extent_buffer *buf = NULL;
1009 struct inode *btree_inode = root->fs_info->btree_inode;
1010 int ret = 0;
1011
1012 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1013 if (!buf)
1014 return 0;
1015 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1016 buf, 0, WAIT_NONE, btree_get_extent, 0);
1017 free_extent_buffer(buf);
1018 return ret;
1019}
1020
1021int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1022 int mirror_num, struct extent_buffer **eb)
1023{
1024 struct extent_buffer *buf = NULL;
1025 struct inode *btree_inode = root->fs_info->btree_inode;
1026 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1027 int ret;
1028
1029 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1030 if (!buf)
1031 return 0;
1032
1033 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1034
1035 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1036 btree_get_extent, mirror_num);
1037 if (ret) {
1038 free_extent_buffer(buf);
1039 return ret;
1040 }
1041
1042 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1043 free_extent_buffer(buf);
1044 return -EIO;
1045 } else if (extent_buffer_uptodate(buf)) {
1046 *eb = buf;
1047 } else {
1048 free_extent_buffer(buf);
1049 }
1050 return 0;
1051}
1052
1053struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1054 u64 bytenr, u32 blocksize)
1055{
1056 struct inode *btree_inode = root->fs_info->btree_inode;
1057 struct extent_buffer *eb;
1058 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1059 bytenr, blocksize);
1060 return eb;
1061}
1062
1063struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1064 u64 bytenr, u32 blocksize)
1065{
1066 struct inode *btree_inode = root->fs_info->btree_inode;
1067 struct extent_buffer *eb;
1068
1069 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1070 bytenr, blocksize);
1071 return eb;
1072}
1073
1074
1075int btrfs_write_tree_block(struct extent_buffer *buf)
1076{
1077 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1078 buf->start + buf->len - 1);
1079}
1080
1081int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1082{
1083 return filemap_fdatawait_range(buf->pages[0]->mapping,
1084 buf->start, buf->start + buf->len - 1);
1085}
1086
1087struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1088 u32 blocksize, u64 parent_transid)
1089{
1090 struct extent_buffer *buf = NULL;
1091 int ret;
1092
1093 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1094 if (!buf)
1095 return NULL;
1096
1097 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1098 return buf;
1099
1100}
1101
1102void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1103 struct extent_buffer *buf)
1104{
1105 if (btrfs_header_generation(buf) ==
1106 root->fs_info->running_transaction->transid) {
1107 btrfs_assert_tree_locked(buf);
1108
1109 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1110 spin_lock(&root->fs_info->delalloc_lock);
1111 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1112 root->fs_info->dirty_metadata_bytes -= buf->len;
1113 else {
1114 spin_unlock(&root->fs_info->delalloc_lock);
1115 btrfs_panic(root->fs_info, -EOVERFLOW,
1116 "Can't clear %lu bytes from "
1117 " dirty_mdatadata_bytes (%lu)",
1118 buf->len,
1119 root->fs_info->dirty_metadata_bytes);
1120 }
1121 spin_unlock(&root->fs_info->delalloc_lock);
1122 }
1123
1124 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1125 btrfs_set_lock_blocking(buf);
1126 clear_extent_buffer_dirty(buf);
1127 }
1128}
1129
1130static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1131 u32 stripesize, struct btrfs_root *root,
1132 struct btrfs_fs_info *fs_info,
1133 u64 objectid)
1134{
1135 root->node = NULL;
1136 root->commit_root = NULL;
1137 root->sectorsize = sectorsize;
1138 root->nodesize = nodesize;
1139 root->leafsize = leafsize;
1140 root->stripesize = stripesize;
1141 root->ref_cows = 0;
1142 root->track_dirty = 0;
1143 root->in_radix = 0;
1144 root->orphan_item_inserted = 0;
1145 root->orphan_cleanup_state = 0;
1146
1147 root->objectid = objectid;
1148 root->last_trans = 0;
1149 root->highest_objectid = 0;
1150 root->name = NULL;
1151 root->inode_tree = RB_ROOT;
1152 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1153 root->block_rsv = NULL;
1154 root->orphan_block_rsv = NULL;
1155
1156 INIT_LIST_HEAD(&root->dirty_list);
1157 INIT_LIST_HEAD(&root->root_list);
1158 spin_lock_init(&root->orphan_lock);
1159 spin_lock_init(&root->inode_lock);
1160 spin_lock_init(&root->accounting_lock);
1161 mutex_init(&root->objectid_mutex);
1162 mutex_init(&root->log_mutex);
1163 init_waitqueue_head(&root->log_writer_wait);
1164 init_waitqueue_head(&root->log_commit_wait[0]);
1165 init_waitqueue_head(&root->log_commit_wait[1]);
1166 atomic_set(&root->log_commit[0], 0);
1167 atomic_set(&root->log_commit[1], 0);
1168 atomic_set(&root->log_writers, 0);
1169 atomic_set(&root->orphan_inodes, 0);
1170 root->log_batch = 0;
1171 root->log_transid = 0;
1172 root->last_log_commit = 0;
1173 extent_io_tree_init(&root->dirty_log_pages,
1174 fs_info->btree_inode->i_mapping);
1175
1176 memset(&root->root_key, 0, sizeof(root->root_key));
1177 memset(&root->root_item, 0, sizeof(root->root_item));
1178 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1179 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1180 root->defrag_trans_start = fs_info->generation;
1181 init_completion(&root->kobj_unregister);
1182 root->defrag_running = 0;
1183 root->root_key.objectid = objectid;
1184 root->anon_dev = 0;
1185}
1186
1187static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1188 struct btrfs_fs_info *fs_info,
1189 u64 objectid,
1190 struct btrfs_root *root)
1191{
1192 int ret;
1193 u32 blocksize;
1194 u64 generation;
1195
1196 __setup_root(tree_root->nodesize, tree_root->leafsize,
1197 tree_root->sectorsize, tree_root->stripesize,
1198 root, fs_info, objectid);
1199 ret = btrfs_find_last_root(tree_root, objectid,
1200 &root->root_item, &root->root_key);
1201 if (ret > 0)
1202 return -ENOENT;
1203 else if (ret < 0)
1204 return ret;
1205
1206 generation = btrfs_root_generation(&root->root_item);
1207 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1208 root->commit_root = NULL;
1209 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1210 blocksize, generation);
1211 if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1212 free_extent_buffer(root->node);
1213 root->node = NULL;
1214 return -EIO;
1215 }
1216 root->commit_root = btrfs_root_node(root);
1217 return 0;
1218}
1219
1220static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1221{
1222 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1223 if (root)
1224 root->fs_info = fs_info;
1225 return root;
1226}
1227
1228static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1229 struct btrfs_fs_info *fs_info)
1230{
1231 struct btrfs_root *root;
1232 struct btrfs_root *tree_root = fs_info->tree_root;
1233 struct extent_buffer *leaf;
1234
1235 root = btrfs_alloc_root(fs_info);
1236 if (!root)
1237 return ERR_PTR(-ENOMEM);
1238
1239 __setup_root(tree_root->nodesize, tree_root->leafsize,
1240 tree_root->sectorsize, tree_root->stripesize,
1241 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1242
1243 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1244 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1245 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1246 /*
1247 * log trees do not get reference counted because they go away
1248 * before a real commit is actually done. They do store pointers
1249 * to file data extents, and those reference counts still get
1250 * updated (along with back refs to the log tree).
1251 */
1252 root->ref_cows = 0;
1253
1254 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1255 BTRFS_TREE_LOG_OBJECTID, NULL,
1256 0, 0, 0);
1257 if (IS_ERR(leaf)) {
1258 kfree(root);
1259 return ERR_CAST(leaf);
1260 }
1261
1262 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1263 btrfs_set_header_bytenr(leaf, leaf->start);
1264 btrfs_set_header_generation(leaf, trans->transid);
1265 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1266 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1267 root->node = leaf;
1268
1269 write_extent_buffer(root->node, root->fs_info->fsid,
1270 (unsigned long)btrfs_header_fsid(root->node),
1271 BTRFS_FSID_SIZE);
1272 btrfs_mark_buffer_dirty(root->node);
1273 btrfs_tree_unlock(root->node);
1274 return root;
1275}
1276
1277int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1278 struct btrfs_fs_info *fs_info)
1279{
1280 struct btrfs_root *log_root;
1281
1282 log_root = alloc_log_tree(trans, fs_info);
1283 if (IS_ERR(log_root))
1284 return PTR_ERR(log_root);
1285 WARN_ON(fs_info->log_root_tree);
1286 fs_info->log_root_tree = log_root;
1287 return 0;
1288}
1289
1290int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1291 struct btrfs_root *root)
1292{
1293 struct btrfs_root *log_root;
1294 struct btrfs_inode_item *inode_item;
1295
1296 log_root = alloc_log_tree(trans, root->fs_info);
1297 if (IS_ERR(log_root))
1298 return PTR_ERR(log_root);
1299
1300 log_root->last_trans = trans->transid;
1301 log_root->root_key.offset = root->root_key.objectid;
1302
1303 inode_item = &log_root->root_item.inode;
1304 inode_item->generation = cpu_to_le64(1);
1305 inode_item->size = cpu_to_le64(3);
1306 inode_item->nlink = cpu_to_le32(1);
1307 inode_item->nbytes = cpu_to_le64(root->leafsize);
1308 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1309
1310 btrfs_set_root_node(&log_root->root_item, log_root->node);
1311
1312 WARN_ON(root->log_root);
1313 root->log_root = log_root;
1314 root->log_transid = 0;
1315 root->last_log_commit = 0;
1316 return 0;
1317}
1318
1319struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1320 struct btrfs_key *location)
1321{
1322 struct btrfs_root *root;
1323 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1324 struct btrfs_path *path;
1325 struct extent_buffer *l;
1326 u64 generation;
1327 u32 blocksize;
1328 int ret = 0;
1329
1330 root = btrfs_alloc_root(fs_info);
1331 if (!root)
1332 return ERR_PTR(-ENOMEM);
1333 if (location->offset == (u64)-1) {
1334 ret = find_and_setup_root(tree_root, fs_info,
1335 location->objectid, root);
1336 if (ret) {
1337 kfree(root);
1338 return ERR_PTR(ret);
1339 }
1340 goto out;
1341 }
1342
1343 __setup_root(tree_root->nodesize, tree_root->leafsize,
1344 tree_root->sectorsize, tree_root->stripesize,
1345 root, fs_info, location->objectid);
1346
1347 path = btrfs_alloc_path();
1348 if (!path) {
1349 kfree(root);
1350 return ERR_PTR(-ENOMEM);
1351 }
1352 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1353 if (ret == 0) {
1354 l = path->nodes[0];
1355 read_extent_buffer(l, &root->root_item,
1356 btrfs_item_ptr_offset(l, path->slots[0]),
1357 sizeof(root->root_item));
1358 memcpy(&root->root_key, location, sizeof(*location));
1359 }
1360 btrfs_free_path(path);
1361 if (ret) {
1362 kfree(root);
1363 if (ret > 0)
1364 ret = -ENOENT;
1365 return ERR_PTR(ret);
1366 }
1367
1368 generation = btrfs_root_generation(&root->root_item);
1369 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1370 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1371 blocksize, generation);
1372 root->commit_root = btrfs_root_node(root);
1373 BUG_ON(!root->node); /* -ENOMEM */
1374out:
1375 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1376 root->ref_cows = 1;
1377 btrfs_check_and_init_root_item(&root->root_item);
1378 }
1379
1380 return root;
1381}
1382
1383struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1384 struct btrfs_key *location)
1385{
1386 struct btrfs_root *root;
1387 int ret;
1388
1389 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1390 return fs_info->tree_root;
1391 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1392 return fs_info->extent_root;
1393 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1394 return fs_info->chunk_root;
1395 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1396 return fs_info->dev_root;
1397 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1398 return fs_info->csum_root;
1399again:
1400 spin_lock(&fs_info->fs_roots_radix_lock);
1401 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1402 (unsigned long)location->objectid);
1403 spin_unlock(&fs_info->fs_roots_radix_lock);
1404 if (root)
1405 return root;
1406
1407 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1408 if (IS_ERR(root))
1409 return root;
1410
1411 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1412 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1413 GFP_NOFS);
1414 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1415 ret = -ENOMEM;
1416 goto fail;
1417 }
1418
1419 btrfs_init_free_ino_ctl(root);
1420 mutex_init(&root->fs_commit_mutex);
1421 spin_lock_init(&root->cache_lock);
1422 init_waitqueue_head(&root->cache_wait);
1423
1424 ret = get_anon_bdev(&root->anon_dev);
1425 if (ret)
1426 goto fail;
1427
1428 if (btrfs_root_refs(&root->root_item) == 0) {
1429 ret = -ENOENT;
1430 goto fail;
1431 }
1432
1433 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1434 if (ret < 0)
1435 goto fail;
1436 if (ret == 0)
1437 root->orphan_item_inserted = 1;
1438
1439 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1440 if (ret)
1441 goto fail;
1442
1443 spin_lock(&fs_info->fs_roots_radix_lock);
1444 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1445 (unsigned long)root->root_key.objectid,
1446 root);
1447 if (ret == 0)
1448 root->in_radix = 1;
1449
1450 spin_unlock(&fs_info->fs_roots_radix_lock);
1451 radix_tree_preload_end();
1452 if (ret) {
1453 if (ret == -EEXIST) {
1454 free_fs_root(root);
1455 goto again;
1456 }
1457 goto fail;
1458 }
1459
1460 ret = btrfs_find_dead_roots(fs_info->tree_root,
1461 root->root_key.objectid);
1462 WARN_ON(ret);
1463 return root;
1464fail:
1465 free_fs_root(root);
1466 return ERR_PTR(ret);
1467}
1468
1469static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1470{
1471 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1472 int ret = 0;
1473 struct btrfs_device *device;
1474 struct backing_dev_info *bdi;
1475
1476 rcu_read_lock();
1477 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1478 if (!device->bdev)
1479 continue;
1480 bdi = blk_get_backing_dev_info(device->bdev);
1481 if (bdi && bdi_congested(bdi, bdi_bits)) {
1482 ret = 1;
1483 break;
1484 }
1485 }
1486 rcu_read_unlock();
1487 return ret;
1488}
1489
1490/*
1491 * If this fails, caller must call bdi_destroy() to get rid of the
1492 * bdi again.
1493 */
1494static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1495{
1496 int err;
1497
1498 bdi->capabilities = BDI_CAP_MAP_COPY;
1499 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1500 if (err)
1501 return err;
1502
1503 bdi->ra_pages = default_backing_dev_info.ra_pages;
1504 bdi->congested_fn = btrfs_congested_fn;
1505 bdi->congested_data = info;
1506 return 0;
1507}
1508
1509/*
1510 * called by the kthread helper functions to finally call the bio end_io
1511 * functions. This is where read checksum verification actually happens
1512 */
1513static void end_workqueue_fn(struct btrfs_work *work)
1514{
1515 struct bio *bio;
1516 struct end_io_wq *end_io_wq;
1517 struct btrfs_fs_info *fs_info;
1518 int error;
1519
1520 end_io_wq = container_of(work, struct end_io_wq, work);
1521 bio = end_io_wq->bio;
1522 fs_info = end_io_wq->info;
1523
1524 error = end_io_wq->error;
1525 bio->bi_private = end_io_wq->private;
1526 bio->bi_end_io = end_io_wq->end_io;
1527 kfree(end_io_wq);
1528 bio_endio(bio, error);
1529}
1530
1531static int cleaner_kthread(void *arg)
1532{
1533 struct btrfs_root *root = arg;
1534
1535 do {
1536 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1537
1538 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1539 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1540 btrfs_run_delayed_iputs(root);
1541 btrfs_clean_old_snapshots(root);
1542 mutex_unlock(&root->fs_info->cleaner_mutex);
1543 btrfs_run_defrag_inodes(root->fs_info);
1544 }
1545
1546 if (!try_to_freeze()) {
1547 set_current_state(TASK_INTERRUPTIBLE);
1548 if (!kthread_should_stop())
1549 schedule();
1550 __set_current_state(TASK_RUNNING);
1551 }
1552 } while (!kthread_should_stop());
1553 return 0;
1554}
1555
1556static int transaction_kthread(void *arg)
1557{
1558 struct btrfs_root *root = arg;
1559 struct btrfs_trans_handle *trans;
1560 struct btrfs_transaction *cur;
1561 u64 transid;
1562 unsigned long now;
1563 unsigned long delay;
1564 bool cannot_commit;
1565
1566 do {
1567 cannot_commit = false;
1568 delay = HZ * 30;
1569 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1570 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1571
1572 spin_lock(&root->fs_info->trans_lock);
1573 cur = root->fs_info->running_transaction;
1574 if (!cur) {
1575 spin_unlock(&root->fs_info->trans_lock);
1576 goto sleep;
1577 }
1578
1579 now = get_seconds();
1580 if (!cur->blocked &&
1581 (now < cur->start_time || now - cur->start_time < 30)) {
1582 spin_unlock(&root->fs_info->trans_lock);
1583 delay = HZ * 5;
1584 goto sleep;
1585 }
1586 transid = cur->transid;
1587 spin_unlock(&root->fs_info->trans_lock);
1588
1589 /* If the file system is aborted, this will always fail. */
1590 trans = btrfs_join_transaction(root);
1591 if (IS_ERR(trans)) {
1592 cannot_commit = true;
1593 goto sleep;
1594 }
1595 if (transid == trans->transid) {
1596 btrfs_commit_transaction(trans, root);
1597 } else {
1598 btrfs_end_transaction(trans, root);
1599 }
1600sleep:
1601 wake_up_process(root->fs_info->cleaner_kthread);
1602 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1603
1604 if (!try_to_freeze()) {
1605 set_current_state(TASK_INTERRUPTIBLE);
1606 if (!kthread_should_stop() &&
1607 (!btrfs_transaction_blocked(root->fs_info) ||
1608 cannot_commit))
1609 schedule_timeout(delay);
1610 __set_current_state(TASK_RUNNING);
1611 }
1612 } while (!kthread_should_stop());
1613 return 0;
1614}
1615
1616/*
1617 * this will find the highest generation in the array of
1618 * root backups. The index of the highest array is returned,
1619 * or -1 if we can't find anything.
1620 *
1621 * We check to make sure the array is valid by comparing the
1622 * generation of the latest root in the array with the generation
1623 * in the super block. If they don't match we pitch it.
1624 */
1625static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1626{
1627 u64 cur;
1628 int newest_index = -1;
1629 struct btrfs_root_backup *root_backup;
1630 int i;
1631
1632 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1633 root_backup = info->super_copy->super_roots + i;
1634 cur = btrfs_backup_tree_root_gen(root_backup);
1635 if (cur == newest_gen)
1636 newest_index = i;
1637 }
1638
1639 /* check to see if we actually wrapped around */
1640 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1641 root_backup = info->super_copy->super_roots;
1642 cur = btrfs_backup_tree_root_gen(root_backup);
1643 if (cur == newest_gen)
1644 newest_index = 0;
1645 }
1646 return newest_index;
1647}
1648
1649
1650/*
1651 * find the oldest backup so we know where to store new entries
1652 * in the backup array. This will set the backup_root_index
1653 * field in the fs_info struct
1654 */
1655static void find_oldest_super_backup(struct btrfs_fs_info *info,
1656 u64 newest_gen)
1657{
1658 int newest_index = -1;
1659
1660 newest_index = find_newest_super_backup(info, newest_gen);
1661 /* if there was garbage in there, just move along */
1662 if (newest_index == -1) {
1663 info->backup_root_index = 0;
1664 } else {
1665 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1666 }
1667}
1668
1669/*
1670 * copy all the root pointers into the super backup array.
1671 * this will bump the backup pointer by one when it is
1672 * done
1673 */
1674static void backup_super_roots(struct btrfs_fs_info *info)
1675{
1676 int next_backup;
1677 struct btrfs_root_backup *root_backup;
1678 int last_backup;
1679
1680 next_backup = info->backup_root_index;
1681 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1682 BTRFS_NUM_BACKUP_ROOTS;
1683
1684 /*
1685 * just overwrite the last backup if we're at the same generation
1686 * this happens only at umount
1687 */
1688 root_backup = info->super_for_commit->super_roots + last_backup;
1689 if (btrfs_backup_tree_root_gen(root_backup) ==
1690 btrfs_header_generation(info->tree_root->node))
1691 next_backup = last_backup;
1692
1693 root_backup = info->super_for_commit->super_roots + next_backup;
1694
1695 /*
1696 * make sure all of our padding and empty slots get zero filled
1697 * regardless of which ones we use today
1698 */
1699 memset(root_backup, 0, sizeof(*root_backup));
1700
1701 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1702
1703 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1704 btrfs_set_backup_tree_root_gen(root_backup,
1705 btrfs_header_generation(info->tree_root->node));
1706
1707 btrfs_set_backup_tree_root_level(root_backup,
1708 btrfs_header_level(info->tree_root->node));
1709
1710 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1711 btrfs_set_backup_chunk_root_gen(root_backup,
1712 btrfs_header_generation(info->chunk_root->node));
1713 btrfs_set_backup_chunk_root_level(root_backup,
1714 btrfs_header_level(info->chunk_root->node));
1715
1716 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1717 btrfs_set_backup_extent_root_gen(root_backup,
1718 btrfs_header_generation(info->extent_root->node));
1719 btrfs_set_backup_extent_root_level(root_backup,
1720 btrfs_header_level(info->extent_root->node));
1721
1722 /*
1723 * we might commit during log recovery, which happens before we set
1724 * the fs_root. Make sure it is valid before we fill it in.
1725 */
1726 if (info->fs_root && info->fs_root->node) {
1727 btrfs_set_backup_fs_root(root_backup,
1728 info->fs_root->node->start);
1729 btrfs_set_backup_fs_root_gen(root_backup,
1730 btrfs_header_generation(info->fs_root->node));
1731 btrfs_set_backup_fs_root_level(root_backup,
1732 btrfs_header_level(info->fs_root->node));
1733 }
1734
1735 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1736 btrfs_set_backup_dev_root_gen(root_backup,
1737 btrfs_header_generation(info->dev_root->node));
1738 btrfs_set_backup_dev_root_level(root_backup,
1739 btrfs_header_level(info->dev_root->node));
1740
1741 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1742 btrfs_set_backup_csum_root_gen(root_backup,
1743 btrfs_header_generation(info->csum_root->node));
1744 btrfs_set_backup_csum_root_level(root_backup,
1745 btrfs_header_level(info->csum_root->node));
1746
1747 btrfs_set_backup_total_bytes(root_backup,
1748 btrfs_super_total_bytes(info->super_copy));
1749 btrfs_set_backup_bytes_used(root_backup,
1750 btrfs_super_bytes_used(info->super_copy));
1751 btrfs_set_backup_num_devices(root_backup,
1752 btrfs_super_num_devices(info->super_copy));
1753
1754 /*
1755 * if we don't copy this out to the super_copy, it won't get remembered
1756 * for the next commit
1757 */
1758 memcpy(&info->super_copy->super_roots,
1759 &info->super_for_commit->super_roots,
1760 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1761}
1762
1763/*
1764 * this copies info out of the root backup array and back into
1765 * the in-memory super block. It is meant to help iterate through
1766 * the array, so you send it the number of backups you've already
1767 * tried and the last backup index you used.
1768 *
1769 * this returns -1 when it has tried all the backups
1770 */
1771static noinline int next_root_backup(struct btrfs_fs_info *info,
1772 struct btrfs_super_block *super,
1773 int *num_backups_tried, int *backup_index)
1774{
1775 struct btrfs_root_backup *root_backup;
1776 int newest = *backup_index;
1777
1778 if (*num_backups_tried == 0) {
1779 u64 gen = btrfs_super_generation(super);
1780
1781 newest = find_newest_super_backup(info, gen);
1782 if (newest == -1)
1783 return -1;
1784
1785 *backup_index = newest;
1786 *num_backups_tried = 1;
1787 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1788 /* we've tried all the backups, all done */
1789 return -1;
1790 } else {
1791 /* jump to the next oldest backup */
1792 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1793 BTRFS_NUM_BACKUP_ROOTS;
1794 *backup_index = newest;
1795 *num_backups_tried += 1;
1796 }
1797 root_backup = super->super_roots + newest;
1798
1799 btrfs_set_super_generation(super,
1800 btrfs_backup_tree_root_gen(root_backup));
1801 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1802 btrfs_set_super_root_level(super,
1803 btrfs_backup_tree_root_level(root_backup));
1804 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1805
1806 /*
1807 * fixme: the total bytes and num_devices need to match or we should
1808 * need a fsck
1809 */
1810 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1811 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1812 return 0;
1813}
1814
1815/* helper to cleanup tree roots */
1816static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1817{
1818 free_extent_buffer(info->tree_root->node);
1819 free_extent_buffer(info->tree_root->commit_root);
1820 free_extent_buffer(info->dev_root->node);
1821 free_extent_buffer(info->dev_root->commit_root);
1822 free_extent_buffer(info->extent_root->node);
1823 free_extent_buffer(info->extent_root->commit_root);
1824 free_extent_buffer(info->csum_root->node);
1825 free_extent_buffer(info->csum_root->commit_root);
1826
1827 info->tree_root->node = NULL;
1828 info->tree_root->commit_root = NULL;
1829 info->dev_root->node = NULL;
1830 info->dev_root->commit_root = NULL;
1831 info->extent_root->node = NULL;
1832 info->extent_root->commit_root = NULL;
1833 info->csum_root->node = NULL;
1834 info->csum_root->commit_root = NULL;
1835
1836 if (chunk_root) {
1837 free_extent_buffer(info->chunk_root->node);
1838 free_extent_buffer(info->chunk_root->commit_root);
1839 info->chunk_root->node = NULL;
1840 info->chunk_root->commit_root = NULL;
1841 }
1842}
1843
1844
1845int open_ctree(struct super_block *sb,
1846 struct btrfs_fs_devices *fs_devices,
1847 char *options)
1848{
1849 u32 sectorsize;
1850 u32 nodesize;
1851 u32 leafsize;
1852 u32 blocksize;
1853 u32 stripesize;
1854 u64 generation;
1855 u64 features;
1856 struct btrfs_key location;
1857 struct buffer_head *bh;
1858 struct btrfs_super_block *disk_super;
1859 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1860 struct btrfs_root *tree_root;
1861 struct btrfs_root *extent_root;
1862 struct btrfs_root *csum_root;
1863 struct btrfs_root *chunk_root;
1864 struct btrfs_root *dev_root;
1865 struct btrfs_root *log_tree_root;
1866 int ret;
1867 int err = -EINVAL;
1868 int num_backups_tried = 0;
1869 int backup_index = 0;
1870
1871 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1872 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1873 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1874 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1875 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1876
1877 if (!tree_root || !extent_root || !csum_root ||
1878 !chunk_root || !dev_root) {
1879 err = -ENOMEM;
1880 goto fail;
1881 }
1882
1883 ret = init_srcu_struct(&fs_info->subvol_srcu);
1884 if (ret) {
1885 err = ret;
1886 goto fail;
1887 }
1888
1889 ret = setup_bdi(fs_info, &fs_info->bdi);
1890 if (ret) {
1891 err = ret;
1892 goto fail_srcu;
1893 }
1894
1895 fs_info->btree_inode = new_inode(sb);
1896 if (!fs_info->btree_inode) {
1897 err = -ENOMEM;
1898 goto fail_bdi;
1899 }
1900
1901 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1902
1903 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1904 INIT_LIST_HEAD(&fs_info->trans_list);
1905 INIT_LIST_HEAD(&fs_info->dead_roots);
1906 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1907 INIT_LIST_HEAD(&fs_info->hashers);
1908 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1909 INIT_LIST_HEAD(&fs_info->ordered_operations);
1910 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1911 spin_lock_init(&fs_info->delalloc_lock);
1912 spin_lock_init(&fs_info->trans_lock);
1913 spin_lock_init(&fs_info->ref_cache_lock);
1914 spin_lock_init(&fs_info->fs_roots_radix_lock);
1915 spin_lock_init(&fs_info->delayed_iput_lock);
1916 spin_lock_init(&fs_info->defrag_inodes_lock);
1917 spin_lock_init(&fs_info->free_chunk_lock);
1918 spin_lock_init(&fs_info->tree_mod_seq_lock);
1919 rwlock_init(&fs_info->tree_mod_log_lock);
1920 mutex_init(&fs_info->reloc_mutex);
1921
1922 init_completion(&fs_info->kobj_unregister);
1923 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1924 INIT_LIST_HEAD(&fs_info->space_info);
1925 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
1926 btrfs_mapping_init(&fs_info->mapping_tree);
1927 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1928 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1929 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1930 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1931 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1932 btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1933 atomic_set(&fs_info->nr_async_submits, 0);
1934 atomic_set(&fs_info->async_delalloc_pages, 0);
1935 atomic_set(&fs_info->async_submit_draining, 0);
1936 atomic_set(&fs_info->nr_async_bios, 0);
1937 atomic_set(&fs_info->defrag_running, 0);
1938 atomic_set(&fs_info->tree_mod_seq, 0);
1939 fs_info->sb = sb;
1940 fs_info->max_inline = 8192 * 1024;
1941 fs_info->metadata_ratio = 0;
1942 fs_info->defrag_inodes = RB_ROOT;
1943 fs_info->trans_no_join = 0;
1944 fs_info->free_chunk_space = 0;
1945 fs_info->tree_mod_log = RB_ROOT;
1946
1947 /* readahead state */
1948 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1949 spin_lock_init(&fs_info->reada_lock);
1950
1951 fs_info->thread_pool_size = min_t(unsigned long,
1952 num_online_cpus() + 2, 8);
1953
1954 INIT_LIST_HEAD(&fs_info->ordered_extents);
1955 spin_lock_init(&fs_info->ordered_extent_lock);
1956 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1957 GFP_NOFS);
1958 if (!fs_info->delayed_root) {
1959 err = -ENOMEM;
1960 goto fail_iput;
1961 }
1962 btrfs_init_delayed_root(fs_info->delayed_root);
1963
1964 mutex_init(&fs_info->scrub_lock);
1965 atomic_set(&fs_info->scrubs_running, 0);
1966 atomic_set(&fs_info->scrub_pause_req, 0);
1967 atomic_set(&fs_info->scrubs_paused, 0);
1968 atomic_set(&fs_info->scrub_cancel_req, 0);
1969 init_waitqueue_head(&fs_info->scrub_pause_wait);
1970 init_rwsem(&fs_info->scrub_super_lock);
1971 fs_info->scrub_workers_refcnt = 0;
1972#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1973 fs_info->check_integrity_print_mask = 0;
1974#endif
1975
1976 spin_lock_init(&fs_info->balance_lock);
1977 mutex_init(&fs_info->balance_mutex);
1978 atomic_set(&fs_info->balance_running, 0);
1979 atomic_set(&fs_info->balance_pause_req, 0);
1980 atomic_set(&fs_info->balance_cancel_req, 0);
1981 fs_info->balance_ctl = NULL;
1982 init_waitqueue_head(&fs_info->balance_wait_q);
1983
1984 sb->s_blocksize = 4096;
1985 sb->s_blocksize_bits = blksize_bits(4096);
1986 sb->s_bdi = &fs_info->bdi;
1987
1988 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1989 set_nlink(fs_info->btree_inode, 1);
1990 /*
1991 * we set the i_size on the btree inode to the max possible int.
1992 * the real end of the address space is determined by all of
1993 * the devices in the system
1994 */
1995 fs_info->btree_inode->i_size = OFFSET_MAX;
1996 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1997 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1998
1999 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2000 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2001 fs_info->btree_inode->i_mapping);
2002 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2003 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2004
2005 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2006
2007 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2008 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2009 sizeof(struct btrfs_key));
2010 set_bit(BTRFS_INODE_DUMMY,
2011 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2012 insert_inode_hash(fs_info->btree_inode);
2013
2014 spin_lock_init(&fs_info->block_group_cache_lock);
2015 fs_info->block_group_cache_tree = RB_ROOT;
2016
2017 extent_io_tree_init(&fs_info->freed_extents[0],
2018 fs_info->btree_inode->i_mapping);
2019 extent_io_tree_init(&fs_info->freed_extents[1],
2020 fs_info->btree_inode->i_mapping);
2021 fs_info->pinned_extents = &fs_info->freed_extents[0];
2022 fs_info->do_barriers = 1;
2023
2024
2025 mutex_init(&fs_info->ordered_operations_mutex);
2026 mutex_init(&fs_info->tree_log_mutex);
2027 mutex_init(&fs_info->chunk_mutex);
2028 mutex_init(&fs_info->transaction_kthread_mutex);
2029 mutex_init(&fs_info->cleaner_mutex);
2030 mutex_init(&fs_info->volume_mutex);
2031 init_rwsem(&fs_info->extent_commit_sem);
2032 init_rwsem(&fs_info->cleanup_work_sem);
2033 init_rwsem(&fs_info->subvol_sem);
2034
2035 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2036 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2037
2038 init_waitqueue_head(&fs_info->transaction_throttle);
2039 init_waitqueue_head(&fs_info->transaction_wait);
2040 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2041 init_waitqueue_head(&fs_info->async_submit_wait);
2042
2043 __setup_root(4096, 4096, 4096, 4096, tree_root,
2044 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2045
2046 invalidate_bdev(fs_devices->latest_bdev);
2047 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2048 if (!bh) {
2049 err = -EINVAL;
2050 goto fail_alloc;
2051 }
2052
2053 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2054 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2055 sizeof(*fs_info->super_for_commit));
2056 brelse(bh);
2057
2058 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2059
2060 disk_super = fs_info->super_copy;
2061 if (!btrfs_super_root(disk_super))
2062 goto fail_alloc;
2063
2064 /* check FS state, whether FS is broken. */
2065 fs_info->fs_state |= btrfs_super_flags(disk_super);
2066
2067 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2068 if (ret) {
2069 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2070 err = ret;
2071 goto fail_alloc;
2072 }
2073
2074 /*
2075 * run through our array of backup supers and setup
2076 * our ring pointer to the oldest one
2077 */
2078 generation = btrfs_super_generation(disk_super);
2079 find_oldest_super_backup(fs_info, generation);
2080
2081 /*
2082 * In the long term, we'll store the compression type in the super
2083 * block, and it'll be used for per file compression control.
2084 */
2085 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2086
2087 ret = btrfs_parse_options(tree_root, options);
2088 if (ret) {
2089 err = ret;
2090 goto fail_alloc;
2091 }
2092
2093 features = btrfs_super_incompat_flags(disk_super) &
2094 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2095 if (features) {
2096 printk(KERN_ERR "BTRFS: couldn't mount because of "
2097 "unsupported optional features (%Lx).\n",
2098 (unsigned long long)features);
2099 err = -EINVAL;
2100 goto fail_alloc;
2101 }
2102
2103 if (btrfs_super_leafsize(disk_super) !=
2104 btrfs_super_nodesize(disk_super)) {
2105 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2106 "blocksizes don't match. node %d leaf %d\n",
2107 btrfs_super_nodesize(disk_super),
2108 btrfs_super_leafsize(disk_super));
2109 err = -EINVAL;
2110 goto fail_alloc;
2111 }
2112 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2113 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2114 "blocksize (%d) was too large\n",
2115 btrfs_super_leafsize(disk_super));
2116 err = -EINVAL;
2117 goto fail_alloc;
2118 }
2119
2120 features = btrfs_super_incompat_flags(disk_super);
2121 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2122 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2123 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2124
2125 /*
2126 * flag our filesystem as having big metadata blocks if
2127 * they are bigger than the page size
2128 */
2129 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2130 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2131 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2132 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2133 }
2134
2135 nodesize = btrfs_super_nodesize(disk_super);
2136 leafsize = btrfs_super_leafsize(disk_super);
2137 sectorsize = btrfs_super_sectorsize(disk_super);
2138 stripesize = btrfs_super_stripesize(disk_super);
2139
2140 /*
2141 * mixed block groups end up with duplicate but slightly offset
2142 * extent buffers for the same range. It leads to corruptions
2143 */
2144 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2145 (sectorsize != leafsize)) {
2146 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2147 "are not allowed for mixed block groups on %s\n",
2148 sb->s_id);
2149 goto fail_alloc;
2150 }
2151
2152 btrfs_set_super_incompat_flags(disk_super, features);
2153
2154 features = btrfs_super_compat_ro_flags(disk_super) &
2155 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2156 if (!(sb->s_flags & MS_RDONLY) && features) {
2157 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2158 "unsupported option features (%Lx).\n",
2159 (unsigned long long)features);
2160 err = -EINVAL;
2161 goto fail_alloc;
2162 }
2163
2164 btrfs_init_workers(&fs_info->generic_worker,
2165 "genwork", 1, NULL);
2166
2167 btrfs_init_workers(&fs_info->workers, "worker",
2168 fs_info->thread_pool_size,
2169 &fs_info->generic_worker);
2170
2171 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2172 fs_info->thread_pool_size,
2173 &fs_info->generic_worker);
2174
2175 btrfs_init_workers(&fs_info->submit_workers, "submit",
2176 min_t(u64, fs_devices->num_devices,
2177 fs_info->thread_pool_size),
2178 &fs_info->generic_worker);
2179
2180 btrfs_init_workers(&fs_info->caching_workers, "cache",
2181 2, &fs_info->generic_worker);
2182
2183 /* a higher idle thresh on the submit workers makes it much more
2184 * likely that bios will be send down in a sane order to the
2185 * devices
2186 */
2187 fs_info->submit_workers.idle_thresh = 64;
2188
2189 fs_info->workers.idle_thresh = 16;
2190 fs_info->workers.ordered = 1;
2191
2192 fs_info->delalloc_workers.idle_thresh = 2;
2193 fs_info->delalloc_workers.ordered = 1;
2194
2195 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2196 &fs_info->generic_worker);
2197 btrfs_init_workers(&fs_info->endio_workers, "endio",
2198 fs_info->thread_pool_size,
2199 &fs_info->generic_worker);
2200 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2201 fs_info->thread_pool_size,
2202 &fs_info->generic_worker);
2203 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2204 "endio-meta-write", fs_info->thread_pool_size,
2205 &fs_info->generic_worker);
2206 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2207 fs_info->thread_pool_size,
2208 &fs_info->generic_worker);
2209 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2210 1, &fs_info->generic_worker);
2211 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2212 fs_info->thread_pool_size,
2213 &fs_info->generic_worker);
2214 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2215 fs_info->thread_pool_size,
2216 &fs_info->generic_worker);
2217
2218 /*
2219 * endios are largely parallel and should have a very
2220 * low idle thresh
2221 */
2222 fs_info->endio_workers.idle_thresh = 4;
2223 fs_info->endio_meta_workers.idle_thresh = 4;
2224
2225 fs_info->endio_write_workers.idle_thresh = 2;
2226 fs_info->endio_meta_write_workers.idle_thresh = 2;
2227 fs_info->readahead_workers.idle_thresh = 2;
2228
2229 /*
2230 * btrfs_start_workers can really only fail because of ENOMEM so just
2231 * return -ENOMEM if any of these fail.
2232 */
2233 ret = btrfs_start_workers(&fs_info->workers);
2234 ret |= btrfs_start_workers(&fs_info->generic_worker);
2235 ret |= btrfs_start_workers(&fs_info->submit_workers);
2236 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2237 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2238 ret |= btrfs_start_workers(&fs_info->endio_workers);
2239 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2240 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2241 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2242 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2243 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2244 ret |= btrfs_start_workers(&fs_info->caching_workers);
2245 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2246 if (ret) {
2247 ret = -ENOMEM;
2248 goto fail_sb_buffer;
2249 }
2250
2251 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2252 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2253 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2254
2255 tree_root->nodesize = nodesize;
2256 tree_root->leafsize = leafsize;
2257 tree_root->sectorsize = sectorsize;
2258 tree_root->stripesize = stripesize;
2259
2260 sb->s_blocksize = sectorsize;
2261 sb->s_blocksize_bits = blksize_bits(sectorsize);
2262
2263 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2264 sizeof(disk_super->magic))) {
2265 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2266 goto fail_sb_buffer;
2267 }
2268
2269 if (sectorsize != PAGE_SIZE) {
2270 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2271 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2272 goto fail_sb_buffer;
2273 }
2274
2275 mutex_lock(&fs_info->chunk_mutex);
2276 ret = btrfs_read_sys_array(tree_root);
2277 mutex_unlock(&fs_info->chunk_mutex);
2278 if (ret) {
2279 printk(KERN_WARNING "btrfs: failed to read the system "
2280 "array on %s\n", sb->s_id);
2281 goto fail_sb_buffer;
2282 }
2283
2284 blocksize = btrfs_level_size(tree_root,
2285 btrfs_super_chunk_root_level(disk_super));
2286 generation = btrfs_super_chunk_root_generation(disk_super);
2287
2288 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2289 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2290
2291 chunk_root->node = read_tree_block(chunk_root,
2292 btrfs_super_chunk_root(disk_super),
2293 blocksize, generation);
2294 BUG_ON(!chunk_root->node); /* -ENOMEM */
2295 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2296 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2297 sb->s_id);
2298 goto fail_tree_roots;
2299 }
2300 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2301 chunk_root->commit_root = btrfs_root_node(chunk_root);
2302
2303 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2304 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2305 BTRFS_UUID_SIZE);
2306
2307 ret = btrfs_read_chunk_tree(chunk_root);
2308 if (ret) {
2309 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2310 sb->s_id);
2311 goto fail_tree_roots;
2312 }
2313
2314 btrfs_close_extra_devices(fs_devices);
2315
2316 if (!fs_devices->latest_bdev) {
2317 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2318 sb->s_id);
2319 goto fail_tree_roots;
2320 }
2321
2322retry_root_backup:
2323 blocksize = btrfs_level_size(tree_root,
2324 btrfs_super_root_level(disk_super));
2325 generation = btrfs_super_generation(disk_super);
2326
2327 tree_root->node = read_tree_block(tree_root,
2328 btrfs_super_root(disk_super),
2329 blocksize, generation);
2330 if (!tree_root->node ||
2331 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2332 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2333 sb->s_id);
2334
2335 goto recovery_tree_root;
2336 }
2337
2338 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2339 tree_root->commit_root = btrfs_root_node(tree_root);
2340
2341 ret = find_and_setup_root(tree_root, fs_info,
2342 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2343 if (ret)
2344 goto recovery_tree_root;
2345 extent_root->track_dirty = 1;
2346
2347 ret = find_and_setup_root(tree_root, fs_info,
2348 BTRFS_DEV_TREE_OBJECTID, dev_root);
2349 if (ret)
2350 goto recovery_tree_root;
2351 dev_root->track_dirty = 1;
2352
2353 ret = find_and_setup_root(tree_root, fs_info,
2354 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2355 if (ret)
2356 goto recovery_tree_root;
2357 csum_root->track_dirty = 1;
2358
2359 fs_info->generation = generation;
2360 fs_info->last_trans_committed = generation;
2361
2362 ret = btrfs_recover_balance(fs_info);
2363 if (ret) {
2364 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2365 goto fail_block_groups;
2366 }
2367
2368 ret = btrfs_init_dev_stats(fs_info);
2369 if (ret) {
2370 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2371 ret);
2372 goto fail_block_groups;
2373 }
2374
2375 ret = btrfs_init_space_info(fs_info);
2376 if (ret) {
2377 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2378 goto fail_block_groups;
2379 }
2380
2381 ret = btrfs_read_block_groups(extent_root);
2382 if (ret) {
2383 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2384 goto fail_block_groups;
2385 }
2386
2387 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2388 "btrfs-cleaner");
2389 if (IS_ERR(fs_info->cleaner_kthread))
2390 goto fail_block_groups;
2391
2392 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2393 tree_root,
2394 "btrfs-transaction");
2395 if (IS_ERR(fs_info->transaction_kthread))
2396 goto fail_cleaner;
2397
2398 if (!btrfs_test_opt(tree_root, SSD) &&
2399 !btrfs_test_opt(tree_root, NOSSD) &&
2400 !fs_info->fs_devices->rotating) {
2401 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2402 "mode\n");
2403 btrfs_set_opt(fs_info->mount_opt, SSD);
2404 }
2405
2406#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2407 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2408 ret = btrfsic_mount(tree_root, fs_devices,
2409 btrfs_test_opt(tree_root,
2410 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2411 1 : 0,
2412 fs_info->check_integrity_print_mask);
2413 if (ret)
2414 printk(KERN_WARNING "btrfs: failed to initialize"
2415 " integrity check module %s\n", sb->s_id);
2416 }
2417#endif
2418
2419 /* do not make disk changes in broken FS */
2420 if (btrfs_super_log_root(disk_super) != 0 &&
2421 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2422 u64 bytenr = btrfs_super_log_root(disk_super);
2423
2424 if (fs_devices->rw_devices == 0) {
2425 printk(KERN_WARNING "Btrfs log replay required "
2426 "on RO media\n");
2427 err = -EIO;
2428 goto fail_trans_kthread;
2429 }
2430 blocksize =
2431 btrfs_level_size(tree_root,
2432 btrfs_super_log_root_level(disk_super));
2433
2434 log_tree_root = btrfs_alloc_root(fs_info);
2435 if (!log_tree_root) {
2436 err = -ENOMEM;
2437 goto fail_trans_kthread;
2438 }
2439
2440 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2441 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2442
2443 log_tree_root->node = read_tree_block(tree_root, bytenr,
2444 blocksize,
2445 generation + 1);
2446 /* returns with log_tree_root freed on success */
2447 ret = btrfs_recover_log_trees(log_tree_root);
2448 if (ret) {
2449 btrfs_error(tree_root->fs_info, ret,
2450 "Failed to recover log tree");
2451 free_extent_buffer(log_tree_root->node);
2452 kfree(log_tree_root);
2453 goto fail_trans_kthread;
2454 }
2455
2456 if (sb->s_flags & MS_RDONLY) {
2457 ret = btrfs_commit_super(tree_root);
2458 if (ret)
2459 goto fail_trans_kthread;
2460 }
2461 }
2462
2463 ret = btrfs_find_orphan_roots(tree_root);
2464 if (ret)
2465 goto fail_trans_kthread;
2466
2467 if (!(sb->s_flags & MS_RDONLY)) {
2468 ret = btrfs_cleanup_fs_roots(fs_info);
2469 if (ret) {
2470 }
2471
2472 ret = btrfs_recover_relocation(tree_root);
2473 if (ret < 0) {
2474 printk(KERN_WARNING
2475 "btrfs: failed to recover relocation\n");
2476 err = -EINVAL;
2477 goto fail_trans_kthread;
2478 }
2479 }
2480
2481 location.objectid = BTRFS_FS_TREE_OBJECTID;
2482 location.type = BTRFS_ROOT_ITEM_KEY;
2483 location.offset = (u64)-1;
2484
2485 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2486 if (!fs_info->fs_root)
2487 goto fail_trans_kthread;
2488 if (IS_ERR(fs_info->fs_root)) {
2489 err = PTR_ERR(fs_info->fs_root);
2490 goto fail_trans_kthread;
2491 }
2492
2493 if (sb->s_flags & MS_RDONLY)
2494 return 0;
2495
2496 down_read(&fs_info->cleanup_work_sem);
2497 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2498 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2499 up_read(&fs_info->cleanup_work_sem);
2500 close_ctree(tree_root);
2501 return ret;
2502 }
2503 up_read(&fs_info->cleanup_work_sem);
2504
2505 ret = btrfs_resume_balance_async(fs_info);
2506 if (ret) {
2507 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2508 close_ctree(tree_root);
2509 return ret;
2510 }
2511
2512 return 0;
2513
2514fail_trans_kthread:
2515 kthread_stop(fs_info->transaction_kthread);
2516fail_cleaner:
2517 kthread_stop(fs_info->cleaner_kthread);
2518
2519 /*
2520 * make sure we're done with the btree inode before we stop our
2521 * kthreads
2522 */
2523 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2524 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2525
2526fail_block_groups:
2527 btrfs_free_block_groups(fs_info);
2528
2529fail_tree_roots:
2530 free_root_pointers(fs_info, 1);
2531
2532fail_sb_buffer:
2533 btrfs_stop_workers(&fs_info->generic_worker);
2534 btrfs_stop_workers(&fs_info->readahead_workers);
2535 btrfs_stop_workers(&fs_info->fixup_workers);
2536 btrfs_stop_workers(&fs_info->delalloc_workers);
2537 btrfs_stop_workers(&fs_info->workers);
2538 btrfs_stop_workers(&fs_info->endio_workers);
2539 btrfs_stop_workers(&fs_info->endio_meta_workers);
2540 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2541 btrfs_stop_workers(&fs_info->endio_write_workers);
2542 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2543 btrfs_stop_workers(&fs_info->submit_workers);
2544 btrfs_stop_workers(&fs_info->delayed_workers);
2545 btrfs_stop_workers(&fs_info->caching_workers);
2546fail_alloc:
2547fail_iput:
2548 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2549
2550 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2551 iput(fs_info->btree_inode);
2552fail_bdi:
2553 bdi_destroy(&fs_info->bdi);
2554fail_srcu:
2555 cleanup_srcu_struct(&fs_info->subvol_srcu);
2556fail:
2557 btrfs_close_devices(fs_info->fs_devices);
2558 return err;
2559
2560recovery_tree_root:
2561 if (!btrfs_test_opt(tree_root, RECOVERY))
2562 goto fail_tree_roots;
2563
2564 free_root_pointers(fs_info, 0);
2565
2566 /* don't use the log in recovery mode, it won't be valid */
2567 btrfs_set_super_log_root(disk_super, 0);
2568
2569 /* we can't trust the free space cache either */
2570 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2571
2572 ret = next_root_backup(fs_info, fs_info->super_copy,
2573 &num_backups_tried, &backup_index);
2574 if (ret == -1)
2575 goto fail_block_groups;
2576 goto retry_root_backup;
2577}
2578
2579static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2580{
2581 if (uptodate) {
2582 set_buffer_uptodate(bh);
2583 } else {
2584 struct btrfs_device *device = (struct btrfs_device *)
2585 bh->b_private;
2586
2587 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2588 "I/O error on %s\n",
2589 rcu_str_deref(device->name));
2590 /* note, we dont' set_buffer_write_io_error because we have
2591 * our own ways of dealing with the IO errors
2592 */
2593 clear_buffer_uptodate(bh);
2594 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2595 }
2596 unlock_buffer(bh);
2597 put_bh(bh);
2598}
2599
2600struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2601{
2602 struct buffer_head *bh;
2603 struct buffer_head *latest = NULL;
2604 struct btrfs_super_block *super;
2605 int i;
2606 u64 transid = 0;
2607 u64 bytenr;
2608
2609 /* we would like to check all the supers, but that would make
2610 * a btrfs mount succeed after a mkfs from a different FS.
2611 * So, we need to add a special mount option to scan for
2612 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2613 */
2614 for (i = 0; i < 1; i++) {
2615 bytenr = btrfs_sb_offset(i);
2616 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2617 break;
2618 bh = __bread(bdev, bytenr / 4096, 4096);
2619 if (!bh)
2620 continue;
2621
2622 super = (struct btrfs_super_block *)bh->b_data;
2623 if (btrfs_super_bytenr(super) != bytenr ||
2624 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2625 sizeof(super->magic))) {
2626 brelse(bh);
2627 continue;
2628 }
2629
2630 if (!latest || btrfs_super_generation(super) > transid) {
2631 brelse(latest);
2632 latest = bh;
2633 transid = btrfs_super_generation(super);
2634 } else {
2635 brelse(bh);
2636 }
2637 }
2638 return latest;
2639}
2640
2641/*
2642 * this should be called twice, once with wait == 0 and
2643 * once with wait == 1. When wait == 0 is done, all the buffer heads
2644 * we write are pinned.
2645 *
2646 * They are released when wait == 1 is done.
2647 * max_mirrors must be the same for both runs, and it indicates how
2648 * many supers on this one device should be written.
2649 *
2650 * max_mirrors == 0 means to write them all.
2651 */
2652static int write_dev_supers(struct btrfs_device *device,
2653 struct btrfs_super_block *sb,
2654 int do_barriers, int wait, int max_mirrors)
2655{
2656 struct buffer_head *bh;
2657 int i;
2658 int ret;
2659 int errors = 0;
2660 u32 crc;
2661 u64 bytenr;
2662
2663 if (max_mirrors == 0)
2664 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2665
2666 for (i = 0; i < max_mirrors; i++) {
2667 bytenr = btrfs_sb_offset(i);
2668 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2669 break;
2670
2671 if (wait) {
2672 bh = __find_get_block(device->bdev, bytenr / 4096,
2673 BTRFS_SUPER_INFO_SIZE);
2674 BUG_ON(!bh);
2675 wait_on_buffer(bh);
2676 if (!buffer_uptodate(bh))
2677 errors++;
2678
2679 /* drop our reference */
2680 brelse(bh);
2681
2682 /* drop the reference from the wait == 0 run */
2683 brelse(bh);
2684 continue;
2685 } else {
2686 btrfs_set_super_bytenr(sb, bytenr);
2687
2688 crc = ~(u32)0;
2689 crc = btrfs_csum_data(NULL, (char *)sb +
2690 BTRFS_CSUM_SIZE, crc,
2691 BTRFS_SUPER_INFO_SIZE -
2692 BTRFS_CSUM_SIZE);
2693 btrfs_csum_final(crc, sb->csum);
2694
2695 /*
2696 * one reference for us, and we leave it for the
2697 * caller
2698 */
2699 bh = __getblk(device->bdev, bytenr / 4096,
2700 BTRFS_SUPER_INFO_SIZE);
2701 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2702
2703 /* one reference for submit_bh */
2704 get_bh(bh);
2705
2706 set_buffer_uptodate(bh);
2707 lock_buffer(bh);
2708 bh->b_end_io = btrfs_end_buffer_write_sync;
2709 bh->b_private = device;
2710 }
2711
2712 /*
2713 * we fua the first super. The others we allow
2714 * to go down lazy.
2715 */
2716 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2717 if (ret)
2718 errors++;
2719 }
2720 return errors < i ? 0 : -1;
2721}
2722
2723/*
2724 * endio for the write_dev_flush, this will wake anyone waiting
2725 * for the barrier when it is done
2726 */
2727static void btrfs_end_empty_barrier(struct bio *bio, int err)
2728{
2729 if (err) {
2730 if (err == -EOPNOTSUPP)
2731 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2732 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2733 }
2734 if (bio->bi_private)
2735 complete(bio->bi_private);
2736 bio_put(bio);
2737}
2738
2739/*
2740 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2741 * sent down. With wait == 1, it waits for the previous flush.
2742 *
2743 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2744 * capable
2745 */
2746static int write_dev_flush(struct btrfs_device *device, int wait)
2747{
2748 struct bio *bio;
2749 int ret = 0;
2750
2751 if (device->nobarriers)
2752 return 0;
2753
2754 if (wait) {
2755 bio = device->flush_bio;
2756 if (!bio)
2757 return 0;
2758
2759 wait_for_completion(&device->flush_wait);
2760
2761 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2762 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2763 rcu_str_deref(device->name));
2764 device->nobarriers = 1;
2765 }
2766 if (!bio_flagged(bio, BIO_UPTODATE)) {
2767 ret = -EIO;
2768 if (!bio_flagged(bio, BIO_EOPNOTSUPP))
2769 btrfs_dev_stat_inc_and_print(device,
2770 BTRFS_DEV_STAT_FLUSH_ERRS);
2771 }
2772
2773 /* drop the reference from the wait == 0 run */
2774 bio_put(bio);
2775 device->flush_bio = NULL;
2776
2777 return ret;
2778 }
2779
2780 /*
2781 * one reference for us, and we leave it for the
2782 * caller
2783 */
2784 device->flush_bio = NULL;
2785 bio = bio_alloc(GFP_NOFS, 0);
2786 if (!bio)
2787 return -ENOMEM;
2788
2789 bio->bi_end_io = btrfs_end_empty_barrier;
2790 bio->bi_bdev = device->bdev;
2791 init_completion(&device->flush_wait);
2792 bio->bi_private = &device->flush_wait;
2793 device->flush_bio = bio;
2794
2795 bio_get(bio);
2796 btrfsic_submit_bio(WRITE_FLUSH, bio);
2797
2798 return 0;
2799}
2800
2801/*
2802 * send an empty flush down to each device in parallel,
2803 * then wait for them
2804 */
2805static int barrier_all_devices(struct btrfs_fs_info *info)
2806{
2807 struct list_head *head;
2808 struct btrfs_device *dev;
2809 int errors = 0;
2810 int ret;
2811
2812 /* send down all the barriers */
2813 head = &info->fs_devices->devices;
2814 list_for_each_entry_rcu(dev, head, dev_list) {
2815 if (!dev->bdev) {
2816 errors++;
2817 continue;
2818 }
2819 if (!dev->in_fs_metadata || !dev->writeable)
2820 continue;
2821
2822 ret = write_dev_flush(dev, 0);
2823 if (ret)
2824 errors++;
2825 }
2826
2827 /* wait for all the barriers */
2828 list_for_each_entry_rcu(dev, head, dev_list) {
2829 if (!dev->bdev) {
2830 errors++;
2831 continue;
2832 }
2833 if (!dev->in_fs_metadata || !dev->writeable)
2834 continue;
2835
2836 ret = write_dev_flush(dev, 1);
2837 if (ret)
2838 errors++;
2839 }
2840 if (errors)
2841 return -EIO;
2842 return 0;
2843}
2844
2845int write_all_supers(struct btrfs_root *root, int max_mirrors)
2846{
2847 struct list_head *head;
2848 struct btrfs_device *dev;
2849 struct btrfs_super_block *sb;
2850 struct btrfs_dev_item *dev_item;
2851 int ret;
2852 int do_barriers;
2853 int max_errors;
2854 int total_errors = 0;
2855 u64 flags;
2856
2857 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2858 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2859 backup_super_roots(root->fs_info);
2860
2861 sb = root->fs_info->super_for_commit;
2862 dev_item = &sb->dev_item;
2863
2864 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2865 head = &root->fs_info->fs_devices->devices;
2866
2867 if (do_barriers)
2868 barrier_all_devices(root->fs_info);
2869
2870 list_for_each_entry_rcu(dev, head, dev_list) {
2871 if (!dev->bdev) {
2872 total_errors++;
2873 continue;
2874 }
2875 if (!dev->in_fs_metadata || !dev->writeable)
2876 continue;
2877
2878 btrfs_set_stack_device_generation(dev_item, 0);
2879 btrfs_set_stack_device_type(dev_item, dev->type);
2880 btrfs_set_stack_device_id(dev_item, dev->devid);
2881 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2882 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2883 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2884 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2885 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2886 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2887 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2888
2889 flags = btrfs_super_flags(sb);
2890 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2891
2892 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2893 if (ret)
2894 total_errors++;
2895 }
2896 if (total_errors > max_errors) {
2897 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2898 total_errors);
2899
2900 /* This shouldn't happen. FUA is masked off if unsupported */
2901 BUG();
2902 }
2903
2904 total_errors = 0;
2905 list_for_each_entry_rcu(dev, head, dev_list) {
2906 if (!dev->bdev)
2907 continue;
2908 if (!dev->in_fs_metadata || !dev->writeable)
2909 continue;
2910
2911 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2912 if (ret)
2913 total_errors++;
2914 }
2915 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2916 if (total_errors > max_errors) {
2917 btrfs_error(root->fs_info, -EIO,
2918 "%d errors while writing supers", total_errors);
2919 return -EIO;
2920 }
2921 return 0;
2922}
2923
2924int write_ctree_super(struct btrfs_trans_handle *trans,
2925 struct btrfs_root *root, int max_mirrors)
2926{
2927 int ret;
2928
2929 ret = write_all_supers(root, max_mirrors);
2930 return ret;
2931}
2932
2933void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2934{
2935 spin_lock(&fs_info->fs_roots_radix_lock);
2936 radix_tree_delete(&fs_info->fs_roots_radix,
2937 (unsigned long)root->root_key.objectid);
2938 spin_unlock(&fs_info->fs_roots_radix_lock);
2939
2940 if (btrfs_root_refs(&root->root_item) == 0)
2941 synchronize_srcu(&fs_info->subvol_srcu);
2942
2943 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2944 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2945 free_fs_root(root);
2946}
2947
2948static void free_fs_root(struct btrfs_root *root)
2949{
2950 iput(root->cache_inode);
2951 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2952 if (root->anon_dev)
2953 free_anon_bdev(root->anon_dev);
2954 free_extent_buffer(root->node);
2955 free_extent_buffer(root->commit_root);
2956 kfree(root->free_ino_ctl);
2957 kfree(root->free_ino_pinned);
2958 kfree(root->name);
2959 kfree(root);
2960}
2961
2962static void del_fs_roots(struct btrfs_fs_info *fs_info)
2963{
2964 int ret;
2965 struct btrfs_root *gang[8];
2966 int i;
2967
2968 while (!list_empty(&fs_info->dead_roots)) {
2969 gang[0] = list_entry(fs_info->dead_roots.next,
2970 struct btrfs_root, root_list);
2971 list_del(&gang[0]->root_list);
2972
2973 if (gang[0]->in_radix) {
2974 btrfs_free_fs_root(fs_info, gang[0]);
2975 } else {
2976 free_extent_buffer(gang[0]->node);
2977 free_extent_buffer(gang[0]->commit_root);
2978 kfree(gang[0]);
2979 }
2980 }
2981
2982 while (1) {
2983 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2984 (void **)gang, 0,
2985 ARRAY_SIZE(gang));
2986 if (!ret)
2987 break;
2988 for (i = 0; i < ret; i++)
2989 btrfs_free_fs_root(fs_info, gang[i]);
2990 }
2991}
2992
2993int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2994{
2995 u64 root_objectid = 0;
2996 struct btrfs_root *gang[8];
2997 int i;
2998 int ret;
2999
3000 while (1) {
3001 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3002 (void **)gang, root_objectid,
3003 ARRAY_SIZE(gang));
3004 if (!ret)
3005 break;
3006
3007 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3008 for (i = 0; i < ret; i++) {
3009 int err;
3010
3011 root_objectid = gang[i]->root_key.objectid;
3012 err = btrfs_orphan_cleanup(gang[i]);
3013 if (err)
3014 return err;
3015 }
3016 root_objectid++;
3017 }
3018 return 0;
3019}
3020
3021int btrfs_commit_super(struct btrfs_root *root)
3022{
3023 struct btrfs_trans_handle *trans;
3024 int ret;
3025
3026 mutex_lock(&root->fs_info->cleaner_mutex);
3027 btrfs_run_delayed_iputs(root);
3028 btrfs_clean_old_snapshots(root);
3029 mutex_unlock(&root->fs_info->cleaner_mutex);
3030
3031 /* wait until ongoing cleanup work done */
3032 down_write(&root->fs_info->cleanup_work_sem);
3033 up_write(&root->fs_info->cleanup_work_sem);
3034
3035 trans = btrfs_join_transaction(root);
3036 if (IS_ERR(trans))
3037 return PTR_ERR(trans);
3038 ret = btrfs_commit_transaction(trans, root);
3039 if (ret)
3040 return ret;
3041 /* run commit again to drop the original snapshot */
3042 trans = btrfs_join_transaction(root);
3043 if (IS_ERR(trans))
3044 return PTR_ERR(trans);
3045 ret = btrfs_commit_transaction(trans, root);
3046 if (ret)
3047 return ret;
3048 ret = btrfs_write_and_wait_transaction(NULL, root);
3049 if (ret) {
3050 btrfs_error(root->fs_info, ret,
3051 "Failed to sync btree inode to disk.");
3052 return ret;
3053 }
3054
3055 ret = write_ctree_super(NULL, root, 0);
3056 return ret;
3057}
3058
3059int close_ctree(struct btrfs_root *root)
3060{
3061 struct btrfs_fs_info *fs_info = root->fs_info;
3062 int ret;
3063
3064 fs_info->closing = 1;
3065 smp_mb();
3066
3067 /* pause restriper - we want to resume on mount */
3068 btrfs_pause_balance(root->fs_info);
3069
3070 btrfs_scrub_cancel(root);
3071
3072 /* wait for any defraggers to finish */
3073 wait_event(fs_info->transaction_wait,
3074 (atomic_read(&fs_info->defrag_running) == 0));
3075
3076 /* clear out the rbtree of defraggable inodes */
3077 btrfs_run_defrag_inodes(fs_info);
3078
3079 /*
3080 * Here come 2 situations when btrfs is broken to flip readonly:
3081 *
3082 * 1. when btrfs flips readonly somewhere else before
3083 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
3084 * and btrfs will skip to write sb directly to keep
3085 * ERROR state on disk.
3086 *
3087 * 2. when btrfs flips readonly just in btrfs_commit_super,
3088 * and in such case, btrfs cannot write sb via btrfs_commit_super,
3089 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3090 * btrfs will cleanup all FS resources first and write sb then.
3091 */
3092 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3093 ret = btrfs_commit_super(root);
3094 if (ret)
3095 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3096 }
3097
3098 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3099 ret = btrfs_error_commit_super(root);
3100 if (ret)
3101 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3102 }
3103
3104 btrfs_put_block_group_cache(fs_info);
3105
3106 kthread_stop(fs_info->transaction_kthread);
3107 kthread_stop(fs_info->cleaner_kthread);
3108
3109 fs_info->closing = 2;
3110 smp_mb();
3111
3112 if (fs_info->delalloc_bytes) {
3113 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3114 (unsigned long long)fs_info->delalloc_bytes);
3115 }
3116 if (fs_info->total_ref_cache_size) {
3117 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3118 (unsigned long long)fs_info->total_ref_cache_size);
3119 }
3120
3121 free_extent_buffer(fs_info->extent_root->node);
3122 free_extent_buffer(fs_info->extent_root->commit_root);
3123 free_extent_buffer(fs_info->tree_root->node);
3124 free_extent_buffer(fs_info->tree_root->commit_root);
3125 free_extent_buffer(fs_info->chunk_root->node);
3126 free_extent_buffer(fs_info->chunk_root->commit_root);
3127 free_extent_buffer(fs_info->dev_root->node);
3128 free_extent_buffer(fs_info->dev_root->commit_root);
3129 free_extent_buffer(fs_info->csum_root->node);
3130 free_extent_buffer(fs_info->csum_root->commit_root);
3131
3132 btrfs_free_block_groups(fs_info);
3133
3134 del_fs_roots(fs_info);
3135
3136 iput(fs_info->btree_inode);
3137
3138 btrfs_stop_workers(&fs_info->generic_worker);
3139 btrfs_stop_workers(&fs_info->fixup_workers);
3140 btrfs_stop_workers(&fs_info->delalloc_workers);
3141 btrfs_stop_workers(&fs_info->workers);
3142 btrfs_stop_workers(&fs_info->endio_workers);
3143 btrfs_stop_workers(&fs_info->endio_meta_workers);
3144 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3145 btrfs_stop_workers(&fs_info->endio_write_workers);
3146 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3147 btrfs_stop_workers(&fs_info->submit_workers);
3148 btrfs_stop_workers(&fs_info->delayed_workers);
3149 btrfs_stop_workers(&fs_info->caching_workers);
3150 btrfs_stop_workers(&fs_info->readahead_workers);
3151
3152#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3153 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3154 btrfsic_unmount(root, fs_info->fs_devices);
3155#endif
3156
3157 btrfs_close_devices(fs_info->fs_devices);
3158 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3159
3160 bdi_destroy(&fs_info->bdi);
3161 cleanup_srcu_struct(&fs_info->subvol_srcu);
3162
3163 return 0;
3164}
3165
3166int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3167 int atomic)
3168{
3169 int ret;
3170 struct inode *btree_inode = buf->pages[0]->mapping->host;
3171
3172 ret = extent_buffer_uptodate(buf);
3173 if (!ret)
3174 return ret;
3175
3176 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3177 parent_transid, atomic);
3178 if (ret == -EAGAIN)
3179 return ret;
3180 return !ret;
3181}
3182
3183int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3184{
3185 return set_extent_buffer_uptodate(buf);
3186}
3187
3188void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3189{
3190 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3191 u64 transid = btrfs_header_generation(buf);
3192 int was_dirty;
3193
3194 btrfs_assert_tree_locked(buf);
3195 if (transid != root->fs_info->generation) {
3196 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3197 "found %llu running %llu\n",
3198 (unsigned long long)buf->start,
3199 (unsigned long long)transid,
3200 (unsigned long long)root->fs_info->generation);
3201 WARN_ON(1);
3202 }
3203 was_dirty = set_extent_buffer_dirty(buf);
3204 if (!was_dirty) {
3205 spin_lock(&root->fs_info->delalloc_lock);
3206 root->fs_info->dirty_metadata_bytes += buf->len;
3207 spin_unlock(&root->fs_info->delalloc_lock);
3208 }
3209}
3210
3211void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3212{
3213 /*
3214 * looks as though older kernels can get into trouble with
3215 * this code, they end up stuck in balance_dirty_pages forever
3216 */
3217 u64 num_dirty;
3218 unsigned long thresh = 32 * 1024 * 1024;
3219
3220 if (current->flags & PF_MEMALLOC)
3221 return;
3222
3223 btrfs_balance_delayed_items(root);
3224
3225 num_dirty = root->fs_info->dirty_metadata_bytes;
3226
3227 if (num_dirty > thresh) {
3228 balance_dirty_pages_ratelimited_nr(
3229 root->fs_info->btree_inode->i_mapping, 1);
3230 }
3231 return;
3232}
3233
3234void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3235{
3236 /*
3237 * looks as though older kernels can get into trouble with
3238 * this code, they end up stuck in balance_dirty_pages forever
3239 */
3240 u64 num_dirty;
3241 unsigned long thresh = 32 * 1024 * 1024;
3242
3243 if (current->flags & PF_MEMALLOC)
3244 return;
3245
3246 num_dirty = root->fs_info->dirty_metadata_bytes;
3247
3248 if (num_dirty > thresh) {
3249 balance_dirty_pages_ratelimited_nr(
3250 root->fs_info->btree_inode->i_mapping, 1);
3251 }
3252 return;
3253}
3254
3255int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3256{
3257 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3258 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3259}
3260
3261static int btree_lock_page_hook(struct page *page, void *data,
3262 void (*flush_fn)(void *))
3263{
3264 struct inode *inode = page->mapping->host;
3265 struct btrfs_root *root = BTRFS_I(inode)->root;
3266 struct extent_buffer *eb;
3267
3268 /*
3269 * We culled this eb but the page is still hanging out on the mapping,
3270 * carry on.
3271 */
3272 if (!PagePrivate(page))
3273 goto out;
3274
3275 eb = (struct extent_buffer *)page->private;
3276 if (!eb) {
3277 WARN_ON(1);
3278 goto out;
3279 }
3280 if (page != eb->pages[0])
3281 goto out;
3282
3283 if (!btrfs_try_tree_write_lock(eb)) {
3284 flush_fn(data);
3285 btrfs_tree_lock(eb);
3286 }
3287 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3288
3289 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3290 spin_lock(&root->fs_info->delalloc_lock);
3291 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3292 root->fs_info->dirty_metadata_bytes -= eb->len;
3293 else
3294 WARN_ON(1);
3295 spin_unlock(&root->fs_info->delalloc_lock);
3296 }
3297
3298 btrfs_tree_unlock(eb);
3299out:
3300 if (!trylock_page(page)) {
3301 flush_fn(data);
3302 lock_page(page);
3303 }
3304 return 0;
3305}
3306
3307static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3308 int read_only)
3309{
3310 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3311 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3312 return -EINVAL;
3313 }
3314
3315 if (read_only)
3316 return 0;
3317
3318 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3319 printk(KERN_WARNING "warning: mount fs with errors, "
3320 "running btrfsck is recommended\n");
3321 }
3322
3323 return 0;
3324}
3325
3326int btrfs_error_commit_super(struct btrfs_root *root)
3327{
3328 int ret;
3329
3330 mutex_lock(&root->fs_info->cleaner_mutex);
3331 btrfs_run_delayed_iputs(root);
3332 mutex_unlock(&root->fs_info->cleaner_mutex);
3333
3334 down_write(&root->fs_info->cleanup_work_sem);
3335 up_write(&root->fs_info->cleanup_work_sem);
3336
3337 /* cleanup FS via transaction */
3338 btrfs_cleanup_transaction(root);
3339
3340 ret = write_ctree_super(NULL, root, 0);
3341
3342 return ret;
3343}
3344
3345static void btrfs_destroy_ordered_operations(struct btrfs_root *root)
3346{
3347 struct btrfs_inode *btrfs_inode;
3348 struct list_head splice;
3349
3350 INIT_LIST_HEAD(&splice);
3351
3352 mutex_lock(&root->fs_info->ordered_operations_mutex);
3353 spin_lock(&root->fs_info->ordered_extent_lock);
3354
3355 list_splice_init(&root->fs_info->ordered_operations, &splice);
3356 while (!list_empty(&splice)) {
3357 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3358 ordered_operations);
3359
3360 list_del_init(&btrfs_inode->ordered_operations);
3361
3362 btrfs_invalidate_inodes(btrfs_inode->root);
3363 }
3364
3365 spin_unlock(&root->fs_info->ordered_extent_lock);
3366 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3367}
3368
3369static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3370{
3371 struct list_head splice;
3372 struct btrfs_ordered_extent *ordered;
3373 struct inode *inode;
3374
3375 INIT_LIST_HEAD(&splice);
3376
3377 spin_lock(&root->fs_info->ordered_extent_lock);
3378
3379 list_splice_init(&root->fs_info->ordered_extents, &splice);
3380 while (!list_empty(&splice)) {
3381 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3382 root_extent_list);
3383
3384 list_del_init(&ordered->root_extent_list);
3385 atomic_inc(&ordered->refs);
3386
3387 /* the inode may be getting freed (in sys_unlink path). */
3388 inode = igrab(ordered->inode);
3389
3390 spin_unlock(&root->fs_info->ordered_extent_lock);
3391 if (inode)
3392 iput(inode);
3393
3394 atomic_set(&ordered->refs, 1);
3395 btrfs_put_ordered_extent(ordered);
3396
3397 spin_lock(&root->fs_info->ordered_extent_lock);
3398 }
3399
3400 spin_unlock(&root->fs_info->ordered_extent_lock);
3401}
3402
3403int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3404 struct btrfs_root *root)
3405{
3406 struct rb_node *node;
3407 struct btrfs_delayed_ref_root *delayed_refs;
3408 struct btrfs_delayed_ref_node *ref;
3409 int ret = 0;
3410
3411 delayed_refs = &trans->delayed_refs;
3412
3413 spin_lock(&delayed_refs->lock);
3414 if (delayed_refs->num_entries == 0) {
3415 spin_unlock(&delayed_refs->lock);
3416 printk(KERN_INFO "delayed_refs has NO entry\n");
3417 return ret;
3418 }
3419
3420 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3421 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3422
3423 atomic_set(&ref->refs, 1);
3424 if (btrfs_delayed_ref_is_head(ref)) {
3425 struct btrfs_delayed_ref_head *head;
3426
3427 head = btrfs_delayed_node_to_head(ref);
3428 if (!mutex_trylock(&head->mutex)) {
3429 atomic_inc(&ref->refs);
3430 spin_unlock(&delayed_refs->lock);
3431
3432 /* Need to wait for the delayed ref to run */
3433 mutex_lock(&head->mutex);
3434 mutex_unlock(&head->mutex);
3435 btrfs_put_delayed_ref(ref);
3436
3437 spin_lock(&delayed_refs->lock);
3438 continue;
3439 }
3440
3441 kfree(head->extent_op);
3442 delayed_refs->num_heads--;
3443 if (list_empty(&head->cluster))
3444 delayed_refs->num_heads_ready--;
3445 list_del_init(&head->cluster);
3446 }
3447 ref->in_tree = 0;
3448 rb_erase(&ref->rb_node, &delayed_refs->root);
3449 delayed_refs->num_entries--;
3450
3451 spin_unlock(&delayed_refs->lock);
3452 btrfs_put_delayed_ref(ref);
3453
3454 cond_resched();
3455 spin_lock(&delayed_refs->lock);
3456 }
3457
3458 spin_unlock(&delayed_refs->lock);
3459
3460 return ret;
3461}
3462
3463static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3464{
3465 struct btrfs_pending_snapshot *snapshot;
3466 struct list_head splice;
3467
3468 INIT_LIST_HEAD(&splice);
3469
3470 list_splice_init(&t->pending_snapshots, &splice);
3471
3472 while (!list_empty(&splice)) {
3473 snapshot = list_entry(splice.next,
3474 struct btrfs_pending_snapshot,
3475 list);
3476
3477 list_del_init(&snapshot->list);
3478
3479 kfree(snapshot);
3480 }
3481}
3482
3483static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3484{
3485 struct btrfs_inode *btrfs_inode;
3486 struct list_head splice;
3487
3488 INIT_LIST_HEAD(&splice);
3489
3490 spin_lock(&root->fs_info->delalloc_lock);
3491 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3492
3493 while (!list_empty(&splice)) {
3494 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3495 delalloc_inodes);
3496
3497 list_del_init(&btrfs_inode->delalloc_inodes);
3498
3499 btrfs_invalidate_inodes(btrfs_inode->root);
3500 }
3501
3502 spin_unlock(&root->fs_info->delalloc_lock);
3503}
3504
3505static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3506 struct extent_io_tree *dirty_pages,
3507 int mark)
3508{
3509 int ret;
3510 struct page *page;
3511 struct inode *btree_inode = root->fs_info->btree_inode;
3512 struct extent_buffer *eb;
3513 u64 start = 0;
3514 u64 end;
3515 u64 offset;
3516 unsigned long index;
3517
3518 while (1) {
3519 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3520 mark);
3521 if (ret)
3522 break;
3523
3524 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3525 while (start <= end) {
3526 index = start >> PAGE_CACHE_SHIFT;
3527 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3528 page = find_get_page(btree_inode->i_mapping, index);
3529 if (!page)
3530 continue;
3531 offset = page_offset(page);
3532
3533 spin_lock(&dirty_pages->buffer_lock);
3534 eb = radix_tree_lookup(
3535 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3536 offset >> PAGE_CACHE_SHIFT);
3537 spin_unlock(&dirty_pages->buffer_lock);
3538 if (eb)
3539 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3540 &eb->bflags);
3541 if (PageWriteback(page))
3542 end_page_writeback(page);
3543
3544 lock_page(page);
3545 if (PageDirty(page)) {
3546 clear_page_dirty_for_io(page);
3547 spin_lock_irq(&page->mapping->tree_lock);
3548 radix_tree_tag_clear(&page->mapping->page_tree,
3549 page_index(page),
3550 PAGECACHE_TAG_DIRTY);
3551 spin_unlock_irq(&page->mapping->tree_lock);
3552 }
3553
3554 unlock_page(page);
3555 page_cache_release(page);
3556 }
3557 }
3558
3559 return ret;
3560}
3561
3562static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3563 struct extent_io_tree *pinned_extents)
3564{
3565 struct extent_io_tree *unpin;
3566 u64 start;
3567 u64 end;
3568 int ret;
3569 bool loop = true;
3570
3571 unpin = pinned_extents;
3572again:
3573 while (1) {
3574 ret = find_first_extent_bit(unpin, 0, &start, &end,
3575 EXTENT_DIRTY);
3576 if (ret)
3577 break;
3578
3579 /* opt_discard */
3580 if (btrfs_test_opt(root, DISCARD))
3581 ret = btrfs_error_discard_extent(root, start,
3582 end + 1 - start,
3583 NULL);
3584
3585 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3586 btrfs_error_unpin_extent_range(root, start, end);
3587 cond_resched();
3588 }
3589
3590 if (loop) {
3591 if (unpin == &root->fs_info->freed_extents[0])
3592 unpin = &root->fs_info->freed_extents[1];
3593 else
3594 unpin = &root->fs_info->freed_extents[0];
3595 loop = false;
3596 goto again;
3597 }
3598
3599 return 0;
3600}
3601
3602void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3603 struct btrfs_root *root)
3604{
3605 btrfs_destroy_delayed_refs(cur_trans, root);
3606 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3607 cur_trans->dirty_pages.dirty_bytes);
3608
3609 /* FIXME: cleanup wait for commit */
3610 cur_trans->in_commit = 1;
3611 cur_trans->blocked = 1;
3612 wake_up(&root->fs_info->transaction_blocked_wait);
3613
3614 cur_trans->blocked = 0;
3615 wake_up(&root->fs_info->transaction_wait);
3616
3617 cur_trans->commit_done = 1;
3618 wake_up(&cur_trans->commit_wait);
3619
3620 btrfs_destroy_delayed_inodes(root);
3621 btrfs_assert_delayed_root_empty(root);
3622
3623 btrfs_destroy_pending_snapshots(cur_trans);
3624
3625 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3626 EXTENT_DIRTY);
3627 btrfs_destroy_pinned_extent(root,
3628 root->fs_info->pinned_extents);
3629
3630 /*
3631 memset(cur_trans, 0, sizeof(*cur_trans));
3632 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3633 */
3634}
3635
3636int btrfs_cleanup_transaction(struct btrfs_root *root)
3637{
3638 struct btrfs_transaction *t;
3639 LIST_HEAD(list);
3640
3641 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3642
3643 spin_lock(&root->fs_info->trans_lock);
3644 list_splice_init(&root->fs_info->trans_list, &list);
3645 root->fs_info->trans_no_join = 1;
3646 spin_unlock(&root->fs_info->trans_lock);
3647
3648 while (!list_empty(&list)) {
3649 t = list_entry(list.next, struct btrfs_transaction, list);
3650 if (!t)
3651 break;
3652
3653 btrfs_destroy_ordered_operations(root);
3654
3655 btrfs_destroy_ordered_extents(root);
3656
3657 btrfs_destroy_delayed_refs(t, root);
3658
3659 btrfs_block_rsv_release(root,
3660 &root->fs_info->trans_block_rsv,
3661 t->dirty_pages.dirty_bytes);
3662
3663 /* FIXME: cleanup wait for commit */
3664 t->in_commit = 1;
3665 t->blocked = 1;
3666 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3667 wake_up(&root->fs_info->transaction_blocked_wait);
3668
3669 t->blocked = 0;
3670 if (waitqueue_active(&root->fs_info->transaction_wait))
3671 wake_up(&root->fs_info->transaction_wait);
3672
3673 t->commit_done = 1;
3674 if (waitqueue_active(&t->commit_wait))
3675 wake_up(&t->commit_wait);
3676
3677 btrfs_destroy_delayed_inodes(root);
3678 btrfs_assert_delayed_root_empty(root);
3679
3680 btrfs_destroy_pending_snapshots(t);
3681
3682 btrfs_destroy_delalloc_inodes(root);
3683
3684 spin_lock(&root->fs_info->trans_lock);
3685 root->fs_info->running_transaction = NULL;
3686 spin_unlock(&root->fs_info->trans_lock);
3687
3688 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3689 EXTENT_DIRTY);
3690
3691 btrfs_destroy_pinned_extent(root,
3692 root->fs_info->pinned_extents);
3693
3694 atomic_set(&t->use_count, 0);
3695 list_del_init(&t->list);
3696 memset(t, 0, sizeof(*t));
3697 kmem_cache_free(btrfs_transaction_cachep, t);
3698 }
3699
3700 spin_lock(&root->fs_info->trans_lock);
3701 root->fs_info->trans_no_join = 0;
3702 spin_unlock(&root->fs_info->trans_lock);
3703 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3704
3705 return 0;
3706}
3707
3708static struct extent_io_ops btree_extent_io_ops = {
3709 .write_cache_pages_lock_hook = btree_lock_page_hook,
3710 .readpage_end_io_hook = btree_readpage_end_io_hook,
3711 .readpage_io_failed_hook = btree_io_failed_hook,
3712 .submit_bio_hook = btree_submit_bio_hook,
3713 /* note we're sharing with inode.c for the merge bio hook */
3714 .merge_bio_hook = btrfs_merge_bio_hook,
3715};