projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jesse/openvswitch
[deliverable/linux.git] / fs / btrfs / disk-io.c
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
48 static struct extent_io_ops btree_extent_io_ops;
49 static void end_workqueue_fn(struct btrfs_work *work);
50 static void free_fs_root(struct btrfs_root *root);
51 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
52 int read_only);
53 static void btrfs_destroy_ordered_operations(struct btrfs_root *root);
54 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_root *root);
57 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
58 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
60 struct extent_io_tree *dirty_pages,
61 int mark);
62 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
63 struct extent_io_tree *pinned_extents);
64
65 /*
66 * end_io_wq structs are used to do processing in task context when an IO is
67 * complete. This is used during reads to verify checksums, and it is used
68 * by writes to insert metadata for new file extents after IO is complete.
69 */
70 struct end_io_wq {
71 struct bio *bio;
72 bio_end_io_t *end_io;
73 void *private;
74 struct btrfs_fs_info *info;
75 int error;
76 int metadata;
77 struct list_head list;
78 struct btrfs_work work;
79 };
80
81 /*
82 * async submit bios are used to offload expensive checksumming
83 * onto the worker threads. They checksum file and metadata bios
84 * just before they are sent down the IO stack.
85 */
86 struct async_submit_bio {
87 struct inode *inode;
88 struct bio *bio;
89 struct list_head list;
90 extent_submit_bio_hook_t *submit_bio_start;
91 extent_submit_bio_hook_t *submit_bio_done;
92 int rw;
93 int mirror_num;
94 unsigned long bio_flags;
95 /*
96 * bio_offset is optional, can be used if the pages in the bio
97 * can't tell us where in the file the bio should go
98 */
99 u64 bio_offset;
100 struct btrfs_work work;
101 int error;
102 };
103
104 /*
105 * Lockdep class keys for extent_buffer->lock's in this root. For a given
106 * eb, the lockdep key is determined by the btrfs_root it belongs to and
107 * the level the eb occupies in the tree.
108 *
109 * Different roots are used for different purposes and may nest inside each
110 * other and they require separate keysets. As lockdep keys should be
111 * static, assign keysets according to the purpose of the root as indicated
112 * by btrfs_root->objectid. This ensures that all special purpose roots
113 * have separate keysets.
114 *
115 * Lock-nesting across peer nodes is always done with the immediate parent
116 * node locked thus preventing deadlock. As lockdep doesn't know this, use
117 * subclass to avoid triggering lockdep warning in such cases.
118 *
119 * The key is set by the readpage_end_io_hook after the buffer has passed
120 * csum validation but before the pages are unlocked. It is also set by
121 * btrfs_init_new_buffer on freshly allocated blocks.
122 *
123 * We also add a check to make sure the highest level of the tree is the
124 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
125 * needs update as well.
126 */
127 #ifdef CONFIG_DEBUG_LOCK_ALLOC
128 # if BTRFS_MAX_LEVEL != 8
129 # error
130 # endif
131
132 static struct btrfs_lockdep_keyset {
133 u64 id; /* root objectid */
134 const char *name_stem; /* lock name stem */
135 char names[BTRFS_MAX_LEVEL + 1][20];
136 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
137 } btrfs_lockdep_keysets[] = {
138 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
139 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
140 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
141 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
142 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
143 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
144 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
145 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
146 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
147 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
148 { .id = 0, .name_stem = "tree" },
149 };
150
151 void __init btrfs_init_lockdep(void)
152 {
153 int i, j;
154
155 /* initialize lockdep class names */
156 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
157 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
158
159 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
160 snprintf(ks->names[j], sizeof(ks->names[j]),
161 "btrfs-%s-%02d", ks->name_stem, j);
162 }
163 }
164
165 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
166 int level)
167 {
168 struct btrfs_lockdep_keyset *ks;
169
170 BUG_ON(level >= ARRAY_SIZE(ks->keys));
171
172 /* find the matching keyset, id 0 is the default entry */
173 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
174 if (ks->id == objectid)
175 break;
176
177 lockdep_set_class_and_name(&eb->lock,
178 &ks->keys[level], ks->names[level]);
179 }
180
181 #endif
182
183 /*
184 * extents on the btree inode are pretty simple, there's one extent
185 * that covers the entire device
186 */
187 static struct extent_map *btree_get_extent(struct inode *inode,
188 struct page *page, size_t pg_offset, u64 start, u64 len,
189 int create)
190 {
191 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
192 struct extent_map *em;
193 int ret;
194
195 read_lock(&em_tree->lock);
196 em = lookup_extent_mapping(em_tree, start, len);
197 if (em) {
198 em->bdev =
199 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
200 read_unlock(&em_tree->lock);
201 goto out;
202 }
203 read_unlock(&em_tree->lock);
204
205 em = alloc_extent_map();
206 if (!em) {
207 em = ERR_PTR(-ENOMEM);
208 goto out;
209 }
210 em->start = 0;
211 em->len = (u64)-1;
212 em->block_len = (u64)-1;
213 em->block_start = 0;
214 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
215
216 write_lock(&em_tree->lock);
217 ret = add_extent_mapping(em_tree, em);
218 if (ret == -EEXIST) {
219 u64 failed_start = em->start;
220 u64 failed_len = em->len;
221
222 free_extent_map(em);
223 em = lookup_extent_mapping(em_tree, start, len);
224 if (em) {
225 ret = 0;
226 } else {
227 em = lookup_extent_mapping(em_tree, failed_start,
228 failed_len);
229 ret = -EIO;
230 }
231 } else if (ret) {
232 free_extent_map(em);
233 em = NULL;
234 }
235 write_unlock(&em_tree->lock);
236
237 if (ret)
238 em = ERR_PTR(ret);
239 out:
240 return em;
241 }
242
243 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
244 {
245 return crc32c(seed, data, len);
246 }
247
248 void btrfs_csum_final(u32 crc, char *result)
249 {
250 put_unaligned_le32(~crc, result);
251 }
252
253 /*
254 * compute the csum for a btree block, and either verify it or write it
255 * into the csum field of the block.
256 */
257 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
258 int verify)
259 {
260 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
261 char *result = NULL;
262 unsigned long len;
263 unsigned long cur_len;
264 unsigned long offset = BTRFS_CSUM_SIZE;
265 char *kaddr;
266 unsigned long map_start;
267 unsigned long map_len;
268 int err;
269 u32 crc = ~(u32)0;
270 unsigned long inline_result;
271
272 len = buf->len - offset;
273 while (len > 0) {
274 err = map_private_extent_buffer(buf, offset, 32,
275 &kaddr, &map_start, &map_len);
276 if (err)
277 return 1;
278 cur_len = min(len, map_len - (offset - map_start));
279 crc = btrfs_csum_data(root, kaddr + offset - map_start,
280 crc, cur_len);
281 len -= cur_len;
282 offset += cur_len;
283 }
284 if (csum_size > sizeof(inline_result)) {
285 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
286 if (!result)
287 return 1;
288 } else {
289 result = (char *)&inline_result;
290 }
291
292 btrfs_csum_final(crc, result);
293
294 if (verify) {
295 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
296 u32 val;
297 u32 found = 0;
298 memcpy(&found, result, csum_size);
299
300 read_extent_buffer(buf, &val, 0, csum_size);
301 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
302 "failed on %llu wanted %X found %X "
303 "level %d\n",
304 root->fs_info->sb->s_id,
305 (unsigned long long)buf->start, val, found,
306 btrfs_header_level(buf));
307 if (result != (char *)&inline_result)
308 kfree(result);
309 return 1;
310 }
311 } else {
312 write_extent_buffer(buf, result, 0, csum_size);
313 }
314 if (result != (char *)&inline_result)
315 kfree(result);
316 return 0;
317 }
318
319 /*
320 * we can't consider a given block up to date unless the transid of the
321 * block matches the transid in the parent node's pointer. This is how we
322 * detect blocks that either didn't get written at all or got written
323 * in the wrong place.
324 */
325 static int verify_parent_transid(struct extent_io_tree *io_tree,
326 struct extent_buffer *eb, u64 parent_transid)
327 {
328 struct extent_state *cached_state = NULL;
329 int ret;
330
331 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
332 return 0;
333
334 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
335 0, &cached_state);
336 if (extent_buffer_uptodate(eb) &&
337 btrfs_header_generation(eb) == parent_transid) {
338 ret = 0;
339 goto out;
340 }
341 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
342 "found %llu\n",
343 (unsigned long long)eb->start,
344 (unsigned long long)parent_transid,
345 (unsigned long long)btrfs_header_generation(eb));
346 ret = 1;
347 clear_extent_buffer_uptodate(eb);
348 out:
349 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
350 &cached_state, GFP_NOFS);
351 return ret;
352 }
353
354 /*
355 * helper to read a given tree block, doing retries as required when
356 * the checksums don't match and we have alternate mirrors to try.
357 */
358 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
359 struct extent_buffer *eb,
360 u64 start, u64 parent_transid)
361 {
362 struct extent_io_tree *io_tree;
363 int failed = 0;
364 int ret;
365 int num_copies = 0;
366 int mirror_num = 0;
367 int failed_mirror = 0;
368
369 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
370 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
371 while (1) {
372 ret = read_extent_buffer_pages(io_tree, eb, start,
373 WAIT_COMPLETE,
374 btree_get_extent, mirror_num);
375 if (!ret && !verify_parent_transid(io_tree, eb, parent_transid))
376 break;
377
378 /*
379 * This buffer's crc is fine, but its contents are corrupted, so
380 * there is no reason to read the other copies, they won't be
381 * any less wrong.
382 */
383 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
384 break;
385
386 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
387 eb->start, eb->len);
388 if (num_copies == 1)
389 break;
390
391 if (!failed_mirror) {
392 failed = 1;
393 failed_mirror = eb->read_mirror;
394 }
395
396 mirror_num++;
397 if (mirror_num == failed_mirror)
398 mirror_num++;
399
400 if (mirror_num > num_copies)
401 break;
402 }
403
404 if (failed && !ret)
405 repair_eb_io_failure(root, eb, failed_mirror);
406
407 return ret;
408 }
409
410 /*
411 * checksum a dirty tree block before IO. This has extra checks to make sure
412 * we only fill in the checksum field in the first page of a multi-page block
413 */
414
415 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
416 {
417 struct extent_io_tree *tree;
418 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
419 u64 found_start;
420 struct extent_buffer *eb;
421
422 tree = &BTRFS_I(page->mapping->host)->io_tree;
423
424 eb = (struct extent_buffer *)page->private;
425 if (page != eb->pages[0])
426 return 0;
427 found_start = btrfs_header_bytenr(eb);
428 if (found_start != start) {
429 WARN_ON(1);
430 return 0;
431 }
432 if (eb->pages[0] != page) {
433 WARN_ON(1);
434 return 0;
435 }
436 if (!PageUptodate(page)) {
437 WARN_ON(1);
438 return 0;
439 }
440 csum_tree_block(root, eb, 0);
441 return 0;
442 }
443
444 static int check_tree_block_fsid(struct btrfs_root *root,
445 struct extent_buffer *eb)
446 {
447 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
448 u8 fsid[BTRFS_UUID_SIZE];
449 int ret = 1;
450
451 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
452 BTRFS_FSID_SIZE);
453 while (fs_devices) {
454 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
455 ret = 0;
456 break;
457 }
458 fs_devices = fs_devices->seed;
459 }
460 return ret;
461 }
462
463 #define CORRUPT(reason, eb, root, slot) \
464 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
465 "root=%llu, slot=%d\n", reason, \
466 (unsigned long long)btrfs_header_bytenr(eb), \
467 (unsigned long long)root->objectid, slot)
468
469 static noinline int check_leaf(struct btrfs_root *root,
470 struct extent_buffer *leaf)
471 {
472 struct btrfs_key key;
473 struct btrfs_key leaf_key;
474 u32 nritems = btrfs_header_nritems(leaf);
475 int slot;
476
477 if (nritems == 0)
478 return 0;
479
480 /* Check the 0 item */
481 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
482 BTRFS_LEAF_DATA_SIZE(root)) {
483 CORRUPT("invalid item offset size pair", leaf, root, 0);
484 return -EIO;
485 }
486
487 /*
488 * Check to make sure each items keys are in the correct order and their
489 * offsets make sense. We only have to loop through nritems-1 because
490 * we check the current slot against the next slot, which verifies the
491 * next slot's offset+size makes sense and that the current's slot
492 * offset is correct.
493 */
494 for (slot = 0; slot < nritems - 1; slot++) {
495 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
496 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
497
498 /* Make sure the keys are in the right order */
499 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
500 CORRUPT("bad key order", leaf, root, slot);
501 return -EIO;
502 }
503
504 /*
505 * Make sure the offset and ends are right, remember that the
506 * item data starts at the end of the leaf and grows towards the
507 * front.
508 */
509 if (btrfs_item_offset_nr(leaf, slot) !=
510 btrfs_item_end_nr(leaf, slot + 1)) {
511 CORRUPT("slot offset bad", leaf, root, slot);
512 return -EIO;
513 }
514
515 /*
516 * Check to make sure that we don't point outside of the leaf,
517 * just incase all the items are consistent to eachother, but
518 * all point outside of the leaf.
519 */
520 if (btrfs_item_end_nr(leaf, slot) >
521 BTRFS_LEAF_DATA_SIZE(root)) {
522 CORRUPT("slot end outside of leaf", leaf, root, slot);
523 return -EIO;
524 }
525 }
526
527 return 0;
528 }
529
530 struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
531 struct page *page, int max_walk)
532 {
533 struct extent_buffer *eb;
534 u64 start = page_offset(page);
535 u64 target = start;
536 u64 min_start;
537
538 if (start < max_walk)
539 min_start = 0;
540 else
541 min_start = start - max_walk;
542
543 while (start >= min_start) {
544 eb = find_extent_buffer(tree, start, 0);
545 if (eb) {
546 /*
547 * we found an extent buffer and it contains our page
548 * horray!
549 */
550 if (eb->start <= target &&
551 eb->start + eb->len > target)
552 return eb;
553
554 /* we found an extent buffer that wasn't for us */
555 free_extent_buffer(eb);
556 return NULL;
557 }
558 if (start == 0)
559 break;
560 start -= PAGE_CACHE_SIZE;
561 }
562 return NULL;
563 }
564
565 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
566 struct extent_state *state, int mirror)
567 {
568 struct extent_io_tree *tree;
569 u64 found_start;
570 int found_level;
571 struct extent_buffer *eb;
572 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
573 int ret = 0;
574 int reads_done;
575
576 if (!page->private)
577 goto out;
578
579 tree = &BTRFS_I(page->mapping->host)->io_tree;
580 eb = (struct extent_buffer *)page->private;
581
582 /* the pending IO might have been the only thing that kept this buffer
583 * in memory. Make sure we have a ref for all this other checks
584 */
585 extent_buffer_get(eb);
586
587 reads_done = atomic_dec_and_test(&eb->io_pages);
588 if (!reads_done)
589 goto err;
590
591 eb->read_mirror = mirror;
592 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
593 ret = -EIO;
594 goto err;
595 }
596
597 found_start = btrfs_header_bytenr(eb);
598 if (found_start != eb->start) {
599 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
600 "%llu %llu\n",
601 (unsigned long long)found_start,
602 (unsigned long long)eb->start);
603 ret = -EIO;
604 goto err;
605 }
606 if (check_tree_block_fsid(root, eb)) {
607 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
608 (unsigned long long)eb->start);
609 ret = -EIO;
610 goto err;
611 }
612 found_level = btrfs_header_level(eb);
613
614 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
615 eb, found_level);
616
617 ret = csum_tree_block(root, eb, 1);
618 if (ret) {
619 ret = -EIO;
620 goto err;
621 }
622
623 /*
624 * If this is a leaf block and it is corrupt, set the corrupt bit so
625 * that we don't try and read the other copies of this block, just
626 * return -EIO.
627 */
628 if (found_level == 0 && check_leaf(root, eb)) {
629 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
630 ret = -EIO;
631 }
632
633 if (!ret)
634 set_extent_buffer_uptodate(eb);
635 err:
636 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
637 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
638 btree_readahead_hook(root, eb, eb->start, ret);
639 }
640
641 if (ret)
642 clear_extent_buffer_uptodate(eb);
643 free_extent_buffer(eb);
644 out:
645 return ret;
646 }
647
648 static int btree_io_failed_hook(struct page *page, int failed_mirror)
649 {
650 struct extent_buffer *eb;
651 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
652
653 eb = (struct extent_buffer *)page->private;
654 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
655 eb->read_mirror = failed_mirror;
656 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
657 btree_readahead_hook(root, eb, eb->start, -EIO);
658 return -EIO; /* we fixed nothing */
659 }
660
661 static void end_workqueue_bio(struct bio *bio, int err)
662 {
663 struct end_io_wq *end_io_wq = bio->bi_private;
664 struct btrfs_fs_info *fs_info;
665
666 fs_info = end_io_wq->info;
667 end_io_wq->error = err;
668 end_io_wq->work.func = end_workqueue_fn;
669 end_io_wq->work.flags = 0;
670
671 if (bio->bi_rw & REQ_WRITE) {
672 if (end_io_wq->metadata == 1)
673 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
674 &end_io_wq->work);
675 else if (end_io_wq->metadata == 2)
676 btrfs_queue_worker(&fs_info->endio_freespace_worker,
677 &end_io_wq->work);
678 else
679 btrfs_queue_worker(&fs_info->endio_write_workers,
680 &end_io_wq->work);
681 } else {
682 if (end_io_wq->metadata)
683 btrfs_queue_worker(&fs_info->endio_meta_workers,
684 &end_io_wq->work);
685 else
686 btrfs_queue_worker(&fs_info->endio_workers,
687 &end_io_wq->work);
688 }
689 }
690
691 /*
692 * For the metadata arg you want
693 *
694 * 0 - if data
695 * 1 - if normal metadta
696 * 2 - if writing to the free space cache area
697 */
698 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
699 int metadata)
700 {
701 struct end_io_wq *end_io_wq;
702 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
703 if (!end_io_wq)
704 return -ENOMEM;
705
706 end_io_wq->private = bio->bi_private;
707 end_io_wq->end_io = bio->bi_end_io;
708 end_io_wq->info = info;
709 end_io_wq->error = 0;
710 end_io_wq->bio = bio;
711 end_io_wq->metadata = metadata;
712
713 bio->bi_private = end_io_wq;
714 bio->bi_end_io = end_workqueue_bio;
715 return 0;
716 }
717
718 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
719 {
720 unsigned long limit = min_t(unsigned long,
721 info->workers.max_workers,
722 info->fs_devices->open_devices);
723 return 256 * limit;
724 }
725
726 static void run_one_async_start(struct btrfs_work *work)
727 {
728 struct async_submit_bio *async;
729 int ret;
730
731 async = container_of(work, struct async_submit_bio, work);
732 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
733 async->mirror_num, async->bio_flags,
734 async->bio_offset);
735 if (ret)
736 async->error = ret;
737 }
738
739 static void run_one_async_done(struct btrfs_work *work)
740 {
741 struct btrfs_fs_info *fs_info;
742 struct async_submit_bio *async;
743 int limit;
744
745 async = container_of(work, struct async_submit_bio, work);
746 fs_info = BTRFS_I(async->inode)->root->fs_info;
747
748 limit = btrfs_async_submit_limit(fs_info);
749 limit = limit * 2 / 3;
750
751 atomic_dec(&fs_info->nr_async_submits);
752
753 if (atomic_read(&fs_info->nr_async_submits) < limit &&
754 waitqueue_active(&fs_info->async_submit_wait))
755 wake_up(&fs_info->async_submit_wait);
756
757 /* If an error occured we just want to clean up the bio and move on */
758 if (async->error) {
759 bio_endio(async->bio, async->error);
760 return;
761 }
762
763 async->submit_bio_done(async->inode, async->rw, async->bio,
764 async->mirror_num, async->bio_flags,
765 async->bio_offset);
766 }
767
768 static void run_one_async_free(struct btrfs_work *work)
769 {
770 struct async_submit_bio *async;
771
772 async = container_of(work, struct async_submit_bio, work);
773 kfree(async);
774 }
775
776 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
777 int rw, struct bio *bio, int mirror_num,
778 unsigned long bio_flags,
779 u64 bio_offset,
780 extent_submit_bio_hook_t *submit_bio_start,
781 extent_submit_bio_hook_t *submit_bio_done)
782 {
783 struct async_submit_bio *async;
784
785 async = kmalloc(sizeof(*async), GFP_NOFS);
786 if (!async)
787 return -ENOMEM;
788
789 async->inode = inode;
790 async->rw = rw;
791 async->bio = bio;
792 async->mirror_num = mirror_num;
793 async->submit_bio_start = submit_bio_start;
794 async->submit_bio_done = submit_bio_done;
795
796 async->work.func = run_one_async_start;
797 async->work.ordered_func = run_one_async_done;
798 async->work.ordered_free = run_one_async_free;
799
800 async->work.flags = 0;
801 async->bio_flags = bio_flags;
802 async->bio_offset = bio_offset;
803
804 async->error = 0;
805
806 atomic_inc(&fs_info->nr_async_submits);
807
808 if (rw & REQ_SYNC)
809 btrfs_set_work_high_prio(&async->work);
810
811 btrfs_queue_worker(&fs_info->workers, &async->work);
812
813 while (atomic_read(&fs_info->async_submit_draining) &&
814 atomic_read(&fs_info->nr_async_submits)) {
815 wait_event(fs_info->async_submit_wait,
816 (atomic_read(&fs_info->nr_async_submits) == 0));
817 }
818
819 return 0;
820 }
821
822 static int btree_csum_one_bio(struct bio *bio)
823 {
824 struct bio_vec *bvec = bio->bi_io_vec;
825 int bio_index = 0;
826 struct btrfs_root *root;
827 int ret = 0;
828
829 WARN_ON(bio->bi_vcnt <= 0);
830 while (bio_index < bio->bi_vcnt) {
831 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
832 ret = csum_dirty_buffer(root, bvec->bv_page);
833 if (ret)
834 break;
835 bio_index++;
836 bvec++;
837 }
838 return ret;
839 }
840
841 static int __btree_submit_bio_start(struct inode *inode, int rw,
842 struct bio *bio, int mirror_num,
843 unsigned long bio_flags,
844 u64 bio_offset)
845 {
846 /*
847 * when we're called for a write, we're already in the async
848 * submission context. Just jump into btrfs_map_bio
849 */
850 return btree_csum_one_bio(bio);
851 }
852
853 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
854 int mirror_num, unsigned long bio_flags,
855 u64 bio_offset)
856 {
857 /*
858 * when we're called for a write, we're already in the async
859 * submission context. Just jump into btrfs_map_bio
860 */
861 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
862 }
863
864 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
865 int mirror_num, unsigned long bio_flags,
866 u64 bio_offset)
867 {
868 int ret;
869
870 if (!(rw & REQ_WRITE)) {
871
872 /*
873 * called for a read, do the setup so that checksum validation
874 * can happen in the async kernel threads
875 */
876 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
877 bio, 1);
878 if (ret)
879 return ret;
880 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
881 mirror_num, 0);
882 }
883
884 /*
885 * kthread helpers are used to submit writes so that checksumming
886 * can happen in parallel across all CPUs
887 */
888 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
889 inode, rw, bio, mirror_num, 0,
890 bio_offset,
891 __btree_submit_bio_start,
892 __btree_submit_bio_done);
893 }
894
895 #ifdef CONFIG_MIGRATION
896 static int btree_migratepage(struct address_space *mapping,
897 struct page *newpage, struct page *page,
898 enum migrate_mode mode)
899 {
900 /*
901 * we can't safely write a btree page from here,
902 * we haven't done the locking hook
903 */
904 if (PageDirty(page))
905 return -EAGAIN;
906 /*
907 * Buffers may be managed in a filesystem specific way.
908 * We must have no buffers or drop them.
909 */
910 if (page_has_private(page) &&
911 !try_to_release_page(page, GFP_KERNEL))
912 return -EAGAIN;
913 return migrate_page(mapping, newpage, page, mode);
914 }
915 #endif
916
917
918 static int btree_writepages(struct address_space *mapping,
919 struct writeback_control *wbc)
920 {
921 struct extent_io_tree *tree;
922 tree = &BTRFS_I(mapping->host)->io_tree;
923 if (wbc->sync_mode == WB_SYNC_NONE) {
924 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
925 u64 num_dirty;
926 unsigned long thresh = 32 * 1024 * 1024;
927
928 if (wbc->for_kupdate)
929 return 0;
930
931 /* this is a bit racy, but that's ok */
932 num_dirty = root->fs_info->dirty_metadata_bytes;
933 if (num_dirty < thresh)
934 return 0;
935 }
936 return btree_write_cache_pages(mapping, wbc);
937 }
938
939 static int btree_readpage(struct file *file, struct page *page)
940 {
941 struct extent_io_tree *tree;
942 tree = &BTRFS_I(page->mapping->host)->io_tree;
943 return extent_read_full_page(tree, page, btree_get_extent, 0);
944 }
945
946 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
947 {
948 if (PageWriteback(page) || PageDirty(page))
949 return 0;
950 /*
951 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
952 * slab allocation from alloc_extent_state down the callchain where
953 * it'd hit a BUG_ON as those flags are not allowed.
954 */
955 gfp_flags &= ~GFP_SLAB_BUG_MASK;
956
957 return try_release_extent_buffer(page, gfp_flags);
958 }
959
960 static void btree_invalidatepage(struct page *page, unsigned long offset)
961 {
962 struct extent_io_tree *tree;
963 tree = &BTRFS_I(page->mapping->host)->io_tree;
964 extent_invalidatepage(tree, page, offset);
965 btree_releasepage(page, GFP_NOFS);
966 if (PagePrivate(page)) {
967 printk(KERN_WARNING "btrfs warning page private not zero "
968 "on page %llu\n", (unsigned long long)page_offset(page));
969 ClearPagePrivate(page);
970 set_page_private(page, 0);
971 page_cache_release(page);
972 }
973 }
974
975 static int btree_set_page_dirty(struct page *page)
976 {
977 struct extent_buffer *eb;
978
979 BUG_ON(!PagePrivate(page));
980 eb = (struct extent_buffer *)page->private;
981 BUG_ON(!eb);
982 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
983 BUG_ON(!atomic_read(&eb->refs));
984 btrfs_assert_tree_locked(eb);
985 return __set_page_dirty_nobuffers(page);
986 }
987
988 static const struct address_space_operations btree_aops = {
989 .readpage = btree_readpage,
990 .writepages = btree_writepages,
991 .releasepage = btree_releasepage,
992 .invalidatepage = btree_invalidatepage,
993 #ifdef CONFIG_MIGRATION
994 .migratepage = btree_migratepage,
995 #endif
996 .set_page_dirty = btree_set_page_dirty,
997 };
998
999 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1000 u64 parent_transid)
1001 {
1002 struct extent_buffer *buf = NULL;
1003 struct inode *btree_inode = root->fs_info->btree_inode;
1004 int ret = 0;
1005
1006 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1007 if (!buf)
1008 return 0;
1009 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1010 buf, 0, WAIT_NONE, btree_get_extent, 0);
1011 free_extent_buffer(buf);
1012 return ret;
1013 }
1014
1015 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1016 int mirror_num, struct extent_buffer **eb)
1017 {
1018 struct extent_buffer *buf = NULL;
1019 struct inode *btree_inode = root->fs_info->btree_inode;
1020 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1021 int ret;
1022
1023 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1024 if (!buf)
1025 return 0;
1026
1027 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1028
1029 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1030 btree_get_extent, mirror_num);
1031 if (ret) {
1032 free_extent_buffer(buf);
1033 return ret;
1034 }
1035
1036 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1037 free_extent_buffer(buf);
1038 return -EIO;
1039 } else if (extent_buffer_uptodate(buf)) {
1040 *eb = buf;
1041 } else {
1042 free_extent_buffer(buf);
1043 }
1044 return 0;
1045 }
1046
1047 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1048 u64 bytenr, u32 blocksize)
1049 {
1050 struct inode *btree_inode = root->fs_info->btree_inode;
1051 struct extent_buffer *eb;
1052 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1053 bytenr, blocksize);
1054 return eb;
1055 }
1056
1057 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1058 u64 bytenr, u32 blocksize)
1059 {
1060 struct inode *btree_inode = root->fs_info->btree_inode;
1061 struct extent_buffer *eb;
1062
1063 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1064 bytenr, blocksize);
1065 return eb;
1066 }
1067
1068
1069 int btrfs_write_tree_block(struct extent_buffer *buf)
1070 {
1071 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1072 buf->start + buf->len - 1);
1073 }
1074
1075 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1076 {
1077 return filemap_fdatawait_range(buf->pages[0]->mapping,
1078 buf->start, buf->start + buf->len - 1);
1079 }
1080
1081 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1082 u32 blocksize, u64 parent_transid)
1083 {
1084 struct extent_buffer *buf = NULL;
1085 int ret;
1086
1087 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1088 if (!buf)
1089 return NULL;
1090
1091 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1092 return buf;
1093
1094 }
1095
1096 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1097 struct extent_buffer *buf)
1098 {
1099 if (btrfs_header_generation(buf) ==
1100 root->fs_info->running_transaction->transid) {
1101 btrfs_assert_tree_locked(buf);
1102
1103 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1104 spin_lock(&root->fs_info->delalloc_lock);
1105 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1106 root->fs_info->dirty_metadata_bytes -= buf->len;
1107 else {
1108 spin_unlock(&root->fs_info->delalloc_lock);
1109 btrfs_panic(root->fs_info, -EOVERFLOW,
1110 "Can't clear %lu bytes from "
1111 " dirty_mdatadata_bytes (%lu)",
1112 buf->len,
1113 root->fs_info->dirty_metadata_bytes);
1114 }
1115 spin_unlock(&root->fs_info->delalloc_lock);
1116 }
1117
1118 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1119 btrfs_set_lock_blocking(buf);
1120 clear_extent_buffer_dirty(buf);
1121 }
1122 }
1123
1124 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1125 u32 stripesize, struct btrfs_root *root,
1126 struct btrfs_fs_info *fs_info,
1127 u64 objectid)
1128 {
1129 root->node = NULL;
1130 root->commit_root = NULL;
1131 root->sectorsize = sectorsize;
1132 root->nodesize = nodesize;
1133 root->leafsize = leafsize;
1134 root->stripesize = stripesize;
1135 root->ref_cows = 0;
1136 root->track_dirty = 0;
1137 root->in_radix = 0;
1138 root->orphan_item_inserted = 0;
1139 root->orphan_cleanup_state = 0;
1140
1141 root->objectid = objectid;
1142 root->last_trans = 0;
1143 root->highest_objectid = 0;
1144 root->name = NULL;
1145 root->inode_tree = RB_ROOT;
1146 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1147 root->block_rsv = NULL;
1148 root->orphan_block_rsv = NULL;
1149
1150 INIT_LIST_HEAD(&root->dirty_list);
1151 INIT_LIST_HEAD(&root->orphan_list);
1152 INIT_LIST_HEAD(&root->root_list);
1153 spin_lock_init(&root->orphan_lock);
1154 spin_lock_init(&root->inode_lock);
1155 spin_lock_init(&root->accounting_lock);
1156 mutex_init(&root->objectid_mutex);
1157 mutex_init(&root->log_mutex);
1158 init_waitqueue_head(&root->log_writer_wait);
1159 init_waitqueue_head(&root->log_commit_wait[0]);
1160 init_waitqueue_head(&root->log_commit_wait[1]);
1161 atomic_set(&root->log_commit[0], 0);
1162 atomic_set(&root->log_commit[1], 0);
1163 atomic_set(&root->log_writers, 0);
1164 root->log_batch = 0;
1165 root->log_transid = 0;
1166 root->last_log_commit = 0;
1167 extent_io_tree_init(&root->dirty_log_pages,
1168 fs_info->btree_inode->i_mapping);
1169
1170 memset(&root->root_key, 0, sizeof(root->root_key));
1171 memset(&root->root_item, 0, sizeof(root->root_item));
1172 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1173 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1174 root->defrag_trans_start = fs_info->generation;
1175 init_completion(&root->kobj_unregister);
1176 root->defrag_running = 0;
1177 root->root_key.objectid = objectid;
1178 root->anon_dev = 0;
1179 }
1180
1181 static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1182 struct btrfs_fs_info *fs_info,
1183 u64 objectid,
1184 struct btrfs_root *root)
1185 {
1186 int ret;
1187 u32 blocksize;
1188 u64 generation;
1189
1190 __setup_root(tree_root->nodesize, tree_root->leafsize,
1191 tree_root->sectorsize, tree_root->stripesize,
1192 root, fs_info, objectid);
1193 ret = btrfs_find_last_root(tree_root, objectid,
1194 &root->root_item, &root->root_key);
1195 if (ret > 0)
1196 return -ENOENT;
1197 else if (ret < 0)
1198 return ret;
1199
1200 generation = btrfs_root_generation(&root->root_item);
1201 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1202 root->commit_root = NULL;
1203 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1204 blocksize, generation);
1205 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1206 free_extent_buffer(root->node);
1207 root->node = NULL;
1208 return -EIO;
1209 }
1210 root->commit_root = btrfs_root_node(root);
1211 return 0;
1212 }
1213
1214 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1215 {
1216 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1217 if (root)
1218 root->fs_info = fs_info;
1219 return root;
1220 }
1221
1222 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1223 struct btrfs_fs_info *fs_info)
1224 {
1225 struct btrfs_root *root;
1226 struct btrfs_root *tree_root = fs_info->tree_root;
1227 struct extent_buffer *leaf;
1228
1229 root = btrfs_alloc_root(fs_info);
1230 if (!root)
1231 return ERR_PTR(-ENOMEM);
1232
1233 __setup_root(tree_root->nodesize, tree_root->leafsize,
1234 tree_root->sectorsize, tree_root->stripesize,
1235 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1236
1237 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1238 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1239 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1240 /*
1241 * log trees do not get reference counted because they go away
1242 * before a real commit is actually done. They do store pointers
1243 * to file data extents, and those reference counts still get
1244 * updated (along with back refs to the log tree).
1245 */
1246 root->ref_cows = 0;
1247
1248 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1249 BTRFS_TREE_LOG_OBJECTID, NULL,
1250 0, 0, 0, 0);
1251 if (IS_ERR(leaf)) {
1252 kfree(root);
1253 return ERR_CAST(leaf);
1254 }
1255
1256 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1257 btrfs_set_header_bytenr(leaf, leaf->start);
1258 btrfs_set_header_generation(leaf, trans->transid);
1259 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1260 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1261 root->node = leaf;
1262
1263 write_extent_buffer(root->node, root->fs_info->fsid,
1264 (unsigned long)btrfs_header_fsid(root->node),
1265 BTRFS_FSID_SIZE);
1266 btrfs_mark_buffer_dirty(root->node);
1267 btrfs_tree_unlock(root->node);
1268 return root;
1269 }
1270
1271 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1272 struct btrfs_fs_info *fs_info)
1273 {
1274 struct btrfs_root *log_root;
1275
1276 log_root = alloc_log_tree(trans, fs_info);
1277 if (IS_ERR(log_root))
1278 return PTR_ERR(log_root);
1279 WARN_ON(fs_info->log_root_tree);
1280 fs_info->log_root_tree = log_root;
1281 return 0;
1282 }
1283
1284 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1285 struct btrfs_root *root)
1286 {
1287 struct btrfs_root *log_root;
1288 struct btrfs_inode_item *inode_item;
1289
1290 log_root = alloc_log_tree(trans, root->fs_info);
1291 if (IS_ERR(log_root))
1292 return PTR_ERR(log_root);
1293
1294 log_root->last_trans = trans->transid;
1295 log_root->root_key.offset = root->root_key.objectid;
1296
1297 inode_item = &log_root->root_item.inode;
1298 inode_item->generation = cpu_to_le64(1);
1299 inode_item->size = cpu_to_le64(3);
1300 inode_item->nlink = cpu_to_le32(1);
1301 inode_item->nbytes = cpu_to_le64(root->leafsize);
1302 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1303
1304 btrfs_set_root_node(&log_root->root_item, log_root->node);
1305
1306 WARN_ON(root->log_root);
1307 root->log_root = log_root;
1308 root->log_transid = 0;
1309 root->last_log_commit = 0;
1310 return 0;
1311 }
1312
1313 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1314 struct btrfs_key *location)
1315 {
1316 struct btrfs_root *root;
1317 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1318 struct btrfs_path *path;
1319 struct extent_buffer *l;
1320 u64 generation;
1321 u32 blocksize;
1322 int ret = 0;
1323
1324 root = btrfs_alloc_root(fs_info);
1325 if (!root)
1326 return ERR_PTR(-ENOMEM);
1327 if (location->offset == (u64)-1) {
1328 ret = find_and_setup_root(tree_root, fs_info,
1329 location->objectid, root);
1330 if (ret) {
1331 kfree(root);
1332 return ERR_PTR(ret);
1333 }
1334 goto out;
1335 }
1336
1337 __setup_root(tree_root->nodesize, tree_root->leafsize,
1338 tree_root->sectorsize, tree_root->stripesize,
1339 root, fs_info, location->objectid);
1340
1341 path = btrfs_alloc_path();
1342 if (!path) {
1343 kfree(root);
1344 return ERR_PTR(-ENOMEM);
1345 }
1346 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1347 if (ret == 0) {
1348 l = path->nodes[0];
1349 read_extent_buffer(l, &root->root_item,
1350 btrfs_item_ptr_offset(l, path->slots[0]),
1351 sizeof(root->root_item));
1352 memcpy(&root->root_key, location, sizeof(*location));
1353 }
1354 btrfs_free_path(path);
1355 if (ret) {
1356 kfree(root);
1357 if (ret > 0)
1358 ret = -ENOENT;
1359 return ERR_PTR(ret);
1360 }
1361
1362 generation = btrfs_root_generation(&root->root_item);
1363 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1364 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1365 blocksize, generation);
1366 root->commit_root = btrfs_root_node(root);
1367 BUG_ON(!root->node); /* -ENOMEM */
1368 out:
1369 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1370 root->ref_cows = 1;
1371 btrfs_check_and_init_root_item(&root->root_item);
1372 }
1373
1374 return root;
1375 }
1376
1377 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1378 struct btrfs_key *location)
1379 {
1380 struct btrfs_root *root;
1381 int ret;
1382
1383 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1384 return fs_info->tree_root;
1385 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1386 return fs_info->extent_root;
1387 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1388 return fs_info->chunk_root;
1389 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1390 return fs_info->dev_root;
1391 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1392 return fs_info->csum_root;
1393 again:
1394 spin_lock(&fs_info->fs_roots_radix_lock);
1395 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1396 (unsigned long)location->objectid);
1397 spin_unlock(&fs_info->fs_roots_radix_lock);
1398 if (root)
1399 return root;
1400
1401 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1402 if (IS_ERR(root))
1403 return root;
1404
1405 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1406 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1407 GFP_NOFS);
1408 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1409 ret = -ENOMEM;
1410 goto fail;
1411 }
1412
1413 btrfs_init_free_ino_ctl(root);
1414 mutex_init(&root->fs_commit_mutex);
1415 spin_lock_init(&root->cache_lock);
1416 init_waitqueue_head(&root->cache_wait);
1417
1418 ret = get_anon_bdev(&root->anon_dev);
1419 if (ret)
1420 goto fail;
1421
1422 if (btrfs_root_refs(&root->root_item) == 0) {
1423 ret = -ENOENT;
1424 goto fail;
1425 }
1426
1427 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1428 if (ret < 0)
1429 goto fail;
1430 if (ret == 0)
1431 root->orphan_item_inserted = 1;
1432
1433 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1434 if (ret)
1435 goto fail;
1436
1437 spin_lock(&fs_info->fs_roots_radix_lock);
1438 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1439 (unsigned long)root->root_key.objectid,
1440 root);
1441 if (ret == 0)
1442 root->in_radix = 1;
1443
1444 spin_unlock(&fs_info->fs_roots_radix_lock);
1445 radix_tree_preload_end();
1446 if (ret) {
1447 if (ret == -EEXIST) {
1448 free_fs_root(root);
1449 goto again;
1450 }
1451 goto fail;
1452 }
1453
1454 ret = btrfs_find_dead_roots(fs_info->tree_root,
1455 root->root_key.objectid);
1456 WARN_ON(ret);
1457 return root;
1458 fail:
1459 free_fs_root(root);
1460 return ERR_PTR(ret);
1461 }
1462
1463 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1464 {
1465 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1466 int ret = 0;
1467 struct btrfs_device *device;
1468 struct backing_dev_info *bdi;
1469
1470 rcu_read_lock();
1471 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1472 if (!device->bdev)
1473 continue;
1474 bdi = blk_get_backing_dev_info(device->bdev);
1475 if (bdi && bdi_congested(bdi, bdi_bits)) {
1476 ret = 1;
1477 break;
1478 }
1479 }
1480 rcu_read_unlock();
1481 return ret;
1482 }
1483
1484 /*
1485 * If this fails, caller must call bdi_destroy() to get rid of the
1486 * bdi again.
1487 */
1488 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1489 {
1490 int err;
1491
1492 bdi->capabilities = BDI_CAP_MAP_COPY;
1493 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1494 if (err)
1495 return err;
1496
1497 bdi->ra_pages = default_backing_dev_info.ra_pages;
1498 bdi->congested_fn = btrfs_congested_fn;
1499 bdi->congested_data = info;
1500 return 0;
1501 }
1502
1503 /*
1504 * called by the kthread helper functions to finally call the bio end_io
1505 * functions. This is where read checksum verification actually happens
1506 */
1507 static void end_workqueue_fn(struct btrfs_work *work)
1508 {
1509 struct bio *bio;
1510 struct end_io_wq *end_io_wq;
1511 struct btrfs_fs_info *fs_info;
1512 int error;
1513
1514 end_io_wq = container_of(work, struct end_io_wq, work);
1515 bio = end_io_wq->bio;
1516 fs_info = end_io_wq->info;
1517
1518 error = end_io_wq->error;
1519 bio->bi_private = end_io_wq->private;
1520 bio->bi_end_io = end_io_wq->end_io;
1521 kfree(end_io_wq);
1522 bio_endio(bio, error);
1523 }
1524
1525 static int cleaner_kthread(void *arg)
1526 {
1527 struct btrfs_root *root = arg;
1528
1529 do {
1530 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1531
1532 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1533 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1534 btrfs_run_delayed_iputs(root);
1535 btrfs_clean_old_snapshots(root);
1536 mutex_unlock(&root->fs_info->cleaner_mutex);
1537 btrfs_run_defrag_inodes(root->fs_info);
1538 }
1539
1540 if (!try_to_freeze()) {
1541 set_current_state(TASK_INTERRUPTIBLE);
1542 if (!kthread_should_stop())
1543 schedule();
1544 __set_current_state(TASK_RUNNING);
1545 }
1546 } while (!kthread_should_stop());
1547 return 0;
1548 }
1549
1550 static int transaction_kthread(void *arg)
1551 {
1552 struct btrfs_root *root = arg;
1553 struct btrfs_trans_handle *trans;
1554 struct btrfs_transaction *cur;
1555 u64 transid;
1556 unsigned long now;
1557 unsigned long delay;
1558 bool cannot_commit;
1559
1560 do {
1561 cannot_commit = false;
1562 delay = HZ * 30;
1563 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1564 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1565
1566 spin_lock(&root->fs_info->trans_lock);
1567 cur = root->fs_info->running_transaction;
1568 if (!cur) {
1569 spin_unlock(&root->fs_info->trans_lock);
1570 goto sleep;
1571 }
1572
1573 now = get_seconds();
1574 if (!cur->blocked &&
1575 (now < cur->start_time || now - cur->start_time < 30)) {
1576 spin_unlock(&root->fs_info->trans_lock);
1577 delay = HZ * 5;
1578 goto sleep;
1579 }
1580 transid = cur->transid;
1581 spin_unlock(&root->fs_info->trans_lock);
1582
1583 /* If the file system is aborted, this will always fail. */
1584 trans = btrfs_join_transaction(root);
1585 if (IS_ERR(trans)) {
1586 cannot_commit = true;
1587 goto sleep;
1588 }
1589 if (transid == trans->transid) {
1590 btrfs_commit_transaction(trans, root);
1591 } else {
1592 btrfs_end_transaction(trans, root);
1593 }
1594 sleep:
1595 wake_up_process(root->fs_info->cleaner_kthread);
1596 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1597
1598 if (!try_to_freeze()) {
1599 set_current_state(TASK_INTERRUPTIBLE);
1600 if (!kthread_should_stop() &&
1601 (!btrfs_transaction_blocked(root->fs_info) ||
1602 cannot_commit))
1603 schedule_timeout(delay);
1604 __set_current_state(TASK_RUNNING);
1605 }
1606 } while (!kthread_should_stop());
1607 return 0;
1608 }
1609
1610 /*
1611 * this will find the highest generation in the array of
1612 * root backups. The index of the highest array is returned,
1613 * or -1 if we can't find anything.
1614 *
1615 * We check to make sure the array is valid by comparing the
1616 * generation of the latest root in the array with the generation
1617 * in the super block. If they don't match we pitch it.
1618 */
1619 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1620 {
1621 u64 cur;
1622 int newest_index = -1;
1623 struct btrfs_root_backup *root_backup;
1624 int i;
1625
1626 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1627 root_backup = info->super_copy->super_roots + i;
1628 cur = btrfs_backup_tree_root_gen(root_backup);
1629 if (cur == newest_gen)
1630 newest_index = i;
1631 }
1632
1633 /* check to see if we actually wrapped around */
1634 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1635 root_backup = info->super_copy->super_roots;
1636 cur = btrfs_backup_tree_root_gen(root_backup);
1637 if (cur == newest_gen)
1638 newest_index = 0;
1639 }
1640 return newest_index;
1641 }
1642
1643
1644 /*
1645 * find the oldest backup so we know where to store new entries
1646 * in the backup array. This will set the backup_root_index
1647 * field in the fs_info struct
1648 */
1649 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1650 u64 newest_gen)
1651 {
1652 int newest_index = -1;
1653
1654 newest_index = find_newest_super_backup(info, newest_gen);
1655 /* if there was garbage in there, just move along */
1656 if (newest_index == -1) {
1657 info->backup_root_index = 0;
1658 } else {
1659 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1660 }
1661 }
1662
1663 /*
1664 * copy all the root pointers into the super backup array.
1665 * this will bump the backup pointer by one when it is
1666 * done
1667 */
1668 static void backup_super_roots(struct btrfs_fs_info *info)
1669 {
1670 int next_backup;
1671 struct btrfs_root_backup *root_backup;
1672 int last_backup;
1673
1674 next_backup = info->backup_root_index;
1675 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1676 BTRFS_NUM_BACKUP_ROOTS;
1677
1678 /*
1679 * just overwrite the last backup if we're at the same generation
1680 * this happens only at umount
1681 */
1682 root_backup = info->super_for_commit->super_roots + last_backup;
1683 if (btrfs_backup_tree_root_gen(root_backup) ==
1684 btrfs_header_generation(info->tree_root->node))
1685 next_backup = last_backup;
1686
1687 root_backup = info->super_for_commit->super_roots + next_backup;
1688
1689 /*
1690 * make sure all of our padding and empty slots get zero filled
1691 * regardless of which ones we use today
1692 */
1693 memset(root_backup, 0, sizeof(*root_backup));
1694
1695 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1696
1697 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1698 btrfs_set_backup_tree_root_gen(root_backup,
1699 btrfs_header_generation(info->tree_root->node));
1700
1701 btrfs_set_backup_tree_root_level(root_backup,
1702 btrfs_header_level(info->tree_root->node));
1703
1704 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1705 btrfs_set_backup_chunk_root_gen(root_backup,
1706 btrfs_header_generation(info->chunk_root->node));
1707 btrfs_set_backup_chunk_root_level(root_backup,
1708 btrfs_header_level(info->chunk_root->node));
1709
1710 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1711 btrfs_set_backup_extent_root_gen(root_backup,
1712 btrfs_header_generation(info->extent_root->node));
1713 btrfs_set_backup_extent_root_level(root_backup,
1714 btrfs_header_level(info->extent_root->node));
1715
1716 /*
1717 * we might commit during log recovery, which happens before we set
1718 * the fs_root. Make sure it is valid before we fill it in.
1719 */
1720 if (info->fs_root && info->fs_root->node) {
1721 btrfs_set_backup_fs_root(root_backup,
1722 info->fs_root->node->start);
1723 btrfs_set_backup_fs_root_gen(root_backup,
1724 btrfs_header_generation(info->fs_root->node));
1725 btrfs_set_backup_fs_root_level(root_backup,
1726 btrfs_header_level(info->fs_root->node));
1727 }
1728
1729 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1730 btrfs_set_backup_dev_root_gen(root_backup,
1731 btrfs_header_generation(info->dev_root->node));
1732 btrfs_set_backup_dev_root_level(root_backup,
1733 btrfs_header_level(info->dev_root->node));
1734
1735 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1736 btrfs_set_backup_csum_root_gen(root_backup,
1737 btrfs_header_generation(info->csum_root->node));
1738 btrfs_set_backup_csum_root_level(root_backup,
1739 btrfs_header_level(info->csum_root->node));
1740
1741 btrfs_set_backup_total_bytes(root_backup,
1742 btrfs_super_total_bytes(info->super_copy));
1743 btrfs_set_backup_bytes_used(root_backup,
1744 btrfs_super_bytes_used(info->super_copy));
1745 btrfs_set_backup_num_devices(root_backup,
1746 btrfs_super_num_devices(info->super_copy));
1747
1748 /*
1749 * if we don't copy this out to the super_copy, it won't get remembered
1750 * for the next commit
1751 */
1752 memcpy(&info->super_copy->super_roots,
1753 &info->super_for_commit->super_roots,
1754 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1755 }
1756
1757 /*
1758 * this copies info out of the root backup array and back into
1759 * the in-memory super block. It is meant to help iterate through
1760 * the array, so you send it the number of backups you've already
1761 * tried and the last backup index you used.
1762 *
1763 * this returns -1 when it has tried all the backups
1764 */
1765 static noinline int next_root_backup(struct btrfs_fs_info *info,
1766 struct btrfs_super_block *super,
1767 int *num_backups_tried, int *backup_index)
1768 {
1769 struct btrfs_root_backup *root_backup;
1770 int newest = *backup_index;
1771
1772 if (*num_backups_tried == 0) {
1773 u64 gen = btrfs_super_generation(super);
1774
1775 newest = find_newest_super_backup(info, gen);
1776 if (newest == -1)
1777 return -1;
1778
1779 *backup_index = newest;
1780 *num_backups_tried = 1;
1781 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1782 /* we've tried all the backups, all done */
1783 return -1;
1784 } else {
1785 /* jump to the next oldest backup */
1786 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1787 BTRFS_NUM_BACKUP_ROOTS;
1788 *backup_index = newest;
1789 *num_backups_tried += 1;
1790 }
1791 root_backup = super->super_roots + newest;
1792
1793 btrfs_set_super_generation(super,
1794 btrfs_backup_tree_root_gen(root_backup));
1795 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1796 btrfs_set_super_root_level(super,
1797 btrfs_backup_tree_root_level(root_backup));
1798 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1799
1800 /*
1801 * fixme: the total bytes and num_devices need to match or we should
1802 * need a fsck
1803 */
1804 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1805 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1806 return 0;
1807 }
1808
1809 /* helper to cleanup tree roots */
1810 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1811 {
1812 free_extent_buffer(info->tree_root->node);
1813 free_extent_buffer(info->tree_root->commit_root);
1814 free_extent_buffer(info->dev_root->node);
1815 free_extent_buffer(info->dev_root->commit_root);
1816 free_extent_buffer(info->extent_root->node);
1817 free_extent_buffer(info->extent_root->commit_root);
1818 free_extent_buffer(info->csum_root->node);
1819 free_extent_buffer(info->csum_root->commit_root);
1820
1821 info->tree_root->node = NULL;
1822 info->tree_root->commit_root = NULL;
1823 info->dev_root->node = NULL;
1824 info->dev_root->commit_root = NULL;
1825 info->extent_root->node = NULL;
1826 info->extent_root->commit_root = NULL;
1827 info->csum_root->node = NULL;
1828 info->csum_root->commit_root = NULL;
1829
1830 if (chunk_root) {
1831 free_extent_buffer(info->chunk_root->node);
1832 free_extent_buffer(info->chunk_root->commit_root);
1833 info->chunk_root->node = NULL;
1834 info->chunk_root->commit_root = NULL;
1835 }
1836 }
1837
1838
1839 int open_ctree(struct super_block *sb,
1840 struct btrfs_fs_devices *fs_devices,
1841 char *options)
1842 {
1843 u32 sectorsize;
1844 u32 nodesize;
1845 u32 leafsize;
1846 u32 blocksize;
1847 u32 stripesize;
1848 u64 generation;
1849 u64 features;
1850 struct btrfs_key location;
1851 struct buffer_head *bh;
1852 struct btrfs_super_block *disk_super;
1853 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1854 struct btrfs_root *tree_root;
1855 struct btrfs_root *extent_root;
1856 struct btrfs_root *csum_root;
1857 struct btrfs_root *chunk_root;
1858 struct btrfs_root *dev_root;
1859 struct btrfs_root *log_tree_root;
1860 int ret;
1861 int err = -EINVAL;
1862 int num_backups_tried = 0;
1863 int backup_index = 0;
1864
1865 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1866 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1867 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1868 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1869 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1870
1871 if (!tree_root || !extent_root || !csum_root ||
1872 !chunk_root || !dev_root) {
1873 err = -ENOMEM;
1874 goto fail;
1875 }
1876
1877 ret = init_srcu_struct(&fs_info->subvol_srcu);
1878 if (ret) {
1879 err = ret;
1880 goto fail;
1881 }
1882
1883 ret = setup_bdi(fs_info, &fs_info->bdi);
1884 if (ret) {
1885 err = ret;
1886 goto fail_srcu;
1887 }
1888
1889 fs_info->btree_inode = new_inode(sb);
1890 if (!fs_info->btree_inode) {
1891 err = -ENOMEM;
1892 goto fail_bdi;
1893 }
1894
1895 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1896
1897 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1898 INIT_LIST_HEAD(&fs_info->trans_list);
1899 INIT_LIST_HEAD(&fs_info->dead_roots);
1900 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1901 INIT_LIST_HEAD(&fs_info->hashers);
1902 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1903 INIT_LIST_HEAD(&fs_info->ordered_operations);
1904 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1905 spin_lock_init(&fs_info->delalloc_lock);
1906 spin_lock_init(&fs_info->trans_lock);
1907 spin_lock_init(&fs_info->ref_cache_lock);
1908 spin_lock_init(&fs_info->fs_roots_radix_lock);
1909 spin_lock_init(&fs_info->delayed_iput_lock);
1910 spin_lock_init(&fs_info->defrag_inodes_lock);
1911 spin_lock_init(&fs_info->free_chunk_lock);
1912 mutex_init(&fs_info->reloc_mutex);
1913
1914 init_completion(&fs_info->kobj_unregister);
1915 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1916 INIT_LIST_HEAD(&fs_info->space_info);
1917 btrfs_mapping_init(&fs_info->mapping_tree);
1918 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1919 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1920 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1921 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1922 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1923 btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1924 atomic_set(&fs_info->nr_async_submits, 0);
1925 atomic_set(&fs_info->async_delalloc_pages, 0);
1926 atomic_set(&fs_info->async_submit_draining, 0);
1927 atomic_set(&fs_info->nr_async_bios, 0);
1928 atomic_set(&fs_info->defrag_running, 0);
1929 fs_info->sb = sb;
1930 fs_info->max_inline = 8192 * 1024;
1931 fs_info->metadata_ratio = 0;
1932 fs_info->defrag_inodes = RB_ROOT;
1933 fs_info->trans_no_join = 0;
1934 fs_info->free_chunk_space = 0;
1935
1936 /* readahead state */
1937 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1938 spin_lock_init(&fs_info->reada_lock);
1939
1940 fs_info->thread_pool_size = min_t(unsigned long,
1941 num_online_cpus() + 2, 8);
1942
1943 INIT_LIST_HEAD(&fs_info->ordered_extents);
1944 spin_lock_init(&fs_info->ordered_extent_lock);
1945 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1946 GFP_NOFS);
1947 if (!fs_info->delayed_root) {
1948 err = -ENOMEM;
1949 goto fail_iput;
1950 }
1951 btrfs_init_delayed_root(fs_info->delayed_root);
1952
1953 mutex_init(&fs_info->scrub_lock);
1954 atomic_set(&fs_info->scrubs_running, 0);
1955 atomic_set(&fs_info->scrub_pause_req, 0);
1956 atomic_set(&fs_info->scrubs_paused, 0);
1957 atomic_set(&fs_info->scrub_cancel_req, 0);
1958 init_waitqueue_head(&fs_info->scrub_pause_wait);
1959 init_rwsem(&fs_info->scrub_super_lock);
1960 fs_info->scrub_workers_refcnt = 0;
1961 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1962 fs_info->check_integrity_print_mask = 0;
1963 #endif
1964
1965 spin_lock_init(&fs_info->balance_lock);
1966 mutex_init(&fs_info->balance_mutex);
1967 atomic_set(&fs_info->balance_running, 0);
1968 atomic_set(&fs_info->balance_pause_req, 0);
1969 atomic_set(&fs_info->balance_cancel_req, 0);
1970 fs_info->balance_ctl = NULL;
1971 init_waitqueue_head(&fs_info->balance_wait_q);
1972
1973 sb->s_blocksize = 4096;
1974 sb->s_blocksize_bits = blksize_bits(4096);
1975 sb->s_bdi = &fs_info->bdi;
1976
1977 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1978 set_nlink(fs_info->btree_inode, 1);
1979 /*
1980 * we set the i_size on the btree inode to the max possible int.
1981 * the real end of the address space is determined by all of
1982 * the devices in the system
1983 */
1984 fs_info->btree_inode->i_size = OFFSET_MAX;
1985 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1986 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1987
1988 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1989 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1990 fs_info->btree_inode->i_mapping);
1991 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
1992 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1993
1994 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1995
1996 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1997 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1998 sizeof(struct btrfs_key));
1999 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
2000 insert_inode_hash(fs_info->btree_inode);
2001
2002 spin_lock_init(&fs_info->block_group_cache_lock);
2003 fs_info->block_group_cache_tree = RB_ROOT;
2004
2005 extent_io_tree_init(&fs_info->freed_extents[0],
2006 fs_info->btree_inode->i_mapping);
2007 extent_io_tree_init(&fs_info->freed_extents[1],
2008 fs_info->btree_inode->i_mapping);
2009 fs_info->pinned_extents = &fs_info->freed_extents[0];
2010 fs_info->do_barriers = 1;
2011
2012
2013 mutex_init(&fs_info->ordered_operations_mutex);
2014 mutex_init(&fs_info->tree_log_mutex);
2015 mutex_init(&fs_info->chunk_mutex);
2016 mutex_init(&fs_info->transaction_kthread_mutex);
2017 mutex_init(&fs_info->cleaner_mutex);
2018 mutex_init(&fs_info->volume_mutex);
2019 init_rwsem(&fs_info->extent_commit_sem);
2020 init_rwsem(&fs_info->cleanup_work_sem);
2021 init_rwsem(&fs_info->subvol_sem);
2022
2023 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2024 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2025
2026 init_waitqueue_head(&fs_info->transaction_throttle);
2027 init_waitqueue_head(&fs_info->transaction_wait);
2028 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2029 init_waitqueue_head(&fs_info->async_submit_wait);
2030
2031 __setup_root(4096, 4096, 4096, 4096, tree_root,
2032 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2033
2034 invalidate_bdev(fs_devices->latest_bdev);
2035 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2036 if (!bh) {
2037 err = -EINVAL;
2038 goto fail_alloc;
2039 }
2040
2041 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2042 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2043 sizeof(*fs_info->super_for_commit));
2044 brelse(bh);
2045
2046 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2047
2048 disk_super = fs_info->super_copy;
2049 if (!btrfs_super_root(disk_super))
2050 goto fail_alloc;
2051
2052 /* check FS state, whether FS is broken. */
2053 fs_info->fs_state |= btrfs_super_flags(disk_super);
2054
2055 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2056 if (ret) {
2057 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2058 err = ret;
2059 goto fail_alloc;
2060 }
2061
2062 /*
2063 * run through our array of backup supers and setup
2064 * our ring pointer to the oldest one
2065 */
2066 generation = btrfs_super_generation(disk_super);
2067 find_oldest_super_backup(fs_info, generation);
2068
2069 /*
2070 * In the long term, we'll store the compression type in the super
2071 * block, and it'll be used for per file compression control.
2072 */
2073 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2074
2075 ret = btrfs_parse_options(tree_root, options);
2076 if (ret) {
2077 err = ret;
2078 goto fail_alloc;
2079 }
2080
2081 features = btrfs_super_incompat_flags(disk_super) &
2082 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2083 if (features) {
2084 printk(KERN_ERR "BTRFS: couldn't mount because of "
2085 "unsupported optional features (%Lx).\n",
2086 (unsigned long long)features);
2087 err = -EINVAL;
2088 goto fail_alloc;
2089 }
2090
2091 if (btrfs_super_leafsize(disk_super) !=
2092 btrfs_super_nodesize(disk_super)) {
2093 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2094 "blocksizes don't match. node %d leaf %d\n",
2095 btrfs_super_nodesize(disk_super),
2096 btrfs_super_leafsize(disk_super));
2097 err = -EINVAL;
2098 goto fail_alloc;
2099 }
2100 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2101 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2102 "blocksize (%d) was too large\n",
2103 btrfs_super_leafsize(disk_super));
2104 err = -EINVAL;
2105 goto fail_alloc;
2106 }
2107
2108 features = btrfs_super_incompat_flags(disk_super);
2109 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2110 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2111 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2112
2113 /*
2114 * flag our filesystem as having big metadata blocks if
2115 * they are bigger than the page size
2116 */
2117 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2118 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2119 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2120 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2121 }
2122
2123 nodesize = btrfs_super_nodesize(disk_super);
2124 leafsize = btrfs_super_leafsize(disk_super);
2125 sectorsize = btrfs_super_sectorsize(disk_super);
2126 stripesize = btrfs_super_stripesize(disk_super);
2127
2128 /*
2129 * mixed block groups end up with duplicate but slightly offset
2130 * extent buffers for the same range. It leads to corruptions
2131 */
2132 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2133 (sectorsize != leafsize)) {
2134 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2135 "are not allowed for mixed block groups on %s\n",
2136 sb->s_id);
2137 goto fail_alloc;
2138 }
2139
2140 btrfs_set_super_incompat_flags(disk_super, features);
2141
2142 features = btrfs_super_compat_ro_flags(disk_super) &
2143 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2144 if (!(sb->s_flags & MS_RDONLY) && features) {
2145 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2146 "unsupported option features (%Lx).\n",
2147 (unsigned long long)features);
2148 err = -EINVAL;
2149 goto fail_alloc;
2150 }
2151
2152 btrfs_init_workers(&fs_info->generic_worker,
2153 "genwork", 1, NULL);
2154
2155 btrfs_init_workers(&fs_info->workers, "worker",
2156 fs_info->thread_pool_size,
2157 &fs_info->generic_worker);
2158
2159 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2160 fs_info->thread_pool_size,
2161 &fs_info->generic_worker);
2162
2163 btrfs_init_workers(&fs_info->submit_workers, "submit",
2164 min_t(u64, fs_devices->num_devices,
2165 fs_info->thread_pool_size),
2166 &fs_info->generic_worker);
2167
2168 btrfs_init_workers(&fs_info->caching_workers, "cache",
2169 2, &fs_info->generic_worker);
2170
2171 /* a higher idle thresh on the submit workers makes it much more
2172 * likely that bios will be send down in a sane order to the
2173 * devices
2174 */
2175 fs_info->submit_workers.idle_thresh = 64;
2176
2177 fs_info->workers.idle_thresh = 16;
2178 fs_info->workers.ordered = 1;
2179
2180 fs_info->delalloc_workers.idle_thresh = 2;
2181 fs_info->delalloc_workers.ordered = 1;
2182
2183 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2184 &fs_info->generic_worker);
2185 btrfs_init_workers(&fs_info->endio_workers, "endio",
2186 fs_info->thread_pool_size,
2187 &fs_info->generic_worker);
2188 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2189 fs_info->thread_pool_size,
2190 &fs_info->generic_worker);
2191 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2192 "endio-meta-write", fs_info->thread_pool_size,
2193 &fs_info->generic_worker);
2194 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2195 fs_info->thread_pool_size,
2196 &fs_info->generic_worker);
2197 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2198 1, &fs_info->generic_worker);
2199 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2200 fs_info->thread_pool_size,
2201 &fs_info->generic_worker);
2202 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2203 fs_info->thread_pool_size,
2204 &fs_info->generic_worker);
2205
2206 /*
2207 * endios are largely parallel and should have a very
2208 * low idle thresh
2209 */
2210 fs_info->endio_workers.idle_thresh = 4;
2211 fs_info->endio_meta_workers.idle_thresh = 4;
2212
2213 fs_info->endio_write_workers.idle_thresh = 2;
2214 fs_info->endio_meta_write_workers.idle_thresh = 2;
2215 fs_info->readahead_workers.idle_thresh = 2;
2216
2217 /*
2218 * btrfs_start_workers can really only fail because of ENOMEM so just
2219 * return -ENOMEM if any of these fail.
2220 */
2221 ret = btrfs_start_workers(&fs_info->workers);
2222 ret |= btrfs_start_workers(&fs_info->generic_worker);
2223 ret |= btrfs_start_workers(&fs_info->submit_workers);
2224 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2225 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2226 ret |= btrfs_start_workers(&fs_info->endio_workers);
2227 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2228 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2229 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2230 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2231 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2232 ret |= btrfs_start_workers(&fs_info->caching_workers);
2233 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2234 if (ret) {
2235 ret = -ENOMEM;
2236 goto fail_sb_buffer;
2237 }
2238
2239 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2240 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2241 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2242
2243 tree_root->nodesize = nodesize;
2244 tree_root->leafsize = leafsize;
2245 tree_root->sectorsize = sectorsize;
2246 tree_root->stripesize = stripesize;
2247
2248 sb->s_blocksize = sectorsize;
2249 sb->s_blocksize_bits = blksize_bits(sectorsize);
2250
2251 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2252 sizeof(disk_super->magic))) {
2253 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2254 goto fail_sb_buffer;
2255 }
2256
2257 if (sectorsize != PAGE_SIZE) {
2258 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2259 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2260 goto fail_sb_buffer;
2261 }
2262
2263 mutex_lock(&fs_info->chunk_mutex);
2264 ret = btrfs_read_sys_array(tree_root);
2265 mutex_unlock(&fs_info->chunk_mutex);
2266 if (ret) {
2267 printk(KERN_WARNING "btrfs: failed to read the system "
2268 "array on %s\n", sb->s_id);
2269 goto fail_sb_buffer;
2270 }
2271
2272 blocksize = btrfs_level_size(tree_root,
2273 btrfs_super_chunk_root_level(disk_super));
2274 generation = btrfs_super_chunk_root_generation(disk_super);
2275
2276 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2277 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2278
2279 chunk_root->node = read_tree_block(chunk_root,
2280 btrfs_super_chunk_root(disk_super),
2281 blocksize, generation);
2282 BUG_ON(!chunk_root->node); /* -ENOMEM */
2283 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2284 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2285 sb->s_id);
2286 goto fail_tree_roots;
2287 }
2288 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2289 chunk_root->commit_root = btrfs_root_node(chunk_root);
2290
2291 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2292 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2293 BTRFS_UUID_SIZE);
2294
2295 ret = btrfs_read_chunk_tree(chunk_root);
2296 if (ret) {
2297 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2298 sb->s_id);
2299 goto fail_tree_roots;
2300 }
2301
2302 btrfs_close_extra_devices(fs_devices);
2303
2304 if (!fs_devices->latest_bdev) {
2305 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2306 sb->s_id);
2307 goto fail_tree_roots;
2308 }
2309
2310 retry_root_backup:
2311 blocksize = btrfs_level_size(tree_root,
2312 btrfs_super_root_level(disk_super));
2313 generation = btrfs_super_generation(disk_super);
2314
2315 tree_root->node = read_tree_block(tree_root,
2316 btrfs_super_root(disk_super),
2317 blocksize, generation);
2318 if (!tree_root->node ||
2319 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2320 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2321 sb->s_id);
2322
2323 goto recovery_tree_root;
2324 }
2325
2326 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2327 tree_root->commit_root = btrfs_root_node(tree_root);
2328
2329 ret = find_and_setup_root(tree_root, fs_info,
2330 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2331 if (ret)
2332 goto recovery_tree_root;
2333 extent_root->track_dirty = 1;
2334
2335 ret = find_and_setup_root(tree_root, fs_info,
2336 BTRFS_DEV_TREE_OBJECTID, dev_root);
2337 if (ret)
2338 goto recovery_tree_root;
2339 dev_root->track_dirty = 1;
2340
2341 ret = find_and_setup_root(tree_root, fs_info,
2342 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2343 if (ret)
2344 goto recovery_tree_root;
2345
2346 csum_root->track_dirty = 1;
2347
2348 fs_info->generation = generation;
2349 fs_info->last_trans_committed = generation;
2350
2351 ret = btrfs_init_space_info(fs_info);
2352 if (ret) {
2353 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2354 goto fail_block_groups;
2355 }
2356
2357 ret = btrfs_read_block_groups(extent_root);
2358 if (ret) {
2359 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2360 goto fail_block_groups;
2361 }
2362
2363 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2364 "btrfs-cleaner");
2365 if (IS_ERR(fs_info->cleaner_kthread))
2366 goto fail_block_groups;
2367
2368 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2369 tree_root,
2370 "btrfs-transaction");
2371 if (IS_ERR(fs_info->transaction_kthread))
2372 goto fail_cleaner;
2373
2374 if (!btrfs_test_opt(tree_root, SSD) &&
2375 !btrfs_test_opt(tree_root, NOSSD) &&
2376 !fs_info->fs_devices->rotating) {
2377 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2378 "mode\n");
2379 btrfs_set_opt(fs_info->mount_opt, SSD);
2380 }
2381
2382 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2383 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2384 ret = btrfsic_mount(tree_root, fs_devices,
2385 btrfs_test_opt(tree_root,
2386 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2387 1 : 0,
2388 fs_info->check_integrity_print_mask);
2389 if (ret)
2390 printk(KERN_WARNING "btrfs: failed to initialize"
2391 " integrity check module %s\n", sb->s_id);
2392 }
2393 #endif
2394
2395 /* do not make disk changes in broken FS */
2396 if (btrfs_super_log_root(disk_super) != 0 &&
2397 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2398 u64 bytenr = btrfs_super_log_root(disk_super);
2399
2400 if (fs_devices->rw_devices == 0) {
2401 printk(KERN_WARNING "Btrfs log replay required "
2402 "on RO media\n");
2403 err = -EIO;
2404 goto fail_trans_kthread;
2405 }
2406 blocksize =
2407 btrfs_level_size(tree_root,
2408 btrfs_super_log_root_level(disk_super));
2409
2410 log_tree_root = btrfs_alloc_root(fs_info);
2411 if (!log_tree_root) {
2412 err = -ENOMEM;
2413 goto fail_trans_kthread;
2414 }
2415
2416 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2417 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2418
2419 log_tree_root->node = read_tree_block(tree_root, bytenr,
2420 blocksize,
2421 generation + 1);
2422 /* returns with log_tree_root freed on success */
2423 ret = btrfs_recover_log_trees(log_tree_root);
2424 if (ret) {
2425 btrfs_error(tree_root->fs_info, ret,
2426 "Failed to recover log tree");
2427 free_extent_buffer(log_tree_root->node);
2428 kfree(log_tree_root);
2429 goto fail_trans_kthread;
2430 }
2431
2432 if (sb->s_flags & MS_RDONLY) {
2433 ret = btrfs_commit_super(tree_root);
2434 if (ret)
2435 goto fail_trans_kthread;
2436 }
2437 }
2438
2439 ret = btrfs_find_orphan_roots(tree_root);
2440 if (ret)
2441 goto fail_trans_kthread;
2442
2443 if (!(sb->s_flags & MS_RDONLY)) {
2444 ret = btrfs_cleanup_fs_roots(fs_info);
2445 if (ret) {
2446 }
2447
2448 ret = btrfs_recover_relocation(tree_root);
2449 if (ret < 0) {
2450 printk(KERN_WARNING
2451 "btrfs: failed to recover relocation\n");
2452 err = -EINVAL;
2453 goto fail_trans_kthread;
2454 }
2455 }
2456
2457 location.objectid = BTRFS_FS_TREE_OBJECTID;
2458 location.type = BTRFS_ROOT_ITEM_KEY;
2459 location.offset = (u64)-1;
2460
2461 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2462 if (!fs_info->fs_root)
2463 goto fail_trans_kthread;
2464 if (IS_ERR(fs_info->fs_root)) {
2465 err = PTR_ERR(fs_info->fs_root);
2466 goto fail_trans_kthread;
2467 }
2468
2469 if (!(sb->s_flags & MS_RDONLY)) {
2470 down_read(&fs_info->cleanup_work_sem);
2471 err = btrfs_orphan_cleanup(fs_info->fs_root);
2472 if (!err)
2473 err = btrfs_orphan_cleanup(fs_info->tree_root);
2474 up_read(&fs_info->cleanup_work_sem);
2475
2476 if (!err)
2477 err = btrfs_recover_balance(fs_info->tree_root);
2478
2479 if (err) {
2480 close_ctree(tree_root);
2481 return err;
2482 }
2483 }
2484
2485 return 0;
2486
2487 fail_trans_kthread:
2488 kthread_stop(fs_info->transaction_kthread);
2489 fail_cleaner:
2490 kthread_stop(fs_info->cleaner_kthread);
2491
2492 /*
2493 * make sure we're done with the btree inode before we stop our
2494 * kthreads
2495 */
2496 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2497 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2498
2499 fail_block_groups:
2500 btrfs_free_block_groups(fs_info);
2501
2502 fail_tree_roots:
2503 free_root_pointers(fs_info, 1);
2504
2505 fail_sb_buffer:
2506 btrfs_stop_workers(&fs_info->generic_worker);
2507 btrfs_stop_workers(&fs_info->readahead_workers);
2508 btrfs_stop_workers(&fs_info->fixup_workers);
2509 btrfs_stop_workers(&fs_info->delalloc_workers);
2510 btrfs_stop_workers(&fs_info->workers);
2511 btrfs_stop_workers(&fs_info->endio_workers);
2512 btrfs_stop_workers(&fs_info->endio_meta_workers);
2513 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2514 btrfs_stop_workers(&fs_info->endio_write_workers);
2515 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2516 btrfs_stop_workers(&fs_info->submit_workers);
2517 btrfs_stop_workers(&fs_info->delayed_workers);
2518 btrfs_stop_workers(&fs_info->caching_workers);
2519 fail_alloc:
2520 fail_iput:
2521 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2522
2523 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2524 iput(fs_info->btree_inode);
2525 fail_bdi:
2526 bdi_destroy(&fs_info->bdi);
2527 fail_srcu:
2528 cleanup_srcu_struct(&fs_info->subvol_srcu);
2529 fail:
2530 btrfs_close_devices(fs_info->fs_devices);
2531 return err;
2532
2533 recovery_tree_root:
2534 if (!btrfs_test_opt(tree_root, RECOVERY))
2535 goto fail_tree_roots;
2536
2537 free_root_pointers(fs_info, 0);
2538
2539 /* don't use the log in recovery mode, it won't be valid */
2540 btrfs_set_super_log_root(disk_super, 0);
2541
2542 /* we can't trust the free space cache either */
2543 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2544
2545 ret = next_root_backup(fs_info, fs_info->super_copy,
2546 &num_backups_tried, &backup_index);
2547 if (ret == -1)
2548 goto fail_block_groups;
2549 goto retry_root_backup;
2550 }
2551
2552 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2553 {
2554 char b[BDEVNAME_SIZE];
2555
2556 if (uptodate) {
2557 set_buffer_uptodate(bh);
2558 } else {
2559 printk_ratelimited(KERN_WARNING "lost page write due to "
2560 "I/O error on %s\n",
2561 bdevname(bh->b_bdev, b));
2562 /* note, we dont' set_buffer_write_io_error because we have
2563 * our own ways of dealing with the IO errors
2564 */
2565 clear_buffer_uptodate(bh);
2566 }
2567 unlock_buffer(bh);
2568 put_bh(bh);
2569 }
2570
2571 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2572 {
2573 struct buffer_head *bh;
2574 struct buffer_head *latest = NULL;
2575 struct btrfs_super_block *super;
2576 int i;
2577 u64 transid = 0;
2578 u64 bytenr;
2579
2580 /* we would like to check all the supers, but that would make
2581 * a btrfs mount succeed after a mkfs from a different FS.
2582 * So, we need to add a special mount option to scan for
2583 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2584 */
2585 for (i = 0; i < 1; i++) {
2586 bytenr = btrfs_sb_offset(i);
2587 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2588 break;
2589 bh = __bread(bdev, bytenr / 4096, 4096);
2590 if (!bh)
2591 continue;
2592
2593 super = (struct btrfs_super_block *)bh->b_data;
2594 if (btrfs_super_bytenr(super) != bytenr ||
2595 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2596 sizeof(super->magic))) {
2597 brelse(bh);
2598 continue;
2599 }
2600
2601 if (!latest || btrfs_super_generation(super) > transid) {
2602 brelse(latest);
2603 latest = bh;
2604 transid = btrfs_super_generation(super);
2605 } else {
2606 brelse(bh);
2607 }
2608 }
2609 return latest;
2610 }
2611
2612 /*
2613 * this should be called twice, once with wait == 0 and
2614 * once with wait == 1. When wait == 0 is done, all the buffer heads
2615 * we write are pinned.
2616 *
2617 * They are released when wait == 1 is done.
2618 * max_mirrors must be the same for both runs, and it indicates how
2619 * many supers on this one device should be written.
2620 *
2621 * max_mirrors == 0 means to write them all.
2622 */
2623 static int write_dev_supers(struct btrfs_device *device,
2624 struct btrfs_super_block *sb,
2625 int do_barriers, int wait, int max_mirrors)
2626 {
2627 struct buffer_head *bh;
2628 int i;
2629 int ret;
2630 int errors = 0;
2631 u32 crc;
2632 u64 bytenr;
2633
2634 if (max_mirrors == 0)
2635 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2636
2637 for (i = 0; i < max_mirrors; i++) {
2638 bytenr = btrfs_sb_offset(i);
2639 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2640 break;
2641
2642 if (wait) {
2643 bh = __find_get_block(device->bdev, bytenr / 4096,
2644 BTRFS_SUPER_INFO_SIZE);
2645 BUG_ON(!bh);
2646 wait_on_buffer(bh);
2647 if (!buffer_uptodate(bh))
2648 errors++;
2649
2650 /* drop our reference */
2651 brelse(bh);
2652
2653 /* drop the reference from the wait == 0 run */
2654 brelse(bh);
2655 continue;
2656 } else {
2657 btrfs_set_super_bytenr(sb, bytenr);
2658
2659 crc = ~(u32)0;
2660 crc = btrfs_csum_data(NULL, (char *)sb +
2661 BTRFS_CSUM_SIZE, crc,
2662 BTRFS_SUPER_INFO_SIZE -
2663 BTRFS_CSUM_SIZE);
2664 btrfs_csum_final(crc, sb->csum);
2665
2666 /*
2667 * one reference for us, and we leave it for the
2668 * caller
2669 */
2670 bh = __getblk(device->bdev, bytenr / 4096,
2671 BTRFS_SUPER_INFO_SIZE);
2672 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2673
2674 /* one reference for submit_bh */
2675 get_bh(bh);
2676
2677 set_buffer_uptodate(bh);
2678 lock_buffer(bh);
2679 bh->b_end_io = btrfs_end_buffer_write_sync;
2680 }
2681
2682 /*
2683 * we fua the first super. The others we allow
2684 * to go down lazy.
2685 */
2686 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2687 if (ret)
2688 errors++;
2689 }
2690 return errors < i ? 0 : -1;
2691 }
2692
2693 /*
2694 * endio for the write_dev_flush, this will wake anyone waiting
2695 * for the barrier when it is done
2696 */
2697 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2698 {
2699 if (err) {
2700 if (err == -EOPNOTSUPP)
2701 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2702 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2703 }
2704 if (bio->bi_private)
2705 complete(bio->bi_private);
2706 bio_put(bio);
2707 }
2708
2709 /*
2710 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2711 * sent down. With wait == 1, it waits for the previous flush.
2712 *
2713 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2714 * capable
2715 */
2716 static int write_dev_flush(struct btrfs_device *device, int wait)
2717 {
2718 struct bio *bio;
2719 int ret = 0;
2720
2721 if (device->nobarriers)
2722 return 0;
2723
2724 if (wait) {
2725 bio = device->flush_bio;
2726 if (!bio)
2727 return 0;
2728
2729 wait_for_completion(&device->flush_wait);
2730
2731 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2732 printk("btrfs: disabling barriers on dev %s\n",
2733 device->name);
2734 device->nobarriers = 1;
2735 }
2736 if (!bio_flagged(bio, BIO_UPTODATE)) {
2737 ret = -EIO;
2738 }
2739
2740 /* drop the reference from the wait == 0 run */
2741 bio_put(bio);
2742 device->flush_bio = NULL;
2743
2744 return ret;
2745 }
2746
2747 /*
2748 * one reference for us, and we leave it for the
2749 * caller
2750 */
2751 device->flush_bio = NULL;;
2752 bio = bio_alloc(GFP_NOFS, 0);
2753 if (!bio)
2754 return -ENOMEM;
2755
2756 bio->bi_end_io = btrfs_end_empty_barrier;
2757 bio->bi_bdev = device->bdev;
2758 init_completion(&device->flush_wait);
2759 bio->bi_private = &device->flush_wait;
2760 device->flush_bio = bio;
2761
2762 bio_get(bio);
2763 btrfsic_submit_bio(WRITE_FLUSH, bio);
2764
2765 return 0;
2766 }
2767
2768 /*
2769 * send an empty flush down to each device in parallel,
2770 * then wait for them
2771 */
2772 static int barrier_all_devices(struct btrfs_fs_info *info)
2773 {
2774 struct list_head *head;
2775 struct btrfs_device *dev;
2776 int errors = 0;
2777 int ret;
2778
2779 /* send down all the barriers */
2780 head = &info->fs_devices->devices;
2781 list_for_each_entry_rcu(dev, head, dev_list) {
2782 if (!dev->bdev) {
2783 errors++;
2784 continue;
2785 }
2786 if (!dev->in_fs_metadata || !dev->writeable)
2787 continue;
2788
2789 ret = write_dev_flush(dev, 0);
2790 if (ret)
2791 errors++;
2792 }
2793
2794 /* wait for all the barriers */
2795 list_for_each_entry_rcu(dev, head, dev_list) {
2796 if (!dev->bdev) {
2797 errors++;
2798 continue;
2799 }
2800 if (!dev->in_fs_metadata || !dev->writeable)
2801 continue;
2802
2803 ret = write_dev_flush(dev, 1);
2804 if (ret)
2805 errors++;
2806 }
2807 if (errors)
2808 return -EIO;
2809 return 0;
2810 }
2811
2812 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2813 {
2814 struct list_head *head;
2815 struct btrfs_device *dev;
2816 struct btrfs_super_block *sb;
2817 struct btrfs_dev_item *dev_item;
2818 int ret;
2819 int do_barriers;
2820 int max_errors;
2821 int total_errors = 0;
2822 u64 flags;
2823
2824 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2825 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2826 backup_super_roots(root->fs_info);
2827
2828 sb = root->fs_info->super_for_commit;
2829 dev_item = &sb->dev_item;
2830
2831 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2832 head = &root->fs_info->fs_devices->devices;
2833
2834 if (do_barriers)
2835 barrier_all_devices(root->fs_info);
2836
2837 list_for_each_entry_rcu(dev, head, dev_list) {
2838 if (!dev->bdev) {
2839 total_errors++;
2840 continue;
2841 }
2842 if (!dev->in_fs_metadata || !dev->writeable)
2843 continue;
2844
2845 btrfs_set_stack_device_generation(dev_item, 0);
2846 btrfs_set_stack_device_type(dev_item, dev->type);
2847 btrfs_set_stack_device_id(dev_item, dev->devid);
2848 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2849 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2850 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2851 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2852 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2853 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2854 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2855
2856 flags = btrfs_super_flags(sb);
2857 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2858
2859 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2860 if (ret)
2861 total_errors++;
2862 }
2863 if (total_errors > max_errors) {
2864 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2865 total_errors);
2866
2867 /* This shouldn't happen. FUA is masked off if unsupported */
2868 BUG();
2869 }
2870
2871 total_errors = 0;
2872 list_for_each_entry_rcu(dev, head, dev_list) {
2873 if (!dev->bdev)
2874 continue;
2875 if (!dev->in_fs_metadata || !dev->writeable)
2876 continue;
2877
2878 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2879 if (ret)
2880 total_errors++;
2881 }
2882 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2883 if (total_errors > max_errors) {
2884 btrfs_error(root->fs_info, -EIO,
2885 "%d errors while writing supers", total_errors);
2886 return -EIO;
2887 }
2888 return 0;
2889 }
2890
2891 int write_ctree_super(struct btrfs_trans_handle *trans,
2892 struct btrfs_root *root, int max_mirrors)
2893 {
2894 int ret;
2895
2896 ret = write_all_supers(root, max_mirrors);
2897 return ret;
2898 }
2899
2900 /* Kill all outstanding I/O */
2901 void btrfs_abort_devices(struct btrfs_root *root)
2902 {
2903 struct list_head *head;
2904 struct btrfs_device *dev;
2905 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2906 head = &root->fs_info->fs_devices->devices;
2907 list_for_each_entry_rcu(dev, head, dev_list) {
2908 blk_abort_queue(dev->bdev->bd_disk->queue);
2909 }
2910 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2911 }
2912
2913 void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2914 {
2915 spin_lock(&fs_info->fs_roots_radix_lock);
2916 radix_tree_delete(&fs_info->fs_roots_radix,
2917 (unsigned long)root->root_key.objectid);
2918 spin_unlock(&fs_info->fs_roots_radix_lock);
2919
2920 if (btrfs_root_refs(&root->root_item) == 0)
2921 synchronize_srcu(&fs_info->subvol_srcu);
2922
2923 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2924 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2925 free_fs_root(root);
2926 }
2927
2928 static void free_fs_root(struct btrfs_root *root)
2929 {
2930 iput(root->cache_inode);
2931 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2932 if (root->anon_dev)
2933 free_anon_bdev(root->anon_dev);
2934 free_extent_buffer(root->node);
2935 free_extent_buffer(root->commit_root);
2936 kfree(root->free_ino_ctl);
2937 kfree(root->free_ino_pinned);
2938 kfree(root->name);
2939 kfree(root);
2940 }
2941
2942 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2943 {
2944 int ret;
2945 struct btrfs_root *gang[8];
2946 int i;
2947
2948 while (!list_empty(&fs_info->dead_roots)) {
2949 gang[0] = list_entry(fs_info->dead_roots.next,
2950 struct btrfs_root, root_list);
2951 list_del(&gang[0]->root_list);
2952
2953 if (gang[0]->in_radix) {
2954 btrfs_free_fs_root(fs_info, gang[0]);
2955 } else {
2956 free_extent_buffer(gang[0]->node);
2957 free_extent_buffer(gang[0]->commit_root);
2958 kfree(gang[0]);
2959 }
2960 }
2961
2962 while (1) {
2963 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2964 (void **)gang, 0,
2965 ARRAY_SIZE(gang));
2966 if (!ret)
2967 break;
2968 for (i = 0; i < ret; i++)
2969 btrfs_free_fs_root(fs_info, gang[i]);
2970 }
2971 }
2972
2973 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2974 {
2975 u64 root_objectid = 0;
2976 struct btrfs_root *gang[8];
2977 int i;
2978 int ret;
2979
2980 while (1) {
2981 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2982 (void **)gang, root_objectid,
2983 ARRAY_SIZE(gang));
2984 if (!ret)
2985 break;
2986
2987 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2988 for (i = 0; i < ret; i++) {
2989 int err;
2990
2991 root_objectid = gang[i]->root_key.objectid;
2992 err = btrfs_orphan_cleanup(gang[i]);
2993 if (err)
2994 return err;
2995 }
2996 root_objectid++;
2997 }
2998 return 0;
2999 }
3000
3001 int btrfs_commit_super(struct btrfs_root *root)
3002 {
3003 struct btrfs_trans_handle *trans;
3004 int ret;
3005
3006 mutex_lock(&root->fs_info->cleaner_mutex);
3007 btrfs_run_delayed_iputs(root);
3008 btrfs_clean_old_snapshots(root);
3009 mutex_unlock(&root->fs_info->cleaner_mutex);
3010
3011 /* wait until ongoing cleanup work done */
3012 down_write(&root->fs_info->cleanup_work_sem);
3013 up_write(&root->fs_info->cleanup_work_sem);
3014
3015 trans = btrfs_join_transaction(root);
3016 if (IS_ERR(trans))
3017 return PTR_ERR(trans);
3018 ret = btrfs_commit_transaction(trans, root);
3019 if (ret)
3020 return ret;
3021 /* run commit again to drop the original snapshot */
3022 trans = btrfs_join_transaction(root);
3023 if (IS_ERR(trans))
3024 return PTR_ERR(trans);
3025 ret = btrfs_commit_transaction(trans, root);
3026 if (ret)
3027 return ret;
3028 ret = btrfs_write_and_wait_transaction(NULL, root);
3029 if (ret) {
3030 btrfs_error(root->fs_info, ret,
3031 "Failed to sync btree inode to disk.");
3032 return ret;
3033 }
3034
3035 ret = write_ctree_super(NULL, root, 0);
3036 return ret;
3037 }
3038
3039 int close_ctree(struct btrfs_root *root)
3040 {
3041 struct btrfs_fs_info *fs_info = root->fs_info;
3042 int ret;
3043
3044 fs_info->closing = 1;
3045 smp_mb();
3046
3047 /* pause restriper - we want to resume on mount */
3048 btrfs_pause_balance(root->fs_info);
3049
3050 btrfs_scrub_cancel(root);
3051
3052 /* wait for any defraggers to finish */
3053 wait_event(fs_info->transaction_wait,
3054 (atomic_read(&fs_info->defrag_running) == 0));
3055
3056 /* clear out the rbtree of defraggable inodes */
3057 btrfs_run_defrag_inodes(fs_info);
3058
3059 /*
3060 * Here come 2 situations when btrfs is broken to flip readonly:
3061 *
3062 * 1. when btrfs flips readonly somewhere else before
3063 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
3064 * and btrfs will skip to write sb directly to keep
3065 * ERROR state on disk.
3066 *
3067 * 2. when btrfs flips readonly just in btrfs_commit_super,
3068 * and in such case, btrfs cannot write sb via btrfs_commit_super,
3069 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3070 * btrfs will cleanup all FS resources first and write sb then.
3071 */
3072 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3073 ret = btrfs_commit_super(root);
3074 if (ret)
3075 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3076 }
3077
3078 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3079 ret = btrfs_error_commit_super(root);
3080 if (ret)
3081 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3082 }
3083
3084 btrfs_put_block_group_cache(fs_info);
3085
3086 kthread_stop(fs_info->transaction_kthread);
3087 kthread_stop(fs_info->cleaner_kthread);
3088
3089 fs_info->closing = 2;
3090 smp_mb();
3091
3092 if (fs_info->delalloc_bytes) {
3093 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3094 (unsigned long long)fs_info->delalloc_bytes);
3095 }
3096 if (fs_info->total_ref_cache_size) {
3097 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3098 (unsigned long long)fs_info->total_ref_cache_size);
3099 }
3100
3101 free_extent_buffer(fs_info->extent_root->node);
3102 free_extent_buffer(fs_info->extent_root->commit_root);
3103 free_extent_buffer(fs_info->tree_root->node);
3104 free_extent_buffer(fs_info->tree_root->commit_root);
3105 free_extent_buffer(fs_info->chunk_root->node);
3106 free_extent_buffer(fs_info->chunk_root->commit_root);
3107 free_extent_buffer(fs_info->dev_root->node);
3108 free_extent_buffer(fs_info->dev_root->commit_root);
3109 free_extent_buffer(fs_info->csum_root->node);
3110 free_extent_buffer(fs_info->csum_root->commit_root);
3111
3112 btrfs_free_block_groups(fs_info);
3113
3114 del_fs_roots(fs_info);
3115
3116 iput(fs_info->btree_inode);
3117
3118 btrfs_stop_workers(&fs_info->generic_worker);
3119 btrfs_stop_workers(&fs_info->fixup_workers);
3120 btrfs_stop_workers(&fs_info->delalloc_workers);
3121 btrfs_stop_workers(&fs_info->workers);
3122 btrfs_stop_workers(&fs_info->endio_workers);
3123 btrfs_stop_workers(&fs_info->endio_meta_workers);
3124 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3125 btrfs_stop_workers(&fs_info->endio_write_workers);
3126 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3127 btrfs_stop_workers(&fs_info->submit_workers);
3128 btrfs_stop_workers(&fs_info->delayed_workers);
3129 btrfs_stop_workers(&fs_info->caching_workers);
3130 btrfs_stop_workers(&fs_info->readahead_workers);
3131
3132 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3133 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3134 btrfsic_unmount(root, fs_info->fs_devices);
3135 #endif
3136
3137 btrfs_close_devices(fs_info->fs_devices);
3138 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3139
3140 bdi_destroy(&fs_info->bdi);
3141 cleanup_srcu_struct(&fs_info->subvol_srcu);
3142
3143 return 0;
3144 }
3145
3146 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
3147 {
3148 int ret;
3149 struct inode *btree_inode = buf->pages[0]->mapping->host;
3150
3151 ret = extent_buffer_uptodate(buf);
3152 if (!ret)
3153 return ret;
3154
3155 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3156 parent_transid);
3157 return !ret;
3158 }
3159
3160 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3161 {
3162 return set_extent_buffer_uptodate(buf);
3163 }
3164
3165 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3166 {
3167 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3168 u64 transid = btrfs_header_generation(buf);
3169 int was_dirty;
3170
3171 btrfs_assert_tree_locked(buf);
3172 if (transid != root->fs_info->generation) {
3173 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3174 "found %llu running %llu\n",
3175 (unsigned long long)buf->start,
3176 (unsigned long long)transid,
3177 (unsigned long long)root->fs_info->generation);
3178 WARN_ON(1);
3179 }
3180 was_dirty = set_extent_buffer_dirty(buf);
3181 if (!was_dirty) {
3182 spin_lock(&root->fs_info->delalloc_lock);
3183 root->fs_info->dirty_metadata_bytes += buf->len;
3184 spin_unlock(&root->fs_info->delalloc_lock);
3185 }
3186 }
3187
3188 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3189 {
3190 /*
3191 * looks as though older kernels can get into trouble with
3192 * this code, they end up stuck in balance_dirty_pages forever
3193 */
3194 u64 num_dirty;
3195 unsigned long thresh = 32 * 1024 * 1024;
3196
3197 if (current->flags & PF_MEMALLOC)
3198 return;
3199
3200 btrfs_balance_delayed_items(root);
3201
3202 num_dirty = root->fs_info->dirty_metadata_bytes;
3203
3204 if (num_dirty > thresh) {
3205 balance_dirty_pages_ratelimited_nr(
3206 root->fs_info->btree_inode->i_mapping, 1);
3207 }
3208 return;
3209 }
3210
3211 void __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 num_dirty = root->fs_info->dirty_metadata_bytes;
3224
3225 if (num_dirty > thresh) {
3226 balance_dirty_pages_ratelimited_nr(
3227 root->fs_info->btree_inode->i_mapping, 1);
3228 }
3229 return;
3230 }
3231
3232 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3233 {
3234 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3235 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3236 }
3237
3238 static int btree_lock_page_hook(struct page *page, void *data,
3239 void (*flush_fn)(void *))
3240 {
3241 struct inode *inode = page->mapping->host;
3242 struct btrfs_root *root = BTRFS_I(inode)->root;
3243 struct extent_buffer *eb;
3244
3245 /*
3246 * We culled this eb but the page is still hanging out on the mapping,
3247 * carry on.
3248 */
3249 if (!PagePrivate(page))
3250 goto out;
3251
3252 eb = (struct extent_buffer *)page->private;
3253 if (!eb) {
3254 WARN_ON(1);
3255 goto out;
3256 }
3257 if (page != eb->pages[0])
3258 goto out;
3259
3260 if (!btrfs_try_tree_write_lock(eb)) {
3261 flush_fn(data);
3262 btrfs_tree_lock(eb);
3263 }
3264 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3265
3266 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3267 spin_lock(&root->fs_info->delalloc_lock);
3268 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3269 root->fs_info->dirty_metadata_bytes -= eb->len;
3270 else
3271 WARN_ON(1);
3272 spin_unlock(&root->fs_info->delalloc_lock);
3273 }
3274
3275 btrfs_tree_unlock(eb);
3276 out:
3277 if (!trylock_page(page)) {
3278 flush_fn(data);
3279 lock_page(page);
3280 }
3281 return 0;
3282 }
3283
3284 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3285 int read_only)
3286 {
3287 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3288 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3289 return -EINVAL;
3290 }
3291
3292 if (read_only)
3293 return 0;
3294
3295 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3296 printk(KERN_WARNING "warning: mount fs with errors, "
3297 "running btrfsck is recommended\n");
3298 }
3299
3300 return 0;
3301 }
3302
3303 int btrfs_error_commit_super(struct btrfs_root *root)
3304 {
3305 int ret;
3306
3307 mutex_lock(&root->fs_info->cleaner_mutex);
3308 btrfs_run_delayed_iputs(root);
3309 mutex_unlock(&root->fs_info->cleaner_mutex);
3310
3311 down_write(&root->fs_info->cleanup_work_sem);
3312 up_write(&root->fs_info->cleanup_work_sem);
3313
3314 /* cleanup FS via transaction */
3315 btrfs_cleanup_transaction(root);
3316
3317 ret = write_ctree_super(NULL, root, 0);
3318
3319 return ret;
3320 }
3321
3322 static void btrfs_destroy_ordered_operations(struct btrfs_root *root)
3323 {
3324 struct btrfs_inode *btrfs_inode;
3325 struct list_head splice;
3326
3327 INIT_LIST_HEAD(&splice);
3328
3329 mutex_lock(&root->fs_info->ordered_operations_mutex);
3330 spin_lock(&root->fs_info->ordered_extent_lock);
3331
3332 list_splice_init(&root->fs_info->ordered_operations, &splice);
3333 while (!list_empty(&splice)) {
3334 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3335 ordered_operations);
3336
3337 list_del_init(&btrfs_inode->ordered_operations);
3338
3339 btrfs_invalidate_inodes(btrfs_inode->root);
3340 }
3341
3342 spin_unlock(&root->fs_info->ordered_extent_lock);
3343 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3344 }
3345
3346 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3347 {
3348 struct list_head splice;
3349 struct btrfs_ordered_extent *ordered;
3350 struct inode *inode;
3351
3352 INIT_LIST_HEAD(&splice);
3353
3354 spin_lock(&root->fs_info->ordered_extent_lock);
3355
3356 list_splice_init(&root->fs_info->ordered_extents, &splice);
3357 while (!list_empty(&splice)) {
3358 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3359 root_extent_list);
3360
3361 list_del_init(&ordered->root_extent_list);
3362 atomic_inc(&ordered->refs);
3363
3364 /* the inode may be getting freed (in sys_unlink path). */
3365 inode = igrab(ordered->inode);
3366
3367 spin_unlock(&root->fs_info->ordered_extent_lock);
3368 if (inode)
3369 iput(inode);
3370
3371 atomic_set(&ordered->refs, 1);
3372 btrfs_put_ordered_extent(ordered);
3373
3374 spin_lock(&root->fs_info->ordered_extent_lock);
3375 }
3376
3377 spin_unlock(&root->fs_info->ordered_extent_lock);
3378 }
3379
3380 int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3381 struct btrfs_root *root)
3382 {
3383 struct rb_node *node;
3384 struct btrfs_delayed_ref_root *delayed_refs;
3385 struct btrfs_delayed_ref_node *ref;
3386 int ret = 0;
3387
3388 delayed_refs = &trans->delayed_refs;
3389
3390 again:
3391 spin_lock(&delayed_refs->lock);
3392 if (delayed_refs->num_entries == 0) {
3393 spin_unlock(&delayed_refs->lock);
3394 printk(KERN_INFO "delayed_refs has NO entry\n");
3395 return ret;
3396 }
3397
3398 node = rb_first(&delayed_refs->root);
3399 while (node) {
3400 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3401 node = rb_next(node);
3402
3403 ref->in_tree = 0;
3404 rb_erase(&ref->rb_node, &delayed_refs->root);
3405 delayed_refs->num_entries--;
3406
3407 atomic_set(&ref->refs, 1);
3408 if (btrfs_delayed_ref_is_head(ref)) {
3409 struct btrfs_delayed_ref_head *head;
3410
3411 head = btrfs_delayed_node_to_head(ref);
3412 spin_unlock(&delayed_refs->lock);
3413 mutex_lock(&head->mutex);
3414 kfree(head->extent_op);
3415 delayed_refs->num_heads--;
3416 if (list_empty(&head->cluster))
3417 delayed_refs->num_heads_ready--;
3418 list_del_init(&head->cluster);
3419 mutex_unlock(&head->mutex);
3420 btrfs_put_delayed_ref(ref);
3421 goto again;
3422 }
3423 spin_unlock(&delayed_refs->lock);
3424 btrfs_put_delayed_ref(ref);
3425
3426 cond_resched();
3427 spin_lock(&delayed_refs->lock);
3428 }
3429
3430 spin_unlock(&delayed_refs->lock);
3431
3432 return ret;
3433 }
3434
3435 static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3436 {
3437 struct btrfs_pending_snapshot *snapshot;
3438 struct list_head splice;
3439
3440 INIT_LIST_HEAD(&splice);
3441
3442 list_splice_init(&t->pending_snapshots, &splice);
3443
3444 while (!list_empty(&splice)) {
3445 snapshot = list_entry(splice.next,
3446 struct btrfs_pending_snapshot,
3447 list);
3448
3449 list_del_init(&snapshot->list);
3450
3451 kfree(snapshot);
3452 }
3453 }
3454
3455 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3456 {
3457 struct btrfs_inode *btrfs_inode;
3458 struct list_head splice;
3459
3460 INIT_LIST_HEAD(&splice);
3461
3462 spin_lock(&root->fs_info->delalloc_lock);
3463 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3464
3465 while (!list_empty(&splice)) {
3466 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3467 delalloc_inodes);
3468
3469 list_del_init(&btrfs_inode->delalloc_inodes);
3470
3471 btrfs_invalidate_inodes(btrfs_inode->root);
3472 }
3473
3474 spin_unlock(&root->fs_info->delalloc_lock);
3475 }
3476
3477 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3478 struct extent_io_tree *dirty_pages,
3479 int mark)
3480 {
3481 int ret;
3482 struct page *page;
3483 struct inode *btree_inode = root->fs_info->btree_inode;
3484 struct extent_buffer *eb;
3485 u64 start = 0;
3486 u64 end;
3487 u64 offset;
3488 unsigned long index;
3489
3490 while (1) {
3491 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3492 mark);
3493 if (ret)
3494 break;
3495
3496 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3497 while (start <= end) {
3498 index = start >> PAGE_CACHE_SHIFT;
3499 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3500 page = find_get_page(btree_inode->i_mapping, index);
3501 if (!page)
3502 continue;
3503 offset = page_offset(page);
3504
3505 spin_lock(&dirty_pages->buffer_lock);
3506 eb = radix_tree_lookup(
3507 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3508 offset >> PAGE_CACHE_SHIFT);
3509 spin_unlock(&dirty_pages->buffer_lock);
3510 if (eb) {
3511 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3512 &eb->bflags);
3513 atomic_set(&eb->refs, 1);
3514 }
3515 if (PageWriteback(page))
3516 end_page_writeback(page);
3517
3518 lock_page(page);
3519 if (PageDirty(page)) {
3520 clear_page_dirty_for_io(page);
3521 spin_lock_irq(&page->mapping->tree_lock);
3522 radix_tree_tag_clear(&page->mapping->page_tree,
3523 page_index(page),
3524 PAGECACHE_TAG_DIRTY);
3525 spin_unlock_irq(&page->mapping->tree_lock);
3526 }
3527
3528 page->mapping->a_ops->invalidatepage(page, 0);
3529 unlock_page(page);
3530 }
3531 }
3532
3533 return ret;
3534 }
3535
3536 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3537 struct extent_io_tree *pinned_extents)
3538 {
3539 struct extent_io_tree *unpin;
3540 u64 start;
3541 u64 end;
3542 int ret;
3543
3544 unpin = pinned_extents;
3545 while (1) {
3546 ret = find_first_extent_bit(unpin, 0, &start, &end,
3547 EXTENT_DIRTY);
3548 if (ret)
3549 break;
3550
3551 /* opt_discard */
3552 if (btrfs_test_opt(root, DISCARD))
3553 ret = btrfs_error_discard_extent(root, start,
3554 end + 1 - start,
3555 NULL);
3556
3557 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3558 btrfs_error_unpin_extent_range(root, start, end);
3559 cond_resched();
3560 }
3561
3562 return 0;
3563 }
3564
3565 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3566 struct btrfs_root *root)
3567 {
3568 btrfs_destroy_delayed_refs(cur_trans, root);
3569 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3570 cur_trans->dirty_pages.dirty_bytes);
3571
3572 /* FIXME: cleanup wait for commit */
3573 cur_trans->in_commit = 1;
3574 cur_trans->blocked = 1;
3575 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3576 wake_up(&root->fs_info->transaction_blocked_wait);
3577
3578 cur_trans->blocked = 0;
3579 if (waitqueue_active(&root->fs_info->transaction_wait))
3580 wake_up(&root->fs_info->transaction_wait);
3581
3582 cur_trans->commit_done = 1;
3583 if (waitqueue_active(&cur_trans->commit_wait))
3584 wake_up(&cur_trans->commit_wait);
3585
3586 btrfs_destroy_pending_snapshots(cur_trans);
3587
3588 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3589 EXTENT_DIRTY);
3590
3591 /*
3592 memset(cur_trans, 0, sizeof(*cur_trans));
3593 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3594 */
3595 }
3596
3597 int btrfs_cleanup_transaction(struct btrfs_root *root)
3598 {
3599 struct btrfs_transaction *t;
3600 LIST_HEAD(list);
3601
3602 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3603
3604 spin_lock(&root->fs_info->trans_lock);
3605 list_splice_init(&root->fs_info->trans_list, &list);
3606 root->fs_info->trans_no_join = 1;
3607 spin_unlock(&root->fs_info->trans_lock);
3608
3609 while (!list_empty(&list)) {
3610 t = list_entry(list.next, struct btrfs_transaction, list);
3611 if (!t)
3612 break;
3613
3614 btrfs_destroy_ordered_operations(root);
3615
3616 btrfs_destroy_ordered_extents(root);
3617
3618 btrfs_destroy_delayed_refs(t, root);
3619
3620 btrfs_block_rsv_release(root,
3621 &root->fs_info->trans_block_rsv,
3622 t->dirty_pages.dirty_bytes);
3623
3624 /* FIXME: cleanup wait for commit */
3625 t->in_commit = 1;
3626 t->blocked = 1;
3627 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3628 wake_up(&root->fs_info->transaction_blocked_wait);
3629
3630 t->blocked = 0;
3631 if (waitqueue_active(&root->fs_info->transaction_wait))
3632 wake_up(&root->fs_info->transaction_wait);
3633
3634 t->commit_done = 1;
3635 if (waitqueue_active(&t->commit_wait))
3636 wake_up(&t->commit_wait);
3637
3638 btrfs_destroy_pending_snapshots(t);
3639
3640 btrfs_destroy_delalloc_inodes(root);
3641
3642 spin_lock(&root->fs_info->trans_lock);
3643 root->fs_info->running_transaction = NULL;
3644 spin_unlock(&root->fs_info->trans_lock);
3645
3646 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3647 EXTENT_DIRTY);
3648
3649 btrfs_destroy_pinned_extent(root,
3650 root->fs_info->pinned_extents);
3651
3652 atomic_set(&t->use_count, 0);
3653 list_del_init(&t->list);
3654 memset(t, 0, sizeof(*t));
3655 kmem_cache_free(btrfs_transaction_cachep, t);
3656 }
3657
3658 spin_lock(&root->fs_info->trans_lock);
3659 root->fs_info->trans_no_join = 0;
3660 spin_unlock(&root->fs_info->trans_lock);
3661 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3662
3663 return 0;
3664 }
3665
3666 static int btree_writepage_io_failed_hook(struct bio *bio, struct page *page,
3667 u64 start, u64 end,
3668 struct extent_state *state)
3669 {
3670 struct super_block *sb = page->mapping->host->i_sb;
3671 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
3672 btrfs_error(fs_info, -EIO,
3673 "Error occured while writing out btree at %llu", start);
3674 return -EIO;
3675 }
3676
3677 static struct extent_io_ops btree_extent_io_ops = {
3678 .write_cache_pages_lock_hook = btree_lock_page_hook,
3679 .readpage_end_io_hook = btree_readpage_end_io_hook,
3680 .readpage_io_failed_hook = btree_io_failed_hook,
3681 .submit_bio_hook = btree_submit_bio_hook,
3682 /* note we're sharing with inode.c for the merge bio hook */
3683 .merge_bio_hook = btrfs_merge_bio_hook,
3684 .writepage_io_failed_hook = btree_writepage_io_failed_hook,
3685 };
Linux kernel - Deliverables
This page took 0.104145 seconds and 6 git commands to generate.