projects / deliverable / linux.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: fix enospc error caused by wrong checks of the chunk
[deliverable/linux.git] / fs / btrfs / disk-io.c
... / ...
CommitLineData
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/fs.h>
20#include <linux/blkdev.h>
21#include <linux/scatterlist.h>
22#include <linux/swap.h>
23#include <linux/radix-tree.h>
24#include <linux/writeback.h>
25#include <linux/buffer_head.h>
26#include <linux/workqueue.h>
27#include <linux/kthread.h>
28#include <linux/freezer.h>
29#include <linux/crc32c.h>
30#include <linux/slab.h>
31#include <linux/migrate.h>
32#include <linux/ratelimit.h>
33#include <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
48static struct extent_io_ops btree_extent_io_ops;
49static void end_workqueue_fn(struct btrfs_work *work);
50static void free_fs_root(struct btrfs_root *root);
51static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
52 int read_only);
53static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
54static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
55static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_root *root);
57static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
58static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59static int btrfs_destroy_marked_extents(struct btrfs_root *root,
60 struct extent_io_tree *dirty_pages,
61 int mark);
62static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
63 struct extent_io_tree *pinned_extents);
64static int btrfs_cleanup_transaction(struct btrfs_root *root);
65
66/*
67 * end_io_wq structs are used to do processing in task context when an IO is
68 * complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
70 */
71struct end_io_wq {
72 struct bio *bio;
73 bio_end_io_t *end_io;
74 void *private;
75 struct btrfs_fs_info *info;
76 int error;
77 int metadata;
78 struct list_head list;
79 struct btrfs_work work;
80};
81
82/*
83 * async submit bios are used to offload expensive checksumming
84 * onto the worker threads. They checksum file and metadata bios
85 * just before they are sent down the IO stack.
86 */
87struct async_submit_bio {
88 struct inode *inode;
89 struct bio *bio;
90 struct list_head list;
91 extent_submit_bio_hook_t *submit_bio_start;
92 extent_submit_bio_hook_t *submit_bio_done;
93 int rw;
94 int mirror_num;
95 unsigned long bio_flags;
96 /*
97 * bio_offset is optional, can be used if the pages in the bio
98 * can't tell us where in the file the bio should go
99 */
100 u64 bio_offset;
101 struct btrfs_work work;
102};
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
132static 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
151void __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
165void 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 */
187static 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);
239out:
240 return em;
241}
242
243u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
244{
245 return crc32c(seed, data, len);
246}
247
248void 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 */
257static 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 */
325static 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, GFP_NOFS);
336 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
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(io_tree, eb, &cached_state);
348out:
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 */
358static 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 ret;
364 int num_copies = 0;
365 int mirror_num = 0;
366
367 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
368 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
369 while (1) {
370 ret = read_extent_buffer_pages(io_tree, eb, start,
371 WAIT_COMPLETE,
372 btree_get_extent, mirror_num);
373 if (!ret &&
374 !verify_parent_transid(io_tree, eb, parent_transid))
375 return ret;
376
377 /*
378 * This buffer's crc is fine, but its contents are corrupted, so
379 * there is no reason to read the other copies, they won't be
380 * any less wrong.
381 */
382 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
383 return ret;
384
385 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
386 eb->start, eb->len);
387 if (num_copies == 1)
388 return ret;
389
390 mirror_num++;
391 if (mirror_num > num_copies)
392 return ret;
393 }
394 return -EIO;
395}
396
397/*
398 * checksum a dirty tree block before IO. This has extra checks to make sure
399 * we only fill in the checksum field in the first page of a multi-page block
400 */
401
402static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
403{
404 struct extent_io_tree *tree;
405 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
406 u64 found_start;
407 unsigned long len;
408 struct extent_buffer *eb;
409 int ret;
410
411 tree = &BTRFS_I(page->mapping->host)->io_tree;
412
413 if (page->private == EXTENT_PAGE_PRIVATE) {
414 WARN_ON(1);
415 goto out;
416 }
417 if (!page->private) {
418 WARN_ON(1);
419 goto out;
420 }
421 len = page->private >> 2;
422 WARN_ON(len == 0);
423
424 eb = alloc_extent_buffer(tree, start, len, page);
425 if (eb == NULL) {
426 WARN_ON(1);
427 goto out;
428 }
429 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
430 btrfs_header_generation(eb));
431 BUG_ON(ret);
432 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
433
434 found_start = btrfs_header_bytenr(eb);
435 if (found_start != start) {
436 WARN_ON(1);
437 goto err;
438 }
439 if (eb->first_page != page) {
440 WARN_ON(1);
441 goto err;
442 }
443 if (!PageUptodate(page)) {
444 WARN_ON(1);
445 goto err;
446 }
447 csum_tree_block(root, eb, 0);
448err:
449 free_extent_buffer(eb);
450out:
451 return 0;
452}
453
454static int check_tree_block_fsid(struct btrfs_root *root,
455 struct extent_buffer *eb)
456{
457 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
458 u8 fsid[BTRFS_UUID_SIZE];
459 int ret = 1;
460
461 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
462 BTRFS_FSID_SIZE);
463 while (fs_devices) {
464 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
465 ret = 0;
466 break;
467 }
468 fs_devices = fs_devices->seed;
469 }
470 return ret;
471}
472
473#define CORRUPT(reason, eb, root, slot) \
474 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
475 "root=%llu, slot=%d\n", reason, \
476 (unsigned long long)btrfs_header_bytenr(eb), \
477 (unsigned long long)root->objectid, slot)
478
479static noinline int check_leaf(struct btrfs_root *root,
480 struct extent_buffer *leaf)
481{
482 struct btrfs_key key;
483 struct btrfs_key leaf_key;
484 u32 nritems = btrfs_header_nritems(leaf);
485 int slot;
486
487 if (nritems == 0)
488 return 0;
489
490 /* Check the 0 item */
491 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
492 BTRFS_LEAF_DATA_SIZE(root)) {
493 CORRUPT("invalid item offset size pair", leaf, root, 0);
494 return -EIO;
495 }
496
497 /*
498 * Check to make sure each items keys are in the correct order and their
499 * offsets make sense. We only have to loop through nritems-1 because
500 * we check the current slot against the next slot, which verifies the
501 * next slot's offset+size makes sense and that the current's slot
502 * offset is correct.
503 */
504 for (slot = 0; slot < nritems - 1; slot++) {
505 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
506 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
507
508 /* Make sure the keys are in the right order */
509 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
510 CORRUPT("bad key order", leaf, root, slot);
511 return -EIO;
512 }
513
514 /*
515 * Make sure the offset and ends are right, remember that the
516 * item data starts at the end of the leaf and grows towards the
517 * front.
518 */
519 if (btrfs_item_offset_nr(leaf, slot) !=
520 btrfs_item_end_nr(leaf, slot + 1)) {
521 CORRUPT("slot offset bad", leaf, root, slot);
522 return -EIO;
523 }
524
525 /*
526 * Check to make sure that we don't point outside of the leaf,
527 * just incase all the items are consistent to eachother, but
528 * all point outside of the leaf.
529 */
530 if (btrfs_item_end_nr(leaf, slot) >
531 BTRFS_LEAF_DATA_SIZE(root)) {
532 CORRUPT("slot end outside of leaf", leaf, root, slot);
533 return -EIO;
534 }
535 }
536
537 return 0;
538}
539
540static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
541 struct extent_state *state)
542{
543 struct extent_io_tree *tree;
544 u64 found_start;
545 int found_level;
546 unsigned long len;
547 struct extent_buffer *eb;
548 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
549 int ret = 0;
550
551 tree = &BTRFS_I(page->mapping->host)->io_tree;
552 if (page->private == EXTENT_PAGE_PRIVATE)
553 goto out;
554 if (!page->private)
555 goto out;
556
557 len = page->private >> 2;
558 WARN_ON(len == 0);
559
560 eb = alloc_extent_buffer(tree, start, len, page);
561 if (eb == NULL) {
562 ret = -EIO;
563 goto out;
564 }
565
566 found_start = btrfs_header_bytenr(eb);
567 if (found_start != start) {
568 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
569 "%llu %llu\n",
570 (unsigned long long)found_start,
571 (unsigned long long)eb->start);
572 ret = -EIO;
573 goto err;
574 }
575 if (eb->first_page != page) {
576 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
577 eb->first_page->index, page->index);
578 WARN_ON(1);
579 ret = -EIO;
580 goto err;
581 }
582 if (check_tree_block_fsid(root, eb)) {
583 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
584 (unsigned long long)eb->start);
585 ret = -EIO;
586 goto err;
587 }
588 found_level = btrfs_header_level(eb);
589
590 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
591 eb, found_level);
592
593 ret = csum_tree_block(root, eb, 1);
594 if (ret) {
595 ret = -EIO;
596 goto err;
597 }
598
599 /*
600 * If this is a leaf block and it is corrupt, set the corrupt bit so
601 * that we don't try and read the other copies of this block, just
602 * return -EIO.
603 */
604 if (found_level == 0 && check_leaf(root, eb)) {
605 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
606 ret = -EIO;
607 }
608
609 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
610 end = eb->start + end - 1;
611err:
612 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
613 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
614 btree_readahead_hook(root, eb, eb->start, ret);
615 }
616
617 free_extent_buffer(eb);
618out:
619 return ret;
620}
621
622static int btree_io_failed_hook(struct bio *failed_bio,
623 struct page *page, u64 start, u64 end,
624 int mirror_num, struct extent_state *state)
625{
626 struct extent_io_tree *tree;
627 unsigned long len;
628 struct extent_buffer *eb;
629 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
630
631 tree = &BTRFS_I(page->mapping->host)->io_tree;
632 if (page->private == EXTENT_PAGE_PRIVATE)
633 goto out;
634 if (!page->private)
635 goto out;
636
637 len = page->private >> 2;
638 WARN_ON(len == 0);
639
640 eb = alloc_extent_buffer(tree, start, len, page);
641 if (eb == NULL)
642 goto out;
643
644 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
645 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
646 btree_readahead_hook(root, eb, eb->start, -EIO);
647 }
648 free_extent_buffer(eb);
649
650out:
651 return -EIO; /* we fixed nothing */
652}
653
654static void end_workqueue_bio(struct bio *bio, int err)
655{
656 struct end_io_wq *end_io_wq = bio->bi_private;
657 struct btrfs_fs_info *fs_info;
658
659 fs_info = end_io_wq->info;
660 end_io_wq->error = err;
661 end_io_wq->work.func = end_workqueue_fn;
662 end_io_wq->work.flags = 0;
663
664 if (bio->bi_rw & REQ_WRITE) {
665 if (end_io_wq->metadata == 1)
666 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
667 &end_io_wq->work);
668 else if (end_io_wq->metadata == 2)
669 btrfs_queue_worker(&fs_info->endio_freespace_worker,
670 &end_io_wq->work);
671 else
672 btrfs_queue_worker(&fs_info->endio_write_workers,
673 &end_io_wq->work);
674 } else {
675 if (end_io_wq->metadata)
676 btrfs_queue_worker(&fs_info->endio_meta_workers,
677 &end_io_wq->work);
678 else
679 btrfs_queue_worker(&fs_info->endio_workers,
680 &end_io_wq->work);
681 }
682}
683
684/*
685 * For the metadata arg you want
686 *
687 * 0 - if data
688 * 1 - if normal metadta
689 * 2 - if writing to the free space cache area
690 */
691int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
692 int metadata)
693{
694 struct end_io_wq *end_io_wq;
695 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
696 if (!end_io_wq)
697 return -ENOMEM;
698
699 end_io_wq->private = bio->bi_private;
700 end_io_wq->end_io = bio->bi_end_io;
701 end_io_wq->info = info;
702 end_io_wq->error = 0;
703 end_io_wq->bio = bio;
704 end_io_wq->metadata = metadata;
705
706 bio->bi_private = end_io_wq;
707 bio->bi_end_io = end_workqueue_bio;
708 return 0;
709}
710
711unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
712{
713 unsigned long limit = min_t(unsigned long,
714 info->workers.max_workers,
715 info->fs_devices->open_devices);
716 return 256 * limit;
717}
718
719static void run_one_async_start(struct btrfs_work *work)
720{
721 struct async_submit_bio *async;
722
723 async = container_of(work, struct async_submit_bio, work);
724 async->submit_bio_start(async->inode, async->rw, async->bio,
725 async->mirror_num, async->bio_flags,
726 async->bio_offset);
727}
728
729static void run_one_async_done(struct btrfs_work *work)
730{
731 struct btrfs_fs_info *fs_info;
732 struct async_submit_bio *async;
733 int limit;
734
735 async = container_of(work, struct async_submit_bio, work);
736 fs_info = BTRFS_I(async->inode)->root->fs_info;
737
738 limit = btrfs_async_submit_limit(fs_info);
739 limit = limit * 2 / 3;
740
741 atomic_dec(&fs_info->nr_async_submits);
742
743 if (atomic_read(&fs_info->nr_async_submits) < limit &&
744 waitqueue_active(&fs_info->async_submit_wait))
745 wake_up(&fs_info->async_submit_wait);
746
747 async->submit_bio_done(async->inode, async->rw, async->bio,
748 async->mirror_num, async->bio_flags,
749 async->bio_offset);
750}
751
752static void run_one_async_free(struct btrfs_work *work)
753{
754 struct async_submit_bio *async;
755
756 async = container_of(work, struct async_submit_bio, work);
757 kfree(async);
758}
759
760int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
761 int rw, struct bio *bio, int mirror_num,
762 unsigned long bio_flags,
763 u64 bio_offset,
764 extent_submit_bio_hook_t *submit_bio_start,
765 extent_submit_bio_hook_t *submit_bio_done)
766{
767 struct async_submit_bio *async;
768
769 async = kmalloc(sizeof(*async), GFP_NOFS);
770 if (!async)
771 return -ENOMEM;
772
773 async->inode = inode;
774 async->rw = rw;
775 async->bio = bio;
776 async->mirror_num = mirror_num;
777 async->submit_bio_start = submit_bio_start;
778 async->submit_bio_done = submit_bio_done;
779
780 async->work.func = run_one_async_start;
781 async->work.ordered_func = run_one_async_done;
782 async->work.ordered_free = run_one_async_free;
783
784 async->work.flags = 0;
785 async->bio_flags = bio_flags;
786 async->bio_offset = bio_offset;
787
788 atomic_inc(&fs_info->nr_async_submits);
789
790 if (rw & REQ_SYNC)
791 btrfs_set_work_high_prio(&async->work);
792
793 btrfs_queue_worker(&fs_info->workers, &async->work);
794
795 while (atomic_read(&fs_info->async_submit_draining) &&
796 atomic_read(&fs_info->nr_async_submits)) {
797 wait_event(fs_info->async_submit_wait,
798 (atomic_read(&fs_info->nr_async_submits) == 0));
799 }
800
801 return 0;
802}
803
804static int btree_csum_one_bio(struct bio *bio)
805{
806 struct bio_vec *bvec = bio->bi_io_vec;
807 int bio_index = 0;
808 struct btrfs_root *root;
809
810 WARN_ON(bio->bi_vcnt <= 0);
811 while (bio_index < bio->bi_vcnt) {
812 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
813 csum_dirty_buffer(root, bvec->bv_page);
814 bio_index++;
815 bvec++;
816 }
817 return 0;
818}
819
820static int __btree_submit_bio_start(struct inode *inode, int rw,
821 struct bio *bio, int mirror_num,
822 unsigned long bio_flags,
823 u64 bio_offset)
824{
825 /*
826 * when we're called for a write, we're already in the async
827 * submission context. Just jump into btrfs_map_bio
828 */
829 btree_csum_one_bio(bio);
830 return 0;
831}
832
833static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
834 int mirror_num, unsigned long bio_flags,
835 u64 bio_offset)
836{
837 /*
838 * when we're called for a write, we're already in the async
839 * submission context. Just jump into btrfs_map_bio
840 */
841 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
842}
843
844static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
845 int mirror_num, unsigned long bio_flags,
846 u64 bio_offset)
847{
848 int ret;
849
850 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
851 bio, 1);
852 BUG_ON(ret);
853
854 if (!(rw & REQ_WRITE)) {
855 /*
856 * called for a read, do the setup so that checksum validation
857 * can happen in the async kernel threads
858 */
859 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
860 mirror_num, 0);
861 }
862
863 /*
864 * kthread helpers are used to submit writes so that checksumming
865 * can happen in parallel across all CPUs
866 */
867 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
868 inode, rw, bio, mirror_num, 0,
869 bio_offset,
870 __btree_submit_bio_start,
871 __btree_submit_bio_done);
872}
873
874#ifdef CONFIG_MIGRATION
875static int btree_migratepage(struct address_space *mapping,
876 struct page *newpage, struct page *page)
877{
878 /*
879 * we can't safely write a btree page from here,
880 * we haven't done the locking hook
881 */
882 if (PageDirty(page))
883 return -EAGAIN;
884 /*
885 * Buffers may be managed in a filesystem specific way.
886 * We must have no buffers or drop them.
887 */
888 if (page_has_private(page) &&
889 !try_to_release_page(page, GFP_KERNEL))
890 return -EAGAIN;
891 return migrate_page(mapping, newpage, page);
892}
893#endif
894
895static int btree_writepage(struct page *page, struct writeback_control *wbc)
896{
897 struct extent_io_tree *tree;
898 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
899 struct extent_buffer *eb;
900 int was_dirty;
901
902 tree = &BTRFS_I(page->mapping->host)->io_tree;
903 if (!(current->flags & PF_MEMALLOC)) {
904 return extent_write_full_page(tree, page,
905 btree_get_extent, wbc);
906 }
907
908 redirty_page_for_writepage(wbc, page);
909 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
910 WARN_ON(!eb);
911
912 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
913 if (!was_dirty) {
914 spin_lock(&root->fs_info->delalloc_lock);
915 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
916 spin_unlock(&root->fs_info->delalloc_lock);
917 }
918 free_extent_buffer(eb);
919
920 unlock_page(page);
921 return 0;
922}
923
924static int btree_writepages(struct address_space *mapping,
925 struct writeback_control *wbc)
926{
927 struct extent_io_tree *tree;
928 tree = &BTRFS_I(mapping->host)->io_tree;
929 if (wbc->sync_mode == WB_SYNC_NONE) {
930 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
931 u64 num_dirty;
932 unsigned long thresh = 32 * 1024 * 1024;
933
934 if (wbc->for_kupdate)
935 return 0;
936
937 /* this is a bit racy, but that's ok */
938 num_dirty = root->fs_info->dirty_metadata_bytes;
939 if (num_dirty < thresh)
940 return 0;
941 }
942 return extent_writepages(tree, mapping, btree_get_extent, wbc);
943}
944
945static int btree_readpage(struct file *file, struct page *page)
946{
947 struct extent_io_tree *tree;
948 tree = &BTRFS_I(page->mapping->host)->io_tree;
949 return extent_read_full_page(tree, page, btree_get_extent, 0);
950}
951
952static int btree_releasepage(struct page *page, gfp_t gfp_flags)
953{
954 struct extent_io_tree *tree;
955 struct extent_map_tree *map;
956 int ret;
957
958 if (PageWriteback(page) || PageDirty(page))
959 return 0;
960
961 tree = &BTRFS_I(page->mapping->host)->io_tree;
962 map = &BTRFS_I(page->mapping->host)->extent_tree;
963
964 ret = try_release_extent_state(map, tree, page, gfp_flags);
965 if (!ret)
966 return 0;
967
968 ret = try_release_extent_buffer(tree, page);
969 if (ret == 1) {
970 ClearPagePrivate(page);
971 set_page_private(page, 0);
972 page_cache_release(page);
973 }
974
975 return ret;
976}
977
978static void btree_invalidatepage(struct page *page, unsigned long offset)
979{
980 struct extent_io_tree *tree;
981 tree = &BTRFS_I(page->mapping->host)->io_tree;
982 extent_invalidatepage(tree, page, offset);
983 btree_releasepage(page, GFP_NOFS);
984 if (PagePrivate(page)) {
985 printk(KERN_WARNING "btrfs warning page private not zero "
986 "on page %llu\n", (unsigned long long)page_offset(page));
987 ClearPagePrivate(page);
988 set_page_private(page, 0);
989 page_cache_release(page);
990 }
991}
992
993static const struct address_space_operations btree_aops = {
994 .readpage = btree_readpage,
995 .writepage = btree_writepage,
996 .writepages = btree_writepages,
997 .releasepage = btree_releasepage,
998 .invalidatepage = btree_invalidatepage,
999#ifdef CONFIG_MIGRATION
1000 .migratepage = btree_migratepage,
1001#endif
1002};
1003
1004int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1005 u64 parent_transid)
1006{
1007 struct extent_buffer *buf = NULL;
1008 struct inode *btree_inode = root->fs_info->btree_inode;
1009 int ret = 0;
1010
1011 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1012 if (!buf)
1013 return 0;
1014 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1015 buf, 0, WAIT_NONE, btree_get_extent, 0);
1016 free_extent_buffer(buf);
1017 return ret;
1018}
1019
1020int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1021 int mirror_num, struct extent_buffer **eb)
1022{
1023 struct extent_buffer *buf = NULL;
1024 struct inode *btree_inode = root->fs_info->btree_inode;
1025 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1026 int ret;
1027
1028 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1029 if (!buf)
1030 return 0;
1031
1032 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1033
1034 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1035 btree_get_extent, mirror_num);
1036 if (ret) {
1037 free_extent_buffer(buf);
1038 return ret;
1039 }
1040
1041 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1042 free_extent_buffer(buf);
1043 return -EIO;
1044 } else if (extent_buffer_uptodate(io_tree, buf, NULL)) {
1045 *eb = buf;
1046 } else {
1047 free_extent_buffer(buf);
1048 }
1049 return 0;
1050}
1051
1052struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1053 u64 bytenr, u32 blocksize)
1054{
1055 struct inode *btree_inode = root->fs_info->btree_inode;
1056 struct extent_buffer *eb;
1057 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1058 bytenr, blocksize);
1059 return eb;
1060}
1061
1062struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1063 u64 bytenr, u32 blocksize)
1064{
1065 struct inode *btree_inode = root->fs_info->btree_inode;
1066 struct extent_buffer *eb;
1067
1068 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1069 bytenr, blocksize, NULL);
1070 return eb;
1071}
1072
1073
1074int btrfs_write_tree_block(struct extent_buffer *buf)
1075{
1076 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1077 buf->start + buf->len - 1);
1078}
1079
1080int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1081{
1082 return filemap_fdatawait_range(buf->first_page->mapping,
1083 buf->start, buf->start + buf->len - 1);
1084}
1085
1086struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1087 u32 blocksize, u64 parent_transid)
1088{
1089 struct extent_buffer *buf = NULL;
1090 int ret;
1091
1092 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1093 if (!buf)
1094 return NULL;
1095
1096 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1097
1098 if (ret == 0)
1099 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1100 return buf;
1101
1102}
1103
1104int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1105 struct extent_buffer *buf)
1106{
1107 struct inode *btree_inode = root->fs_info->btree_inode;
1108 if (btrfs_header_generation(buf) ==
1109 root->fs_info->running_transaction->transid) {
1110 btrfs_assert_tree_locked(buf);
1111
1112 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1113 spin_lock(&root->fs_info->delalloc_lock);
1114 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1115 root->fs_info->dirty_metadata_bytes -= buf->len;
1116 else
1117 WARN_ON(1);
1118 spin_unlock(&root->fs_info->delalloc_lock);
1119 }
1120
1121 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1122 btrfs_set_lock_blocking(buf);
1123 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1124 buf);
1125 }
1126 return 0;
1127}
1128
1129static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1130 u32 stripesize, struct btrfs_root *root,
1131 struct btrfs_fs_info *fs_info,
1132 u64 objectid)
1133{
1134 root->node = NULL;
1135 root->commit_root = NULL;
1136 root->sectorsize = sectorsize;
1137 root->nodesize = nodesize;
1138 root->leafsize = leafsize;
1139 root->stripesize = stripesize;
1140 root->ref_cows = 0;
1141 root->track_dirty = 0;
1142 root->in_radix = 0;
1143 root->orphan_item_inserted = 0;
1144 root->orphan_cleanup_state = 0;
1145
1146 root->fs_info = fs_info;
1147 root->objectid = objectid;
1148 root->last_trans = 0;
1149 root->highest_objectid = 0;
1150 root->name = NULL;
1151 root->inode_tree = RB_ROOT;
1152 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1153 root->block_rsv = NULL;
1154 root->orphan_block_rsv = NULL;
1155
1156 INIT_LIST_HEAD(&root->dirty_list);
1157 INIT_LIST_HEAD(&root->orphan_list);
1158 INIT_LIST_HEAD(&root->root_list);
1159 spin_lock_init(&root->orphan_lock);
1160 spin_lock_init(&root->inode_lock);
1161 spin_lock_init(&root->accounting_lock);
1162 mutex_init(&root->objectid_mutex);
1163 mutex_init(&root->log_mutex);
1164 init_waitqueue_head(&root->log_writer_wait);
1165 init_waitqueue_head(&root->log_commit_wait[0]);
1166 init_waitqueue_head(&root->log_commit_wait[1]);
1167 atomic_set(&root->log_commit[0], 0);
1168 atomic_set(&root->log_commit[1], 0);
1169 atomic_set(&root->log_writers, 0);
1170 root->log_batch = 0;
1171 root->log_transid = 0;
1172 root->last_log_commit = 0;
1173 extent_io_tree_init(&root->dirty_log_pages,
1174 fs_info->btree_inode->i_mapping);
1175
1176 memset(&root->root_key, 0, sizeof(root->root_key));
1177 memset(&root->root_item, 0, sizeof(root->root_item));
1178 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1179 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1180 root->defrag_trans_start = fs_info->generation;
1181 init_completion(&root->kobj_unregister);
1182 root->defrag_running = 0;
1183 root->root_key.objectid = objectid;
1184 root->anon_dev = 0;
1185 return 0;
1186}
1187
1188static int find_and_setup_root(struct btrfs_root *tree_root,
1189 struct btrfs_fs_info *fs_info,
1190 u64 objectid,
1191 struct btrfs_root *root)
1192{
1193 int ret;
1194 u32 blocksize;
1195 u64 generation;
1196
1197 __setup_root(tree_root->nodesize, tree_root->leafsize,
1198 tree_root->sectorsize, tree_root->stripesize,
1199 root, fs_info, objectid);
1200 ret = btrfs_find_last_root(tree_root, objectid,
1201 &root->root_item, &root->root_key);
1202 if (ret > 0)
1203 return -ENOENT;
1204 BUG_ON(ret);
1205
1206 generation = btrfs_root_generation(&root->root_item);
1207 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1208 root->commit_root = NULL;
1209 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1210 blocksize, generation);
1211 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1212 free_extent_buffer(root->node);
1213 root->node = NULL;
1214 return -EIO;
1215 }
1216 root->commit_root = btrfs_root_node(root);
1217 return 0;
1218}
1219
1220static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1221 struct btrfs_fs_info *fs_info)
1222{
1223 struct btrfs_root *root;
1224 struct btrfs_root *tree_root = fs_info->tree_root;
1225 struct extent_buffer *leaf;
1226
1227 root = kzalloc(sizeof(*root), GFP_NOFS);
1228 if (!root)
1229 return ERR_PTR(-ENOMEM);
1230
1231 __setup_root(tree_root->nodesize, tree_root->leafsize,
1232 tree_root->sectorsize, tree_root->stripesize,
1233 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1234
1235 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1236 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1237 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1238 /*
1239 * log trees do not get reference counted because they go away
1240 * before a real commit is actually done. They do store pointers
1241 * to file data extents, and those reference counts still get
1242 * updated (along with back refs to the log tree).
1243 */
1244 root->ref_cows = 0;
1245
1246 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1247 BTRFS_TREE_LOG_OBJECTID, NULL,
1248 0, 0, 0, 0);
1249 if (IS_ERR(leaf)) {
1250 kfree(root);
1251 return ERR_CAST(leaf);
1252 }
1253
1254 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1255 btrfs_set_header_bytenr(leaf, leaf->start);
1256 btrfs_set_header_generation(leaf, trans->transid);
1257 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1258 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1259 root->node = leaf;
1260
1261 write_extent_buffer(root->node, root->fs_info->fsid,
1262 (unsigned long)btrfs_header_fsid(root->node),
1263 BTRFS_FSID_SIZE);
1264 btrfs_mark_buffer_dirty(root->node);
1265 btrfs_tree_unlock(root->node);
1266 return root;
1267}
1268
1269int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1270 struct btrfs_fs_info *fs_info)
1271{
1272 struct btrfs_root *log_root;
1273
1274 log_root = alloc_log_tree(trans, fs_info);
1275 if (IS_ERR(log_root))
1276 return PTR_ERR(log_root);
1277 WARN_ON(fs_info->log_root_tree);
1278 fs_info->log_root_tree = log_root;
1279 return 0;
1280}
1281
1282int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1283 struct btrfs_root *root)
1284{
1285 struct btrfs_root *log_root;
1286 struct btrfs_inode_item *inode_item;
1287
1288 log_root = alloc_log_tree(trans, root->fs_info);
1289 if (IS_ERR(log_root))
1290 return PTR_ERR(log_root);
1291
1292 log_root->last_trans = trans->transid;
1293 log_root->root_key.offset = root->root_key.objectid;
1294
1295 inode_item = &log_root->root_item.inode;
1296 inode_item->generation = cpu_to_le64(1);
1297 inode_item->size = cpu_to_le64(3);
1298 inode_item->nlink = cpu_to_le32(1);
1299 inode_item->nbytes = cpu_to_le64(root->leafsize);
1300 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1301
1302 btrfs_set_root_node(&log_root->root_item, log_root->node);
1303
1304 WARN_ON(root->log_root);
1305 root->log_root = log_root;
1306 root->log_transid = 0;
1307 root->last_log_commit = 0;
1308 return 0;
1309}
1310
1311struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1312 struct btrfs_key *location)
1313{
1314 struct btrfs_root *root;
1315 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1316 struct btrfs_path *path;
1317 struct extent_buffer *l;
1318 u64 generation;
1319 u32 blocksize;
1320 int ret = 0;
1321
1322 root = kzalloc(sizeof(*root), GFP_NOFS);
1323 if (!root)
1324 return ERR_PTR(-ENOMEM);
1325 if (location->offset == (u64)-1) {
1326 ret = find_and_setup_root(tree_root, fs_info,
1327 location->objectid, root);
1328 if (ret) {
1329 kfree(root);
1330 return ERR_PTR(ret);
1331 }
1332 goto out;
1333 }
1334
1335 __setup_root(tree_root->nodesize, tree_root->leafsize,
1336 tree_root->sectorsize, tree_root->stripesize,
1337 root, fs_info, location->objectid);
1338
1339 path = btrfs_alloc_path();
1340 if (!path) {
1341 kfree(root);
1342 return ERR_PTR(-ENOMEM);
1343 }
1344 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1345 if (ret == 0) {
1346 l = path->nodes[0];
1347 read_extent_buffer(l, &root->root_item,
1348 btrfs_item_ptr_offset(l, path->slots[0]),
1349 sizeof(root->root_item));
1350 memcpy(&root->root_key, location, sizeof(*location));
1351 }
1352 btrfs_free_path(path);
1353 if (ret) {
1354 kfree(root);
1355 if (ret > 0)
1356 ret = -ENOENT;
1357 return ERR_PTR(ret);
1358 }
1359
1360 generation = btrfs_root_generation(&root->root_item);
1361 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1362 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1363 blocksize, generation);
1364 root->commit_root = btrfs_root_node(root);
1365 BUG_ON(!root->node);
1366out:
1367 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1368 root->ref_cows = 1;
1369 btrfs_check_and_init_root_item(&root->root_item);
1370 }
1371
1372 return root;
1373}
1374
1375struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1376 struct btrfs_key *location)
1377{
1378 struct btrfs_root *root;
1379 int ret;
1380
1381 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1382 return fs_info->tree_root;
1383 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1384 return fs_info->extent_root;
1385 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1386 return fs_info->chunk_root;
1387 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1388 return fs_info->dev_root;
1389 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1390 return fs_info->csum_root;
1391again:
1392 spin_lock(&fs_info->fs_roots_radix_lock);
1393 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1394 (unsigned long)location->objectid);
1395 spin_unlock(&fs_info->fs_roots_radix_lock);
1396 if (root)
1397 return root;
1398
1399 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1400 if (IS_ERR(root))
1401 return root;
1402
1403 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1404 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1405 GFP_NOFS);
1406 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1407 ret = -ENOMEM;
1408 goto fail;
1409 }
1410
1411 btrfs_init_free_ino_ctl(root);
1412 mutex_init(&root->fs_commit_mutex);
1413 spin_lock_init(&root->cache_lock);
1414 init_waitqueue_head(&root->cache_wait);
1415
1416 ret = get_anon_bdev(&root->anon_dev);
1417 if (ret)
1418 goto fail;
1419
1420 if (btrfs_root_refs(&root->root_item) == 0) {
1421 ret = -ENOENT;
1422 goto fail;
1423 }
1424
1425 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1426 if (ret < 0)
1427 goto fail;
1428 if (ret == 0)
1429 root->orphan_item_inserted = 1;
1430
1431 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1432 if (ret)
1433 goto fail;
1434
1435 spin_lock(&fs_info->fs_roots_radix_lock);
1436 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1437 (unsigned long)root->root_key.objectid,
1438 root);
1439 if (ret == 0)
1440 root->in_radix = 1;
1441
1442 spin_unlock(&fs_info->fs_roots_radix_lock);
1443 radix_tree_preload_end();
1444 if (ret) {
1445 if (ret == -EEXIST) {
1446 free_fs_root(root);
1447 goto again;
1448 }
1449 goto fail;
1450 }
1451
1452 ret = btrfs_find_dead_roots(fs_info->tree_root,
1453 root->root_key.objectid);
1454 WARN_ON(ret);
1455 return root;
1456fail:
1457 free_fs_root(root);
1458 return ERR_PTR(ret);
1459}
1460
1461static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1462{
1463 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1464 int ret = 0;
1465 struct btrfs_device *device;
1466 struct backing_dev_info *bdi;
1467
1468 rcu_read_lock();
1469 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1470 if (!device->bdev)
1471 continue;
1472 bdi = blk_get_backing_dev_info(device->bdev);
1473 if (bdi && bdi_congested(bdi, bdi_bits)) {
1474 ret = 1;
1475 break;
1476 }
1477 }
1478 rcu_read_unlock();
1479 return ret;
1480}
1481
1482/*
1483 * If this fails, caller must call bdi_destroy() to get rid of the
1484 * bdi again.
1485 */
1486static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1487{
1488 int err;
1489
1490 bdi->capabilities = BDI_CAP_MAP_COPY;
1491 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1492 if (err)
1493 return err;
1494
1495 bdi->ra_pages = default_backing_dev_info.ra_pages;
1496 bdi->congested_fn = btrfs_congested_fn;
1497 bdi->congested_data = info;
1498 return 0;
1499}
1500
1501static int bio_ready_for_csum(struct bio *bio)
1502{
1503 u64 length = 0;
1504 u64 buf_len = 0;
1505 u64 start = 0;
1506 struct page *page;
1507 struct extent_io_tree *io_tree = NULL;
1508 struct bio_vec *bvec;
1509 int i;
1510 int ret;
1511
1512 bio_for_each_segment(bvec, bio, i) {
1513 page = bvec->bv_page;
1514 if (page->private == EXTENT_PAGE_PRIVATE) {
1515 length += bvec->bv_len;
1516 continue;
1517 }
1518 if (!page->private) {
1519 length += bvec->bv_len;
1520 continue;
1521 }
1522 length = bvec->bv_len;
1523 buf_len = page->private >> 2;
1524 start = page_offset(page) + bvec->bv_offset;
1525 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1526 }
1527 /* are we fully contained in this bio? */
1528 if (buf_len <= length)
1529 return 1;
1530
1531 ret = extent_range_uptodate(io_tree, start + length,
1532 start + buf_len - 1);
1533 return ret;
1534}
1535
1536/*
1537 * called by the kthread helper functions to finally call the bio end_io
1538 * functions. This is where read checksum verification actually happens
1539 */
1540static void end_workqueue_fn(struct btrfs_work *work)
1541{
1542 struct bio *bio;
1543 struct end_io_wq *end_io_wq;
1544 struct btrfs_fs_info *fs_info;
1545 int error;
1546
1547 end_io_wq = container_of(work, struct end_io_wq, work);
1548 bio = end_io_wq->bio;
1549 fs_info = end_io_wq->info;
1550
1551 /* metadata bio reads are special because the whole tree block must
1552 * be checksummed at once. This makes sure the entire block is in
1553 * ram and up to date before trying to verify things. For
1554 * blocksize <= pagesize, it is basically a noop
1555 */
1556 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1557 !bio_ready_for_csum(bio)) {
1558 btrfs_queue_worker(&fs_info->endio_meta_workers,
1559 &end_io_wq->work);
1560 return;
1561 }
1562 error = end_io_wq->error;
1563 bio->bi_private = end_io_wq->private;
1564 bio->bi_end_io = end_io_wq->end_io;
1565 kfree(end_io_wq);
1566 bio_endio(bio, error);
1567}
1568
1569static int cleaner_kthread(void *arg)
1570{
1571 struct btrfs_root *root = arg;
1572
1573 do {
1574 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1575
1576 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1577 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1578 btrfs_run_delayed_iputs(root);
1579 btrfs_clean_old_snapshots(root);
1580 mutex_unlock(&root->fs_info->cleaner_mutex);
1581 btrfs_run_defrag_inodes(root->fs_info);
1582 }
1583
1584 if (freezing(current)) {
1585 refrigerator();
1586 } else {
1587 set_current_state(TASK_INTERRUPTIBLE);
1588 if (!kthread_should_stop())
1589 schedule();
1590 __set_current_state(TASK_RUNNING);
1591 }
1592 } while (!kthread_should_stop());
1593 return 0;
1594}
1595
1596static int transaction_kthread(void *arg)
1597{
1598 struct btrfs_root *root = arg;
1599 struct btrfs_trans_handle *trans;
1600 struct btrfs_transaction *cur;
1601 u64 transid;
1602 unsigned long now;
1603 unsigned long delay;
1604 int ret;
1605
1606 do {
1607 delay = HZ * 30;
1608 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1609 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1610
1611 spin_lock(&root->fs_info->trans_lock);
1612 cur = root->fs_info->running_transaction;
1613 if (!cur) {
1614 spin_unlock(&root->fs_info->trans_lock);
1615 goto sleep;
1616 }
1617
1618 now = get_seconds();
1619 if (!cur->blocked &&
1620 (now < cur->start_time || now - cur->start_time < 30)) {
1621 spin_unlock(&root->fs_info->trans_lock);
1622 delay = HZ * 5;
1623 goto sleep;
1624 }
1625 transid = cur->transid;
1626 spin_unlock(&root->fs_info->trans_lock);
1627
1628 trans = btrfs_join_transaction(root);
1629 BUG_ON(IS_ERR(trans));
1630 if (transid == trans->transid) {
1631 ret = btrfs_commit_transaction(trans, root);
1632 BUG_ON(ret);
1633 } else {
1634 btrfs_end_transaction(trans, root);
1635 }
1636sleep:
1637 wake_up_process(root->fs_info->cleaner_kthread);
1638 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1639
1640 if (freezing(current)) {
1641 refrigerator();
1642 } else {
1643 set_current_state(TASK_INTERRUPTIBLE);
1644 if (!kthread_should_stop() &&
1645 !btrfs_transaction_blocked(root->fs_info))
1646 schedule_timeout(delay);
1647 __set_current_state(TASK_RUNNING);
1648 }
1649 } while (!kthread_should_stop());
1650 return 0;
1651}
1652
1653/*
1654 * this will find the highest generation in the array of
1655 * root backups. The index of the highest array is returned,
1656 * or -1 if we can't find anything.
1657 *
1658 * We check to make sure the array is valid by comparing the
1659 * generation of the latest root in the array with the generation
1660 * in the super block. If they don't match we pitch it.
1661 */
1662static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1663{
1664 u64 cur;
1665 int newest_index = -1;
1666 struct btrfs_root_backup *root_backup;
1667 int i;
1668
1669 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1670 root_backup = info->super_copy->super_roots + i;
1671 cur = btrfs_backup_tree_root_gen(root_backup);
1672 if (cur == newest_gen)
1673 newest_index = i;
1674 }
1675
1676 /* check to see if we actually wrapped around */
1677 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1678 root_backup = info->super_copy->super_roots;
1679 cur = btrfs_backup_tree_root_gen(root_backup);
1680 if (cur == newest_gen)
1681 newest_index = 0;
1682 }
1683 return newest_index;
1684}
1685
1686
1687/*
1688 * find the oldest backup so we know where to store new entries
1689 * in the backup array. This will set the backup_root_index
1690 * field in the fs_info struct
1691 */
1692static void find_oldest_super_backup(struct btrfs_fs_info *info,
1693 u64 newest_gen)
1694{
1695 int newest_index = -1;
1696
1697 newest_index = find_newest_super_backup(info, newest_gen);
1698 /* if there was garbage in there, just move along */
1699 if (newest_index == -1) {
1700 info->backup_root_index = 0;
1701 } else {
1702 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1703 }
1704}
1705
1706/*
1707 * copy all the root pointers into the super backup array.
1708 * this will bump the backup pointer by one when it is
1709 * done
1710 */
1711static void backup_super_roots(struct btrfs_fs_info *info)
1712{
1713 int next_backup;
1714 struct btrfs_root_backup *root_backup;
1715 int last_backup;
1716
1717 next_backup = info->backup_root_index;
1718 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1719 BTRFS_NUM_BACKUP_ROOTS;
1720
1721 /*
1722 * just overwrite the last backup if we're at the same generation
1723 * this happens only at umount
1724 */
1725 root_backup = info->super_for_commit->super_roots + last_backup;
1726 if (btrfs_backup_tree_root_gen(root_backup) ==
1727 btrfs_header_generation(info->tree_root->node))
1728 next_backup = last_backup;
1729
1730 root_backup = info->super_for_commit->super_roots + next_backup;
1731
1732 /*
1733 * make sure all of our padding and empty slots get zero filled
1734 * regardless of which ones we use today
1735 */
1736 memset(root_backup, 0, sizeof(*root_backup));
1737
1738 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1739
1740 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1741 btrfs_set_backup_tree_root_gen(root_backup,
1742 btrfs_header_generation(info->tree_root->node));
1743
1744 btrfs_set_backup_tree_root_level(root_backup,
1745 btrfs_header_level(info->tree_root->node));
1746
1747 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1748 btrfs_set_backup_chunk_root_gen(root_backup,
1749 btrfs_header_generation(info->chunk_root->node));
1750 btrfs_set_backup_chunk_root_level(root_backup,
1751 btrfs_header_level(info->chunk_root->node));
1752
1753 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1754 btrfs_set_backup_extent_root_gen(root_backup,
1755 btrfs_header_generation(info->extent_root->node));
1756 btrfs_set_backup_extent_root_level(root_backup,
1757 btrfs_header_level(info->extent_root->node));
1758
1759 /*
1760 * we might commit during log recovery, which happens before we set
1761 * the fs_root. Make sure it is valid before we fill it in.
1762 */
1763 if (info->fs_root && info->fs_root->node) {
1764 btrfs_set_backup_fs_root(root_backup,
1765 info->fs_root->node->start);
1766 btrfs_set_backup_fs_root_gen(root_backup,
1767 btrfs_header_generation(info->fs_root->node));
1768 btrfs_set_backup_fs_root_level(root_backup,
1769 btrfs_header_level(info->fs_root->node));
1770 }
1771
1772 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1773 btrfs_set_backup_dev_root_gen(root_backup,
1774 btrfs_header_generation(info->dev_root->node));
1775 btrfs_set_backup_dev_root_level(root_backup,
1776 btrfs_header_level(info->dev_root->node));
1777
1778 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1779 btrfs_set_backup_csum_root_gen(root_backup,
1780 btrfs_header_generation(info->csum_root->node));
1781 btrfs_set_backup_csum_root_level(root_backup,
1782 btrfs_header_level(info->csum_root->node));
1783
1784 btrfs_set_backup_total_bytes(root_backup,
1785 btrfs_super_total_bytes(info->super_copy));
1786 btrfs_set_backup_bytes_used(root_backup,
1787 btrfs_super_bytes_used(info->super_copy));
1788 btrfs_set_backup_num_devices(root_backup,
1789 btrfs_super_num_devices(info->super_copy));
1790
1791 /*
1792 * if we don't copy this out to the super_copy, it won't get remembered
1793 * for the next commit
1794 */
1795 memcpy(&info->super_copy->super_roots,
1796 &info->super_for_commit->super_roots,
1797 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1798}
1799
1800/*
1801 * this copies info out of the root backup array and back into
1802 * the in-memory super block. It is meant to help iterate through
1803 * the array, so you send it the number of backups you've already
1804 * tried and the last backup index you used.
1805 *
1806 * this returns -1 when it has tried all the backups
1807 */
1808static noinline int next_root_backup(struct btrfs_fs_info *info,
1809 struct btrfs_super_block *super,
1810 int *num_backups_tried, int *backup_index)
1811{
1812 struct btrfs_root_backup *root_backup;
1813 int newest = *backup_index;
1814
1815 if (*num_backups_tried == 0) {
1816 u64 gen = btrfs_super_generation(super);
1817
1818 newest = find_newest_super_backup(info, gen);
1819 if (newest == -1)
1820 return -1;
1821
1822 *backup_index = newest;
1823 *num_backups_tried = 1;
1824 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1825 /* we've tried all the backups, all done */
1826 return -1;
1827 } else {
1828 /* jump to the next oldest backup */
1829 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1830 BTRFS_NUM_BACKUP_ROOTS;
1831 *backup_index = newest;
1832 *num_backups_tried += 1;
1833 }
1834 root_backup = super->super_roots + newest;
1835
1836 btrfs_set_super_generation(super,
1837 btrfs_backup_tree_root_gen(root_backup));
1838 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1839 btrfs_set_super_root_level(super,
1840 btrfs_backup_tree_root_level(root_backup));
1841 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1842
1843 /*
1844 * fixme: the total bytes and num_devices need to match or we should
1845 * need a fsck
1846 */
1847 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1848 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1849 return 0;
1850}
1851
1852/* helper to cleanup tree roots */
1853static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1854{
1855 free_extent_buffer(info->tree_root->node);
1856 free_extent_buffer(info->tree_root->commit_root);
1857 free_extent_buffer(info->dev_root->node);
1858 free_extent_buffer(info->dev_root->commit_root);
1859 free_extent_buffer(info->extent_root->node);
1860 free_extent_buffer(info->extent_root->commit_root);
1861 free_extent_buffer(info->csum_root->node);
1862 free_extent_buffer(info->csum_root->commit_root);
1863
1864 info->tree_root->node = NULL;
1865 info->tree_root->commit_root = NULL;
1866 info->dev_root->node = NULL;
1867 info->dev_root->commit_root = NULL;
1868 info->extent_root->node = NULL;
1869 info->extent_root->commit_root = NULL;
1870 info->csum_root->node = NULL;
1871 info->csum_root->commit_root = NULL;
1872
1873 if (chunk_root) {
1874 free_extent_buffer(info->chunk_root->node);
1875 free_extent_buffer(info->chunk_root->commit_root);
1876 info->chunk_root->node = NULL;
1877 info->chunk_root->commit_root = NULL;
1878 }
1879}
1880
1881
1882struct btrfs_root *open_ctree(struct super_block *sb,
1883 struct btrfs_fs_devices *fs_devices,
1884 char *options)
1885{
1886 u32 sectorsize;
1887 u32 nodesize;
1888 u32 leafsize;
1889 u32 blocksize;
1890 u32 stripesize;
1891 u64 generation;
1892 u64 features;
1893 struct btrfs_key location;
1894 struct buffer_head *bh;
1895 struct btrfs_super_block *disk_super;
1896 struct btrfs_root *tree_root = btrfs_sb(sb);
1897 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1898 struct btrfs_root *extent_root;
1899 struct btrfs_root *csum_root;
1900 struct btrfs_root *chunk_root;
1901 struct btrfs_root *dev_root;
1902 struct btrfs_root *log_tree_root;
1903 int ret;
1904 int err = -EINVAL;
1905 int num_backups_tried = 0;
1906 int backup_index = 0;
1907
1908 extent_root = fs_info->extent_root =
1909 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1910 csum_root = fs_info->csum_root =
1911 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1912 chunk_root = fs_info->chunk_root =
1913 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1914 dev_root = fs_info->dev_root =
1915 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1916
1917 if (!extent_root || !csum_root || !chunk_root || !dev_root) {
1918 err = -ENOMEM;
1919 goto fail;
1920 }
1921
1922 ret = init_srcu_struct(&fs_info->subvol_srcu);
1923 if (ret) {
1924 err = ret;
1925 goto fail;
1926 }
1927
1928 ret = setup_bdi(fs_info, &fs_info->bdi);
1929 if (ret) {
1930 err = ret;
1931 goto fail_srcu;
1932 }
1933
1934 fs_info->btree_inode = new_inode(sb);
1935 if (!fs_info->btree_inode) {
1936 err = -ENOMEM;
1937 goto fail_bdi;
1938 }
1939
1940 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1941
1942 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1943 INIT_LIST_HEAD(&fs_info->trans_list);
1944 INIT_LIST_HEAD(&fs_info->dead_roots);
1945 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1946 INIT_LIST_HEAD(&fs_info->hashers);
1947 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1948 INIT_LIST_HEAD(&fs_info->ordered_operations);
1949 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1950 spin_lock_init(&fs_info->delalloc_lock);
1951 spin_lock_init(&fs_info->trans_lock);
1952 spin_lock_init(&fs_info->ref_cache_lock);
1953 spin_lock_init(&fs_info->fs_roots_radix_lock);
1954 spin_lock_init(&fs_info->delayed_iput_lock);
1955 spin_lock_init(&fs_info->defrag_inodes_lock);
1956 spin_lock_init(&fs_info->free_chunk_lock);
1957 mutex_init(&fs_info->reloc_mutex);
1958
1959 init_completion(&fs_info->kobj_unregister);
1960 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1961 INIT_LIST_HEAD(&fs_info->space_info);
1962 btrfs_mapping_init(&fs_info->mapping_tree);
1963 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1964 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1965 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1966 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1967 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1968 btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1969 atomic_set(&fs_info->nr_async_submits, 0);
1970 atomic_set(&fs_info->async_delalloc_pages, 0);
1971 atomic_set(&fs_info->async_submit_draining, 0);
1972 atomic_set(&fs_info->nr_async_bios, 0);
1973 atomic_set(&fs_info->defrag_running, 0);
1974 fs_info->sb = sb;
1975 fs_info->max_inline = 8192 * 1024;
1976 fs_info->metadata_ratio = 0;
1977 fs_info->defrag_inodes = RB_ROOT;
1978 fs_info->trans_no_join = 0;
1979 fs_info->free_chunk_space = 0;
1980
1981 /* readahead state */
1982 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1983 spin_lock_init(&fs_info->reada_lock);
1984
1985 fs_info->thread_pool_size = min_t(unsigned long,
1986 num_online_cpus() + 2, 8);
1987
1988 INIT_LIST_HEAD(&fs_info->ordered_extents);
1989 spin_lock_init(&fs_info->ordered_extent_lock);
1990 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1991 GFP_NOFS);
1992 if (!fs_info->delayed_root) {
1993 err = -ENOMEM;
1994 goto fail_iput;
1995 }
1996 btrfs_init_delayed_root(fs_info->delayed_root);
1997
1998 mutex_init(&fs_info->scrub_lock);
1999 atomic_set(&fs_info->scrubs_running, 0);
2000 atomic_set(&fs_info->scrub_pause_req, 0);
2001 atomic_set(&fs_info->scrubs_paused, 0);
2002 atomic_set(&fs_info->scrub_cancel_req, 0);
2003 init_waitqueue_head(&fs_info->scrub_pause_wait);
2004 init_rwsem(&fs_info->scrub_super_lock);
2005 fs_info->scrub_workers_refcnt = 0;
2006#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2007 fs_info->check_integrity_print_mask = 0;
2008#endif
2009
2010 spin_lock_init(&fs_info->balance_lock);
2011 mutex_init(&fs_info->balance_mutex);
2012 atomic_set(&fs_info->balance_running, 0);
2013 atomic_set(&fs_info->balance_pause_req, 0);
2014 atomic_set(&fs_info->balance_cancel_req, 0);
2015 fs_info->balance_ctl = NULL;
2016 init_waitqueue_head(&fs_info->balance_wait_q);
2017
2018 sb->s_blocksize = 4096;
2019 sb->s_blocksize_bits = blksize_bits(4096);
2020 sb->s_bdi = &fs_info->bdi;
2021
2022 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2023 set_nlink(fs_info->btree_inode, 1);
2024 /*
2025 * we set the i_size on the btree inode to the max possible int.
2026 * the real end of the address space is determined by all of
2027 * the devices in the system
2028 */
2029 fs_info->btree_inode->i_size = OFFSET_MAX;
2030 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2031 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2032
2033 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2034 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2035 fs_info->btree_inode->i_mapping);
2036 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2037
2038 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2039
2040 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2041 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2042 sizeof(struct btrfs_key));
2043 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
2044 insert_inode_hash(fs_info->btree_inode);
2045
2046 spin_lock_init(&fs_info->block_group_cache_lock);
2047 fs_info->block_group_cache_tree = RB_ROOT;
2048
2049 extent_io_tree_init(&fs_info->freed_extents[0],
2050 fs_info->btree_inode->i_mapping);
2051 extent_io_tree_init(&fs_info->freed_extents[1],
2052 fs_info->btree_inode->i_mapping);
2053 fs_info->pinned_extents = &fs_info->freed_extents[0];
2054 fs_info->do_barriers = 1;
2055
2056
2057 mutex_init(&fs_info->ordered_operations_mutex);
2058 mutex_init(&fs_info->tree_log_mutex);
2059 mutex_init(&fs_info->chunk_mutex);
2060 mutex_init(&fs_info->transaction_kthread_mutex);
2061 mutex_init(&fs_info->cleaner_mutex);
2062 mutex_init(&fs_info->volume_mutex);
2063 init_rwsem(&fs_info->extent_commit_sem);
2064 init_rwsem(&fs_info->cleanup_work_sem);
2065 init_rwsem(&fs_info->subvol_sem);
2066
2067 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2068 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2069
2070 init_waitqueue_head(&fs_info->transaction_throttle);
2071 init_waitqueue_head(&fs_info->transaction_wait);
2072 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2073 init_waitqueue_head(&fs_info->async_submit_wait);
2074
2075 __setup_root(4096, 4096, 4096, 4096, tree_root,
2076 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2077
2078 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2079 if (!bh) {
2080 err = -EINVAL;
2081 goto fail_alloc;
2082 }
2083
2084 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2085 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2086 sizeof(*fs_info->super_for_commit));
2087 brelse(bh);
2088
2089 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2090
2091 disk_super = fs_info->super_copy;
2092 if (!btrfs_super_root(disk_super))
2093 goto fail_alloc;
2094
2095 /* check FS state, whether FS is broken. */
2096 fs_info->fs_state |= btrfs_super_flags(disk_super);
2097
2098 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2099
2100 /*
2101 * run through our array of backup supers and setup
2102 * our ring pointer to the oldest one
2103 */
2104 generation = btrfs_super_generation(disk_super);
2105 find_oldest_super_backup(fs_info, generation);
2106
2107 /*
2108 * In the long term, we'll store the compression type in the super
2109 * block, and it'll be used for per file compression control.
2110 */
2111 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2112
2113 ret = btrfs_parse_options(tree_root, options);
2114 if (ret) {
2115 err = ret;
2116 goto fail_alloc;
2117 }
2118
2119 features = btrfs_super_incompat_flags(disk_super) &
2120 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2121 if (features) {
2122 printk(KERN_ERR "BTRFS: couldn't mount because of "
2123 "unsupported optional features (%Lx).\n",
2124 (unsigned long long)features);
2125 err = -EINVAL;
2126 goto fail_alloc;
2127 }
2128
2129 features = btrfs_super_incompat_flags(disk_super);
2130 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2131 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2132 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2133 btrfs_set_super_incompat_flags(disk_super, features);
2134
2135 features = btrfs_super_compat_ro_flags(disk_super) &
2136 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2137 if (!(sb->s_flags & MS_RDONLY) && features) {
2138 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2139 "unsupported option features (%Lx).\n",
2140 (unsigned long long)features);
2141 err = -EINVAL;
2142 goto fail_alloc;
2143 }
2144
2145 btrfs_init_workers(&fs_info->generic_worker,
2146 "genwork", 1, NULL);
2147
2148 btrfs_init_workers(&fs_info->workers, "worker",
2149 fs_info->thread_pool_size,
2150 &fs_info->generic_worker);
2151
2152 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2153 fs_info->thread_pool_size,
2154 &fs_info->generic_worker);
2155
2156 btrfs_init_workers(&fs_info->submit_workers, "submit",
2157 min_t(u64, fs_devices->num_devices,
2158 fs_info->thread_pool_size),
2159 &fs_info->generic_worker);
2160
2161 btrfs_init_workers(&fs_info->caching_workers, "cache",
2162 2, &fs_info->generic_worker);
2163
2164 /* a higher idle thresh on the submit workers makes it much more
2165 * likely that bios will be send down in a sane order to the
2166 * devices
2167 */
2168 fs_info->submit_workers.idle_thresh = 64;
2169
2170 fs_info->workers.idle_thresh = 16;
2171 fs_info->workers.ordered = 1;
2172
2173 fs_info->delalloc_workers.idle_thresh = 2;
2174 fs_info->delalloc_workers.ordered = 1;
2175
2176 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2177 &fs_info->generic_worker);
2178 btrfs_init_workers(&fs_info->endio_workers, "endio",
2179 fs_info->thread_pool_size,
2180 &fs_info->generic_worker);
2181 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2182 fs_info->thread_pool_size,
2183 &fs_info->generic_worker);
2184 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2185 "endio-meta-write", fs_info->thread_pool_size,
2186 &fs_info->generic_worker);
2187 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2188 fs_info->thread_pool_size,
2189 &fs_info->generic_worker);
2190 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2191 1, &fs_info->generic_worker);
2192 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2193 fs_info->thread_pool_size,
2194 &fs_info->generic_worker);
2195 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2196 fs_info->thread_pool_size,
2197 &fs_info->generic_worker);
2198
2199 /*
2200 * endios are largely parallel and should have a very
2201 * low idle thresh
2202 */
2203 fs_info->endio_workers.idle_thresh = 4;
2204 fs_info->endio_meta_workers.idle_thresh = 4;
2205
2206 fs_info->endio_write_workers.idle_thresh = 2;
2207 fs_info->endio_meta_write_workers.idle_thresh = 2;
2208 fs_info->readahead_workers.idle_thresh = 2;
2209
2210 /*
2211 * btrfs_start_workers can really only fail because of ENOMEM so just
2212 * return -ENOMEM if any of these fail.
2213 */
2214 ret = btrfs_start_workers(&fs_info->workers);
2215 ret |= btrfs_start_workers(&fs_info->generic_worker);
2216 ret |= btrfs_start_workers(&fs_info->submit_workers);
2217 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2218 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2219 ret |= btrfs_start_workers(&fs_info->endio_workers);
2220 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2221 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2222 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2223 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2224 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2225 ret |= btrfs_start_workers(&fs_info->caching_workers);
2226 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2227 if (ret) {
2228 ret = -ENOMEM;
2229 goto fail_sb_buffer;
2230 }
2231
2232 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2233 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2234 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2235
2236 nodesize = btrfs_super_nodesize(disk_super);
2237 leafsize = btrfs_super_leafsize(disk_super);
2238 sectorsize = btrfs_super_sectorsize(disk_super);
2239 stripesize = btrfs_super_stripesize(disk_super);
2240 tree_root->nodesize = nodesize;
2241 tree_root->leafsize = leafsize;
2242 tree_root->sectorsize = sectorsize;
2243 tree_root->stripesize = stripesize;
2244
2245 sb->s_blocksize = sectorsize;
2246 sb->s_blocksize_bits = blksize_bits(sectorsize);
2247
2248 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2249 sizeof(disk_super->magic))) {
2250 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2251 goto fail_sb_buffer;
2252 }
2253
2254 mutex_lock(&fs_info->chunk_mutex);
2255 ret = btrfs_read_sys_array(tree_root);
2256 mutex_unlock(&fs_info->chunk_mutex);
2257 if (ret) {
2258 printk(KERN_WARNING "btrfs: failed to read the system "
2259 "array on %s\n", sb->s_id);
2260 goto fail_sb_buffer;
2261 }
2262
2263 blocksize = btrfs_level_size(tree_root,
2264 btrfs_super_chunk_root_level(disk_super));
2265 generation = btrfs_super_chunk_root_generation(disk_super);
2266
2267 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2268 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2269
2270 chunk_root->node = read_tree_block(chunk_root,
2271 btrfs_super_chunk_root(disk_super),
2272 blocksize, generation);
2273 BUG_ON(!chunk_root->node);
2274 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2275 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2276 sb->s_id);
2277 goto fail_tree_roots;
2278 }
2279 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2280 chunk_root->commit_root = btrfs_root_node(chunk_root);
2281
2282 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2283 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2284 BTRFS_UUID_SIZE);
2285
2286 ret = btrfs_read_chunk_tree(chunk_root);
2287 if (ret) {
2288 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2289 sb->s_id);
2290 goto fail_tree_roots;
2291 }
2292
2293 btrfs_close_extra_devices(fs_devices);
2294
2295retry_root_backup:
2296 blocksize = btrfs_level_size(tree_root,
2297 btrfs_super_root_level(disk_super));
2298 generation = btrfs_super_generation(disk_super);
2299
2300 tree_root->node = read_tree_block(tree_root,
2301 btrfs_super_root(disk_super),
2302 blocksize, generation);
2303 if (!tree_root->node ||
2304 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2305 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2306 sb->s_id);
2307
2308 goto recovery_tree_root;
2309 }
2310
2311 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2312 tree_root->commit_root = btrfs_root_node(tree_root);
2313
2314 ret = find_and_setup_root(tree_root, fs_info,
2315 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2316 if (ret)
2317 goto recovery_tree_root;
2318 extent_root->track_dirty = 1;
2319
2320 ret = find_and_setup_root(tree_root, fs_info,
2321 BTRFS_DEV_TREE_OBJECTID, dev_root);
2322 if (ret)
2323 goto recovery_tree_root;
2324 dev_root->track_dirty = 1;
2325
2326 ret = find_and_setup_root(tree_root, fs_info,
2327 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2328 if (ret)
2329 goto recovery_tree_root;
2330
2331 csum_root->track_dirty = 1;
2332
2333 fs_info->generation = generation;
2334 fs_info->last_trans_committed = generation;
2335
2336 ret = btrfs_init_space_info(fs_info);
2337 if (ret) {
2338 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2339 goto fail_block_groups;
2340 }
2341
2342 ret = btrfs_read_block_groups(extent_root);
2343 if (ret) {
2344 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2345 goto fail_block_groups;
2346 }
2347
2348 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2349 "btrfs-cleaner");
2350 if (IS_ERR(fs_info->cleaner_kthread))
2351 goto fail_block_groups;
2352
2353 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2354 tree_root,
2355 "btrfs-transaction");
2356 if (IS_ERR(fs_info->transaction_kthread))
2357 goto fail_cleaner;
2358
2359 if (!btrfs_test_opt(tree_root, SSD) &&
2360 !btrfs_test_opt(tree_root, NOSSD) &&
2361 !fs_info->fs_devices->rotating) {
2362 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2363 "mode\n");
2364 btrfs_set_opt(fs_info->mount_opt, SSD);
2365 }
2366
2367#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2368 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2369 ret = btrfsic_mount(tree_root, fs_devices,
2370 btrfs_test_opt(tree_root,
2371 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2372 1 : 0,
2373 fs_info->check_integrity_print_mask);
2374 if (ret)
2375 printk(KERN_WARNING "btrfs: failed to initialize"
2376 " integrity check module %s\n", sb->s_id);
2377 }
2378#endif
2379
2380 /* do not make disk changes in broken FS */
2381 if (btrfs_super_log_root(disk_super) != 0 &&
2382 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2383 u64 bytenr = btrfs_super_log_root(disk_super);
2384
2385 if (fs_devices->rw_devices == 0) {
2386 printk(KERN_WARNING "Btrfs log replay required "
2387 "on RO media\n");
2388 err = -EIO;
2389 goto fail_trans_kthread;
2390 }
2391 blocksize =
2392 btrfs_level_size(tree_root,
2393 btrfs_super_log_root_level(disk_super));
2394
2395 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2396 if (!log_tree_root) {
2397 err = -ENOMEM;
2398 goto fail_trans_kthread;
2399 }
2400
2401 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2402 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2403
2404 log_tree_root->node = read_tree_block(tree_root, bytenr,
2405 blocksize,
2406 generation + 1);
2407 ret = btrfs_recover_log_trees(log_tree_root);
2408 BUG_ON(ret);
2409
2410 if (sb->s_flags & MS_RDONLY) {
2411 ret = btrfs_commit_super(tree_root);
2412 BUG_ON(ret);
2413 }
2414 }
2415
2416 ret = btrfs_find_orphan_roots(tree_root);
2417 BUG_ON(ret);
2418
2419 if (!(sb->s_flags & MS_RDONLY)) {
2420 ret = btrfs_cleanup_fs_roots(fs_info);
2421 BUG_ON(ret);
2422
2423 ret = btrfs_recover_relocation(tree_root);
2424 if (ret < 0) {
2425 printk(KERN_WARNING
2426 "btrfs: failed to recover relocation\n");
2427 err = -EINVAL;
2428 goto fail_trans_kthread;
2429 }
2430 }
2431
2432 location.objectid = BTRFS_FS_TREE_OBJECTID;
2433 location.type = BTRFS_ROOT_ITEM_KEY;
2434 location.offset = (u64)-1;
2435
2436 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2437 if (!fs_info->fs_root)
2438 goto fail_trans_kthread;
2439 if (IS_ERR(fs_info->fs_root)) {
2440 err = PTR_ERR(fs_info->fs_root);
2441 goto fail_trans_kthread;
2442 }
2443
2444 if (!(sb->s_flags & MS_RDONLY)) {
2445 down_read(&fs_info->cleanup_work_sem);
2446 err = btrfs_orphan_cleanup(fs_info->fs_root);
2447 if (!err)
2448 err = btrfs_orphan_cleanup(fs_info->tree_root);
2449 up_read(&fs_info->cleanup_work_sem);
2450
2451 if (!err)
2452 err = btrfs_recover_balance(fs_info->tree_root);
2453
2454 if (err) {
2455 close_ctree(tree_root);
2456 return ERR_PTR(err);
2457 }
2458 }
2459
2460 return tree_root;
2461
2462fail_trans_kthread:
2463 kthread_stop(fs_info->transaction_kthread);
2464fail_cleaner:
2465 kthread_stop(fs_info->cleaner_kthread);
2466
2467 /*
2468 * make sure we're done with the btree inode before we stop our
2469 * kthreads
2470 */
2471 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2472 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2473
2474fail_block_groups:
2475 btrfs_free_block_groups(fs_info);
2476
2477fail_tree_roots:
2478 free_root_pointers(fs_info, 1);
2479
2480fail_sb_buffer:
2481 btrfs_stop_workers(&fs_info->generic_worker);
2482 btrfs_stop_workers(&fs_info->readahead_workers);
2483 btrfs_stop_workers(&fs_info->fixup_workers);
2484 btrfs_stop_workers(&fs_info->delalloc_workers);
2485 btrfs_stop_workers(&fs_info->workers);
2486 btrfs_stop_workers(&fs_info->endio_workers);
2487 btrfs_stop_workers(&fs_info->endio_meta_workers);
2488 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2489 btrfs_stop_workers(&fs_info->endio_write_workers);
2490 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2491 btrfs_stop_workers(&fs_info->submit_workers);
2492 btrfs_stop_workers(&fs_info->delayed_workers);
2493 btrfs_stop_workers(&fs_info->caching_workers);
2494fail_alloc:
2495fail_iput:
2496 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2497
2498 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2499 iput(fs_info->btree_inode);
2500fail_bdi:
2501 bdi_destroy(&fs_info->bdi);
2502fail_srcu:
2503 cleanup_srcu_struct(&fs_info->subvol_srcu);
2504fail:
2505 btrfs_close_devices(fs_info->fs_devices);
2506 free_fs_info(fs_info);
2507 return ERR_PTR(err);
2508
2509recovery_tree_root:
2510 if (!btrfs_test_opt(tree_root, RECOVERY))
2511 goto fail_tree_roots;
2512
2513 free_root_pointers(fs_info, 0);
2514
2515 /* don't use the log in recovery mode, it won't be valid */
2516 btrfs_set_super_log_root(disk_super, 0);
2517
2518 /* we can't trust the free space cache either */
2519 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2520
2521 ret = next_root_backup(fs_info, fs_info->super_copy,
2522 &num_backups_tried, &backup_index);
2523 if (ret == -1)
2524 goto fail_block_groups;
2525 goto retry_root_backup;
2526}
2527
2528static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2529{
2530 char b[BDEVNAME_SIZE];
2531
2532 if (uptodate) {
2533 set_buffer_uptodate(bh);
2534 } else {
2535 printk_ratelimited(KERN_WARNING "lost page write due to "
2536 "I/O error on %s\n",
2537 bdevname(bh->b_bdev, b));
2538 /* note, we dont' set_buffer_write_io_error because we have
2539 * our own ways of dealing with the IO errors
2540 */
2541 clear_buffer_uptodate(bh);
2542 }
2543 unlock_buffer(bh);
2544 put_bh(bh);
2545}
2546
2547struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2548{
2549 struct buffer_head *bh;
2550 struct buffer_head *latest = NULL;
2551 struct btrfs_super_block *super;
2552 int i;
2553 u64 transid = 0;
2554 u64 bytenr;
2555
2556 /* we would like to check all the supers, but that would make
2557 * a btrfs mount succeed after a mkfs from a different FS.
2558 * So, we need to add a special mount option to scan for
2559 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2560 */
2561 for (i = 0; i < 1; i++) {
2562 bytenr = btrfs_sb_offset(i);
2563 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2564 break;
2565 bh = __bread(bdev, bytenr / 4096, 4096);
2566 if (!bh)
2567 continue;
2568
2569 super = (struct btrfs_super_block *)bh->b_data;
2570 if (btrfs_super_bytenr(super) != bytenr ||
2571 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2572 sizeof(super->magic))) {
2573 brelse(bh);
2574 continue;
2575 }
2576
2577 if (!latest || btrfs_super_generation(super) > transid) {
2578 brelse(latest);
2579 latest = bh;
2580 transid = btrfs_super_generation(super);
2581 } else {
2582 brelse(bh);
2583 }
2584 }
2585 return latest;
2586}
2587
2588/*
2589 * this should be called twice, once with wait == 0 and
2590 * once with wait == 1. When wait == 0 is done, all the buffer heads
2591 * we write are pinned.
2592 *
2593 * They are released when wait == 1 is done.
2594 * max_mirrors must be the same for both runs, and it indicates how
2595 * many supers on this one device should be written.
2596 *
2597 * max_mirrors == 0 means to write them all.
2598 */
2599static int write_dev_supers(struct btrfs_device *device,
2600 struct btrfs_super_block *sb,
2601 int do_barriers, int wait, int max_mirrors)
2602{
2603 struct buffer_head *bh;
2604 int i;
2605 int ret;
2606 int errors = 0;
2607 u32 crc;
2608 u64 bytenr;
2609
2610 if (max_mirrors == 0)
2611 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2612
2613 for (i = 0; i < max_mirrors; i++) {
2614 bytenr = btrfs_sb_offset(i);
2615 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2616 break;
2617
2618 if (wait) {
2619 bh = __find_get_block(device->bdev, bytenr / 4096,
2620 BTRFS_SUPER_INFO_SIZE);
2621 BUG_ON(!bh);
2622 wait_on_buffer(bh);
2623 if (!buffer_uptodate(bh))
2624 errors++;
2625
2626 /* drop our reference */
2627 brelse(bh);
2628
2629 /* drop the reference from the wait == 0 run */
2630 brelse(bh);
2631 continue;
2632 } else {
2633 btrfs_set_super_bytenr(sb, bytenr);
2634
2635 crc = ~(u32)0;
2636 crc = btrfs_csum_data(NULL, (char *)sb +
2637 BTRFS_CSUM_SIZE, crc,
2638 BTRFS_SUPER_INFO_SIZE -
2639 BTRFS_CSUM_SIZE);
2640 btrfs_csum_final(crc, sb->csum);
2641
2642 /*
2643 * one reference for us, and we leave it for the
2644 * caller
2645 */
2646 bh = __getblk(device->bdev, bytenr / 4096,
2647 BTRFS_SUPER_INFO_SIZE);
2648 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2649
2650 /* one reference for submit_bh */
2651 get_bh(bh);
2652
2653 set_buffer_uptodate(bh);
2654 lock_buffer(bh);
2655 bh->b_end_io = btrfs_end_buffer_write_sync;
2656 }
2657
2658 /*
2659 * we fua the first super. The others we allow
2660 * to go down lazy.
2661 */
2662 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2663 if (ret)
2664 errors++;
2665 }
2666 return errors < i ? 0 : -1;
2667}
2668
2669/*
2670 * endio for the write_dev_flush, this will wake anyone waiting
2671 * for the barrier when it is done
2672 */
2673static void btrfs_end_empty_barrier(struct bio *bio, int err)
2674{
2675 if (err) {
2676 if (err == -EOPNOTSUPP)
2677 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2678 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2679 }
2680 if (bio->bi_private)
2681 complete(bio->bi_private);
2682 bio_put(bio);
2683}
2684
2685/*
2686 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2687 * sent down. With wait == 1, it waits for the previous flush.
2688 *
2689 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2690 * capable
2691 */
2692static int write_dev_flush(struct btrfs_device *device, int wait)
2693{
2694 struct bio *bio;
2695 int ret = 0;
2696
2697 if (device->nobarriers)
2698 return 0;
2699
2700 if (wait) {
2701 bio = device->flush_bio;
2702 if (!bio)
2703 return 0;
2704
2705 wait_for_completion(&device->flush_wait);
2706
2707 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2708 printk("btrfs: disabling barriers on dev %s\n",
2709 device->name);
2710 device->nobarriers = 1;
2711 }
2712 if (!bio_flagged(bio, BIO_UPTODATE)) {
2713 ret = -EIO;
2714 }
2715
2716 /* drop the reference from the wait == 0 run */
2717 bio_put(bio);
2718 device->flush_bio = NULL;
2719
2720 return ret;
2721 }
2722
2723 /*
2724 * one reference for us, and we leave it for the
2725 * caller
2726 */
2727 device->flush_bio = NULL;;
2728 bio = bio_alloc(GFP_NOFS, 0);
2729 if (!bio)
2730 return -ENOMEM;
2731
2732 bio->bi_end_io = btrfs_end_empty_barrier;
2733 bio->bi_bdev = device->bdev;
2734 init_completion(&device->flush_wait);
2735 bio->bi_private = &device->flush_wait;
2736 device->flush_bio = bio;
2737
2738 bio_get(bio);
2739 btrfsic_submit_bio(WRITE_FLUSH, bio);
2740
2741 return 0;
2742}
2743
2744/*
2745 * send an empty flush down to each device in parallel,
2746 * then wait for them
2747 */
2748static int barrier_all_devices(struct btrfs_fs_info *info)
2749{
2750 struct list_head *head;
2751 struct btrfs_device *dev;
2752 int errors = 0;
2753 int ret;
2754
2755 /* send down all the barriers */
2756 head = &info->fs_devices->devices;
2757 list_for_each_entry_rcu(dev, head, dev_list) {
2758 if (!dev->bdev) {
2759 errors++;
2760 continue;
2761 }
2762 if (!dev->in_fs_metadata || !dev->writeable)
2763 continue;
2764
2765 ret = write_dev_flush(dev, 0);
2766 if (ret)
2767 errors++;
2768 }
2769
2770 /* wait for all the barriers */
2771 list_for_each_entry_rcu(dev, head, dev_list) {
2772 if (!dev->bdev) {
2773 errors++;
2774 continue;
2775 }
2776 if (!dev->in_fs_metadata || !dev->writeable)
2777 continue;
2778
2779 ret = write_dev_flush(dev, 1);
2780 if (ret)
2781 errors++;
2782 }
2783 if (errors)
2784 return -EIO;
2785 return 0;
2786}
2787
2788int write_all_supers(struct btrfs_root *root, int max_mirrors)
2789{
2790 struct list_head *head;
2791 struct btrfs_device *dev;
2792 struct btrfs_super_block *sb;
2793 struct btrfs_dev_item *dev_item;
2794 int ret;
2795 int do_barriers;
2796 int max_errors;
2797 int total_errors = 0;
2798 u64 flags;
2799
2800 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2801 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2802 backup_super_roots(root->fs_info);
2803
2804 sb = root->fs_info->super_for_commit;
2805 dev_item = &sb->dev_item;
2806
2807 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2808 head = &root->fs_info->fs_devices->devices;
2809
2810 if (do_barriers)
2811 barrier_all_devices(root->fs_info);
2812
2813 list_for_each_entry_rcu(dev, head, dev_list) {
2814 if (!dev->bdev) {
2815 total_errors++;
2816 continue;
2817 }
2818 if (!dev->in_fs_metadata || !dev->writeable)
2819 continue;
2820
2821 btrfs_set_stack_device_generation(dev_item, 0);
2822 btrfs_set_stack_device_type(dev_item, dev->type);
2823 btrfs_set_stack_device_id(dev_item, dev->devid);
2824 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2825 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2826 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2827 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2828 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2829 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2830 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2831
2832 flags = btrfs_super_flags(sb);
2833 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2834
2835 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2836 if (ret)
2837 total_errors++;
2838 }
2839 if (total_errors > max_errors) {
2840 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2841 total_errors);
2842 BUG();
2843 }
2844
2845 total_errors = 0;
2846 list_for_each_entry_rcu(dev, head, dev_list) {
2847 if (!dev->bdev)
2848 continue;
2849 if (!dev->in_fs_metadata || !dev->writeable)
2850 continue;
2851
2852 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2853 if (ret)
2854 total_errors++;
2855 }
2856 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2857 if (total_errors > max_errors) {
2858 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2859 total_errors);
2860 BUG();
2861 }
2862 return 0;
2863}
2864
2865int write_ctree_super(struct btrfs_trans_handle *trans,
2866 struct btrfs_root *root, int max_mirrors)
2867{
2868 int ret;
2869
2870 ret = write_all_supers(root, max_mirrors);
2871 return ret;
2872}
2873
2874int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2875{
2876 spin_lock(&fs_info->fs_roots_radix_lock);
2877 radix_tree_delete(&fs_info->fs_roots_radix,
2878 (unsigned long)root->root_key.objectid);
2879 spin_unlock(&fs_info->fs_roots_radix_lock);
2880
2881 if (btrfs_root_refs(&root->root_item) == 0)
2882 synchronize_srcu(&fs_info->subvol_srcu);
2883
2884 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2885 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2886 free_fs_root(root);
2887 return 0;
2888}
2889
2890static void free_fs_root(struct btrfs_root *root)
2891{
2892 iput(root->cache_inode);
2893 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2894 if (root->anon_dev)
2895 free_anon_bdev(root->anon_dev);
2896 free_extent_buffer(root->node);
2897 free_extent_buffer(root->commit_root);
2898 kfree(root->free_ino_ctl);
2899 kfree(root->free_ino_pinned);
2900 kfree(root->name);
2901 kfree(root);
2902}
2903
2904static int del_fs_roots(struct btrfs_fs_info *fs_info)
2905{
2906 int ret;
2907 struct btrfs_root *gang[8];
2908 int i;
2909
2910 while (!list_empty(&fs_info->dead_roots)) {
2911 gang[0] = list_entry(fs_info->dead_roots.next,
2912 struct btrfs_root, root_list);
2913 list_del(&gang[0]->root_list);
2914
2915 if (gang[0]->in_radix) {
2916 btrfs_free_fs_root(fs_info, gang[0]);
2917 } else {
2918 free_extent_buffer(gang[0]->node);
2919 free_extent_buffer(gang[0]->commit_root);
2920 kfree(gang[0]);
2921 }
2922 }
2923
2924 while (1) {
2925 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2926 (void **)gang, 0,
2927 ARRAY_SIZE(gang));
2928 if (!ret)
2929 break;
2930 for (i = 0; i < ret; i++)
2931 btrfs_free_fs_root(fs_info, gang[i]);
2932 }
2933 return 0;
2934}
2935
2936int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2937{
2938 u64 root_objectid = 0;
2939 struct btrfs_root *gang[8];
2940 int i;
2941 int ret;
2942
2943 while (1) {
2944 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2945 (void **)gang, root_objectid,
2946 ARRAY_SIZE(gang));
2947 if (!ret)
2948 break;
2949
2950 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2951 for (i = 0; i < ret; i++) {
2952 int err;
2953
2954 root_objectid = gang[i]->root_key.objectid;
2955 err = btrfs_orphan_cleanup(gang[i]);
2956 if (err)
2957 return err;
2958 }
2959 root_objectid++;
2960 }
2961 return 0;
2962}
2963
2964int btrfs_commit_super(struct btrfs_root *root)
2965{
2966 struct btrfs_trans_handle *trans;
2967 int ret;
2968
2969 mutex_lock(&root->fs_info->cleaner_mutex);
2970 btrfs_run_delayed_iputs(root);
2971 btrfs_clean_old_snapshots(root);
2972 mutex_unlock(&root->fs_info->cleaner_mutex);
2973
2974 /* wait until ongoing cleanup work done */
2975 down_write(&root->fs_info->cleanup_work_sem);
2976 up_write(&root->fs_info->cleanup_work_sem);
2977
2978 trans = btrfs_join_transaction(root);
2979 if (IS_ERR(trans))
2980 return PTR_ERR(trans);
2981 ret = btrfs_commit_transaction(trans, root);
2982 BUG_ON(ret);
2983 /* run commit again to drop the original snapshot */
2984 trans = btrfs_join_transaction(root);
2985 if (IS_ERR(trans))
2986 return PTR_ERR(trans);
2987 btrfs_commit_transaction(trans, root);
2988 ret = btrfs_write_and_wait_transaction(NULL, root);
2989 BUG_ON(ret);
2990
2991 ret = write_ctree_super(NULL, root, 0);
2992 return ret;
2993}
2994
2995int close_ctree(struct btrfs_root *root)
2996{
2997 struct btrfs_fs_info *fs_info = root->fs_info;
2998 int ret;
2999
3000 fs_info->closing = 1;
3001 smp_mb();
3002
3003 /* pause restriper - we want to resume on mount */
3004 btrfs_pause_balance(root->fs_info);
3005
3006 btrfs_scrub_cancel(root);
3007
3008 /* wait for any defraggers to finish */
3009 wait_event(fs_info->transaction_wait,
3010 (atomic_read(&fs_info->defrag_running) == 0));
3011
3012 /* clear out the rbtree of defraggable inodes */
3013 btrfs_run_defrag_inodes(root->fs_info);
3014
3015 /*
3016 * Here come 2 situations when btrfs is broken to flip readonly:
3017 *
3018 * 1. when btrfs flips readonly somewhere else before
3019 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
3020 * and btrfs will skip to write sb directly to keep
3021 * ERROR state on disk.
3022 *
3023 * 2. when btrfs flips readonly just in btrfs_commit_super,
3024 * and in such case, btrfs cannot write sb via btrfs_commit_super,
3025 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3026 * btrfs will cleanup all FS resources first and write sb then.
3027 */
3028 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3029 ret = btrfs_commit_super(root);
3030 if (ret)
3031 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3032 }
3033
3034 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3035 ret = btrfs_error_commit_super(root);
3036 if (ret)
3037 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3038 }
3039
3040 btrfs_put_block_group_cache(fs_info);
3041
3042 kthread_stop(root->fs_info->transaction_kthread);
3043 kthread_stop(root->fs_info->cleaner_kthread);
3044
3045 fs_info->closing = 2;
3046 smp_mb();
3047
3048 if (fs_info->delalloc_bytes) {
3049 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3050 (unsigned long long)fs_info->delalloc_bytes);
3051 }
3052 if (fs_info->total_ref_cache_size) {
3053 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3054 (unsigned long long)fs_info->total_ref_cache_size);
3055 }
3056
3057 free_extent_buffer(fs_info->extent_root->node);
3058 free_extent_buffer(fs_info->extent_root->commit_root);
3059 free_extent_buffer(fs_info->tree_root->node);
3060 free_extent_buffer(fs_info->tree_root->commit_root);
3061 free_extent_buffer(root->fs_info->chunk_root->node);
3062 free_extent_buffer(root->fs_info->chunk_root->commit_root);
3063 free_extent_buffer(root->fs_info->dev_root->node);
3064 free_extent_buffer(root->fs_info->dev_root->commit_root);
3065 free_extent_buffer(root->fs_info->csum_root->node);
3066 free_extent_buffer(root->fs_info->csum_root->commit_root);
3067
3068 btrfs_free_block_groups(root->fs_info);
3069
3070 del_fs_roots(fs_info);
3071
3072 iput(fs_info->btree_inode);
3073
3074 btrfs_stop_workers(&fs_info->generic_worker);
3075 btrfs_stop_workers(&fs_info->fixup_workers);
3076 btrfs_stop_workers(&fs_info->delalloc_workers);
3077 btrfs_stop_workers(&fs_info->workers);
3078 btrfs_stop_workers(&fs_info->endio_workers);
3079 btrfs_stop_workers(&fs_info->endio_meta_workers);
3080 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3081 btrfs_stop_workers(&fs_info->endio_write_workers);
3082 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3083 btrfs_stop_workers(&fs_info->submit_workers);
3084 btrfs_stop_workers(&fs_info->delayed_workers);
3085 btrfs_stop_workers(&fs_info->caching_workers);
3086 btrfs_stop_workers(&fs_info->readahead_workers);
3087
3088#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3089 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3090 btrfsic_unmount(root, fs_info->fs_devices);
3091#endif
3092
3093 btrfs_close_devices(fs_info->fs_devices);
3094 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3095
3096 bdi_destroy(&fs_info->bdi);
3097 cleanup_srcu_struct(&fs_info->subvol_srcu);
3098
3099 free_fs_info(fs_info);
3100
3101 return 0;
3102}
3103
3104int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
3105{
3106 int ret;
3107 struct inode *btree_inode = buf->first_page->mapping->host;
3108
3109 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
3110 NULL);
3111 if (!ret)
3112 return ret;
3113
3114 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3115 parent_transid);
3116 return !ret;
3117}
3118
3119int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3120{
3121 struct inode *btree_inode = buf->first_page->mapping->host;
3122 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
3123 buf);
3124}
3125
3126void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3127{
3128 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3129 u64 transid = btrfs_header_generation(buf);
3130 struct inode *btree_inode = root->fs_info->btree_inode;
3131 int was_dirty;
3132
3133 btrfs_assert_tree_locked(buf);
3134 if (transid != root->fs_info->generation) {
3135 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3136 "found %llu running %llu\n",
3137 (unsigned long long)buf->start,
3138 (unsigned long long)transid,
3139 (unsigned long long)root->fs_info->generation);
3140 WARN_ON(1);
3141 }
3142 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
3143 buf);
3144 if (!was_dirty) {
3145 spin_lock(&root->fs_info->delalloc_lock);
3146 root->fs_info->dirty_metadata_bytes += buf->len;
3147 spin_unlock(&root->fs_info->delalloc_lock);
3148 }
3149}
3150
3151void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3152{
3153 /*
3154 * looks as though older kernels can get into trouble with
3155 * this code, they end up stuck in balance_dirty_pages forever
3156 */
3157 u64 num_dirty;
3158 unsigned long thresh = 32 * 1024 * 1024;
3159
3160 if (current->flags & PF_MEMALLOC)
3161 return;
3162
3163 btrfs_balance_delayed_items(root);
3164
3165 num_dirty = root->fs_info->dirty_metadata_bytes;
3166
3167 if (num_dirty > thresh) {
3168 balance_dirty_pages_ratelimited_nr(
3169 root->fs_info->btree_inode->i_mapping, 1);
3170 }
3171 return;
3172}
3173
3174void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3175{
3176 /*
3177 * looks as though older kernels can get into trouble with
3178 * this code, they end up stuck in balance_dirty_pages forever
3179 */
3180 u64 num_dirty;
3181 unsigned long thresh = 32 * 1024 * 1024;
3182
3183 if (current->flags & PF_MEMALLOC)
3184 return;
3185
3186 num_dirty = root->fs_info->dirty_metadata_bytes;
3187
3188 if (num_dirty > thresh) {
3189 balance_dirty_pages_ratelimited_nr(
3190 root->fs_info->btree_inode->i_mapping, 1);
3191 }
3192 return;
3193}
3194
3195int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3196{
3197 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3198 int ret;
3199 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3200 if (ret == 0)
3201 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
3202 return ret;
3203}
3204
3205static int btree_lock_page_hook(struct page *page, void *data,
3206 void (*flush_fn)(void *))
3207{
3208 struct inode *inode = page->mapping->host;
3209 struct btrfs_root *root = BTRFS_I(inode)->root;
3210 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3211 struct extent_buffer *eb;
3212 unsigned long len;
3213 u64 bytenr = page_offset(page);
3214
3215 if (page->private == EXTENT_PAGE_PRIVATE)
3216 goto out;
3217
3218 len = page->private >> 2;
3219 eb = find_extent_buffer(io_tree, bytenr, len);
3220 if (!eb)
3221 goto out;
3222
3223 if (!btrfs_try_tree_write_lock(eb)) {
3224 flush_fn(data);
3225 btrfs_tree_lock(eb);
3226 }
3227 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3228
3229 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3230 spin_lock(&root->fs_info->delalloc_lock);
3231 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3232 root->fs_info->dirty_metadata_bytes -= eb->len;
3233 else
3234 WARN_ON(1);
3235 spin_unlock(&root->fs_info->delalloc_lock);
3236 }
3237
3238 btrfs_tree_unlock(eb);
3239 free_extent_buffer(eb);
3240out:
3241 if (!trylock_page(page)) {
3242 flush_fn(data);
3243 lock_page(page);
3244 }
3245 return 0;
3246}
3247
3248static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3249 int read_only)
3250{
3251 if (read_only)
3252 return;
3253
3254 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3255 printk(KERN_WARNING "warning: mount fs with errors, "
3256 "running btrfsck is recommended\n");
3257}
3258
3259int btrfs_error_commit_super(struct btrfs_root *root)
3260{
3261 int ret;
3262
3263 mutex_lock(&root->fs_info->cleaner_mutex);
3264 btrfs_run_delayed_iputs(root);
3265 mutex_unlock(&root->fs_info->cleaner_mutex);
3266
3267 down_write(&root->fs_info->cleanup_work_sem);
3268 up_write(&root->fs_info->cleanup_work_sem);
3269
3270 /* cleanup FS via transaction */
3271 btrfs_cleanup_transaction(root);
3272
3273 ret = write_ctree_super(NULL, root, 0);
3274
3275 return ret;
3276}
3277
3278static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
3279{
3280 struct btrfs_inode *btrfs_inode;
3281 struct list_head splice;
3282
3283 INIT_LIST_HEAD(&splice);
3284
3285 mutex_lock(&root->fs_info->ordered_operations_mutex);
3286 spin_lock(&root->fs_info->ordered_extent_lock);
3287
3288 list_splice_init(&root->fs_info->ordered_operations, &splice);
3289 while (!list_empty(&splice)) {
3290 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3291 ordered_operations);
3292
3293 list_del_init(&btrfs_inode->ordered_operations);
3294
3295 btrfs_invalidate_inodes(btrfs_inode->root);
3296 }
3297
3298 spin_unlock(&root->fs_info->ordered_extent_lock);
3299 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3300
3301 return 0;
3302}
3303
3304static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
3305{
3306 struct list_head splice;
3307 struct btrfs_ordered_extent *ordered;
3308 struct inode *inode;
3309
3310 INIT_LIST_HEAD(&splice);
3311
3312 spin_lock(&root->fs_info->ordered_extent_lock);
3313
3314 list_splice_init(&root->fs_info->ordered_extents, &splice);
3315 while (!list_empty(&splice)) {
3316 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3317 root_extent_list);
3318
3319 list_del_init(&ordered->root_extent_list);
3320 atomic_inc(&ordered->refs);
3321
3322 /* the inode may be getting freed (in sys_unlink path). */
3323 inode = igrab(ordered->inode);
3324
3325 spin_unlock(&root->fs_info->ordered_extent_lock);
3326 if (inode)
3327 iput(inode);
3328
3329 atomic_set(&ordered->refs, 1);
3330 btrfs_put_ordered_extent(ordered);
3331
3332 spin_lock(&root->fs_info->ordered_extent_lock);
3333 }
3334
3335 spin_unlock(&root->fs_info->ordered_extent_lock);
3336
3337 return 0;
3338}
3339
3340static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3341 struct btrfs_root *root)
3342{
3343 struct rb_node *node;
3344 struct btrfs_delayed_ref_root *delayed_refs;
3345 struct btrfs_delayed_ref_node *ref;
3346 int ret = 0;
3347
3348 delayed_refs = &trans->delayed_refs;
3349
3350 spin_lock(&delayed_refs->lock);
3351 if (delayed_refs->num_entries == 0) {
3352 spin_unlock(&delayed_refs->lock);
3353 printk(KERN_INFO "delayed_refs has NO entry\n");
3354 return ret;
3355 }
3356
3357 node = rb_first(&delayed_refs->root);
3358 while (node) {
3359 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3360 node = rb_next(node);
3361
3362 ref->in_tree = 0;
3363 rb_erase(&ref->rb_node, &delayed_refs->root);
3364 delayed_refs->num_entries--;
3365
3366 atomic_set(&ref->refs, 1);
3367 if (btrfs_delayed_ref_is_head(ref)) {
3368 struct btrfs_delayed_ref_head *head;
3369
3370 head = btrfs_delayed_node_to_head(ref);
3371 mutex_lock(&head->mutex);
3372 kfree(head->extent_op);
3373 delayed_refs->num_heads--;
3374 if (list_empty(&head->cluster))
3375 delayed_refs->num_heads_ready--;
3376 list_del_init(&head->cluster);
3377 mutex_unlock(&head->mutex);
3378 }
3379
3380 spin_unlock(&delayed_refs->lock);
3381 btrfs_put_delayed_ref(ref);
3382
3383 cond_resched();
3384 spin_lock(&delayed_refs->lock);
3385 }
3386
3387 spin_unlock(&delayed_refs->lock);
3388
3389 return ret;
3390}
3391
3392static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3393{
3394 struct btrfs_pending_snapshot *snapshot;
3395 struct list_head splice;
3396
3397 INIT_LIST_HEAD(&splice);
3398
3399 list_splice_init(&t->pending_snapshots, &splice);
3400
3401 while (!list_empty(&splice)) {
3402 snapshot = list_entry(splice.next,
3403 struct btrfs_pending_snapshot,
3404 list);
3405
3406 list_del_init(&snapshot->list);
3407
3408 kfree(snapshot);
3409 }
3410
3411 return 0;
3412}
3413
3414static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3415{
3416 struct btrfs_inode *btrfs_inode;
3417 struct list_head splice;
3418
3419 INIT_LIST_HEAD(&splice);
3420
3421 spin_lock(&root->fs_info->delalloc_lock);
3422 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3423
3424 while (!list_empty(&splice)) {
3425 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3426 delalloc_inodes);
3427
3428 list_del_init(&btrfs_inode->delalloc_inodes);
3429
3430 btrfs_invalidate_inodes(btrfs_inode->root);
3431 }
3432
3433 spin_unlock(&root->fs_info->delalloc_lock);
3434
3435 return 0;
3436}
3437
3438static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3439 struct extent_io_tree *dirty_pages,
3440 int mark)
3441{
3442 int ret;
3443 struct page *page;
3444 struct inode *btree_inode = root->fs_info->btree_inode;
3445 struct extent_buffer *eb;
3446 u64 start = 0;
3447 u64 end;
3448 u64 offset;
3449 unsigned long index;
3450
3451 while (1) {
3452 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3453 mark);
3454 if (ret)
3455 break;
3456
3457 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3458 while (start <= end) {
3459 index = start >> PAGE_CACHE_SHIFT;
3460 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3461 page = find_get_page(btree_inode->i_mapping, index);
3462 if (!page)
3463 continue;
3464 offset = page_offset(page);
3465
3466 spin_lock(&dirty_pages->buffer_lock);
3467 eb = radix_tree_lookup(
3468 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3469 offset >> PAGE_CACHE_SHIFT);
3470 spin_unlock(&dirty_pages->buffer_lock);
3471 if (eb) {
3472 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3473 &eb->bflags);
3474 atomic_set(&eb->refs, 1);
3475 }
3476 if (PageWriteback(page))
3477 end_page_writeback(page);
3478
3479 lock_page(page);
3480 if (PageDirty(page)) {
3481 clear_page_dirty_for_io(page);
3482 spin_lock_irq(&page->mapping->tree_lock);
3483 radix_tree_tag_clear(&page->mapping->page_tree,
3484 page_index(page),
3485 PAGECACHE_TAG_DIRTY);
3486 spin_unlock_irq(&page->mapping->tree_lock);
3487 }
3488
3489 page->mapping->a_ops->invalidatepage(page, 0);
3490 unlock_page(page);
3491 }
3492 }
3493
3494 return ret;
3495}
3496
3497static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3498 struct extent_io_tree *pinned_extents)
3499{
3500 struct extent_io_tree *unpin;
3501 u64 start;
3502 u64 end;
3503 int ret;
3504
3505 unpin = pinned_extents;
3506 while (1) {
3507 ret = find_first_extent_bit(unpin, 0, &start, &end,
3508 EXTENT_DIRTY);
3509 if (ret)
3510 break;
3511
3512 /* opt_discard */
3513 if (btrfs_test_opt(root, DISCARD))
3514 ret = btrfs_error_discard_extent(root, start,
3515 end + 1 - start,
3516 NULL);
3517
3518 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3519 btrfs_error_unpin_extent_range(root, start, end);
3520 cond_resched();
3521 }
3522
3523 return 0;
3524}
3525
3526static int btrfs_cleanup_transaction(struct btrfs_root *root)
3527{
3528 struct btrfs_transaction *t;
3529 LIST_HEAD(list);
3530
3531 WARN_ON(1);
3532
3533 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3534
3535 spin_lock(&root->fs_info->trans_lock);
3536 list_splice_init(&root->fs_info->trans_list, &list);
3537 root->fs_info->trans_no_join = 1;
3538 spin_unlock(&root->fs_info->trans_lock);
3539
3540 while (!list_empty(&list)) {
3541 t = list_entry(list.next, struct btrfs_transaction, list);
3542 if (!t)
3543 break;
3544
3545 btrfs_destroy_ordered_operations(root);
3546
3547 btrfs_destroy_ordered_extents(root);
3548
3549 btrfs_destroy_delayed_refs(t, root);
3550
3551 btrfs_block_rsv_release(root,
3552 &root->fs_info->trans_block_rsv,
3553 t->dirty_pages.dirty_bytes);
3554
3555 /* FIXME: cleanup wait for commit */
3556 t->in_commit = 1;
3557 t->blocked = 1;
3558 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3559 wake_up(&root->fs_info->transaction_blocked_wait);
3560
3561 t->blocked = 0;
3562 if (waitqueue_active(&root->fs_info->transaction_wait))
3563 wake_up(&root->fs_info->transaction_wait);
3564
3565 t->commit_done = 1;
3566 if (waitqueue_active(&t->commit_wait))
3567 wake_up(&t->commit_wait);
3568
3569 btrfs_destroy_pending_snapshots(t);
3570
3571 btrfs_destroy_delalloc_inodes(root);
3572
3573 spin_lock(&root->fs_info->trans_lock);
3574 root->fs_info->running_transaction = NULL;
3575 spin_unlock(&root->fs_info->trans_lock);
3576
3577 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3578 EXTENT_DIRTY);
3579
3580 btrfs_destroy_pinned_extent(root,
3581 root->fs_info->pinned_extents);
3582
3583 atomic_set(&t->use_count, 0);
3584 list_del_init(&t->list);
3585 memset(t, 0, sizeof(*t));
3586 kmem_cache_free(btrfs_transaction_cachep, t);
3587 }
3588
3589 spin_lock(&root->fs_info->trans_lock);
3590 root->fs_info->trans_no_join = 0;
3591 spin_unlock(&root->fs_info->trans_lock);
3592 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3593
3594 return 0;
3595}
3596
3597static struct extent_io_ops btree_extent_io_ops = {
3598 .write_cache_pages_lock_hook = btree_lock_page_hook,
3599 .readpage_end_io_hook = btree_readpage_end_io_hook,
3600 .readpage_io_failed_hook = btree_io_failed_hook,
3601 .submit_bio_hook = btree_submit_bio_hook,
3602 /* note we're sharing with inode.c for the merge bio hook */
3603 .merge_bio_hook = btrfs_merge_bio_hook,
3604};
Linux kernel - Deliverables
This page took 0.036788 seconds and 5 git commands to generate.