projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: only use the existing eb if it's count isn't 0
[linux-2.6-block.git] / fs / btrfs / disk-io.c
... / ...
CommitLineData
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#include <linux/fs.h>
20#include <linux/blkdev.h>
21#include <linux/scatterlist.h>
22#include <linux/swap.h>
23#include <linux/radix-tree.h>
24#include <linux/writeback.h>
25#include <linux/buffer_head.h>
26#include <linux/workqueue.h>
27#include <linux/kthread.h>
28#include <linux/freezer.h>
29#include <linux/crc32c.h>
30#include <linux/slab.h>
31#include <linux/migrate.h>
32#include <linux/ratelimit.h>
33#include <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 && !verify_parent_transid(io_tree, eb, parent_transid))
374 return ret;
375
376 /*
377 * This buffer's crc is fine, but its contents are corrupted, so
378 * there is no reason to read the other copies, they won't be
379 * any less wrong.
380 */
381 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
382 return ret;
383
384 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
385 eb->start, eb->len);
386 if (num_copies == 1)
387 return ret;
388
389 mirror_num++;
390 if (mirror_num > num_copies)
391 return ret;
392 }
393 return -EIO;
394}
395
396/*
397 * checksum a dirty tree block before IO. This has extra checks to make sure
398 * we only fill in the checksum field in the first page of a multi-page block
399 */
400
401static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
402{
403 struct extent_io_tree *tree;
404 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
405 u64 found_start;
406 struct extent_buffer *eb;
407
408 tree = &BTRFS_I(page->mapping->host)->io_tree;
409
410 eb = (struct extent_buffer *)page->private;
411 if (page != eb->pages[0])
412 return 0;
413
414 found_start = btrfs_header_bytenr(eb);
415 if (found_start != start) {
416 WARN_ON(1);
417 return 0;
418 }
419 if (eb->pages[0] != page) {
420 WARN_ON(1);
421 return 0;
422 }
423 if (!PageUptodate(page)) {
424 WARN_ON(1);
425 return 0;
426 }
427 csum_tree_block(root, eb, 0);
428 return 0;
429}
430
431static int check_tree_block_fsid(struct btrfs_root *root,
432 struct extent_buffer *eb)
433{
434 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
435 u8 fsid[BTRFS_UUID_SIZE];
436 int ret = 1;
437
438 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
439 BTRFS_FSID_SIZE);
440 while (fs_devices) {
441 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
442 ret = 0;
443 break;
444 }
445 fs_devices = fs_devices->seed;
446 }
447 return ret;
448}
449
450#define CORRUPT(reason, eb, root, slot) \
451 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
452 "root=%llu, slot=%d\n", reason, \
453 (unsigned long long)btrfs_header_bytenr(eb), \
454 (unsigned long long)root->objectid, slot)
455
456static noinline int check_leaf(struct btrfs_root *root,
457 struct extent_buffer *leaf)
458{
459 struct btrfs_key key;
460 struct btrfs_key leaf_key;
461 u32 nritems = btrfs_header_nritems(leaf);
462 int slot;
463
464 if (nritems == 0)
465 return 0;
466
467 /* Check the 0 item */
468 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
469 BTRFS_LEAF_DATA_SIZE(root)) {
470 CORRUPT("invalid item offset size pair", leaf, root, 0);
471 return -EIO;
472 }
473
474 /*
475 * Check to make sure each items keys are in the correct order and their
476 * offsets make sense. We only have to loop through nritems-1 because
477 * we check the current slot against the next slot, which verifies the
478 * next slot's offset+size makes sense and that the current's slot
479 * offset is correct.
480 */
481 for (slot = 0; slot < nritems - 1; slot++) {
482 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
483 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
484
485 /* Make sure the keys are in the right order */
486 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
487 CORRUPT("bad key order", leaf, root, slot);
488 return -EIO;
489 }
490
491 /*
492 * Make sure the offset and ends are right, remember that the
493 * item data starts at the end of the leaf and grows towards the
494 * front.
495 */
496 if (btrfs_item_offset_nr(leaf, slot) !=
497 btrfs_item_end_nr(leaf, slot + 1)) {
498 CORRUPT("slot offset bad", leaf, root, slot);
499 return -EIO;
500 }
501
502 /*
503 * Check to make sure that we don't point outside of the leaf,
504 * just incase all the items are consistent to eachother, but
505 * all point outside of the leaf.
506 */
507 if (btrfs_item_end_nr(leaf, slot) >
508 BTRFS_LEAF_DATA_SIZE(root)) {
509 CORRUPT("slot end outside of leaf", leaf, root, slot);
510 return -EIO;
511 }
512 }
513
514 return 0;
515}
516
517struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
518 struct page *page, int max_walk)
519{
520 struct extent_buffer *eb;
521 u64 start = page_offset(page);
522 u64 target = start;
523 u64 min_start;
524
525 if (start < max_walk)
526 min_start = 0;
527 else
528 min_start = start - max_walk;
529
530 while (start >= min_start) {
531 eb = find_extent_buffer(tree, start, 0);
532 if (eb) {
533 /*
534 * we found an extent buffer and it contains our page
535 * horray!
536 */
537 if (eb->start <= target &&
538 eb->start + eb->len > target)
539 return eb;
540
541 /* we found an extent buffer that wasn't for us */
542 free_extent_buffer(eb);
543 return NULL;
544 }
545 if (start == 0)
546 break;
547 start -= PAGE_CACHE_SIZE;
548 }
549 return NULL;
550}
551
552static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
553 struct extent_state *state)
554{
555 struct extent_io_tree *tree;
556 u64 found_start;
557 int found_level;
558 struct extent_buffer *eb;
559 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
560 int ret = 0;
561 int reads_done;
562
563 if (!page->private)
564 goto out;
565
566 tree = &BTRFS_I(page->mapping->host)->io_tree;
567 eb = (struct extent_buffer *)page->private;
568
569 reads_done = atomic_dec_and_test(&eb->pages_reading);
570 if (!reads_done)
571 goto err;
572
573 found_start = btrfs_header_bytenr(eb);
574 if (found_start != eb->start) {
575 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
576 "%llu %llu\n",
577 (unsigned long long)found_start,
578 (unsigned long long)eb->start);
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
609err:
610 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
611 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
612 btree_readahead_hook(root, eb, eb->start, ret);
613 }
614
615 if (ret && eb)
616 clear_extent_buffer_uptodate(tree, eb, NULL);
617out:
618 return ret;
619}
620
621static int btree_io_failed_hook(struct bio *failed_bio,
622 struct page *page, u64 start, u64 end,
623 int mirror_num, struct extent_state *state)
624{
625 struct extent_buffer *eb;
626 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
627
628 eb = (struct extent_buffer *)page->private;
629 if (page != eb->pages[0])
630 return -EIO;
631
632 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
633 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
634 btree_readahead_hook(root, eb, eb->start, -EIO);
635 }
636 return -EIO; /* we fixed nothing */
637}
638
639static void end_workqueue_bio(struct bio *bio, int err)
640{
641 struct end_io_wq *end_io_wq = bio->bi_private;
642 struct btrfs_fs_info *fs_info;
643
644 fs_info = end_io_wq->info;
645 end_io_wq->error = err;
646 end_io_wq->work.func = end_workqueue_fn;
647 end_io_wq->work.flags = 0;
648
649 if (bio->bi_rw & REQ_WRITE) {
650 if (end_io_wq->metadata == 1)
651 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
652 &end_io_wq->work);
653 else if (end_io_wq->metadata == 2)
654 btrfs_queue_worker(&fs_info->endio_freespace_worker,
655 &end_io_wq->work);
656 else
657 btrfs_queue_worker(&fs_info->endio_write_workers,
658 &end_io_wq->work);
659 } else {
660 if (end_io_wq->metadata)
661 btrfs_queue_worker(&fs_info->endio_meta_workers,
662 &end_io_wq->work);
663 else
664 btrfs_queue_worker(&fs_info->endio_workers,
665 &end_io_wq->work);
666 }
667}
668
669/*
670 * For the metadata arg you want
671 *
672 * 0 - if data
673 * 1 - if normal metadta
674 * 2 - if writing to the free space cache area
675 */
676int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
677 int metadata)
678{
679 struct end_io_wq *end_io_wq;
680 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
681 if (!end_io_wq)
682 return -ENOMEM;
683
684 end_io_wq->private = bio->bi_private;
685 end_io_wq->end_io = bio->bi_end_io;
686 end_io_wq->info = info;
687 end_io_wq->error = 0;
688 end_io_wq->bio = bio;
689 end_io_wq->metadata = metadata;
690
691 bio->bi_private = end_io_wq;
692 bio->bi_end_io = end_workqueue_bio;
693 return 0;
694}
695
696unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
697{
698 unsigned long limit = min_t(unsigned long,
699 info->workers.max_workers,
700 info->fs_devices->open_devices);
701 return 256 * limit;
702}
703
704static void run_one_async_start(struct btrfs_work *work)
705{
706 struct async_submit_bio *async;
707
708 async = container_of(work, struct async_submit_bio, work);
709 async->submit_bio_start(async->inode, async->rw, async->bio,
710 async->mirror_num, async->bio_flags,
711 async->bio_offset);
712}
713
714static void run_one_async_done(struct btrfs_work *work)
715{
716 struct btrfs_fs_info *fs_info;
717 struct async_submit_bio *async;
718 int limit;
719
720 async = container_of(work, struct async_submit_bio, work);
721 fs_info = BTRFS_I(async->inode)->root->fs_info;
722
723 limit = btrfs_async_submit_limit(fs_info);
724 limit = limit * 2 / 3;
725
726 atomic_dec(&fs_info->nr_async_submits);
727
728 if (atomic_read(&fs_info->nr_async_submits) < limit &&
729 waitqueue_active(&fs_info->async_submit_wait))
730 wake_up(&fs_info->async_submit_wait);
731
732 async->submit_bio_done(async->inode, async->rw, async->bio,
733 async->mirror_num, async->bio_flags,
734 async->bio_offset);
735}
736
737static void run_one_async_free(struct btrfs_work *work)
738{
739 struct async_submit_bio *async;
740
741 async = container_of(work, struct async_submit_bio, work);
742 kfree(async);
743}
744
745int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
746 int rw, struct bio *bio, int mirror_num,
747 unsigned long bio_flags,
748 u64 bio_offset,
749 extent_submit_bio_hook_t *submit_bio_start,
750 extent_submit_bio_hook_t *submit_bio_done)
751{
752 struct async_submit_bio *async;
753
754 async = kmalloc(sizeof(*async), GFP_NOFS);
755 if (!async)
756 return -ENOMEM;
757
758 async->inode = inode;
759 async->rw = rw;
760 async->bio = bio;
761 async->mirror_num = mirror_num;
762 async->submit_bio_start = submit_bio_start;
763 async->submit_bio_done = submit_bio_done;
764
765 async->work.func = run_one_async_start;
766 async->work.ordered_func = run_one_async_done;
767 async->work.ordered_free = run_one_async_free;
768
769 async->work.flags = 0;
770 async->bio_flags = bio_flags;
771 async->bio_offset = bio_offset;
772
773 atomic_inc(&fs_info->nr_async_submits);
774
775 if (rw & REQ_SYNC)
776 btrfs_set_work_high_prio(&async->work);
777
778 btrfs_queue_worker(&fs_info->workers, &async->work);
779
780 while (atomic_read(&fs_info->async_submit_draining) &&
781 atomic_read(&fs_info->nr_async_submits)) {
782 wait_event(fs_info->async_submit_wait,
783 (atomic_read(&fs_info->nr_async_submits) == 0));
784 }
785
786 return 0;
787}
788
789static int btree_csum_one_bio(struct bio *bio)
790{
791 struct bio_vec *bvec = bio->bi_io_vec;
792 int bio_index = 0;
793 struct btrfs_root *root;
794
795 WARN_ON(bio->bi_vcnt <= 0);
796 while (bio_index < bio->bi_vcnt) {
797 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
798 csum_dirty_buffer(root, bvec->bv_page);
799 bio_index++;
800 bvec++;
801 }
802 return 0;
803}
804
805static int __btree_submit_bio_start(struct inode *inode, int rw,
806 struct bio *bio, int mirror_num,
807 unsigned long bio_flags,
808 u64 bio_offset)
809{
810 /*
811 * when we're called for a write, we're already in the async
812 * submission context. Just jump into btrfs_map_bio
813 */
814 btree_csum_one_bio(bio);
815 return 0;
816}
817
818static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
819 int mirror_num, unsigned long bio_flags,
820 u64 bio_offset)
821{
822 /*
823 * when we're called for a write, we're already in the async
824 * submission context. Just jump into btrfs_map_bio
825 */
826 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
827}
828
829static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
830 int mirror_num, unsigned long bio_flags,
831 u64 bio_offset)
832{
833 int ret;
834
835 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
836 bio, 1);
837 BUG_ON(ret);
838
839 if (!(rw & REQ_WRITE)) {
840 /*
841 * called for a read, do the setup so that checksum validation
842 * can happen in the async kernel threads
843 */
844 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
845 mirror_num, 0);
846 }
847
848 /*
849 * kthread helpers are used to submit writes so that checksumming
850 * can happen in parallel across all CPUs
851 */
852 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
853 inode, rw, bio, mirror_num, 0,
854 bio_offset,
855 __btree_submit_bio_start,
856 __btree_submit_bio_done);
857}
858
859#ifdef CONFIG_MIGRATION
860static int btree_migratepage(struct address_space *mapping,
861 struct page *newpage, struct page *page,
862 enum migrate_mode mode)
863{
864 /*
865 * we can't safely write a btree page from here,
866 * we haven't done the locking hook
867 */
868 if (PageDirty(page))
869 return -EAGAIN;
870 /*
871 * Buffers may be managed in a filesystem specific way.
872 * We must have no buffers or drop them.
873 */
874 if (page_has_private(page) &&
875 !try_to_release_page(page, GFP_KERNEL))
876 return -EAGAIN;
877 return migrate_page(mapping, newpage, page, mode);
878}
879#endif
880
881static int btree_writepage(struct page *page, struct writeback_control *wbc)
882{
883 struct extent_io_tree *tree;
884 tree = &BTRFS_I(page->mapping->host)->io_tree;
885
886 if (!(current->flags & PF_MEMALLOC)) {
887 return extent_write_full_page(tree, page,
888 btree_get_extent, wbc);
889 }
890
891 redirty_page_for_writepage(wbc, page);
892 unlock_page(page);
893 return 0;
894}
895
896static int btree_writepages(struct address_space *mapping,
897 struct writeback_control *wbc)
898{
899 struct extent_io_tree *tree;
900 tree = &BTRFS_I(mapping->host)->io_tree;
901 if (wbc->sync_mode == WB_SYNC_NONE) {
902 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
903 u64 num_dirty;
904 unsigned long thresh = 32 * 1024 * 1024;
905
906 if (wbc->for_kupdate)
907 return 0;
908
909 /* this is a bit racy, but that's ok */
910 num_dirty = root->fs_info->dirty_metadata_bytes;
911 if (num_dirty < thresh)
912 return 0;
913 }
914 return extent_writepages(tree, mapping, btree_get_extent, wbc);
915}
916
917static int btree_readpage(struct file *file, struct page *page)
918{
919 struct extent_io_tree *tree;
920 tree = &BTRFS_I(page->mapping->host)->io_tree;
921 return extent_read_full_page(tree, page, btree_get_extent, 0);
922}
923
924static int btree_releasepage(struct page *page, gfp_t gfp_flags)
925{
926 struct extent_map_tree *map;
927 struct extent_io_tree *tree;
928 int ret;
929
930 if (PageWriteback(page) || PageDirty(page))
931 return 0;
932
933 tree = &BTRFS_I(page->mapping->host)->io_tree;
934 map = &BTRFS_I(page->mapping->host)->extent_tree;
935
936 /*
937 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
938 * slab allocation from alloc_extent_state down the callchain where
939 * it'd hit a BUG_ON as those flags are not allowed.
940 */
941 gfp_flags &= ~GFP_SLAB_BUG_MASK;
942
943 ret = try_release_extent_state(map, tree, page, gfp_flags);
944 if (!ret)
945 return 0;
946
947 return try_release_extent_buffer(tree, page);
948}
949
950static void btree_invalidatepage(struct page *page, unsigned long offset)
951{
952 struct extent_io_tree *tree;
953 tree = &BTRFS_I(page->mapping->host)->io_tree;
954 extent_invalidatepage(tree, page, offset);
955 btree_releasepage(page, GFP_NOFS);
956 if (PagePrivate(page)) {
957 printk(KERN_WARNING "btrfs warning page private not zero "
958 "on page %llu\n", (unsigned long long)page_offset(page));
959 ClearPagePrivate(page);
960 set_page_private(page, 0);
961 page_cache_release(page);
962 }
963}
964
965static const struct address_space_operations btree_aops = {
966 .readpage = btree_readpage,
967 .writepage = btree_writepage,
968 .writepages = btree_writepages,
969 .releasepage = btree_releasepage,
970 .invalidatepage = btree_invalidatepage,
971#ifdef CONFIG_MIGRATION
972 .migratepage = btree_migratepage,
973#endif
974};
975
976int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
977 u64 parent_transid)
978{
979 struct extent_buffer *buf = NULL;
980 struct inode *btree_inode = root->fs_info->btree_inode;
981 int ret = 0;
982
983 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
984 if (!buf)
985 return 0;
986 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
987 buf, 0, WAIT_NONE, btree_get_extent, 0);
988 free_extent_buffer(buf);
989 return ret;
990}
991
992int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
993 int mirror_num, struct extent_buffer **eb)
994{
995 struct extent_buffer *buf = NULL;
996 struct inode *btree_inode = root->fs_info->btree_inode;
997 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
998 int ret;
999
1000 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1001 if (!buf)
1002 return 0;
1003
1004 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1005
1006 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1007 btree_get_extent, mirror_num);
1008 if (ret) {
1009 free_extent_buffer(buf);
1010 return ret;
1011 }
1012
1013 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1014 free_extent_buffer(buf);
1015 return -EIO;
1016 } else if (extent_buffer_uptodate(io_tree, buf, NULL)) {
1017 *eb = buf;
1018 } else {
1019 free_extent_buffer(buf);
1020 }
1021 return 0;
1022}
1023
1024struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1025 u64 bytenr, u32 blocksize)
1026{
1027 struct inode *btree_inode = root->fs_info->btree_inode;
1028 struct extent_buffer *eb;
1029 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1030 bytenr, blocksize);
1031 return eb;
1032}
1033
1034struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1035 u64 bytenr, u32 blocksize)
1036{
1037 struct inode *btree_inode = root->fs_info->btree_inode;
1038 struct extent_buffer *eb;
1039
1040 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1041 bytenr, blocksize);
1042 return eb;
1043}
1044
1045
1046int btrfs_write_tree_block(struct extent_buffer *buf)
1047{
1048 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1049 buf->start + buf->len - 1);
1050}
1051
1052int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1053{
1054 return filemap_fdatawait_range(buf->pages[0]->mapping,
1055 buf->start, buf->start + buf->len - 1);
1056}
1057
1058struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1059 u32 blocksize, u64 parent_transid)
1060{
1061 struct extent_buffer *buf = NULL;
1062 int ret;
1063
1064 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1065 if (!buf)
1066 return NULL;
1067
1068 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1069
1070 if (ret == 0)
1071 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1072 return buf;
1073
1074}
1075
1076int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1077 struct extent_buffer *buf)
1078{
1079 struct inode *btree_inode = root->fs_info->btree_inode;
1080 if (btrfs_header_generation(buf) ==
1081 root->fs_info->running_transaction->transid) {
1082 btrfs_assert_tree_locked(buf);
1083
1084 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1085 spin_lock(&root->fs_info->delalloc_lock);
1086 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1087 root->fs_info->dirty_metadata_bytes -= buf->len;
1088 else
1089 WARN_ON(1);
1090 spin_unlock(&root->fs_info->delalloc_lock);
1091 }
1092
1093 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1094 btrfs_set_lock_blocking(buf);
1095 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1096 buf);
1097 }
1098 return 0;
1099}
1100
1101static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1102 u32 stripesize, struct btrfs_root *root,
1103 struct btrfs_fs_info *fs_info,
1104 u64 objectid)
1105{
1106 root->node = NULL;
1107 root->commit_root = NULL;
1108 root->sectorsize = sectorsize;
1109 root->nodesize = nodesize;
1110 root->leafsize = leafsize;
1111 root->stripesize = stripesize;
1112 root->ref_cows = 0;
1113 root->track_dirty = 0;
1114 root->in_radix = 0;
1115 root->orphan_item_inserted = 0;
1116 root->orphan_cleanup_state = 0;
1117
1118 root->objectid = objectid;
1119 root->last_trans = 0;
1120 root->highest_objectid = 0;
1121 root->name = NULL;
1122 root->inode_tree = RB_ROOT;
1123 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1124 root->block_rsv = NULL;
1125 root->orphan_block_rsv = NULL;
1126
1127 INIT_LIST_HEAD(&root->dirty_list);
1128 INIT_LIST_HEAD(&root->orphan_list);
1129 INIT_LIST_HEAD(&root->root_list);
1130 spin_lock_init(&root->orphan_lock);
1131 spin_lock_init(&root->inode_lock);
1132 spin_lock_init(&root->accounting_lock);
1133 mutex_init(&root->objectid_mutex);
1134 mutex_init(&root->log_mutex);
1135 init_waitqueue_head(&root->log_writer_wait);
1136 init_waitqueue_head(&root->log_commit_wait[0]);
1137 init_waitqueue_head(&root->log_commit_wait[1]);
1138 atomic_set(&root->log_commit[0], 0);
1139 atomic_set(&root->log_commit[1], 0);
1140 atomic_set(&root->log_writers, 0);
1141 root->log_batch = 0;
1142 root->log_transid = 0;
1143 root->last_log_commit = 0;
1144 extent_io_tree_init(&root->dirty_log_pages,
1145 fs_info->btree_inode->i_mapping);
1146
1147 memset(&root->root_key, 0, sizeof(root->root_key));
1148 memset(&root->root_item, 0, sizeof(root->root_item));
1149 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1150 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1151 root->defrag_trans_start = fs_info->generation;
1152 init_completion(&root->kobj_unregister);
1153 root->defrag_running = 0;
1154 root->root_key.objectid = objectid;
1155 root->anon_dev = 0;
1156 return 0;
1157}
1158
1159static int find_and_setup_root(struct btrfs_root *tree_root,
1160 struct btrfs_fs_info *fs_info,
1161 u64 objectid,
1162 struct btrfs_root *root)
1163{
1164 int ret;
1165 u32 blocksize;
1166 u64 generation;
1167
1168 __setup_root(tree_root->nodesize, tree_root->leafsize,
1169 tree_root->sectorsize, tree_root->stripesize,
1170 root, fs_info, objectid);
1171 ret = btrfs_find_last_root(tree_root, objectid,
1172 &root->root_item, &root->root_key);
1173 if (ret > 0)
1174 return -ENOENT;
1175 BUG_ON(ret);
1176
1177 generation = btrfs_root_generation(&root->root_item);
1178 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1179 root->commit_root = NULL;
1180 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1181 blocksize, generation);
1182 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1183 free_extent_buffer(root->node);
1184 root->node = NULL;
1185 return -EIO;
1186 }
1187 root->commit_root = btrfs_root_node(root);
1188 return 0;
1189}
1190
1191static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1192{
1193 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1194 if (root)
1195 root->fs_info = fs_info;
1196 return root;
1197}
1198
1199static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1200 struct btrfs_fs_info *fs_info)
1201{
1202 struct btrfs_root *root;
1203 struct btrfs_root *tree_root = fs_info->tree_root;
1204 struct extent_buffer *leaf;
1205
1206 root = btrfs_alloc_root(fs_info);
1207 if (!root)
1208 return ERR_PTR(-ENOMEM);
1209
1210 __setup_root(tree_root->nodesize, tree_root->leafsize,
1211 tree_root->sectorsize, tree_root->stripesize,
1212 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1213
1214 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1215 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1216 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1217 /*
1218 * log trees do not get reference counted because they go away
1219 * before a real commit is actually done. They do store pointers
1220 * to file data extents, and those reference counts still get
1221 * updated (along with back refs to the log tree).
1222 */
1223 root->ref_cows = 0;
1224
1225 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1226 BTRFS_TREE_LOG_OBJECTID, NULL,
1227 0, 0, 0, 0);
1228 if (IS_ERR(leaf)) {
1229 kfree(root);
1230 return ERR_CAST(leaf);
1231 }
1232
1233 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1234 btrfs_set_header_bytenr(leaf, leaf->start);
1235 btrfs_set_header_generation(leaf, trans->transid);
1236 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1237 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1238 root->node = leaf;
1239
1240 write_extent_buffer(root->node, root->fs_info->fsid,
1241 (unsigned long)btrfs_header_fsid(root->node),
1242 BTRFS_FSID_SIZE);
1243 btrfs_mark_buffer_dirty(root->node);
1244 btrfs_tree_unlock(root->node);
1245 return root;
1246}
1247
1248int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1249 struct btrfs_fs_info *fs_info)
1250{
1251 struct btrfs_root *log_root;
1252
1253 log_root = alloc_log_tree(trans, fs_info);
1254 if (IS_ERR(log_root))
1255 return PTR_ERR(log_root);
1256 WARN_ON(fs_info->log_root_tree);
1257 fs_info->log_root_tree = log_root;
1258 return 0;
1259}
1260
1261int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1262 struct btrfs_root *root)
1263{
1264 struct btrfs_root *log_root;
1265 struct btrfs_inode_item *inode_item;
1266
1267 log_root = alloc_log_tree(trans, root->fs_info);
1268 if (IS_ERR(log_root))
1269 return PTR_ERR(log_root);
1270
1271 log_root->last_trans = trans->transid;
1272 log_root->root_key.offset = root->root_key.objectid;
1273
1274 inode_item = &log_root->root_item.inode;
1275 inode_item->generation = cpu_to_le64(1);
1276 inode_item->size = cpu_to_le64(3);
1277 inode_item->nlink = cpu_to_le32(1);
1278 inode_item->nbytes = cpu_to_le64(root->leafsize);
1279 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1280
1281 btrfs_set_root_node(&log_root->root_item, log_root->node);
1282
1283 WARN_ON(root->log_root);
1284 root->log_root = log_root;
1285 root->log_transid = 0;
1286 root->last_log_commit = 0;
1287 return 0;
1288}
1289
1290struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1291 struct btrfs_key *location)
1292{
1293 struct btrfs_root *root;
1294 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1295 struct btrfs_path *path;
1296 struct extent_buffer *l;
1297 u64 generation;
1298 u32 blocksize;
1299 int ret = 0;
1300
1301 root = btrfs_alloc_root(fs_info);
1302 if (!root)
1303 return ERR_PTR(-ENOMEM);
1304 if (location->offset == (u64)-1) {
1305 ret = find_and_setup_root(tree_root, fs_info,
1306 location->objectid, root);
1307 if (ret) {
1308 kfree(root);
1309 return ERR_PTR(ret);
1310 }
1311 goto out;
1312 }
1313
1314 __setup_root(tree_root->nodesize, tree_root->leafsize,
1315 tree_root->sectorsize, tree_root->stripesize,
1316 root, fs_info, location->objectid);
1317
1318 path = btrfs_alloc_path();
1319 if (!path) {
1320 kfree(root);
1321 return ERR_PTR(-ENOMEM);
1322 }
1323 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1324 if (ret == 0) {
1325 l = path->nodes[0];
1326 read_extent_buffer(l, &root->root_item,
1327 btrfs_item_ptr_offset(l, path->slots[0]),
1328 sizeof(root->root_item));
1329 memcpy(&root->root_key, location, sizeof(*location));
1330 }
1331 btrfs_free_path(path);
1332 if (ret) {
1333 kfree(root);
1334 if (ret > 0)
1335 ret = -ENOENT;
1336 return ERR_PTR(ret);
1337 }
1338
1339 generation = btrfs_root_generation(&root->root_item);
1340 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1341 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1342 blocksize, generation);
1343 root->commit_root = btrfs_root_node(root);
1344 BUG_ON(!root->node);
1345out:
1346 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1347 root->ref_cows = 1;
1348 btrfs_check_and_init_root_item(&root->root_item);
1349 }
1350
1351 return root;
1352}
1353
1354struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1355 struct btrfs_key *location)
1356{
1357 struct btrfs_root *root;
1358 int ret;
1359
1360 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1361 return fs_info->tree_root;
1362 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1363 return fs_info->extent_root;
1364 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1365 return fs_info->chunk_root;
1366 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1367 return fs_info->dev_root;
1368 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1369 return fs_info->csum_root;
1370again:
1371 spin_lock(&fs_info->fs_roots_radix_lock);
1372 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1373 (unsigned long)location->objectid);
1374 spin_unlock(&fs_info->fs_roots_radix_lock);
1375 if (root)
1376 return root;
1377
1378 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1379 if (IS_ERR(root))
1380 return root;
1381
1382 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1383 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1384 GFP_NOFS);
1385 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1386 ret = -ENOMEM;
1387 goto fail;
1388 }
1389
1390 btrfs_init_free_ino_ctl(root);
1391 mutex_init(&root->fs_commit_mutex);
1392 spin_lock_init(&root->cache_lock);
1393 init_waitqueue_head(&root->cache_wait);
1394
1395 ret = get_anon_bdev(&root->anon_dev);
1396 if (ret)
1397 goto fail;
1398
1399 if (btrfs_root_refs(&root->root_item) == 0) {
1400 ret = -ENOENT;
1401 goto fail;
1402 }
1403
1404 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1405 if (ret < 0)
1406 goto fail;
1407 if (ret == 0)
1408 root->orphan_item_inserted = 1;
1409
1410 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1411 if (ret)
1412 goto fail;
1413
1414 spin_lock(&fs_info->fs_roots_radix_lock);
1415 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1416 (unsigned long)root->root_key.objectid,
1417 root);
1418 if (ret == 0)
1419 root->in_radix = 1;
1420
1421 spin_unlock(&fs_info->fs_roots_radix_lock);
1422 radix_tree_preload_end();
1423 if (ret) {
1424 if (ret == -EEXIST) {
1425 free_fs_root(root);
1426 goto again;
1427 }
1428 goto fail;
1429 }
1430
1431 ret = btrfs_find_dead_roots(fs_info->tree_root,
1432 root->root_key.objectid);
1433 WARN_ON(ret);
1434 return root;
1435fail:
1436 free_fs_root(root);
1437 return ERR_PTR(ret);
1438}
1439
1440static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1441{
1442 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1443 int ret = 0;
1444 struct btrfs_device *device;
1445 struct backing_dev_info *bdi;
1446
1447 rcu_read_lock();
1448 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1449 if (!device->bdev)
1450 continue;
1451 bdi = blk_get_backing_dev_info(device->bdev);
1452 if (bdi && bdi_congested(bdi, bdi_bits)) {
1453 ret = 1;
1454 break;
1455 }
1456 }
1457 rcu_read_unlock();
1458 return ret;
1459}
1460
1461/*
1462 * If this fails, caller must call bdi_destroy() to get rid of the
1463 * bdi again.
1464 */
1465static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1466{
1467 int err;
1468
1469 bdi->capabilities = BDI_CAP_MAP_COPY;
1470 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1471 if (err)
1472 return err;
1473
1474 bdi->ra_pages = default_backing_dev_info.ra_pages;
1475 bdi->congested_fn = btrfs_congested_fn;
1476 bdi->congested_data = info;
1477 return 0;
1478}
1479
1480/*
1481 * called by the kthread helper functions to finally call the bio end_io
1482 * functions. This is where read checksum verification actually happens
1483 */
1484static void end_workqueue_fn(struct btrfs_work *work)
1485{
1486 struct bio *bio;
1487 struct end_io_wq *end_io_wq;
1488 struct btrfs_fs_info *fs_info;
1489 int error;
1490
1491 end_io_wq = container_of(work, struct end_io_wq, work);
1492 bio = end_io_wq->bio;
1493 fs_info = end_io_wq->info;
1494
1495 error = end_io_wq->error;
1496 bio->bi_private = end_io_wq->private;
1497 bio->bi_end_io = end_io_wq->end_io;
1498 kfree(end_io_wq);
1499 bio_endio(bio, error);
1500}
1501
1502static int cleaner_kthread(void *arg)
1503{
1504 struct btrfs_root *root = arg;
1505
1506 do {
1507 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1508
1509 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1510 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1511 btrfs_run_delayed_iputs(root);
1512 btrfs_clean_old_snapshots(root);
1513 mutex_unlock(&root->fs_info->cleaner_mutex);
1514 btrfs_run_defrag_inodes(root->fs_info);
1515 }
1516
1517 if (!try_to_freeze()) {
1518 set_current_state(TASK_INTERRUPTIBLE);
1519 if (!kthread_should_stop())
1520 schedule();
1521 __set_current_state(TASK_RUNNING);
1522 }
1523 } while (!kthread_should_stop());
1524 return 0;
1525}
1526
1527static int transaction_kthread(void *arg)
1528{
1529 struct btrfs_root *root = arg;
1530 struct btrfs_trans_handle *trans;
1531 struct btrfs_transaction *cur;
1532 u64 transid;
1533 unsigned long now;
1534 unsigned long delay;
1535 int ret;
1536
1537 do {
1538 delay = HZ * 30;
1539 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1540 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1541
1542 spin_lock(&root->fs_info->trans_lock);
1543 cur = root->fs_info->running_transaction;
1544 if (!cur) {
1545 spin_unlock(&root->fs_info->trans_lock);
1546 goto sleep;
1547 }
1548
1549 now = get_seconds();
1550 if (!cur->blocked &&
1551 (now < cur->start_time || now - cur->start_time < 30)) {
1552 spin_unlock(&root->fs_info->trans_lock);
1553 delay = HZ * 5;
1554 goto sleep;
1555 }
1556 transid = cur->transid;
1557 spin_unlock(&root->fs_info->trans_lock);
1558
1559 trans = btrfs_join_transaction(root);
1560 BUG_ON(IS_ERR(trans));
1561 if (transid == trans->transid) {
1562 ret = btrfs_commit_transaction(trans, root);
1563 BUG_ON(ret);
1564 } else {
1565 btrfs_end_transaction(trans, root);
1566 }
1567sleep:
1568 wake_up_process(root->fs_info->cleaner_kthread);
1569 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1570
1571 if (!try_to_freeze()) {
1572 set_current_state(TASK_INTERRUPTIBLE);
1573 if (!kthread_should_stop() &&
1574 !btrfs_transaction_blocked(root->fs_info))
1575 schedule_timeout(delay);
1576 __set_current_state(TASK_RUNNING);
1577 }
1578 } while (!kthread_should_stop());
1579 return 0;
1580}
1581
1582/*
1583 * this will find the highest generation in the array of
1584 * root backups. The index of the highest array is returned,
1585 * or -1 if we can't find anything.
1586 *
1587 * We check to make sure the array is valid by comparing the
1588 * generation of the latest root in the array with the generation
1589 * in the super block. If they don't match we pitch it.
1590 */
1591static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1592{
1593 u64 cur;
1594 int newest_index = -1;
1595 struct btrfs_root_backup *root_backup;
1596 int i;
1597
1598 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1599 root_backup = info->super_copy->super_roots + i;
1600 cur = btrfs_backup_tree_root_gen(root_backup);
1601 if (cur == newest_gen)
1602 newest_index = i;
1603 }
1604
1605 /* check to see if we actually wrapped around */
1606 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1607 root_backup = info->super_copy->super_roots;
1608 cur = btrfs_backup_tree_root_gen(root_backup);
1609 if (cur == newest_gen)
1610 newest_index = 0;
1611 }
1612 return newest_index;
1613}
1614
1615
1616/*
1617 * find the oldest backup so we know where to store new entries
1618 * in the backup array. This will set the backup_root_index
1619 * field in the fs_info struct
1620 */
1621static void find_oldest_super_backup(struct btrfs_fs_info *info,
1622 u64 newest_gen)
1623{
1624 int newest_index = -1;
1625
1626 newest_index = find_newest_super_backup(info, newest_gen);
1627 /* if there was garbage in there, just move along */
1628 if (newest_index == -1) {
1629 info->backup_root_index = 0;
1630 } else {
1631 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1632 }
1633}
1634
1635/*
1636 * copy all the root pointers into the super backup array.
1637 * this will bump the backup pointer by one when it is
1638 * done
1639 */
1640static void backup_super_roots(struct btrfs_fs_info *info)
1641{
1642 int next_backup;
1643 struct btrfs_root_backup *root_backup;
1644 int last_backup;
1645
1646 next_backup = info->backup_root_index;
1647 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1648 BTRFS_NUM_BACKUP_ROOTS;
1649
1650 /*
1651 * just overwrite the last backup if we're at the same generation
1652 * this happens only at umount
1653 */
1654 root_backup = info->super_for_commit->super_roots + last_backup;
1655 if (btrfs_backup_tree_root_gen(root_backup) ==
1656 btrfs_header_generation(info->tree_root->node))
1657 next_backup = last_backup;
1658
1659 root_backup = info->super_for_commit->super_roots + next_backup;
1660
1661 /*
1662 * make sure all of our padding and empty slots get zero filled
1663 * regardless of which ones we use today
1664 */
1665 memset(root_backup, 0, sizeof(*root_backup));
1666
1667 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1668
1669 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1670 btrfs_set_backup_tree_root_gen(root_backup,
1671 btrfs_header_generation(info->tree_root->node));
1672
1673 btrfs_set_backup_tree_root_level(root_backup,
1674 btrfs_header_level(info->tree_root->node));
1675
1676 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1677 btrfs_set_backup_chunk_root_gen(root_backup,
1678 btrfs_header_generation(info->chunk_root->node));
1679 btrfs_set_backup_chunk_root_level(root_backup,
1680 btrfs_header_level(info->chunk_root->node));
1681
1682 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1683 btrfs_set_backup_extent_root_gen(root_backup,
1684 btrfs_header_generation(info->extent_root->node));
1685 btrfs_set_backup_extent_root_level(root_backup,
1686 btrfs_header_level(info->extent_root->node));
1687
1688 /*
1689 * we might commit during log recovery, which happens before we set
1690 * the fs_root. Make sure it is valid before we fill it in.
1691 */
1692 if (info->fs_root && info->fs_root->node) {
1693 btrfs_set_backup_fs_root(root_backup,
1694 info->fs_root->node->start);
1695 btrfs_set_backup_fs_root_gen(root_backup,
1696 btrfs_header_generation(info->fs_root->node));
1697 btrfs_set_backup_fs_root_level(root_backup,
1698 btrfs_header_level(info->fs_root->node));
1699 }
1700
1701 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1702 btrfs_set_backup_dev_root_gen(root_backup,
1703 btrfs_header_generation(info->dev_root->node));
1704 btrfs_set_backup_dev_root_level(root_backup,
1705 btrfs_header_level(info->dev_root->node));
1706
1707 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1708 btrfs_set_backup_csum_root_gen(root_backup,
1709 btrfs_header_generation(info->csum_root->node));
1710 btrfs_set_backup_csum_root_level(root_backup,
1711 btrfs_header_level(info->csum_root->node));
1712
1713 btrfs_set_backup_total_bytes(root_backup,
1714 btrfs_super_total_bytes(info->super_copy));
1715 btrfs_set_backup_bytes_used(root_backup,
1716 btrfs_super_bytes_used(info->super_copy));
1717 btrfs_set_backup_num_devices(root_backup,
1718 btrfs_super_num_devices(info->super_copy));
1719
1720 /*
1721 * if we don't copy this out to the super_copy, it won't get remembered
1722 * for the next commit
1723 */
1724 memcpy(&info->super_copy->super_roots,
1725 &info->super_for_commit->super_roots,
1726 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1727}
1728
1729/*
1730 * this copies info out of the root backup array and back into
1731 * the in-memory super block. It is meant to help iterate through
1732 * the array, so you send it the number of backups you've already
1733 * tried and the last backup index you used.
1734 *
1735 * this returns -1 when it has tried all the backups
1736 */
1737static noinline int next_root_backup(struct btrfs_fs_info *info,
1738 struct btrfs_super_block *super,
1739 int *num_backups_tried, int *backup_index)
1740{
1741 struct btrfs_root_backup *root_backup;
1742 int newest = *backup_index;
1743
1744 if (*num_backups_tried == 0) {
1745 u64 gen = btrfs_super_generation(super);
1746
1747 newest = find_newest_super_backup(info, gen);
1748 if (newest == -1)
1749 return -1;
1750
1751 *backup_index = newest;
1752 *num_backups_tried = 1;
1753 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1754 /* we've tried all the backups, all done */
1755 return -1;
1756 } else {
1757 /* jump to the next oldest backup */
1758 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1759 BTRFS_NUM_BACKUP_ROOTS;
1760 *backup_index = newest;
1761 *num_backups_tried += 1;
1762 }
1763 root_backup = super->super_roots + newest;
1764
1765 btrfs_set_super_generation(super,
1766 btrfs_backup_tree_root_gen(root_backup));
1767 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1768 btrfs_set_super_root_level(super,
1769 btrfs_backup_tree_root_level(root_backup));
1770 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1771
1772 /*
1773 * fixme: the total bytes and num_devices need to match or we should
1774 * need a fsck
1775 */
1776 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1777 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1778 return 0;
1779}
1780
1781/* helper to cleanup tree roots */
1782static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1783{
1784 free_extent_buffer(info->tree_root->node);
1785 free_extent_buffer(info->tree_root->commit_root);
1786 free_extent_buffer(info->dev_root->node);
1787 free_extent_buffer(info->dev_root->commit_root);
1788 free_extent_buffer(info->extent_root->node);
1789 free_extent_buffer(info->extent_root->commit_root);
1790 free_extent_buffer(info->csum_root->node);
1791 free_extent_buffer(info->csum_root->commit_root);
1792
1793 info->tree_root->node = NULL;
1794 info->tree_root->commit_root = NULL;
1795 info->dev_root->node = NULL;
1796 info->dev_root->commit_root = NULL;
1797 info->extent_root->node = NULL;
1798 info->extent_root->commit_root = NULL;
1799 info->csum_root->node = NULL;
1800 info->csum_root->commit_root = NULL;
1801
1802 if (chunk_root) {
1803 free_extent_buffer(info->chunk_root->node);
1804 free_extent_buffer(info->chunk_root->commit_root);
1805 info->chunk_root->node = NULL;
1806 info->chunk_root->commit_root = NULL;
1807 }
1808}
1809
1810
1811int open_ctree(struct super_block *sb,
1812 struct btrfs_fs_devices *fs_devices,
1813 char *options)
1814{
1815 u32 sectorsize;
1816 u32 nodesize;
1817 u32 leafsize;
1818 u32 blocksize;
1819 u32 stripesize;
1820 u64 generation;
1821 u64 features;
1822 struct btrfs_key location;
1823 struct buffer_head *bh;
1824 struct btrfs_super_block *disk_super;
1825 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1826 struct btrfs_root *tree_root;
1827 struct btrfs_root *extent_root;
1828 struct btrfs_root *csum_root;
1829 struct btrfs_root *chunk_root;
1830 struct btrfs_root *dev_root;
1831 struct btrfs_root *log_tree_root;
1832 int ret;
1833 int err = -EINVAL;
1834 int num_backups_tried = 0;
1835 int backup_index = 0;
1836
1837 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1838 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1839 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1840 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1841 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1842
1843 if (!tree_root || !extent_root || !csum_root ||
1844 !chunk_root || !dev_root) {
1845 err = -ENOMEM;
1846 goto fail;
1847 }
1848
1849 ret = init_srcu_struct(&fs_info->subvol_srcu);
1850 if (ret) {
1851 err = ret;
1852 goto fail;
1853 }
1854
1855 ret = setup_bdi(fs_info, &fs_info->bdi);
1856 if (ret) {
1857 err = ret;
1858 goto fail_srcu;
1859 }
1860
1861 fs_info->btree_inode = new_inode(sb);
1862 if (!fs_info->btree_inode) {
1863 err = -ENOMEM;
1864 goto fail_bdi;
1865 }
1866
1867 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1868
1869 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1870 INIT_LIST_HEAD(&fs_info->trans_list);
1871 INIT_LIST_HEAD(&fs_info->dead_roots);
1872 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1873 INIT_LIST_HEAD(&fs_info->hashers);
1874 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1875 INIT_LIST_HEAD(&fs_info->ordered_operations);
1876 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1877 spin_lock_init(&fs_info->delalloc_lock);
1878 spin_lock_init(&fs_info->trans_lock);
1879 spin_lock_init(&fs_info->ref_cache_lock);
1880 spin_lock_init(&fs_info->fs_roots_radix_lock);
1881 spin_lock_init(&fs_info->delayed_iput_lock);
1882 spin_lock_init(&fs_info->defrag_inodes_lock);
1883 spin_lock_init(&fs_info->free_chunk_lock);
1884 mutex_init(&fs_info->reloc_mutex);
1885
1886 init_completion(&fs_info->kobj_unregister);
1887 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1888 INIT_LIST_HEAD(&fs_info->space_info);
1889 btrfs_mapping_init(&fs_info->mapping_tree);
1890 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1891 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1892 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1893 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1894 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1895 btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1896 atomic_set(&fs_info->nr_async_submits, 0);
1897 atomic_set(&fs_info->async_delalloc_pages, 0);
1898 atomic_set(&fs_info->async_submit_draining, 0);
1899 atomic_set(&fs_info->nr_async_bios, 0);
1900 atomic_set(&fs_info->defrag_running, 0);
1901 fs_info->sb = sb;
1902 fs_info->max_inline = 8192 * 1024;
1903 fs_info->metadata_ratio = 0;
1904 fs_info->defrag_inodes = RB_ROOT;
1905 fs_info->trans_no_join = 0;
1906 fs_info->free_chunk_space = 0;
1907
1908 /* readahead state */
1909 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1910 spin_lock_init(&fs_info->reada_lock);
1911
1912 fs_info->thread_pool_size = min_t(unsigned long,
1913 num_online_cpus() + 2, 8);
1914
1915 INIT_LIST_HEAD(&fs_info->ordered_extents);
1916 spin_lock_init(&fs_info->ordered_extent_lock);
1917 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1918 GFP_NOFS);
1919 if (!fs_info->delayed_root) {
1920 err = -ENOMEM;
1921 goto fail_iput;
1922 }
1923 btrfs_init_delayed_root(fs_info->delayed_root);
1924
1925 mutex_init(&fs_info->scrub_lock);
1926 atomic_set(&fs_info->scrubs_running, 0);
1927 atomic_set(&fs_info->scrub_pause_req, 0);
1928 atomic_set(&fs_info->scrubs_paused, 0);
1929 atomic_set(&fs_info->scrub_cancel_req, 0);
1930 init_waitqueue_head(&fs_info->scrub_pause_wait);
1931 init_rwsem(&fs_info->scrub_super_lock);
1932 fs_info->scrub_workers_refcnt = 0;
1933#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
1934 fs_info->check_integrity_print_mask = 0;
1935#endif
1936
1937 spin_lock_init(&fs_info->balance_lock);
1938 mutex_init(&fs_info->balance_mutex);
1939 atomic_set(&fs_info->balance_running, 0);
1940 atomic_set(&fs_info->balance_pause_req, 0);
1941 atomic_set(&fs_info->balance_cancel_req, 0);
1942 fs_info->balance_ctl = NULL;
1943 init_waitqueue_head(&fs_info->balance_wait_q);
1944
1945 sb->s_blocksize = 4096;
1946 sb->s_blocksize_bits = blksize_bits(4096);
1947 sb->s_bdi = &fs_info->bdi;
1948
1949 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1950 set_nlink(fs_info->btree_inode, 1);
1951 /*
1952 * we set the i_size on the btree inode to the max possible int.
1953 * the real end of the address space is determined by all of
1954 * the devices in the system
1955 */
1956 fs_info->btree_inode->i_size = OFFSET_MAX;
1957 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1958 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1959
1960 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1961 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1962 fs_info->btree_inode->i_mapping);
1963 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1964
1965 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1966
1967 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1968 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1969 sizeof(struct btrfs_key));
1970 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1971 insert_inode_hash(fs_info->btree_inode);
1972
1973 spin_lock_init(&fs_info->block_group_cache_lock);
1974 fs_info->block_group_cache_tree = RB_ROOT;
1975
1976 extent_io_tree_init(&fs_info->freed_extents[0],
1977 fs_info->btree_inode->i_mapping);
1978 extent_io_tree_init(&fs_info->freed_extents[1],
1979 fs_info->btree_inode->i_mapping);
1980 fs_info->pinned_extents = &fs_info->freed_extents[0];
1981 fs_info->do_barriers = 1;
1982
1983
1984 mutex_init(&fs_info->ordered_operations_mutex);
1985 mutex_init(&fs_info->tree_log_mutex);
1986 mutex_init(&fs_info->chunk_mutex);
1987 mutex_init(&fs_info->transaction_kthread_mutex);
1988 mutex_init(&fs_info->cleaner_mutex);
1989 mutex_init(&fs_info->volume_mutex);
1990 init_rwsem(&fs_info->extent_commit_sem);
1991 init_rwsem(&fs_info->cleanup_work_sem);
1992 init_rwsem(&fs_info->subvol_sem);
1993
1994 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1995 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1996
1997 init_waitqueue_head(&fs_info->transaction_throttle);
1998 init_waitqueue_head(&fs_info->transaction_wait);
1999 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2000 init_waitqueue_head(&fs_info->async_submit_wait);
2001
2002 __setup_root(4096, 4096, 4096, 4096, tree_root,
2003 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2004
2005 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2006 if (!bh) {
2007 err = -EINVAL;
2008 goto fail_alloc;
2009 }
2010
2011 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2012 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2013 sizeof(*fs_info->super_for_commit));
2014 brelse(bh);
2015
2016 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2017
2018 disk_super = fs_info->super_copy;
2019 if (!btrfs_super_root(disk_super))
2020 goto fail_alloc;
2021
2022 /* check FS state, whether FS is broken. */
2023 fs_info->fs_state |= btrfs_super_flags(disk_super);
2024
2025 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2026
2027 /*
2028 * run through our array of backup supers and setup
2029 * our ring pointer to the oldest one
2030 */
2031 generation = btrfs_super_generation(disk_super);
2032 find_oldest_super_backup(fs_info, generation);
2033
2034 /*
2035 * In the long term, we'll store the compression type in the super
2036 * block, and it'll be used for per file compression control.
2037 */
2038 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2039
2040 ret = btrfs_parse_options(tree_root, options);
2041 if (ret) {
2042 err = ret;
2043 goto fail_alloc;
2044 }
2045
2046 features = btrfs_super_incompat_flags(disk_super) &
2047 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2048 if (features) {
2049 printk(KERN_ERR "BTRFS: couldn't mount because of "
2050 "unsupported optional features (%Lx).\n",
2051 (unsigned long long)features);
2052 err = -EINVAL;
2053 goto fail_alloc;
2054 }
2055
2056 if (btrfs_super_leafsize(disk_super) !=
2057 btrfs_super_nodesize(disk_super)) {
2058 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2059 "blocksizes don't match. node %d leaf %d\n",
2060 btrfs_super_nodesize(disk_super),
2061 btrfs_super_leafsize(disk_super));
2062 err = -EINVAL;
2063 goto fail_alloc;
2064 }
2065 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2066 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2067 "blocksize (%d) was too large\n",
2068 btrfs_super_leafsize(disk_super));
2069 err = -EINVAL;
2070 goto fail_alloc;
2071 }
2072
2073 features = btrfs_super_incompat_flags(disk_super);
2074 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2075 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2076 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2077
2078 /*
2079 * flag our filesystem as having big metadata blocks if
2080 * they are bigger than the page size
2081 */
2082 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2083 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2084 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2085 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2086 }
2087
2088 btrfs_set_super_incompat_flags(disk_super, features);
2089
2090 features = btrfs_super_compat_ro_flags(disk_super) &
2091 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2092 if (!(sb->s_flags & MS_RDONLY) && features) {
2093 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2094 "unsupported option features (%Lx).\n",
2095 (unsigned long long)features);
2096 err = -EINVAL;
2097 goto fail_alloc;
2098 }
2099
2100 btrfs_init_workers(&fs_info->generic_worker,
2101 "genwork", 1, NULL);
2102
2103 btrfs_init_workers(&fs_info->workers, "worker",
2104 fs_info->thread_pool_size,
2105 &fs_info->generic_worker);
2106
2107 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2108 fs_info->thread_pool_size,
2109 &fs_info->generic_worker);
2110
2111 btrfs_init_workers(&fs_info->submit_workers, "submit",
2112 min_t(u64, fs_devices->num_devices,
2113 fs_info->thread_pool_size),
2114 &fs_info->generic_worker);
2115
2116 btrfs_init_workers(&fs_info->caching_workers, "cache",
2117 2, &fs_info->generic_worker);
2118
2119 /* a higher idle thresh on the submit workers makes it much more
2120 * likely that bios will be send down in a sane order to the
2121 * devices
2122 */
2123 fs_info->submit_workers.idle_thresh = 64;
2124
2125 fs_info->workers.idle_thresh = 16;
2126 fs_info->workers.ordered = 1;
2127
2128 fs_info->delalloc_workers.idle_thresh = 2;
2129 fs_info->delalloc_workers.ordered = 1;
2130
2131 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2132 &fs_info->generic_worker);
2133 btrfs_init_workers(&fs_info->endio_workers, "endio",
2134 fs_info->thread_pool_size,
2135 &fs_info->generic_worker);
2136 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2137 fs_info->thread_pool_size,
2138 &fs_info->generic_worker);
2139 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2140 "endio-meta-write", fs_info->thread_pool_size,
2141 &fs_info->generic_worker);
2142 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2143 fs_info->thread_pool_size,
2144 &fs_info->generic_worker);
2145 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2146 1, &fs_info->generic_worker);
2147 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2148 fs_info->thread_pool_size,
2149 &fs_info->generic_worker);
2150 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2151 fs_info->thread_pool_size,
2152 &fs_info->generic_worker);
2153
2154 /*
2155 * endios are largely parallel and should have a very
2156 * low idle thresh
2157 */
2158 fs_info->endio_workers.idle_thresh = 4;
2159 fs_info->endio_meta_workers.idle_thresh = 4;
2160
2161 fs_info->endio_write_workers.idle_thresh = 2;
2162 fs_info->endio_meta_write_workers.idle_thresh = 2;
2163 fs_info->readahead_workers.idle_thresh = 2;
2164
2165 /*
2166 * btrfs_start_workers can really only fail because of ENOMEM so just
2167 * return -ENOMEM if any of these fail.
2168 */
2169 ret = btrfs_start_workers(&fs_info->workers);
2170 ret |= btrfs_start_workers(&fs_info->generic_worker);
2171 ret |= btrfs_start_workers(&fs_info->submit_workers);
2172 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2173 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2174 ret |= btrfs_start_workers(&fs_info->endio_workers);
2175 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2176 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2177 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2178 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2179 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2180 ret |= btrfs_start_workers(&fs_info->caching_workers);
2181 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2182 if (ret) {
2183 ret = -ENOMEM;
2184 goto fail_sb_buffer;
2185 }
2186
2187 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2188 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2189 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2190
2191 nodesize = btrfs_super_nodesize(disk_super);
2192 leafsize = btrfs_super_leafsize(disk_super);
2193 sectorsize = btrfs_super_sectorsize(disk_super);
2194 stripesize = btrfs_super_stripesize(disk_super);
2195 tree_root->nodesize = nodesize;
2196 tree_root->leafsize = leafsize;
2197 tree_root->sectorsize = sectorsize;
2198 tree_root->stripesize = stripesize;
2199
2200 sb->s_blocksize = sectorsize;
2201 sb->s_blocksize_bits = blksize_bits(sectorsize);
2202
2203 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2204 sizeof(disk_super->magic))) {
2205 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2206 goto fail_sb_buffer;
2207 }
2208
2209 if (sectorsize < PAGE_SIZE) {
2210 printk(KERN_WARNING "btrfs: Incompatible sector size "
2211 "found on %s\n", sb->s_id);
2212 goto fail_sb_buffer;
2213 }
2214
2215 mutex_lock(&fs_info->chunk_mutex);
2216 ret = btrfs_read_sys_array(tree_root);
2217 mutex_unlock(&fs_info->chunk_mutex);
2218 if (ret) {
2219 printk(KERN_WARNING "btrfs: failed to read the system "
2220 "array on %s\n", sb->s_id);
2221 goto fail_sb_buffer;
2222 }
2223
2224 blocksize = btrfs_level_size(tree_root,
2225 btrfs_super_chunk_root_level(disk_super));
2226 generation = btrfs_super_chunk_root_generation(disk_super);
2227
2228 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2229 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2230
2231 chunk_root->node = read_tree_block(chunk_root,
2232 btrfs_super_chunk_root(disk_super),
2233 blocksize, generation);
2234 BUG_ON(!chunk_root->node);
2235 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2236 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2237 sb->s_id);
2238 goto fail_tree_roots;
2239 }
2240 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2241 chunk_root->commit_root = btrfs_root_node(chunk_root);
2242
2243 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2244 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2245 BTRFS_UUID_SIZE);
2246
2247 ret = btrfs_read_chunk_tree(chunk_root);
2248 if (ret) {
2249 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2250 sb->s_id);
2251 goto fail_tree_roots;
2252 }
2253
2254 btrfs_close_extra_devices(fs_devices);
2255
2256 if (!fs_devices->latest_bdev) {
2257 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2258 sb->s_id);
2259 goto fail_tree_roots;
2260 }
2261
2262retry_root_backup:
2263 blocksize = btrfs_level_size(tree_root,
2264 btrfs_super_root_level(disk_super));
2265 generation = btrfs_super_generation(disk_super);
2266
2267 tree_root->node = read_tree_block(tree_root,
2268 btrfs_super_root(disk_super),
2269 blocksize, generation);
2270 if (!tree_root->node ||
2271 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2272 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2273 sb->s_id);
2274
2275 goto recovery_tree_root;
2276 }
2277
2278 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2279 tree_root->commit_root = btrfs_root_node(tree_root);
2280
2281 ret = find_and_setup_root(tree_root, fs_info,
2282 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2283 if (ret)
2284 goto recovery_tree_root;
2285 extent_root->track_dirty = 1;
2286
2287 ret = find_and_setup_root(tree_root, fs_info,
2288 BTRFS_DEV_TREE_OBJECTID, dev_root);
2289 if (ret)
2290 goto recovery_tree_root;
2291 dev_root->track_dirty = 1;
2292
2293 ret = find_and_setup_root(tree_root, fs_info,
2294 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2295 if (ret)
2296 goto recovery_tree_root;
2297
2298 csum_root->track_dirty = 1;
2299
2300 fs_info->generation = generation;
2301 fs_info->last_trans_committed = generation;
2302
2303 ret = btrfs_init_space_info(fs_info);
2304 if (ret) {
2305 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2306 goto fail_block_groups;
2307 }
2308
2309 ret = btrfs_read_block_groups(extent_root);
2310 if (ret) {
2311 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2312 goto fail_block_groups;
2313 }
2314
2315 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2316 "btrfs-cleaner");
2317 if (IS_ERR(fs_info->cleaner_kthread))
2318 goto fail_block_groups;
2319
2320 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2321 tree_root,
2322 "btrfs-transaction");
2323 if (IS_ERR(fs_info->transaction_kthread))
2324 goto fail_cleaner;
2325
2326 if (!btrfs_test_opt(tree_root, SSD) &&
2327 !btrfs_test_opt(tree_root, NOSSD) &&
2328 !fs_info->fs_devices->rotating) {
2329 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2330 "mode\n");
2331 btrfs_set_opt(fs_info->mount_opt, SSD);
2332 }
2333
2334#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2335 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2336 ret = btrfsic_mount(tree_root, fs_devices,
2337 btrfs_test_opt(tree_root,
2338 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2339 1 : 0,
2340 fs_info->check_integrity_print_mask);
2341 if (ret)
2342 printk(KERN_WARNING "btrfs: failed to initialize"
2343 " integrity check module %s\n", sb->s_id);
2344 }
2345#endif
2346
2347 /* do not make disk changes in broken FS */
2348 if (btrfs_super_log_root(disk_super) != 0 &&
2349 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2350 u64 bytenr = btrfs_super_log_root(disk_super);
2351
2352 if (fs_devices->rw_devices == 0) {
2353 printk(KERN_WARNING "Btrfs log replay required "
2354 "on RO media\n");
2355 err = -EIO;
2356 goto fail_trans_kthread;
2357 }
2358 blocksize =
2359 btrfs_level_size(tree_root,
2360 btrfs_super_log_root_level(disk_super));
2361
2362 log_tree_root = btrfs_alloc_root(fs_info);
2363 if (!log_tree_root) {
2364 err = -ENOMEM;
2365 goto fail_trans_kthread;
2366 }
2367
2368 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2369 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2370
2371 log_tree_root->node = read_tree_block(tree_root, bytenr,
2372 blocksize,
2373 generation + 1);
2374 ret = btrfs_recover_log_trees(log_tree_root);
2375 BUG_ON(ret);
2376
2377 if (sb->s_flags & MS_RDONLY) {
2378 ret = btrfs_commit_super(tree_root);
2379 BUG_ON(ret);
2380 }
2381 }
2382
2383 ret = btrfs_find_orphan_roots(tree_root);
2384 BUG_ON(ret);
2385
2386 if (!(sb->s_flags & MS_RDONLY)) {
2387 ret = btrfs_cleanup_fs_roots(fs_info);
2388 BUG_ON(ret);
2389
2390 ret = btrfs_recover_relocation(tree_root);
2391 if (ret < 0) {
2392 printk(KERN_WARNING
2393 "btrfs: failed to recover relocation\n");
2394 err = -EINVAL;
2395 goto fail_trans_kthread;
2396 }
2397 }
2398
2399 location.objectid = BTRFS_FS_TREE_OBJECTID;
2400 location.type = BTRFS_ROOT_ITEM_KEY;
2401 location.offset = (u64)-1;
2402
2403 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2404 if (!fs_info->fs_root)
2405 goto fail_trans_kthread;
2406 if (IS_ERR(fs_info->fs_root)) {
2407 err = PTR_ERR(fs_info->fs_root);
2408 goto fail_trans_kthread;
2409 }
2410
2411 if (!(sb->s_flags & MS_RDONLY)) {
2412 down_read(&fs_info->cleanup_work_sem);
2413 err = btrfs_orphan_cleanup(fs_info->fs_root);
2414 if (!err)
2415 err = btrfs_orphan_cleanup(fs_info->tree_root);
2416 up_read(&fs_info->cleanup_work_sem);
2417
2418 if (!err)
2419 err = btrfs_recover_balance(fs_info->tree_root);
2420
2421 if (err) {
2422 close_ctree(tree_root);
2423 return err;
2424 }
2425 }
2426
2427 return 0;
2428
2429fail_trans_kthread:
2430 kthread_stop(fs_info->transaction_kthread);
2431fail_cleaner:
2432 kthread_stop(fs_info->cleaner_kthread);
2433
2434 /*
2435 * make sure we're done with the btree inode before we stop our
2436 * kthreads
2437 */
2438 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2439 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2440
2441fail_block_groups:
2442 btrfs_free_block_groups(fs_info);
2443
2444fail_tree_roots:
2445 free_root_pointers(fs_info, 1);
2446
2447fail_sb_buffer:
2448 btrfs_stop_workers(&fs_info->generic_worker);
2449 btrfs_stop_workers(&fs_info->readahead_workers);
2450 btrfs_stop_workers(&fs_info->fixup_workers);
2451 btrfs_stop_workers(&fs_info->delalloc_workers);
2452 btrfs_stop_workers(&fs_info->workers);
2453 btrfs_stop_workers(&fs_info->endio_workers);
2454 btrfs_stop_workers(&fs_info->endio_meta_workers);
2455 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2456 btrfs_stop_workers(&fs_info->endio_write_workers);
2457 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2458 btrfs_stop_workers(&fs_info->submit_workers);
2459 btrfs_stop_workers(&fs_info->delayed_workers);
2460 btrfs_stop_workers(&fs_info->caching_workers);
2461fail_alloc:
2462fail_iput:
2463 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2464
2465 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2466 iput(fs_info->btree_inode);
2467fail_bdi:
2468 bdi_destroy(&fs_info->bdi);
2469fail_srcu:
2470 cleanup_srcu_struct(&fs_info->subvol_srcu);
2471fail:
2472 btrfs_close_devices(fs_info->fs_devices);
2473 return err;
2474
2475recovery_tree_root:
2476 if (!btrfs_test_opt(tree_root, RECOVERY))
2477 goto fail_tree_roots;
2478
2479 free_root_pointers(fs_info, 0);
2480
2481 /* don't use the log in recovery mode, it won't be valid */
2482 btrfs_set_super_log_root(disk_super, 0);
2483
2484 /* we can't trust the free space cache either */
2485 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2486
2487 ret = next_root_backup(fs_info, fs_info->super_copy,
2488 &num_backups_tried, &backup_index);
2489 if (ret == -1)
2490 goto fail_block_groups;
2491 goto retry_root_backup;
2492}
2493
2494static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2495{
2496 char b[BDEVNAME_SIZE];
2497
2498 if (uptodate) {
2499 set_buffer_uptodate(bh);
2500 } else {
2501 printk_ratelimited(KERN_WARNING "lost page write due to "
2502 "I/O error on %s\n",
2503 bdevname(bh->b_bdev, b));
2504 /* note, we dont' set_buffer_write_io_error because we have
2505 * our own ways of dealing with the IO errors
2506 */
2507 clear_buffer_uptodate(bh);
2508 }
2509 unlock_buffer(bh);
2510 put_bh(bh);
2511}
2512
2513struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2514{
2515 struct buffer_head *bh;
2516 struct buffer_head *latest = NULL;
2517 struct btrfs_super_block *super;
2518 int i;
2519 u64 transid = 0;
2520 u64 bytenr;
2521
2522 /* we would like to check all the supers, but that would make
2523 * a btrfs mount succeed after a mkfs from a different FS.
2524 * So, we need to add a special mount option to scan for
2525 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2526 */
2527 for (i = 0; i < 1; i++) {
2528 bytenr = btrfs_sb_offset(i);
2529 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2530 break;
2531 bh = __bread(bdev, bytenr / 4096, 4096);
2532 if (!bh)
2533 continue;
2534
2535 super = (struct btrfs_super_block *)bh->b_data;
2536 if (btrfs_super_bytenr(super) != bytenr ||
2537 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2538 sizeof(super->magic))) {
2539 brelse(bh);
2540 continue;
2541 }
2542
2543 if (!latest || btrfs_super_generation(super) > transid) {
2544 brelse(latest);
2545 latest = bh;
2546 transid = btrfs_super_generation(super);
2547 } else {
2548 brelse(bh);
2549 }
2550 }
2551 return latest;
2552}
2553
2554/*
2555 * this should be called twice, once with wait == 0 and
2556 * once with wait == 1. When wait == 0 is done, all the buffer heads
2557 * we write are pinned.
2558 *
2559 * They are released when wait == 1 is done.
2560 * max_mirrors must be the same for both runs, and it indicates how
2561 * many supers on this one device should be written.
2562 *
2563 * max_mirrors == 0 means to write them all.
2564 */
2565static int write_dev_supers(struct btrfs_device *device,
2566 struct btrfs_super_block *sb,
2567 int do_barriers, int wait, int max_mirrors)
2568{
2569 struct buffer_head *bh;
2570 int i;
2571 int ret;
2572 int errors = 0;
2573 u32 crc;
2574 u64 bytenr;
2575
2576 if (max_mirrors == 0)
2577 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2578
2579 for (i = 0; i < max_mirrors; i++) {
2580 bytenr = btrfs_sb_offset(i);
2581 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2582 break;
2583
2584 if (wait) {
2585 bh = __find_get_block(device->bdev, bytenr / 4096,
2586 BTRFS_SUPER_INFO_SIZE);
2587 BUG_ON(!bh);
2588 wait_on_buffer(bh);
2589 if (!buffer_uptodate(bh))
2590 errors++;
2591
2592 /* drop our reference */
2593 brelse(bh);
2594
2595 /* drop the reference from the wait == 0 run */
2596 brelse(bh);
2597 continue;
2598 } else {
2599 btrfs_set_super_bytenr(sb, bytenr);
2600
2601 crc = ~(u32)0;
2602 crc = btrfs_csum_data(NULL, (char *)sb +
2603 BTRFS_CSUM_SIZE, crc,
2604 BTRFS_SUPER_INFO_SIZE -
2605 BTRFS_CSUM_SIZE);
2606 btrfs_csum_final(crc, sb->csum);
2607
2608 /*
2609 * one reference for us, and we leave it for the
2610 * caller
2611 */
2612 bh = __getblk(device->bdev, bytenr / 4096,
2613 BTRFS_SUPER_INFO_SIZE);
2614 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2615
2616 /* one reference for submit_bh */
2617 get_bh(bh);
2618
2619 set_buffer_uptodate(bh);
2620 lock_buffer(bh);
2621 bh->b_end_io = btrfs_end_buffer_write_sync;
2622 }
2623
2624 /*
2625 * we fua the first super. The others we allow
2626 * to go down lazy.
2627 */
2628 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2629 if (ret)
2630 errors++;
2631 }
2632 return errors < i ? 0 : -1;
2633}
2634
2635/*
2636 * endio for the write_dev_flush, this will wake anyone waiting
2637 * for the barrier when it is done
2638 */
2639static void btrfs_end_empty_barrier(struct bio *bio, int err)
2640{
2641 if (err) {
2642 if (err == -EOPNOTSUPP)
2643 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2644 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2645 }
2646 if (bio->bi_private)
2647 complete(bio->bi_private);
2648 bio_put(bio);
2649}
2650
2651/*
2652 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2653 * sent down. With wait == 1, it waits for the previous flush.
2654 *
2655 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2656 * capable
2657 */
2658static int write_dev_flush(struct btrfs_device *device, int wait)
2659{
2660 struct bio *bio;
2661 int ret = 0;
2662
2663 if (device->nobarriers)
2664 return 0;
2665
2666 if (wait) {
2667 bio = device->flush_bio;
2668 if (!bio)
2669 return 0;
2670
2671 wait_for_completion(&device->flush_wait);
2672
2673 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2674 printk("btrfs: disabling barriers on dev %s\n",
2675 device->name);
2676 device->nobarriers = 1;
2677 }
2678 if (!bio_flagged(bio, BIO_UPTODATE)) {
2679 ret = -EIO;
2680 }
2681
2682 /* drop the reference from the wait == 0 run */
2683 bio_put(bio);
2684 device->flush_bio = NULL;
2685
2686 return ret;
2687 }
2688
2689 /*
2690 * one reference for us, and we leave it for the
2691 * caller
2692 */
2693 device->flush_bio = NULL;;
2694 bio = bio_alloc(GFP_NOFS, 0);
2695 if (!bio)
2696 return -ENOMEM;
2697
2698 bio->bi_end_io = btrfs_end_empty_barrier;
2699 bio->bi_bdev = device->bdev;
2700 init_completion(&device->flush_wait);
2701 bio->bi_private = &device->flush_wait;
2702 device->flush_bio = bio;
2703
2704 bio_get(bio);
2705 btrfsic_submit_bio(WRITE_FLUSH, bio);
2706
2707 return 0;
2708}
2709
2710/*
2711 * send an empty flush down to each device in parallel,
2712 * then wait for them
2713 */
2714static int barrier_all_devices(struct btrfs_fs_info *info)
2715{
2716 struct list_head *head;
2717 struct btrfs_device *dev;
2718 int errors = 0;
2719 int ret;
2720
2721 /* send down all the barriers */
2722 head = &info->fs_devices->devices;
2723 list_for_each_entry_rcu(dev, head, dev_list) {
2724 if (!dev->bdev) {
2725 errors++;
2726 continue;
2727 }
2728 if (!dev->in_fs_metadata || !dev->writeable)
2729 continue;
2730
2731 ret = write_dev_flush(dev, 0);
2732 if (ret)
2733 errors++;
2734 }
2735
2736 /* wait for all the barriers */
2737 list_for_each_entry_rcu(dev, head, dev_list) {
2738 if (!dev->bdev) {
2739 errors++;
2740 continue;
2741 }
2742 if (!dev->in_fs_metadata || !dev->writeable)
2743 continue;
2744
2745 ret = write_dev_flush(dev, 1);
2746 if (ret)
2747 errors++;
2748 }
2749 if (errors)
2750 return -EIO;
2751 return 0;
2752}
2753
2754int write_all_supers(struct btrfs_root *root, int max_mirrors)
2755{
2756 struct list_head *head;
2757 struct btrfs_device *dev;
2758 struct btrfs_super_block *sb;
2759 struct btrfs_dev_item *dev_item;
2760 int ret;
2761 int do_barriers;
2762 int max_errors;
2763 int total_errors = 0;
2764 u64 flags;
2765
2766 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2767 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2768 backup_super_roots(root->fs_info);
2769
2770 sb = root->fs_info->super_for_commit;
2771 dev_item = &sb->dev_item;
2772
2773 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2774 head = &root->fs_info->fs_devices->devices;
2775
2776 if (do_barriers)
2777 barrier_all_devices(root->fs_info);
2778
2779 list_for_each_entry_rcu(dev, head, dev_list) {
2780 if (!dev->bdev) {
2781 total_errors++;
2782 continue;
2783 }
2784 if (!dev->in_fs_metadata || !dev->writeable)
2785 continue;
2786
2787 btrfs_set_stack_device_generation(dev_item, 0);
2788 btrfs_set_stack_device_type(dev_item, dev->type);
2789 btrfs_set_stack_device_id(dev_item, dev->devid);
2790 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2791 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2792 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2793 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2794 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2795 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2796 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2797
2798 flags = btrfs_super_flags(sb);
2799 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2800
2801 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2802 if (ret)
2803 total_errors++;
2804 }
2805 if (total_errors > max_errors) {
2806 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2807 total_errors);
2808 BUG();
2809 }
2810
2811 total_errors = 0;
2812 list_for_each_entry_rcu(dev, head, dev_list) {
2813 if (!dev->bdev)
2814 continue;
2815 if (!dev->in_fs_metadata || !dev->writeable)
2816 continue;
2817
2818 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2819 if (ret)
2820 total_errors++;
2821 }
2822 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2823 if (total_errors > max_errors) {
2824 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2825 total_errors);
2826 BUG();
2827 }
2828 return 0;
2829}
2830
2831int write_ctree_super(struct btrfs_trans_handle *trans,
2832 struct btrfs_root *root, int max_mirrors)
2833{
2834 int ret;
2835
2836 ret = write_all_supers(root, max_mirrors);
2837 return ret;
2838}
2839
2840int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2841{
2842 spin_lock(&fs_info->fs_roots_radix_lock);
2843 radix_tree_delete(&fs_info->fs_roots_radix,
2844 (unsigned long)root->root_key.objectid);
2845 spin_unlock(&fs_info->fs_roots_radix_lock);
2846
2847 if (btrfs_root_refs(&root->root_item) == 0)
2848 synchronize_srcu(&fs_info->subvol_srcu);
2849
2850 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2851 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2852 free_fs_root(root);
2853 return 0;
2854}
2855
2856static void free_fs_root(struct btrfs_root *root)
2857{
2858 iput(root->cache_inode);
2859 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2860 if (root->anon_dev)
2861 free_anon_bdev(root->anon_dev);
2862 free_extent_buffer(root->node);
2863 free_extent_buffer(root->commit_root);
2864 kfree(root->free_ino_ctl);
2865 kfree(root->free_ino_pinned);
2866 kfree(root->name);
2867 kfree(root);
2868}
2869
2870static int del_fs_roots(struct btrfs_fs_info *fs_info)
2871{
2872 int ret;
2873 struct btrfs_root *gang[8];
2874 int i;
2875
2876 while (!list_empty(&fs_info->dead_roots)) {
2877 gang[0] = list_entry(fs_info->dead_roots.next,
2878 struct btrfs_root, root_list);
2879 list_del(&gang[0]->root_list);
2880
2881 if (gang[0]->in_radix) {
2882 btrfs_free_fs_root(fs_info, gang[0]);
2883 } else {
2884 free_extent_buffer(gang[0]->node);
2885 free_extent_buffer(gang[0]->commit_root);
2886 kfree(gang[0]);
2887 }
2888 }
2889
2890 while (1) {
2891 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2892 (void **)gang, 0,
2893 ARRAY_SIZE(gang));
2894 if (!ret)
2895 break;
2896 for (i = 0; i < ret; i++)
2897 btrfs_free_fs_root(fs_info, gang[i]);
2898 }
2899 return 0;
2900}
2901
2902int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2903{
2904 u64 root_objectid = 0;
2905 struct btrfs_root *gang[8];
2906 int i;
2907 int ret;
2908
2909 while (1) {
2910 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2911 (void **)gang, root_objectid,
2912 ARRAY_SIZE(gang));
2913 if (!ret)
2914 break;
2915
2916 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2917 for (i = 0; i < ret; i++) {
2918 int err;
2919
2920 root_objectid = gang[i]->root_key.objectid;
2921 err = btrfs_orphan_cleanup(gang[i]);
2922 if (err)
2923 return err;
2924 }
2925 root_objectid++;
2926 }
2927 return 0;
2928}
2929
2930int btrfs_commit_super(struct btrfs_root *root)
2931{
2932 struct btrfs_trans_handle *trans;
2933 int ret;
2934
2935 mutex_lock(&root->fs_info->cleaner_mutex);
2936 btrfs_run_delayed_iputs(root);
2937 btrfs_clean_old_snapshots(root);
2938 mutex_unlock(&root->fs_info->cleaner_mutex);
2939
2940 /* wait until ongoing cleanup work done */
2941 down_write(&root->fs_info->cleanup_work_sem);
2942 up_write(&root->fs_info->cleanup_work_sem);
2943
2944 trans = btrfs_join_transaction(root);
2945 if (IS_ERR(trans))
2946 return PTR_ERR(trans);
2947 ret = btrfs_commit_transaction(trans, root);
2948 BUG_ON(ret);
2949 /* run commit again to drop the original snapshot */
2950 trans = btrfs_join_transaction(root);
2951 if (IS_ERR(trans))
2952 return PTR_ERR(trans);
2953 btrfs_commit_transaction(trans, root);
2954 ret = btrfs_write_and_wait_transaction(NULL, root);
2955 BUG_ON(ret);
2956
2957 ret = write_ctree_super(NULL, root, 0);
2958 return ret;
2959}
2960
2961int close_ctree(struct btrfs_root *root)
2962{
2963 struct btrfs_fs_info *fs_info = root->fs_info;
2964 int ret;
2965
2966 fs_info->closing = 1;
2967 smp_mb();
2968
2969 /* pause restriper - we want to resume on mount */
2970 btrfs_pause_balance(root->fs_info);
2971
2972 btrfs_scrub_cancel(root);
2973
2974 /* wait for any defraggers to finish */
2975 wait_event(fs_info->transaction_wait,
2976 (atomic_read(&fs_info->defrag_running) == 0));
2977
2978 /* clear out the rbtree of defraggable inodes */
2979 btrfs_run_defrag_inodes(fs_info);
2980
2981 /*
2982 * Here come 2 situations when btrfs is broken to flip readonly:
2983 *
2984 * 1. when btrfs flips readonly somewhere else before
2985 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2986 * and btrfs will skip to write sb directly to keep
2987 * ERROR state on disk.
2988 *
2989 * 2. when btrfs flips readonly just in btrfs_commit_super,
2990 * and in such case, btrfs cannot write sb via btrfs_commit_super,
2991 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2992 * btrfs will cleanup all FS resources first and write sb then.
2993 */
2994 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2995 ret = btrfs_commit_super(root);
2996 if (ret)
2997 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2998 }
2999
3000 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3001 ret = btrfs_error_commit_super(root);
3002 if (ret)
3003 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3004 }
3005
3006 btrfs_put_block_group_cache(fs_info);
3007
3008 kthread_stop(fs_info->transaction_kthread);
3009 kthread_stop(fs_info->cleaner_kthread);
3010
3011 fs_info->closing = 2;
3012 smp_mb();
3013
3014 if (fs_info->delalloc_bytes) {
3015 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3016 (unsigned long long)fs_info->delalloc_bytes);
3017 }
3018 if (fs_info->total_ref_cache_size) {
3019 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3020 (unsigned long long)fs_info->total_ref_cache_size);
3021 }
3022
3023 free_extent_buffer(fs_info->extent_root->node);
3024 free_extent_buffer(fs_info->extent_root->commit_root);
3025 free_extent_buffer(fs_info->tree_root->node);
3026 free_extent_buffer(fs_info->tree_root->commit_root);
3027 free_extent_buffer(fs_info->chunk_root->node);
3028 free_extent_buffer(fs_info->chunk_root->commit_root);
3029 free_extent_buffer(fs_info->dev_root->node);
3030 free_extent_buffer(fs_info->dev_root->commit_root);
3031 free_extent_buffer(fs_info->csum_root->node);
3032 free_extent_buffer(fs_info->csum_root->commit_root);
3033
3034 btrfs_free_block_groups(fs_info);
3035
3036 del_fs_roots(fs_info);
3037
3038 iput(fs_info->btree_inode);
3039
3040 btrfs_stop_workers(&fs_info->generic_worker);
3041 btrfs_stop_workers(&fs_info->fixup_workers);
3042 btrfs_stop_workers(&fs_info->delalloc_workers);
3043 btrfs_stop_workers(&fs_info->workers);
3044 btrfs_stop_workers(&fs_info->endio_workers);
3045 btrfs_stop_workers(&fs_info->endio_meta_workers);
3046 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3047 btrfs_stop_workers(&fs_info->endio_write_workers);
3048 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3049 btrfs_stop_workers(&fs_info->submit_workers);
3050 btrfs_stop_workers(&fs_info->delayed_workers);
3051 btrfs_stop_workers(&fs_info->caching_workers);
3052 btrfs_stop_workers(&fs_info->readahead_workers);
3053
3054#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3055 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3056 btrfsic_unmount(root, fs_info->fs_devices);
3057#endif
3058
3059 btrfs_close_devices(fs_info->fs_devices);
3060 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3061
3062 bdi_destroy(&fs_info->bdi);
3063 cleanup_srcu_struct(&fs_info->subvol_srcu);
3064
3065 return 0;
3066}
3067
3068int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
3069{
3070 int ret;
3071 struct inode *btree_inode = buf->pages[0]->mapping->host;
3072
3073 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
3074 NULL);
3075 if (!ret)
3076 return ret;
3077
3078 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3079 parent_transid);
3080 return !ret;
3081}
3082
3083int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3084{
3085 struct inode *btree_inode = buf->pages[0]->mapping->host;
3086 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
3087 buf);
3088}
3089
3090void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3091{
3092 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3093 u64 transid = btrfs_header_generation(buf);
3094 struct inode *btree_inode = root->fs_info->btree_inode;
3095 int was_dirty;
3096
3097 btrfs_assert_tree_locked(buf);
3098 if (transid != root->fs_info->generation) {
3099 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3100 "found %llu running %llu\n",
3101 (unsigned long long)buf->start,
3102 (unsigned long long)transid,
3103 (unsigned long long)root->fs_info->generation);
3104 WARN_ON(1);
3105 }
3106 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
3107 buf);
3108 if (!was_dirty) {
3109 spin_lock(&root->fs_info->delalloc_lock);
3110 root->fs_info->dirty_metadata_bytes += buf->len;
3111 spin_unlock(&root->fs_info->delalloc_lock);
3112 }
3113}
3114
3115void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3116{
3117 /*
3118 * looks as though older kernels can get into trouble with
3119 * this code, they end up stuck in balance_dirty_pages forever
3120 */
3121 u64 num_dirty;
3122 unsigned long thresh = 32 * 1024 * 1024;
3123
3124 if (current->flags & PF_MEMALLOC)
3125 return;
3126
3127 btrfs_balance_delayed_items(root);
3128
3129 num_dirty = root->fs_info->dirty_metadata_bytes;
3130
3131 if (num_dirty > thresh) {
3132 balance_dirty_pages_ratelimited_nr(
3133 root->fs_info->btree_inode->i_mapping, 1);
3134 }
3135 return;
3136}
3137
3138void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3139{
3140 /*
3141 * looks as though older kernels can get into trouble with
3142 * this code, they end up stuck in balance_dirty_pages forever
3143 */
3144 u64 num_dirty;
3145 unsigned long thresh = 32 * 1024 * 1024;
3146
3147 if (current->flags & PF_MEMALLOC)
3148 return;
3149
3150 num_dirty = root->fs_info->dirty_metadata_bytes;
3151
3152 if (num_dirty > thresh) {
3153 balance_dirty_pages_ratelimited_nr(
3154 root->fs_info->btree_inode->i_mapping, 1);
3155 }
3156 return;
3157}
3158
3159int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3160{
3161 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3162 int ret;
3163 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3164 if (ret == 0)
3165 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
3166 return ret;
3167}
3168
3169static int btree_lock_page_hook(struct page *page, void *data,
3170 void (*flush_fn)(void *))
3171{
3172 struct inode *inode = page->mapping->host;
3173 struct btrfs_root *root = BTRFS_I(inode)->root;
3174 struct extent_buffer *eb;
3175
3176 /*
3177 * We culled this eb but the page is still hanging out on the mapping,
3178 * carry on.
3179 */
3180 if (!PagePrivate(page))
3181 goto out;
3182
3183 eb = (struct extent_buffer *)page->private;
3184 if (!eb) {
3185 WARN_ON(1);
3186 goto out;
3187 }
3188 if (page != eb->pages[0])
3189 goto out;
3190
3191 if (!btrfs_try_tree_write_lock(eb)) {
3192 flush_fn(data);
3193 btrfs_tree_lock(eb);
3194 }
3195 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3196
3197 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3198 spin_lock(&root->fs_info->delalloc_lock);
3199 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3200 root->fs_info->dirty_metadata_bytes -= eb->len;
3201 else
3202 WARN_ON(1);
3203 spin_unlock(&root->fs_info->delalloc_lock);
3204 }
3205
3206 btrfs_tree_unlock(eb);
3207out:
3208 if (!trylock_page(page)) {
3209 flush_fn(data);
3210 lock_page(page);
3211 }
3212 return 0;
3213}
3214
3215static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3216 int read_only)
3217{
3218 if (read_only)
3219 return;
3220
3221 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3222 printk(KERN_WARNING "warning: mount fs with errors, "
3223 "running btrfsck is recommended\n");
3224}
3225
3226int btrfs_error_commit_super(struct btrfs_root *root)
3227{
3228 int ret;
3229
3230 mutex_lock(&root->fs_info->cleaner_mutex);
3231 btrfs_run_delayed_iputs(root);
3232 mutex_unlock(&root->fs_info->cleaner_mutex);
3233
3234 down_write(&root->fs_info->cleanup_work_sem);
3235 up_write(&root->fs_info->cleanup_work_sem);
3236
3237 /* cleanup FS via transaction */
3238 btrfs_cleanup_transaction(root);
3239
3240 ret = write_ctree_super(NULL, root, 0);
3241
3242 return ret;
3243}
3244
3245static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
3246{
3247 struct btrfs_inode *btrfs_inode;
3248 struct list_head splice;
3249
3250 INIT_LIST_HEAD(&splice);
3251
3252 mutex_lock(&root->fs_info->ordered_operations_mutex);
3253 spin_lock(&root->fs_info->ordered_extent_lock);
3254
3255 list_splice_init(&root->fs_info->ordered_operations, &splice);
3256 while (!list_empty(&splice)) {
3257 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3258 ordered_operations);
3259
3260 list_del_init(&btrfs_inode->ordered_operations);
3261
3262 btrfs_invalidate_inodes(btrfs_inode->root);
3263 }
3264
3265 spin_unlock(&root->fs_info->ordered_extent_lock);
3266 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3267
3268 return 0;
3269}
3270
3271static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
3272{
3273 struct list_head splice;
3274 struct btrfs_ordered_extent *ordered;
3275 struct inode *inode;
3276
3277 INIT_LIST_HEAD(&splice);
3278
3279 spin_lock(&root->fs_info->ordered_extent_lock);
3280
3281 list_splice_init(&root->fs_info->ordered_extents, &splice);
3282 while (!list_empty(&splice)) {
3283 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3284 root_extent_list);
3285
3286 list_del_init(&ordered->root_extent_list);
3287 atomic_inc(&ordered->refs);
3288
3289 /* the inode may be getting freed (in sys_unlink path). */
3290 inode = igrab(ordered->inode);
3291
3292 spin_unlock(&root->fs_info->ordered_extent_lock);
3293 if (inode)
3294 iput(inode);
3295
3296 atomic_set(&ordered->refs, 1);
3297 btrfs_put_ordered_extent(ordered);
3298
3299 spin_lock(&root->fs_info->ordered_extent_lock);
3300 }
3301
3302 spin_unlock(&root->fs_info->ordered_extent_lock);
3303
3304 return 0;
3305}
3306
3307static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3308 struct btrfs_root *root)
3309{
3310 struct rb_node *node;
3311 struct btrfs_delayed_ref_root *delayed_refs;
3312 struct btrfs_delayed_ref_node *ref;
3313 int ret = 0;
3314
3315 delayed_refs = &trans->delayed_refs;
3316
3317 spin_lock(&delayed_refs->lock);
3318 if (delayed_refs->num_entries == 0) {
3319 spin_unlock(&delayed_refs->lock);
3320 printk(KERN_INFO "delayed_refs has NO entry\n");
3321 return ret;
3322 }
3323
3324 node = rb_first(&delayed_refs->root);
3325 while (node) {
3326 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3327 node = rb_next(node);
3328
3329 ref->in_tree = 0;
3330 rb_erase(&ref->rb_node, &delayed_refs->root);
3331 delayed_refs->num_entries--;
3332
3333 atomic_set(&ref->refs, 1);
3334 if (btrfs_delayed_ref_is_head(ref)) {
3335 struct btrfs_delayed_ref_head *head;
3336
3337 head = btrfs_delayed_node_to_head(ref);
3338 mutex_lock(&head->mutex);
3339 kfree(head->extent_op);
3340 delayed_refs->num_heads--;
3341 if (list_empty(&head->cluster))
3342 delayed_refs->num_heads_ready--;
3343 list_del_init(&head->cluster);
3344 mutex_unlock(&head->mutex);
3345 }
3346
3347 spin_unlock(&delayed_refs->lock);
3348 btrfs_put_delayed_ref(ref);
3349
3350 cond_resched();
3351 spin_lock(&delayed_refs->lock);
3352 }
3353
3354 spin_unlock(&delayed_refs->lock);
3355
3356 return ret;
3357}
3358
3359static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3360{
3361 struct btrfs_pending_snapshot *snapshot;
3362 struct list_head splice;
3363
3364 INIT_LIST_HEAD(&splice);
3365
3366 list_splice_init(&t->pending_snapshots, &splice);
3367
3368 while (!list_empty(&splice)) {
3369 snapshot = list_entry(splice.next,
3370 struct btrfs_pending_snapshot,
3371 list);
3372
3373 list_del_init(&snapshot->list);
3374
3375 kfree(snapshot);
3376 }
3377
3378 return 0;
3379}
3380
3381static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3382{
3383 struct btrfs_inode *btrfs_inode;
3384 struct list_head splice;
3385
3386 INIT_LIST_HEAD(&splice);
3387
3388 spin_lock(&root->fs_info->delalloc_lock);
3389 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3390
3391 while (!list_empty(&splice)) {
3392 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3393 delalloc_inodes);
3394
3395 list_del_init(&btrfs_inode->delalloc_inodes);
3396
3397 btrfs_invalidate_inodes(btrfs_inode->root);
3398 }
3399
3400 spin_unlock(&root->fs_info->delalloc_lock);
3401
3402 return 0;
3403}
3404
3405static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3406 struct extent_io_tree *dirty_pages,
3407 int mark)
3408{
3409 int ret;
3410 struct page *page;
3411 struct inode *btree_inode = root->fs_info->btree_inode;
3412 struct extent_buffer *eb;
3413 u64 start = 0;
3414 u64 end;
3415 u64 offset;
3416 unsigned long index;
3417
3418 while (1) {
3419 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3420 mark);
3421 if (ret)
3422 break;
3423
3424 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3425 while (start <= end) {
3426 index = start >> PAGE_CACHE_SHIFT;
3427 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3428 page = find_get_page(btree_inode->i_mapping, index);
3429 if (!page)
3430 continue;
3431 offset = page_offset(page);
3432
3433 spin_lock(&dirty_pages->buffer_lock);
3434 eb = radix_tree_lookup(
3435 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3436 offset >> PAGE_CACHE_SHIFT);
3437 spin_unlock(&dirty_pages->buffer_lock);
3438 if (eb) {
3439 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3440 &eb->bflags);
3441 atomic_set(&eb->refs, 1);
3442 }
3443 if (PageWriteback(page))
3444 end_page_writeback(page);
3445
3446 lock_page(page);
3447 if (PageDirty(page)) {
3448 clear_page_dirty_for_io(page);
3449 spin_lock_irq(&page->mapping->tree_lock);
3450 radix_tree_tag_clear(&page->mapping->page_tree,
3451 page_index(page),
3452 PAGECACHE_TAG_DIRTY);
3453 spin_unlock_irq(&page->mapping->tree_lock);
3454 }
3455
3456 page->mapping->a_ops->invalidatepage(page, 0);
3457 unlock_page(page);
3458 }
3459 }
3460
3461 return ret;
3462}
3463
3464static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3465 struct extent_io_tree *pinned_extents)
3466{
3467 struct extent_io_tree *unpin;
3468 u64 start;
3469 u64 end;
3470 int ret;
3471
3472 unpin = pinned_extents;
3473 while (1) {
3474 ret = find_first_extent_bit(unpin, 0, &start, &end,
3475 EXTENT_DIRTY);
3476 if (ret)
3477 break;
3478
3479 /* opt_discard */
3480 if (btrfs_test_opt(root, DISCARD))
3481 ret = btrfs_error_discard_extent(root, start,
3482 end + 1 - start,
3483 NULL);
3484
3485 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3486 btrfs_error_unpin_extent_range(root, start, end);
3487 cond_resched();
3488 }
3489
3490 return 0;
3491}
3492
3493static int btrfs_cleanup_transaction(struct btrfs_root *root)
3494{
3495 struct btrfs_transaction *t;
3496 LIST_HEAD(list);
3497
3498 WARN_ON(1);
3499
3500 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3501
3502 spin_lock(&root->fs_info->trans_lock);
3503 list_splice_init(&root->fs_info->trans_list, &list);
3504 root->fs_info->trans_no_join = 1;
3505 spin_unlock(&root->fs_info->trans_lock);
3506
3507 while (!list_empty(&list)) {
3508 t = list_entry(list.next, struct btrfs_transaction, list);
3509 if (!t)
3510 break;
3511
3512 btrfs_destroy_ordered_operations(root);
3513
3514 btrfs_destroy_ordered_extents(root);
3515
3516 btrfs_destroy_delayed_refs(t, root);
3517
3518 btrfs_block_rsv_release(root,
3519 &root->fs_info->trans_block_rsv,
3520 t->dirty_pages.dirty_bytes);
3521
3522 /* FIXME: cleanup wait for commit */
3523 t->in_commit = 1;
3524 t->blocked = 1;
3525 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3526 wake_up(&root->fs_info->transaction_blocked_wait);
3527
3528 t->blocked = 0;
3529 if (waitqueue_active(&root->fs_info->transaction_wait))
3530 wake_up(&root->fs_info->transaction_wait);
3531
3532 t->commit_done = 1;
3533 if (waitqueue_active(&t->commit_wait))
3534 wake_up(&t->commit_wait);
3535
3536 btrfs_destroy_pending_snapshots(t);
3537
3538 btrfs_destroy_delalloc_inodes(root);
3539
3540 spin_lock(&root->fs_info->trans_lock);
3541 root->fs_info->running_transaction = NULL;
3542 spin_unlock(&root->fs_info->trans_lock);
3543
3544 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3545 EXTENT_DIRTY);
3546
3547 btrfs_destroy_pinned_extent(root,
3548 root->fs_info->pinned_extents);
3549
3550 atomic_set(&t->use_count, 0);
3551 list_del_init(&t->list);
3552 memset(t, 0, sizeof(*t));
3553 kmem_cache_free(btrfs_transaction_cachep, t);
3554 }
3555
3556 spin_lock(&root->fs_info->trans_lock);
3557 root->fs_info->trans_no_join = 0;
3558 spin_unlock(&root->fs_info->trans_lock);
3559 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3560
3561 return 0;
3562}
3563
3564static struct extent_io_ops btree_extent_io_ops = {
3565 .write_cache_pages_lock_hook = btree_lock_page_hook,
3566 .readpage_end_io_hook = btree_readpage_end_io_hook,
3567 .readpage_io_failed_hook = btree_io_failed_hook,
3568 .submit_bio_hook = btree_submit_bio_hook,
3569 /* note we're sharing with inode.c for the merge bio hook */
3570 .merge_bio_hook = btrfs_merge_bio_hook,
3571};
Linux 6.x block layer and io_uring tree(s)