projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
btrfs: fix min csum item size warnings in 32bit
[GitHub/mt8127/android_kernel_alcatel_ttab.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#include "rcu-string.h"
48
49static struct extent_io_ops btree_extent_io_ops;
50static void end_workqueue_fn(struct btrfs_work *work);
51static void free_fs_root(struct btrfs_root *root);
52static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
53 int read_only);
54static void btrfs_destroy_ordered_operations(struct btrfs_root *root);
55static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
56static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
57 struct btrfs_root *root);
58static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
59static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
60static int btrfs_destroy_marked_extents(struct btrfs_root *root,
61 struct extent_io_tree *dirty_pages,
62 int mark);
63static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
64 struct extent_io_tree *pinned_extents);
65
66/*
67 * end_io_wq structs are used to do processing in task context when an IO is
68 * complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
70 */
71struct end_io_wq {
72 struct bio *bio;
73 bio_end_io_t *end_io;
74 void *private;
75 struct btrfs_fs_info *info;
76 int error;
77 int metadata;
78 struct list_head list;
79 struct btrfs_work work;
80};
81
82/*
83 * async submit bios are used to offload expensive checksumming
84 * onto the worker threads. They checksum file and metadata bios
85 * just before they are sent down the IO stack.
86 */
87struct async_submit_bio {
88 struct inode *inode;
89 struct bio *bio;
90 struct list_head list;
91 extent_submit_bio_hook_t *submit_bio_start;
92 extent_submit_bio_hook_t *submit_bio_done;
93 int rw;
94 int mirror_num;
95 unsigned long bio_flags;
96 /*
97 * bio_offset is optional, can be used if the pages in the bio
98 * can't tell us where in the file the bio should go
99 */
100 u64 bio_offset;
101 struct btrfs_work work;
102 int error;
103};
104
105/*
106 * Lockdep class keys for extent_buffer->lock's in this root. For a given
107 * eb, the lockdep key is determined by the btrfs_root it belongs to and
108 * the level the eb occupies in the tree.
109 *
110 * Different roots are used for different purposes and may nest inside each
111 * other and they require separate keysets. As lockdep keys should be
112 * static, assign keysets according to the purpose of the root as indicated
113 * by btrfs_root->objectid. This ensures that all special purpose roots
114 * have separate keysets.
115 *
116 * Lock-nesting across peer nodes is always done with the immediate parent
117 * node locked thus preventing deadlock. As lockdep doesn't know this, use
118 * subclass to avoid triggering lockdep warning in such cases.
119 *
120 * The key is set by the readpage_end_io_hook after the buffer has passed
121 * csum validation but before the pages are unlocked. It is also set by
122 * btrfs_init_new_buffer on freshly allocated blocks.
123 *
124 * We also add a check to make sure the highest level of the tree is the
125 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
126 * needs update as well.
127 */
128#ifdef CONFIG_DEBUG_LOCK_ALLOC
129# if BTRFS_MAX_LEVEL != 8
130# error
131# endif
132
133static struct btrfs_lockdep_keyset {
134 u64 id; /* root objectid */
135 const char *name_stem; /* lock name stem */
136 char names[BTRFS_MAX_LEVEL + 1][20];
137 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
138} btrfs_lockdep_keysets[] = {
139 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
140 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
141 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
142 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
143 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
144 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
145 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
146 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
147 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
148 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
149 { .id = 0, .name_stem = "tree" },
150};
151
152void __init btrfs_init_lockdep(void)
153{
154 int i, j;
155
156 /* initialize lockdep class names */
157 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
158 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
159
160 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
161 snprintf(ks->names[j], sizeof(ks->names[j]),
162 "btrfs-%s-%02d", ks->name_stem, j);
163 }
164}
165
166void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
167 int level)
168{
169 struct btrfs_lockdep_keyset *ks;
170
171 BUG_ON(level >= ARRAY_SIZE(ks->keys));
172
173 /* find the matching keyset, id 0 is the default entry */
174 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
175 if (ks->id == objectid)
176 break;
177
178 lockdep_set_class_and_name(&eb->lock,
179 &ks->keys[level], ks->names[level]);
180}
181
182#endif
183
184/*
185 * extents on the btree inode are pretty simple, there's one extent
186 * that covers the entire device
187 */
188static struct extent_map *btree_get_extent(struct inode *inode,
189 struct page *page, size_t pg_offset, u64 start, u64 len,
190 int create)
191{
192 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
193 struct extent_map *em;
194 int ret;
195
196 read_lock(&em_tree->lock);
197 em = lookup_extent_mapping(em_tree, start, len);
198 if (em) {
199 em->bdev =
200 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
201 read_unlock(&em_tree->lock);
202 goto out;
203 }
204 read_unlock(&em_tree->lock);
205
206 em = alloc_extent_map();
207 if (!em) {
208 em = ERR_PTR(-ENOMEM);
209 goto out;
210 }
211 em->start = 0;
212 em->len = (u64)-1;
213 em->block_len = (u64)-1;
214 em->block_start = 0;
215 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
216
217 write_lock(&em_tree->lock);
218 ret = add_extent_mapping(em_tree, em);
219 if (ret == -EEXIST) {
220 free_extent_map(em);
221 em = lookup_extent_mapping(em_tree, start, len);
222 if (!em)
223 em = ERR_PTR(-EIO);
224 } else if (ret) {
225 free_extent_map(em);
226 em = ERR_PTR(ret);
227 }
228 write_unlock(&em_tree->lock);
229
230out:
231 return em;
232}
233
234u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
235{
236 return crc32c(seed, data, len);
237}
238
239void btrfs_csum_final(u32 crc, char *result)
240{
241 put_unaligned_le32(~crc, result);
242}
243
244/*
245 * compute the csum for a btree block, and either verify it or write it
246 * into the csum field of the block.
247 */
248static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
249 int verify)
250{
251 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
252 char *result = NULL;
253 unsigned long len;
254 unsigned long cur_len;
255 unsigned long offset = BTRFS_CSUM_SIZE;
256 char *kaddr;
257 unsigned long map_start;
258 unsigned long map_len;
259 int err;
260 u32 crc = ~(u32)0;
261 unsigned long inline_result;
262
263 len = buf->len - offset;
264 while (len > 0) {
265 err = map_private_extent_buffer(buf, offset, 32,
266 &kaddr, &map_start, &map_len);
267 if (err)
268 return 1;
269 cur_len = min(len, map_len - (offset - map_start));
270 crc = btrfs_csum_data(root, kaddr + offset - map_start,
271 crc, cur_len);
272 len -= cur_len;
273 offset += cur_len;
274 }
275 if (csum_size > sizeof(inline_result)) {
276 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
277 if (!result)
278 return 1;
279 } else {
280 result = (char *)&inline_result;
281 }
282
283 btrfs_csum_final(crc, result);
284
285 if (verify) {
286 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
287 u32 val;
288 u32 found = 0;
289 memcpy(&found, result, csum_size);
290
291 read_extent_buffer(buf, &val, 0, csum_size);
292 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
293 "failed on %llu wanted %X found %X "
294 "level %d\n",
295 root->fs_info->sb->s_id,
296 (unsigned long long)buf->start, val, found,
297 btrfs_header_level(buf));
298 if (result != (char *)&inline_result)
299 kfree(result);
300 return 1;
301 }
302 } else {
303 write_extent_buffer(buf, result, 0, csum_size);
304 }
305 if (result != (char *)&inline_result)
306 kfree(result);
307 return 0;
308}
309
310/*
311 * we can't consider a given block up to date unless the transid of the
312 * block matches the transid in the parent node's pointer. This is how we
313 * detect blocks that either didn't get written at all or got written
314 * in the wrong place.
315 */
316static int verify_parent_transid(struct extent_io_tree *io_tree,
317 struct extent_buffer *eb, u64 parent_transid,
318 int atomic)
319{
320 struct extent_state *cached_state = NULL;
321 int ret;
322
323 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
324 return 0;
325
326 if (atomic)
327 return -EAGAIN;
328
329 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
330 0, &cached_state);
331 if (extent_buffer_uptodate(eb) &&
332 btrfs_header_generation(eb) == parent_transid) {
333 ret = 0;
334 goto out;
335 }
336 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
337 "found %llu\n",
338 (unsigned long long)eb->start,
339 (unsigned long long)parent_transid,
340 (unsigned long long)btrfs_header_generation(eb));
341 ret = 1;
342 clear_extent_buffer_uptodate(eb);
343out:
344 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
345 &cached_state, GFP_NOFS);
346 return ret;
347}
348
349/*
350 * helper to read a given tree block, doing retries as required when
351 * the checksums don't match and we have alternate mirrors to try.
352 */
353static int btree_read_extent_buffer_pages(struct btrfs_root *root,
354 struct extent_buffer *eb,
355 u64 start, u64 parent_transid)
356{
357 struct extent_io_tree *io_tree;
358 int failed = 0;
359 int ret;
360 int num_copies = 0;
361 int mirror_num = 0;
362 int failed_mirror = 0;
363
364 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
365 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
366 while (1) {
367 ret = read_extent_buffer_pages(io_tree, eb, start,
368 WAIT_COMPLETE,
369 btree_get_extent, mirror_num);
370 if (!ret) {
371 if (!verify_parent_transid(io_tree, eb,
372 parent_transid, 0))
373 break;
374 else
375 ret = -EIO;
376 }
377
378 /*
379 * This buffer's crc is fine, but its contents are corrupted, so
380 * there is no reason to read the other copies, they won't be
381 * any less wrong.
382 */
383 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
384 break;
385
386 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
387 eb->start, eb->len);
388 if (num_copies == 1)
389 break;
390
391 if (!failed_mirror) {
392 failed = 1;
393 failed_mirror = eb->read_mirror;
394 }
395
396 mirror_num++;
397 if (mirror_num == failed_mirror)
398 mirror_num++;
399
400 if (mirror_num > num_copies)
401 break;
402 }
403
404 if (failed && !ret && failed_mirror)
405 repair_eb_io_failure(root, eb, failed_mirror);
406
407 return ret;
408}
409
410/*
411 * checksum a dirty tree block before IO. This has extra checks to make sure
412 * we only fill in the checksum field in the first page of a multi-page block
413 */
414
415static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
416{
417 struct extent_io_tree *tree;
418 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
419 u64 found_start;
420 struct extent_buffer *eb;
421
422 tree = &BTRFS_I(page->mapping->host)->io_tree;
423
424 eb = (struct extent_buffer *)page->private;
425 if (page != eb->pages[0])
426 return 0;
427 found_start = btrfs_header_bytenr(eb);
428 if (found_start != start) {
429 WARN_ON(1);
430 return 0;
431 }
432 if (eb->pages[0] != page) {
433 WARN_ON(1);
434 return 0;
435 }
436 if (!PageUptodate(page)) {
437 WARN_ON(1);
438 return 0;
439 }
440 csum_tree_block(root, eb, 0);
441 return 0;
442}
443
444static int check_tree_block_fsid(struct btrfs_root *root,
445 struct extent_buffer *eb)
446{
447 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
448 u8 fsid[BTRFS_UUID_SIZE];
449 int ret = 1;
450
451 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
452 BTRFS_FSID_SIZE);
453 while (fs_devices) {
454 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
455 ret = 0;
456 break;
457 }
458 fs_devices = fs_devices->seed;
459 }
460 return ret;
461}
462
463#define CORRUPT(reason, eb, root, slot) \
464 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
465 "root=%llu, slot=%d\n", reason, \
466 (unsigned long long)btrfs_header_bytenr(eb), \
467 (unsigned long long)root->objectid, slot)
468
469static noinline int check_leaf(struct btrfs_root *root,
470 struct extent_buffer *leaf)
471{
472 struct btrfs_key key;
473 struct btrfs_key leaf_key;
474 u32 nritems = btrfs_header_nritems(leaf);
475 int slot;
476
477 if (nritems == 0)
478 return 0;
479
480 /* Check the 0 item */
481 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
482 BTRFS_LEAF_DATA_SIZE(root)) {
483 CORRUPT("invalid item offset size pair", leaf, root, 0);
484 return -EIO;
485 }
486
487 /*
488 * Check to make sure each items keys are in the correct order and their
489 * offsets make sense. We only have to loop through nritems-1 because
490 * we check the current slot against the next slot, which verifies the
491 * next slot's offset+size makes sense and that the current's slot
492 * offset is correct.
493 */
494 for (slot = 0; slot < nritems - 1; slot++) {
495 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
496 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
497
498 /* Make sure the keys are in the right order */
499 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
500 CORRUPT("bad key order", leaf, root, slot);
501 return -EIO;
502 }
503
504 /*
505 * Make sure the offset and ends are right, remember that the
506 * item data starts at the end of the leaf and grows towards the
507 * front.
508 */
509 if (btrfs_item_offset_nr(leaf, slot) !=
510 btrfs_item_end_nr(leaf, slot + 1)) {
511 CORRUPT("slot offset bad", leaf, root, slot);
512 return -EIO;
513 }
514
515 /*
516 * Check to make sure that we don't point outside of the leaf,
517 * just incase all the items are consistent to eachother, but
518 * all point outside of the leaf.
519 */
520 if (btrfs_item_end_nr(leaf, slot) >
521 BTRFS_LEAF_DATA_SIZE(root)) {
522 CORRUPT("slot end outside of leaf", leaf, root, slot);
523 return -EIO;
524 }
525 }
526
527 return 0;
528}
529
530struct extent_buffer *find_eb_for_page(struct extent_io_tree *tree,
531 struct page *page, int max_walk)
532{
533 struct extent_buffer *eb;
534 u64 start = page_offset(page);
535 u64 target = start;
536 u64 min_start;
537
538 if (start < max_walk)
539 min_start = 0;
540 else
541 min_start = start - max_walk;
542
543 while (start >= min_start) {
544 eb = find_extent_buffer(tree, start, 0);
545 if (eb) {
546 /*
547 * we found an extent buffer and it contains our page
548 * horray!
549 */
550 if (eb->start <= target &&
551 eb->start + eb->len > target)
552 return eb;
553
554 /* we found an extent buffer that wasn't for us */
555 free_extent_buffer(eb);
556 return NULL;
557 }
558 if (start == 0)
559 break;
560 start -= PAGE_CACHE_SIZE;
561 }
562 return NULL;
563}
564
565static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
566 struct extent_state *state, int mirror)
567{
568 struct extent_io_tree *tree;
569 u64 found_start;
570 int found_level;
571 struct extent_buffer *eb;
572 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
573 int ret = 0;
574 int reads_done;
575
576 if (!page->private)
577 goto out;
578
579 tree = &BTRFS_I(page->mapping->host)->io_tree;
580 eb = (struct extent_buffer *)page->private;
581
582 /* the pending IO might have been the only thing that kept this buffer
583 * in memory. Make sure we have a ref for all this other checks
584 */
585 extent_buffer_get(eb);
586
587 reads_done = atomic_dec_and_test(&eb->io_pages);
588 if (!reads_done)
589 goto err;
590
591 eb->read_mirror = mirror;
592 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
593 ret = -EIO;
594 goto err;
595 }
596
597 found_start = btrfs_header_bytenr(eb);
598 if (found_start != eb->start) {
599 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
600 "%llu %llu\n",
601 (unsigned long long)found_start,
602 (unsigned long long)eb->start);
603 ret = -EIO;
604 goto err;
605 }
606 if (check_tree_block_fsid(root, eb)) {
607 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
608 (unsigned long long)eb->start);
609 ret = -EIO;
610 goto err;
611 }
612 found_level = btrfs_header_level(eb);
613
614 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
615 eb, found_level);
616
617 ret = csum_tree_block(root, eb, 1);
618 if (ret) {
619 ret = -EIO;
620 goto err;
621 }
622
623 /*
624 * If this is a leaf block and it is corrupt, set the corrupt bit so
625 * that we don't try and read the other copies of this block, just
626 * return -EIO.
627 */
628 if (found_level == 0 && check_leaf(root, eb)) {
629 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
630 ret = -EIO;
631 }
632
633 if (!ret)
634 set_extent_buffer_uptodate(eb);
635err:
636 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
637 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
638 btree_readahead_hook(root, eb, eb->start, ret);
639 }
640
641 if (ret)
642 clear_extent_buffer_uptodate(eb);
643 free_extent_buffer(eb);
644out:
645 return ret;
646}
647
648static int btree_io_failed_hook(struct page *page, int failed_mirror)
649{
650 struct extent_buffer *eb;
651 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
652
653 eb = (struct extent_buffer *)page->private;
654 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
655 eb->read_mirror = failed_mirror;
656 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
657 btree_readahead_hook(root, eb, eb->start, -EIO);
658 return -EIO; /* we fixed nothing */
659}
660
661static void end_workqueue_bio(struct bio *bio, int err)
662{
663 struct end_io_wq *end_io_wq = bio->bi_private;
664 struct btrfs_fs_info *fs_info;
665
666 fs_info = end_io_wq->info;
667 end_io_wq->error = err;
668 end_io_wq->work.func = end_workqueue_fn;
669 end_io_wq->work.flags = 0;
670
671 if (bio->bi_rw & REQ_WRITE) {
672 if (end_io_wq->metadata == 1)
673 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
674 &end_io_wq->work);
675 else if (end_io_wq->metadata == 2)
676 btrfs_queue_worker(&fs_info->endio_freespace_worker,
677 &end_io_wq->work);
678 else
679 btrfs_queue_worker(&fs_info->endio_write_workers,
680 &end_io_wq->work);
681 } else {
682 if (end_io_wq->metadata)
683 btrfs_queue_worker(&fs_info->endio_meta_workers,
684 &end_io_wq->work);
685 else
686 btrfs_queue_worker(&fs_info->endio_workers,
687 &end_io_wq->work);
688 }
689}
690
691/*
692 * For the metadata arg you want
693 *
694 * 0 - if data
695 * 1 - if normal metadta
696 * 2 - if writing to the free space cache area
697 */
698int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
699 int metadata)
700{
701 struct end_io_wq *end_io_wq;
702 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
703 if (!end_io_wq)
704 return -ENOMEM;
705
706 end_io_wq->private = bio->bi_private;
707 end_io_wq->end_io = bio->bi_end_io;
708 end_io_wq->info = info;
709 end_io_wq->error = 0;
710 end_io_wq->bio = bio;
711 end_io_wq->metadata = metadata;
712
713 bio->bi_private = end_io_wq;
714 bio->bi_end_io = end_workqueue_bio;
715 return 0;
716}
717
718unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
719{
720 unsigned long limit = min_t(unsigned long,
721 info->workers.max_workers,
722 info->fs_devices->open_devices);
723 return 256 * limit;
724}
725
726static void run_one_async_start(struct btrfs_work *work)
727{
728 struct async_submit_bio *async;
729 int ret;
730
731 async = container_of(work, struct async_submit_bio, work);
732 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
733 async->mirror_num, async->bio_flags,
734 async->bio_offset);
735 if (ret)
736 async->error = ret;
737}
738
739static void run_one_async_done(struct btrfs_work *work)
740{
741 struct btrfs_fs_info *fs_info;
742 struct async_submit_bio *async;
743 int limit;
744
745 async = container_of(work, struct async_submit_bio, work);
746 fs_info = BTRFS_I(async->inode)->root->fs_info;
747
748 limit = btrfs_async_submit_limit(fs_info);
749 limit = limit * 2 / 3;
750
751 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
752 waitqueue_active(&fs_info->async_submit_wait))
753 wake_up(&fs_info->async_submit_wait);
754
755 /* If an error occured we just want to clean up the bio and move on */
756 if (async->error) {
757 bio_endio(async->bio, async->error);
758 return;
759 }
760
761 async->submit_bio_done(async->inode, async->rw, async->bio,
762 async->mirror_num, async->bio_flags,
763 async->bio_offset);
764}
765
766static void run_one_async_free(struct btrfs_work *work)
767{
768 struct async_submit_bio *async;
769
770 async = container_of(work, struct async_submit_bio, work);
771 kfree(async);
772}
773
774int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
775 int rw, struct bio *bio, int mirror_num,
776 unsigned long bio_flags,
777 u64 bio_offset,
778 extent_submit_bio_hook_t *submit_bio_start,
779 extent_submit_bio_hook_t *submit_bio_done)
780{
781 struct async_submit_bio *async;
782
783 async = kmalloc(sizeof(*async), GFP_NOFS);
784 if (!async)
785 return -ENOMEM;
786
787 async->inode = inode;
788 async->rw = rw;
789 async->bio = bio;
790 async->mirror_num = mirror_num;
791 async->submit_bio_start = submit_bio_start;
792 async->submit_bio_done = submit_bio_done;
793
794 async->work.func = run_one_async_start;
795 async->work.ordered_func = run_one_async_done;
796 async->work.ordered_free = run_one_async_free;
797
798 async->work.flags = 0;
799 async->bio_flags = bio_flags;
800 async->bio_offset = bio_offset;
801
802 async->error = 0;
803
804 atomic_inc(&fs_info->nr_async_submits);
805
806 if (rw & REQ_SYNC)
807 btrfs_set_work_high_prio(&async->work);
808
809 btrfs_queue_worker(&fs_info->workers, &async->work);
810
811 while (atomic_read(&fs_info->async_submit_draining) &&
812 atomic_read(&fs_info->nr_async_submits)) {
813 wait_event(fs_info->async_submit_wait,
814 (atomic_read(&fs_info->nr_async_submits) == 0));
815 }
816
817 return 0;
818}
819
820static int btree_csum_one_bio(struct bio *bio)
821{
822 struct bio_vec *bvec = bio->bi_io_vec;
823 int bio_index = 0;
824 struct btrfs_root *root;
825 int ret = 0;
826
827 WARN_ON(bio->bi_vcnt <= 0);
828 while (bio_index < bio->bi_vcnt) {
829 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
830 ret = csum_dirty_buffer(root, bvec->bv_page);
831 if (ret)
832 break;
833 bio_index++;
834 bvec++;
835 }
836 return ret;
837}
838
839static int __btree_submit_bio_start(struct inode *inode, int rw,
840 struct bio *bio, int mirror_num,
841 unsigned long bio_flags,
842 u64 bio_offset)
843{
844 /*
845 * when we're called for a write, we're already in the async
846 * submission context. Just jump into btrfs_map_bio
847 */
848 return btree_csum_one_bio(bio);
849}
850
851static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
852 int mirror_num, unsigned long bio_flags,
853 u64 bio_offset)
854{
855 /*
856 * when we're called for a write, we're already in the async
857 * submission context. Just jump into btrfs_map_bio
858 */
859 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
860}
861
862static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
863 int mirror_num, unsigned long bio_flags,
864 u64 bio_offset)
865{
866 int ret;
867
868 if (!(rw & REQ_WRITE)) {
869
870 /*
871 * called for a read, do the setup so that checksum validation
872 * can happen in the async kernel threads
873 */
874 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
875 bio, 1);
876 if (ret)
877 return ret;
878 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
879 mirror_num, 0);
880 }
881
882 /*
883 * kthread helpers are used to submit writes so that checksumming
884 * can happen in parallel across all CPUs
885 */
886 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
887 inode, rw, bio, mirror_num, 0,
888 bio_offset,
889 __btree_submit_bio_start,
890 __btree_submit_bio_done);
891}
892
893#ifdef CONFIG_MIGRATION
894static int btree_migratepage(struct address_space *mapping,
895 struct page *newpage, struct page *page,
896 enum migrate_mode mode)
897{
898 /*
899 * we can't safely write a btree page from here,
900 * we haven't done the locking hook
901 */
902 if (PageDirty(page))
903 return -EAGAIN;
904 /*
905 * Buffers may be managed in a filesystem specific way.
906 * We must have no buffers or drop them.
907 */
908 if (page_has_private(page) &&
909 !try_to_release_page(page, GFP_KERNEL))
910 return -EAGAIN;
911 return migrate_page(mapping, newpage, page, mode);
912}
913#endif
914
915
916static int btree_writepages(struct address_space *mapping,
917 struct writeback_control *wbc)
918{
919 struct extent_io_tree *tree;
920 tree = &BTRFS_I(mapping->host)->io_tree;
921 if (wbc->sync_mode == WB_SYNC_NONE) {
922 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
923 u64 num_dirty;
924 unsigned long thresh = 32 * 1024 * 1024;
925
926 if (wbc->for_kupdate)
927 return 0;
928
929 /* this is a bit racy, but that's ok */
930 num_dirty = root->fs_info->dirty_metadata_bytes;
931 if (num_dirty < thresh)
932 return 0;
933 }
934 return btree_write_cache_pages(mapping, wbc);
935}
936
937static int btree_readpage(struct file *file, struct page *page)
938{
939 struct extent_io_tree *tree;
940 tree = &BTRFS_I(page->mapping->host)->io_tree;
941 return extent_read_full_page(tree, page, btree_get_extent, 0);
942}
943
944static int btree_releasepage(struct page *page, gfp_t gfp_flags)
945{
946 if (PageWriteback(page) || PageDirty(page))
947 return 0;
948 /*
949 * We need to mask out eg. __GFP_HIGHMEM and __GFP_DMA32 as we're doing
950 * slab allocation from alloc_extent_state down the callchain where
951 * it'd hit a BUG_ON as those flags are not allowed.
952 */
953 gfp_flags &= ~GFP_SLAB_BUG_MASK;
954
955 return try_release_extent_buffer(page, gfp_flags);
956}
957
958static void btree_invalidatepage(struct page *page, unsigned long offset)
959{
960 struct extent_io_tree *tree;
961 tree = &BTRFS_I(page->mapping->host)->io_tree;
962 extent_invalidatepage(tree, page, offset);
963 btree_releasepage(page, GFP_NOFS);
964 if (PagePrivate(page)) {
965 printk(KERN_WARNING "btrfs warning page private not zero "
966 "on page %llu\n", (unsigned long long)page_offset(page));
967 ClearPagePrivate(page);
968 set_page_private(page, 0);
969 page_cache_release(page);
970 }
971}
972
973static int btree_set_page_dirty(struct page *page)
974{
975 struct extent_buffer *eb;
976
977 BUG_ON(!PagePrivate(page));
978 eb = (struct extent_buffer *)page->private;
979 BUG_ON(!eb);
980 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
981 BUG_ON(!atomic_read(&eb->refs));
982 btrfs_assert_tree_locked(eb);
983 return __set_page_dirty_nobuffers(page);
984}
985
986static const struct address_space_operations btree_aops = {
987 .readpage = btree_readpage,
988 .writepages = btree_writepages,
989 .releasepage = btree_releasepage,
990 .invalidatepage = btree_invalidatepage,
991#ifdef CONFIG_MIGRATION
992 .migratepage = btree_migratepage,
993#endif
994 .set_page_dirty = btree_set_page_dirty,
995};
996
997int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
998 u64 parent_transid)
999{
1000 struct extent_buffer *buf = NULL;
1001 struct inode *btree_inode = root->fs_info->btree_inode;
1002 int ret = 0;
1003
1004 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1005 if (!buf)
1006 return 0;
1007 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1008 buf, 0, WAIT_NONE, btree_get_extent, 0);
1009 free_extent_buffer(buf);
1010 return ret;
1011}
1012
1013int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1014 int mirror_num, struct extent_buffer **eb)
1015{
1016 struct extent_buffer *buf = NULL;
1017 struct inode *btree_inode = root->fs_info->btree_inode;
1018 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1019 int ret;
1020
1021 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1022 if (!buf)
1023 return 0;
1024
1025 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1026
1027 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1028 btree_get_extent, mirror_num);
1029 if (ret) {
1030 free_extent_buffer(buf);
1031 return ret;
1032 }
1033
1034 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1035 free_extent_buffer(buf);
1036 return -EIO;
1037 } else if (extent_buffer_uptodate(buf)) {
1038 *eb = buf;
1039 } else {
1040 free_extent_buffer(buf);
1041 }
1042 return 0;
1043}
1044
1045struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1046 u64 bytenr, u32 blocksize)
1047{
1048 struct inode *btree_inode = root->fs_info->btree_inode;
1049 struct extent_buffer *eb;
1050 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1051 bytenr, blocksize);
1052 return eb;
1053}
1054
1055struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1056 u64 bytenr, u32 blocksize)
1057{
1058 struct inode *btree_inode = root->fs_info->btree_inode;
1059 struct extent_buffer *eb;
1060
1061 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1062 bytenr, blocksize);
1063 return eb;
1064}
1065
1066
1067int btrfs_write_tree_block(struct extent_buffer *buf)
1068{
1069 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1070 buf->start + buf->len - 1);
1071}
1072
1073int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1074{
1075 return filemap_fdatawait_range(buf->pages[0]->mapping,
1076 buf->start, buf->start + buf->len - 1);
1077}
1078
1079struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1080 u32 blocksize, u64 parent_transid)
1081{
1082 struct extent_buffer *buf = NULL;
1083 int ret;
1084
1085 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1086 if (!buf)
1087 return NULL;
1088
1089 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1090 return buf;
1091
1092}
1093
1094void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1095 struct extent_buffer *buf)
1096{
1097 if (btrfs_header_generation(buf) ==
1098 root->fs_info->running_transaction->transid) {
1099 btrfs_assert_tree_locked(buf);
1100
1101 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1102 spin_lock(&root->fs_info->delalloc_lock);
1103 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1104 root->fs_info->dirty_metadata_bytes -= buf->len;
1105 else {
1106 spin_unlock(&root->fs_info->delalloc_lock);
1107 btrfs_panic(root->fs_info, -EOVERFLOW,
1108 "Can't clear %lu bytes from "
1109 " dirty_mdatadata_bytes (%llu)",
1110 buf->len,
1111 root->fs_info->dirty_metadata_bytes);
1112 }
1113 spin_unlock(&root->fs_info->delalloc_lock);
1114 }
1115
1116 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1117 btrfs_set_lock_blocking(buf);
1118 clear_extent_buffer_dirty(buf);
1119 }
1120}
1121
1122static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1123 u32 stripesize, struct btrfs_root *root,
1124 struct btrfs_fs_info *fs_info,
1125 u64 objectid)
1126{
1127 root->node = NULL;
1128 root->commit_root = NULL;
1129 root->sectorsize = sectorsize;
1130 root->nodesize = nodesize;
1131 root->leafsize = leafsize;
1132 root->stripesize = stripesize;
1133 root->ref_cows = 0;
1134 root->track_dirty = 0;
1135 root->in_radix = 0;
1136 root->orphan_item_inserted = 0;
1137 root->orphan_cleanup_state = 0;
1138
1139 root->objectid = objectid;
1140 root->last_trans = 0;
1141 root->highest_objectid = 0;
1142 root->name = NULL;
1143 root->inode_tree = RB_ROOT;
1144 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1145 root->block_rsv = NULL;
1146 root->orphan_block_rsv = NULL;
1147
1148 INIT_LIST_HEAD(&root->dirty_list);
1149 INIT_LIST_HEAD(&root->root_list);
1150 spin_lock_init(&root->orphan_lock);
1151 spin_lock_init(&root->inode_lock);
1152 spin_lock_init(&root->accounting_lock);
1153 mutex_init(&root->objectid_mutex);
1154 mutex_init(&root->log_mutex);
1155 init_waitqueue_head(&root->log_writer_wait);
1156 init_waitqueue_head(&root->log_commit_wait[0]);
1157 init_waitqueue_head(&root->log_commit_wait[1]);
1158 atomic_set(&root->log_commit[0], 0);
1159 atomic_set(&root->log_commit[1], 0);
1160 atomic_set(&root->log_writers, 0);
1161 atomic_set(&root->log_batch, 0);
1162 atomic_set(&root->orphan_inodes, 0);
1163 root->log_transid = 0;
1164 root->last_log_commit = 0;
1165 extent_io_tree_init(&root->dirty_log_pages,
1166 fs_info->btree_inode->i_mapping);
1167
1168 memset(&root->root_key, 0, sizeof(root->root_key));
1169 memset(&root->root_item, 0, sizeof(root->root_item));
1170 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1171 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1172 root->defrag_trans_start = fs_info->generation;
1173 init_completion(&root->kobj_unregister);
1174 root->defrag_running = 0;
1175 root->root_key.objectid = objectid;
1176 root->anon_dev = 0;
1177
1178 spin_lock_init(&root->root_times_lock);
1179}
1180
1181static int __must_check find_and_setup_root(struct btrfs_root *tree_root,
1182 struct btrfs_fs_info *fs_info,
1183 u64 objectid,
1184 struct btrfs_root *root)
1185{
1186 int ret;
1187 u32 blocksize;
1188 u64 generation;
1189
1190 __setup_root(tree_root->nodesize, tree_root->leafsize,
1191 tree_root->sectorsize, tree_root->stripesize,
1192 root, fs_info, objectid);
1193 ret = btrfs_find_last_root(tree_root, objectid,
1194 &root->root_item, &root->root_key);
1195 if (ret > 0)
1196 return -ENOENT;
1197 else if (ret < 0)
1198 return ret;
1199
1200 generation = btrfs_root_generation(&root->root_item);
1201 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1202 root->commit_root = NULL;
1203 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1204 blocksize, generation);
1205 if (!root->node || !btrfs_buffer_uptodate(root->node, generation, 0)) {
1206 free_extent_buffer(root->node);
1207 root->node = NULL;
1208 return -EIO;
1209 }
1210 root->commit_root = btrfs_root_node(root);
1211 return 0;
1212}
1213
1214static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1215{
1216 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1217 if (root)
1218 root->fs_info = fs_info;
1219 return root;
1220}
1221
1222struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1223 struct btrfs_fs_info *fs_info,
1224 u64 objectid)
1225{
1226 struct extent_buffer *leaf;
1227 struct btrfs_root *tree_root = fs_info->tree_root;
1228 struct btrfs_root *root;
1229 struct btrfs_key key;
1230 int ret = 0;
1231 u64 bytenr;
1232
1233 root = btrfs_alloc_root(fs_info);
1234 if (!root)
1235 return ERR_PTR(-ENOMEM);
1236
1237 __setup_root(tree_root->nodesize, tree_root->leafsize,
1238 tree_root->sectorsize, tree_root->stripesize,
1239 root, fs_info, objectid);
1240 root->root_key.objectid = objectid;
1241 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1242 root->root_key.offset = 0;
1243
1244 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1245 0, objectid, NULL, 0, 0, 0);
1246 if (IS_ERR(leaf)) {
1247 ret = PTR_ERR(leaf);
1248 goto fail;
1249 }
1250
1251 bytenr = leaf->start;
1252 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1253 btrfs_set_header_bytenr(leaf, leaf->start);
1254 btrfs_set_header_generation(leaf, trans->transid);
1255 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1256 btrfs_set_header_owner(leaf, objectid);
1257 root->node = leaf;
1258
1259 write_extent_buffer(leaf, fs_info->fsid,
1260 (unsigned long)btrfs_header_fsid(leaf),
1261 BTRFS_FSID_SIZE);
1262 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1263 (unsigned long)btrfs_header_chunk_tree_uuid(leaf),
1264 BTRFS_UUID_SIZE);
1265 btrfs_mark_buffer_dirty(leaf);
1266
1267 root->commit_root = btrfs_root_node(root);
1268 root->track_dirty = 1;
1269
1270
1271 root->root_item.flags = 0;
1272 root->root_item.byte_limit = 0;
1273 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1274 btrfs_set_root_generation(&root->root_item, trans->transid);
1275 btrfs_set_root_level(&root->root_item, 0);
1276 btrfs_set_root_refs(&root->root_item, 1);
1277 btrfs_set_root_used(&root->root_item, leaf->len);
1278 btrfs_set_root_last_snapshot(&root->root_item, 0);
1279 btrfs_set_root_dirid(&root->root_item, 0);
1280 root->root_item.drop_level = 0;
1281
1282 key.objectid = objectid;
1283 key.type = BTRFS_ROOT_ITEM_KEY;
1284 key.offset = 0;
1285 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1286 if (ret)
1287 goto fail;
1288
1289 btrfs_tree_unlock(leaf);
1290
1291fail:
1292 if (ret)
1293 return ERR_PTR(ret);
1294
1295 return root;
1296}
1297
1298static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1299 struct btrfs_fs_info *fs_info)
1300{
1301 struct btrfs_root *root;
1302 struct btrfs_root *tree_root = fs_info->tree_root;
1303 struct extent_buffer *leaf;
1304
1305 root = btrfs_alloc_root(fs_info);
1306 if (!root)
1307 return ERR_PTR(-ENOMEM);
1308
1309 __setup_root(tree_root->nodesize, tree_root->leafsize,
1310 tree_root->sectorsize, tree_root->stripesize,
1311 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1312
1313 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1314 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1315 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1316 /*
1317 * log trees do not get reference counted because they go away
1318 * before a real commit is actually done. They do store pointers
1319 * to file data extents, and those reference counts still get
1320 * updated (along with back refs to the log tree).
1321 */
1322 root->ref_cows = 0;
1323
1324 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1325 BTRFS_TREE_LOG_OBJECTID, NULL,
1326 0, 0, 0);
1327 if (IS_ERR(leaf)) {
1328 kfree(root);
1329 return ERR_CAST(leaf);
1330 }
1331
1332 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1333 btrfs_set_header_bytenr(leaf, leaf->start);
1334 btrfs_set_header_generation(leaf, trans->transid);
1335 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1336 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1337 root->node = leaf;
1338
1339 write_extent_buffer(root->node, root->fs_info->fsid,
1340 (unsigned long)btrfs_header_fsid(root->node),
1341 BTRFS_FSID_SIZE);
1342 btrfs_mark_buffer_dirty(root->node);
1343 btrfs_tree_unlock(root->node);
1344 return root;
1345}
1346
1347int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1348 struct btrfs_fs_info *fs_info)
1349{
1350 struct btrfs_root *log_root;
1351
1352 log_root = alloc_log_tree(trans, fs_info);
1353 if (IS_ERR(log_root))
1354 return PTR_ERR(log_root);
1355 WARN_ON(fs_info->log_root_tree);
1356 fs_info->log_root_tree = log_root;
1357 return 0;
1358}
1359
1360int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1361 struct btrfs_root *root)
1362{
1363 struct btrfs_root *log_root;
1364 struct btrfs_inode_item *inode_item;
1365
1366 log_root = alloc_log_tree(trans, root->fs_info);
1367 if (IS_ERR(log_root))
1368 return PTR_ERR(log_root);
1369
1370 log_root->last_trans = trans->transid;
1371 log_root->root_key.offset = root->root_key.objectid;
1372
1373 inode_item = &log_root->root_item.inode;
1374 inode_item->generation = cpu_to_le64(1);
1375 inode_item->size = cpu_to_le64(3);
1376 inode_item->nlink = cpu_to_le32(1);
1377 inode_item->nbytes = cpu_to_le64(root->leafsize);
1378 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1379
1380 btrfs_set_root_node(&log_root->root_item, log_root->node);
1381
1382 WARN_ON(root->log_root);
1383 root->log_root = log_root;
1384 root->log_transid = 0;
1385 root->last_log_commit = 0;
1386 return 0;
1387}
1388
1389struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1390 struct btrfs_key *location)
1391{
1392 struct btrfs_root *root;
1393 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1394 struct btrfs_path *path;
1395 struct extent_buffer *l;
1396 u64 generation;
1397 u32 blocksize;
1398 int ret = 0;
1399 int slot;
1400
1401 root = btrfs_alloc_root(fs_info);
1402 if (!root)
1403 return ERR_PTR(-ENOMEM);
1404 if (location->offset == (u64)-1) {
1405 ret = find_and_setup_root(tree_root, fs_info,
1406 location->objectid, root);
1407 if (ret) {
1408 kfree(root);
1409 return ERR_PTR(ret);
1410 }
1411 goto out;
1412 }
1413
1414 __setup_root(tree_root->nodesize, tree_root->leafsize,
1415 tree_root->sectorsize, tree_root->stripesize,
1416 root, fs_info, location->objectid);
1417
1418 path = btrfs_alloc_path();
1419 if (!path) {
1420 kfree(root);
1421 return ERR_PTR(-ENOMEM);
1422 }
1423 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1424 if (ret == 0) {
1425 l = path->nodes[0];
1426 slot = path->slots[0];
1427 btrfs_read_root_item(tree_root, l, slot, &root->root_item);
1428 memcpy(&root->root_key, location, sizeof(*location));
1429 }
1430 btrfs_free_path(path);
1431 if (ret) {
1432 kfree(root);
1433 if (ret > 0)
1434 ret = -ENOENT;
1435 return ERR_PTR(ret);
1436 }
1437
1438 generation = btrfs_root_generation(&root->root_item);
1439 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1440 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1441 blocksize, generation);
1442 root->commit_root = btrfs_root_node(root);
1443 BUG_ON(!root->node); /* -ENOMEM */
1444out:
1445 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1446 root->ref_cows = 1;
1447 btrfs_check_and_init_root_item(&root->root_item);
1448 }
1449
1450 return root;
1451}
1452
1453struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1454 struct btrfs_key *location)
1455{
1456 struct btrfs_root *root;
1457 int ret;
1458
1459 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1460 return fs_info->tree_root;
1461 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1462 return fs_info->extent_root;
1463 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1464 return fs_info->chunk_root;
1465 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1466 return fs_info->dev_root;
1467 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1468 return fs_info->csum_root;
1469 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1470 return fs_info->quota_root ? fs_info->quota_root :
1471 ERR_PTR(-ENOENT);
1472again:
1473 spin_lock(&fs_info->fs_roots_radix_lock);
1474 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1475 (unsigned long)location->objectid);
1476 spin_unlock(&fs_info->fs_roots_radix_lock);
1477 if (root)
1478 return root;
1479
1480 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1481 if (IS_ERR(root))
1482 return root;
1483
1484 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1485 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1486 GFP_NOFS);
1487 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1488 ret = -ENOMEM;
1489 goto fail;
1490 }
1491
1492 btrfs_init_free_ino_ctl(root);
1493 mutex_init(&root->fs_commit_mutex);
1494 spin_lock_init(&root->cache_lock);
1495 init_waitqueue_head(&root->cache_wait);
1496
1497 ret = get_anon_bdev(&root->anon_dev);
1498 if (ret)
1499 goto fail;
1500
1501 if (btrfs_root_refs(&root->root_item) == 0) {
1502 ret = -ENOENT;
1503 goto fail;
1504 }
1505
1506 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1507 if (ret < 0)
1508 goto fail;
1509 if (ret == 0)
1510 root->orphan_item_inserted = 1;
1511
1512 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1513 if (ret)
1514 goto fail;
1515
1516 spin_lock(&fs_info->fs_roots_radix_lock);
1517 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1518 (unsigned long)root->root_key.objectid,
1519 root);
1520 if (ret == 0)
1521 root->in_radix = 1;
1522
1523 spin_unlock(&fs_info->fs_roots_radix_lock);
1524 radix_tree_preload_end();
1525 if (ret) {
1526 if (ret == -EEXIST) {
1527 free_fs_root(root);
1528 goto again;
1529 }
1530 goto fail;
1531 }
1532
1533 ret = btrfs_find_dead_roots(fs_info->tree_root,
1534 root->root_key.objectid);
1535 WARN_ON(ret);
1536 return root;
1537fail:
1538 free_fs_root(root);
1539 return ERR_PTR(ret);
1540}
1541
1542static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1543{
1544 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1545 int ret = 0;
1546 struct btrfs_device *device;
1547 struct backing_dev_info *bdi;
1548
1549 rcu_read_lock();
1550 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1551 if (!device->bdev)
1552 continue;
1553 bdi = blk_get_backing_dev_info(device->bdev);
1554 if (bdi && bdi_congested(bdi, bdi_bits)) {
1555 ret = 1;
1556 break;
1557 }
1558 }
1559 rcu_read_unlock();
1560 return ret;
1561}
1562
1563/*
1564 * If this fails, caller must call bdi_destroy() to get rid of the
1565 * bdi again.
1566 */
1567static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1568{
1569 int err;
1570
1571 bdi->capabilities = BDI_CAP_MAP_COPY;
1572 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1573 if (err)
1574 return err;
1575
1576 bdi->ra_pages = default_backing_dev_info.ra_pages;
1577 bdi->congested_fn = btrfs_congested_fn;
1578 bdi->congested_data = info;
1579 return 0;
1580}
1581
1582/*
1583 * called by the kthread helper functions to finally call the bio end_io
1584 * functions. This is where read checksum verification actually happens
1585 */
1586static void end_workqueue_fn(struct btrfs_work *work)
1587{
1588 struct bio *bio;
1589 struct end_io_wq *end_io_wq;
1590 struct btrfs_fs_info *fs_info;
1591 int error;
1592
1593 end_io_wq = container_of(work, struct end_io_wq, work);
1594 bio = end_io_wq->bio;
1595 fs_info = end_io_wq->info;
1596
1597 error = end_io_wq->error;
1598 bio->bi_private = end_io_wq->private;
1599 bio->bi_end_io = end_io_wq->end_io;
1600 kfree(end_io_wq);
1601 bio_endio(bio, error);
1602}
1603
1604static int cleaner_kthread(void *arg)
1605{
1606 struct btrfs_root *root = arg;
1607
1608 do {
1609 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1610 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1611 btrfs_run_delayed_iputs(root);
1612 btrfs_clean_old_snapshots(root);
1613 mutex_unlock(&root->fs_info->cleaner_mutex);
1614 btrfs_run_defrag_inodes(root->fs_info);
1615 }
1616
1617 if (!try_to_freeze()) {
1618 set_current_state(TASK_INTERRUPTIBLE);
1619 if (!kthread_should_stop())
1620 schedule();
1621 __set_current_state(TASK_RUNNING);
1622 }
1623 } while (!kthread_should_stop());
1624 return 0;
1625}
1626
1627static int transaction_kthread(void *arg)
1628{
1629 struct btrfs_root *root = arg;
1630 struct btrfs_trans_handle *trans;
1631 struct btrfs_transaction *cur;
1632 u64 transid;
1633 unsigned long now;
1634 unsigned long delay;
1635 bool cannot_commit;
1636
1637 do {
1638 cannot_commit = false;
1639 delay = HZ * 30;
1640 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1641
1642 spin_lock(&root->fs_info->trans_lock);
1643 cur = root->fs_info->running_transaction;
1644 if (!cur) {
1645 spin_unlock(&root->fs_info->trans_lock);
1646 goto sleep;
1647 }
1648
1649 now = get_seconds();
1650 if (!cur->blocked &&
1651 (now < cur->start_time || now - cur->start_time < 30)) {
1652 spin_unlock(&root->fs_info->trans_lock);
1653 delay = HZ * 5;
1654 goto sleep;
1655 }
1656 transid = cur->transid;
1657 spin_unlock(&root->fs_info->trans_lock);
1658
1659 /* If the file system is aborted, this will always fail. */
1660 trans = btrfs_attach_transaction(root);
1661 if (IS_ERR(trans)) {
1662 if (PTR_ERR(trans) != -ENOENT)
1663 cannot_commit = true;
1664 goto sleep;
1665 }
1666 if (transid == trans->transid) {
1667 btrfs_commit_transaction(trans, root);
1668 } else {
1669 btrfs_end_transaction(trans, root);
1670 }
1671sleep:
1672 wake_up_process(root->fs_info->cleaner_kthread);
1673 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1674
1675 if (!try_to_freeze()) {
1676 set_current_state(TASK_INTERRUPTIBLE);
1677 if (!kthread_should_stop() &&
1678 (!btrfs_transaction_blocked(root->fs_info) ||
1679 cannot_commit))
1680 schedule_timeout(delay);
1681 __set_current_state(TASK_RUNNING);
1682 }
1683 } while (!kthread_should_stop());
1684 return 0;
1685}
1686
1687/*
1688 * this will find the highest generation in the array of
1689 * root backups. The index of the highest array is returned,
1690 * or -1 if we can't find anything.
1691 *
1692 * We check to make sure the array is valid by comparing the
1693 * generation of the latest root in the array with the generation
1694 * in the super block. If they don't match we pitch it.
1695 */
1696static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1697{
1698 u64 cur;
1699 int newest_index = -1;
1700 struct btrfs_root_backup *root_backup;
1701 int i;
1702
1703 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1704 root_backup = info->super_copy->super_roots + i;
1705 cur = btrfs_backup_tree_root_gen(root_backup);
1706 if (cur == newest_gen)
1707 newest_index = i;
1708 }
1709
1710 /* check to see if we actually wrapped around */
1711 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1712 root_backup = info->super_copy->super_roots;
1713 cur = btrfs_backup_tree_root_gen(root_backup);
1714 if (cur == newest_gen)
1715 newest_index = 0;
1716 }
1717 return newest_index;
1718}
1719
1720
1721/*
1722 * find the oldest backup so we know where to store new entries
1723 * in the backup array. This will set the backup_root_index
1724 * field in the fs_info struct
1725 */
1726static void find_oldest_super_backup(struct btrfs_fs_info *info,
1727 u64 newest_gen)
1728{
1729 int newest_index = -1;
1730
1731 newest_index = find_newest_super_backup(info, newest_gen);
1732 /* if there was garbage in there, just move along */
1733 if (newest_index == -1) {
1734 info->backup_root_index = 0;
1735 } else {
1736 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1737 }
1738}
1739
1740/*
1741 * copy all the root pointers into the super backup array.
1742 * this will bump the backup pointer by one when it is
1743 * done
1744 */
1745static void backup_super_roots(struct btrfs_fs_info *info)
1746{
1747 int next_backup;
1748 struct btrfs_root_backup *root_backup;
1749 int last_backup;
1750
1751 next_backup = info->backup_root_index;
1752 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1753 BTRFS_NUM_BACKUP_ROOTS;
1754
1755 /*
1756 * just overwrite the last backup if we're at the same generation
1757 * this happens only at umount
1758 */
1759 root_backup = info->super_for_commit->super_roots + last_backup;
1760 if (btrfs_backup_tree_root_gen(root_backup) ==
1761 btrfs_header_generation(info->tree_root->node))
1762 next_backup = last_backup;
1763
1764 root_backup = info->super_for_commit->super_roots + next_backup;
1765
1766 /*
1767 * make sure all of our padding and empty slots get zero filled
1768 * regardless of which ones we use today
1769 */
1770 memset(root_backup, 0, sizeof(*root_backup));
1771
1772 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1773
1774 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1775 btrfs_set_backup_tree_root_gen(root_backup,
1776 btrfs_header_generation(info->tree_root->node));
1777
1778 btrfs_set_backup_tree_root_level(root_backup,
1779 btrfs_header_level(info->tree_root->node));
1780
1781 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1782 btrfs_set_backup_chunk_root_gen(root_backup,
1783 btrfs_header_generation(info->chunk_root->node));
1784 btrfs_set_backup_chunk_root_level(root_backup,
1785 btrfs_header_level(info->chunk_root->node));
1786
1787 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1788 btrfs_set_backup_extent_root_gen(root_backup,
1789 btrfs_header_generation(info->extent_root->node));
1790 btrfs_set_backup_extent_root_level(root_backup,
1791 btrfs_header_level(info->extent_root->node));
1792
1793 /*
1794 * we might commit during log recovery, which happens before we set
1795 * the fs_root. Make sure it is valid before we fill it in.
1796 */
1797 if (info->fs_root && info->fs_root->node) {
1798 btrfs_set_backup_fs_root(root_backup,
1799 info->fs_root->node->start);
1800 btrfs_set_backup_fs_root_gen(root_backup,
1801 btrfs_header_generation(info->fs_root->node));
1802 btrfs_set_backup_fs_root_level(root_backup,
1803 btrfs_header_level(info->fs_root->node));
1804 }
1805
1806 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1807 btrfs_set_backup_dev_root_gen(root_backup,
1808 btrfs_header_generation(info->dev_root->node));
1809 btrfs_set_backup_dev_root_level(root_backup,
1810 btrfs_header_level(info->dev_root->node));
1811
1812 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1813 btrfs_set_backup_csum_root_gen(root_backup,
1814 btrfs_header_generation(info->csum_root->node));
1815 btrfs_set_backup_csum_root_level(root_backup,
1816 btrfs_header_level(info->csum_root->node));
1817
1818 btrfs_set_backup_total_bytes(root_backup,
1819 btrfs_super_total_bytes(info->super_copy));
1820 btrfs_set_backup_bytes_used(root_backup,
1821 btrfs_super_bytes_used(info->super_copy));
1822 btrfs_set_backup_num_devices(root_backup,
1823 btrfs_super_num_devices(info->super_copy));
1824
1825 /*
1826 * if we don't copy this out to the super_copy, it won't get remembered
1827 * for the next commit
1828 */
1829 memcpy(&info->super_copy->super_roots,
1830 &info->super_for_commit->super_roots,
1831 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1832}
1833
1834/*
1835 * this copies info out of the root backup array and back into
1836 * the in-memory super block. It is meant to help iterate through
1837 * the array, so you send it the number of backups you've already
1838 * tried and the last backup index you used.
1839 *
1840 * this returns -1 when it has tried all the backups
1841 */
1842static noinline int next_root_backup(struct btrfs_fs_info *info,
1843 struct btrfs_super_block *super,
1844 int *num_backups_tried, int *backup_index)
1845{
1846 struct btrfs_root_backup *root_backup;
1847 int newest = *backup_index;
1848
1849 if (*num_backups_tried == 0) {
1850 u64 gen = btrfs_super_generation(super);
1851
1852 newest = find_newest_super_backup(info, gen);
1853 if (newest == -1)
1854 return -1;
1855
1856 *backup_index = newest;
1857 *num_backups_tried = 1;
1858 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1859 /* we've tried all the backups, all done */
1860 return -1;
1861 } else {
1862 /* jump to the next oldest backup */
1863 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1864 BTRFS_NUM_BACKUP_ROOTS;
1865 *backup_index = newest;
1866 *num_backups_tried += 1;
1867 }
1868 root_backup = super->super_roots + newest;
1869
1870 btrfs_set_super_generation(super,
1871 btrfs_backup_tree_root_gen(root_backup));
1872 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1873 btrfs_set_super_root_level(super,
1874 btrfs_backup_tree_root_level(root_backup));
1875 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1876
1877 /*
1878 * fixme: the total bytes and num_devices need to match or we should
1879 * need a fsck
1880 */
1881 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1882 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1883 return 0;
1884}
1885
1886/* helper to cleanup tree roots */
1887static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1888{
1889 free_extent_buffer(info->tree_root->node);
1890 free_extent_buffer(info->tree_root->commit_root);
1891 free_extent_buffer(info->dev_root->node);
1892 free_extent_buffer(info->dev_root->commit_root);
1893 free_extent_buffer(info->extent_root->node);
1894 free_extent_buffer(info->extent_root->commit_root);
1895 free_extent_buffer(info->csum_root->node);
1896 free_extent_buffer(info->csum_root->commit_root);
1897 if (info->quota_root) {
1898 free_extent_buffer(info->quota_root->node);
1899 free_extent_buffer(info->quota_root->commit_root);
1900 }
1901
1902 info->tree_root->node = NULL;
1903 info->tree_root->commit_root = NULL;
1904 info->dev_root->node = NULL;
1905 info->dev_root->commit_root = NULL;
1906 info->extent_root->node = NULL;
1907 info->extent_root->commit_root = NULL;
1908 info->csum_root->node = NULL;
1909 info->csum_root->commit_root = NULL;
1910 if (info->quota_root) {
1911 info->quota_root->node = NULL;
1912 info->quota_root->commit_root = NULL;
1913 }
1914
1915 if (chunk_root) {
1916 free_extent_buffer(info->chunk_root->node);
1917 free_extent_buffer(info->chunk_root->commit_root);
1918 info->chunk_root->node = NULL;
1919 info->chunk_root->commit_root = NULL;
1920 }
1921}
1922
1923
1924int open_ctree(struct super_block *sb,
1925 struct btrfs_fs_devices *fs_devices,
1926 char *options)
1927{
1928 u32 sectorsize;
1929 u32 nodesize;
1930 u32 leafsize;
1931 u32 blocksize;
1932 u32 stripesize;
1933 u64 generation;
1934 u64 features;
1935 struct btrfs_key location;
1936 struct buffer_head *bh;
1937 struct btrfs_super_block *disk_super;
1938 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
1939 struct btrfs_root *tree_root;
1940 struct btrfs_root *extent_root;
1941 struct btrfs_root *csum_root;
1942 struct btrfs_root *chunk_root;
1943 struct btrfs_root *dev_root;
1944 struct btrfs_root *quota_root;
1945 struct btrfs_root *log_tree_root;
1946 int ret;
1947 int err = -EINVAL;
1948 int num_backups_tried = 0;
1949 int backup_index = 0;
1950
1951 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
1952 extent_root = fs_info->extent_root = btrfs_alloc_root(fs_info);
1953 csum_root = fs_info->csum_root = btrfs_alloc_root(fs_info);
1954 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
1955 dev_root = fs_info->dev_root = btrfs_alloc_root(fs_info);
1956 quota_root = fs_info->quota_root = btrfs_alloc_root(fs_info);
1957
1958 if (!tree_root || !extent_root || !csum_root ||
1959 !chunk_root || !dev_root || !quota_root) {
1960 err = -ENOMEM;
1961 goto fail;
1962 }
1963
1964 ret = init_srcu_struct(&fs_info->subvol_srcu);
1965 if (ret) {
1966 err = ret;
1967 goto fail;
1968 }
1969
1970 ret = setup_bdi(fs_info, &fs_info->bdi);
1971 if (ret) {
1972 err = ret;
1973 goto fail_srcu;
1974 }
1975
1976 fs_info->btree_inode = new_inode(sb);
1977 if (!fs_info->btree_inode) {
1978 err = -ENOMEM;
1979 goto fail_bdi;
1980 }
1981
1982 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1983
1984 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1985 INIT_LIST_HEAD(&fs_info->trans_list);
1986 INIT_LIST_HEAD(&fs_info->dead_roots);
1987 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1988 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1989 INIT_LIST_HEAD(&fs_info->ordered_operations);
1990 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1991 spin_lock_init(&fs_info->delalloc_lock);
1992 spin_lock_init(&fs_info->trans_lock);
1993 spin_lock_init(&fs_info->fs_roots_radix_lock);
1994 spin_lock_init(&fs_info->delayed_iput_lock);
1995 spin_lock_init(&fs_info->defrag_inodes_lock);
1996 spin_lock_init(&fs_info->free_chunk_lock);
1997 spin_lock_init(&fs_info->tree_mod_seq_lock);
1998 rwlock_init(&fs_info->tree_mod_log_lock);
1999 mutex_init(&fs_info->reloc_mutex);
2000
2001 init_completion(&fs_info->kobj_unregister);
2002 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2003 INIT_LIST_HEAD(&fs_info->space_info);
2004 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2005 btrfs_mapping_init(&fs_info->mapping_tree);
2006 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2007 BTRFS_BLOCK_RSV_GLOBAL);
2008 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2009 BTRFS_BLOCK_RSV_DELALLOC);
2010 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2011 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2012 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2013 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2014 BTRFS_BLOCK_RSV_DELOPS);
2015 atomic_set(&fs_info->nr_async_submits, 0);
2016 atomic_set(&fs_info->async_delalloc_pages, 0);
2017 atomic_set(&fs_info->async_submit_draining, 0);
2018 atomic_set(&fs_info->nr_async_bios, 0);
2019 atomic_set(&fs_info->defrag_running, 0);
2020 atomic_set(&fs_info->tree_mod_seq, 0);
2021 fs_info->sb = sb;
2022 fs_info->max_inline = 8192 * 1024;
2023 fs_info->metadata_ratio = 0;
2024 fs_info->defrag_inodes = RB_ROOT;
2025 fs_info->trans_no_join = 0;
2026 fs_info->free_chunk_space = 0;
2027 fs_info->tree_mod_log = RB_ROOT;
2028
2029 /* readahead state */
2030 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2031 spin_lock_init(&fs_info->reada_lock);
2032
2033 fs_info->thread_pool_size = min_t(unsigned long,
2034 num_online_cpus() + 2, 8);
2035
2036 INIT_LIST_HEAD(&fs_info->ordered_extents);
2037 spin_lock_init(&fs_info->ordered_extent_lock);
2038 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2039 GFP_NOFS);
2040 if (!fs_info->delayed_root) {
2041 err = -ENOMEM;
2042 goto fail_iput;
2043 }
2044 btrfs_init_delayed_root(fs_info->delayed_root);
2045
2046 mutex_init(&fs_info->scrub_lock);
2047 atomic_set(&fs_info->scrubs_running, 0);
2048 atomic_set(&fs_info->scrub_pause_req, 0);
2049 atomic_set(&fs_info->scrubs_paused, 0);
2050 atomic_set(&fs_info->scrub_cancel_req, 0);
2051 init_waitqueue_head(&fs_info->scrub_pause_wait);
2052 init_rwsem(&fs_info->scrub_super_lock);
2053 fs_info->scrub_workers_refcnt = 0;
2054#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2055 fs_info->check_integrity_print_mask = 0;
2056#endif
2057
2058 spin_lock_init(&fs_info->balance_lock);
2059 mutex_init(&fs_info->balance_mutex);
2060 atomic_set(&fs_info->balance_running, 0);
2061 atomic_set(&fs_info->balance_pause_req, 0);
2062 atomic_set(&fs_info->balance_cancel_req, 0);
2063 fs_info->balance_ctl = NULL;
2064 init_waitqueue_head(&fs_info->balance_wait_q);
2065
2066 sb->s_blocksize = 4096;
2067 sb->s_blocksize_bits = blksize_bits(4096);
2068 sb->s_bdi = &fs_info->bdi;
2069
2070 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2071 set_nlink(fs_info->btree_inode, 1);
2072 /*
2073 * we set the i_size on the btree inode to the max possible int.
2074 * the real end of the address space is determined by all of
2075 * the devices in the system
2076 */
2077 fs_info->btree_inode->i_size = OFFSET_MAX;
2078 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2079 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2080
2081 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2082 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2083 fs_info->btree_inode->i_mapping);
2084 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2085 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2086
2087 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2088
2089 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2090 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2091 sizeof(struct btrfs_key));
2092 set_bit(BTRFS_INODE_DUMMY,
2093 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2094 insert_inode_hash(fs_info->btree_inode);
2095
2096 spin_lock_init(&fs_info->block_group_cache_lock);
2097 fs_info->block_group_cache_tree = RB_ROOT;
2098
2099 extent_io_tree_init(&fs_info->freed_extents[0],
2100 fs_info->btree_inode->i_mapping);
2101 extent_io_tree_init(&fs_info->freed_extents[1],
2102 fs_info->btree_inode->i_mapping);
2103 fs_info->pinned_extents = &fs_info->freed_extents[0];
2104 fs_info->do_barriers = 1;
2105
2106
2107 mutex_init(&fs_info->ordered_operations_mutex);
2108 mutex_init(&fs_info->tree_log_mutex);
2109 mutex_init(&fs_info->chunk_mutex);
2110 mutex_init(&fs_info->transaction_kthread_mutex);
2111 mutex_init(&fs_info->cleaner_mutex);
2112 mutex_init(&fs_info->volume_mutex);
2113 init_rwsem(&fs_info->extent_commit_sem);
2114 init_rwsem(&fs_info->cleanup_work_sem);
2115 init_rwsem(&fs_info->subvol_sem);
2116
2117 spin_lock_init(&fs_info->qgroup_lock);
2118 fs_info->qgroup_tree = RB_ROOT;
2119 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2120 fs_info->qgroup_seq = 1;
2121 fs_info->quota_enabled = 0;
2122 fs_info->pending_quota_state = 0;
2123
2124 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2125 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2126
2127 init_waitqueue_head(&fs_info->transaction_throttle);
2128 init_waitqueue_head(&fs_info->transaction_wait);
2129 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2130 init_waitqueue_head(&fs_info->async_submit_wait);
2131
2132 __setup_root(4096, 4096, 4096, 4096, tree_root,
2133 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2134
2135 invalidate_bdev(fs_devices->latest_bdev);
2136 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2137 if (!bh) {
2138 err = -EINVAL;
2139 goto fail_alloc;
2140 }
2141
2142 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2143 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2144 sizeof(*fs_info->super_for_commit));
2145 brelse(bh);
2146
2147 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2148
2149 disk_super = fs_info->super_copy;
2150 if (!btrfs_super_root(disk_super))
2151 goto fail_alloc;
2152
2153 /* check FS state, whether FS is broken. */
2154 fs_info->fs_state |= btrfs_super_flags(disk_super);
2155
2156 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2157 if (ret) {
2158 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2159 err = ret;
2160 goto fail_alloc;
2161 }
2162
2163 /*
2164 * run through our array of backup supers and setup
2165 * our ring pointer to the oldest one
2166 */
2167 generation = btrfs_super_generation(disk_super);
2168 find_oldest_super_backup(fs_info, generation);
2169
2170 /*
2171 * In the long term, we'll store the compression type in the super
2172 * block, and it'll be used for per file compression control.
2173 */
2174 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2175
2176 ret = btrfs_parse_options(tree_root, options);
2177 if (ret) {
2178 err = ret;
2179 goto fail_alloc;
2180 }
2181
2182 features = btrfs_super_incompat_flags(disk_super) &
2183 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2184 if (features) {
2185 printk(KERN_ERR "BTRFS: couldn't mount because of "
2186 "unsupported optional features (%Lx).\n",
2187 (unsigned long long)features);
2188 err = -EINVAL;
2189 goto fail_alloc;
2190 }
2191
2192 if (btrfs_super_leafsize(disk_super) !=
2193 btrfs_super_nodesize(disk_super)) {
2194 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2195 "blocksizes don't match. node %d leaf %d\n",
2196 btrfs_super_nodesize(disk_super),
2197 btrfs_super_leafsize(disk_super));
2198 err = -EINVAL;
2199 goto fail_alloc;
2200 }
2201 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2202 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2203 "blocksize (%d) was too large\n",
2204 btrfs_super_leafsize(disk_super));
2205 err = -EINVAL;
2206 goto fail_alloc;
2207 }
2208
2209 features = btrfs_super_incompat_flags(disk_super);
2210 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2211 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2212 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2213
2214 /*
2215 * flag our filesystem as having big metadata blocks if
2216 * they are bigger than the page size
2217 */
2218 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2219 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2220 printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2221 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2222 }
2223
2224 nodesize = btrfs_super_nodesize(disk_super);
2225 leafsize = btrfs_super_leafsize(disk_super);
2226 sectorsize = btrfs_super_sectorsize(disk_super);
2227 stripesize = btrfs_super_stripesize(disk_super);
2228
2229 /*
2230 * mixed block groups end up with duplicate but slightly offset
2231 * extent buffers for the same range. It leads to corruptions
2232 */
2233 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2234 (sectorsize != leafsize)) {
2235 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2236 "are not allowed for mixed block groups on %s\n",
2237 sb->s_id);
2238 goto fail_alloc;
2239 }
2240
2241 btrfs_set_super_incompat_flags(disk_super, features);
2242
2243 features = btrfs_super_compat_ro_flags(disk_super) &
2244 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2245 if (!(sb->s_flags & MS_RDONLY) && features) {
2246 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2247 "unsupported option features (%Lx).\n",
2248 (unsigned long long)features);
2249 err = -EINVAL;
2250 goto fail_alloc;
2251 }
2252
2253 btrfs_init_workers(&fs_info->generic_worker,
2254 "genwork", 1, NULL);
2255
2256 btrfs_init_workers(&fs_info->workers, "worker",
2257 fs_info->thread_pool_size,
2258 &fs_info->generic_worker);
2259
2260 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2261 fs_info->thread_pool_size,
2262 &fs_info->generic_worker);
2263
2264 btrfs_init_workers(&fs_info->submit_workers, "submit",
2265 min_t(u64, fs_devices->num_devices,
2266 fs_info->thread_pool_size),
2267 &fs_info->generic_worker);
2268
2269 btrfs_init_workers(&fs_info->caching_workers, "cache",
2270 2, &fs_info->generic_worker);
2271
2272 /* a higher idle thresh on the submit workers makes it much more
2273 * likely that bios will be send down in a sane order to the
2274 * devices
2275 */
2276 fs_info->submit_workers.idle_thresh = 64;
2277
2278 fs_info->workers.idle_thresh = 16;
2279 fs_info->workers.ordered = 1;
2280
2281 fs_info->delalloc_workers.idle_thresh = 2;
2282 fs_info->delalloc_workers.ordered = 1;
2283
2284 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2285 &fs_info->generic_worker);
2286 btrfs_init_workers(&fs_info->endio_workers, "endio",
2287 fs_info->thread_pool_size,
2288 &fs_info->generic_worker);
2289 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2290 fs_info->thread_pool_size,
2291 &fs_info->generic_worker);
2292 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2293 "endio-meta-write", fs_info->thread_pool_size,
2294 &fs_info->generic_worker);
2295 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2296 fs_info->thread_pool_size,
2297 &fs_info->generic_worker);
2298 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2299 1, &fs_info->generic_worker);
2300 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2301 fs_info->thread_pool_size,
2302 &fs_info->generic_worker);
2303 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2304 fs_info->thread_pool_size,
2305 &fs_info->generic_worker);
2306
2307 /*
2308 * endios are largely parallel and should have a very
2309 * low idle thresh
2310 */
2311 fs_info->endio_workers.idle_thresh = 4;
2312 fs_info->endio_meta_workers.idle_thresh = 4;
2313
2314 fs_info->endio_write_workers.idle_thresh = 2;
2315 fs_info->endio_meta_write_workers.idle_thresh = 2;
2316 fs_info->readahead_workers.idle_thresh = 2;
2317
2318 /*
2319 * btrfs_start_workers can really only fail because of ENOMEM so just
2320 * return -ENOMEM if any of these fail.
2321 */
2322 ret = btrfs_start_workers(&fs_info->workers);
2323 ret |= btrfs_start_workers(&fs_info->generic_worker);
2324 ret |= btrfs_start_workers(&fs_info->submit_workers);
2325 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2326 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2327 ret |= btrfs_start_workers(&fs_info->endio_workers);
2328 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2329 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2330 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2331 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2332 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2333 ret |= btrfs_start_workers(&fs_info->caching_workers);
2334 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2335 if (ret) {
2336 err = -ENOMEM;
2337 goto fail_sb_buffer;
2338 }
2339
2340 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2341 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2342 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2343
2344 tree_root->nodesize = nodesize;
2345 tree_root->leafsize = leafsize;
2346 tree_root->sectorsize = sectorsize;
2347 tree_root->stripesize = stripesize;
2348
2349 sb->s_blocksize = sectorsize;
2350 sb->s_blocksize_bits = blksize_bits(sectorsize);
2351
2352 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2353 sizeof(disk_super->magic))) {
2354 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2355 goto fail_sb_buffer;
2356 }
2357
2358 if (sectorsize != PAGE_SIZE) {
2359 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2360 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2361 goto fail_sb_buffer;
2362 }
2363
2364 mutex_lock(&fs_info->chunk_mutex);
2365 ret = btrfs_read_sys_array(tree_root);
2366 mutex_unlock(&fs_info->chunk_mutex);
2367 if (ret) {
2368 printk(KERN_WARNING "btrfs: failed to read the system "
2369 "array on %s\n", sb->s_id);
2370 goto fail_sb_buffer;
2371 }
2372
2373 blocksize = btrfs_level_size(tree_root,
2374 btrfs_super_chunk_root_level(disk_super));
2375 generation = btrfs_super_chunk_root_generation(disk_super);
2376
2377 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2378 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2379
2380 chunk_root->node = read_tree_block(chunk_root,
2381 btrfs_super_chunk_root(disk_super),
2382 blocksize, generation);
2383 BUG_ON(!chunk_root->node); /* -ENOMEM */
2384 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2385 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2386 sb->s_id);
2387 goto fail_tree_roots;
2388 }
2389 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2390 chunk_root->commit_root = btrfs_root_node(chunk_root);
2391
2392 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2393 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2394 BTRFS_UUID_SIZE);
2395
2396 ret = btrfs_read_chunk_tree(chunk_root);
2397 if (ret) {
2398 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2399 sb->s_id);
2400 goto fail_tree_roots;
2401 }
2402
2403 btrfs_close_extra_devices(fs_devices);
2404
2405 if (!fs_devices->latest_bdev) {
2406 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2407 sb->s_id);
2408 goto fail_tree_roots;
2409 }
2410
2411retry_root_backup:
2412 blocksize = btrfs_level_size(tree_root,
2413 btrfs_super_root_level(disk_super));
2414 generation = btrfs_super_generation(disk_super);
2415
2416 tree_root->node = read_tree_block(tree_root,
2417 btrfs_super_root(disk_super),
2418 blocksize, generation);
2419 if (!tree_root->node ||
2420 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2421 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2422 sb->s_id);
2423
2424 goto recovery_tree_root;
2425 }
2426
2427 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2428 tree_root->commit_root = btrfs_root_node(tree_root);
2429
2430 ret = find_and_setup_root(tree_root, fs_info,
2431 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2432 if (ret)
2433 goto recovery_tree_root;
2434 extent_root->track_dirty = 1;
2435
2436 ret = find_and_setup_root(tree_root, fs_info,
2437 BTRFS_DEV_TREE_OBJECTID, dev_root);
2438 if (ret)
2439 goto recovery_tree_root;
2440 dev_root->track_dirty = 1;
2441
2442 ret = find_and_setup_root(tree_root, fs_info,
2443 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2444 if (ret)
2445 goto recovery_tree_root;
2446 csum_root->track_dirty = 1;
2447
2448 ret = find_and_setup_root(tree_root, fs_info,
2449 BTRFS_QUOTA_TREE_OBJECTID, quota_root);
2450 if (ret) {
2451 kfree(quota_root);
2452 quota_root = fs_info->quota_root = NULL;
2453 } else {
2454 quota_root->track_dirty = 1;
2455 fs_info->quota_enabled = 1;
2456 fs_info->pending_quota_state = 1;
2457 }
2458
2459 fs_info->generation = generation;
2460 fs_info->last_trans_committed = generation;
2461
2462 ret = btrfs_recover_balance(fs_info);
2463 if (ret) {
2464 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2465 goto fail_block_groups;
2466 }
2467
2468 ret = btrfs_init_dev_stats(fs_info);
2469 if (ret) {
2470 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2471 ret);
2472 goto fail_block_groups;
2473 }
2474
2475 ret = btrfs_init_space_info(fs_info);
2476 if (ret) {
2477 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2478 goto fail_block_groups;
2479 }
2480
2481 ret = btrfs_read_block_groups(extent_root);
2482 if (ret) {
2483 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2484 goto fail_block_groups;
2485 }
2486
2487 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2488 "btrfs-cleaner");
2489 if (IS_ERR(fs_info->cleaner_kthread))
2490 goto fail_block_groups;
2491
2492 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2493 tree_root,
2494 "btrfs-transaction");
2495 if (IS_ERR(fs_info->transaction_kthread))
2496 goto fail_cleaner;
2497
2498 if (!btrfs_test_opt(tree_root, SSD) &&
2499 !btrfs_test_opt(tree_root, NOSSD) &&
2500 !fs_info->fs_devices->rotating) {
2501 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2502 "mode\n");
2503 btrfs_set_opt(fs_info->mount_opt, SSD);
2504 }
2505
2506#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2507 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2508 ret = btrfsic_mount(tree_root, fs_devices,
2509 btrfs_test_opt(tree_root,
2510 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2511 1 : 0,
2512 fs_info->check_integrity_print_mask);
2513 if (ret)
2514 printk(KERN_WARNING "btrfs: failed to initialize"
2515 " integrity check module %s\n", sb->s_id);
2516 }
2517#endif
2518 ret = btrfs_read_qgroup_config(fs_info);
2519 if (ret)
2520 goto fail_trans_kthread;
2521
2522 /* do not make disk changes in broken FS */
2523 if (btrfs_super_log_root(disk_super) != 0) {
2524 u64 bytenr = btrfs_super_log_root(disk_super);
2525
2526 if (fs_devices->rw_devices == 0) {
2527 printk(KERN_WARNING "Btrfs log replay required "
2528 "on RO media\n");
2529 err = -EIO;
2530 goto fail_qgroup;
2531 }
2532 blocksize =
2533 btrfs_level_size(tree_root,
2534 btrfs_super_log_root_level(disk_super));
2535
2536 log_tree_root = btrfs_alloc_root(fs_info);
2537 if (!log_tree_root) {
2538 err = -ENOMEM;
2539 goto fail_qgroup;
2540 }
2541
2542 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2543 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2544
2545 log_tree_root->node = read_tree_block(tree_root, bytenr,
2546 blocksize,
2547 generation + 1);
2548 /* returns with log_tree_root freed on success */
2549 ret = btrfs_recover_log_trees(log_tree_root);
2550 if (ret) {
2551 btrfs_error(tree_root->fs_info, ret,
2552 "Failed to recover log tree");
2553 free_extent_buffer(log_tree_root->node);
2554 kfree(log_tree_root);
2555 goto fail_trans_kthread;
2556 }
2557
2558 if (sb->s_flags & MS_RDONLY) {
2559 ret = btrfs_commit_super(tree_root);
2560 if (ret)
2561 goto fail_trans_kthread;
2562 }
2563 }
2564
2565 ret = btrfs_find_orphan_roots(tree_root);
2566 if (ret)
2567 goto fail_trans_kthread;
2568
2569 if (!(sb->s_flags & MS_RDONLY)) {
2570 ret = btrfs_cleanup_fs_roots(fs_info);
2571 if (ret)
2572 goto fail_trans_kthread;
2573
2574 ret = btrfs_recover_relocation(tree_root);
2575 if (ret < 0) {
2576 printk(KERN_WARNING
2577 "btrfs: failed to recover relocation\n");
2578 err = -EINVAL;
2579 goto fail_qgroup;
2580 }
2581 }
2582
2583 location.objectid = BTRFS_FS_TREE_OBJECTID;
2584 location.type = BTRFS_ROOT_ITEM_KEY;
2585 location.offset = (u64)-1;
2586
2587 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2588 if (!fs_info->fs_root)
2589 goto fail_qgroup;
2590 if (IS_ERR(fs_info->fs_root)) {
2591 err = PTR_ERR(fs_info->fs_root);
2592 goto fail_qgroup;
2593 }
2594
2595 if (sb->s_flags & MS_RDONLY)
2596 return 0;
2597
2598 down_read(&fs_info->cleanup_work_sem);
2599 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2600 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2601 up_read(&fs_info->cleanup_work_sem);
2602 close_ctree(tree_root);
2603 return ret;
2604 }
2605 up_read(&fs_info->cleanup_work_sem);
2606
2607 ret = btrfs_resume_balance_async(fs_info);
2608 if (ret) {
2609 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2610 close_ctree(tree_root);
2611 return ret;
2612 }
2613
2614 return 0;
2615
2616fail_qgroup:
2617 btrfs_free_qgroup_config(fs_info);
2618fail_trans_kthread:
2619 kthread_stop(fs_info->transaction_kthread);
2620fail_cleaner:
2621 kthread_stop(fs_info->cleaner_kthread);
2622
2623 /*
2624 * make sure we're done with the btree inode before we stop our
2625 * kthreads
2626 */
2627 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2628 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2629
2630fail_block_groups:
2631 btrfs_free_block_groups(fs_info);
2632
2633fail_tree_roots:
2634 free_root_pointers(fs_info, 1);
2635
2636fail_sb_buffer:
2637 btrfs_stop_workers(&fs_info->generic_worker);
2638 btrfs_stop_workers(&fs_info->readahead_workers);
2639 btrfs_stop_workers(&fs_info->fixup_workers);
2640 btrfs_stop_workers(&fs_info->delalloc_workers);
2641 btrfs_stop_workers(&fs_info->workers);
2642 btrfs_stop_workers(&fs_info->endio_workers);
2643 btrfs_stop_workers(&fs_info->endio_meta_workers);
2644 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2645 btrfs_stop_workers(&fs_info->endio_write_workers);
2646 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2647 btrfs_stop_workers(&fs_info->submit_workers);
2648 btrfs_stop_workers(&fs_info->delayed_workers);
2649 btrfs_stop_workers(&fs_info->caching_workers);
2650fail_alloc:
2651fail_iput:
2652 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2653
2654 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2655 iput(fs_info->btree_inode);
2656fail_bdi:
2657 bdi_destroy(&fs_info->bdi);
2658fail_srcu:
2659 cleanup_srcu_struct(&fs_info->subvol_srcu);
2660fail:
2661 btrfs_close_devices(fs_info->fs_devices);
2662 return err;
2663
2664recovery_tree_root:
2665 if (!btrfs_test_opt(tree_root, RECOVERY))
2666 goto fail_tree_roots;
2667
2668 free_root_pointers(fs_info, 0);
2669
2670 /* don't use the log in recovery mode, it won't be valid */
2671 btrfs_set_super_log_root(disk_super, 0);
2672
2673 /* we can't trust the free space cache either */
2674 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2675
2676 ret = next_root_backup(fs_info, fs_info->super_copy,
2677 &num_backups_tried, &backup_index);
2678 if (ret == -1)
2679 goto fail_block_groups;
2680 goto retry_root_backup;
2681}
2682
2683static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2684{
2685 if (uptodate) {
2686 set_buffer_uptodate(bh);
2687 } else {
2688 struct btrfs_device *device = (struct btrfs_device *)
2689 bh->b_private;
2690
2691 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
2692 "I/O error on %s\n",
2693 rcu_str_deref(device->name));
2694 /* note, we dont' set_buffer_write_io_error because we have
2695 * our own ways of dealing with the IO errors
2696 */
2697 clear_buffer_uptodate(bh);
2698 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
2699 }
2700 unlock_buffer(bh);
2701 put_bh(bh);
2702}
2703
2704struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2705{
2706 struct buffer_head *bh;
2707 struct buffer_head *latest = NULL;
2708 struct btrfs_super_block *super;
2709 int i;
2710 u64 transid = 0;
2711 u64 bytenr;
2712
2713 /* we would like to check all the supers, but that would make
2714 * a btrfs mount succeed after a mkfs from a different FS.
2715 * So, we need to add a special mount option to scan for
2716 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2717 */
2718 for (i = 0; i < 1; i++) {
2719 bytenr = btrfs_sb_offset(i);
2720 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2721 break;
2722 bh = __bread(bdev, bytenr / 4096, 4096);
2723 if (!bh)
2724 continue;
2725
2726 super = (struct btrfs_super_block *)bh->b_data;
2727 if (btrfs_super_bytenr(super) != bytenr ||
2728 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2729 sizeof(super->magic))) {
2730 brelse(bh);
2731 continue;
2732 }
2733
2734 if (!latest || btrfs_super_generation(super) > transid) {
2735 brelse(latest);
2736 latest = bh;
2737 transid = btrfs_super_generation(super);
2738 } else {
2739 brelse(bh);
2740 }
2741 }
2742 return latest;
2743}
2744
2745/*
2746 * this should be called twice, once with wait == 0 and
2747 * once with wait == 1. When wait == 0 is done, all the buffer heads
2748 * we write are pinned.
2749 *
2750 * They are released when wait == 1 is done.
2751 * max_mirrors must be the same for both runs, and it indicates how
2752 * many supers on this one device should be written.
2753 *
2754 * max_mirrors == 0 means to write them all.
2755 */
2756static int write_dev_supers(struct btrfs_device *device,
2757 struct btrfs_super_block *sb,
2758 int do_barriers, int wait, int max_mirrors)
2759{
2760 struct buffer_head *bh;
2761 int i;
2762 int ret;
2763 int errors = 0;
2764 u32 crc;
2765 u64 bytenr;
2766
2767 if (max_mirrors == 0)
2768 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2769
2770 for (i = 0; i < max_mirrors; i++) {
2771 bytenr = btrfs_sb_offset(i);
2772 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2773 break;
2774
2775 if (wait) {
2776 bh = __find_get_block(device->bdev, bytenr / 4096,
2777 BTRFS_SUPER_INFO_SIZE);
2778 BUG_ON(!bh);
2779 wait_on_buffer(bh);
2780 if (!buffer_uptodate(bh))
2781 errors++;
2782
2783 /* drop our reference */
2784 brelse(bh);
2785
2786 /* drop the reference from the wait == 0 run */
2787 brelse(bh);
2788 continue;
2789 } else {
2790 btrfs_set_super_bytenr(sb, bytenr);
2791
2792 crc = ~(u32)0;
2793 crc = btrfs_csum_data(NULL, (char *)sb +
2794 BTRFS_CSUM_SIZE, crc,
2795 BTRFS_SUPER_INFO_SIZE -
2796 BTRFS_CSUM_SIZE);
2797 btrfs_csum_final(crc, sb->csum);
2798
2799 /*
2800 * one reference for us, and we leave it for the
2801 * caller
2802 */
2803 bh = __getblk(device->bdev, bytenr / 4096,
2804 BTRFS_SUPER_INFO_SIZE);
2805 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2806
2807 /* one reference for submit_bh */
2808 get_bh(bh);
2809
2810 set_buffer_uptodate(bh);
2811 lock_buffer(bh);
2812 bh->b_end_io = btrfs_end_buffer_write_sync;
2813 bh->b_private = device;
2814 }
2815
2816 /*
2817 * we fua the first super. The others we allow
2818 * to go down lazy.
2819 */
2820 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2821 if (ret)
2822 errors++;
2823 }
2824 return errors < i ? 0 : -1;
2825}
2826
2827/*
2828 * endio for the write_dev_flush, this will wake anyone waiting
2829 * for the barrier when it is done
2830 */
2831static void btrfs_end_empty_barrier(struct bio *bio, int err)
2832{
2833 if (err) {
2834 if (err == -EOPNOTSUPP)
2835 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2836 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2837 }
2838 if (bio->bi_private)
2839 complete(bio->bi_private);
2840 bio_put(bio);
2841}
2842
2843/*
2844 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2845 * sent down. With wait == 1, it waits for the previous flush.
2846 *
2847 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2848 * capable
2849 */
2850static int write_dev_flush(struct btrfs_device *device, int wait)
2851{
2852 struct bio *bio;
2853 int ret = 0;
2854
2855 if (device->nobarriers)
2856 return 0;
2857
2858 if (wait) {
2859 bio = device->flush_bio;
2860 if (!bio)
2861 return 0;
2862
2863 wait_for_completion(&device->flush_wait);
2864
2865 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2866 printk_in_rcu("btrfs: disabling barriers on dev %s\n",
2867 rcu_str_deref(device->name));
2868 device->nobarriers = 1;
2869 }
2870 if (!bio_flagged(bio, BIO_UPTODATE)) {
2871 ret = -EIO;
2872 if (!bio_flagged(bio, BIO_EOPNOTSUPP))
2873 btrfs_dev_stat_inc_and_print(device,
2874 BTRFS_DEV_STAT_FLUSH_ERRS);
2875 }
2876
2877 /* drop the reference from the wait == 0 run */
2878 bio_put(bio);
2879 device->flush_bio = NULL;
2880
2881 return ret;
2882 }
2883
2884 /*
2885 * one reference for us, and we leave it for the
2886 * caller
2887 */
2888 device->flush_bio = NULL;
2889 bio = bio_alloc(GFP_NOFS, 0);
2890 if (!bio)
2891 return -ENOMEM;
2892
2893 bio->bi_end_io = btrfs_end_empty_barrier;
2894 bio->bi_bdev = device->bdev;
2895 init_completion(&device->flush_wait);
2896 bio->bi_private = &device->flush_wait;
2897 device->flush_bio = bio;
2898
2899 bio_get(bio);
2900 btrfsic_submit_bio(WRITE_FLUSH, bio);
2901
2902 return 0;
2903}
2904
2905/*
2906 * send an empty flush down to each device in parallel,
2907 * then wait for them
2908 */
2909static int barrier_all_devices(struct btrfs_fs_info *info)
2910{
2911 struct list_head *head;
2912 struct btrfs_device *dev;
2913 int errors = 0;
2914 int ret;
2915
2916 /* send down all the barriers */
2917 head = &info->fs_devices->devices;
2918 list_for_each_entry_rcu(dev, head, dev_list) {
2919 if (!dev->bdev) {
2920 errors++;
2921 continue;
2922 }
2923 if (!dev->in_fs_metadata || !dev->writeable)
2924 continue;
2925
2926 ret = write_dev_flush(dev, 0);
2927 if (ret)
2928 errors++;
2929 }
2930
2931 /* wait for all the barriers */
2932 list_for_each_entry_rcu(dev, head, dev_list) {
2933 if (!dev->bdev) {
2934 errors++;
2935 continue;
2936 }
2937 if (!dev->in_fs_metadata || !dev->writeable)
2938 continue;
2939
2940 ret = write_dev_flush(dev, 1);
2941 if (ret)
2942 errors++;
2943 }
2944 if (errors)
2945 return -EIO;
2946 return 0;
2947}
2948
2949int write_all_supers(struct btrfs_root *root, int max_mirrors)
2950{
2951 struct list_head *head;
2952 struct btrfs_device *dev;
2953 struct btrfs_super_block *sb;
2954 struct btrfs_dev_item *dev_item;
2955 int ret;
2956 int do_barriers;
2957 int max_errors;
2958 int total_errors = 0;
2959 u64 flags;
2960
2961 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2962 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2963 backup_super_roots(root->fs_info);
2964
2965 sb = root->fs_info->super_for_commit;
2966 dev_item = &sb->dev_item;
2967
2968 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2969 head = &root->fs_info->fs_devices->devices;
2970
2971 if (do_barriers)
2972 barrier_all_devices(root->fs_info);
2973
2974 list_for_each_entry_rcu(dev, head, dev_list) {
2975 if (!dev->bdev) {
2976 total_errors++;
2977 continue;
2978 }
2979 if (!dev->in_fs_metadata || !dev->writeable)
2980 continue;
2981
2982 btrfs_set_stack_device_generation(dev_item, 0);
2983 btrfs_set_stack_device_type(dev_item, dev->type);
2984 btrfs_set_stack_device_id(dev_item, dev->devid);
2985 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2986 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2987 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2988 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2989 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2990 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2991 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2992
2993 flags = btrfs_super_flags(sb);
2994 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2995
2996 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2997 if (ret)
2998 total_errors++;
2999 }
3000 if (total_errors > max_errors) {
3001 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3002 total_errors);
3003
3004 /* This shouldn't happen. FUA is masked off if unsupported */
3005 BUG();
3006 }
3007
3008 total_errors = 0;
3009 list_for_each_entry_rcu(dev, head, dev_list) {
3010 if (!dev->bdev)
3011 continue;
3012 if (!dev->in_fs_metadata || !dev->writeable)
3013 continue;
3014
3015 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3016 if (ret)
3017 total_errors++;
3018 }
3019 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3020 if (total_errors > max_errors) {
3021 btrfs_error(root->fs_info, -EIO,
3022 "%d errors while writing supers", total_errors);
3023 return -EIO;
3024 }
3025 return 0;
3026}
3027
3028int write_ctree_super(struct btrfs_trans_handle *trans,
3029 struct btrfs_root *root, int max_mirrors)
3030{
3031 int ret;
3032
3033 ret = write_all_supers(root, max_mirrors);
3034 return ret;
3035}
3036
3037void btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
3038{
3039 spin_lock(&fs_info->fs_roots_radix_lock);
3040 radix_tree_delete(&fs_info->fs_roots_radix,
3041 (unsigned long)root->root_key.objectid);
3042 spin_unlock(&fs_info->fs_roots_radix_lock);
3043
3044 if (btrfs_root_refs(&root->root_item) == 0)
3045 synchronize_srcu(&fs_info->subvol_srcu);
3046
3047 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3048 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3049 free_fs_root(root);
3050}
3051
3052static void free_fs_root(struct btrfs_root *root)
3053{
3054 iput(root->cache_inode);
3055 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3056 if (root->anon_dev)
3057 free_anon_bdev(root->anon_dev);
3058 free_extent_buffer(root->node);
3059 free_extent_buffer(root->commit_root);
3060 kfree(root->free_ino_ctl);
3061 kfree(root->free_ino_pinned);
3062 kfree(root->name);
3063 kfree(root);
3064}
3065
3066static void del_fs_roots(struct btrfs_fs_info *fs_info)
3067{
3068 int ret;
3069 struct btrfs_root *gang[8];
3070 int i;
3071
3072 while (!list_empty(&fs_info->dead_roots)) {
3073 gang[0] = list_entry(fs_info->dead_roots.next,
3074 struct btrfs_root, root_list);
3075 list_del(&gang[0]->root_list);
3076
3077 if (gang[0]->in_radix) {
3078 btrfs_free_fs_root(fs_info, gang[0]);
3079 } else {
3080 free_extent_buffer(gang[0]->node);
3081 free_extent_buffer(gang[0]->commit_root);
3082 kfree(gang[0]);
3083 }
3084 }
3085
3086 while (1) {
3087 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3088 (void **)gang, 0,
3089 ARRAY_SIZE(gang));
3090 if (!ret)
3091 break;
3092 for (i = 0; i < ret; i++)
3093 btrfs_free_fs_root(fs_info, gang[i]);
3094 }
3095}
3096
3097int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3098{
3099 u64 root_objectid = 0;
3100 struct btrfs_root *gang[8];
3101 int i;
3102 int ret;
3103
3104 while (1) {
3105 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3106 (void **)gang, root_objectid,
3107 ARRAY_SIZE(gang));
3108 if (!ret)
3109 break;
3110
3111 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3112 for (i = 0; i < ret; i++) {
3113 int err;
3114
3115 root_objectid = gang[i]->root_key.objectid;
3116 err = btrfs_orphan_cleanup(gang[i]);
3117 if (err)
3118 return err;
3119 }
3120 root_objectid++;
3121 }
3122 return 0;
3123}
3124
3125int btrfs_commit_super(struct btrfs_root *root)
3126{
3127 struct btrfs_trans_handle *trans;
3128 int ret;
3129
3130 mutex_lock(&root->fs_info->cleaner_mutex);
3131 btrfs_run_delayed_iputs(root);
3132 btrfs_clean_old_snapshots(root);
3133 mutex_unlock(&root->fs_info->cleaner_mutex);
3134
3135 /* wait until ongoing cleanup work done */
3136 down_write(&root->fs_info->cleanup_work_sem);
3137 up_write(&root->fs_info->cleanup_work_sem);
3138
3139 trans = btrfs_join_transaction(root);
3140 if (IS_ERR(trans))
3141 return PTR_ERR(trans);
3142 ret = btrfs_commit_transaction(trans, root);
3143 if (ret)
3144 return ret;
3145 /* run commit again to drop the original snapshot */
3146 trans = btrfs_join_transaction(root);
3147 if (IS_ERR(trans))
3148 return PTR_ERR(trans);
3149 ret = btrfs_commit_transaction(trans, root);
3150 if (ret)
3151 return ret;
3152 ret = btrfs_write_and_wait_transaction(NULL, root);
3153 if (ret) {
3154 btrfs_error(root->fs_info, ret,
3155 "Failed to sync btree inode to disk.");
3156 return ret;
3157 }
3158
3159 ret = write_ctree_super(NULL, root, 0);
3160 return ret;
3161}
3162
3163int close_ctree(struct btrfs_root *root)
3164{
3165 struct btrfs_fs_info *fs_info = root->fs_info;
3166 int ret;
3167
3168 fs_info->closing = 1;
3169 smp_mb();
3170
3171 /* pause restriper - we want to resume on mount */
3172 btrfs_pause_balance(root->fs_info);
3173
3174 btrfs_scrub_cancel(root);
3175
3176 /* wait for any defraggers to finish */
3177 wait_event(fs_info->transaction_wait,
3178 (atomic_read(&fs_info->defrag_running) == 0));
3179
3180 /* clear out the rbtree of defraggable inodes */
3181 btrfs_run_defrag_inodes(fs_info);
3182
3183 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3184 ret = btrfs_commit_super(root);
3185 if (ret)
3186 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3187 }
3188
3189 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3190 btrfs_error_commit_super(root);
3191
3192 btrfs_put_block_group_cache(fs_info);
3193
3194 kthread_stop(fs_info->transaction_kthread);
3195 kthread_stop(fs_info->cleaner_kthread);
3196
3197 fs_info->closing = 2;
3198 smp_mb();
3199
3200 btrfs_free_qgroup_config(root->fs_info);
3201
3202 if (fs_info->delalloc_bytes) {
3203 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3204 (unsigned long long)fs_info->delalloc_bytes);
3205 }
3206
3207 free_extent_buffer(fs_info->extent_root->node);
3208 free_extent_buffer(fs_info->extent_root->commit_root);
3209 free_extent_buffer(fs_info->tree_root->node);
3210 free_extent_buffer(fs_info->tree_root->commit_root);
3211 free_extent_buffer(fs_info->chunk_root->node);
3212 free_extent_buffer(fs_info->chunk_root->commit_root);
3213 free_extent_buffer(fs_info->dev_root->node);
3214 free_extent_buffer(fs_info->dev_root->commit_root);
3215 free_extent_buffer(fs_info->csum_root->node);
3216 free_extent_buffer(fs_info->csum_root->commit_root);
3217 if (fs_info->quota_root) {
3218 free_extent_buffer(fs_info->quota_root->node);
3219 free_extent_buffer(fs_info->quota_root->commit_root);
3220 }
3221
3222 btrfs_free_block_groups(fs_info);
3223
3224 del_fs_roots(fs_info);
3225
3226 iput(fs_info->btree_inode);
3227
3228 btrfs_stop_workers(&fs_info->generic_worker);
3229 btrfs_stop_workers(&fs_info->fixup_workers);
3230 btrfs_stop_workers(&fs_info->delalloc_workers);
3231 btrfs_stop_workers(&fs_info->workers);
3232 btrfs_stop_workers(&fs_info->endio_workers);
3233 btrfs_stop_workers(&fs_info->endio_meta_workers);
3234 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3235 btrfs_stop_workers(&fs_info->endio_write_workers);
3236 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3237 btrfs_stop_workers(&fs_info->submit_workers);
3238 btrfs_stop_workers(&fs_info->delayed_workers);
3239 btrfs_stop_workers(&fs_info->caching_workers);
3240 btrfs_stop_workers(&fs_info->readahead_workers);
3241
3242#ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3243 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3244 btrfsic_unmount(root, fs_info->fs_devices);
3245#endif
3246
3247 btrfs_close_devices(fs_info->fs_devices);
3248 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3249
3250 bdi_destroy(&fs_info->bdi);
3251 cleanup_srcu_struct(&fs_info->subvol_srcu);
3252
3253 return 0;
3254}
3255
3256int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3257 int atomic)
3258{
3259 int ret;
3260 struct inode *btree_inode = buf->pages[0]->mapping->host;
3261
3262 ret = extent_buffer_uptodate(buf);
3263 if (!ret)
3264 return ret;
3265
3266 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3267 parent_transid, atomic);
3268 if (ret == -EAGAIN)
3269 return ret;
3270 return !ret;
3271}
3272
3273int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3274{
3275 return set_extent_buffer_uptodate(buf);
3276}
3277
3278void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3279{
3280 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3281 u64 transid = btrfs_header_generation(buf);
3282 int was_dirty;
3283
3284 btrfs_assert_tree_locked(buf);
3285 if (transid != root->fs_info->generation) {
3286 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3287 "found %llu running %llu\n",
3288 (unsigned long long)buf->start,
3289 (unsigned long long)transid,
3290 (unsigned long long)root->fs_info->generation);
3291 WARN_ON(1);
3292 }
3293 was_dirty = set_extent_buffer_dirty(buf);
3294 if (!was_dirty) {
3295 spin_lock(&root->fs_info->delalloc_lock);
3296 root->fs_info->dirty_metadata_bytes += buf->len;
3297 spin_unlock(&root->fs_info->delalloc_lock);
3298 }
3299}
3300
3301void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3302{
3303 /*
3304 * looks as though older kernels can get into trouble with
3305 * this code, they end up stuck in balance_dirty_pages forever
3306 */
3307 u64 num_dirty;
3308 unsigned long thresh = 32 * 1024 * 1024;
3309
3310 if (current->flags & PF_MEMALLOC)
3311 return;
3312
3313 btrfs_balance_delayed_items(root);
3314
3315 num_dirty = root->fs_info->dirty_metadata_bytes;
3316
3317 if (num_dirty > thresh) {
3318 balance_dirty_pages_ratelimited_nr(
3319 root->fs_info->btree_inode->i_mapping, 1);
3320 }
3321 return;
3322}
3323
3324void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3325{
3326 /*
3327 * looks as though older kernels can get into trouble with
3328 * this code, they end up stuck in balance_dirty_pages forever
3329 */
3330 u64 num_dirty;
3331 unsigned long thresh = 32 * 1024 * 1024;
3332
3333 if (current->flags & PF_MEMALLOC)
3334 return;
3335
3336 num_dirty = root->fs_info->dirty_metadata_bytes;
3337
3338 if (num_dirty > thresh) {
3339 balance_dirty_pages_ratelimited_nr(
3340 root->fs_info->btree_inode->i_mapping, 1);
3341 }
3342 return;
3343}
3344
3345int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3346{
3347 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3348 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3349}
3350
3351static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3352 int read_only)
3353{
3354 if (btrfs_super_csum_type(fs_info->super_copy) >= ARRAY_SIZE(btrfs_csum_sizes)) {
3355 printk(KERN_ERR "btrfs: unsupported checksum algorithm\n");
3356 return -EINVAL;
3357 }
3358
3359 if (read_only)
3360 return 0;
3361
3362 return 0;
3363}
3364
3365void btrfs_error_commit_super(struct btrfs_root *root)
3366{
3367 mutex_lock(&root->fs_info->cleaner_mutex);
3368 btrfs_run_delayed_iputs(root);
3369 mutex_unlock(&root->fs_info->cleaner_mutex);
3370
3371 down_write(&root->fs_info->cleanup_work_sem);
3372 up_write(&root->fs_info->cleanup_work_sem);
3373
3374 /* cleanup FS via transaction */
3375 btrfs_cleanup_transaction(root);
3376}
3377
3378static void btrfs_destroy_ordered_operations(struct btrfs_root *root)
3379{
3380 struct btrfs_inode *btrfs_inode;
3381 struct list_head splice;
3382
3383 INIT_LIST_HEAD(&splice);
3384
3385 mutex_lock(&root->fs_info->ordered_operations_mutex);
3386 spin_lock(&root->fs_info->ordered_extent_lock);
3387
3388 list_splice_init(&root->fs_info->ordered_operations, &splice);
3389 while (!list_empty(&splice)) {
3390 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3391 ordered_operations);
3392
3393 list_del_init(&btrfs_inode->ordered_operations);
3394
3395 btrfs_invalidate_inodes(btrfs_inode->root);
3396 }
3397
3398 spin_unlock(&root->fs_info->ordered_extent_lock);
3399 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3400}
3401
3402static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3403{
3404 struct list_head splice;
3405 struct btrfs_ordered_extent *ordered;
3406 struct inode *inode;
3407
3408 INIT_LIST_HEAD(&splice);
3409
3410 spin_lock(&root->fs_info->ordered_extent_lock);
3411
3412 list_splice_init(&root->fs_info->ordered_extents, &splice);
3413 while (!list_empty(&splice)) {
3414 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3415 root_extent_list);
3416
3417 list_del_init(&ordered->root_extent_list);
3418 atomic_inc(&ordered->refs);
3419
3420 /* the inode may be getting freed (in sys_unlink path). */
3421 inode = igrab(ordered->inode);
3422
3423 spin_unlock(&root->fs_info->ordered_extent_lock);
3424 if (inode)
3425 iput(inode);
3426
3427 atomic_set(&ordered->refs, 1);
3428 btrfs_put_ordered_extent(ordered);
3429
3430 spin_lock(&root->fs_info->ordered_extent_lock);
3431 }
3432
3433 spin_unlock(&root->fs_info->ordered_extent_lock);
3434}
3435
3436int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3437 struct btrfs_root *root)
3438{
3439 struct rb_node *node;
3440 struct btrfs_delayed_ref_root *delayed_refs;
3441 struct btrfs_delayed_ref_node *ref;
3442 int ret = 0;
3443
3444 delayed_refs = &trans->delayed_refs;
3445
3446 spin_lock(&delayed_refs->lock);
3447 if (delayed_refs->num_entries == 0) {
3448 spin_unlock(&delayed_refs->lock);
3449 printk(KERN_INFO "delayed_refs has NO entry\n");
3450 return ret;
3451 }
3452
3453 while ((node = rb_first(&delayed_refs->root)) != NULL) {
3454 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3455
3456 atomic_set(&ref->refs, 1);
3457 if (btrfs_delayed_ref_is_head(ref)) {
3458 struct btrfs_delayed_ref_head *head;
3459
3460 head = btrfs_delayed_node_to_head(ref);
3461 if (!mutex_trylock(&head->mutex)) {
3462 atomic_inc(&ref->refs);
3463 spin_unlock(&delayed_refs->lock);
3464
3465 /* Need to wait for the delayed ref to run */
3466 mutex_lock(&head->mutex);
3467 mutex_unlock(&head->mutex);
3468 btrfs_put_delayed_ref(ref);
3469
3470 spin_lock(&delayed_refs->lock);
3471 continue;
3472 }
3473
3474 kfree(head->extent_op);
3475 delayed_refs->num_heads--;
3476 if (list_empty(&head->cluster))
3477 delayed_refs->num_heads_ready--;
3478 list_del_init(&head->cluster);
3479 }
3480 ref->in_tree = 0;
3481 rb_erase(&ref->rb_node, &delayed_refs->root);
3482 delayed_refs->num_entries--;
3483
3484 spin_unlock(&delayed_refs->lock);
3485 btrfs_put_delayed_ref(ref);
3486
3487 cond_resched();
3488 spin_lock(&delayed_refs->lock);
3489 }
3490
3491 spin_unlock(&delayed_refs->lock);
3492
3493 return ret;
3494}
3495
3496static void btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3497{
3498 struct btrfs_pending_snapshot *snapshot;
3499 struct list_head splice;
3500
3501 INIT_LIST_HEAD(&splice);
3502
3503 list_splice_init(&t->pending_snapshots, &splice);
3504
3505 while (!list_empty(&splice)) {
3506 snapshot = list_entry(splice.next,
3507 struct btrfs_pending_snapshot,
3508 list);
3509
3510 list_del_init(&snapshot->list);
3511
3512 kfree(snapshot);
3513 }
3514}
3515
3516static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3517{
3518 struct btrfs_inode *btrfs_inode;
3519 struct list_head splice;
3520
3521 INIT_LIST_HEAD(&splice);
3522
3523 spin_lock(&root->fs_info->delalloc_lock);
3524 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3525
3526 while (!list_empty(&splice)) {
3527 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3528 delalloc_inodes);
3529
3530 list_del_init(&btrfs_inode->delalloc_inodes);
3531
3532 btrfs_invalidate_inodes(btrfs_inode->root);
3533 }
3534
3535 spin_unlock(&root->fs_info->delalloc_lock);
3536}
3537
3538static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3539 struct extent_io_tree *dirty_pages,
3540 int mark)
3541{
3542 int ret;
3543 struct page *page;
3544 struct inode *btree_inode = root->fs_info->btree_inode;
3545 struct extent_buffer *eb;
3546 u64 start = 0;
3547 u64 end;
3548 u64 offset;
3549 unsigned long index;
3550
3551 while (1) {
3552 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3553 mark);
3554 if (ret)
3555 break;
3556
3557 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3558 while (start <= end) {
3559 index = start >> PAGE_CACHE_SHIFT;
3560 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3561 page = find_get_page(btree_inode->i_mapping, index);
3562 if (!page)
3563 continue;
3564 offset = page_offset(page);
3565
3566 spin_lock(&dirty_pages->buffer_lock);
3567 eb = radix_tree_lookup(
3568 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3569 offset >> PAGE_CACHE_SHIFT);
3570 spin_unlock(&dirty_pages->buffer_lock);
3571 if (eb)
3572 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3573 &eb->bflags);
3574 if (PageWriteback(page))
3575 end_page_writeback(page);
3576
3577 lock_page(page);
3578 if (PageDirty(page)) {
3579 clear_page_dirty_for_io(page);
3580 spin_lock_irq(&page->mapping->tree_lock);
3581 radix_tree_tag_clear(&page->mapping->page_tree,
3582 page_index(page),
3583 PAGECACHE_TAG_DIRTY);
3584 spin_unlock_irq(&page->mapping->tree_lock);
3585 }
3586
3587 unlock_page(page);
3588 page_cache_release(page);
3589 }
3590 }
3591
3592 return ret;
3593}
3594
3595static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3596 struct extent_io_tree *pinned_extents)
3597{
3598 struct extent_io_tree *unpin;
3599 u64 start;
3600 u64 end;
3601 int ret;
3602 bool loop = true;
3603
3604 unpin = pinned_extents;
3605again:
3606 while (1) {
3607 ret = find_first_extent_bit(unpin, 0, &start, &end,
3608 EXTENT_DIRTY);
3609 if (ret)
3610 break;
3611
3612 /* opt_discard */
3613 if (btrfs_test_opt(root, DISCARD))
3614 ret = btrfs_error_discard_extent(root, start,
3615 end + 1 - start,
3616 NULL);
3617
3618 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3619 btrfs_error_unpin_extent_range(root, start, end);
3620 cond_resched();
3621 }
3622
3623 if (loop) {
3624 if (unpin == &root->fs_info->freed_extents[0])
3625 unpin = &root->fs_info->freed_extents[1];
3626 else
3627 unpin = &root->fs_info->freed_extents[0];
3628 loop = false;
3629 goto again;
3630 }
3631
3632 return 0;
3633}
3634
3635void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
3636 struct btrfs_root *root)
3637{
3638 btrfs_destroy_delayed_refs(cur_trans, root);
3639 btrfs_block_rsv_release(root, &root->fs_info->trans_block_rsv,
3640 cur_trans->dirty_pages.dirty_bytes);
3641
3642 /* FIXME: cleanup wait for commit */
3643 cur_trans->in_commit = 1;
3644 cur_trans->blocked = 1;
3645 wake_up(&root->fs_info->transaction_blocked_wait);
3646
3647 cur_trans->blocked = 0;
3648 wake_up(&root->fs_info->transaction_wait);
3649
3650 cur_trans->commit_done = 1;
3651 wake_up(&cur_trans->commit_wait);
3652
3653 btrfs_destroy_delayed_inodes(root);
3654 btrfs_assert_delayed_root_empty(root);
3655
3656 btrfs_destroy_pending_snapshots(cur_trans);
3657
3658 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
3659 EXTENT_DIRTY);
3660 btrfs_destroy_pinned_extent(root,
3661 root->fs_info->pinned_extents);
3662
3663 /*
3664 memset(cur_trans, 0, sizeof(*cur_trans));
3665 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
3666 */
3667}
3668
3669int btrfs_cleanup_transaction(struct btrfs_root *root)
3670{
3671 struct btrfs_transaction *t;
3672 LIST_HEAD(list);
3673
3674 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3675
3676 spin_lock(&root->fs_info->trans_lock);
3677 list_splice_init(&root->fs_info->trans_list, &list);
3678 root->fs_info->trans_no_join = 1;
3679 spin_unlock(&root->fs_info->trans_lock);
3680
3681 while (!list_empty(&list)) {
3682 t = list_entry(list.next, struct btrfs_transaction, list);
3683 if (!t)
3684 break;
3685
3686 btrfs_destroy_ordered_operations(root);
3687
3688 btrfs_destroy_ordered_extents(root);
3689
3690 btrfs_destroy_delayed_refs(t, root);
3691
3692 btrfs_block_rsv_release(root,
3693 &root->fs_info->trans_block_rsv,
3694 t->dirty_pages.dirty_bytes);
3695
3696 /* FIXME: cleanup wait for commit */
3697 t->in_commit = 1;
3698 t->blocked = 1;
3699 smp_mb();
3700 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3701 wake_up(&root->fs_info->transaction_blocked_wait);
3702
3703 t->blocked = 0;
3704 smp_mb();
3705 if (waitqueue_active(&root->fs_info->transaction_wait))
3706 wake_up(&root->fs_info->transaction_wait);
3707
3708 t->commit_done = 1;
3709 smp_mb();
3710 if (waitqueue_active(&t->commit_wait))
3711 wake_up(&t->commit_wait);
3712
3713 btrfs_destroy_delayed_inodes(root);
3714 btrfs_assert_delayed_root_empty(root);
3715
3716 btrfs_destroy_pending_snapshots(t);
3717
3718 btrfs_destroy_delalloc_inodes(root);
3719
3720 spin_lock(&root->fs_info->trans_lock);
3721 root->fs_info->running_transaction = NULL;
3722 spin_unlock(&root->fs_info->trans_lock);
3723
3724 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3725 EXTENT_DIRTY);
3726
3727 btrfs_destroy_pinned_extent(root,
3728 root->fs_info->pinned_extents);
3729
3730 atomic_set(&t->use_count, 0);
3731 list_del_init(&t->list);
3732 memset(t, 0, sizeof(*t));
3733 kmem_cache_free(btrfs_transaction_cachep, t);
3734 }
3735
3736 spin_lock(&root->fs_info->trans_lock);
3737 root->fs_info->trans_no_join = 0;
3738 spin_unlock(&root->fs_info->trans_lock);
3739 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3740
3741 return 0;
3742}
3743
3744static struct extent_io_ops btree_extent_io_ops = {
3745 .readpage_end_io_hook = btree_readpage_end_io_hook,
3746 .readpage_io_failed_hook = btree_io_failed_hook,
3747 .submit_bio_hook = btree_submit_bio_hook,
3748 /* note we're sharing with inode.c for the merge bio hook */
3749 .merge_bio_hook = btrfs_merge_bio_hook,
3750};
kernel source for telekom puls (alcatel/mediatek)