?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/linux...
[linux-2.6.git] / fs / btrfs / disk-io.c
blob89e99eb384db54a8006ae991bd83ea2e61abe0eb
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <asm/unaligned.h>
34 #include "compat.h"
35 #include "ctree.h"
36 #include "disk-io.h"
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "volumes.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
42 #include "locking.h"
43 #include "tree-log.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47
48 static struct extent_io_ops btree_extent_io_ops;
49 static void end_workqueue_fn(struct btrfs_work *work);
50 static void free_fs_root(struct btrfs_root *root);
51 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
52 int read_only);
53 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
54 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
55 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
56 struct btrfs_root *root);
57 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
58 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
59 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
60 struct extent_io_tree *dirty_pages,
61 int mark);
62 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
63 struct extent_io_tree *pinned_extents);
64 static int btrfs_cleanup_transaction(struct btrfs_root *root);
65
66 /*
67 * end_io_wq structs are used to do processing in task context when an IO is
68 * complete. This is used during reads to verify checksums, and it is used
69 * by writes to insert metadata for new file extents after IO is complete.
70 */
71 struct 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 */
87 struct async_submit_bio {
88 struct inode *inode;
89 struct bio *bio;
90 struct list_head list;
91 extent_submit_bio_hook_t *submit_bio_start;
92 extent_submit_bio_hook_t *submit_bio_done;
93 int rw;
94 int mirror_num;
95 unsigned long bio_flags;
96 /*
97 * bio_offset is optional, can be used if the pages in the bio
98 * can't tell us where in the file the bio should go
99 */
100 u64 bio_offset;
101 struct btrfs_work work;
102 };
103
104 /*
105 * Lockdep class keys for extent_buffer->lock's in this root. For a given
106 * eb, the lockdep key is determined by the btrfs_root it belongs to and
107 * the level the eb occupies in the tree.
108 *
109 * Different roots are used for different purposes and may nest inside each
110 * other and they require separate keysets. As lockdep keys should be
111 * static, assign keysets according to the purpose of the root as indicated
112 * by btrfs_root->objectid. This ensures that all special purpose roots
113 * have separate keysets.
114 *
115 * Lock-nesting across peer nodes is always done with the immediate parent
116 * node locked thus preventing deadlock. As lockdep doesn't know this, use
117 * subclass to avoid triggering lockdep warning in such cases.
118 *
119 * The key is set by the readpage_end_io_hook after the buffer has passed
120 * csum validation but before the pages are unlocked. It is also set by
121 * btrfs_init_new_buffer on freshly allocated blocks.
122 *
123 * We also add a check to make sure the highest level of the tree is the
124 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
125 * needs update as well.
126 */
127 #ifdef CONFIG_DEBUG_LOCK_ALLOC
128 # if BTRFS_MAX_LEVEL != 8
129 # error
130 # endif
131
132 static struct btrfs_lockdep_keyset {
133 u64 id; /* root objectid */
134 const char *name_stem; /* lock name stem */
135 char names[BTRFS_MAX_LEVEL + 1][20];
136 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
137 } btrfs_lockdep_keysets[] = {
138 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
139 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
140 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
141 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
142 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
143 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
144 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
145 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
146 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
147 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
148 { .id = 0, .name_stem = "tree" },
149 };
150
151 void __init btrfs_init_lockdep(void)
152 {
153 int i, j;
154
155 /* initialize lockdep class names */
156 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
157 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
158
159 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
160 snprintf(ks->names[j], sizeof(ks->names[j]),
161 "btrfs-%s-%02d", ks->name_stem, j);
162 }
163 }
164
165 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
166 int level)
167 {
168 struct btrfs_lockdep_keyset *ks;
169
170 BUG_ON(level >= ARRAY_SIZE(ks->keys));
171
172 /* find the matching keyset, id 0 is the default entry */
173 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
174 if (ks->id == objectid)
175 break;
176
177 lockdep_set_class_and_name(&eb->lock,
178 &ks->keys[level], ks->names[level]);
179 }
180
181 #endif
182
183 /*
184 * extents on the btree inode are pretty simple, there's one extent
185 * that covers the entire device
186 */
187 static struct extent_map *btree_get_extent(struct inode *inode,
188 struct page *page, size_t pg_offset, u64 start, u64 len,
189 int create)
190 {
191 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
192 struct extent_map *em;
193 int ret;
194
195 read_lock(&em_tree->lock);
196 em = lookup_extent_mapping(em_tree, start, len);
197 if (em) {
198 em->bdev =
199 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
200 read_unlock(&em_tree->lock);
201 goto out;
202 }
203 read_unlock(&em_tree->lock);
204
205 em = alloc_extent_map();
206 if (!em) {
207 em = ERR_PTR(-ENOMEM);
208 goto out;
209 }
210 em->start = 0;
211 em->len = (u64)-1;
212 em->block_len = (u64)-1;
213 em->block_start = 0;
214 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
215
216 write_lock(&em_tree->lock);
217 ret = add_extent_mapping(em_tree, em);
218 if (ret == -EEXIST) {
219 u64 failed_start = em->start;
220 u64 failed_len = em->len;
221
222 free_extent_map(em);
223 em = lookup_extent_mapping(em_tree, start, len);
224 if (em) {
225 ret = 0;
226 } else {
227 em = lookup_extent_mapping(em_tree, failed_start,
228 failed_len);
229 ret = -EIO;
230 }
231 } else if (ret) {
232 free_extent_map(em);
233 em = NULL;
234 }
235 write_unlock(&em_tree->lock);
236
237 if (ret)
238 em = ERR_PTR(ret);
239 out:
240 return em;
241 }
242
243 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
244 {
245 return crc32c(seed, data, len);
246 }
247
248 void btrfs_csum_final(u32 crc, char *result)
249 {
250 put_unaligned_le32(~crc, result);
251 }
252
253 /*
254 * compute the csum for a btree block, and either verify it or write it
255 * into the csum field of the block.
256 */
257 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
258 int verify)
259 {
260 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
261 char *result = NULL;
262 unsigned long len;
263 unsigned long cur_len;
264 unsigned long offset = BTRFS_CSUM_SIZE;
265 char *kaddr;
266 unsigned long map_start;
267 unsigned long map_len;
268 int err;
269 u32 crc = ~(u32)0;
270 unsigned long inline_result;
271
272 len = buf->len - offset;
273 while (len > 0) {
274 err = map_private_extent_buffer(buf, offset, 32,
275 &kaddr, &map_start, &map_len);
276 if (err)
277 return 1;
278 cur_len = min(len, map_len - (offset - map_start));
279 crc = btrfs_csum_data(root, kaddr + offset - map_start,
280 crc, cur_len);
281 len -= cur_len;
282 offset += cur_len;
283 }
284 if (csum_size > sizeof(inline_result)) {
285 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
286 if (!result)
287 return 1;
288 } else {
289 result = (char *)&inline_result;
290 }
291
292 btrfs_csum_final(crc, result);
293
294 if (verify) {
295 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
296 u32 val;
297 u32 found = 0;
298 memcpy(&found, result, csum_size);
299
300 read_extent_buffer(buf, &val, 0, csum_size);
301 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
302 "failed on %llu wanted %X found %X "
303 "level %d\n",
304 root->fs_info->sb->s_id,
305 (unsigned long long)buf->start, val, found,
306 btrfs_header_level(buf));
307 if (result != (char *)&inline_result)
308 kfree(result);
309 return 1;
310 }
311 } else {
312 write_extent_buffer(buf, result, 0, csum_size);
313 }
314 if (result != (char *)&inline_result)
315 kfree(result);
316 return 0;
317 }
318
319 /*
320 * we can't consider a given block up to date unless the transid of the
321 * block matches the transid in the parent node's pointer. This is how we
322 * detect blocks that either didn't get written at all or got written
323 * in the wrong place.
324 */
325 static int verify_parent_transid(struct extent_io_tree *io_tree,
326 struct extent_buffer *eb, u64 parent_transid)
327 {
328 struct extent_state *cached_state = NULL;
329 int ret;
330
331 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
332 return 0;
333
334 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
335 0, &cached_state, GFP_NOFS);
336 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
337 btrfs_header_generation(eb) == parent_transid) {
338 ret = 0;
339 goto out;
340 }
341 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
342 "found %llu\n",
343 (unsigned long long)eb->start,
344 (unsigned long long)parent_transid,
345 (unsigned long long)btrfs_header_generation(eb));
346 ret = 1;
347 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
348 out:
349 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
350 &cached_state, GFP_NOFS);
351 return ret;
352 }
353
354 /*
355 * helper to read a given tree block, doing retries as required when
356 * the checksums don't match and we have alternate mirrors to try.
357 */
358 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
359 struct extent_buffer *eb,
360 u64 start, u64 parent_transid)
361 {
362 struct extent_io_tree *io_tree;
363 int ret;
364 int num_copies = 0;
365 int mirror_num = 0;
366
367 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
368 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
369 while (1) {
370 ret = read_extent_buffer_pages(io_tree, eb, start,
371 WAIT_COMPLETE,
372 btree_get_extent, mirror_num);
373 if (!ret &&
374 !verify_parent_transid(io_tree, eb, parent_transid))
375 return ret;
376
377 /*
378 * This buffer's crc is fine, but its contents are corrupted, so
379 * there is no reason to read the other copies, they won't be
380 * any less wrong.
381 */
382 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
383 return ret;
384
385 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
386 eb->start, eb->len);
387 if (num_copies == 1)
388 return ret;
389
390 mirror_num++;
391 if (mirror_num > num_copies)
392 return ret;
393 }
394 return -EIO;
395 }
396
397 /*
398 * checksum a dirty tree block before IO. This has extra checks to make sure
399 * we only fill in the checksum field in the first page of a multi-page block
400 */
401
402 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
403 {
404 struct extent_io_tree *tree;
405 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
406 u64 found_start;
407 unsigned long len;
408 struct extent_buffer *eb;
409 int ret;
410
411 tree = &BTRFS_I(page->mapping->host)->io_tree;
412
413 if (page->private == EXTENT_PAGE_PRIVATE) {
414 WARN_ON(1);
415 goto out;
416 }
417 if (!page->private) {
418 WARN_ON(1);
419 goto out;
420 }
421 len = page->private >> 2;
422 WARN_ON(len == 0);
423
424 eb = alloc_extent_buffer(tree, start, len, page);
425 if (eb == NULL) {
426 WARN_ON(1);
427 goto out;
428 }
429 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
430 btrfs_header_generation(eb));
431 BUG_ON(ret);
432 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
433
434 found_start = btrfs_header_bytenr(eb);
435 if (found_start != start) {
436 WARN_ON(1);
437 goto err;
438 }
439 if (eb->first_page != page) {
440 WARN_ON(1);
441 goto err;
442 }
443 if (!PageUptodate(page)) {
444 WARN_ON(1);
445 goto err;
446 }
447 csum_tree_block(root, eb, 0);
448 err:
449 free_extent_buffer(eb);
450 out:
451 return 0;
452 }
453
454 static int check_tree_block_fsid(struct btrfs_root *root,
455 struct extent_buffer *eb)
456 {
457 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
458 u8 fsid[BTRFS_UUID_SIZE];
459 int ret = 1;
460
461 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
462 BTRFS_FSID_SIZE);
463 while (fs_devices) {
464 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
465 ret = 0;
466 break;
467 }
468 fs_devices = fs_devices->seed;
469 }
470 return ret;
471 }
472
473 #define CORRUPT(reason, eb, root, slot) \
474 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
475 "root=%llu, slot=%d\n", reason, \
476 (unsigned long long)btrfs_header_bytenr(eb), \
477 (unsigned long long)root->objectid, slot)
478
479 static noinline int check_leaf(struct btrfs_root *root,
480 struct extent_buffer *leaf)
481 {
482 struct btrfs_key key;
483 struct btrfs_key leaf_key;
484 u32 nritems = btrfs_header_nritems(leaf);
485 int slot;
486
487 if (nritems == 0)
488 return 0;
489
490 /* Check the 0 item */
491 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
492 BTRFS_LEAF_DATA_SIZE(root)) {
493 CORRUPT("invalid item offset size pair", leaf, root, 0);
494 return -EIO;
495 }
496
497 /*
498 * Check to make sure each items keys are in the correct order and their
499 * offsets make sense. We only have to loop through nritems-1 because
500 * we check the current slot against the next slot, which verifies the
501 * next slot's offset+size makes sense and that the current's slot
502 * offset is correct.
503 */
504 for (slot = 0; slot < nritems - 1; slot++) {
505 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
506 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
507
508 /* Make sure the keys are in the right order */
509 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
510 CORRUPT("bad key order", leaf, root, slot);
511 return -EIO;
512 }
513
514 /*
515 * Make sure the offset and ends are right, remember that the
516 * item data starts at the end of the leaf and grows towards the
517 * front.
518 */
519 if (btrfs_item_offset_nr(leaf, slot) !=
520 btrfs_item_end_nr(leaf, slot + 1)) {
521 CORRUPT("slot offset bad", leaf, root, slot);
522 return -EIO;
523 }
524
525 /*
526 * Check to make sure that we don't point outside of the leaf,
527 * just incase all the items are consistent to eachother, but
528 * all point outside of the leaf.
529 */
530 if (btrfs_item_end_nr(leaf, slot) >
531 BTRFS_LEAF_DATA_SIZE(root)) {
532 CORRUPT("slot end outside of leaf", leaf, root, slot);
533 return -EIO;
534 }
535 }
536
537 return 0;
538 }
539
540 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
541 struct extent_state *state)
542 {
543 struct extent_io_tree *tree;
544 u64 found_start;
545 int found_level;
546 unsigned long len;
547 struct extent_buffer *eb;
548 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
549 int ret = 0;
550
551 tree = &BTRFS_I(page->mapping->host)->io_tree;
552 if (page->private == EXTENT_PAGE_PRIVATE)
553 goto out;
554 if (!page->private)
555 goto out;
556
557 len = page->private >> 2;
558 WARN_ON(len == 0);
559
560 eb = alloc_extent_buffer(tree, start, len, page);
561 if (eb == NULL) {
562 ret = -EIO;
563 goto out;
564 }
565
566 found_start = btrfs_header_bytenr(eb);
567 if (found_start != start) {
568 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
569 "%llu %llu\n",
570 (unsigned long long)found_start,
571 (unsigned long long)eb->start);
572 ret = -EIO;
573 goto err;
574 }
575 if (eb->first_page != page) {
576 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
577 eb->first_page->index, page->index);
578 WARN_ON(1);
579 ret = -EIO;
580 goto err;
581 }
582 if (check_tree_block_fsid(root, eb)) {
583 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
584 (unsigned long long)eb->start);
585 ret = -EIO;
586 goto err;
587 }
588 found_level = btrfs_header_level(eb);
589
590 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
591 eb, found_level);
592
593 ret = csum_tree_block(root, eb, 1);
594 if (ret) {
595 ret = -EIO;
596 goto err;
597 }
598
599 /*
600 * If this is a leaf block and it is corrupt, set the corrupt bit so
601 * that we don't try and read the other copies of this block, just
602 * return -EIO.
603 */
604 if (found_level == 0 && check_leaf(root, eb)) {
605 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
606 ret = -EIO;
607 }
608
609 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
610 end = eb->start + end - 1;
611 err:
612 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
613 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
614 btree_readahead_hook(root, eb, eb->start, ret);
615 }
616
617 free_extent_buffer(eb);
618 out:
619 return ret;
620 }
621
622 static int btree_io_failed_hook(struct bio *failed_bio,
623 struct page *page, u64 start, u64 end,
624 int mirror_num, struct extent_state *state)
625 {
626 struct extent_io_tree *tree;
627 unsigned long len;
628 struct extent_buffer *eb;
629 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
630
631 tree = &BTRFS_I(page->mapping->host)->io_tree;
632 if (page->private == EXTENT_PAGE_PRIVATE)
633 goto out;
634 if (!page->private)
635 goto out;
636
637 len = page->private >> 2;
638 WARN_ON(len == 0);
639
640 eb = alloc_extent_buffer(tree, start, len, page);
641 if (eb == NULL)
642 goto out;
643
644 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
645 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
646 btree_readahead_hook(root, eb, eb->start, -EIO);
647 }
648 free_extent_buffer(eb);
649
650 out:
651 return -EIO; /* we fixed nothing */
652 }
653
654 static void end_workqueue_bio(struct bio *bio, int err)
655 {
656 struct end_io_wq *end_io_wq = bio->bi_private;
657 struct btrfs_fs_info *fs_info;
658
659 fs_info = end_io_wq->info;
660 end_io_wq->error = err;
661 end_io_wq->work.func = end_workqueue_fn;
662 end_io_wq->work.flags = 0;
663
664 if (bio->bi_rw & REQ_WRITE) {
665 if (end_io_wq->metadata == 1)
666 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
667 &end_io_wq->work);
668 else if (end_io_wq->metadata == 2)
669 btrfs_queue_worker(&fs_info->endio_freespace_worker,
670 &end_io_wq->work);
671 else
672 btrfs_queue_worker(&fs_info->endio_write_workers,
673 &end_io_wq->work);
674 } else {
675 if (end_io_wq->metadata)
676 btrfs_queue_worker(&fs_info->endio_meta_workers,
677 &end_io_wq->work);
678 else
679 btrfs_queue_worker(&fs_info->endio_workers,
680 &end_io_wq->work);
681 }
682 }
683
684 /*
685 * For the metadata arg you want
686 *
687 * 0 - if data
688 * 1 - if normal metadta
689 * 2 - if writing to the free space cache area
690 */
691 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
692 int metadata)
693 {
694 struct end_io_wq *end_io_wq;
695 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
696 if (!end_io_wq)
697 return -ENOMEM;
698
699 end_io_wq->private = bio->bi_private;
700 end_io_wq->end_io = bio->bi_end_io;
701 end_io_wq->info = info;
702 end_io_wq->error = 0;
703 end_io_wq->bio = bio;
704 end_io_wq->metadata = metadata;
705
706 bio->bi_private = end_io_wq;
707 bio->bi_end_io = end_workqueue_bio;
708 return 0;
709 }
710
711 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
712 {
713 unsigned long limit = min_t(unsigned long,
714 info->workers.max_workers,
715 info->fs_devices->open_devices);
716 return 256 * limit;
717 }
718
719 static void run_one_async_start(struct btrfs_work *work)
720 {
721 struct async_submit_bio *async;
722
723 async = container_of(work, struct async_submit_bio, work);
724 async->submit_bio_start(async->inode, async->rw, async->bio,
725 async->mirror_num, async->bio_flags,
726 async->bio_offset);
727 }
728
729 static void run_one_async_done(struct btrfs_work *work)
730 {
731 struct btrfs_fs_info *fs_info;
732 struct async_submit_bio *async;
733 int limit;
734
735 async = container_of(work, struct async_submit_bio, work);
736 fs_info = BTRFS_I(async->inode)->root->fs_info;
737
738 limit = btrfs_async_submit_limit(fs_info);
739 limit = limit * 2 / 3;
740
741 atomic_dec(&fs_info->nr_async_submits);
742
743 if (atomic_read(&fs_info->nr_async_submits) < limit &&
744 waitqueue_active(&fs_info->async_submit_wait))
745 wake_up(&fs_info->async_submit_wait);
746
747 async->submit_bio_done(async->inode, async->rw, async->bio,
748 async->mirror_num, async->bio_flags,
749 async->bio_offset);
750 }
751
752 static void run_one_async_free(struct btrfs_work *work)
753 {
754 struct async_submit_bio *async;
755
756 async = container_of(work, struct async_submit_bio, work);
757 kfree(async);
758 }
759
760 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
761 int rw, struct bio *bio, int mirror_num,
762 unsigned long bio_flags,
763 u64 bio_offset,
764 extent_submit_bio_hook_t *submit_bio_start,
765 extent_submit_bio_hook_t *submit_bio_done)
766 {
767 struct async_submit_bio *async;
768
769 async = kmalloc(sizeof(*async), GFP_NOFS);
770 if (!async)
771 return -ENOMEM;
772
773 async->inode = inode;
774 async->rw = rw;
775 async->bio = bio;
776 async->mirror_num = mirror_num;
777 async->submit_bio_start = submit_bio_start;
778 async->submit_bio_done = submit_bio_done;
779
780 async->work.func = run_one_async_start;
781 async->work.ordered_func = run_one_async_done;
782 async->work.ordered_free = run_one_async_free;
783
784 async->work.flags = 0;
785 async->bio_flags = bio_flags;
786 async->bio_offset = bio_offset;
787
788 atomic_inc(&fs_info->nr_async_submits);
789
790 if (rw & REQ_SYNC)
791 btrfs_set_work_high_prio(&async->work);
792
793 btrfs_queue_worker(&fs_info->workers, &async->work);
794
795 while (atomic_read(&fs_info->async_submit_draining) &&
796 atomic_read(&fs_info->nr_async_submits)) {
797 wait_event(fs_info->async_submit_wait,
798 (atomic_read(&fs_info->nr_async_submits) == 0));
799 }
800
801 return 0;
802 }
803
804 static int btree_csum_one_bio(struct bio *bio)
805 {
806 struct bio_vec *bvec = bio->bi_io_vec;
807 int bio_index = 0;
808 struct btrfs_root *root;
809
810 WARN_ON(bio->bi_vcnt <= 0);
811 while (bio_index < bio->bi_vcnt) {
812 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
813 csum_dirty_buffer(root, bvec->bv_page);
814 bio_index++;
815 bvec++;
816 }
817 return 0;
818 }
819
820 static int __btree_submit_bio_start(struct inode *inode, int rw,
821 struct bio *bio, int mirror_num,
822 unsigned long bio_flags,
823 u64 bio_offset)
824 {
825 /*
826 * when we're called for a write, we're already in the async
827 * submission context. Just jump into btrfs_map_bio
828 */
829 btree_csum_one_bio(bio);
830 return 0;
831 }
832
833 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
834 int mirror_num, unsigned long bio_flags,
835 u64 bio_offset)
836 {
837 /*
838 * when we're called for a write, we're already in the async
839 * submission context. Just jump into btrfs_map_bio
840 */
841 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
842 }
843
844 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
845 int mirror_num, unsigned long bio_flags,
846 u64 bio_offset)
847 {
848 int ret;
849
850 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
851 bio, 1);
852 BUG_ON(ret);
853
854 if (!(rw & REQ_WRITE)) {
855 /*
856 * called for a read, do the setup so that checksum validation
857 * can happen in the async kernel threads
858 */
859 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
860 mirror_num, 0);
861 }
862
863 /*
864 * kthread helpers are used to submit writes so that checksumming
865 * can happen in parallel across all CPUs
866 */
867 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
868 inode, rw, bio, mirror_num, 0,
869 bio_offset,
870 __btree_submit_bio_start,
871 __btree_submit_bio_done);
872 }
873
874 #ifdef CONFIG_MIGRATION
875 static int btree_migratepage(struct address_space *mapping,
876 struct page *newpage, struct page *page,
877 enum migrate_mode mode)
878 {
879 /*
880 * we can't safely write a btree page from here,
881 * we haven't done the locking hook
882 */
883 if (PageDirty(page))
884 return -EAGAIN;
885 /*
886 * Buffers may be managed in a filesystem specific way.
887 * We must have no buffers or drop them.
888 */
889 if (page_has_private(page) &&
890 !try_to_release_page(page, GFP_KERNEL))
891 return -EAGAIN;
892 return migrate_page(mapping, newpage, page, mode);
893 }
894 #endif
895
896 static int btree_writepage(struct page *page, struct writeback_control *wbc)
897 {
898 struct extent_io_tree *tree;
899 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
900 struct extent_buffer *eb;
901 int was_dirty;
902
903 tree = &BTRFS_I(page->mapping->host)->io_tree;
904 if (!(current->flags & PF_MEMALLOC)) {
905 return extent_write_full_page(tree, page,
906 btree_get_extent, wbc);
907 }
908
909 redirty_page_for_writepage(wbc, page);
910 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
911 WARN_ON(!eb);
912
913 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
914 if (!was_dirty) {
915 spin_lock(&root->fs_info->delalloc_lock);
916 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
917 spin_unlock(&root->fs_info->delalloc_lock);
918 }
919 free_extent_buffer(eb);
920
921 unlock_page(page);
922 return 0;
923 }
924
925 static int btree_writepages(struct address_space *mapping,
926 struct writeback_control *wbc)
927 {
928 struct extent_io_tree *tree;
929 tree = &BTRFS_I(mapping->host)->io_tree;
930 if (wbc->sync_mode == WB_SYNC_NONE) {
931 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
932 u64 num_dirty;
933 unsigned long thresh = 32 * 1024 * 1024;
934
935 if (wbc->for_kupdate)
936 return 0;
937
938 /* this is a bit racy, but that's ok */
939 num_dirty = root->fs_info->dirty_metadata_bytes;
940 if (num_dirty < thresh)
941 return 0;
942 }
943 return extent_writepages(tree, mapping, btree_get_extent, wbc);
944 }
945
946 static int btree_readpage(struct file *file, struct page *page)
947 {
948 struct extent_io_tree *tree;
949 tree = &BTRFS_I(page->mapping->host)->io_tree;
950 return extent_read_full_page(tree, page, btree_get_extent, 0);
951 }
952
953 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
954 {
955 struct extent_io_tree *tree;
956 struct extent_map_tree *map;
957 int ret;
958
959 if (PageWriteback(page) || PageDirty(page))
960 return 0;
961
962 tree = &BTRFS_I(page->mapping->host)->io_tree;
963 map = &BTRFS_I(page->mapping->host)->extent_tree;
964
965 ret = try_release_extent_state(map, tree, page, gfp_flags);
966 if (!ret)
967 return 0;
968
969 ret = try_release_extent_buffer(tree, page);
970 if (ret == 1) {
971 ClearPagePrivate(page);
972 set_page_private(page, 0);
973 page_cache_release(page);
974 }
975
976 return ret;
977 }
978
979 static void btree_invalidatepage(struct page *page, unsigned long offset)
980 {
981 struct extent_io_tree *tree;
982 tree = &BTRFS_I(page->mapping->host)->io_tree;
983 extent_invalidatepage(tree, page, offset);
984 btree_releasepage(page, GFP_NOFS);
985 if (PagePrivate(page)) {
986 printk(KERN_WARNING "btrfs warning page private not zero "
987 "on page %llu\n", (unsigned long long)page_offset(page));
988 ClearPagePrivate(page);
989 set_page_private(page, 0);
990 page_cache_release(page);
991 }
992 }
993
994 static const struct address_space_operations btree_aops = {
995 .readpage = btree_readpage,
996 .writepage = btree_writepage,
997 .writepages = btree_writepages,
998 .releasepage = btree_releasepage,
999 .invalidatepage = btree_invalidatepage,
1000 #ifdef CONFIG_MIGRATION
1001 .migratepage = btree_migratepage,
1002 #endif
1003 };
1004
1005 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1006 u64 parent_transid)
1007 {
1008 struct extent_buffer *buf = NULL;
1009 struct inode *btree_inode = root->fs_info->btree_inode;
1010 int ret = 0;
1011
1012 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1013 if (!buf)
1014 return 0;
1015 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1016 buf, 0, WAIT_NONE, btree_get_extent, 0);
1017 free_extent_buffer(buf);
1018 return ret;
1019 }
1020
1021 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1022 int mirror_num, struct extent_buffer **eb)
1023 {
1024 struct extent_buffer *buf = NULL;
1025 struct inode *btree_inode = root->fs_info->btree_inode;
1026 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1027 int ret;
1028
1029 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1030 if (!buf)
1031 return 0;
1032
1033 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1034
1035 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1036 btree_get_extent, mirror_num);
1037 if (ret) {
1038 free_extent_buffer(buf);
1039 return ret;
1040 }
1041
1042 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1043 free_extent_buffer(buf);
1044 return -EIO;
1045 } else if (extent_buffer_uptodate(io_tree, buf, NULL)) {
1046 *eb = buf;
1047 } else {
1048 free_extent_buffer(buf);
1049 }
1050 return 0;
1051 }
1052
1053 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1054 u64 bytenr, u32 blocksize)
1055 {
1056 struct inode *btree_inode = root->fs_info->btree_inode;
1057 struct extent_buffer *eb;
1058 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1059 bytenr, blocksize);
1060 return eb;
1061 }
1062
1063 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1064 u64 bytenr, u32 blocksize)
1065 {
1066 struct inode *btree_inode = root->fs_info->btree_inode;
1067 struct extent_buffer *eb;
1068
1069 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1070 bytenr, blocksize, NULL);
1071 return eb;
1072 }
1073
1074
1075 int btrfs_write_tree_block(struct extent_buffer *buf)
1076 {
1077 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1078 buf->start + buf->len - 1);
1079 }
1080
1081 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1082 {
1083 return filemap_fdatawait_range(buf->first_page->mapping,
1084 buf->start, buf->start + buf->len - 1);
1085 }
1086
1087 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1088 u32 blocksize, u64 parent_transid)
1089 {
1090 struct extent_buffer *buf = NULL;
1091 int ret;
1092
1093 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1094 if (!buf)
1095 return NULL;
1096
1097 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1098
1099 if (ret == 0)
1100 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1101 return buf;
1102
1103 }
1104
1105 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1106 struct extent_buffer *buf)
1107 {
1108 struct inode *btree_inode = root->fs_info->btree_inode;
1109 if (btrfs_header_generation(buf) ==
1110 root->fs_info->running_transaction->transid) {
1111 btrfs_assert_tree_locked(buf);
1112
1113 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1114 spin_lock(&root->fs_info->delalloc_lock);
1115 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1116 root->fs_info->dirty_metadata_bytes -= buf->len;
1117 else
1118 WARN_ON(1);
1119 spin_unlock(&root->fs_info->delalloc_lock);
1120 }
1121
1122 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1123 btrfs_set_lock_blocking(buf);
1124 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1125 buf);
1126 }
1127 return 0;
1128 }
1129
1130 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1131 u32 stripesize, struct btrfs_root *root,
1132 struct btrfs_fs_info *fs_info,
1133 u64 objectid)
1134 {
1135 root->node = NULL;
1136 root->commit_root = NULL;
1137 root->sectorsize = sectorsize;
1138 root->nodesize = nodesize;
1139 root->leafsize = leafsize;
1140 root->stripesize = stripesize;
1141 root->ref_cows = 0;
1142 root->track_dirty = 0;
1143 root->in_radix = 0;
1144 root->orphan_item_inserted = 0;
1145 root->orphan_cleanup_state = 0;
1146
1147 root->fs_info = fs_info;
1148 root->objectid = objectid;
1149 root->last_trans = 0;
1150 root->highest_objectid = 0;
1151 root->name = NULL;
1152 root->inode_tree = RB_ROOT;
1153 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1154 root->block_rsv = NULL;
1155 root->orphan_block_rsv = NULL;
1156
1157 INIT_LIST_HEAD(&root->dirty_list);
1158 INIT_LIST_HEAD(&root->orphan_list);
1159 INIT_LIST_HEAD(&root->root_list);
1160 spin_lock_init(&root->orphan_lock);
1161 spin_lock_init(&root->inode_lock);
1162 spin_lock_init(&root->accounting_lock);
1163 mutex_init(&root->objectid_mutex);
1164 mutex_init(&root->log_mutex);
1165 init_waitqueue_head(&root->log_writer_wait);
1166 init_waitqueue_head(&root->log_commit_wait[0]);
1167 init_waitqueue_head(&root->log_commit_wait[1]);
1168 atomic_set(&root->log_commit[0], 0);
1169 atomic_set(&root->log_commit[1], 0);
1170 atomic_set(&root->log_writers, 0);
1171 root->log_batch = 0;
1172 root->log_transid = 0;
1173 root->last_log_commit = 0;
1174 extent_io_tree_init(&root->dirty_log_pages,
1175 fs_info->btree_inode->i_mapping);
1176
1177 memset(&root->root_key, 0, sizeof(root->root_key));
1178 memset(&root->root_item, 0, sizeof(root->root_item));
1179 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1180 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1181 root->defrag_trans_start = fs_info->generation;
1182 init_completion(&root->kobj_unregister);
1183 root->defrag_running = 0;
1184 root->root_key.objectid = objectid;
1185 root->anon_dev = 0;
1186 return 0;
1187 }
1188
1189 static int find_and_setup_root(struct btrfs_root *tree_root,
1190 struct btrfs_fs_info *fs_info,
1191 u64 objectid,
1192 struct btrfs_root *root)
1193 {
1194 int ret;
1195 u32 blocksize;
1196 u64 generation;
1197
1198 __setup_root(tree_root->nodesize, tree_root->leafsize,
1199 tree_root->sectorsize, tree_root->stripesize,
1200 root, fs_info, objectid);
1201 ret = btrfs_find_last_root(tree_root, objectid,
1202 &root->root_item, &root->root_key);
1203 if (ret > 0)
1204 return -ENOENT;
1205 BUG_ON(ret);
1206
1207 generation = btrfs_root_generation(&root->root_item);
1208 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1209 root->commit_root = NULL;
1210 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1211 blocksize, generation);
1212 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1213 free_extent_buffer(root->node);
1214 root->node = NULL;
1215 return -EIO;
1216 }
1217 root->commit_root = btrfs_root_node(root);
1218 return 0;
1219 }
1220
1221 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1222 struct btrfs_fs_info *fs_info)
1223 {
1224 struct btrfs_root *root;
1225 struct btrfs_root *tree_root = fs_info->tree_root;
1226 struct extent_buffer *leaf;
1227
1228 root = kzalloc(sizeof(*root), GFP_NOFS);
1229 if (!root)
1230 return ERR_PTR(-ENOMEM);
1231
1232 __setup_root(tree_root->nodesize, tree_root->leafsize,
1233 tree_root->sectorsize, tree_root->stripesize,
1234 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1235
1236 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1237 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1238 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1239 /*
1240 * log trees do not get reference counted because they go away
1241 * before a real commit is actually done. They do store pointers
1242 * to file data extents, and those reference counts still get
1243 * updated (along with back refs to the log tree).
1244 */
1245 root->ref_cows = 0;
1246
1247 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1248 BTRFS_TREE_LOG_OBJECTID, NULL,
1249 0, 0, 0, 0);
1250 if (IS_ERR(leaf)) {
1251 kfree(root);
1252 return ERR_CAST(leaf);
1253 }
1254
1255 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1256 btrfs_set_header_bytenr(leaf, leaf->start);
1257 btrfs_set_header_generation(leaf, trans->transid);
1258 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1259 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1260 root->node = leaf;
1261
1262 write_extent_buffer(root->node, root->fs_info->fsid,
1263 (unsigned long)btrfs_header_fsid(root->node),
1264 BTRFS_FSID_SIZE);
1265 btrfs_mark_buffer_dirty(root->node);
1266 btrfs_tree_unlock(root->node);
1267 return root;
1268 }
1269
1270 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1271 struct btrfs_fs_info *fs_info)
1272 {
1273 struct btrfs_root *log_root;
1274
1275 log_root = alloc_log_tree(trans, fs_info);
1276 if (IS_ERR(log_root))
1277 return PTR_ERR(log_root);
1278 WARN_ON(fs_info->log_root_tree);
1279 fs_info->log_root_tree = log_root;
1280 return 0;
1281 }
1282
1283 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1284 struct btrfs_root *root)
1285 {
1286 struct btrfs_root *log_root;
1287 struct btrfs_inode_item *inode_item;
1288
1289 log_root = alloc_log_tree(trans, root->fs_info);
1290 if (IS_ERR(log_root))
1291 return PTR_ERR(log_root);
1292
1293 log_root->last_trans = trans->transid;
1294 log_root->root_key.offset = root->root_key.objectid;
1295
1296 inode_item = &log_root->root_item.inode;
1297 inode_item->generation = cpu_to_le64(1);
1298 inode_item->size = cpu_to_le64(3);
1299 inode_item->nlink = cpu_to_le32(1);
1300 inode_item->nbytes = cpu_to_le64(root->leafsize);
1301 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1302
1303 btrfs_set_root_node(&log_root->root_item, log_root->node);
1304
1305 WARN_ON(root->log_root);
1306 root->log_root = log_root;
1307 root->log_transid = 0;
1308 root->last_log_commit = 0;
1309 return 0;
1310 }
1311
1312 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1313 struct btrfs_key *location)
1314 {
1315 struct btrfs_root *root;
1316 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1317 struct btrfs_path *path;
1318 struct extent_buffer *l;
1319 u64 generation;
1320 u32 blocksize;
1321 int ret = 0;
1322
1323 root = kzalloc(sizeof(*root), GFP_NOFS);
1324 if (!root)
1325 return ERR_PTR(-ENOMEM);
1326 if (location->offset == (u64)-1) {
1327 ret = find_and_setup_root(tree_root, fs_info,
1328 location->objectid, root);
1329 if (ret) {
1330 kfree(root);
1331 return ERR_PTR(ret);
1332 }
1333 goto out;
1334 }
1335
1336 __setup_root(tree_root->nodesize, tree_root->leafsize,
1337 tree_root->sectorsize, tree_root->stripesize,
1338 root, fs_info, location->objectid);
1339
1340 path = btrfs_alloc_path();
1341 if (!path) {
1342 kfree(root);
1343 return ERR_PTR(-ENOMEM);
1344 }
1345 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1346 if (ret == 0) {
1347 l = path->nodes[0];
1348 read_extent_buffer(l, &root->root_item,
1349 btrfs_item_ptr_offset(l, path->slots[0]),
1350 sizeof(root->root_item));
1351 memcpy(&root->root_key, location, sizeof(*location));
1352 }
1353 btrfs_free_path(path);
1354 if (ret) {
1355 kfree(root);
1356 if (ret > 0)
1357 ret = -ENOENT;
1358 return ERR_PTR(ret);
1359 }
1360
1361 generation = btrfs_root_generation(&root->root_item);
1362 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1363 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1364 blocksize, generation);
1365 root->commit_root = btrfs_root_node(root);
1366 BUG_ON(!root->node);
1367 out:
1368 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1369 root->ref_cows = 1;
1370 btrfs_check_and_init_root_item(&root->root_item);
1371 }
1372
1373 return root;
1374 }
1375
1376 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1377 struct btrfs_key *location)
1378 {
1379 struct btrfs_root *root;
1380 int ret;
1381
1382 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1383 return fs_info->tree_root;
1384 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1385 return fs_info->extent_root;
1386 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1387 return fs_info->chunk_root;
1388 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1389 return fs_info->dev_root;
1390 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1391 return fs_info->csum_root;
1392 again:
1393 spin_lock(&fs_info->fs_roots_radix_lock);
1394 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1395 (unsigned long)location->objectid);
1396 spin_unlock(&fs_info->fs_roots_radix_lock);
1397 if (root)
1398 return root;
1399
1400 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1401 if (IS_ERR(root))
1402 return root;
1403
1404 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1405 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1406 GFP_NOFS);
1407 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1408 ret = -ENOMEM;
1409 goto fail;
1410 }
1411
1412 btrfs_init_free_ino_ctl(root);
1413 mutex_init(&root->fs_commit_mutex);
1414 spin_lock_init(&root->cache_lock);
1415 init_waitqueue_head(&root->cache_wait);
1416
1417 ret = get_anon_bdev(&root->anon_dev);
1418 if (ret)
1419 goto fail;
1420
1421 if (btrfs_root_refs(&root->root_item) == 0) {
1422 ret = -ENOENT;
1423 goto fail;
1424 }
1425
1426 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1427 if (ret < 0)
1428 goto fail;
1429 if (ret == 0)
1430 root->orphan_item_inserted = 1;
1431
1432 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1433 if (ret)
1434 goto fail;
1435
1436 spin_lock(&fs_info->fs_roots_radix_lock);
1437 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1438 (unsigned long)root->root_key.objectid,
1439 root);
1440 if (ret == 0)
1441 root->in_radix = 1;
1442
1443 spin_unlock(&fs_info->fs_roots_radix_lock);
1444 radix_tree_preload_end();
1445 if (ret) {
1446 if (ret == -EEXIST) {
1447 free_fs_root(root);
1448 goto again;
1449 }
1450 goto fail;
1451 }
1452
1453 ret = btrfs_find_dead_roots(fs_info->tree_root,
1454 root->root_key.objectid);
1455 WARN_ON(ret);
1456 return root;
1457 fail:
1458 free_fs_root(root);
1459 return ERR_PTR(ret);
1460 }
1461
1462 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1463 {
1464 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1465 int ret = 0;
1466 struct btrfs_device *device;
1467 struct backing_dev_info *bdi;
1468
1469 rcu_read_lock();
1470 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1471 if (!device->bdev)
1472 continue;
1473 bdi = blk_get_backing_dev_info(device->bdev);
1474 if (bdi && bdi_congested(bdi, bdi_bits)) {
1475 ret = 1;
1476 break;
1477 }
1478 }
1479 rcu_read_unlock();
1480 return ret;
1481 }
1482
1483 /*
1484 * If this fails, caller must call bdi_destroy() to get rid of the
1485 * bdi again.
1486 */
1487 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1488 {
1489 int err;
1490
1491 bdi->capabilities = BDI_CAP_MAP_COPY;
1492 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1493 if (err)
1494 return err;
1495
1496 bdi->ra_pages = default_backing_dev_info.ra_pages;
1497 bdi->congested_fn = btrfs_congested_fn;
1498 bdi->congested_data = info;
1499 return 0;
1500 }
1501
1502 static int bio_ready_for_csum(struct bio *bio)
1503 {
1504 u64 length = 0;
1505 u64 buf_len = 0;
1506 u64 start = 0;
1507 struct page *page;
1508 struct extent_io_tree *io_tree = NULL;
1509 struct bio_vec *bvec;
1510 int i;
1511 int ret;
1512
1513 bio_for_each_segment(bvec, bio, i) {
1514 page = bvec->bv_page;
1515 if (page->private == EXTENT_PAGE_PRIVATE) {
1516 length += bvec->bv_len;
1517 continue;
1518 }
1519 if (!page->private) {
1520 length += bvec->bv_len;
1521 continue;
1522 }
1523 length = bvec->bv_len;
1524 buf_len = page->private >> 2;
1525 start = page_offset(page) + bvec->bv_offset;
1526 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1527 }
1528 /* are we fully contained in this bio? */
1529 if (buf_len <= length)
1530 return 1;
1531
1532 ret = extent_range_uptodate(io_tree, start + length,
1533 start + buf_len - 1);
1534 return ret;
1535 }
1536
1537 /*
1538 * called by the kthread helper functions to finally call the bio end_io
1539 * functions. This is where read checksum verification actually happens
1540 */
1541 static void end_workqueue_fn(struct btrfs_work *work)
1542 {
1543 struct bio *bio;
1544 struct end_io_wq *end_io_wq;
1545 struct btrfs_fs_info *fs_info;
1546 int error;
1547
1548 end_io_wq = container_of(work, struct end_io_wq, work);
1549 bio = end_io_wq->bio;
1550 fs_info = end_io_wq->info;
1551
1552 /* metadata bio reads are special because the whole tree block must
1553 * be checksummed at once. This makes sure the entire block is in
1554 * ram and up to date before trying to verify things. For
1555 * blocksize <= pagesize, it is basically a noop
1556 */
1557 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1558 !bio_ready_for_csum(bio)) {
1559 btrfs_queue_worker(&fs_info->endio_meta_workers,
1560 &end_io_wq->work);
1561 return;
1562 }
1563 error = end_io_wq->error;
1564 bio->bi_private = end_io_wq->private;
1565 bio->bi_end_io = end_io_wq->end_io;
1566 kfree(end_io_wq);
1567 bio_endio(bio, error);
1568 }
1569
1570 static int cleaner_kthread(void *arg)
1571 {
1572 struct btrfs_root *root = arg;
1573
1574 do {
1575 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1576
1577 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1578 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1579 btrfs_run_delayed_iputs(root);
1580 btrfs_clean_old_snapshots(root);
1581 mutex_unlock(&root->fs_info->cleaner_mutex);
1582 btrfs_run_defrag_inodes(root->fs_info);
1583 }
1584
1585 if (!try_to_freeze()) {
1586 set_current_state(TASK_INTERRUPTIBLE);
1587 if (!kthread_should_stop())
1588 schedule();
1589 __set_current_state(TASK_RUNNING);
1590 }
1591 } while (!kthread_should_stop());
1592 return 0;
1593 }
1594
1595 static int transaction_kthread(void *arg)
1596 {
1597 struct btrfs_root *root = arg;
1598 struct btrfs_trans_handle *trans;
1599 struct btrfs_transaction *cur;
1600 u64 transid;
1601 unsigned long now;
1602 unsigned long delay;
1603 int ret;
1604
1605 do {
1606 delay = HZ * 30;
1607 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1608 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1609
1610 spin_lock(&root->fs_info->trans_lock);
1611 cur = root->fs_info->running_transaction;
1612 if (!cur) {
1613 spin_unlock(&root->fs_info->trans_lock);
1614 goto sleep;
1615 }
1616
1617 now = get_seconds();
1618 if (!cur->blocked &&
1619 (now < cur->start_time || now - cur->start_time < 30)) {
1620 spin_unlock(&root->fs_info->trans_lock);
1621 delay = HZ * 5;
1622 goto sleep;
1623 }
1624 transid = cur->transid;
1625 spin_unlock(&root->fs_info->trans_lock);
1626
1627 trans = btrfs_join_transaction(root);
1628 BUG_ON(IS_ERR(trans));
1629 if (transid == trans->transid) {
1630 ret = btrfs_commit_transaction(trans, root);
1631 BUG_ON(ret);
1632 } else {
1633 btrfs_end_transaction(trans, root);
1634 }
1635 sleep:
1636 wake_up_process(root->fs_info->cleaner_kthread);
1637 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1638
1639 if (!try_to_freeze()) {
1640 set_current_state(TASK_INTERRUPTIBLE);
1641 if (!kthread_should_stop() &&
1642 !btrfs_transaction_blocked(root->fs_info))
1643 schedule_timeout(delay);
1644 __set_current_state(TASK_RUNNING);
1645 }
1646 } while (!kthread_should_stop());
1647 return 0;
1648 }
1649
1650 /*
1651 * this will find the highest generation in the array of
1652 * root backups. The index of the highest array is returned,
1653 * or -1 if we can't find anything.
1654 *
1655 * We check to make sure the array is valid by comparing the
1656 * generation of the latest root in the array with the generation
1657 * in the super block. If they don't match we pitch it.
1658 */
1659 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1660 {
1661 u64 cur;
1662 int newest_index = -1;
1663 struct btrfs_root_backup *root_backup;
1664 int i;
1665
1666 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1667 root_backup = info->super_copy->super_roots + i;
1668 cur = btrfs_backup_tree_root_gen(root_backup);
1669 if (cur == newest_gen)
1670 newest_index = i;
1671 }
1672
1673 /* check to see if we actually wrapped around */
1674 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1675 root_backup = info->super_copy->super_roots;
1676 cur = btrfs_backup_tree_root_gen(root_backup);
1677 if (cur == newest_gen)
1678 newest_index = 0;
1679 }
1680 return newest_index;
1681 }
1682
1683
1684 /*
1685 * find the oldest backup so we know where to store new entries
1686 * in the backup array. This will set the backup_root_index
1687 * field in the fs_info struct
1688 */
1689 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1690 u64 newest_gen)
1691 {
1692 int newest_index = -1;
1693
1694 newest_index = find_newest_super_backup(info, newest_gen);
1695 /* if there was garbage in there, just move along */
1696 if (newest_index == -1) {
1697 info->backup_root_index = 0;
1698 } else {
1699 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1700 }
1701 }
1702
1703 /*
1704 * copy all the root pointers into the super backup array.
1705 * this will bump the backup pointer by one when it is
1706 * done
1707 */
1708 static void backup_super_roots(struct btrfs_fs_info *info)
1709 {
1710 int next_backup;
1711 struct btrfs_root_backup *root_backup;
1712 int last_backup;
1713
1714 next_backup = info->backup_root_index;
1715 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1716 BTRFS_NUM_BACKUP_ROOTS;
1717
1718 /*
1719 * just overwrite the last backup if we're at the same generation
1720 * this happens only at umount
1721 */
1722 root_backup = info->super_for_commit->super_roots + last_backup;
1723 if (btrfs_backup_tree_root_gen(root_backup) ==
1724 btrfs_header_generation(info->tree_root->node))
1725 next_backup = last_backup;
1726
1727 root_backup = info->super_for_commit->super_roots + next_backup;
1728
1729 /*
1730 * make sure all of our padding and empty slots get zero filled
1731 * regardless of which ones we use today
1732 */
1733 memset(root_backup, 0, sizeof(*root_backup));
1734
1735 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1736
1737 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1738 btrfs_set_backup_tree_root_gen(root_backup,
1739 btrfs_header_generation(info->tree_root->node));
1740
1741 btrfs_set_backup_tree_root_level(root_backup,
1742 btrfs_header_level(info->tree_root->node));
1743
1744 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1745 btrfs_set_backup_chunk_root_gen(root_backup,
1746 btrfs_header_generation(info->chunk_root->node));
1747 btrfs_set_backup_chunk_root_level(root_backup,
1748 btrfs_header_level(info->chunk_root->node));
1749
1750 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1751 btrfs_set_backup_extent_root_gen(root_backup,
1752 btrfs_header_generation(info->extent_root->node));
1753 btrfs_set_backup_extent_root_level(root_backup,
1754 btrfs_header_level(info->extent_root->node));
1755
1756 /*
1757 * we might commit during log recovery, which happens before we set
1758 * the fs_root. Make sure it is valid before we fill it in.
1759 */
1760 if (info->fs_root && info->fs_root->node) {
1761 btrfs_set_backup_fs_root(root_backup,
1762 info->fs_root->node->start);
1763 btrfs_set_backup_fs_root_gen(root_backup,
1764 btrfs_header_generation(info->fs_root->node));
1765 btrfs_set_backup_fs_root_level(root_backup,
1766 btrfs_header_level(info->fs_root->node));
1767 }
1768
1769 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1770 btrfs_set_backup_dev_root_gen(root_backup,
1771 btrfs_header_generation(info->dev_root->node));
1772 btrfs_set_backup_dev_root_level(root_backup,
1773 btrfs_header_level(info->dev_root->node));
1774
1775 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1776 btrfs_set_backup_csum_root_gen(root_backup,
1777 btrfs_header_generation(info->csum_root->node));
1778 btrfs_set_backup_csum_root_level(root_backup,
1779 btrfs_header_level(info->csum_root->node));
1780
1781 btrfs_set_backup_total_bytes(root_backup,
1782 btrfs_super_total_bytes(info->super_copy));
1783 btrfs_set_backup_bytes_used(root_backup,
1784 btrfs_super_bytes_used(info->super_copy));
1785 btrfs_set_backup_num_devices(root_backup,
1786 btrfs_super_num_devices(info->super_copy));
1787
1788 /*
1789 * if we don't copy this out to the super_copy, it won't get remembered
1790 * for the next commit
1791 */
1792 memcpy(&info->super_copy->super_roots,
1793 &info->super_for_commit->super_roots,
1794 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1795 }
1796
1797 /*
1798 * this copies info out of the root backup array and back into
1799 * the in-memory super block. It is meant to help iterate through
1800 * the array, so you send it the number of backups you've already
1801 * tried and the last backup index you used.
1802 *
1803 * this returns -1 when it has tried all the backups
1804 */
1805 static noinline int next_root_backup(struct btrfs_fs_info *info,
1806 struct btrfs_super_block *super,
1807 int *num_backups_tried, int *backup_index)
1808 {
1809 struct btrfs_root_backup *root_backup;
1810 int newest = *backup_index;
1811
1812 if (*num_backups_tried == 0) {
1813 u64 gen = btrfs_super_generation(super);
1814
1815 newest = find_newest_super_backup(info, gen);
1816 if (newest == -1)
1817 return -1;
1818
1819 *backup_index = newest;
1820 *num_backups_tried = 1;
1821 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1822 /* we've tried all the backups, all done */
1823 return -1;
1824 } else {
1825 /* jump to the next oldest backup */
1826 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1827 BTRFS_NUM_BACKUP_ROOTS;
1828 *backup_index = newest;
1829 *num_backups_tried += 1;
1830 }
1831 root_backup = super->super_roots + newest;
1832
1833 btrfs_set_super_generation(super,
1834 btrfs_backup_tree_root_gen(root_backup));
1835 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1836 btrfs_set_super_root_level(super,
1837 btrfs_backup_tree_root_level(root_backup));
1838 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1839
1840 /*
1841 * fixme: the total bytes and num_devices need to match or we should
1842 * need a fsck
1843 */
1844 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1845 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1846 return 0;
1847 }
1848
1849 /* helper to cleanup tree roots */
1850 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1851 {
1852 free_extent_buffer(info->tree_root->node);
1853 free_extent_buffer(info->tree_root->commit_root);
1854 free_extent_buffer(info->dev_root->node);
1855 free_extent_buffer(info->dev_root->commit_root);
1856 free_extent_buffer(info->extent_root->node);
1857 free_extent_buffer(info->extent_root->commit_root);
1858 free_extent_buffer(info->csum_root->node);
1859 free_extent_buffer(info->csum_root->commit_root);
1860
1861 info->tree_root->node = NULL;
1862 info->tree_root->commit_root = NULL;
1863 info->dev_root->node = NULL;
1864 info->dev_root->commit_root = NULL;
1865 info->extent_root->node = NULL;
1866 info->extent_root->commit_root = NULL;
1867 info->csum_root->node = NULL;
1868 info->csum_root->commit_root = NULL;
1869
1870 if (chunk_root) {
1871 free_extent_buffer(info->chunk_root->node);
1872 free_extent_buffer(info->chunk_root->commit_root);
1873 info->chunk_root->node = NULL;
1874 info->chunk_root->commit_root = NULL;
1875 }
1876 }
1877
1878
1879 struct btrfs_root *open_ctree(struct super_block *sb,
1880 struct btrfs_fs_devices *fs_devices,
1881 char *options)
1882 {
1883 u32 sectorsize;
1884 u32 nodesize;
1885 u32 leafsize;
1886 u32 blocksize;
1887 u32 stripesize;
1888 u64 generation;
1889 u64 features;
1890 struct btrfs_key location;
1891 struct buffer_head *bh;
1892 struct btrfs_super_block *disk_super;
1893 struct btrfs_root *tree_root = btrfs_sb(sb);
1894 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1895 struct btrfs_root *extent_root;
1896 struct btrfs_root *csum_root;
1897 struct btrfs_root *chunk_root;
1898 struct btrfs_root *dev_root;
1899 struct btrfs_root *log_tree_root;
1900 int ret;
1901 int err = -EINVAL;
1902 int num_backups_tried = 0;
1903 int backup_index = 0;
1904
1905 extent_root = fs_info->extent_root =
1906 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1907 csum_root = fs_info->csum_root =
1908 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1909 chunk_root = fs_info->chunk_root =
1910 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1911 dev_root = fs_info->dev_root =
1912 kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
1913
1914 if (!extent_root || !csum_root || !chunk_root || !dev_root) {
1915 err = -ENOMEM;
1916 goto fail;
1917 }
1918
1919 ret = init_srcu_struct(&fs_info->subvol_srcu);
1920 if (ret) {
1921 err = ret;
1922 goto fail;
1923 }
1924
1925 ret = setup_bdi(fs_info, &fs_info->bdi);
1926 if (ret) {
1927 err = ret;
1928 goto fail_srcu;
1929 }
1930
1931 fs_info->btree_inode = new_inode(sb);
1932 if (!fs_info->btree_inode) {
1933 err = -ENOMEM;
1934 goto fail_bdi;
1935 }
1936
1937 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1938
1939 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1940 INIT_LIST_HEAD(&fs_info->trans_list);
1941 INIT_LIST_HEAD(&fs_info->dead_roots);
1942 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1943 INIT_LIST_HEAD(&fs_info->hashers);
1944 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1945 INIT_LIST_HEAD(&fs_info->ordered_operations);
1946 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1947 spin_lock_init(&fs_info->delalloc_lock);
1948 spin_lock_init(&fs_info->trans_lock);
1949 spin_lock_init(&fs_info->ref_cache_lock);
1950 spin_lock_init(&fs_info->fs_roots_radix_lock);
1951 spin_lock_init(&fs_info->delayed_iput_lock);
1952 spin_lock_init(&fs_info->defrag_inodes_lock);
1953 spin_lock_init(&fs_info->free_chunk_lock);
1954 mutex_init(&fs_info->reloc_mutex);
1955
1956 init_completion(&fs_info->kobj_unregister);
1957 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1958 INIT_LIST_HEAD(&fs_info->space_info);
1959 btrfs_mapping_init(&fs_info->mapping_tree);
1960 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1961 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1962 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1963 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1964 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1965 btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1966 atomic_set(&fs_info->nr_async_submits, 0);
1967 atomic_set(&fs_info->async_delalloc_pages, 0);
1968 atomic_set(&fs_info->async_submit_draining, 0);
1969 atomic_set(&fs_info->nr_async_bios, 0);
1970 atomic_set(&fs_info->defrag_running, 0);
1971 fs_info->sb = sb;
1972 fs_info->max_inline = 8192 * 1024;
1973 fs_info->metadata_ratio = 0;
1974 fs_info->defrag_inodes = RB_ROOT;
1975 fs_info->trans_no_join = 0;
1976 fs_info->free_chunk_space = 0;
1977
1978 /* readahead state */
1979 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1980 spin_lock_init(&fs_info->reada_lock);
1981
1982 fs_info->thread_pool_size = min_t(unsigned long,
1983 num_online_cpus() + 2, 8);
1984
1985 INIT_LIST_HEAD(&fs_info->ordered_extents);
1986 spin_lock_init(&fs_info->ordered_extent_lock);
1987 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1988 GFP_NOFS);
1989 if (!fs_info->delayed_root) {
1990 err = -ENOMEM;
1991 goto fail_iput;
1992 }
1993 btrfs_init_delayed_root(fs_info->delayed_root);
1994
1995 mutex_init(&fs_info->scrub_lock);
1996 atomic_set(&fs_info->scrubs_running, 0);
1997 atomic_set(&fs_info->scrub_pause_req, 0);
1998 atomic_set(&fs_info->scrubs_paused, 0);
1999 atomic_set(&fs_info->scrub_cancel_req, 0);
2000 init_waitqueue_head(&fs_info->scrub_pause_wait);
2001 init_rwsem(&fs_info->scrub_super_lock);
2002 fs_info->scrub_workers_refcnt = 0;
2003 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2004 fs_info->check_integrity_print_mask = 0;
2005 #endif
2006
2007 spin_lock_init(&fs_info->balance_lock);
2008 mutex_init(&fs_info->balance_mutex);
2009 atomic_set(&fs_info->balance_running, 0);
2010 atomic_set(&fs_info->balance_pause_req, 0);
2011 atomic_set(&fs_info->balance_cancel_req, 0);
2012 fs_info->balance_ctl = NULL;
2013 init_waitqueue_head(&fs_info->balance_wait_q);
2014
2015 sb->s_blocksize = 4096;
2016 sb->s_blocksize_bits = blksize_bits(4096);
2017 sb->s_bdi = &fs_info->bdi;
2018
2019 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2020 set_nlink(fs_info->btree_inode, 1);
2021 /*
2022 * we set the i_size on the btree inode to the max possible int.
2023 * the real end of the address space is determined by all of
2024 * the devices in the system
2025 */
2026 fs_info->btree_inode->i_size = OFFSET_MAX;
2027 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2028 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2029
2030 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2031 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2032 fs_info->btree_inode->i_mapping);
2033 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2034
2035 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2036
2037 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2038 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2039 sizeof(struct btrfs_key));
2040 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
2041 insert_inode_hash(fs_info->btree_inode);
2042
2043 spin_lock_init(&fs_info->block_group_cache_lock);
2044 fs_info->block_group_cache_tree = RB_ROOT;
2045
2046 extent_io_tree_init(&fs_info->freed_extents[0],
2047 fs_info->btree_inode->i_mapping);
2048 extent_io_tree_init(&fs_info->freed_extents[1],
2049 fs_info->btree_inode->i_mapping);
2050 fs_info->pinned_extents = &fs_info->freed_extents[0];
2051 fs_info->do_barriers = 1;
2052
2053
2054 mutex_init(&fs_info->ordered_operations_mutex);
2055 mutex_init(&fs_info->tree_log_mutex);
2056 mutex_init(&fs_info->chunk_mutex);
2057 mutex_init(&fs_info->transaction_kthread_mutex);
2058 mutex_init(&fs_info->cleaner_mutex);
2059 mutex_init(&fs_info->volume_mutex);
2060 init_rwsem(&fs_info->extent_commit_sem);
2061 init_rwsem(&fs_info->cleanup_work_sem);
2062 init_rwsem(&fs_info->subvol_sem);
2063
2064 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2065 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2066
2067 init_waitqueue_head(&fs_info->transaction_throttle);
2068 init_waitqueue_head(&fs_info->transaction_wait);
2069 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2070 init_waitqueue_head(&fs_info->async_submit_wait);
2071
2072 __setup_root(4096, 4096, 4096, 4096, tree_root,
2073 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2074
2075 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2076 if (!bh) {
2077 err = -EINVAL;
2078 goto fail_alloc;
2079 }
2080
2081 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2082 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2083 sizeof(*fs_info->super_for_commit));
2084 brelse(bh);
2085
2086 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2087
2088 disk_super = fs_info->super_copy;
2089 if (!btrfs_super_root(disk_super))
2090 goto fail_alloc;
2091
2092 /* check FS state, whether FS is broken. */
2093 fs_info->fs_state |= btrfs_super_flags(disk_super);
2094
2095 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2096
2097 /*
2098 * run through our array of backup supers and setup
2099 * our ring pointer to the oldest one
2100 */
2101 generation = btrfs_super_generation(disk_super);
2102 find_oldest_super_backup(fs_info, generation);
2103
2104 /*
2105 * In the long term, we'll store the compression type in the super
2106 * block, and it'll be used for per file compression control.
2107 */
2108 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2109
2110 ret = btrfs_parse_options(tree_root, options);
2111 if (ret) {
2112 err = ret;
2113 goto fail_alloc;
2114 }
2115
2116 features = btrfs_super_incompat_flags(disk_super) &
2117 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2118 if (features) {
2119 printk(KERN_ERR "BTRFS: couldn't mount because of "
2120 "unsupported optional features (%Lx).\n",
2121 (unsigned long long)features);
2122 err = -EINVAL;
2123 goto fail_alloc;
2124 }
2125
2126 features = btrfs_super_incompat_flags(disk_super);
2127 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2128 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2129 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2130 btrfs_set_super_incompat_flags(disk_super, features);
2131
2132 features = btrfs_super_compat_ro_flags(disk_super) &
2133 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2134 if (!(sb->s_flags & MS_RDONLY) && features) {
2135 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2136 "unsupported option features (%Lx).\n",
2137 (unsigned long long)features);
2138 err = -EINVAL;
2139 goto fail_alloc;
2140 }
2141
2142 btrfs_init_workers(&fs_info->generic_worker,
2143 "genwork", 1, NULL);
2144
2145 btrfs_init_workers(&fs_info->workers, "worker",
2146 fs_info->thread_pool_size,
2147 &fs_info->generic_worker);
2148
2149 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2150 fs_info->thread_pool_size,
2151 &fs_info->generic_worker);
2152
2153 btrfs_init_workers(&fs_info->submit_workers, "submit",
2154 min_t(u64, fs_devices->num_devices,
2155 fs_info->thread_pool_size),
2156 &fs_info->generic_worker);
2157
2158 btrfs_init_workers(&fs_info->caching_workers, "cache",
2159 2, &fs_info->generic_worker);
2160
2161 /* a higher idle thresh on the submit workers makes it much more
2162 * likely that bios will be send down in a sane order to the
2163 * devices
2164 */
2165 fs_info->submit_workers.idle_thresh = 64;
2166
2167 fs_info->workers.idle_thresh = 16;
2168 fs_info->workers.ordered = 1;
2169
2170 fs_info->delalloc_workers.idle_thresh = 2;
2171 fs_info->delalloc_workers.ordered = 1;
2172
2173 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2174 &fs_info->generic_worker);
2175 btrfs_init_workers(&fs_info->endio_workers, "endio",
2176 fs_info->thread_pool_size,
2177 &fs_info->generic_worker);
2178 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2179 fs_info->thread_pool_size,
2180 &fs_info->generic_worker);
2181 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2182 "endio-meta-write", fs_info->thread_pool_size,
2183 &fs_info->generic_worker);
2184 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2185 fs_info->thread_pool_size,
2186 &fs_info->generic_worker);
2187 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2188 1, &fs_info->generic_worker);
2189 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2190 fs_info->thread_pool_size,
2191 &fs_info->generic_worker);
2192 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2193 fs_info->thread_pool_size,
2194 &fs_info->generic_worker);
2195
2196 /*
2197 * endios are largely parallel and should have a very
2198 * low idle thresh
2199 */
2200 fs_info->endio_workers.idle_thresh = 4;
2201 fs_info->endio_meta_workers.idle_thresh = 4;
2202
2203 fs_info->endio_write_workers.idle_thresh = 2;
2204 fs_info->endio_meta_write_workers.idle_thresh = 2;
2205 fs_info->readahead_workers.idle_thresh = 2;
2206
2207 /*
2208 * btrfs_start_workers can really only fail because of ENOMEM so just
2209 * return -ENOMEM if any of these fail.
2210 */
2211 ret = btrfs_start_workers(&fs_info->workers);
2212 ret |= btrfs_start_workers(&fs_info->generic_worker);
2213 ret |= btrfs_start_workers(&fs_info->submit_workers);
2214 ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2215 ret |= btrfs_start_workers(&fs_info->fixup_workers);
2216 ret |= btrfs_start_workers(&fs_info->endio_workers);
2217 ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2218 ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2219 ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2220 ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2221 ret |= btrfs_start_workers(&fs_info->delayed_workers);
2222 ret |= btrfs_start_workers(&fs_info->caching_workers);
2223 ret |= btrfs_start_workers(&fs_info->readahead_workers);
2224 if (ret) {
2225 ret = -ENOMEM;
2226 goto fail_sb_buffer;
2227 }
2228
2229 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2230 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2231 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2232
2233 nodesize = btrfs_super_nodesize(disk_super);
2234 leafsize = btrfs_super_leafsize(disk_super);
2235 sectorsize = btrfs_super_sectorsize(disk_super);
2236 stripesize = btrfs_super_stripesize(disk_super);
2237 tree_root->nodesize = nodesize;
2238 tree_root->leafsize = leafsize;
2239 tree_root->sectorsize = sectorsize;
2240 tree_root->stripesize = stripesize;
2241
2242 sb->s_blocksize = sectorsize;
2243 sb->s_blocksize_bits = blksize_bits(sectorsize);
2244
2245 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2246 sizeof(disk_super->magic))) {
2247 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2248 goto fail_sb_buffer;
2249 }
2250
2251 mutex_lock(&fs_info->chunk_mutex);
2252 ret = btrfs_read_sys_array(tree_root);
2253 mutex_unlock(&fs_info->chunk_mutex);
2254 if (ret) {
2255 printk(KERN_WARNING "btrfs: failed to read the system "
2256 "array on %s\n", sb->s_id);
2257 goto fail_sb_buffer;
2258 }
2259
2260 blocksize = btrfs_level_size(tree_root,
2261 btrfs_super_chunk_root_level(disk_super));
2262 generation = btrfs_super_chunk_root_generation(disk_super);
2263
2264 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2265 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2266
2267 chunk_root->node = read_tree_block(chunk_root,
2268 btrfs_super_chunk_root(disk_super),
2269 blocksize, generation);
2270 BUG_ON(!chunk_root->node);
2271 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2272 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2273 sb->s_id);
2274 goto fail_tree_roots;
2275 }
2276 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2277 chunk_root->commit_root = btrfs_root_node(chunk_root);
2278
2279 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2280 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2281 BTRFS_UUID_SIZE);
2282
2283 ret = btrfs_read_chunk_tree(chunk_root);
2284 if (ret) {
2285 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2286 sb->s_id);
2287 goto fail_tree_roots;
2288 }
2289
2290 btrfs_close_extra_devices(fs_devices);
2291
2292 retry_root_backup:
2293 blocksize = btrfs_level_size(tree_root,
2294 btrfs_super_root_level(disk_super));
2295 generation = btrfs_super_generation(disk_super);
2296
2297 tree_root->node = read_tree_block(tree_root,
2298 btrfs_super_root(disk_super),
2299 blocksize, generation);
2300 if (!tree_root->node ||
2301 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2302 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2303 sb->s_id);
2304
2305 goto recovery_tree_root;
2306 }
2307
2308 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2309 tree_root->commit_root = btrfs_root_node(tree_root);
2310
2311 ret = find_and_setup_root(tree_root, fs_info,
2312 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2313 if (ret)
2314 goto recovery_tree_root;
2315 extent_root->track_dirty = 1;
2316
2317 ret = find_and_setup_root(tree_root, fs_info,
2318 BTRFS_DEV_TREE_OBJECTID, dev_root);
2319 if (ret)
2320 goto recovery_tree_root;
2321 dev_root->track_dirty = 1;
2322
2323 ret = find_and_setup_root(tree_root, fs_info,
2324 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2325 if (ret)
2326 goto recovery_tree_root;
2327
2328 csum_root->track_dirty = 1;
2329
2330 fs_info->generation = generation;
2331 fs_info->last_trans_committed = generation;
2332
2333 ret = btrfs_init_space_info(fs_info);
2334 if (ret) {
2335 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2336 goto fail_block_groups;
2337 }
2338
2339 ret = btrfs_read_block_groups(extent_root);
2340 if (ret) {
2341 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2342 goto fail_block_groups;
2343 }
2344
2345 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2346 "btrfs-cleaner");
2347 if (IS_ERR(fs_info->cleaner_kthread))
2348 goto fail_block_groups;
2349
2350 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2351 tree_root,
2352 "btrfs-transaction");
2353 if (IS_ERR(fs_info->transaction_kthread))
2354 goto fail_cleaner;
2355
2356 if (!btrfs_test_opt(tree_root, SSD) &&
2357 !btrfs_test_opt(tree_root, NOSSD) &&
2358 !fs_info->fs_devices->rotating) {
2359 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2360 "mode\n");
2361 btrfs_set_opt(fs_info->mount_opt, SSD);
2362 }
2363
2364 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2365 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2366 ret = btrfsic_mount(tree_root, fs_devices,
2367 btrfs_test_opt(tree_root,
2368 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2369 1 : 0,
2370 fs_info->check_integrity_print_mask);
2371 if (ret)
2372 printk(KERN_WARNING "btrfs: failed to initialize"
2373 " integrity check module %s\n", sb->s_id);
2374 }
2375 #endif
2376
2377 /* do not make disk changes in broken FS */
2378 if (btrfs_super_log_root(disk_super) != 0 &&
2379 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2380 u64 bytenr = btrfs_super_log_root(disk_super);
2381
2382 if (fs_devices->rw_devices == 0) {
2383 printk(KERN_WARNING "Btrfs log replay required "
2384 "on RO media\n");
2385 err = -EIO;
2386 goto fail_trans_kthread;
2387 }
2388 blocksize =
2389 btrfs_level_size(tree_root,
2390 btrfs_super_log_root_level(disk_super));
2391
2392 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2393 if (!log_tree_root) {
2394 err = -ENOMEM;
2395 goto fail_trans_kthread;
2396 }
2397
2398 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2399 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2400
2401 log_tree_root->node = read_tree_block(tree_root, bytenr,
2402 blocksize,
2403 generation + 1);
2404 ret = btrfs_recover_log_trees(log_tree_root);
2405 BUG_ON(ret);
2406
2407 if (sb->s_flags & MS_RDONLY) {
2408 ret = btrfs_commit_super(tree_root);
2409 BUG_ON(ret);
2410 }
2411 }
2412
2413 ret = btrfs_find_orphan_roots(tree_root);
2414 BUG_ON(ret);
2415
2416 if (!(sb->s_flags & MS_RDONLY)) {
2417 ret = btrfs_cleanup_fs_roots(fs_info);
2418 BUG_ON(ret);
2419
2420 ret = btrfs_recover_relocation(tree_root);
2421 if (ret < 0) {
2422 printk(KERN_WARNING
2423 "btrfs: failed to recover relocation\n");
2424 err = -EINVAL;
2425 goto fail_trans_kthread;
2426 }
2427 }
2428
2429 location.objectid = BTRFS_FS_TREE_OBJECTID;
2430 location.type = BTRFS_ROOT_ITEM_KEY;
2431 location.offset = (u64)-1;
2432
2433 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2434 if (!fs_info->fs_root)
2435 goto fail_trans_kthread;
2436 if (IS_ERR(fs_info->fs_root)) {
2437 err = PTR_ERR(fs_info->fs_root);
2438 goto fail_trans_kthread;
2439 }
2440
2441 if (!(sb->s_flags & MS_RDONLY)) {
2442 down_read(&fs_info->cleanup_work_sem);
2443 err = btrfs_orphan_cleanup(fs_info->fs_root);
2444 if (!err)
2445 err = btrfs_orphan_cleanup(fs_info->tree_root);
2446 up_read(&fs_info->cleanup_work_sem);
2447
2448 if (!err)
2449 err = btrfs_recover_balance(fs_info->tree_root);
2450
2451 if (err) {
2452 close_ctree(tree_root);
2453 return ERR_PTR(err);
2454 }
2455 }
2456
2457 return tree_root;
2458
2459 fail_trans_kthread:
2460 kthread_stop(fs_info->transaction_kthread);
2461 fail_cleaner:
2462 kthread_stop(fs_info->cleaner_kthread);
2463
2464 /*
2465 * make sure we're done with the btree inode before we stop our
2466 * kthreads
2467 */
2468 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2469 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2470
2471 fail_block_groups:
2472 btrfs_free_block_groups(fs_info);
2473
2474 fail_tree_roots:
2475 free_root_pointers(fs_info, 1);
2476
2477 fail_sb_buffer:
2478 btrfs_stop_workers(&fs_info->generic_worker);
2479 btrfs_stop_workers(&fs_info->readahead_workers);
2480 btrfs_stop_workers(&fs_info->fixup_workers);
2481 btrfs_stop_workers(&fs_info->delalloc_workers);
2482 btrfs_stop_workers(&fs_info->workers);
2483 btrfs_stop_workers(&fs_info->endio_workers);
2484 btrfs_stop_workers(&fs_info->endio_meta_workers);
2485 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2486 btrfs_stop_workers(&fs_info->endio_write_workers);
2487 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2488 btrfs_stop_workers(&fs_info->submit_workers);
2489 btrfs_stop_workers(&fs_info->delayed_workers);
2490 btrfs_stop_workers(&fs_info->caching_workers);
2491 fail_alloc:
2492 fail_iput:
2493 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2494
2495 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2496 iput(fs_info->btree_inode);
2497 fail_bdi:
2498 bdi_destroy(&fs_info->bdi);
2499 fail_srcu:
2500 cleanup_srcu_struct(&fs_info->subvol_srcu);
2501 fail:
2502 btrfs_close_devices(fs_info->fs_devices);
2503 free_fs_info(fs_info);
2504 return ERR_PTR(err);
2505
2506 recovery_tree_root:
2507 if (!btrfs_test_opt(tree_root, RECOVERY))
2508 goto fail_tree_roots;
2509
2510 free_root_pointers(fs_info, 0);
2511
2512 /* don't use the log in recovery mode, it won't be valid */
2513 btrfs_set_super_log_root(disk_super, 0);
2514
2515 /* we can't trust the free space cache either */
2516 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2517
2518 ret = next_root_backup(fs_info, fs_info->super_copy,
2519 &num_backups_tried, &backup_index);
2520 if (ret == -1)
2521 goto fail_block_groups;
2522 goto retry_root_backup;
2523 }
2524
2525 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2526 {
2527 char b[BDEVNAME_SIZE];
2528
2529 if (uptodate) {
2530 set_buffer_uptodate(bh);
2531 } else {
2532 printk_ratelimited(KERN_WARNING "lost page write due to "
2533 "I/O error on %s\n",
2534 bdevname(bh->b_bdev, b));
2535 /* note, we dont' set_buffer_write_io_error because we have
2536 * our own ways of dealing with the IO errors
2537 */
2538 clear_buffer_uptodate(bh);
2539 }
2540 unlock_buffer(bh);
2541 put_bh(bh);
2542 }
2543
2544 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2545 {
2546 struct buffer_head *bh;
2547 struct buffer_head *latest = NULL;
2548 struct btrfs_super_block *super;
2549 int i;
2550 u64 transid = 0;
2551 u64 bytenr;
2552
2553 /* we would like to check all the supers, but that would make
2554 * a btrfs mount succeed after a mkfs from a different FS.
2555 * So, we need to add a special mount option to scan for
2556 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2557 */
2558 for (i = 0; i < 1; i++) {
2559 bytenr = btrfs_sb_offset(i);
2560 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2561 break;
2562 bh = __bread(bdev, bytenr / 4096, 4096);
2563 if (!bh)
2564 continue;
2565
2566 super = (struct btrfs_super_block *)bh->b_data;
2567 if (btrfs_super_bytenr(super) != bytenr ||
2568 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2569 sizeof(super->magic))) {
2570 brelse(bh);
2571 continue;
2572 }
2573
2574 if (!latest || btrfs_super_generation(super) > transid) {
2575 brelse(latest);
2576 latest = bh;
2577 transid = btrfs_super_generation(super);
2578 } else {
2579 brelse(bh);
2580 }
2581 }
2582 return latest;
2583 }
2584
2585 /*
2586 * this should be called twice, once with wait == 0 and
2587 * once with wait == 1. When wait == 0 is done, all the buffer heads
2588 * we write are pinned.
2589 *
2590 * They are released when wait == 1 is done.
2591 * max_mirrors must be the same for both runs, and it indicates how
2592 * many supers on this one device should be written.
2593 *
2594 * max_mirrors == 0 means to write them all.
2595 */
2596 static int write_dev_supers(struct btrfs_device *device,
2597 struct btrfs_super_block *sb,
2598 int do_barriers, int wait, int max_mirrors)
2599 {
2600 struct buffer_head *bh;
2601 int i;
2602 int ret;
2603 int errors = 0;
2604 u32 crc;
2605 u64 bytenr;
2606
2607 if (max_mirrors == 0)
2608 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2609
2610 for (i = 0; i < max_mirrors; i++) {
2611 bytenr = btrfs_sb_offset(i);
2612 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2613 break;
2614
2615 if (wait) {
2616 bh = __find_get_block(device->bdev, bytenr / 4096,
2617 BTRFS_SUPER_INFO_SIZE);
2618 BUG_ON(!bh);
2619 wait_on_buffer(bh);
2620 if (!buffer_uptodate(bh))
2621 errors++;
2622
2623 /* drop our reference */
2624 brelse(bh);
2625
2626 /* drop the reference from the wait == 0 run */
2627 brelse(bh);
2628 continue;
2629 } else {
2630 btrfs_set_super_bytenr(sb, bytenr);
2631
2632 crc = ~(u32)0;
2633 crc = btrfs_csum_data(NULL, (char *)sb +
2634 BTRFS_CSUM_SIZE, crc,
2635 BTRFS_SUPER_INFO_SIZE -
2636 BTRFS_CSUM_SIZE);
2637 btrfs_csum_final(crc, sb->csum);
2638
2639 /*
2640 * one reference for us, and we leave it for the
2641 * caller
2642 */
2643 bh = __getblk(device->bdev, bytenr / 4096,
2644 BTRFS_SUPER_INFO_SIZE);
2645 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2646
2647 /* one reference for submit_bh */
2648 get_bh(bh);
2649
2650 set_buffer_uptodate(bh);
2651 lock_buffer(bh);
2652 bh->b_end_io = btrfs_end_buffer_write_sync;
2653 }
2654
2655 /*
2656 * we fua the first super. The others we allow
2657 * to go down lazy.
2658 */
2659 ret = btrfsic_submit_bh(WRITE_FUA, bh);
2660 if (ret)
2661 errors++;
2662 }
2663 return errors < i ? 0 : -1;
2664 }
2665
2666 /*
2667 * endio for the write_dev_flush, this will wake anyone waiting
2668 * for the barrier when it is done
2669 */
2670 static void btrfs_end_empty_barrier(struct bio *bio, int err)
2671 {
2672 if (err) {
2673 if (err == -EOPNOTSUPP)
2674 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2675 clear_bit(BIO_UPTODATE, &bio->bi_flags);
2676 }
2677 if (bio->bi_private)
2678 complete(bio->bi_private);
2679 bio_put(bio);
2680 }
2681
2682 /*
2683 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
2684 * sent down. With wait == 1, it waits for the previous flush.
2685 *
2686 * any device where the flush fails with eopnotsupp are flagged as not-barrier
2687 * capable
2688 */
2689 static int write_dev_flush(struct btrfs_device *device, int wait)
2690 {
2691 struct bio *bio;
2692 int ret = 0;
2693
2694 if (device->nobarriers)
2695 return 0;
2696
2697 if (wait) {
2698 bio = device->flush_bio;
2699 if (!bio)
2700 return 0;
2701
2702 wait_for_completion(&device->flush_wait);
2703
2704 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
2705 printk("btrfs: disabling barriers on dev %s\n",
2706 device->name);
2707 device->nobarriers = 1;
2708 }
2709 if (!bio_flagged(bio, BIO_UPTODATE)) {
2710 ret = -EIO;
2711 }
2712
2713 /* drop the reference from the wait == 0 run */
2714 bio_put(bio);
2715 device->flush_bio = NULL;
2716
2717 return ret;
2718 }
2719
2720 /*
2721 * one reference for us, and we leave it for the
2722 * caller
2723 */
2724 device->flush_bio = NULL;;
2725 bio = bio_alloc(GFP_NOFS, 0);
2726 if (!bio)
2727 return -ENOMEM;
2728
2729 bio->bi_end_io = btrfs_end_empty_barrier;
2730 bio->bi_bdev = device->bdev;
2731 init_completion(&device->flush_wait);
2732 bio->bi_private = &device->flush_wait;
2733 device->flush_bio = bio;
2734
2735 bio_get(bio);
2736 btrfsic_submit_bio(WRITE_FLUSH, bio);
2737
2738 return 0;
2739 }
2740
2741 /*
2742 * send an empty flush down to each device in parallel,
2743 * then wait for them
2744 */
2745 static int barrier_all_devices(struct btrfs_fs_info *info)
2746 {
2747 struct list_head *head;
2748 struct btrfs_device *dev;
2749 int errors = 0;
2750 int ret;
2751
2752 /* send down all the barriers */
2753 head = &info->fs_devices->devices;
2754 list_for_each_entry_rcu(dev, head, dev_list) {
2755 if (!dev->bdev) {
2756 errors++;
2757 continue;
2758 }
2759 if (!dev->in_fs_metadata || !dev->writeable)
2760 continue;
2761
2762 ret = write_dev_flush(dev, 0);
2763 if (ret)
2764 errors++;
2765 }
2766
2767 /* wait for all the barriers */
2768 list_for_each_entry_rcu(dev, head, dev_list) {
2769 if (!dev->bdev) {
2770 errors++;
2771 continue;
2772 }
2773 if (!dev->in_fs_metadata || !dev->writeable)
2774 continue;
2775
2776 ret = write_dev_flush(dev, 1);
2777 if (ret)
2778 errors++;
2779 }
2780 if (errors)
2781 return -EIO;
2782 return 0;
2783 }
2784
2785 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2786 {
2787 struct list_head *head;
2788 struct btrfs_device *dev;
2789 struct btrfs_super_block *sb;
2790 struct btrfs_dev_item *dev_item;
2791 int ret;
2792 int do_barriers;
2793 int max_errors;
2794 int total_errors = 0;
2795 u64 flags;
2796
2797 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2798 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2799 backup_super_roots(root->fs_info);
2800
2801 sb = root->fs_info->super_for_commit;
2802 dev_item = &sb->dev_item;
2803
2804 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2805 head = &root->fs_info->fs_devices->devices;
2806
2807 if (do_barriers)
2808 barrier_all_devices(root->fs_info);
2809
2810 list_for_each_entry_rcu(dev, head, dev_list) {
2811 if (!dev->bdev) {
2812 total_errors++;
2813 continue;
2814 }
2815 if (!dev->in_fs_metadata || !dev->writeable)
2816 continue;
2817
2818 btrfs_set_stack_device_generation(dev_item, 0);
2819 btrfs_set_stack_device_type(dev_item, dev->type);
2820 btrfs_set_stack_device_id(dev_item, dev->devid);
2821 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2822 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2823 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2824 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2825 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2826 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2827 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2828
2829 flags = btrfs_super_flags(sb);
2830 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2831
2832 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2833 if (ret)
2834 total_errors++;
2835 }
2836 if (total_errors > max_errors) {
2837 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2838 total_errors);
2839 BUG();
2840 }
2841
2842 total_errors = 0;
2843 list_for_each_entry_rcu(dev, head, dev_list) {
2844 if (!dev->bdev)
2845 continue;
2846 if (!dev->in_fs_metadata || !dev->writeable)
2847 continue;
2848
2849 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2850 if (ret)
2851 total_errors++;
2852 }
2853 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2854 if (total_errors > max_errors) {
2855 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2856 total_errors);
2857 BUG();
2858 }
2859 return 0;
2860 }
2861
2862 int write_ctree_super(struct btrfs_trans_handle *trans,
2863 struct btrfs_root *root, int max_mirrors)
2864 {
2865 int ret;
2866
2867 ret = write_all_supers(root, max_mirrors);
2868 return ret;
2869 }
2870
2871 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2872 {
2873 spin_lock(&fs_info->fs_roots_radix_lock);
2874 radix_tree_delete(&fs_info->fs_roots_radix,
2875 (unsigned long)root->root_key.objectid);
2876 spin_unlock(&fs_info->fs_roots_radix_lock);
2877
2878 if (btrfs_root_refs(&root->root_item) == 0)
2879 synchronize_srcu(&fs_info->subvol_srcu);
2880
2881 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2882 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2883 free_fs_root(root);
2884 return 0;
2885 }
2886
2887 static void free_fs_root(struct btrfs_root *root)
2888 {
2889 iput(root->cache_inode);
2890 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2891 if (root->anon_dev)
2892 free_anon_bdev(root->anon_dev);
2893 free_extent_buffer(root->node);
2894 free_extent_buffer(root->commit_root);
2895 kfree(root->free_ino_ctl);
2896 kfree(root->free_ino_pinned);
2897 kfree(root->name);
2898 kfree(root);
2899 }
2900
2901 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2902 {
2903 int ret;
2904 struct btrfs_root *gang[8];
2905 int i;
2906
2907 while (!list_empty(&fs_info->dead_roots)) {
2908 gang[0] = list_entry(fs_info->dead_roots.next,
2909 struct btrfs_root, root_list);
2910 list_del(&gang[0]->root_list);
2911
2912 if (gang[0]->in_radix) {
2913 btrfs_free_fs_root(fs_info, gang[0]);
2914 } else {
2915 free_extent_buffer(gang[0]->node);
2916 free_extent_buffer(gang[0]->commit_root);
2917 kfree(gang[0]);
2918 }
2919 }
2920
2921 while (1) {
2922 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2923 (void **)gang, 0,
2924 ARRAY_SIZE(gang));
2925 if (!ret)
2926 break;
2927 for (i = 0; i < ret; i++)
2928 btrfs_free_fs_root(fs_info, gang[i]);
2929 }
2930 return 0;
2931 }
2932
2933 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2934 {
2935 u64 root_objectid = 0;
2936 struct btrfs_root *gang[8];
2937 int i;
2938 int ret;
2939
2940 while (1) {
2941 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2942 (void **)gang, root_objectid,
2943 ARRAY_SIZE(gang));
2944 if (!ret)
2945 break;
2946
2947 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2948 for (i = 0; i < ret; i++) {
2949 int err;
2950
2951 root_objectid = gang[i]->root_key.objectid;
2952 err = btrfs_orphan_cleanup(gang[i]);
2953 if (err)
2954 return err;
2955 }
2956 root_objectid++;
2957 }
2958 return 0;
2959 }
2960
2961 int btrfs_commit_super(struct btrfs_root *root)
2962 {
2963 struct btrfs_trans_handle *trans;
2964 int ret;
2965
2966 mutex_lock(&root->fs_info->cleaner_mutex);
2967 btrfs_run_delayed_iputs(root);
2968 btrfs_clean_old_snapshots(root);
2969 mutex_unlock(&root->fs_info->cleaner_mutex);
2970
2971 /* wait until ongoing cleanup work done */
2972 down_write(&root->fs_info->cleanup_work_sem);
2973 up_write(&root->fs_info->cleanup_work_sem);
2974
2975 trans = btrfs_join_transaction(root);
2976 if (IS_ERR(trans))
2977 return PTR_ERR(trans);
2978 ret = btrfs_commit_transaction(trans, root);
2979 BUG_ON(ret);
2980 /* run commit again to drop the original snapshot */
2981 trans = btrfs_join_transaction(root);
2982 if (IS_ERR(trans))
2983 return PTR_ERR(trans);
2984 btrfs_commit_transaction(trans, root);
2985 ret = btrfs_write_and_wait_transaction(NULL, root);
2986 BUG_ON(ret);
2987
2988 ret = write_ctree_super(NULL, root, 0);
2989 return ret;
2990 }
2991
2992 int close_ctree(struct btrfs_root *root)
2993 {
2994 struct btrfs_fs_info *fs_info = root->fs_info;
2995 int ret;
2996
2997 fs_info->closing = 1;
2998 smp_mb();
2999
3000 /* pause restriper - we want to resume on mount */
3001 btrfs_pause_balance(root->fs_info);
3002
3003 btrfs_scrub_cancel(root);
3004
3005 /* wait for any defraggers to finish */
3006 wait_event(fs_info->transaction_wait,
3007 (atomic_read(&fs_info->defrag_running) == 0));
3008
3009 /* clear out the rbtree of defraggable inodes */
3010 btrfs_run_defrag_inodes(root->fs_info);
3011
3012 /*
3013 * Here come 2 situations when btrfs is broken to flip readonly:
3014 *
3015 * 1. when btrfs flips readonly somewhere else before
3016 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
3017 * and btrfs will skip to write sb directly to keep
3018 * ERROR state on disk.
3019 *
3020 * 2. when btrfs flips readonly just in btrfs_commit_super,
3021 * and in such case, btrfs cannot write sb via btrfs_commit_super,
3022 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
3023 * btrfs will cleanup all FS resources first and write sb then.
3024 */
3025 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3026 ret = btrfs_commit_super(root);
3027 if (ret)
3028 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3029 }
3030
3031 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
3032 ret = btrfs_error_commit_super(root);
3033 if (ret)
3034 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3035 }
3036
3037 btrfs_put_block_group_cache(fs_info);
3038
3039 kthread_stop(root->fs_info->transaction_kthread);
3040 kthread_stop(root->fs_info->cleaner_kthread);
3041
3042 fs_info->closing = 2;
3043 smp_mb();
3044
3045 if (fs_info->delalloc_bytes) {
3046 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
3047 (unsigned long long)fs_info->delalloc_bytes);
3048 }
3049 if (fs_info->total_ref_cache_size) {
3050 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
3051 (unsigned long long)fs_info->total_ref_cache_size);
3052 }
3053
3054 free_extent_buffer(fs_info->extent_root->node);
3055 free_extent_buffer(fs_info->extent_root->commit_root);
3056 free_extent_buffer(fs_info->tree_root->node);
3057 free_extent_buffer(fs_info->tree_root->commit_root);
3058 free_extent_buffer(root->fs_info->chunk_root->node);
3059 free_extent_buffer(root->fs_info->chunk_root->commit_root);
3060 free_extent_buffer(root->fs_info->dev_root->node);
3061 free_extent_buffer(root->fs_info->dev_root->commit_root);
3062 free_extent_buffer(root->fs_info->csum_root->node);
3063 free_extent_buffer(root->fs_info->csum_root->commit_root);
3064
3065 btrfs_free_block_groups(root->fs_info);
3066
3067 del_fs_roots(fs_info);
3068
3069 iput(fs_info->btree_inode);
3070
3071 btrfs_stop_workers(&fs_info->generic_worker);
3072 btrfs_stop_workers(&fs_info->fixup_workers);
3073 btrfs_stop_workers(&fs_info->delalloc_workers);
3074 btrfs_stop_workers(&fs_info->workers);
3075 btrfs_stop_workers(&fs_info->endio_workers);
3076 btrfs_stop_workers(&fs_info->endio_meta_workers);
3077 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
3078 btrfs_stop_workers(&fs_info->endio_write_workers);
3079 btrfs_stop_workers(&fs_info->endio_freespace_worker);
3080 btrfs_stop_workers(&fs_info->submit_workers);
3081 btrfs_stop_workers(&fs_info->delayed_workers);
3082 btrfs_stop_workers(&fs_info->caching_workers);
3083 btrfs_stop_workers(&fs_info->readahead_workers);
3084
3085 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3086 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3087 btrfsic_unmount(root, fs_info->fs_devices);
3088 #endif
3089
3090 btrfs_close_devices(fs_info->fs_devices);
3091 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3092
3093 bdi_destroy(&fs_info->bdi);
3094 cleanup_srcu_struct(&fs_info->subvol_srcu);
3095
3096 free_fs_info(fs_info);
3097
3098 return 0;
3099 }
3100
3101 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
3102 {
3103 int ret;
3104 struct inode *btree_inode = buf->first_page->mapping->host;
3105
3106 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
3107 NULL);
3108 if (!ret)
3109 return ret;
3110
3111 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3112 parent_transid);
3113 return !ret;
3114 }
3115
3116 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3117 {
3118 struct inode *btree_inode = buf->first_page->mapping->host;
3119 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
3120 buf);
3121 }
3122
3123 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3124 {
3125 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3126 u64 transid = btrfs_header_generation(buf);
3127 struct inode *btree_inode = root->fs_info->btree_inode;
3128 int was_dirty;
3129
3130 btrfs_assert_tree_locked(buf);
3131 if (transid != root->fs_info->generation) {
3132 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
3133 "found %llu running %llu\n",
3134 (unsigned long long)buf->start,
3135 (unsigned long long)transid,
3136 (unsigned long long)root->fs_info->generation);
3137 WARN_ON(1);
3138 }
3139 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
3140 buf);
3141 if (!was_dirty) {
3142 spin_lock(&root->fs_info->delalloc_lock);
3143 root->fs_info->dirty_metadata_bytes += buf->len;
3144 spin_unlock(&root->fs_info->delalloc_lock);
3145 }
3146 }
3147
3148 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3149 {
3150 /*
3151 * looks as though older kernels can get into trouble with
3152 * this code, they end up stuck in balance_dirty_pages forever
3153 */
3154 u64 num_dirty;
3155 unsigned long thresh = 32 * 1024 * 1024;
3156
3157 if (current->flags & PF_MEMALLOC)
3158 return;
3159
3160 btrfs_balance_delayed_items(root);
3161
3162 num_dirty = root->fs_info->dirty_metadata_bytes;
3163
3164 if (num_dirty > thresh) {
3165 balance_dirty_pages_ratelimited_nr(
3166 root->fs_info->btree_inode->i_mapping, 1);
3167 }
3168 return;
3169 }
3170
3171 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3172 {
3173 /*
3174 * looks as though older kernels can get into trouble with
3175 * this code, they end up stuck in balance_dirty_pages forever
3176 */
3177 u64 num_dirty;
3178 unsigned long thresh = 32 * 1024 * 1024;
3179
3180 if (current->flags & PF_MEMALLOC)
3181 return;
3182
3183 num_dirty = root->fs_info->dirty_metadata_bytes;
3184
3185 if (num_dirty > thresh) {
3186 balance_dirty_pages_ratelimited_nr(
3187 root->fs_info->btree_inode->i_mapping, 1);
3188 }
3189 return;
3190 }
3191
3192 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3193 {
3194 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3195 int ret;
3196 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3197 if (ret == 0)
3198 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
3199 return ret;
3200 }
3201
3202 static int btree_lock_page_hook(struct page *page, void *data,
3203 void (*flush_fn)(void *))
3204 {
3205 struct inode *inode = page->mapping->host;
3206 struct btrfs_root *root = BTRFS_I(inode)->root;
3207 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3208 struct extent_buffer *eb;
3209 unsigned long len;
3210 u64 bytenr = page_offset(page);
3211
3212 if (page->private == EXTENT_PAGE_PRIVATE)
3213 goto out;
3214
3215 len = page->private >> 2;
3216 eb = find_extent_buffer(io_tree, bytenr, len);
3217 if (!eb)
3218 goto out;
3219
3220 if (!btrfs_try_tree_write_lock(eb)) {
3221 flush_fn(data);
3222 btrfs_tree_lock(eb);
3223 }
3224 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3225
3226 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3227 spin_lock(&root->fs_info->delalloc_lock);
3228 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3229 root->fs_info->dirty_metadata_bytes -= eb->len;
3230 else
3231 WARN_ON(1);
3232 spin_unlock(&root->fs_info->delalloc_lock);
3233 }
3234
3235 btrfs_tree_unlock(eb);
3236 free_extent_buffer(eb);
3237 out:
3238 if (!trylock_page(page)) {
3239 flush_fn(data);
3240 lock_page(page);
3241 }
3242 return 0;
3243 }
3244
3245 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3246 int read_only)
3247 {
3248 if (read_only)
3249 return;
3250
3251 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3252 printk(KERN_WARNING "warning: mount fs with errors, "
3253 "running btrfsck is recommended\n");
3254 }
3255
3256 int btrfs_error_commit_super(struct btrfs_root *root)
3257 {
3258 int ret;
3259
3260 mutex_lock(&root->fs_info->cleaner_mutex);
3261 btrfs_run_delayed_iputs(root);
3262 mutex_unlock(&root->fs_info->cleaner_mutex);
3263
3264 down_write(&root->fs_info->cleanup_work_sem);
3265 up_write(&root->fs_info->cleanup_work_sem);
3266
3267 /* cleanup FS via transaction */
3268 btrfs_cleanup_transaction(root);
3269
3270 ret = write_ctree_super(NULL, root, 0);
3271
3272 return ret;
3273 }
3274
3275 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
3276 {
3277 struct btrfs_inode *btrfs_inode;
3278 struct list_head splice;
3279
3280 INIT_LIST_HEAD(&splice);
3281
3282 mutex_lock(&root->fs_info->ordered_operations_mutex);
3283 spin_lock(&root->fs_info->ordered_extent_lock);
3284
3285 list_splice_init(&root->fs_info->ordered_operations, &splice);
3286 while (!list_empty(&splice)) {
3287 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3288 ordered_operations);
3289
3290 list_del_init(&btrfs_inode->ordered_operations);
3291
3292 btrfs_invalidate_inodes(btrfs_inode->root);
3293 }
3294
3295 spin_unlock(&root->fs_info->ordered_extent_lock);
3296 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3297
3298 return 0;
3299 }
3300
3301 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
3302 {
3303 struct list_head splice;
3304 struct btrfs_ordered_extent *ordered;
3305 struct inode *inode;
3306
3307 INIT_LIST_HEAD(&splice);
3308
3309 spin_lock(&root->fs_info->ordered_extent_lock);
3310
3311 list_splice_init(&root->fs_info->ordered_extents, &splice);
3312 while (!list_empty(&splice)) {
3313 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3314 root_extent_list);
3315
3316 list_del_init(&ordered->root_extent_list);
3317 atomic_inc(&ordered->refs);
3318
3319 /* the inode may be getting freed (in sys_unlink path). */
3320 inode = igrab(ordered->inode);
3321
3322 spin_unlock(&root->fs_info->ordered_extent_lock);
3323 if (inode)
3324 iput(inode);
3325
3326 atomic_set(&ordered->refs, 1);
3327 btrfs_put_ordered_extent(ordered);
3328
3329 spin_lock(&root->fs_info->ordered_extent_lock);
3330 }
3331
3332 spin_unlock(&root->fs_info->ordered_extent_lock);
3333
3334 return 0;
3335 }
3336
3337 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3338 struct btrfs_root *root)
3339 {
3340 struct rb_node *node;
3341 struct btrfs_delayed_ref_root *delayed_refs;
3342 struct btrfs_delayed_ref_node *ref;
3343 int ret = 0;
3344
3345 delayed_refs = &trans->delayed_refs;
3346
3347 spin_lock(&delayed_refs->lock);
3348 if (delayed_refs->num_entries == 0) {
3349 spin_unlock(&delayed_refs->lock);
3350 printk(KERN_INFO "delayed_refs has NO entry\n");
3351 return ret;
3352 }
3353
3354 node = rb_first(&delayed_refs->root);
3355 while (node) {
3356 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3357 node = rb_next(node);
3358
3359 ref->in_tree = 0;
3360 rb_erase(&ref->rb_node, &delayed_refs->root);
3361 delayed_refs->num_entries--;
3362
3363 atomic_set(&ref->refs, 1);
3364 if (btrfs_delayed_ref_is_head(ref)) {
3365 struct btrfs_delayed_ref_head *head;
3366
3367 head = btrfs_delayed_node_to_head(ref);
3368 mutex_lock(&head->mutex);
3369 kfree(head->extent_op);
3370 delayed_refs->num_heads--;
3371 if (list_empty(&head->cluster))
3372 delayed_refs->num_heads_ready--;
3373 list_del_init(&head->cluster);
3374 mutex_unlock(&head->mutex);
3375 }
3376
3377 spin_unlock(&delayed_refs->lock);
3378 btrfs_put_delayed_ref(ref);
3379
3380 cond_resched();
3381 spin_lock(&delayed_refs->lock);
3382 }
3383
3384 spin_unlock(&delayed_refs->lock);
3385
3386 return ret;
3387 }
3388
3389 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3390 {
3391 struct btrfs_pending_snapshot *snapshot;
3392 struct list_head splice;
3393
3394 INIT_LIST_HEAD(&splice);
3395
3396 list_splice_init(&t->pending_snapshots, &splice);
3397
3398 while (!list_empty(&splice)) {
3399 snapshot = list_entry(splice.next,
3400 struct btrfs_pending_snapshot,
3401 list);
3402
3403 list_del_init(&snapshot->list);
3404
3405 kfree(snapshot);
3406 }
3407
3408 return 0;
3409 }
3410
3411 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3412 {
3413 struct btrfs_inode *btrfs_inode;
3414 struct list_head splice;
3415
3416 INIT_LIST_HEAD(&splice);
3417
3418 spin_lock(&root->fs_info->delalloc_lock);
3419 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3420
3421 while (!list_empty(&splice)) {
3422 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3423 delalloc_inodes);
3424
3425 list_del_init(&btrfs_inode->delalloc_inodes);
3426
3427 btrfs_invalidate_inodes(btrfs_inode->root);
3428 }
3429
3430 spin_unlock(&root->fs_info->delalloc_lock);
3431
3432 return 0;
3433 }
3434
3435 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3436 struct extent_io_tree *dirty_pages,
3437 int mark)
3438 {
3439 int ret;
3440 struct page *page;
3441 struct inode *btree_inode = root->fs_info->btree_inode;
3442 struct extent_buffer *eb;
3443 u64 start = 0;
3444 u64 end;
3445 u64 offset;
3446 unsigned long index;
3447
3448 while (1) {
3449 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3450 mark);
3451 if (ret)
3452 break;
3453
3454 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3455 while (start <= end) {
3456 index = start >> PAGE_CACHE_SHIFT;
3457 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3458 page = find_get_page(btree_inode->i_mapping, index);
3459 if (!page)
3460 continue;
3461 offset = page_offset(page);
3462
3463 spin_lock(&dirty_pages->buffer_lock);
3464 eb = radix_tree_lookup(
3465 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3466 offset >> PAGE_CACHE_SHIFT);
3467 spin_unlock(&dirty_pages->buffer_lock);
3468 if (eb) {
3469 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3470 &eb->bflags);
3471 atomic_set(&eb->refs, 1);
3472 }
3473 if (PageWriteback(page))
3474 end_page_writeback(page);
3475
3476 lock_page(page);
3477 if (PageDirty(page)) {
3478 clear_page_dirty_for_io(page);
3479 spin_lock_irq(&page->mapping->tree_lock);
3480 radix_tree_tag_clear(&page->mapping->page_tree,
3481 page_index(page),
3482 PAGECACHE_TAG_DIRTY);
3483 spin_unlock_irq(&page->mapping->tree_lock);
3484 }
3485
3486 page->mapping->a_ops->invalidatepage(page, 0);
3487 unlock_page(page);
3488 }
3489 }
3490
3491 return ret;
3492 }
3493
3494 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3495 struct extent_io_tree *pinned_extents)
3496 {
3497 struct extent_io_tree *unpin;
3498 u64 start;
3499 u64 end;
3500 int ret;
3501
3502 unpin = pinned_extents;
3503 while (1) {
3504 ret = find_first_extent_bit(unpin, 0, &start, &end,
3505 EXTENT_DIRTY);
3506 if (ret)
3507 break;
3508
3509 /* opt_discard */
3510 if (btrfs_test_opt(root, DISCARD))
3511 ret = btrfs_error_discard_extent(root, start,
3512 end + 1 - start,
3513 NULL);
3514
3515 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3516 btrfs_error_unpin_extent_range(root, start, end);
3517 cond_resched();
3518 }
3519
3520 return 0;
3521 }
3522
3523 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3524 {
3525 struct btrfs_transaction *t;
3526 LIST_HEAD(list);
3527
3528 WARN_ON(1);
3529
3530 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3531
3532 spin_lock(&root->fs_info->trans_lock);
3533 list_splice_init(&root->fs_info->trans_list, &list);
3534 root->fs_info->trans_no_join = 1;
3535 spin_unlock(&root->fs_info->trans_lock);
3536
3537 while (!list_empty(&list)) {
3538 t = list_entry(list.next, struct btrfs_transaction, list);
3539 if (!t)
3540 break;
3541
3542 btrfs_destroy_ordered_operations(root);
3543
3544 btrfs_destroy_ordered_extents(root);
3545
3546 btrfs_destroy_delayed_refs(t, root);
3547
3548 btrfs_block_rsv_release(root,
3549 &root->fs_info->trans_block_rsv,
3550 t->dirty_pages.dirty_bytes);
3551
3552 /* FIXME: cleanup wait for commit */
3553 t->in_commit = 1;
3554 t->blocked = 1;
3555 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3556 wake_up(&root->fs_info->transaction_blocked_wait);
3557
3558 t->blocked = 0;
3559 if (waitqueue_active(&root->fs_info->transaction_wait))
3560 wake_up(&root->fs_info->transaction_wait);
3561
3562 t->commit_done = 1;
3563 if (waitqueue_active(&t->commit_wait))
3564 wake_up(&t->commit_wait);
3565
3566 btrfs_destroy_pending_snapshots(t);
3567
3568 btrfs_destroy_delalloc_inodes(root);
3569
3570 spin_lock(&root->fs_info->trans_lock);
3571 root->fs_info->running_transaction = NULL;
3572 spin_unlock(&root->fs_info->trans_lock);
3573
3574 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3575 EXTENT_DIRTY);
3576
3577 btrfs_destroy_pinned_extent(root,
3578 root->fs_info->pinned_extents);
3579
3580 atomic_set(&t->use_count, 0);
3581 list_del_init(&t->list);
3582 memset(t, 0, sizeof(*t));
3583 kmem_cache_free(btrfs_transaction_cachep, t);
3584 }
3585
3586 spin_lock(&root->fs_info->trans_lock);
3587 root->fs_info->trans_no_join = 0;
3588 spin_unlock(&root->fs_info->trans_lock);
3589 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3590
3591 return 0;
3592 }
3593
3594 static struct extent_io_ops btree_extent_io_ops = {
3595 .write_cache_pages_lock_hook = btree_lock_page_hook,
3596 .readpage_end_io_hook = btree_readpage_end_io_hook,
3597 .readpage_io_failed_hook = btree_io_failed_hook,
3598 .submit_bio_hook = btree_submit_bio_hook,
3599 /* note we're sharing with inode.c for the merge bio hook */
3600 .merge_bio_hook = btrfs_merge_bio_hook,
3601 };
Linus' kernel tree