?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Add linux-next specific files for 20110831
[linux-2.6/next.git] / fs / btrfs / disk-io.c
blob64b1c071bd647bc17aa8de9317c9d9407f9ebd55
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
47 static struct extent_io_ops btree_extent_io_ops;
48 static void end_workqueue_fn(struct btrfs_work *work);
49 static void free_fs_root(struct btrfs_root *root);
50 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
51 int read_only);
52 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
53 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
55 struct btrfs_root *root);
56 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
57 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
59 struct extent_io_tree *dirty_pages,
60 int mark);
61 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
62 struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_root *root);
64
65 /*
66 * end_io_wq structs are used to do processing in task context when an IO is
67 * complete. This is used during reads to verify checksums, and it is used
68 * by writes to insert metadata for new file extents after IO is complete.
69 */
70 struct end_io_wq {
71 struct bio *bio;
72 bio_end_io_t *end_io;
73 void *private;
74 struct btrfs_fs_info *info;
75 int error;
76 int metadata;
77 struct list_head list;
78 struct btrfs_work work;
79 };
80
81 /*
82 * async submit bios are used to offload expensive checksumming
83 * onto the worker threads. They checksum file and metadata bios
84 * just before they are sent down the IO stack.
85 */
86 struct async_submit_bio {
87 struct inode *inode;
88 struct bio *bio;
89 struct list_head list;
90 extent_submit_bio_hook_t *submit_bio_start;
91 extent_submit_bio_hook_t *submit_bio_done;
92 int rw;
93 int mirror_num;
94 unsigned long bio_flags;
95 /*
96 * bio_offset is optional, can be used if the pages in the bio
97 * can't tell us where in the file the bio should go
98 */
99 u64 bio_offset;
100 struct btrfs_work work;
101 };
102
103 /*
104 * Lockdep class keys for extent_buffer->lock's in this root. For a given
105 * eb, the lockdep key is determined by the btrfs_root it belongs to and
106 * the level the eb occupies in the tree.
107 *
108 * Different roots are used for different purposes and may nest inside each
109 * other and they require separate keysets. As lockdep keys should be
110 * static, assign keysets according to the purpose of the root as indicated
111 * by btrfs_root->objectid. This ensures that all special purpose roots
112 * have separate keysets.
113 *
114 * Lock-nesting across peer nodes is always done with the immediate parent
115 * node locked thus preventing deadlock. As lockdep doesn't know this, use
116 * subclass to avoid triggering lockdep warning in such cases.
117 *
118 * The key is set by the readpage_end_io_hook after the buffer has passed
119 * csum validation but before the pages are unlocked. It is also set by
120 * btrfs_init_new_buffer on freshly allocated blocks.
121 *
122 * We also add a check to make sure the highest level of the tree is the
123 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
124 * needs update as well.
125 */
126 #ifdef CONFIG_DEBUG_LOCK_ALLOC
127 # if BTRFS_MAX_LEVEL != 8
128 # error
129 # endif
130
131 static struct btrfs_lockdep_keyset {
132 u64 id; /* root objectid */
133 const char *name_stem; /* lock name stem */
134 char names[BTRFS_MAX_LEVEL + 1][20];
135 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
136 } btrfs_lockdep_keysets[] = {
137 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
138 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
139 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
140 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
141 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
142 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
143 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
144 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
145 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
146 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
147 { .id = 0, .name_stem = "tree" },
148 };
149
150 void __init btrfs_init_lockdep(void)
151 {
152 int i, j;
153
154 /* initialize lockdep class names */
155 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
156 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
157
158 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
159 snprintf(ks->names[j], sizeof(ks->names[j]),
160 "btrfs-%s-%02d", ks->name_stem, j);
161 }
162 }
163
164 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
165 int level)
166 {
167 struct btrfs_lockdep_keyset *ks;
168
169 BUG_ON(level >= ARRAY_SIZE(ks->keys));
170
171 /* find the matching keyset, id 0 is the default entry */
172 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
173 if (ks->id == objectid)
174 break;
175
176 lockdep_set_class_and_name(&eb->lock,
177 &ks->keys[level], ks->names[level]);
178 }
179
180 #endif
181
182 /*
183 * extents on the btree inode are pretty simple, there's one extent
184 * that covers the entire device
185 */
186 static struct extent_map *btree_get_extent(struct inode *inode,
187 struct page *page, size_t pg_offset, u64 start, u64 len,
188 int create)
189 {
190 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
191 struct extent_map *em;
192 int ret;
193
194 read_lock(&em_tree->lock);
195 em = lookup_extent_mapping(em_tree, start, len);
196 if (em) {
197 em->bdev =
198 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
199 read_unlock(&em_tree->lock);
200 goto out;
201 }
202 read_unlock(&em_tree->lock);
203
204 em = alloc_extent_map();
205 if (!em) {
206 em = ERR_PTR(-ENOMEM);
207 goto out;
208 }
209 em->start = 0;
210 em->len = (u64)-1;
211 em->block_len = (u64)-1;
212 em->block_start = 0;
213 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
214
215 write_lock(&em_tree->lock);
216 ret = add_extent_mapping(em_tree, em);
217 if (ret == -EEXIST) {
218 u64 failed_start = em->start;
219 u64 failed_len = em->len;
220
221 free_extent_map(em);
222 em = lookup_extent_mapping(em_tree, start, len);
223 if (em) {
224 ret = 0;
225 } else {
226 em = lookup_extent_mapping(em_tree, failed_start,
227 failed_len);
228 ret = -EIO;
229 }
230 } else if (ret) {
231 free_extent_map(em);
232 em = NULL;
233 }
234 write_unlock(&em_tree->lock);
235
236 if (ret)
237 em = ERR_PTR(ret);
238 out:
239 return em;
240 }
241
242 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
243 {
244 return crc32c(seed, data, len);
245 }
246
247 void btrfs_csum_final(u32 crc, char *result)
248 {
249 put_unaligned_le32(~crc, result);
250 }
251
252 /*
253 * compute the csum for a btree block, and either verify it or write it
254 * into the csum field of the block.
255 */
256 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
257 int verify)
258 {
259 u16 csum_size =
260 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, 1,
371 btree_get_extent, mirror_num);
372 if (!ret &&
373 !verify_parent_transid(io_tree, eb, parent_transid))
374 return ret;
375
376 /*
377 * This buffer's crc is fine, but its contents are corrupted, so
378 * there is no reason to read the other copies, they won't be
379 * any less wrong.
380 */
381 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
382 return ret;
383
384 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
385 eb->start, eb->len);
386 if (num_copies == 1)
387 return ret;
388
389 mirror_num++;
390 if (mirror_num > num_copies)
391 return ret;
392 }
393 return -EIO;
394 }
395
396 /*
397 * checksum a dirty tree block before IO. This has extra checks to make sure
398 * we only fill in the checksum field in the first page of a multi-page block
399 */
400
401 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
402 {
403 struct extent_io_tree *tree;
404 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
405 u64 found_start;
406 unsigned long len;
407 struct extent_buffer *eb;
408 int ret;
409
410 tree = &BTRFS_I(page->mapping->host)->io_tree;
411
412 if (page->private == EXTENT_PAGE_PRIVATE) {
413 WARN_ON(1);
414 goto out;
415 }
416 if (!page->private) {
417 WARN_ON(1);
418 goto out;
419 }
420 len = page->private >> 2;
421 WARN_ON(len == 0);
422
423 eb = alloc_extent_buffer(tree, start, len, page);
424 if (eb == NULL) {
425 WARN_ON(1);
426 goto out;
427 }
428 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
429 btrfs_header_generation(eb));
430 BUG_ON(ret);
431 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
432
433 found_start = btrfs_header_bytenr(eb);
434 if (found_start != start) {
435 WARN_ON(1);
436 goto err;
437 }
438 if (eb->first_page != page) {
439 WARN_ON(1);
440 goto err;
441 }
442 if (!PageUptodate(page)) {
443 WARN_ON(1);
444 goto err;
445 }
446 csum_tree_block(root, eb, 0);
447 err:
448 free_extent_buffer(eb);
449 out:
450 return 0;
451 }
452
453 static int check_tree_block_fsid(struct btrfs_root *root,
454 struct extent_buffer *eb)
455 {
456 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
457 u8 fsid[BTRFS_UUID_SIZE];
458 int ret = 1;
459
460 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
461 BTRFS_FSID_SIZE);
462 while (fs_devices) {
463 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
464 ret = 0;
465 break;
466 }
467 fs_devices = fs_devices->seed;
468 }
469 return ret;
470 }
471
472 #define CORRUPT(reason, eb, root, slot) \
473 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
474 "root=%llu, slot=%d\n", reason, \
475 (unsigned long long)btrfs_header_bytenr(eb), \
476 (unsigned long long)root->objectid, slot)
477
478 static noinline int check_leaf(struct btrfs_root *root,
479 struct extent_buffer *leaf)
480 {
481 struct btrfs_key key;
482 struct btrfs_key leaf_key;
483 u32 nritems = btrfs_header_nritems(leaf);
484 int slot;
485
486 if (nritems == 0)
487 return 0;
488
489 /* Check the 0 item */
490 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
491 BTRFS_LEAF_DATA_SIZE(root)) {
492 CORRUPT("invalid item offset size pair", leaf, root, 0);
493 return -EIO;
494 }
495
496 /*
497 * Check to make sure each items keys are in the correct order and their
498 * offsets make sense. We only have to loop through nritems-1 because
499 * we check the current slot against the next slot, which verifies the
500 * next slot's offset+size makes sense and that the current's slot
501 * offset is correct.
502 */
503 for (slot = 0; slot < nritems - 1; slot++) {
504 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
505 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
506
507 /* Make sure the keys are in the right order */
508 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
509 CORRUPT("bad key order", leaf, root, slot);
510 return -EIO;
511 }
512
513 /*
514 * Make sure the offset and ends are right, remember that the
515 * item data starts at the end of the leaf and grows towards the
516 * front.
517 */
518 if (btrfs_item_offset_nr(leaf, slot) !=
519 btrfs_item_end_nr(leaf, slot + 1)) {
520 CORRUPT("slot offset bad", leaf, root, slot);
521 return -EIO;
522 }
523
524 /*
525 * Check to make sure that we don't point outside of the leaf,
526 * just incase all the items are consistent to eachother, but
527 * all point outside of the leaf.
528 */
529 if (btrfs_item_end_nr(leaf, slot) >
530 BTRFS_LEAF_DATA_SIZE(root)) {
531 CORRUPT("slot end outside of leaf", leaf, root, slot);
532 return -EIO;
533 }
534 }
535
536 return 0;
537 }
538
539 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
540 struct extent_state *state)
541 {
542 struct extent_io_tree *tree;
543 u64 found_start;
544 int found_level;
545 unsigned long len;
546 struct extent_buffer *eb;
547 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
548 int ret = 0;
549
550 tree = &BTRFS_I(page->mapping->host)->io_tree;
551 if (page->private == EXTENT_PAGE_PRIVATE)
552 goto out;
553 if (!page->private)
554 goto out;
555
556 len = page->private >> 2;
557 WARN_ON(len == 0);
558
559 eb = alloc_extent_buffer(tree, start, len, page);
560 if (eb == NULL) {
561 ret = -EIO;
562 goto out;
563 }
564
565 found_start = btrfs_header_bytenr(eb);
566 if (found_start != start) {
567 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
568 "%llu %llu\n",
569 (unsigned long long)found_start,
570 (unsigned long long)eb->start);
571 ret = -EIO;
572 goto err;
573 }
574 if (eb->first_page != page) {
575 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
576 eb->first_page->index, page->index);
577 WARN_ON(1);
578 ret = -EIO;
579 goto err;
580 }
581 if (check_tree_block_fsid(root, eb)) {
582 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
583 (unsigned long long)eb->start);
584 ret = -EIO;
585 goto err;
586 }
587 found_level = btrfs_header_level(eb);
588
589 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
590 eb, found_level);
591
592 ret = csum_tree_block(root, eb, 1);
593 if (ret) {
594 ret = -EIO;
595 goto err;
596 }
597
598 /*
599 * If this is a leaf block and it is corrupt, set the corrupt bit so
600 * that we don't try and read the other copies of this block, just
601 * return -EIO.
602 */
603 if (found_level == 0 && check_leaf(root, eb)) {
604 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
605 ret = -EIO;
606 }
607
608 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
609 end = eb->start + end - 1;
610 err:
611 free_extent_buffer(eb);
612 out:
613 return ret;
614 }
615
616 static void end_workqueue_bio(struct bio *bio, int err)
617 {
618 struct end_io_wq *end_io_wq = bio->bi_private;
619 struct btrfs_fs_info *fs_info;
620
621 fs_info = end_io_wq->info;
622 end_io_wq->error = err;
623 end_io_wq->work.func = end_workqueue_fn;
624 end_io_wq->work.flags = 0;
625
626 if (bio->bi_rw & REQ_WRITE) {
627 if (end_io_wq->metadata == 1)
628 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
629 &end_io_wq->work);
630 else if (end_io_wq->metadata == 2)
631 btrfs_queue_worker(&fs_info->endio_freespace_worker,
632 &end_io_wq->work);
633 else
634 btrfs_queue_worker(&fs_info->endio_write_workers,
635 &end_io_wq->work);
636 } else {
637 if (end_io_wq->metadata)
638 btrfs_queue_worker(&fs_info->endio_meta_workers,
639 &end_io_wq->work);
640 else
641 btrfs_queue_worker(&fs_info->endio_workers,
642 &end_io_wq->work);
643 }
644 }
645
646 /*
647 * For the metadata arg you want
648 *
649 * 0 - if data
650 * 1 - if normal metadta
651 * 2 - if writing to the free space cache area
652 */
653 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
654 int metadata)
655 {
656 struct end_io_wq *end_io_wq;
657 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
658 if (!end_io_wq)
659 return -ENOMEM;
660
661 end_io_wq->private = bio->bi_private;
662 end_io_wq->end_io = bio->bi_end_io;
663 end_io_wq->info = info;
664 end_io_wq->error = 0;
665 end_io_wq->bio = bio;
666 end_io_wq->metadata = metadata;
667
668 bio->bi_private = end_io_wq;
669 bio->bi_end_io = end_workqueue_bio;
670 return 0;
671 }
672
673 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
674 {
675 unsigned long limit = min_t(unsigned long,
676 info->workers.max_workers,
677 info->fs_devices->open_devices);
678 return 256 * limit;
679 }
680
681 static void run_one_async_start(struct btrfs_work *work)
682 {
683 struct async_submit_bio *async;
684
685 async = container_of(work, struct async_submit_bio, work);
686 async->submit_bio_start(async->inode, async->rw, async->bio,
687 async->mirror_num, async->bio_flags,
688 async->bio_offset);
689 }
690
691 static void run_one_async_done(struct btrfs_work *work)
692 {
693 struct btrfs_fs_info *fs_info;
694 struct async_submit_bio *async;
695 int limit;
696
697 async = container_of(work, struct async_submit_bio, work);
698 fs_info = BTRFS_I(async->inode)->root->fs_info;
699
700 limit = btrfs_async_submit_limit(fs_info);
701 limit = limit * 2 / 3;
702
703 atomic_dec(&fs_info->nr_async_submits);
704
705 if (atomic_read(&fs_info->nr_async_submits) < limit &&
706 waitqueue_active(&fs_info->async_submit_wait))
707 wake_up(&fs_info->async_submit_wait);
708
709 async->submit_bio_done(async->inode, async->rw, async->bio,
710 async->mirror_num, async->bio_flags,
711 async->bio_offset);
712 }
713
714 static void run_one_async_free(struct btrfs_work *work)
715 {
716 struct async_submit_bio *async;
717
718 async = container_of(work, struct async_submit_bio, work);
719 kfree(async);
720 }
721
722 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
723 int rw, struct bio *bio, int mirror_num,
724 unsigned long bio_flags,
725 u64 bio_offset,
726 extent_submit_bio_hook_t *submit_bio_start,
727 extent_submit_bio_hook_t *submit_bio_done)
728 {
729 struct async_submit_bio *async;
730
731 async = kmalloc(sizeof(*async), GFP_NOFS);
732 if (!async)
733 return -ENOMEM;
734
735 async->inode = inode;
736 async->rw = rw;
737 async->bio = bio;
738 async->mirror_num = mirror_num;
739 async->submit_bio_start = submit_bio_start;
740 async->submit_bio_done = submit_bio_done;
741
742 async->work.func = run_one_async_start;
743 async->work.ordered_func = run_one_async_done;
744 async->work.ordered_free = run_one_async_free;
745
746 async->work.flags = 0;
747 async->bio_flags = bio_flags;
748 async->bio_offset = bio_offset;
749
750 atomic_inc(&fs_info->nr_async_submits);
751
752 if (rw & REQ_SYNC)
753 btrfs_set_work_high_prio(&async->work);
754
755 btrfs_queue_worker(&fs_info->workers, &async->work);
756
757 while (atomic_read(&fs_info->async_submit_draining) &&
758 atomic_read(&fs_info->nr_async_submits)) {
759 wait_event(fs_info->async_submit_wait,
760 (atomic_read(&fs_info->nr_async_submits) == 0));
761 }
762
763 return 0;
764 }
765
766 static int btree_csum_one_bio(struct bio *bio)
767 {
768 struct bio_vec *bvec = bio->bi_io_vec;
769 int bio_index = 0;
770 struct btrfs_root *root;
771
772 WARN_ON(bio->bi_vcnt <= 0);
773 while (bio_index < bio->bi_vcnt) {
774 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
775 csum_dirty_buffer(root, bvec->bv_page);
776 bio_index++;
777 bvec++;
778 }
779 return 0;
780 }
781
782 static int __btree_submit_bio_start(struct inode *inode, int rw,
783 struct bio *bio, int mirror_num,
784 unsigned long bio_flags,
785 u64 bio_offset)
786 {
787 /*
788 * when we're called for a write, we're already in the async
789 * submission context. Just jump into btrfs_map_bio
790 */
791 btree_csum_one_bio(bio);
792 return 0;
793 }
794
795 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
796 int mirror_num, unsigned long bio_flags,
797 u64 bio_offset)
798 {
799 /*
800 * when we're called for a write, we're already in the async
801 * submission context. Just jump into btrfs_map_bio
802 */
803 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
804 }
805
806 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
807 int mirror_num, unsigned long bio_flags,
808 u64 bio_offset)
809 {
810 int ret;
811
812 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
813 bio, 1);
814 BUG_ON(ret);
815
816 if (!(rw & REQ_WRITE)) {
817 /*
818 * called for a read, do the setup so that checksum validation
819 * can happen in the async kernel threads
820 */
821 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
822 mirror_num, 0);
823 }
824
825 /*
826 * kthread helpers are used to submit writes so that checksumming
827 * can happen in parallel across all CPUs
828 */
829 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
830 inode, rw, bio, mirror_num, 0,
831 bio_offset,
832 __btree_submit_bio_start,
833 __btree_submit_bio_done);
834 }
835
836 #ifdef CONFIG_MIGRATION
837 static int btree_migratepage(struct address_space *mapping,
838 struct page *newpage, struct page *page)
839 {
840 /*
841 * we can't safely write a btree page from here,
842 * we haven't done the locking hook
843 */
844 if (PageDirty(page))
845 return -EAGAIN;
846 /*
847 * Buffers may be managed in a filesystem specific way.
848 * We must have no buffers or drop them.
849 */
850 if (page_has_private(page) &&
851 !try_to_release_page(page, GFP_KERNEL))
852 return -EAGAIN;
853 return migrate_page(mapping, newpage, page);
854 }
855 #endif
856
857 static int btree_writepage(struct page *page, struct writeback_control *wbc)
858 {
859 struct extent_io_tree *tree;
860 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
861 struct extent_buffer *eb;
862 int was_dirty;
863
864 tree = &BTRFS_I(page->mapping->host)->io_tree;
865 if (!(current->flags & PF_MEMALLOC)) {
866 return extent_write_full_page(tree, page,
867 btree_get_extent, wbc);
868 }
869
870 redirty_page_for_writepage(wbc, page);
871 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
872 WARN_ON(!eb);
873
874 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
875 if (!was_dirty) {
876 spin_lock(&root->fs_info->delalloc_lock);
877 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
878 spin_unlock(&root->fs_info->delalloc_lock);
879 }
880 free_extent_buffer(eb);
881
882 unlock_page(page);
883 return 0;
884 }
885
886 static int btree_writepages(struct address_space *mapping,
887 struct writeback_control *wbc)
888 {
889 struct extent_io_tree *tree;
890 tree = &BTRFS_I(mapping->host)->io_tree;
891 if (wbc->sync_mode == WB_SYNC_NONE) {
892 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
893 u64 num_dirty;
894 unsigned long thresh = 32 * 1024 * 1024;
895
896 if (wbc->for_kupdate)
897 return 0;
898
899 /* this is a bit racy, but that's ok */
900 num_dirty = root->fs_info->dirty_metadata_bytes;
901 if (num_dirty < thresh)
902 return 0;
903 }
904 return extent_writepages(tree, mapping, btree_get_extent, wbc);
905 }
906
907 static int btree_readpage(struct file *file, struct page *page)
908 {
909 struct extent_io_tree *tree;
910 tree = &BTRFS_I(page->mapping->host)->io_tree;
911 return extent_read_full_page(tree, page, btree_get_extent);
912 }
913
914 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
915 {
916 struct extent_io_tree *tree;
917 struct extent_map_tree *map;
918 int ret;
919
920 if (PageWriteback(page) || PageDirty(page))
921 return 0;
922
923 tree = &BTRFS_I(page->mapping->host)->io_tree;
924 map = &BTRFS_I(page->mapping->host)->extent_tree;
925
926 ret = try_release_extent_state(map, tree, page, gfp_flags);
927 if (!ret)
928 return 0;
929
930 ret = try_release_extent_buffer(tree, page);
931 if (ret == 1) {
932 ClearPagePrivate(page);
933 set_page_private(page, 0);
934 page_cache_release(page);
935 }
936
937 return ret;
938 }
939
940 static void btree_invalidatepage(struct page *page, unsigned long offset)
941 {
942 struct extent_io_tree *tree;
943 tree = &BTRFS_I(page->mapping->host)->io_tree;
944 extent_invalidatepage(tree, page, offset);
945 btree_releasepage(page, GFP_NOFS);
946 if (PagePrivate(page)) {
947 printk(KERN_WARNING "btrfs warning page private not zero "
948 "on page %llu\n", (unsigned long long)page_offset(page));
949 ClearPagePrivate(page);
950 set_page_private(page, 0);
951 page_cache_release(page);
952 }
953 }
954
955 static const struct address_space_operations btree_aops = {
956 .readpage = btree_readpage,
957 .writepage = btree_writepage,
958 .writepages = btree_writepages,
959 .releasepage = btree_releasepage,
960 .invalidatepage = btree_invalidatepage,
961 #ifdef CONFIG_MIGRATION
962 .migratepage = btree_migratepage,
963 #endif
964 };
965
966 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
967 u64 parent_transid)
968 {
969 struct extent_buffer *buf = NULL;
970 struct inode *btree_inode = root->fs_info->btree_inode;
971 int ret = 0;
972
973 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
974 if (!buf)
975 return 0;
976 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
977 buf, 0, 0, btree_get_extent, 0);
978 free_extent_buffer(buf);
979 return ret;
980 }
981
982 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
983 u64 bytenr, u32 blocksize)
984 {
985 struct inode *btree_inode = root->fs_info->btree_inode;
986 struct extent_buffer *eb;
987 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
988 bytenr, blocksize);
989 return eb;
990 }
991
992 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
993 u64 bytenr, u32 blocksize)
994 {
995 struct inode *btree_inode = root->fs_info->btree_inode;
996 struct extent_buffer *eb;
997
998 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
999 bytenr, blocksize, NULL);
1000 return eb;
1001 }
1002
1003
1004 int btrfs_write_tree_block(struct extent_buffer *buf)
1005 {
1006 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1007 buf->start + buf->len - 1);
1008 }
1009
1010 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1011 {
1012 return filemap_fdatawait_range(buf->first_page->mapping,
1013 buf->start, buf->start + buf->len - 1);
1014 }
1015
1016 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1017 u32 blocksize, u64 parent_transid)
1018 {
1019 struct extent_buffer *buf = NULL;
1020 int ret;
1021
1022 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1023 if (!buf)
1024 return NULL;
1025
1026 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1027
1028 if (ret == 0)
1029 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1030 return buf;
1031
1032 }
1033
1034 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1035 struct extent_buffer *buf)
1036 {
1037 struct inode *btree_inode = root->fs_info->btree_inode;
1038 if (btrfs_header_generation(buf) ==
1039 root->fs_info->running_transaction->transid) {
1040 btrfs_assert_tree_locked(buf);
1041
1042 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1043 spin_lock(&root->fs_info->delalloc_lock);
1044 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1045 root->fs_info->dirty_metadata_bytes -= buf->len;
1046 else
1047 WARN_ON(1);
1048 spin_unlock(&root->fs_info->delalloc_lock);
1049 }
1050
1051 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1052 btrfs_set_lock_blocking(buf);
1053 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1054 buf);
1055 }
1056 return 0;
1057 }
1058
1059 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1060 u32 stripesize, struct btrfs_root *root,
1061 struct btrfs_fs_info *fs_info,
1062 u64 objectid)
1063 {
1064 root->node = NULL;
1065 root->commit_root = NULL;
1066 root->sectorsize = sectorsize;
1067 root->nodesize = nodesize;
1068 root->leafsize = leafsize;
1069 root->stripesize = stripesize;
1070 root->ref_cows = 0;
1071 root->track_dirty = 0;
1072 root->in_radix = 0;
1073 root->orphan_item_inserted = 0;
1074 root->orphan_cleanup_state = 0;
1075
1076 root->fs_info = fs_info;
1077 root->objectid = objectid;
1078 root->last_trans = 0;
1079 root->highest_objectid = 0;
1080 root->name = NULL;
1081 root->inode_tree = RB_ROOT;
1082 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1083 root->block_rsv = NULL;
1084 root->orphan_block_rsv = NULL;
1085
1086 INIT_LIST_HEAD(&root->dirty_list);
1087 INIT_LIST_HEAD(&root->orphan_list);
1088 INIT_LIST_HEAD(&root->root_list);
1089 spin_lock_init(&root->orphan_lock);
1090 spin_lock_init(&root->inode_lock);
1091 spin_lock_init(&root->accounting_lock);
1092 mutex_init(&root->objectid_mutex);
1093 mutex_init(&root->log_mutex);
1094 init_waitqueue_head(&root->log_writer_wait);
1095 init_waitqueue_head(&root->log_commit_wait[0]);
1096 init_waitqueue_head(&root->log_commit_wait[1]);
1097 atomic_set(&root->log_commit[0], 0);
1098 atomic_set(&root->log_commit[1], 0);
1099 atomic_set(&root->log_writers, 0);
1100 root->log_batch = 0;
1101 root->log_transid = 0;
1102 root->last_log_commit = 0;
1103 extent_io_tree_init(&root->dirty_log_pages,
1104 fs_info->btree_inode->i_mapping);
1105
1106 memset(&root->root_key, 0, sizeof(root->root_key));
1107 memset(&root->root_item, 0, sizeof(root->root_item));
1108 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1109 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1110 root->defrag_trans_start = fs_info->generation;
1111 init_completion(&root->kobj_unregister);
1112 root->defrag_running = 0;
1113 root->root_key.objectid = objectid;
1114 root->anon_dev = 0;
1115 return 0;
1116 }
1117
1118 static int find_and_setup_root(struct btrfs_root *tree_root,
1119 struct btrfs_fs_info *fs_info,
1120 u64 objectid,
1121 struct btrfs_root *root)
1122 {
1123 int ret;
1124 u32 blocksize;
1125 u64 generation;
1126
1127 __setup_root(tree_root->nodesize, tree_root->leafsize,
1128 tree_root->sectorsize, tree_root->stripesize,
1129 root, fs_info, objectid);
1130 ret = btrfs_find_last_root(tree_root, objectid,
1131 &root->root_item, &root->root_key);
1132 if (ret > 0)
1133 return -ENOENT;
1134 BUG_ON(ret);
1135
1136 generation = btrfs_root_generation(&root->root_item);
1137 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1138 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1139 blocksize, generation);
1140 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1141 free_extent_buffer(root->node);
1142 return -EIO;
1143 }
1144 root->commit_root = btrfs_root_node(root);
1145 return 0;
1146 }
1147
1148 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1149 struct btrfs_fs_info *fs_info)
1150 {
1151 struct btrfs_root *root;
1152 struct btrfs_root *tree_root = fs_info->tree_root;
1153 struct extent_buffer *leaf;
1154
1155 root = kzalloc(sizeof(*root), GFP_NOFS);
1156 if (!root)
1157 return ERR_PTR(-ENOMEM);
1158
1159 __setup_root(tree_root->nodesize, tree_root->leafsize,
1160 tree_root->sectorsize, tree_root->stripesize,
1161 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1162
1163 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1164 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1165 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1166 /*
1167 * log trees do not get reference counted because they go away
1168 * before a real commit is actually done. They do store pointers
1169 * to file data extents, and those reference counts still get
1170 * updated (along with back refs to the log tree).
1171 */
1172 root->ref_cows = 0;
1173
1174 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1175 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1176 if (IS_ERR(leaf)) {
1177 kfree(root);
1178 return ERR_CAST(leaf);
1179 }
1180
1181 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1182 btrfs_set_header_bytenr(leaf, leaf->start);
1183 btrfs_set_header_generation(leaf, trans->transid);
1184 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1185 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1186 root->node = leaf;
1187
1188 write_extent_buffer(root->node, root->fs_info->fsid,
1189 (unsigned long)btrfs_header_fsid(root->node),
1190 BTRFS_FSID_SIZE);
1191 btrfs_mark_buffer_dirty(root->node);
1192 btrfs_tree_unlock(root->node);
1193 return root;
1194 }
1195
1196 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1197 struct btrfs_fs_info *fs_info)
1198 {
1199 struct btrfs_root *log_root;
1200
1201 log_root = alloc_log_tree(trans, fs_info);
1202 if (IS_ERR(log_root))
1203 return PTR_ERR(log_root);
1204 WARN_ON(fs_info->log_root_tree);
1205 fs_info->log_root_tree = log_root;
1206 return 0;
1207 }
1208
1209 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1210 struct btrfs_root *root)
1211 {
1212 struct btrfs_root *log_root;
1213 struct btrfs_inode_item *inode_item;
1214
1215 log_root = alloc_log_tree(trans, root->fs_info);
1216 if (IS_ERR(log_root))
1217 return PTR_ERR(log_root);
1218
1219 log_root->last_trans = trans->transid;
1220 log_root->root_key.offset = root->root_key.objectid;
1221
1222 inode_item = &log_root->root_item.inode;
1223 inode_item->generation = cpu_to_le64(1);
1224 inode_item->size = cpu_to_le64(3);
1225 inode_item->nlink = cpu_to_le32(1);
1226 inode_item->nbytes = cpu_to_le64(root->leafsize);
1227 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1228
1229 btrfs_set_root_node(&log_root->root_item, log_root->node);
1230
1231 WARN_ON(root->log_root);
1232 root->log_root = log_root;
1233 root->log_transid = 0;
1234 root->last_log_commit = 0;
1235 return 0;
1236 }
1237
1238 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1239 struct btrfs_key *location)
1240 {
1241 struct btrfs_root *root;
1242 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1243 struct btrfs_path *path;
1244 struct extent_buffer *l;
1245 u64 generation;
1246 u32 blocksize;
1247 int ret = 0;
1248
1249 root = kzalloc(sizeof(*root), GFP_NOFS);
1250 if (!root)
1251 return ERR_PTR(-ENOMEM);
1252 if (location->offset == (u64)-1) {
1253 ret = find_and_setup_root(tree_root, fs_info,
1254 location->objectid, root);
1255 if (ret) {
1256 kfree(root);
1257 return ERR_PTR(ret);
1258 }
1259 goto out;
1260 }
1261
1262 __setup_root(tree_root->nodesize, tree_root->leafsize,
1263 tree_root->sectorsize, tree_root->stripesize,
1264 root, fs_info, location->objectid);
1265
1266 path = btrfs_alloc_path();
1267 if (!path) {
1268 kfree(root);
1269 return ERR_PTR(-ENOMEM);
1270 }
1271 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1272 if (ret == 0) {
1273 l = path->nodes[0];
1274 read_extent_buffer(l, &root->root_item,
1275 btrfs_item_ptr_offset(l, path->slots[0]),
1276 sizeof(root->root_item));
1277 memcpy(&root->root_key, location, sizeof(*location));
1278 }
1279 btrfs_free_path(path);
1280 if (ret) {
1281 kfree(root);
1282 if (ret > 0)
1283 ret = -ENOENT;
1284 return ERR_PTR(ret);
1285 }
1286
1287 generation = btrfs_root_generation(&root->root_item);
1288 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1289 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1290 blocksize, generation);
1291 root->commit_root = btrfs_root_node(root);
1292 BUG_ON(!root->node);
1293 out:
1294 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1295 root->ref_cows = 1;
1296 btrfs_check_and_init_root_item(&root->root_item);
1297 }
1298
1299 return root;
1300 }
1301
1302 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1303 struct btrfs_key *location)
1304 {
1305 struct btrfs_root *root;
1306 int ret;
1307
1308 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1309 return fs_info->tree_root;
1310 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1311 return fs_info->extent_root;
1312 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1313 return fs_info->chunk_root;
1314 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1315 return fs_info->dev_root;
1316 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1317 return fs_info->csum_root;
1318 again:
1319 spin_lock(&fs_info->fs_roots_radix_lock);
1320 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1321 (unsigned long)location->objectid);
1322 spin_unlock(&fs_info->fs_roots_radix_lock);
1323 if (root)
1324 return root;
1325
1326 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1327 if (IS_ERR(root))
1328 return root;
1329
1330 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1331 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1332 GFP_NOFS);
1333 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1334 ret = -ENOMEM;
1335 goto fail;
1336 }
1337
1338 btrfs_init_free_ino_ctl(root);
1339 mutex_init(&root->fs_commit_mutex);
1340 spin_lock_init(&root->cache_lock);
1341 init_waitqueue_head(&root->cache_wait);
1342
1343 ret = get_anon_bdev(&root->anon_dev);
1344 if (ret)
1345 goto fail;
1346
1347 if (btrfs_root_refs(&root->root_item) == 0) {
1348 ret = -ENOENT;
1349 goto fail;
1350 }
1351
1352 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1353 if (ret < 0)
1354 goto fail;
1355 if (ret == 0)
1356 root->orphan_item_inserted = 1;
1357
1358 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1359 if (ret)
1360 goto fail;
1361
1362 spin_lock(&fs_info->fs_roots_radix_lock);
1363 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1364 (unsigned long)root->root_key.objectid,
1365 root);
1366 if (ret == 0)
1367 root->in_radix = 1;
1368
1369 spin_unlock(&fs_info->fs_roots_radix_lock);
1370 radix_tree_preload_end();
1371 if (ret) {
1372 if (ret == -EEXIST) {
1373 free_fs_root(root);
1374 goto again;
1375 }
1376 goto fail;
1377 }
1378
1379 ret = btrfs_find_dead_roots(fs_info->tree_root,
1380 root->root_key.objectid);
1381 WARN_ON(ret);
1382 return root;
1383 fail:
1384 free_fs_root(root);
1385 return ERR_PTR(ret);
1386 }
1387
1388 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1389 {
1390 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1391 int ret = 0;
1392 struct btrfs_device *device;
1393 struct backing_dev_info *bdi;
1394
1395 rcu_read_lock();
1396 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1397 if (!device->bdev)
1398 continue;
1399 bdi = blk_get_backing_dev_info(device->bdev);
1400 if (bdi && bdi_congested(bdi, bdi_bits)) {
1401 ret = 1;
1402 break;
1403 }
1404 }
1405 rcu_read_unlock();
1406 return ret;
1407 }
1408
1409 /*
1410 * If this fails, caller must call bdi_destroy() to get rid of the
1411 * bdi again.
1412 */
1413 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1414 {
1415 int err;
1416
1417 bdi->capabilities = BDI_CAP_MAP_COPY;
1418 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1419 if (err)
1420 return err;
1421
1422 bdi->ra_pages = default_backing_dev_info.ra_pages;
1423 bdi->congested_fn = btrfs_congested_fn;
1424 bdi->congested_data = info;
1425 return 0;
1426 }
1427
1428 static int bio_ready_for_csum(struct bio *bio)
1429 {
1430 u64 length = 0;
1431 u64 buf_len = 0;
1432 u64 start = 0;
1433 struct page *page;
1434 struct extent_io_tree *io_tree = NULL;
1435 struct bio_vec *bvec;
1436 int i;
1437 int ret;
1438
1439 bio_for_each_segment(bvec, bio, i) {
1440 page = bvec->bv_page;
1441 if (page->private == EXTENT_PAGE_PRIVATE) {
1442 length += bvec->bv_len;
1443 continue;
1444 }
1445 if (!page->private) {
1446 length += bvec->bv_len;
1447 continue;
1448 }
1449 length = bvec->bv_len;
1450 buf_len = page->private >> 2;
1451 start = page_offset(page) + bvec->bv_offset;
1452 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1453 }
1454 /* are we fully contained in this bio? */
1455 if (buf_len <= length)
1456 return 1;
1457
1458 ret = extent_range_uptodate(io_tree, start + length,
1459 start + buf_len - 1);
1460 return ret;
1461 }
1462
1463 /*
1464 * called by the kthread helper functions to finally call the bio end_io
1465 * functions. This is where read checksum verification actually happens
1466 */
1467 static void end_workqueue_fn(struct btrfs_work *work)
1468 {
1469 struct bio *bio;
1470 struct end_io_wq *end_io_wq;
1471 struct btrfs_fs_info *fs_info;
1472 int error;
1473
1474 end_io_wq = container_of(work, struct end_io_wq, work);
1475 bio = end_io_wq->bio;
1476 fs_info = end_io_wq->info;
1477
1478 /* metadata bio reads are special because the whole tree block must
1479 * be checksummed at once. This makes sure the entire block is in
1480 * ram and up to date before trying to verify things. For
1481 * blocksize <= pagesize, it is basically a noop
1482 */
1483 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1484 !bio_ready_for_csum(bio)) {
1485 btrfs_queue_worker(&fs_info->endio_meta_workers,
1486 &end_io_wq->work);
1487 return;
1488 }
1489 error = end_io_wq->error;
1490 bio->bi_private = end_io_wq->private;
1491 bio->bi_end_io = end_io_wq->end_io;
1492 kfree(end_io_wq);
1493 bio_endio(bio, error);
1494 }
1495
1496 static int cleaner_kthread(void *arg)
1497 {
1498 struct btrfs_root *root = arg;
1499
1500 do {
1501 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1502
1503 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1504 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1505 btrfs_run_delayed_iputs(root);
1506 btrfs_clean_old_snapshots(root);
1507 mutex_unlock(&root->fs_info->cleaner_mutex);
1508 btrfs_run_defrag_inodes(root->fs_info);
1509 }
1510
1511 if (!try_to_freeze()) {
1512 set_current_state(TASK_INTERRUPTIBLE);
1513 if (!kthread_should_stop())
1514 schedule();
1515 __set_current_state(TASK_RUNNING);
1516 }
1517 } while (!kthread_should_stop());
1518 return 0;
1519 }
1520
1521 static int transaction_kthread(void *arg)
1522 {
1523 struct btrfs_root *root = arg;
1524 struct btrfs_trans_handle *trans;
1525 struct btrfs_transaction *cur;
1526 u64 transid;
1527 unsigned long now;
1528 unsigned long delay;
1529 int ret;
1530
1531 do {
1532 delay = HZ * 30;
1533 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1534 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1535
1536 spin_lock(&root->fs_info->trans_lock);
1537 cur = root->fs_info->running_transaction;
1538 if (!cur) {
1539 spin_unlock(&root->fs_info->trans_lock);
1540 goto sleep;
1541 }
1542
1543 now = get_seconds();
1544 if (!cur->blocked &&
1545 (now < cur->start_time || now - cur->start_time < 30)) {
1546 spin_unlock(&root->fs_info->trans_lock);
1547 delay = HZ * 5;
1548 goto sleep;
1549 }
1550 transid = cur->transid;
1551 spin_unlock(&root->fs_info->trans_lock);
1552
1553 trans = btrfs_join_transaction(root);
1554 BUG_ON(IS_ERR(trans));
1555 if (transid == trans->transid) {
1556 ret = btrfs_commit_transaction(trans, root);
1557 BUG_ON(ret);
1558 } else {
1559 btrfs_end_transaction(trans, root);
1560 }
1561 sleep:
1562 wake_up_process(root->fs_info->cleaner_kthread);
1563 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1564
1565 if (!try_to_freeze()) {
1566 set_current_state(TASK_INTERRUPTIBLE);
1567 if (!kthread_should_stop() &&
1568 !btrfs_transaction_blocked(root->fs_info))
1569 schedule_timeout(delay);
1570 __set_current_state(TASK_RUNNING);
1571 }
1572 } while (!kthread_should_stop());
1573 return 0;
1574 }
1575
1576 struct btrfs_root *open_ctree(struct super_block *sb,
1577 struct btrfs_fs_devices *fs_devices,
1578 char *options)
1579 {
1580 u32 sectorsize;
1581 u32 nodesize;
1582 u32 leafsize;
1583 u32 blocksize;
1584 u32 stripesize;
1585 u64 generation;
1586 u64 features;
1587 struct btrfs_key location;
1588 struct buffer_head *bh;
1589 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1590 GFP_NOFS);
1591 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1592 GFP_NOFS);
1593 struct btrfs_root *tree_root = btrfs_sb(sb);
1594 struct btrfs_fs_info *fs_info = NULL;
1595 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1596 GFP_NOFS);
1597 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1598 GFP_NOFS);
1599 struct btrfs_root *log_tree_root;
1600
1601 int ret;
1602 int err = -EINVAL;
1603
1604 struct btrfs_super_block *disk_super;
1605
1606 if (!extent_root || !tree_root || !tree_root->fs_info ||
1607 !chunk_root || !dev_root || !csum_root) {
1608 err = -ENOMEM;
1609 goto fail;
1610 }
1611 fs_info = tree_root->fs_info;
1612
1613 ret = init_srcu_struct(&fs_info->subvol_srcu);
1614 if (ret) {
1615 err = ret;
1616 goto fail;
1617 }
1618
1619 ret = setup_bdi(fs_info, &fs_info->bdi);
1620 if (ret) {
1621 err = ret;
1622 goto fail_srcu;
1623 }
1624
1625 fs_info->btree_inode = new_inode(sb);
1626 if (!fs_info->btree_inode) {
1627 err = -ENOMEM;
1628 goto fail_bdi;
1629 }
1630
1631 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1632
1633 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1634 INIT_LIST_HEAD(&fs_info->trans_list);
1635 INIT_LIST_HEAD(&fs_info->dead_roots);
1636 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1637 INIT_LIST_HEAD(&fs_info->hashers);
1638 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1639 INIT_LIST_HEAD(&fs_info->ordered_operations);
1640 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1641 spin_lock_init(&fs_info->delalloc_lock);
1642 spin_lock_init(&fs_info->trans_lock);
1643 spin_lock_init(&fs_info->ref_cache_lock);
1644 spin_lock_init(&fs_info->fs_roots_radix_lock);
1645 spin_lock_init(&fs_info->delayed_iput_lock);
1646 spin_lock_init(&fs_info->defrag_inodes_lock);
1647 mutex_init(&fs_info->reloc_mutex);
1648
1649 init_completion(&fs_info->kobj_unregister);
1650 fs_info->tree_root = tree_root;
1651 fs_info->extent_root = extent_root;
1652 fs_info->csum_root = csum_root;
1653 fs_info->chunk_root = chunk_root;
1654 fs_info->dev_root = dev_root;
1655 fs_info->fs_devices = fs_devices;
1656 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1657 INIT_LIST_HEAD(&fs_info->space_info);
1658 btrfs_mapping_init(&fs_info->mapping_tree);
1659 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1660 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1661 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1662 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1663 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1664 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1665 mutex_init(&fs_info->durable_block_rsv_mutex);
1666 atomic_set(&fs_info->nr_async_submits, 0);
1667 atomic_set(&fs_info->async_delalloc_pages, 0);
1668 atomic_set(&fs_info->async_submit_draining, 0);
1669 atomic_set(&fs_info->nr_async_bios, 0);
1670 atomic_set(&fs_info->defrag_running, 0);
1671 fs_info->sb = sb;
1672 fs_info->max_inline = 8192 * 1024;
1673 fs_info->metadata_ratio = 0;
1674 fs_info->defrag_inodes = RB_ROOT;
1675 fs_info->trans_no_join = 0;
1676
1677 fs_info->thread_pool_size = min_t(unsigned long,
1678 num_online_cpus() + 2, 8);
1679
1680 INIT_LIST_HEAD(&fs_info->ordered_extents);
1681 spin_lock_init(&fs_info->ordered_extent_lock);
1682 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1683 GFP_NOFS);
1684 if (!fs_info->delayed_root) {
1685 err = -ENOMEM;
1686 goto fail_iput;
1687 }
1688 btrfs_init_delayed_root(fs_info->delayed_root);
1689
1690 mutex_init(&fs_info->scrub_lock);
1691 atomic_set(&fs_info->scrubs_running, 0);
1692 atomic_set(&fs_info->scrub_pause_req, 0);
1693 atomic_set(&fs_info->scrubs_paused, 0);
1694 atomic_set(&fs_info->scrub_cancel_req, 0);
1695 init_waitqueue_head(&fs_info->scrub_pause_wait);
1696 init_rwsem(&fs_info->scrub_super_lock);
1697 fs_info->scrub_workers_refcnt = 0;
1698
1699 sb->s_blocksize = 4096;
1700 sb->s_blocksize_bits = blksize_bits(4096);
1701 sb->s_bdi = &fs_info->bdi;
1702
1703 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1704 fs_info->btree_inode->i_nlink = 1;
1705 /*
1706 * we set the i_size on the btree inode to the max possible int.
1707 * the real end of the address space is determined by all of
1708 * the devices in the system
1709 */
1710 fs_info->btree_inode->i_size = OFFSET_MAX;
1711 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1712 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1713
1714 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1715 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1716 fs_info->btree_inode->i_mapping);
1717 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1718
1719 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1720
1721 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1722 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1723 sizeof(struct btrfs_key));
1724 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1725 insert_inode_hash(fs_info->btree_inode);
1726
1727 spin_lock_init(&fs_info->block_group_cache_lock);
1728 fs_info->block_group_cache_tree = RB_ROOT;
1729
1730 extent_io_tree_init(&fs_info->freed_extents[0],
1731 fs_info->btree_inode->i_mapping);
1732 extent_io_tree_init(&fs_info->freed_extents[1],
1733 fs_info->btree_inode->i_mapping);
1734 fs_info->pinned_extents = &fs_info->freed_extents[0];
1735 fs_info->do_barriers = 1;
1736
1737
1738 mutex_init(&fs_info->ordered_operations_mutex);
1739 mutex_init(&fs_info->tree_log_mutex);
1740 mutex_init(&fs_info->chunk_mutex);
1741 mutex_init(&fs_info->transaction_kthread_mutex);
1742 mutex_init(&fs_info->cleaner_mutex);
1743 mutex_init(&fs_info->volume_mutex);
1744 init_rwsem(&fs_info->extent_commit_sem);
1745 init_rwsem(&fs_info->cleanup_work_sem);
1746 init_rwsem(&fs_info->subvol_sem);
1747
1748 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1749 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1750
1751 init_waitqueue_head(&fs_info->transaction_throttle);
1752 init_waitqueue_head(&fs_info->transaction_wait);
1753 init_waitqueue_head(&fs_info->transaction_blocked_wait);
1754 init_waitqueue_head(&fs_info->async_submit_wait);
1755
1756 __setup_root(4096, 4096, 4096, 4096, tree_root,
1757 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1758
1759 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1760 if (!bh) {
1761 err = -EINVAL;
1762 goto fail_alloc;
1763 }
1764
1765 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1766 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1767 sizeof(fs_info->super_for_commit));
1768 brelse(bh);
1769
1770 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1771
1772 disk_super = &fs_info->super_copy;
1773 if (!btrfs_super_root(disk_super))
1774 goto fail_alloc;
1775
1776 /* check FS state, whether FS is broken. */
1777 fs_info->fs_state |= btrfs_super_flags(disk_super);
1778
1779 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1780
1781 /*
1782 * In the long term, we'll store the compression type in the super
1783 * block, and it'll be used for per file compression control.
1784 */
1785 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1786
1787 ret = btrfs_parse_options(tree_root, options);
1788 if (ret) {
1789 err = ret;
1790 goto fail_alloc;
1791 }
1792
1793 features = btrfs_super_incompat_flags(disk_super) &
1794 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1795 if (features) {
1796 printk(KERN_ERR "BTRFS: couldn't mount because of "
1797 "unsupported optional features (%Lx).\n",
1798 (unsigned long long)features);
1799 err = -EINVAL;
1800 goto fail_alloc;
1801 }
1802
1803 features = btrfs_super_incompat_flags(disk_super);
1804 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1805 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1806 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1807 btrfs_set_super_incompat_flags(disk_super, features);
1808
1809 features = btrfs_super_compat_ro_flags(disk_super) &
1810 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1811 if (!(sb->s_flags & MS_RDONLY) && features) {
1812 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1813 "unsupported option features (%Lx).\n",
1814 (unsigned long long)features);
1815 err = -EINVAL;
1816 goto fail_alloc;
1817 }
1818
1819 btrfs_init_workers(&fs_info->generic_worker,
1820 "genwork", 1, NULL);
1821
1822 btrfs_init_workers(&fs_info->workers, "worker",
1823 fs_info->thread_pool_size,
1824 &fs_info->generic_worker);
1825
1826 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1827 fs_info->thread_pool_size,
1828 &fs_info->generic_worker);
1829
1830 btrfs_init_workers(&fs_info->submit_workers, "submit",
1831 min_t(u64, fs_devices->num_devices,
1832 fs_info->thread_pool_size),
1833 &fs_info->generic_worker);
1834
1835 btrfs_init_workers(&fs_info->caching_workers, "cache",
1836 2, &fs_info->generic_worker);
1837
1838 /* a higher idle thresh on the submit workers makes it much more
1839 * likely that bios will be send down in a sane order to the
1840 * devices
1841 */
1842 fs_info->submit_workers.idle_thresh = 64;
1843
1844 fs_info->workers.idle_thresh = 16;
1845 fs_info->workers.ordered = 1;
1846
1847 fs_info->delalloc_workers.idle_thresh = 2;
1848 fs_info->delalloc_workers.ordered = 1;
1849
1850 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1851 &fs_info->generic_worker);
1852 btrfs_init_workers(&fs_info->endio_workers, "endio",
1853 fs_info->thread_pool_size,
1854 &fs_info->generic_worker);
1855 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1856 fs_info->thread_pool_size,
1857 &fs_info->generic_worker);
1858 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1859 "endio-meta-write", fs_info->thread_pool_size,
1860 &fs_info->generic_worker);
1861 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1862 fs_info->thread_pool_size,
1863 &fs_info->generic_worker);
1864 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1865 1, &fs_info->generic_worker);
1866 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
1867 fs_info->thread_pool_size,
1868 &fs_info->generic_worker);
1869
1870 /*
1871 * endios are largely parallel and should have a very
1872 * low idle thresh
1873 */
1874 fs_info->endio_workers.idle_thresh = 4;
1875 fs_info->endio_meta_workers.idle_thresh = 4;
1876
1877 fs_info->endio_write_workers.idle_thresh = 2;
1878 fs_info->endio_meta_write_workers.idle_thresh = 2;
1879
1880 btrfs_start_workers(&fs_info->workers, 1);
1881 btrfs_start_workers(&fs_info->generic_worker, 1);
1882 btrfs_start_workers(&fs_info->submit_workers, 1);
1883 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1884 btrfs_start_workers(&fs_info->fixup_workers, 1);
1885 btrfs_start_workers(&fs_info->endio_workers, 1);
1886 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1887 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1888 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1889 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1890 btrfs_start_workers(&fs_info->delayed_workers, 1);
1891 btrfs_start_workers(&fs_info->caching_workers, 1);
1892
1893 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1894 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1895 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1896
1897 nodesize = btrfs_super_nodesize(disk_super);
1898 leafsize = btrfs_super_leafsize(disk_super);
1899 sectorsize = btrfs_super_sectorsize(disk_super);
1900 stripesize = btrfs_super_stripesize(disk_super);
1901 tree_root->nodesize = nodesize;
1902 tree_root->leafsize = leafsize;
1903 tree_root->sectorsize = sectorsize;
1904 tree_root->stripesize = stripesize;
1905
1906 sb->s_blocksize = sectorsize;
1907 sb->s_blocksize_bits = blksize_bits(sectorsize);
1908
1909 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1910 sizeof(disk_super->magic))) {
1911 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1912 goto fail_sb_buffer;
1913 }
1914
1915 mutex_lock(&fs_info->chunk_mutex);
1916 ret = btrfs_read_sys_array(tree_root);
1917 mutex_unlock(&fs_info->chunk_mutex);
1918 if (ret) {
1919 printk(KERN_WARNING "btrfs: failed to read the system "
1920 "array on %s\n", sb->s_id);
1921 goto fail_sb_buffer;
1922 }
1923
1924 blocksize = btrfs_level_size(tree_root,
1925 btrfs_super_chunk_root_level(disk_super));
1926 generation = btrfs_super_chunk_root_generation(disk_super);
1927
1928 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1929 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1930
1931 chunk_root->node = read_tree_block(chunk_root,
1932 btrfs_super_chunk_root(disk_super),
1933 blocksize, generation);
1934 BUG_ON(!chunk_root->node);
1935 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1936 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1937 sb->s_id);
1938 goto fail_chunk_root;
1939 }
1940 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1941 chunk_root->commit_root = btrfs_root_node(chunk_root);
1942
1943 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1944 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1945 BTRFS_UUID_SIZE);
1946
1947 mutex_lock(&fs_info->chunk_mutex);
1948 ret = btrfs_read_chunk_tree(chunk_root);
1949 mutex_unlock(&fs_info->chunk_mutex);
1950 if (ret) {
1951 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1952 sb->s_id);
1953 goto fail_chunk_root;
1954 }
1955
1956 btrfs_close_extra_devices(fs_devices);
1957
1958 blocksize = btrfs_level_size(tree_root,
1959 btrfs_super_root_level(disk_super));
1960 generation = btrfs_super_generation(disk_super);
1961
1962 tree_root->node = read_tree_block(tree_root,
1963 btrfs_super_root(disk_super),
1964 blocksize, generation);
1965 if (!tree_root->node)
1966 goto fail_chunk_root;
1967 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1968 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1969 sb->s_id);
1970 goto fail_tree_root;
1971 }
1972 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1973 tree_root->commit_root = btrfs_root_node(tree_root);
1974
1975 ret = find_and_setup_root(tree_root, fs_info,
1976 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1977 if (ret)
1978 goto fail_tree_root;
1979 extent_root->track_dirty = 1;
1980
1981 ret = find_and_setup_root(tree_root, fs_info,
1982 BTRFS_DEV_TREE_OBJECTID, dev_root);
1983 if (ret)
1984 goto fail_extent_root;
1985 dev_root->track_dirty = 1;
1986
1987 ret = find_and_setup_root(tree_root, fs_info,
1988 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1989 if (ret)
1990 goto fail_dev_root;
1991
1992 csum_root->track_dirty = 1;
1993
1994 fs_info->generation = generation;
1995 fs_info->last_trans_committed = generation;
1996 fs_info->data_alloc_profile = (u64)-1;
1997 fs_info->metadata_alloc_profile = (u64)-1;
1998 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1999
2000 ret = btrfs_init_space_info(fs_info);
2001 if (ret) {
2002 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2003 goto fail_block_groups;
2004 }
2005
2006 ret = btrfs_read_block_groups(extent_root);
2007 if (ret) {
2008 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2009 goto fail_block_groups;
2010 }
2011
2012 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2013 "btrfs-cleaner");
2014 if (IS_ERR(fs_info->cleaner_kthread))
2015 goto fail_block_groups;
2016
2017 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2018 tree_root,
2019 "btrfs-transaction");
2020 if (IS_ERR(fs_info->transaction_kthread))
2021 goto fail_cleaner;
2022
2023 if (!btrfs_test_opt(tree_root, SSD) &&
2024 !btrfs_test_opt(tree_root, NOSSD) &&
2025 !fs_info->fs_devices->rotating) {
2026 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2027 "mode\n");
2028 btrfs_set_opt(fs_info->mount_opt, SSD);
2029 }
2030
2031 /* do not make disk changes in broken FS */
2032 if (btrfs_super_log_root(disk_super) != 0 &&
2033 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2034 u64 bytenr = btrfs_super_log_root(disk_super);
2035
2036 if (fs_devices->rw_devices == 0) {
2037 printk(KERN_WARNING "Btrfs log replay required "
2038 "on RO media\n");
2039 err = -EIO;
2040 goto fail_trans_kthread;
2041 }
2042 blocksize =
2043 btrfs_level_size(tree_root,
2044 btrfs_super_log_root_level(disk_super));
2045
2046 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2047 if (!log_tree_root) {
2048 err = -ENOMEM;
2049 goto fail_trans_kthread;
2050 }
2051
2052 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2053 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2054
2055 log_tree_root->node = read_tree_block(tree_root, bytenr,
2056 blocksize,
2057 generation + 1);
2058 ret = btrfs_recover_log_trees(log_tree_root);
2059 BUG_ON(ret);
2060
2061 if (sb->s_flags & MS_RDONLY) {
2062 ret = btrfs_commit_super(tree_root);
2063 BUG_ON(ret);
2064 }
2065 }
2066
2067 ret = btrfs_find_orphan_roots(tree_root);
2068 BUG_ON(ret);
2069
2070 if (!(sb->s_flags & MS_RDONLY)) {
2071 ret = btrfs_cleanup_fs_roots(fs_info);
2072 BUG_ON(ret);
2073
2074 ret = btrfs_recover_relocation(tree_root);
2075 if (ret < 0) {
2076 printk(KERN_WARNING
2077 "btrfs: failed to recover relocation\n");
2078 err = -EINVAL;
2079 goto fail_trans_kthread;
2080 }
2081 }
2082
2083 location.objectid = BTRFS_FS_TREE_OBJECTID;
2084 location.type = BTRFS_ROOT_ITEM_KEY;
2085 location.offset = (u64)-1;
2086
2087 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2088 if (!fs_info->fs_root)
2089 goto fail_trans_kthread;
2090 if (IS_ERR(fs_info->fs_root)) {
2091 err = PTR_ERR(fs_info->fs_root);
2092 goto fail_trans_kthread;
2093 }
2094
2095 if (!(sb->s_flags & MS_RDONLY)) {
2096 down_read(&fs_info->cleanup_work_sem);
2097 err = btrfs_orphan_cleanup(fs_info->fs_root);
2098 if (!err)
2099 err = btrfs_orphan_cleanup(fs_info->tree_root);
2100 up_read(&fs_info->cleanup_work_sem);
2101 if (err) {
2102 close_ctree(tree_root);
2103 return ERR_PTR(err);
2104 }
2105 }
2106
2107 return tree_root;
2108
2109 fail_trans_kthread:
2110 kthread_stop(fs_info->transaction_kthread);
2111 fail_cleaner:
2112 kthread_stop(fs_info->cleaner_kthread);
2113
2114 /*
2115 * make sure we're done with the btree inode before we stop our
2116 * kthreads
2117 */
2118 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2119 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2120
2121 fail_block_groups:
2122 btrfs_free_block_groups(fs_info);
2123 free_extent_buffer(csum_root->node);
2124 free_extent_buffer(csum_root->commit_root);
2125 fail_dev_root:
2126 free_extent_buffer(dev_root->node);
2127 free_extent_buffer(dev_root->commit_root);
2128 fail_extent_root:
2129 free_extent_buffer(extent_root->node);
2130 free_extent_buffer(extent_root->commit_root);
2131 fail_tree_root:
2132 free_extent_buffer(tree_root->node);
2133 free_extent_buffer(tree_root->commit_root);
2134 fail_chunk_root:
2135 free_extent_buffer(chunk_root->node);
2136 free_extent_buffer(chunk_root->commit_root);
2137 fail_sb_buffer:
2138 btrfs_stop_workers(&fs_info->generic_worker);
2139 btrfs_stop_workers(&fs_info->fixup_workers);
2140 btrfs_stop_workers(&fs_info->delalloc_workers);
2141 btrfs_stop_workers(&fs_info->workers);
2142 btrfs_stop_workers(&fs_info->endio_workers);
2143 btrfs_stop_workers(&fs_info->endio_meta_workers);
2144 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2145 btrfs_stop_workers(&fs_info->endio_write_workers);
2146 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2147 btrfs_stop_workers(&fs_info->submit_workers);
2148 btrfs_stop_workers(&fs_info->delayed_workers);
2149 btrfs_stop_workers(&fs_info->caching_workers);
2150 fail_alloc:
2151 kfree(fs_info->delayed_root);
2152 fail_iput:
2153 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2154 iput(fs_info->btree_inode);
2155
2156 btrfs_close_devices(fs_info->fs_devices);
2157 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2158 fail_bdi:
2159 bdi_destroy(&fs_info->bdi);
2160 fail_srcu:
2161 cleanup_srcu_struct(&fs_info->subvol_srcu);
2162 fail:
2163 kfree(extent_root);
2164 kfree(tree_root);
2165 kfree(fs_info);
2166 kfree(chunk_root);
2167 kfree(dev_root);
2168 kfree(csum_root);
2169 return ERR_PTR(err);
2170 }
2171
2172 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2173 {
2174 char b[BDEVNAME_SIZE];
2175
2176 if (uptodate) {
2177 set_buffer_uptodate(bh);
2178 } else {
2179 printk_ratelimited(KERN_WARNING "lost page write due to "
2180 "I/O error on %s\n",
2181 bdevname(bh->b_bdev, b));
2182 /* note, we dont' set_buffer_write_io_error because we have
2183 * our own ways of dealing with the IO errors
2184 */
2185 clear_buffer_uptodate(bh);
2186 }
2187 unlock_buffer(bh);
2188 put_bh(bh);
2189 }
2190
2191 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2192 {
2193 struct buffer_head *bh;
2194 struct buffer_head *latest = NULL;
2195 struct btrfs_super_block *super;
2196 int i;
2197 u64 transid = 0;
2198 u64 bytenr;
2199
2200 /* we would like to check all the supers, but that would make
2201 * a btrfs mount succeed after a mkfs from a different FS.
2202 * So, we need to add a special mount option to scan for
2203 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2204 */
2205 for (i = 0; i < 1; i++) {
2206 bytenr = btrfs_sb_offset(i);
2207 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2208 break;
2209 bh = __bread(bdev, bytenr / 4096, 4096);
2210 if (!bh)
2211 continue;
2212
2213 super = (struct btrfs_super_block *)bh->b_data;
2214 if (btrfs_super_bytenr(super) != bytenr ||
2215 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2216 sizeof(super->magic))) {
2217 brelse(bh);
2218 continue;
2219 }
2220
2221 if (!latest || btrfs_super_generation(super) > transid) {
2222 brelse(latest);
2223 latest = bh;
2224 transid = btrfs_super_generation(super);
2225 } else {
2226 brelse(bh);
2227 }
2228 }
2229 return latest;
2230 }
2231
2232 /*
2233 * this should be called twice, once with wait == 0 and
2234 * once with wait == 1. When wait == 0 is done, all the buffer heads
2235 * we write are pinned.
2236 *
2237 * They are released when wait == 1 is done.
2238 * max_mirrors must be the same for both runs, and it indicates how
2239 * many supers on this one device should be written.
2240 *
2241 * max_mirrors == 0 means to write them all.
2242 */
2243 static int write_dev_supers(struct btrfs_device *device,
2244 struct btrfs_super_block *sb,
2245 int do_barriers, int wait, int max_mirrors)
2246 {
2247 struct buffer_head *bh;
2248 int i;
2249 int ret;
2250 int errors = 0;
2251 u32 crc;
2252 u64 bytenr;
2253 int last_barrier = 0;
2254
2255 if (max_mirrors == 0)
2256 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2257
2258 /* make sure only the last submit_bh does a barrier */
2259 if (do_barriers) {
2260 for (i = 0; i < max_mirrors; i++) {
2261 bytenr = btrfs_sb_offset(i);
2262 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2263 device->total_bytes)
2264 break;
2265 last_barrier = i;
2266 }
2267 }
2268
2269 for (i = 0; i < max_mirrors; i++) {
2270 bytenr = btrfs_sb_offset(i);
2271 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2272 break;
2273
2274 if (wait) {
2275 bh = __find_get_block(device->bdev, bytenr / 4096,
2276 BTRFS_SUPER_INFO_SIZE);
2277 BUG_ON(!bh);
2278 wait_on_buffer(bh);
2279 if (!buffer_uptodate(bh))
2280 errors++;
2281
2282 /* drop our reference */
2283 brelse(bh);
2284
2285 /* drop the reference from the wait == 0 run */
2286 brelse(bh);
2287 continue;
2288 } else {
2289 btrfs_set_super_bytenr(sb, bytenr);
2290
2291 crc = ~(u32)0;
2292 crc = btrfs_csum_data(NULL, (char *)sb +
2293 BTRFS_CSUM_SIZE, crc,
2294 BTRFS_SUPER_INFO_SIZE -
2295 BTRFS_CSUM_SIZE);
2296 btrfs_csum_final(crc, sb->csum);
2297
2298 /*
2299 * one reference for us, and we leave it for the
2300 * caller
2301 */
2302 bh = __getblk(device->bdev, bytenr / 4096,
2303 BTRFS_SUPER_INFO_SIZE);
2304 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2305
2306 /* one reference for submit_bh */
2307 get_bh(bh);
2308
2309 set_buffer_uptodate(bh);
2310 lock_buffer(bh);
2311 bh->b_end_io = btrfs_end_buffer_write_sync;
2312 }
2313
2314 if (i == last_barrier && do_barriers)
2315 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2316 else
2317 ret = submit_bh(WRITE_SYNC, bh);
2318
2319 if (ret)
2320 errors++;
2321 }
2322 return errors < i ? 0 : -1;
2323 }
2324
2325 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2326 {
2327 struct list_head *head;
2328 struct btrfs_device *dev;
2329 struct btrfs_super_block *sb;
2330 struct btrfs_dev_item *dev_item;
2331 int ret;
2332 int do_barriers;
2333 int max_errors;
2334 int total_errors = 0;
2335 u64 flags;
2336
2337 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2338 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2339
2340 sb = &root->fs_info->super_for_commit;
2341 dev_item = &sb->dev_item;
2342
2343 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2344 head = &root->fs_info->fs_devices->devices;
2345 list_for_each_entry_rcu(dev, head, dev_list) {
2346 if (!dev->bdev) {
2347 total_errors++;
2348 continue;
2349 }
2350 if (!dev->in_fs_metadata || !dev->writeable)
2351 continue;
2352
2353 btrfs_set_stack_device_generation(dev_item, 0);
2354 btrfs_set_stack_device_type(dev_item, dev->type);
2355 btrfs_set_stack_device_id(dev_item, dev->devid);
2356 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2357 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2358 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2359 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2360 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2361 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2362 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2363
2364 flags = btrfs_super_flags(sb);
2365 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2366
2367 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2368 if (ret)
2369 total_errors++;
2370 }
2371 if (total_errors > max_errors) {
2372 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2373 total_errors);
2374 BUG();
2375 }
2376
2377 total_errors = 0;
2378 list_for_each_entry_rcu(dev, head, dev_list) {
2379 if (!dev->bdev)
2380 continue;
2381 if (!dev->in_fs_metadata || !dev->writeable)
2382 continue;
2383
2384 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2385 if (ret)
2386 total_errors++;
2387 }
2388 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2389 if (total_errors > max_errors) {
2390 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2391 total_errors);
2392 BUG();
2393 }
2394 return 0;
2395 }
2396
2397 int write_ctree_super(struct btrfs_trans_handle *trans,
2398 struct btrfs_root *root, int max_mirrors)
2399 {
2400 int ret;
2401
2402 ret = write_all_supers(root, max_mirrors);
2403 return ret;
2404 }
2405
2406 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2407 {
2408 spin_lock(&fs_info->fs_roots_radix_lock);
2409 radix_tree_delete(&fs_info->fs_roots_radix,
2410 (unsigned long)root->root_key.objectid);
2411 spin_unlock(&fs_info->fs_roots_radix_lock);
2412
2413 if (btrfs_root_refs(&root->root_item) == 0)
2414 synchronize_srcu(&fs_info->subvol_srcu);
2415
2416 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2417 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2418 free_fs_root(root);
2419 return 0;
2420 }
2421
2422 static void free_fs_root(struct btrfs_root *root)
2423 {
2424 iput(root->cache_inode);
2425 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2426 if (root->anon_dev)
2427 free_anon_bdev(root->anon_dev);
2428 free_extent_buffer(root->node);
2429 free_extent_buffer(root->commit_root);
2430 kfree(root->free_ino_ctl);
2431 kfree(root->free_ino_pinned);
2432 kfree(root->name);
2433 kfree(root);
2434 }
2435
2436 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2437 {
2438 int ret;
2439 struct btrfs_root *gang[8];
2440 int i;
2441
2442 while (!list_empty(&fs_info->dead_roots)) {
2443 gang[0] = list_entry(fs_info->dead_roots.next,
2444 struct btrfs_root, root_list);
2445 list_del(&gang[0]->root_list);
2446
2447 if (gang[0]->in_radix) {
2448 btrfs_free_fs_root(fs_info, gang[0]);
2449 } else {
2450 free_extent_buffer(gang[0]->node);
2451 free_extent_buffer(gang[0]->commit_root);
2452 kfree(gang[0]);
2453 }
2454 }
2455
2456 while (1) {
2457 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2458 (void **)gang, 0,
2459 ARRAY_SIZE(gang));
2460 if (!ret)
2461 break;
2462 for (i = 0; i < ret; i++)
2463 btrfs_free_fs_root(fs_info, gang[i]);
2464 }
2465 return 0;
2466 }
2467
2468 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2469 {
2470 u64 root_objectid = 0;
2471 struct btrfs_root *gang[8];
2472 int i;
2473 int ret;
2474
2475 while (1) {
2476 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2477 (void **)gang, root_objectid,
2478 ARRAY_SIZE(gang));
2479 if (!ret)
2480 break;
2481
2482 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2483 for (i = 0; i < ret; i++) {
2484 int err;
2485
2486 root_objectid = gang[i]->root_key.objectid;
2487 err = btrfs_orphan_cleanup(gang[i]);
2488 if (err)
2489 return err;
2490 }
2491 root_objectid++;
2492 }
2493 return 0;
2494 }
2495
2496 int btrfs_commit_super(struct btrfs_root *root)
2497 {
2498 struct btrfs_trans_handle *trans;
2499 int ret;
2500
2501 mutex_lock(&root->fs_info->cleaner_mutex);
2502 btrfs_run_delayed_iputs(root);
2503 btrfs_clean_old_snapshots(root);
2504 mutex_unlock(&root->fs_info->cleaner_mutex);
2505
2506 /* wait until ongoing cleanup work done */
2507 down_write(&root->fs_info->cleanup_work_sem);
2508 up_write(&root->fs_info->cleanup_work_sem);
2509
2510 trans = btrfs_join_transaction(root);
2511 if (IS_ERR(trans))
2512 return PTR_ERR(trans);
2513 ret = btrfs_commit_transaction(trans, root);
2514 BUG_ON(ret);
2515 /* run commit again to drop the original snapshot */
2516 trans = btrfs_join_transaction(root);
2517 if (IS_ERR(trans))
2518 return PTR_ERR(trans);
2519 btrfs_commit_transaction(trans, root);
2520 ret = btrfs_write_and_wait_transaction(NULL, root);
2521 BUG_ON(ret);
2522
2523 ret = write_ctree_super(NULL, root, 0);
2524 return ret;
2525 }
2526
2527 int close_ctree(struct btrfs_root *root)
2528 {
2529 struct btrfs_fs_info *fs_info = root->fs_info;
2530 int ret;
2531
2532 fs_info->closing = 1;
2533 smp_mb();
2534
2535 btrfs_scrub_cancel(root);
2536
2537 /* wait for any defraggers to finish */
2538 wait_event(fs_info->transaction_wait,
2539 (atomic_read(&fs_info->defrag_running) == 0));
2540
2541 /* clear out the rbtree of defraggable inodes */
2542 btrfs_run_defrag_inodes(root->fs_info);
2543
2544 btrfs_put_block_group_cache(fs_info);
2545
2546 /*
2547 * Here come 2 situations when btrfs is broken to flip readonly:
2548 *
2549 * 1. when btrfs flips readonly somewhere else before
2550 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2551 * and btrfs will skip to write sb directly to keep
2552 * ERROR state on disk.
2553 *
2554 * 2. when btrfs flips readonly just in btrfs_commit_super,
2555 * and in such case, btrfs cannot write sb via btrfs_commit_super,
2556 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2557 * btrfs will cleanup all FS resources first and write sb then.
2558 */
2559 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2560 ret = btrfs_commit_super(root);
2561 if (ret)
2562 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2563 }
2564
2565 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2566 ret = btrfs_error_commit_super(root);
2567 if (ret)
2568 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2569 }
2570
2571 kthread_stop(root->fs_info->transaction_kthread);
2572 kthread_stop(root->fs_info->cleaner_kthread);
2573
2574 fs_info->closing = 2;
2575 smp_mb();
2576
2577 if (fs_info->delalloc_bytes) {
2578 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2579 (unsigned long long)fs_info->delalloc_bytes);
2580 }
2581 if (fs_info->total_ref_cache_size) {
2582 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2583 (unsigned long long)fs_info->total_ref_cache_size);
2584 }
2585
2586 free_extent_buffer(fs_info->extent_root->node);
2587 free_extent_buffer(fs_info->extent_root->commit_root);
2588 free_extent_buffer(fs_info->tree_root->node);
2589 free_extent_buffer(fs_info->tree_root->commit_root);
2590 free_extent_buffer(root->fs_info->chunk_root->node);
2591 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2592 free_extent_buffer(root->fs_info->dev_root->node);
2593 free_extent_buffer(root->fs_info->dev_root->commit_root);
2594 free_extent_buffer(root->fs_info->csum_root->node);
2595 free_extent_buffer(root->fs_info->csum_root->commit_root);
2596
2597 btrfs_free_block_groups(root->fs_info);
2598
2599 del_fs_roots(fs_info);
2600
2601 iput(fs_info->btree_inode);
2602 kfree(fs_info->delayed_root);
2603
2604 btrfs_stop_workers(&fs_info->generic_worker);
2605 btrfs_stop_workers(&fs_info->fixup_workers);
2606 btrfs_stop_workers(&fs_info->delalloc_workers);
2607 btrfs_stop_workers(&fs_info->workers);
2608 btrfs_stop_workers(&fs_info->endio_workers);
2609 btrfs_stop_workers(&fs_info->endio_meta_workers);
2610 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2611 btrfs_stop_workers(&fs_info->endio_write_workers);
2612 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2613 btrfs_stop_workers(&fs_info->submit_workers);
2614 btrfs_stop_workers(&fs_info->delayed_workers);
2615 btrfs_stop_workers(&fs_info->caching_workers);
2616
2617 btrfs_close_devices(fs_info->fs_devices);
2618 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2619
2620 bdi_destroy(&fs_info->bdi);
2621 cleanup_srcu_struct(&fs_info->subvol_srcu);
2622
2623 kfree(fs_info->extent_root);
2624 kfree(fs_info->tree_root);
2625 kfree(fs_info->chunk_root);
2626 kfree(fs_info->dev_root);
2627 kfree(fs_info->csum_root);
2628 kfree(fs_info);
2629
2630 return 0;
2631 }
2632
2633 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2634 {
2635 int ret;
2636 struct inode *btree_inode = buf->first_page->mapping->host;
2637
2638 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2639 NULL);
2640 if (!ret)
2641 return ret;
2642
2643 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2644 parent_transid);
2645 return !ret;
2646 }
2647
2648 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2649 {
2650 struct inode *btree_inode = buf->first_page->mapping->host;
2651 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2652 buf);
2653 }
2654
2655 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2656 {
2657 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2658 u64 transid = btrfs_header_generation(buf);
2659 struct inode *btree_inode = root->fs_info->btree_inode;
2660 int was_dirty;
2661
2662 btrfs_assert_tree_locked(buf);
2663 if (transid != root->fs_info->generation) {
2664 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2665 "found %llu running %llu\n",
2666 (unsigned long long)buf->start,
2667 (unsigned long long)transid,
2668 (unsigned long long)root->fs_info->generation);
2669 WARN_ON(1);
2670 }
2671 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2672 buf);
2673 if (!was_dirty) {
2674 spin_lock(&root->fs_info->delalloc_lock);
2675 root->fs_info->dirty_metadata_bytes += buf->len;
2676 spin_unlock(&root->fs_info->delalloc_lock);
2677 }
2678 }
2679
2680 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2681 {
2682 /*
2683 * looks as though older kernels can get into trouble with
2684 * this code, they end up stuck in balance_dirty_pages forever
2685 */
2686 u64 num_dirty;
2687 unsigned long thresh = 32 * 1024 * 1024;
2688
2689 if (current->flags & PF_MEMALLOC)
2690 return;
2691
2692 btrfs_balance_delayed_items(root);
2693
2694 num_dirty = root->fs_info->dirty_metadata_bytes;
2695
2696 if (num_dirty > thresh) {
2697 balance_dirty_pages_ratelimited_nr(
2698 root->fs_info->btree_inode->i_mapping, 1);
2699 }
2700 return;
2701 }
2702
2703 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2704 {
2705 /*
2706 * looks as though older kernels can get into trouble with
2707 * this code, they end up stuck in balance_dirty_pages forever
2708 */
2709 u64 num_dirty;
2710 unsigned long thresh = 32 * 1024 * 1024;
2711
2712 if (current->flags & PF_MEMALLOC)
2713 return;
2714
2715 num_dirty = root->fs_info->dirty_metadata_bytes;
2716
2717 if (num_dirty > thresh) {
2718 balance_dirty_pages_ratelimited_nr(
2719 root->fs_info->btree_inode->i_mapping, 1);
2720 }
2721 return;
2722 }
2723
2724 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2725 {
2726 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2727 int ret;
2728 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2729 if (ret == 0)
2730 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2731 return ret;
2732 }
2733
2734 int btree_lock_page_hook(struct page *page)
2735 {
2736 struct inode *inode = page->mapping->host;
2737 struct btrfs_root *root = BTRFS_I(inode)->root;
2738 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2739 struct extent_buffer *eb;
2740 unsigned long len;
2741 u64 bytenr = page_offset(page);
2742
2743 if (page->private == EXTENT_PAGE_PRIVATE)
2744 goto out;
2745
2746 len = page->private >> 2;
2747 eb = find_extent_buffer(io_tree, bytenr, len);
2748 if (!eb)
2749 goto out;
2750
2751 btrfs_tree_lock(eb);
2752 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2753
2754 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2755 spin_lock(&root->fs_info->delalloc_lock);
2756 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2757 root->fs_info->dirty_metadata_bytes -= eb->len;
2758 else
2759 WARN_ON(1);
2760 spin_unlock(&root->fs_info->delalloc_lock);
2761 }
2762
2763 btrfs_tree_unlock(eb);
2764 free_extent_buffer(eb);
2765 out:
2766 lock_page(page);
2767 return 0;
2768 }
2769
2770 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2771 int read_only)
2772 {
2773 if (read_only)
2774 return;
2775
2776 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2777 printk(KERN_WARNING "warning: mount fs with errors, "
2778 "running btrfsck is recommended\n");
2779 }
2780
2781 int btrfs_error_commit_super(struct btrfs_root *root)
2782 {
2783 int ret;
2784
2785 mutex_lock(&root->fs_info->cleaner_mutex);
2786 btrfs_run_delayed_iputs(root);
2787 mutex_unlock(&root->fs_info->cleaner_mutex);
2788
2789 down_write(&root->fs_info->cleanup_work_sem);
2790 up_write(&root->fs_info->cleanup_work_sem);
2791
2792 /* cleanup FS via transaction */
2793 btrfs_cleanup_transaction(root);
2794
2795 ret = write_ctree_super(NULL, root, 0);
2796
2797 return ret;
2798 }
2799
2800 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2801 {
2802 struct btrfs_inode *btrfs_inode;
2803 struct list_head splice;
2804
2805 INIT_LIST_HEAD(&splice);
2806
2807 mutex_lock(&root->fs_info->ordered_operations_mutex);
2808 spin_lock(&root->fs_info->ordered_extent_lock);
2809
2810 list_splice_init(&root->fs_info->ordered_operations, &splice);
2811 while (!list_empty(&splice)) {
2812 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2813 ordered_operations);
2814
2815 list_del_init(&btrfs_inode->ordered_operations);
2816
2817 btrfs_invalidate_inodes(btrfs_inode->root);
2818 }
2819
2820 spin_unlock(&root->fs_info->ordered_extent_lock);
2821 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2822
2823 return 0;
2824 }
2825
2826 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2827 {
2828 struct list_head splice;
2829 struct btrfs_ordered_extent *ordered;
2830 struct inode *inode;
2831
2832 INIT_LIST_HEAD(&splice);
2833
2834 spin_lock(&root->fs_info->ordered_extent_lock);
2835
2836 list_splice_init(&root->fs_info->ordered_extents, &splice);
2837 while (!list_empty(&splice)) {
2838 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2839 root_extent_list);
2840
2841 list_del_init(&ordered->root_extent_list);
2842 atomic_inc(&ordered->refs);
2843
2844 /* the inode may be getting freed (in sys_unlink path). */
2845 inode = igrab(ordered->inode);
2846
2847 spin_unlock(&root->fs_info->ordered_extent_lock);
2848 if (inode)
2849 iput(inode);
2850
2851 atomic_set(&ordered->refs, 1);
2852 btrfs_put_ordered_extent(ordered);
2853
2854 spin_lock(&root->fs_info->ordered_extent_lock);
2855 }
2856
2857 spin_unlock(&root->fs_info->ordered_extent_lock);
2858
2859 return 0;
2860 }
2861
2862 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2863 struct btrfs_root *root)
2864 {
2865 struct rb_node *node;
2866 struct btrfs_delayed_ref_root *delayed_refs;
2867 struct btrfs_delayed_ref_node *ref;
2868 int ret = 0;
2869
2870 delayed_refs = &trans->delayed_refs;
2871
2872 spin_lock(&delayed_refs->lock);
2873 if (delayed_refs->num_entries == 0) {
2874 spin_unlock(&delayed_refs->lock);
2875 printk(KERN_INFO "delayed_refs has NO entry\n");
2876 return ret;
2877 }
2878
2879 node = rb_first(&delayed_refs->root);
2880 while (node) {
2881 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2882 node = rb_next(node);
2883
2884 ref->in_tree = 0;
2885 rb_erase(&ref->rb_node, &delayed_refs->root);
2886 delayed_refs->num_entries--;
2887
2888 atomic_set(&ref->refs, 1);
2889 if (btrfs_delayed_ref_is_head(ref)) {
2890 struct btrfs_delayed_ref_head *head;
2891
2892 head = btrfs_delayed_node_to_head(ref);
2893 mutex_lock(&head->mutex);
2894 kfree(head->extent_op);
2895 delayed_refs->num_heads--;
2896 if (list_empty(&head->cluster))
2897 delayed_refs->num_heads_ready--;
2898 list_del_init(&head->cluster);
2899 mutex_unlock(&head->mutex);
2900 }
2901
2902 spin_unlock(&delayed_refs->lock);
2903 btrfs_put_delayed_ref(ref);
2904
2905 cond_resched();
2906 spin_lock(&delayed_refs->lock);
2907 }
2908
2909 spin_unlock(&delayed_refs->lock);
2910
2911 return ret;
2912 }
2913
2914 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2915 {
2916 struct btrfs_pending_snapshot *snapshot;
2917 struct list_head splice;
2918
2919 INIT_LIST_HEAD(&splice);
2920
2921 list_splice_init(&t->pending_snapshots, &splice);
2922
2923 while (!list_empty(&splice)) {
2924 snapshot = list_entry(splice.next,
2925 struct btrfs_pending_snapshot,
2926 list);
2927
2928 list_del_init(&snapshot->list);
2929
2930 kfree(snapshot);
2931 }
2932
2933 return 0;
2934 }
2935
2936 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2937 {
2938 struct btrfs_inode *btrfs_inode;
2939 struct list_head splice;
2940
2941 INIT_LIST_HEAD(&splice);
2942
2943 spin_lock(&root->fs_info->delalloc_lock);
2944 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2945
2946 while (!list_empty(&splice)) {
2947 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2948 delalloc_inodes);
2949
2950 list_del_init(&btrfs_inode->delalloc_inodes);
2951
2952 btrfs_invalidate_inodes(btrfs_inode->root);
2953 }
2954
2955 spin_unlock(&root->fs_info->delalloc_lock);
2956
2957 return 0;
2958 }
2959
2960 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2961 struct extent_io_tree *dirty_pages,
2962 int mark)
2963 {
2964 int ret;
2965 struct page *page;
2966 struct inode *btree_inode = root->fs_info->btree_inode;
2967 struct extent_buffer *eb;
2968 u64 start = 0;
2969 u64 end;
2970 u64 offset;
2971 unsigned long index;
2972
2973 while (1) {
2974 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2975 mark);
2976 if (ret)
2977 break;
2978
2979 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2980 while (start <= end) {
2981 index = start >> PAGE_CACHE_SHIFT;
2982 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2983 page = find_get_page(btree_inode->i_mapping, index);
2984 if (!page)
2985 continue;
2986 offset = page_offset(page);
2987
2988 spin_lock(&dirty_pages->buffer_lock);
2989 eb = radix_tree_lookup(
2990 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2991 offset >> PAGE_CACHE_SHIFT);
2992 spin_unlock(&dirty_pages->buffer_lock);
2993 if (eb) {
2994 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
2995 &eb->bflags);
2996 atomic_set(&eb->refs, 1);
2997 }
2998 if (PageWriteback(page))
2999 end_page_writeback(page);
3000
3001 lock_page(page);
3002 if (PageDirty(page)) {
3003 clear_page_dirty_for_io(page);
3004 spin_lock_irq(&page->mapping->tree_lock);
3005 radix_tree_tag_clear(&page->mapping->page_tree,
3006 page_index(page),
3007 PAGECACHE_TAG_DIRTY);
3008 spin_unlock_irq(&page->mapping->tree_lock);
3009 }
3010
3011 page->mapping->a_ops->invalidatepage(page, 0);
3012 unlock_page(page);
3013 }
3014 }
3015
3016 return ret;
3017 }
3018
3019 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3020 struct extent_io_tree *pinned_extents)
3021 {
3022 struct extent_io_tree *unpin;
3023 u64 start;
3024 u64 end;
3025 int ret;
3026
3027 unpin = pinned_extents;
3028 while (1) {
3029 ret = find_first_extent_bit(unpin, 0, &start, &end,
3030 EXTENT_DIRTY);
3031 if (ret)
3032 break;
3033
3034 /* opt_discard */
3035 if (btrfs_test_opt(root, DISCARD))
3036 ret = btrfs_error_discard_extent(root, start,
3037 end + 1 - start,
3038 NULL);
3039
3040 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3041 btrfs_error_unpin_extent_range(root, start, end);
3042 cond_resched();
3043 }
3044
3045 return 0;
3046 }
3047
3048 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3049 {
3050 struct btrfs_transaction *t;
3051 LIST_HEAD(list);
3052
3053 WARN_ON(1);
3054
3055 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3056
3057 spin_lock(&root->fs_info->trans_lock);
3058 list_splice_init(&root->fs_info->trans_list, &list);
3059 root->fs_info->trans_no_join = 1;
3060 spin_unlock(&root->fs_info->trans_lock);
3061
3062 while (!list_empty(&list)) {
3063 t = list_entry(list.next, struct btrfs_transaction, list);
3064 if (!t)
3065 break;
3066
3067 btrfs_destroy_ordered_operations(root);
3068
3069 btrfs_destroy_ordered_extents(root);
3070
3071 btrfs_destroy_delayed_refs(t, root);
3072
3073 btrfs_block_rsv_release(root,
3074 &root->fs_info->trans_block_rsv,
3075 t->dirty_pages.dirty_bytes);
3076
3077 /* FIXME: cleanup wait for commit */
3078 t->in_commit = 1;
3079 t->blocked = 1;
3080 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3081 wake_up(&root->fs_info->transaction_blocked_wait);
3082
3083 t->blocked = 0;
3084 if (waitqueue_active(&root->fs_info->transaction_wait))
3085 wake_up(&root->fs_info->transaction_wait);
3086
3087 t->commit_done = 1;
3088 if (waitqueue_active(&t->commit_wait))
3089 wake_up(&t->commit_wait);
3090
3091 btrfs_destroy_pending_snapshots(t);
3092
3093 btrfs_destroy_delalloc_inodes(root);
3094
3095 spin_lock(&root->fs_info->trans_lock);
3096 root->fs_info->running_transaction = NULL;
3097 spin_unlock(&root->fs_info->trans_lock);
3098
3099 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3100 EXTENT_DIRTY);
3101
3102 btrfs_destroy_pinned_extent(root,
3103 root->fs_info->pinned_extents);
3104
3105 atomic_set(&t->use_count, 0);
3106 list_del_init(&t->list);
3107 memset(t, 0, sizeof(*t));
3108 kmem_cache_free(btrfs_transaction_cachep, t);
3109 }
3110
3111 spin_lock(&root->fs_info->trans_lock);
3112 root->fs_info->trans_no_join = 0;
3113 spin_unlock(&root->fs_info->trans_lock);
3114 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3115
3116 return 0;
3117 }
3118
3119 static struct extent_io_ops btree_extent_io_ops = {
3120 .write_cache_pages_lock_hook = btree_lock_page_hook,
3121 .readpage_end_io_hook = btree_readpage_end_io_hook,
3122 .submit_bio_hook = btree_submit_bio_hook,
3123 /* note we're sharing with inode.c for the merge bio hook */
3124 .merge_bio_hook = btrfs_merge_bio_hook,
3125 };
linux-next