?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs: S_ISREG(mode) is not mode & S_IFREG...
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / btrfs / disk-io.c
blobb231ae13b2697ebcd5b88d848a81ddc83e01c990
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 /* These are used to set the lockdep class on the extent buffer locks.
104 * The class is set by the readpage_end_io_hook after the buffer has
105 * passed csum validation but before the pages are unlocked.
106 *
107 * The lockdep class is also set by btrfs_init_new_buffer on freshly
108 * allocated blocks.
109 *
110 * The class is based on the level in the tree block, which allows lockdep
111 * to know that lower nodes nest inside the locks of higher nodes.
112 *
113 * We also add a check to make sure the highest level of the tree is
114 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
115 * code needs update as well.
116 */
117 #ifdef CONFIG_DEBUG_LOCK_ALLOC
118 # if BTRFS_MAX_LEVEL != 8
119 # error
120 # endif
121 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
122 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
123 /* leaf */
124 "btrfs-extent-00",
125 "btrfs-extent-01",
126 "btrfs-extent-02",
127 "btrfs-extent-03",
128 "btrfs-extent-04",
129 "btrfs-extent-05",
130 "btrfs-extent-06",
131 "btrfs-extent-07",
132 /* highest possible level */
133 "btrfs-extent-08",
134 };
135 #endif
136
137 /*
138 * extents on the btree inode are pretty simple, there's one extent
139 * that covers the entire device
140 */
141 static struct extent_map *btree_get_extent(struct inode *inode,
142 struct page *page, size_t pg_offset, u64 start, u64 len,
143 int create)
144 {
145 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
146 struct extent_map *em;
147 int ret;
148
149 read_lock(&em_tree->lock);
150 em = lookup_extent_mapping(em_tree, start, len);
151 if (em) {
152 em->bdev =
153 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
154 read_unlock(&em_tree->lock);
155 goto out;
156 }
157 read_unlock(&em_tree->lock);
158
159 em = alloc_extent_map();
160 if (!em) {
161 em = ERR_PTR(-ENOMEM);
162 goto out;
163 }
164 em->start = 0;
165 em->len = (u64)-1;
166 em->block_len = (u64)-1;
167 em->block_start = 0;
168 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
169
170 write_lock(&em_tree->lock);
171 ret = add_extent_mapping(em_tree, em);
172 if (ret == -EEXIST) {
173 u64 failed_start = em->start;
174 u64 failed_len = em->len;
175
176 free_extent_map(em);
177 em = lookup_extent_mapping(em_tree, start, len);
178 if (em) {
179 ret = 0;
180 } else {
181 em = lookup_extent_mapping(em_tree, failed_start,
182 failed_len);
183 ret = -EIO;
184 }
185 } else if (ret) {
186 free_extent_map(em);
187 em = NULL;
188 }
189 write_unlock(&em_tree->lock);
190
191 if (ret)
192 em = ERR_PTR(ret);
193 out:
194 return em;
195 }
196
197 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
198 {
199 return crc32c(seed, data, len);
200 }
201
202 void btrfs_csum_final(u32 crc, char *result)
203 {
204 put_unaligned_le32(~crc, result);
205 }
206
207 /*
208 * compute the csum for a btree block, and either verify it or write it
209 * into the csum field of the block.
210 */
211 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
212 int verify)
213 {
214 u16 csum_size =
215 btrfs_super_csum_size(&root->fs_info->super_copy);
216 char *result = NULL;
217 unsigned long len;
218 unsigned long cur_len;
219 unsigned long offset = BTRFS_CSUM_SIZE;
220 char *map_token = NULL;
221 char *kaddr;
222 unsigned long map_start;
223 unsigned long map_len;
224 int err;
225 u32 crc = ~(u32)0;
226 unsigned long inline_result;
227
228 len = buf->len - offset;
229 while (len > 0) {
230 err = map_private_extent_buffer(buf, offset, 32,
231 &map_token, &kaddr,
232 &map_start, &map_len, KM_USER0);
233 if (err)
234 return 1;
235 cur_len = min(len, map_len - (offset - map_start));
236 crc = btrfs_csum_data(root, kaddr + offset - map_start,
237 crc, cur_len);
238 len -= cur_len;
239 offset += cur_len;
240 unmap_extent_buffer(buf, map_token, KM_USER0);
241 }
242 if (csum_size > sizeof(inline_result)) {
243 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
244 if (!result)
245 return 1;
246 } else {
247 result = (char *)&inline_result;
248 }
249
250 btrfs_csum_final(crc, result);
251
252 if (verify) {
253 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
254 u32 val;
255 u32 found = 0;
256 memcpy(&found, result, csum_size);
257
258 read_extent_buffer(buf, &val, 0, csum_size);
259 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
260 "failed on %llu wanted %X found %X "
261 "level %d\n",
262 root->fs_info->sb->s_id,
263 (unsigned long long)buf->start, val, found,
264 btrfs_header_level(buf));
265 if (result != (char *)&inline_result)
266 kfree(result);
267 return 1;
268 }
269 } else {
270 write_extent_buffer(buf, result, 0, csum_size);
271 }
272 if (result != (char *)&inline_result)
273 kfree(result);
274 return 0;
275 }
276
277 /*
278 * we can't consider a given block up to date unless the transid of the
279 * block matches the transid in the parent node's pointer. This is how we
280 * detect blocks that either didn't get written at all or got written
281 * in the wrong place.
282 */
283 static int verify_parent_transid(struct extent_io_tree *io_tree,
284 struct extent_buffer *eb, u64 parent_transid)
285 {
286 struct extent_state *cached_state = NULL;
287 int ret;
288
289 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
290 return 0;
291
292 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
293 0, &cached_state, GFP_NOFS);
294 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
295 btrfs_header_generation(eb) == parent_transid) {
296 ret = 0;
297 goto out;
298 }
299 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
300 "found %llu\n",
301 (unsigned long long)eb->start,
302 (unsigned long long)parent_transid,
303 (unsigned long long)btrfs_header_generation(eb));
304 ret = 1;
305 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
306 out:
307 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
308 &cached_state, GFP_NOFS);
309 return ret;
310 }
311
312 /*
313 * helper to read a given tree block, doing retries as required when
314 * the checksums don't match and we have alternate mirrors to try.
315 */
316 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
317 struct extent_buffer *eb,
318 u64 start, u64 parent_transid)
319 {
320 struct extent_io_tree *io_tree;
321 int ret;
322 int num_copies = 0;
323 int mirror_num = 0;
324
325 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
326 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
327 while (1) {
328 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
329 btree_get_extent, mirror_num);
330 if (!ret &&
331 !verify_parent_transid(io_tree, eb, parent_transid))
332 return ret;
333
334 /*
335 * This buffer's crc is fine, but its contents are corrupted, so
336 * there is no reason to read the other copies, they won't be
337 * any less wrong.
338 */
339 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
340 return ret;
341
342 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
343 eb->start, eb->len);
344 if (num_copies == 1)
345 return ret;
346
347 mirror_num++;
348 if (mirror_num > num_copies)
349 return ret;
350 }
351 return -EIO;
352 }
353
354 /*
355 * checksum a dirty tree block before IO. This has extra checks to make sure
356 * we only fill in the checksum field in the first page of a multi-page block
357 */
358
359 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
360 {
361 struct extent_io_tree *tree;
362 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
363 u64 found_start;
364 unsigned long len;
365 struct extent_buffer *eb;
366 int ret;
367
368 tree = &BTRFS_I(page->mapping->host)->io_tree;
369
370 if (page->private == EXTENT_PAGE_PRIVATE) {
371 WARN_ON(1);
372 goto out;
373 }
374 if (!page->private) {
375 WARN_ON(1);
376 goto out;
377 }
378 len = page->private >> 2;
379 WARN_ON(len == 0);
380
381 eb = alloc_extent_buffer(tree, start, len, page);
382 if (eb == NULL) {
383 WARN_ON(1);
384 goto out;
385 }
386 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
387 btrfs_header_generation(eb));
388 BUG_ON(ret);
389 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
390
391 found_start = btrfs_header_bytenr(eb);
392 if (found_start != start) {
393 WARN_ON(1);
394 goto err;
395 }
396 if (eb->first_page != page) {
397 WARN_ON(1);
398 goto err;
399 }
400 if (!PageUptodate(page)) {
401 WARN_ON(1);
402 goto err;
403 }
404 csum_tree_block(root, eb, 0);
405 err:
406 free_extent_buffer(eb);
407 out:
408 return 0;
409 }
410
411 static int check_tree_block_fsid(struct btrfs_root *root,
412 struct extent_buffer *eb)
413 {
414 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
415 u8 fsid[BTRFS_UUID_SIZE];
416 int ret = 1;
417
418 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
419 BTRFS_FSID_SIZE);
420 while (fs_devices) {
421 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
422 ret = 0;
423 break;
424 }
425 fs_devices = fs_devices->seed;
426 }
427 return ret;
428 }
429
430 #define CORRUPT(reason, eb, root, slot) \
431 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
432 "root=%llu, slot=%d\n", reason, \
433 (unsigned long long)btrfs_header_bytenr(eb), \
434 (unsigned long long)root->objectid, slot)
435
436 static noinline int check_leaf(struct btrfs_root *root,
437 struct extent_buffer *leaf)
438 {
439 struct btrfs_key key;
440 struct btrfs_key leaf_key;
441 u32 nritems = btrfs_header_nritems(leaf);
442 int slot;
443
444 if (nritems == 0)
445 return 0;
446
447 /* Check the 0 item */
448 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
449 BTRFS_LEAF_DATA_SIZE(root)) {
450 CORRUPT("invalid item offset size pair", leaf, root, 0);
451 return -EIO;
452 }
453
454 /*
455 * Check to make sure each items keys are in the correct order and their
456 * offsets make sense. We only have to loop through nritems-1 because
457 * we check the current slot against the next slot, which verifies the
458 * next slot's offset+size makes sense and that the current's slot
459 * offset is correct.
460 */
461 for (slot = 0; slot < nritems - 1; slot++) {
462 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
463 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
464
465 /* Make sure the keys are in the right order */
466 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
467 CORRUPT("bad key order", leaf, root, slot);
468 return -EIO;
469 }
470
471 /*
472 * Make sure the offset and ends are right, remember that the
473 * item data starts at the end of the leaf and grows towards the
474 * front.
475 */
476 if (btrfs_item_offset_nr(leaf, slot) !=
477 btrfs_item_end_nr(leaf, slot + 1)) {
478 CORRUPT("slot offset bad", leaf, root, slot);
479 return -EIO;
480 }
481
482 /*
483 * Check to make sure that we don't point outside of the leaf,
484 * just incase all the items are consistent to eachother, but
485 * all point outside of the leaf.
486 */
487 if (btrfs_item_end_nr(leaf, slot) >
488 BTRFS_LEAF_DATA_SIZE(root)) {
489 CORRUPT("slot end outside of leaf", leaf, root, slot);
490 return -EIO;
491 }
492 }
493
494 return 0;
495 }
496
497 #ifdef CONFIG_DEBUG_LOCK_ALLOC
498 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
499 {
500 lockdep_set_class_and_name(&eb->lock,
501 &btrfs_eb_class[level],
502 btrfs_eb_name[level]);
503 }
504 #endif
505
506 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
507 struct extent_state *state)
508 {
509 struct extent_io_tree *tree;
510 u64 found_start;
511 int found_level;
512 unsigned long len;
513 struct extent_buffer *eb;
514 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
515 int ret = 0;
516
517 tree = &BTRFS_I(page->mapping->host)->io_tree;
518 if (page->private == EXTENT_PAGE_PRIVATE)
519 goto out;
520 if (!page->private)
521 goto out;
522
523 len = page->private >> 2;
524 WARN_ON(len == 0);
525
526 eb = alloc_extent_buffer(tree, start, len, page);
527 if (eb == NULL) {
528 ret = -EIO;
529 goto out;
530 }
531
532 found_start = btrfs_header_bytenr(eb);
533 if (found_start != start) {
534 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
535 "%llu %llu\n",
536 (unsigned long long)found_start,
537 (unsigned long long)eb->start);
538 ret = -EIO;
539 goto err;
540 }
541 if (eb->first_page != page) {
542 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
543 eb->first_page->index, page->index);
544 WARN_ON(1);
545 ret = -EIO;
546 goto err;
547 }
548 if (check_tree_block_fsid(root, eb)) {
549 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
550 (unsigned long long)eb->start);
551 ret = -EIO;
552 goto err;
553 }
554 found_level = btrfs_header_level(eb);
555
556 btrfs_set_buffer_lockdep_class(eb, found_level);
557
558 ret = csum_tree_block(root, eb, 1);
559 if (ret) {
560 ret = -EIO;
561 goto err;
562 }
563
564 /*
565 * If this is a leaf block and it is corrupt, set the corrupt bit so
566 * that we don't try and read the other copies of this block, just
567 * return -EIO.
568 */
569 if (found_level == 0 && check_leaf(root, eb)) {
570 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
571 ret = -EIO;
572 }
573
574 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
575 end = eb->start + end - 1;
576 err:
577 free_extent_buffer(eb);
578 out:
579 return ret;
580 }
581
582 static void end_workqueue_bio(struct bio *bio, int err)
583 {
584 struct end_io_wq *end_io_wq = bio->bi_private;
585 struct btrfs_fs_info *fs_info;
586
587 fs_info = end_io_wq->info;
588 end_io_wq->error = err;
589 end_io_wq->work.func = end_workqueue_fn;
590 end_io_wq->work.flags = 0;
591
592 if (bio->bi_rw & REQ_WRITE) {
593 if (end_io_wq->metadata == 1)
594 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
595 &end_io_wq->work);
596 else if (end_io_wq->metadata == 2)
597 btrfs_queue_worker(&fs_info->endio_freespace_worker,
598 &end_io_wq->work);
599 else
600 btrfs_queue_worker(&fs_info->endio_write_workers,
601 &end_io_wq->work);
602 } else {
603 if (end_io_wq->metadata)
604 btrfs_queue_worker(&fs_info->endio_meta_workers,
605 &end_io_wq->work);
606 else
607 btrfs_queue_worker(&fs_info->endio_workers,
608 &end_io_wq->work);
609 }
610 }
611
612 /*
613 * For the metadata arg you want
614 *
615 * 0 - if data
616 * 1 - if normal metadta
617 * 2 - if writing to the free space cache area
618 */
619 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
620 int metadata)
621 {
622 struct end_io_wq *end_io_wq;
623 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
624 if (!end_io_wq)
625 return -ENOMEM;
626
627 end_io_wq->private = bio->bi_private;
628 end_io_wq->end_io = bio->bi_end_io;
629 end_io_wq->info = info;
630 end_io_wq->error = 0;
631 end_io_wq->bio = bio;
632 end_io_wq->metadata = metadata;
633
634 bio->bi_private = end_io_wq;
635 bio->bi_end_io = end_workqueue_bio;
636 return 0;
637 }
638
639 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
640 {
641 unsigned long limit = min_t(unsigned long,
642 info->workers.max_workers,
643 info->fs_devices->open_devices);
644 return 256 * limit;
645 }
646
647 static void run_one_async_start(struct btrfs_work *work)
648 {
649 struct async_submit_bio *async;
650
651 async = container_of(work, struct async_submit_bio, work);
652 async->submit_bio_start(async->inode, async->rw, async->bio,
653 async->mirror_num, async->bio_flags,
654 async->bio_offset);
655 }
656
657 static void run_one_async_done(struct btrfs_work *work)
658 {
659 struct btrfs_fs_info *fs_info;
660 struct async_submit_bio *async;
661 int limit;
662
663 async = container_of(work, struct async_submit_bio, work);
664 fs_info = BTRFS_I(async->inode)->root->fs_info;
665
666 limit = btrfs_async_submit_limit(fs_info);
667 limit = limit * 2 / 3;
668
669 atomic_dec(&fs_info->nr_async_submits);
670
671 if (atomic_read(&fs_info->nr_async_submits) < limit &&
672 waitqueue_active(&fs_info->async_submit_wait))
673 wake_up(&fs_info->async_submit_wait);
674
675 async->submit_bio_done(async->inode, async->rw, async->bio,
676 async->mirror_num, async->bio_flags,
677 async->bio_offset);
678 }
679
680 static void run_one_async_free(struct btrfs_work *work)
681 {
682 struct async_submit_bio *async;
683
684 async = container_of(work, struct async_submit_bio, work);
685 kfree(async);
686 }
687
688 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
689 int rw, struct bio *bio, int mirror_num,
690 unsigned long bio_flags,
691 u64 bio_offset,
692 extent_submit_bio_hook_t *submit_bio_start,
693 extent_submit_bio_hook_t *submit_bio_done)
694 {
695 struct async_submit_bio *async;
696
697 async = kmalloc(sizeof(*async), GFP_NOFS);
698 if (!async)
699 return -ENOMEM;
700
701 async->inode = inode;
702 async->rw = rw;
703 async->bio = bio;
704 async->mirror_num = mirror_num;
705 async->submit_bio_start = submit_bio_start;
706 async->submit_bio_done = submit_bio_done;
707
708 async->work.func = run_one_async_start;
709 async->work.ordered_func = run_one_async_done;
710 async->work.ordered_free = run_one_async_free;
711
712 async->work.flags = 0;
713 async->bio_flags = bio_flags;
714 async->bio_offset = bio_offset;
715
716 atomic_inc(&fs_info->nr_async_submits);
717
718 if (rw & REQ_SYNC)
719 btrfs_set_work_high_prio(&async->work);
720
721 btrfs_queue_worker(&fs_info->workers, &async->work);
722
723 while (atomic_read(&fs_info->async_submit_draining) &&
724 atomic_read(&fs_info->nr_async_submits)) {
725 wait_event(fs_info->async_submit_wait,
726 (atomic_read(&fs_info->nr_async_submits) == 0));
727 }
728
729 return 0;
730 }
731
732 static int btree_csum_one_bio(struct bio *bio)
733 {
734 struct bio_vec *bvec = bio->bi_io_vec;
735 int bio_index = 0;
736 struct btrfs_root *root;
737
738 WARN_ON(bio->bi_vcnt <= 0);
739 while (bio_index < bio->bi_vcnt) {
740 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
741 csum_dirty_buffer(root, bvec->bv_page);
742 bio_index++;
743 bvec++;
744 }
745 return 0;
746 }
747
748 static int __btree_submit_bio_start(struct inode *inode, int rw,
749 struct bio *bio, int mirror_num,
750 unsigned long bio_flags,
751 u64 bio_offset)
752 {
753 /*
754 * when we're called for a write, we're already in the async
755 * submission context. Just jump into btrfs_map_bio
756 */
757 btree_csum_one_bio(bio);
758 return 0;
759 }
760
761 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
762 int mirror_num, unsigned long bio_flags,
763 u64 bio_offset)
764 {
765 /*
766 * when we're called for a write, we're already in the async
767 * submission context. Just jump into btrfs_map_bio
768 */
769 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
770 }
771
772 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
773 int mirror_num, unsigned long bio_flags,
774 u64 bio_offset)
775 {
776 int ret;
777
778 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
779 bio, 1);
780 BUG_ON(ret);
781
782 if (!(rw & REQ_WRITE)) {
783 /*
784 * called for a read, do the setup so that checksum validation
785 * can happen in the async kernel threads
786 */
787 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
788 mirror_num, 0);
789 }
790
791 /*
792 * kthread helpers are used to submit writes so that checksumming
793 * can happen in parallel across all CPUs
794 */
795 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
796 inode, rw, bio, mirror_num, 0,
797 bio_offset,
798 __btree_submit_bio_start,
799 __btree_submit_bio_done);
800 }
801
802 #ifdef CONFIG_MIGRATION
803 static int btree_migratepage(struct address_space *mapping,
804 struct page *newpage, struct page *page)
805 {
806 /*
807 * we can't safely write a btree page from here,
808 * we haven't done the locking hook
809 */
810 if (PageDirty(page))
811 return -EAGAIN;
812 /*
813 * Buffers may be managed in a filesystem specific way.
814 * We must have no buffers or drop them.
815 */
816 if (page_has_private(page) &&
817 !try_to_release_page(page, GFP_KERNEL))
818 return -EAGAIN;
819 return migrate_page(mapping, newpage, page);
820 }
821 #endif
822
823 static int btree_writepage(struct page *page, struct writeback_control *wbc)
824 {
825 struct extent_io_tree *tree;
826 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
827 struct extent_buffer *eb;
828 int was_dirty;
829
830 tree = &BTRFS_I(page->mapping->host)->io_tree;
831 if (!(current->flags & PF_MEMALLOC)) {
832 return extent_write_full_page(tree, page,
833 btree_get_extent, wbc);
834 }
835
836 redirty_page_for_writepage(wbc, page);
837 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
838 WARN_ON(!eb);
839
840 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
841 if (!was_dirty) {
842 spin_lock(&root->fs_info->delalloc_lock);
843 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
844 spin_unlock(&root->fs_info->delalloc_lock);
845 }
846 free_extent_buffer(eb);
847
848 unlock_page(page);
849 return 0;
850 }
851
852 static int btree_writepages(struct address_space *mapping,
853 struct writeback_control *wbc)
854 {
855 struct extent_io_tree *tree;
856 tree = &BTRFS_I(mapping->host)->io_tree;
857 if (wbc->sync_mode == WB_SYNC_NONE) {
858 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
859 u64 num_dirty;
860 unsigned long thresh = 32 * 1024 * 1024;
861
862 if (wbc->for_kupdate)
863 return 0;
864
865 /* this is a bit racy, but that's ok */
866 num_dirty = root->fs_info->dirty_metadata_bytes;
867 if (num_dirty < thresh)
868 return 0;
869 }
870 return extent_writepages(tree, mapping, btree_get_extent, wbc);
871 }
872
873 static int btree_readpage(struct file *file, struct page *page)
874 {
875 struct extent_io_tree *tree;
876 tree = &BTRFS_I(page->mapping->host)->io_tree;
877 return extent_read_full_page(tree, page, btree_get_extent);
878 }
879
880 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
881 {
882 struct extent_io_tree *tree;
883 struct extent_map_tree *map;
884 int ret;
885
886 if (PageWriteback(page) || PageDirty(page))
887 return 0;
888
889 tree = &BTRFS_I(page->mapping->host)->io_tree;
890 map = &BTRFS_I(page->mapping->host)->extent_tree;
891
892 ret = try_release_extent_state(map, tree, page, gfp_flags);
893 if (!ret)
894 return 0;
895
896 ret = try_release_extent_buffer(tree, page);
897 if (ret == 1) {
898 ClearPagePrivate(page);
899 set_page_private(page, 0);
900 page_cache_release(page);
901 }
902
903 return ret;
904 }
905
906 static void btree_invalidatepage(struct page *page, unsigned long offset)
907 {
908 struct extent_io_tree *tree;
909 tree = &BTRFS_I(page->mapping->host)->io_tree;
910 extent_invalidatepage(tree, page, offset);
911 btree_releasepage(page, GFP_NOFS);
912 if (PagePrivate(page)) {
913 printk(KERN_WARNING "btrfs warning page private not zero "
914 "on page %llu\n", (unsigned long long)page_offset(page));
915 ClearPagePrivate(page);
916 set_page_private(page, 0);
917 page_cache_release(page);
918 }
919 }
920
921 static const struct address_space_operations btree_aops = {
922 .readpage = btree_readpage,
923 .writepage = btree_writepage,
924 .writepages = btree_writepages,
925 .releasepage = btree_releasepage,
926 .invalidatepage = btree_invalidatepage,
927 #ifdef CONFIG_MIGRATION
928 .migratepage = btree_migratepage,
929 #endif
930 };
931
932 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
933 u64 parent_transid)
934 {
935 struct extent_buffer *buf = NULL;
936 struct inode *btree_inode = root->fs_info->btree_inode;
937 int ret = 0;
938
939 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
940 if (!buf)
941 return 0;
942 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
943 buf, 0, 0, btree_get_extent, 0);
944 free_extent_buffer(buf);
945 return ret;
946 }
947
948 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
949 u64 bytenr, u32 blocksize)
950 {
951 struct inode *btree_inode = root->fs_info->btree_inode;
952 struct extent_buffer *eb;
953 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
954 bytenr, blocksize);
955 return eb;
956 }
957
958 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
959 u64 bytenr, u32 blocksize)
960 {
961 struct inode *btree_inode = root->fs_info->btree_inode;
962 struct extent_buffer *eb;
963
964 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
965 bytenr, blocksize, NULL);
966 return eb;
967 }
968
969
970 int btrfs_write_tree_block(struct extent_buffer *buf)
971 {
972 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
973 buf->start + buf->len - 1);
974 }
975
976 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
977 {
978 return filemap_fdatawait_range(buf->first_page->mapping,
979 buf->start, buf->start + buf->len - 1);
980 }
981
982 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
983 u32 blocksize, u64 parent_transid)
984 {
985 struct extent_buffer *buf = NULL;
986 int ret;
987
988 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
989 if (!buf)
990 return NULL;
991
992 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
993
994 if (ret == 0)
995 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
996 return buf;
997
998 }
999
1000 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1001 struct extent_buffer *buf)
1002 {
1003 struct inode *btree_inode = root->fs_info->btree_inode;
1004 if (btrfs_header_generation(buf) ==
1005 root->fs_info->running_transaction->transid) {
1006 btrfs_assert_tree_locked(buf);
1007
1008 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1009 spin_lock(&root->fs_info->delalloc_lock);
1010 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1011 root->fs_info->dirty_metadata_bytes -= buf->len;
1012 else
1013 WARN_ON(1);
1014 spin_unlock(&root->fs_info->delalloc_lock);
1015 }
1016
1017 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1018 btrfs_set_lock_blocking(buf);
1019 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1020 buf);
1021 }
1022 return 0;
1023 }
1024
1025 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1026 u32 stripesize, struct btrfs_root *root,
1027 struct btrfs_fs_info *fs_info,
1028 u64 objectid)
1029 {
1030 root->node = NULL;
1031 root->commit_root = NULL;
1032 root->sectorsize = sectorsize;
1033 root->nodesize = nodesize;
1034 root->leafsize = leafsize;
1035 root->stripesize = stripesize;
1036 root->ref_cows = 0;
1037 root->track_dirty = 0;
1038 root->in_radix = 0;
1039 root->orphan_item_inserted = 0;
1040 root->orphan_cleanup_state = 0;
1041
1042 root->fs_info = fs_info;
1043 root->objectid = objectid;
1044 root->last_trans = 0;
1045 root->highest_objectid = 0;
1046 root->name = NULL;
1047 root->inode_tree = RB_ROOT;
1048 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1049 root->block_rsv = NULL;
1050 root->orphan_block_rsv = NULL;
1051
1052 INIT_LIST_HEAD(&root->dirty_list);
1053 INIT_LIST_HEAD(&root->orphan_list);
1054 INIT_LIST_HEAD(&root->root_list);
1055 spin_lock_init(&root->orphan_lock);
1056 spin_lock_init(&root->inode_lock);
1057 spin_lock_init(&root->accounting_lock);
1058 mutex_init(&root->objectid_mutex);
1059 mutex_init(&root->log_mutex);
1060 init_waitqueue_head(&root->log_writer_wait);
1061 init_waitqueue_head(&root->log_commit_wait[0]);
1062 init_waitqueue_head(&root->log_commit_wait[1]);
1063 atomic_set(&root->log_commit[0], 0);
1064 atomic_set(&root->log_commit[1], 0);
1065 atomic_set(&root->log_writers, 0);
1066 root->log_batch = 0;
1067 root->log_transid = 0;
1068 root->last_log_commit = 0;
1069 extent_io_tree_init(&root->dirty_log_pages,
1070 fs_info->btree_inode->i_mapping);
1071
1072 memset(&root->root_key, 0, sizeof(root->root_key));
1073 memset(&root->root_item, 0, sizeof(root->root_item));
1074 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1075 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1076 root->defrag_trans_start = fs_info->generation;
1077 init_completion(&root->kobj_unregister);
1078 root->defrag_running = 0;
1079 root->root_key.objectid = objectid;
1080 root->anon_dev = 0;
1081 return 0;
1082 }
1083
1084 static int find_and_setup_root(struct btrfs_root *tree_root,
1085 struct btrfs_fs_info *fs_info,
1086 u64 objectid,
1087 struct btrfs_root *root)
1088 {
1089 int ret;
1090 u32 blocksize;
1091 u64 generation;
1092
1093 __setup_root(tree_root->nodesize, tree_root->leafsize,
1094 tree_root->sectorsize, tree_root->stripesize,
1095 root, fs_info, objectid);
1096 ret = btrfs_find_last_root(tree_root, objectid,
1097 &root->root_item, &root->root_key);
1098 if (ret > 0)
1099 return -ENOENT;
1100 BUG_ON(ret);
1101
1102 generation = btrfs_root_generation(&root->root_item);
1103 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1104 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1105 blocksize, generation);
1106 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1107 free_extent_buffer(root->node);
1108 return -EIO;
1109 }
1110 root->commit_root = btrfs_root_node(root);
1111 return 0;
1112 }
1113
1114 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1115 struct btrfs_fs_info *fs_info)
1116 {
1117 struct btrfs_root *root;
1118 struct btrfs_root *tree_root = fs_info->tree_root;
1119 struct extent_buffer *leaf;
1120
1121 root = kzalloc(sizeof(*root), GFP_NOFS);
1122 if (!root)
1123 return ERR_PTR(-ENOMEM);
1124
1125 __setup_root(tree_root->nodesize, tree_root->leafsize,
1126 tree_root->sectorsize, tree_root->stripesize,
1127 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1128
1129 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1130 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1131 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1132 /*
1133 * log trees do not get reference counted because they go away
1134 * before a real commit is actually done. They do store pointers
1135 * to file data extents, and those reference counts still get
1136 * updated (along with back refs to the log tree).
1137 */
1138 root->ref_cows = 0;
1139
1140 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1141 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1142 if (IS_ERR(leaf)) {
1143 kfree(root);
1144 return ERR_CAST(leaf);
1145 }
1146
1147 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1148 btrfs_set_header_bytenr(leaf, leaf->start);
1149 btrfs_set_header_generation(leaf, trans->transid);
1150 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1151 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1152 root->node = leaf;
1153
1154 write_extent_buffer(root->node, root->fs_info->fsid,
1155 (unsigned long)btrfs_header_fsid(root->node),
1156 BTRFS_FSID_SIZE);
1157 btrfs_mark_buffer_dirty(root->node);
1158 btrfs_tree_unlock(root->node);
1159 return root;
1160 }
1161
1162 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1163 struct btrfs_fs_info *fs_info)
1164 {
1165 struct btrfs_root *log_root;
1166
1167 log_root = alloc_log_tree(trans, fs_info);
1168 if (IS_ERR(log_root))
1169 return PTR_ERR(log_root);
1170 WARN_ON(fs_info->log_root_tree);
1171 fs_info->log_root_tree = log_root;
1172 return 0;
1173 }
1174
1175 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1176 struct btrfs_root *root)
1177 {
1178 struct btrfs_root *log_root;
1179 struct btrfs_inode_item *inode_item;
1180
1181 log_root = alloc_log_tree(trans, root->fs_info);
1182 if (IS_ERR(log_root))
1183 return PTR_ERR(log_root);
1184
1185 log_root->last_trans = trans->transid;
1186 log_root->root_key.offset = root->root_key.objectid;
1187
1188 inode_item = &log_root->root_item.inode;
1189 inode_item->generation = cpu_to_le64(1);
1190 inode_item->size = cpu_to_le64(3);
1191 inode_item->nlink = cpu_to_le32(1);
1192 inode_item->nbytes = cpu_to_le64(root->leafsize);
1193 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1194
1195 btrfs_set_root_node(&log_root->root_item, log_root->node);
1196
1197 WARN_ON(root->log_root);
1198 root->log_root = log_root;
1199 root->log_transid = 0;
1200 root->last_log_commit = 0;
1201 return 0;
1202 }
1203
1204 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1205 struct btrfs_key *location)
1206 {
1207 struct btrfs_root *root;
1208 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1209 struct btrfs_path *path;
1210 struct extent_buffer *l;
1211 u64 generation;
1212 u32 blocksize;
1213 int ret = 0;
1214
1215 root = kzalloc(sizeof(*root), GFP_NOFS);
1216 if (!root)
1217 return ERR_PTR(-ENOMEM);
1218 if (location->offset == (u64)-1) {
1219 ret = find_and_setup_root(tree_root, fs_info,
1220 location->objectid, root);
1221 if (ret) {
1222 kfree(root);
1223 return ERR_PTR(ret);
1224 }
1225 goto out;
1226 }
1227
1228 __setup_root(tree_root->nodesize, tree_root->leafsize,
1229 tree_root->sectorsize, tree_root->stripesize,
1230 root, fs_info, location->objectid);
1231
1232 path = btrfs_alloc_path();
1233 if (!path) {
1234 kfree(root);
1235 return ERR_PTR(-ENOMEM);
1236 }
1237 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1238 if (ret == 0) {
1239 l = path->nodes[0];
1240 read_extent_buffer(l, &root->root_item,
1241 btrfs_item_ptr_offset(l, path->slots[0]),
1242 sizeof(root->root_item));
1243 memcpy(&root->root_key, location, sizeof(*location));
1244 }
1245 btrfs_free_path(path);
1246 if (ret) {
1247 kfree(root);
1248 if (ret > 0)
1249 ret = -ENOENT;
1250 return ERR_PTR(ret);
1251 }
1252
1253 generation = btrfs_root_generation(&root->root_item);
1254 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1255 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1256 blocksize, generation);
1257 root->commit_root = btrfs_root_node(root);
1258 BUG_ON(!root->node);
1259 out:
1260 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1261 root->ref_cows = 1;
1262 btrfs_check_and_init_root_item(&root->root_item);
1263 }
1264
1265 return root;
1266 }
1267
1268 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1269 struct btrfs_key *location)
1270 {
1271 struct btrfs_root *root;
1272 int ret;
1273
1274 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1275 return fs_info->tree_root;
1276 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1277 return fs_info->extent_root;
1278 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1279 return fs_info->chunk_root;
1280 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1281 return fs_info->dev_root;
1282 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1283 return fs_info->csum_root;
1284 again:
1285 spin_lock(&fs_info->fs_roots_radix_lock);
1286 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1287 (unsigned long)location->objectid);
1288 spin_unlock(&fs_info->fs_roots_radix_lock);
1289 if (root)
1290 return root;
1291
1292 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1293 if (IS_ERR(root))
1294 return root;
1295
1296 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1297 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1298 GFP_NOFS);
1299 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1300 ret = -ENOMEM;
1301 goto fail;
1302 }
1303
1304 btrfs_init_free_ino_ctl(root);
1305 mutex_init(&root->fs_commit_mutex);
1306 spin_lock_init(&root->cache_lock);
1307 init_waitqueue_head(&root->cache_wait);
1308
1309 ret = get_anon_bdev(&root->anon_dev);
1310 if (ret)
1311 goto fail;
1312
1313 if (btrfs_root_refs(&root->root_item) == 0) {
1314 ret = -ENOENT;
1315 goto fail;
1316 }
1317
1318 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1319 if (ret < 0)
1320 goto fail;
1321 if (ret == 0)
1322 root->orphan_item_inserted = 1;
1323
1324 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1325 if (ret)
1326 goto fail;
1327
1328 spin_lock(&fs_info->fs_roots_radix_lock);
1329 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1330 (unsigned long)root->root_key.objectid,
1331 root);
1332 if (ret == 0)
1333 root->in_radix = 1;
1334
1335 spin_unlock(&fs_info->fs_roots_radix_lock);
1336 radix_tree_preload_end();
1337 if (ret) {
1338 if (ret == -EEXIST) {
1339 free_fs_root(root);
1340 goto again;
1341 }
1342 goto fail;
1343 }
1344
1345 ret = btrfs_find_dead_roots(fs_info->tree_root,
1346 root->root_key.objectid);
1347 WARN_ON(ret);
1348 return root;
1349 fail:
1350 free_fs_root(root);
1351 return ERR_PTR(ret);
1352 }
1353
1354 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1355 {
1356 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1357 int ret = 0;
1358 struct btrfs_device *device;
1359 struct backing_dev_info *bdi;
1360
1361 rcu_read_lock();
1362 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1363 if (!device->bdev)
1364 continue;
1365 bdi = blk_get_backing_dev_info(device->bdev);
1366 if (bdi && bdi_congested(bdi, bdi_bits)) {
1367 ret = 1;
1368 break;
1369 }
1370 }
1371 rcu_read_unlock();
1372 return ret;
1373 }
1374
1375 /*
1376 * If this fails, caller must call bdi_destroy() to get rid of the
1377 * bdi again.
1378 */
1379 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1380 {
1381 int err;
1382
1383 bdi->capabilities = BDI_CAP_MAP_COPY;
1384 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1385 if (err)
1386 return err;
1387
1388 bdi->ra_pages = default_backing_dev_info.ra_pages;
1389 bdi->congested_fn = btrfs_congested_fn;
1390 bdi->congested_data = info;
1391 return 0;
1392 }
1393
1394 static int bio_ready_for_csum(struct bio *bio)
1395 {
1396 u64 length = 0;
1397 u64 buf_len = 0;
1398 u64 start = 0;
1399 struct page *page;
1400 struct extent_io_tree *io_tree = NULL;
1401 struct bio_vec *bvec;
1402 int i;
1403 int ret;
1404
1405 bio_for_each_segment(bvec, bio, i) {
1406 page = bvec->bv_page;
1407 if (page->private == EXTENT_PAGE_PRIVATE) {
1408 length += bvec->bv_len;
1409 continue;
1410 }
1411 if (!page->private) {
1412 length += bvec->bv_len;
1413 continue;
1414 }
1415 length = bvec->bv_len;
1416 buf_len = page->private >> 2;
1417 start = page_offset(page) + bvec->bv_offset;
1418 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1419 }
1420 /* are we fully contained in this bio? */
1421 if (buf_len <= length)
1422 return 1;
1423
1424 ret = extent_range_uptodate(io_tree, start + length,
1425 start + buf_len - 1);
1426 return ret;
1427 }
1428
1429 /*
1430 * called by the kthread helper functions to finally call the bio end_io
1431 * functions. This is where read checksum verification actually happens
1432 */
1433 static void end_workqueue_fn(struct btrfs_work *work)
1434 {
1435 struct bio *bio;
1436 struct end_io_wq *end_io_wq;
1437 struct btrfs_fs_info *fs_info;
1438 int error;
1439
1440 end_io_wq = container_of(work, struct end_io_wq, work);
1441 bio = end_io_wq->bio;
1442 fs_info = end_io_wq->info;
1443
1444 /* metadata bio reads are special because the whole tree block must
1445 * be checksummed at once. This makes sure the entire block is in
1446 * ram and up to date before trying to verify things. For
1447 * blocksize <= pagesize, it is basically a noop
1448 */
1449 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1450 !bio_ready_for_csum(bio)) {
1451 btrfs_queue_worker(&fs_info->endio_meta_workers,
1452 &end_io_wq->work);
1453 return;
1454 }
1455 error = end_io_wq->error;
1456 bio->bi_private = end_io_wq->private;
1457 bio->bi_end_io = end_io_wq->end_io;
1458 kfree(end_io_wq);
1459 bio_endio(bio, error);
1460 }
1461
1462 static int cleaner_kthread(void *arg)
1463 {
1464 struct btrfs_root *root = arg;
1465
1466 do {
1467 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1468
1469 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1470 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1471 btrfs_run_delayed_iputs(root);
1472 btrfs_clean_old_snapshots(root);
1473 mutex_unlock(&root->fs_info->cleaner_mutex);
1474 btrfs_run_defrag_inodes(root->fs_info);
1475 }
1476
1477 if (freezing(current)) {
1478 refrigerator();
1479 } else {
1480 set_current_state(TASK_INTERRUPTIBLE);
1481 if (!kthread_should_stop())
1482 schedule();
1483 __set_current_state(TASK_RUNNING);
1484 }
1485 } while (!kthread_should_stop());
1486 return 0;
1487 }
1488
1489 static int transaction_kthread(void *arg)
1490 {
1491 struct btrfs_root *root = arg;
1492 struct btrfs_trans_handle *trans;
1493 struct btrfs_transaction *cur;
1494 u64 transid;
1495 unsigned long now;
1496 unsigned long delay;
1497 int ret;
1498
1499 do {
1500 delay = HZ * 30;
1501 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1502 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1503
1504 spin_lock(&root->fs_info->trans_lock);
1505 cur = root->fs_info->running_transaction;
1506 if (!cur) {
1507 spin_unlock(&root->fs_info->trans_lock);
1508 goto sleep;
1509 }
1510
1511 now = get_seconds();
1512 if (!cur->blocked &&
1513 (now < cur->start_time || now - cur->start_time < 30)) {
1514 spin_unlock(&root->fs_info->trans_lock);
1515 delay = HZ * 5;
1516 goto sleep;
1517 }
1518 transid = cur->transid;
1519 spin_unlock(&root->fs_info->trans_lock);
1520
1521 trans = btrfs_join_transaction(root);
1522 BUG_ON(IS_ERR(trans));
1523 if (transid == trans->transid) {
1524 ret = btrfs_commit_transaction(trans, root);
1525 BUG_ON(ret);
1526 } else {
1527 btrfs_end_transaction(trans, root);
1528 }
1529 sleep:
1530 wake_up_process(root->fs_info->cleaner_kthread);
1531 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1532
1533 if (freezing(current)) {
1534 refrigerator();
1535 } else {
1536 set_current_state(TASK_INTERRUPTIBLE);
1537 if (!kthread_should_stop() &&
1538 !btrfs_transaction_blocked(root->fs_info))
1539 schedule_timeout(delay);
1540 __set_current_state(TASK_RUNNING);
1541 }
1542 } while (!kthread_should_stop());
1543 return 0;
1544 }
1545
1546 struct btrfs_root *open_ctree(struct super_block *sb,
1547 struct btrfs_fs_devices *fs_devices,
1548 char *options)
1549 {
1550 u32 sectorsize;
1551 u32 nodesize;
1552 u32 leafsize;
1553 u32 blocksize;
1554 u32 stripesize;
1555 u64 generation;
1556 u64 features;
1557 struct btrfs_key location;
1558 struct buffer_head *bh;
1559 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1560 GFP_NOFS);
1561 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1562 GFP_NOFS);
1563 struct btrfs_root *tree_root = btrfs_sb(sb);
1564 struct btrfs_fs_info *fs_info = NULL;
1565 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1566 GFP_NOFS);
1567 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1568 GFP_NOFS);
1569 struct btrfs_root *log_tree_root;
1570
1571 int ret;
1572 int err = -EINVAL;
1573
1574 struct btrfs_super_block *disk_super;
1575
1576 if (!extent_root || !tree_root || !tree_root->fs_info ||
1577 !chunk_root || !dev_root || !csum_root) {
1578 err = -ENOMEM;
1579 goto fail;
1580 }
1581 fs_info = tree_root->fs_info;
1582
1583 ret = init_srcu_struct(&fs_info->subvol_srcu);
1584 if (ret) {
1585 err = ret;
1586 goto fail;
1587 }
1588
1589 ret = setup_bdi(fs_info, &fs_info->bdi);
1590 if (ret) {
1591 err = ret;
1592 goto fail_srcu;
1593 }
1594
1595 fs_info->btree_inode = new_inode(sb);
1596 if (!fs_info->btree_inode) {
1597 err = -ENOMEM;
1598 goto fail_bdi;
1599 }
1600
1601 fs_info->btree_inode->i_mapping->flags &= ~__GFP_FS;
1602
1603 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1604 INIT_LIST_HEAD(&fs_info->trans_list);
1605 INIT_LIST_HEAD(&fs_info->dead_roots);
1606 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1607 INIT_LIST_HEAD(&fs_info->hashers);
1608 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1609 INIT_LIST_HEAD(&fs_info->ordered_operations);
1610 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1611 spin_lock_init(&fs_info->delalloc_lock);
1612 spin_lock_init(&fs_info->trans_lock);
1613 spin_lock_init(&fs_info->ref_cache_lock);
1614 spin_lock_init(&fs_info->fs_roots_radix_lock);
1615 spin_lock_init(&fs_info->delayed_iput_lock);
1616 spin_lock_init(&fs_info->defrag_inodes_lock);
1617 mutex_init(&fs_info->reloc_mutex);
1618
1619 init_completion(&fs_info->kobj_unregister);
1620 fs_info->tree_root = tree_root;
1621 fs_info->extent_root = extent_root;
1622 fs_info->csum_root = csum_root;
1623 fs_info->chunk_root = chunk_root;
1624 fs_info->dev_root = dev_root;
1625 fs_info->fs_devices = fs_devices;
1626 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1627 INIT_LIST_HEAD(&fs_info->space_info);
1628 btrfs_mapping_init(&fs_info->mapping_tree);
1629 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1630 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1631 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1632 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1633 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1634 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1635 mutex_init(&fs_info->durable_block_rsv_mutex);
1636 atomic_set(&fs_info->nr_async_submits, 0);
1637 atomic_set(&fs_info->async_delalloc_pages, 0);
1638 atomic_set(&fs_info->async_submit_draining, 0);
1639 atomic_set(&fs_info->nr_async_bios, 0);
1640 atomic_set(&fs_info->defrag_running, 0);
1641 fs_info->sb = sb;
1642 fs_info->max_inline = 8192 * 1024;
1643 fs_info->metadata_ratio = 0;
1644 fs_info->defrag_inodes = RB_ROOT;
1645 fs_info->trans_no_join = 0;
1646
1647 fs_info->thread_pool_size = min_t(unsigned long,
1648 num_online_cpus() + 2, 8);
1649
1650 INIT_LIST_HEAD(&fs_info->ordered_extents);
1651 spin_lock_init(&fs_info->ordered_extent_lock);
1652 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1653 GFP_NOFS);
1654 if (!fs_info->delayed_root) {
1655 err = -ENOMEM;
1656 goto fail_iput;
1657 }
1658 btrfs_init_delayed_root(fs_info->delayed_root);
1659
1660 mutex_init(&fs_info->scrub_lock);
1661 atomic_set(&fs_info->scrubs_running, 0);
1662 atomic_set(&fs_info->scrub_pause_req, 0);
1663 atomic_set(&fs_info->scrubs_paused, 0);
1664 atomic_set(&fs_info->scrub_cancel_req, 0);
1665 init_waitqueue_head(&fs_info->scrub_pause_wait);
1666 init_rwsem(&fs_info->scrub_super_lock);
1667 fs_info->scrub_workers_refcnt = 0;
1668
1669 sb->s_blocksize = 4096;
1670 sb->s_blocksize_bits = blksize_bits(4096);
1671 sb->s_bdi = &fs_info->bdi;
1672
1673 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1674 fs_info->btree_inode->i_nlink = 1;
1675 /*
1676 * we set the i_size on the btree inode to the max possible int.
1677 * the real end of the address space is determined by all of
1678 * the devices in the system
1679 */
1680 fs_info->btree_inode->i_size = OFFSET_MAX;
1681 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1682 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1683
1684 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1685 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1686 fs_info->btree_inode->i_mapping);
1687 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1688
1689 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1690
1691 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1692 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1693 sizeof(struct btrfs_key));
1694 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1695 insert_inode_hash(fs_info->btree_inode);
1696
1697 spin_lock_init(&fs_info->block_group_cache_lock);
1698 fs_info->block_group_cache_tree = RB_ROOT;
1699
1700 extent_io_tree_init(&fs_info->freed_extents[0],
1701 fs_info->btree_inode->i_mapping);
1702 extent_io_tree_init(&fs_info->freed_extents[1],
1703 fs_info->btree_inode->i_mapping);
1704 fs_info->pinned_extents = &fs_info->freed_extents[0];
1705 fs_info->do_barriers = 1;
1706
1707
1708 mutex_init(&fs_info->ordered_operations_mutex);
1709 mutex_init(&fs_info->tree_log_mutex);
1710 mutex_init(&fs_info->chunk_mutex);
1711 mutex_init(&fs_info->transaction_kthread_mutex);
1712 mutex_init(&fs_info->cleaner_mutex);
1713 mutex_init(&fs_info->volume_mutex);
1714 init_rwsem(&fs_info->extent_commit_sem);
1715 init_rwsem(&fs_info->cleanup_work_sem);
1716 init_rwsem(&fs_info->subvol_sem);
1717
1718 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1719 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1720
1721 init_waitqueue_head(&fs_info->transaction_throttle);
1722 init_waitqueue_head(&fs_info->transaction_wait);
1723 init_waitqueue_head(&fs_info->transaction_blocked_wait);
1724 init_waitqueue_head(&fs_info->async_submit_wait);
1725
1726 __setup_root(4096, 4096, 4096, 4096, tree_root,
1727 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1728
1729 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1730 if (!bh) {
1731 err = -EINVAL;
1732 goto fail_alloc;
1733 }
1734
1735 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1736 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1737 sizeof(fs_info->super_for_commit));
1738 brelse(bh);
1739
1740 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1741
1742 disk_super = &fs_info->super_copy;
1743 if (!btrfs_super_root(disk_super))
1744 goto fail_alloc;
1745
1746 /* check FS state, whether FS is broken. */
1747 fs_info->fs_state |= btrfs_super_flags(disk_super);
1748
1749 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1750
1751 /*
1752 * In the long term, we'll store the compression type in the super
1753 * block, and it'll be used for per file compression control.
1754 */
1755 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1756
1757 ret = btrfs_parse_options(tree_root, options);
1758 if (ret) {
1759 err = ret;
1760 goto fail_alloc;
1761 }
1762
1763 features = btrfs_super_incompat_flags(disk_super) &
1764 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1765 if (features) {
1766 printk(KERN_ERR "BTRFS: couldn't mount because of "
1767 "unsupported optional features (%Lx).\n",
1768 (unsigned long long)features);
1769 err = -EINVAL;
1770 goto fail_alloc;
1771 }
1772
1773 features = btrfs_super_incompat_flags(disk_super);
1774 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1775 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1776 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1777 btrfs_set_super_incompat_flags(disk_super, features);
1778
1779 features = btrfs_super_compat_ro_flags(disk_super) &
1780 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1781 if (!(sb->s_flags & MS_RDONLY) && features) {
1782 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1783 "unsupported option features (%Lx).\n",
1784 (unsigned long long)features);
1785 err = -EINVAL;
1786 goto fail_alloc;
1787 }
1788
1789 btrfs_init_workers(&fs_info->generic_worker,
1790 "genwork", 1, NULL);
1791
1792 btrfs_init_workers(&fs_info->workers, "worker",
1793 fs_info->thread_pool_size,
1794 &fs_info->generic_worker);
1795
1796 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1797 fs_info->thread_pool_size,
1798 &fs_info->generic_worker);
1799
1800 btrfs_init_workers(&fs_info->submit_workers, "submit",
1801 min_t(u64, fs_devices->num_devices,
1802 fs_info->thread_pool_size),
1803 &fs_info->generic_worker);
1804
1805 /* a higher idle thresh on the submit workers makes it much more
1806 * likely that bios will be send down in a sane order to the
1807 * devices
1808 */
1809 fs_info->submit_workers.idle_thresh = 64;
1810
1811 fs_info->workers.idle_thresh = 16;
1812 fs_info->workers.ordered = 1;
1813
1814 fs_info->delalloc_workers.idle_thresh = 2;
1815 fs_info->delalloc_workers.ordered = 1;
1816
1817 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1818 &fs_info->generic_worker);
1819 btrfs_init_workers(&fs_info->endio_workers, "endio",
1820 fs_info->thread_pool_size,
1821 &fs_info->generic_worker);
1822 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1823 fs_info->thread_pool_size,
1824 &fs_info->generic_worker);
1825 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1826 "endio-meta-write", fs_info->thread_pool_size,
1827 &fs_info->generic_worker);
1828 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1829 fs_info->thread_pool_size,
1830 &fs_info->generic_worker);
1831 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1832 1, &fs_info->generic_worker);
1833 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
1834 fs_info->thread_pool_size,
1835 &fs_info->generic_worker);
1836
1837 /*
1838 * endios are largely parallel and should have a very
1839 * low idle thresh
1840 */
1841 fs_info->endio_workers.idle_thresh = 4;
1842 fs_info->endio_meta_workers.idle_thresh = 4;
1843
1844 fs_info->endio_write_workers.idle_thresh = 2;
1845 fs_info->endio_meta_write_workers.idle_thresh = 2;
1846
1847 btrfs_start_workers(&fs_info->workers, 1);
1848 btrfs_start_workers(&fs_info->generic_worker, 1);
1849 btrfs_start_workers(&fs_info->submit_workers, 1);
1850 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1851 btrfs_start_workers(&fs_info->fixup_workers, 1);
1852 btrfs_start_workers(&fs_info->endio_workers, 1);
1853 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1854 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1855 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1856 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1857 btrfs_start_workers(&fs_info->delayed_workers, 1);
1858
1859 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1860 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1861 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1862
1863 nodesize = btrfs_super_nodesize(disk_super);
1864 leafsize = btrfs_super_leafsize(disk_super);
1865 sectorsize = btrfs_super_sectorsize(disk_super);
1866 stripesize = btrfs_super_stripesize(disk_super);
1867 tree_root->nodesize = nodesize;
1868 tree_root->leafsize = leafsize;
1869 tree_root->sectorsize = sectorsize;
1870 tree_root->stripesize = stripesize;
1871
1872 sb->s_blocksize = sectorsize;
1873 sb->s_blocksize_bits = blksize_bits(sectorsize);
1874
1875 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1876 sizeof(disk_super->magic))) {
1877 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1878 goto fail_sb_buffer;
1879 }
1880
1881 mutex_lock(&fs_info->chunk_mutex);
1882 ret = btrfs_read_sys_array(tree_root);
1883 mutex_unlock(&fs_info->chunk_mutex);
1884 if (ret) {
1885 printk(KERN_WARNING "btrfs: failed to read the system "
1886 "array on %s\n", sb->s_id);
1887 goto fail_sb_buffer;
1888 }
1889
1890 blocksize = btrfs_level_size(tree_root,
1891 btrfs_super_chunk_root_level(disk_super));
1892 generation = btrfs_super_chunk_root_generation(disk_super);
1893
1894 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1895 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1896
1897 chunk_root->node = read_tree_block(chunk_root,
1898 btrfs_super_chunk_root(disk_super),
1899 blocksize, generation);
1900 BUG_ON(!chunk_root->node);
1901 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1902 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1903 sb->s_id);
1904 goto fail_chunk_root;
1905 }
1906 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1907 chunk_root->commit_root = btrfs_root_node(chunk_root);
1908
1909 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1910 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1911 BTRFS_UUID_SIZE);
1912
1913 mutex_lock(&fs_info->chunk_mutex);
1914 ret = btrfs_read_chunk_tree(chunk_root);
1915 mutex_unlock(&fs_info->chunk_mutex);
1916 if (ret) {
1917 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1918 sb->s_id);
1919 goto fail_chunk_root;
1920 }
1921
1922 btrfs_close_extra_devices(fs_devices);
1923
1924 blocksize = btrfs_level_size(tree_root,
1925 btrfs_super_root_level(disk_super));
1926 generation = btrfs_super_generation(disk_super);
1927
1928 tree_root->node = read_tree_block(tree_root,
1929 btrfs_super_root(disk_super),
1930 blocksize, generation);
1931 if (!tree_root->node)
1932 goto fail_chunk_root;
1933 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1934 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1935 sb->s_id);
1936 goto fail_tree_root;
1937 }
1938 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1939 tree_root->commit_root = btrfs_root_node(tree_root);
1940
1941 ret = find_and_setup_root(tree_root, fs_info,
1942 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1943 if (ret)
1944 goto fail_tree_root;
1945 extent_root->track_dirty = 1;
1946
1947 ret = find_and_setup_root(tree_root, fs_info,
1948 BTRFS_DEV_TREE_OBJECTID, dev_root);
1949 if (ret)
1950 goto fail_extent_root;
1951 dev_root->track_dirty = 1;
1952
1953 ret = find_and_setup_root(tree_root, fs_info,
1954 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1955 if (ret)
1956 goto fail_dev_root;
1957
1958 csum_root->track_dirty = 1;
1959
1960 fs_info->generation = generation;
1961 fs_info->last_trans_committed = generation;
1962 fs_info->data_alloc_profile = (u64)-1;
1963 fs_info->metadata_alloc_profile = (u64)-1;
1964 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1965
1966 ret = btrfs_init_space_info(fs_info);
1967 if (ret) {
1968 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
1969 goto fail_block_groups;
1970 }
1971
1972 ret = btrfs_read_block_groups(extent_root);
1973 if (ret) {
1974 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
1975 goto fail_block_groups;
1976 }
1977
1978 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1979 "btrfs-cleaner");
1980 if (IS_ERR(fs_info->cleaner_kthread))
1981 goto fail_block_groups;
1982
1983 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1984 tree_root,
1985 "btrfs-transaction");
1986 if (IS_ERR(fs_info->transaction_kthread))
1987 goto fail_cleaner;
1988
1989 if (!btrfs_test_opt(tree_root, SSD) &&
1990 !btrfs_test_opt(tree_root, NOSSD) &&
1991 !fs_info->fs_devices->rotating) {
1992 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1993 "mode\n");
1994 btrfs_set_opt(fs_info->mount_opt, SSD);
1995 }
1996
1997 /* do not make disk changes in broken FS */
1998 if (btrfs_super_log_root(disk_super) != 0 &&
1999 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2000 u64 bytenr = btrfs_super_log_root(disk_super);
2001
2002 if (fs_devices->rw_devices == 0) {
2003 printk(KERN_WARNING "Btrfs log replay required "
2004 "on RO media\n");
2005 err = -EIO;
2006 goto fail_trans_kthread;
2007 }
2008 blocksize =
2009 btrfs_level_size(tree_root,
2010 btrfs_super_log_root_level(disk_super));
2011
2012 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2013 if (!log_tree_root) {
2014 err = -ENOMEM;
2015 goto fail_trans_kthread;
2016 }
2017
2018 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2019 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2020
2021 log_tree_root->node = read_tree_block(tree_root, bytenr,
2022 blocksize,
2023 generation + 1);
2024 ret = btrfs_recover_log_trees(log_tree_root);
2025 BUG_ON(ret);
2026
2027 if (sb->s_flags & MS_RDONLY) {
2028 ret = btrfs_commit_super(tree_root);
2029 BUG_ON(ret);
2030 }
2031 }
2032
2033 ret = btrfs_find_orphan_roots(tree_root);
2034 BUG_ON(ret);
2035
2036 if (!(sb->s_flags & MS_RDONLY)) {
2037 ret = btrfs_cleanup_fs_roots(fs_info);
2038 BUG_ON(ret);
2039
2040 ret = btrfs_recover_relocation(tree_root);
2041 if (ret < 0) {
2042 printk(KERN_WARNING
2043 "btrfs: failed to recover relocation\n");
2044 err = -EINVAL;
2045 goto fail_trans_kthread;
2046 }
2047 }
2048
2049 location.objectid = BTRFS_FS_TREE_OBJECTID;
2050 location.type = BTRFS_ROOT_ITEM_KEY;
2051 location.offset = (u64)-1;
2052
2053 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2054 if (!fs_info->fs_root)
2055 goto fail_trans_kthread;
2056 if (IS_ERR(fs_info->fs_root)) {
2057 err = PTR_ERR(fs_info->fs_root);
2058 goto fail_trans_kthread;
2059 }
2060
2061 if (!(sb->s_flags & MS_RDONLY)) {
2062 down_read(&fs_info->cleanup_work_sem);
2063 err = btrfs_orphan_cleanup(fs_info->fs_root);
2064 if (!err)
2065 err = btrfs_orphan_cleanup(fs_info->tree_root);
2066 up_read(&fs_info->cleanup_work_sem);
2067 if (err) {
2068 close_ctree(tree_root);
2069 return ERR_PTR(err);
2070 }
2071 }
2072
2073 return tree_root;
2074
2075 fail_trans_kthread:
2076 kthread_stop(fs_info->transaction_kthread);
2077 fail_cleaner:
2078 kthread_stop(fs_info->cleaner_kthread);
2079
2080 /*
2081 * make sure we're done with the btree inode before we stop our
2082 * kthreads
2083 */
2084 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2085 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2086
2087 fail_block_groups:
2088 btrfs_free_block_groups(fs_info);
2089 free_extent_buffer(csum_root->node);
2090 free_extent_buffer(csum_root->commit_root);
2091 fail_dev_root:
2092 free_extent_buffer(dev_root->node);
2093 free_extent_buffer(dev_root->commit_root);
2094 fail_extent_root:
2095 free_extent_buffer(extent_root->node);
2096 free_extent_buffer(extent_root->commit_root);
2097 fail_tree_root:
2098 free_extent_buffer(tree_root->node);
2099 free_extent_buffer(tree_root->commit_root);
2100 fail_chunk_root:
2101 free_extent_buffer(chunk_root->node);
2102 free_extent_buffer(chunk_root->commit_root);
2103 fail_sb_buffer:
2104 btrfs_stop_workers(&fs_info->generic_worker);
2105 btrfs_stop_workers(&fs_info->fixup_workers);
2106 btrfs_stop_workers(&fs_info->delalloc_workers);
2107 btrfs_stop_workers(&fs_info->workers);
2108 btrfs_stop_workers(&fs_info->endio_workers);
2109 btrfs_stop_workers(&fs_info->endio_meta_workers);
2110 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2111 btrfs_stop_workers(&fs_info->endio_write_workers);
2112 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2113 btrfs_stop_workers(&fs_info->submit_workers);
2114 btrfs_stop_workers(&fs_info->delayed_workers);
2115 fail_alloc:
2116 kfree(fs_info->delayed_root);
2117 fail_iput:
2118 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2119 iput(fs_info->btree_inode);
2120
2121 btrfs_close_devices(fs_info->fs_devices);
2122 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2123 fail_bdi:
2124 bdi_destroy(&fs_info->bdi);
2125 fail_srcu:
2126 cleanup_srcu_struct(&fs_info->subvol_srcu);
2127 fail:
2128 kfree(extent_root);
2129 kfree(tree_root);
2130 kfree(fs_info);
2131 kfree(chunk_root);
2132 kfree(dev_root);
2133 kfree(csum_root);
2134 return ERR_PTR(err);
2135 }
2136
2137 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2138 {
2139 char b[BDEVNAME_SIZE];
2140
2141 if (uptodate) {
2142 set_buffer_uptodate(bh);
2143 } else {
2144 printk_ratelimited(KERN_WARNING "lost page write due to "
2145 "I/O error on %s\n",
2146 bdevname(bh->b_bdev, b));
2147 /* note, we dont' set_buffer_write_io_error because we have
2148 * our own ways of dealing with the IO errors
2149 */
2150 clear_buffer_uptodate(bh);
2151 }
2152 unlock_buffer(bh);
2153 put_bh(bh);
2154 }
2155
2156 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2157 {
2158 struct buffer_head *bh;
2159 struct buffer_head *latest = NULL;
2160 struct btrfs_super_block *super;
2161 int i;
2162 u64 transid = 0;
2163 u64 bytenr;
2164
2165 /* we would like to check all the supers, but that would make
2166 * a btrfs mount succeed after a mkfs from a different FS.
2167 * So, we need to add a special mount option to scan for
2168 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2169 */
2170 for (i = 0; i < 1; i++) {
2171 bytenr = btrfs_sb_offset(i);
2172 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2173 break;
2174 bh = __bread(bdev, bytenr / 4096, 4096);
2175 if (!bh)
2176 continue;
2177
2178 super = (struct btrfs_super_block *)bh->b_data;
2179 if (btrfs_super_bytenr(super) != bytenr ||
2180 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2181 sizeof(super->magic))) {
2182 brelse(bh);
2183 continue;
2184 }
2185
2186 if (!latest || btrfs_super_generation(super) > transid) {
2187 brelse(latest);
2188 latest = bh;
2189 transid = btrfs_super_generation(super);
2190 } else {
2191 brelse(bh);
2192 }
2193 }
2194 return latest;
2195 }
2196
2197 /*
2198 * this should be called twice, once with wait == 0 and
2199 * once with wait == 1. When wait == 0 is done, all the buffer heads
2200 * we write are pinned.
2201 *
2202 * They are released when wait == 1 is done.
2203 * max_mirrors must be the same for both runs, and it indicates how
2204 * many supers on this one device should be written.
2205 *
2206 * max_mirrors == 0 means to write them all.
2207 */
2208 static int write_dev_supers(struct btrfs_device *device,
2209 struct btrfs_super_block *sb,
2210 int do_barriers, int wait, int max_mirrors)
2211 {
2212 struct buffer_head *bh;
2213 int i;
2214 int ret;
2215 int errors = 0;
2216 u32 crc;
2217 u64 bytenr;
2218 int last_barrier = 0;
2219
2220 if (max_mirrors == 0)
2221 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2222
2223 /* make sure only the last submit_bh does a barrier */
2224 if (do_barriers) {
2225 for (i = 0; i < max_mirrors; i++) {
2226 bytenr = btrfs_sb_offset(i);
2227 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2228 device->total_bytes)
2229 break;
2230 last_barrier = i;
2231 }
2232 }
2233
2234 for (i = 0; i < max_mirrors; i++) {
2235 bytenr = btrfs_sb_offset(i);
2236 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2237 break;
2238
2239 if (wait) {
2240 bh = __find_get_block(device->bdev, bytenr / 4096,
2241 BTRFS_SUPER_INFO_SIZE);
2242 BUG_ON(!bh);
2243 wait_on_buffer(bh);
2244 if (!buffer_uptodate(bh))
2245 errors++;
2246
2247 /* drop our reference */
2248 brelse(bh);
2249
2250 /* drop the reference from the wait == 0 run */
2251 brelse(bh);
2252 continue;
2253 } else {
2254 btrfs_set_super_bytenr(sb, bytenr);
2255
2256 crc = ~(u32)0;
2257 crc = btrfs_csum_data(NULL, (char *)sb +
2258 BTRFS_CSUM_SIZE, crc,
2259 BTRFS_SUPER_INFO_SIZE -
2260 BTRFS_CSUM_SIZE);
2261 btrfs_csum_final(crc, sb->csum);
2262
2263 /*
2264 * one reference for us, and we leave it for the
2265 * caller
2266 */
2267 bh = __getblk(device->bdev, bytenr / 4096,
2268 BTRFS_SUPER_INFO_SIZE);
2269 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2270
2271 /* one reference for submit_bh */
2272 get_bh(bh);
2273
2274 set_buffer_uptodate(bh);
2275 lock_buffer(bh);
2276 bh->b_end_io = btrfs_end_buffer_write_sync;
2277 }
2278
2279 if (i == last_barrier && do_barriers)
2280 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2281 else
2282 ret = submit_bh(WRITE_SYNC, bh);
2283
2284 if (ret)
2285 errors++;
2286 }
2287 return errors < i ? 0 : -1;
2288 }
2289
2290 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2291 {
2292 struct list_head *head;
2293 struct btrfs_device *dev;
2294 struct btrfs_super_block *sb;
2295 struct btrfs_dev_item *dev_item;
2296 int ret;
2297 int do_barriers;
2298 int max_errors;
2299 int total_errors = 0;
2300 u64 flags;
2301
2302 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2303 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2304
2305 sb = &root->fs_info->super_for_commit;
2306 dev_item = &sb->dev_item;
2307
2308 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2309 head = &root->fs_info->fs_devices->devices;
2310 list_for_each_entry_rcu(dev, head, dev_list) {
2311 if (!dev->bdev) {
2312 total_errors++;
2313 continue;
2314 }
2315 if (!dev->in_fs_metadata || !dev->writeable)
2316 continue;
2317
2318 btrfs_set_stack_device_generation(dev_item, 0);
2319 btrfs_set_stack_device_type(dev_item, dev->type);
2320 btrfs_set_stack_device_id(dev_item, dev->devid);
2321 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2322 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2323 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2324 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2325 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2326 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2327 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2328
2329 flags = btrfs_super_flags(sb);
2330 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2331
2332 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2333 if (ret)
2334 total_errors++;
2335 }
2336 if (total_errors > max_errors) {
2337 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2338 total_errors);
2339 BUG();
2340 }
2341
2342 total_errors = 0;
2343 list_for_each_entry_rcu(dev, head, dev_list) {
2344 if (!dev->bdev)
2345 continue;
2346 if (!dev->in_fs_metadata || !dev->writeable)
2347 continue;
2348
2349 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2350 if (ret)
2351 total_errors++;
2352 }
2353 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2354 if (total_errors > max_errors) {
2355 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2356 total_errors);
2357 BUG();
2358 }
2359 return 0;
2360 }
2361
2362 int write_ctree_super(struct btrfs_trans_handle *trans,
2363 struct btrfs_root *root, int max_mirrors)
2364 {
2365 int ret;
2366
2367 ret = write_all_supers(root, max_mirrors);
2368 return ret;
2369 }
2370
2371 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2372 {
2373 spin_lock(&fs_info->fs_roots_radix_lock);
2374 radix_tree_delete(&fs_info->fs_roots_radix,
2375 (unsigned long)root->root_key.objectid);
2376 spin_unlock(&fs_info->fs_roots_radix_lock);
2377
2378 if (btrfs_root_refs(&root->root_item) == 0)
2379 synchronize_srcu(&fs_info->subvol_srcu);
2380
2381 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2382 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2383 free_fs_root(root);
2384 return 0;
2385 }
2386
2387 static void free_fs_root(struct btrfs_root *root)
2388 {
2389 iput(root->cache_inode);
2390 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2391 if (root->anon_dev)
2392 free_anon_bdev(root->anon_dev);
2393 free_extent_buffer(root->node);
2394 free_extent_buffer(root->commit_root);
2395 kfree(root->free_ino_ctl);
2396 kfree(root->free_ino_pinned);
2397 kfree(root->name);
2398 kfree(root);
2399 }
2400
2401 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2402 {
2403 int ret;
2404 struct btrfs_root *gang[8];
2405 int i;
2406
2407 while (!list_empty(&fs_info->dead_roots)) {
2408 gang[0] = list_entry(fs_info->dead_roots.next,
2409 struct btrfs_root, root_list);
2410 list_del(&gang[0]->root_list);
2411
2412 if (gang[0]->in_radix) {
2413 btrfs_free_fs_root(fs_info, gang[0]);
2414 } else {
2415 free_extent_buffer(gang[0]->node);
2416 free_extent_buffer(gang[0]->commit_root);
2417 kfree(gang[0]);
2418 }
2419 }
2420
2421 while (1) {
2422 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2423 (void **)gang, 0,
2424 ARRAY_SIZE(gang));
2425 if (!ret)
2426 break;
2427 for (i = 0; i < ret; i++)
2428 btrfs_free_fs_root(fs_info, gang[i]);
2429 }
2430 return 0;
2431 }
2432
2433 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2434 {
2435 u64 root_objectid = 0;
2436 struct btrfs_root *gang[8];
2437 int i;
2438 int ret;
2439
2440 while (1) {
2441 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2442 (void **)gang, root_objectid,
2443 ARRAY_SIZE(gang));
2444 if (!ret)
2445 break;
2446
2447 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2448 for (i = 0; i < ret; i++) {
2449 int err;
2450
2451 root_objectid = gang[i]->root_key.objectid;
2452 err = btrfs_orphan_cleanup(gang[i]);
2453 if (err)
2454 return err;
2455 }
2456 root_objectid++;
2457 }
2458 return 0;
2459 }
2460
2461 int btrfs_commit_super(struct btrfs_root *root)
2462 {
2463 struct btrfs_trans_handle *trans;
2464 int ret;
2465
2466 mutex_lock(&root->fs_info->cleaner_mutex);
2467 btrfs_run_delayed_iputs(root);
2468 btrfs_clean_old_snapshots(root);
2469 mutex_unlock(&root->fs_info->cleaner_mutex);
2470
2471 /* wait until ongoing cleanup work done */
2472 down_write(&root->fs_info->cleanup_work_sem);
2473 up_write(&root->fs_info->cleanup_work_sem);
2474
2475 trans = btrfs_join_transaction(root);
2476 if (IS_ERR(trans))
2477 return PTR_ERR(trans);
2478 ret = btrfs_commit_transaction(trans, root);
2479 BUG_ON(ret);
2480 /* run commit again to drop the original snapshot */
2481 trans = btrfs_join_transaction(root);
2482 if (IS_ERR(trans))
2483 return PTR_ERR(trans);
2484 btrfs_commit_transaction(trans, root);
2485 ret = btrfs_write_and_wait_transaction(NULL, root);
2486 BUG_ON(ret);
2487
2488 ret = write_ctree_super(NULL, root, 0);
2489 return ret;
2490 }
2491
2492 int close_ctree(struct btrfs_root *root)
2493 {
2494 struct btrfs_fs_info *fs_info = root->fs_info;
2495 int ret;
2496
2497 fs_info->closing = 1;
2498 smp_mb();
2499
2500 btrfs_scrub_cancel(root);
2501
2502 /* wait for any defraggers to finish */
2503 wait_event(fs_info->transaction_wait,
2504 (atomic_read(&fs_info->defrag_running) == 0));
2505
2506 /* clear out the rbtree of defraggable inodes */
2507 btrfs_run_defrag_inodes(root->fs_info);
2508
2509 btrfs_put_block_group_cache(fs_info);
2510
2511 /*
2512 * Here come 2 situations when btrfs is broken to flip readonly:
2513 *
2514 * 1. when btrfs flips readonly somewhere else before
2515 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2516 * and btrfs will skip to write sb directly to keep
2517 * ERROR state on disk.
2518 *
2519 * 2. when btrfs flips readonly just in btrfs_commit_super,
2520 * and in such case, btrfs cannot write sb via btrfs_commit_super,
2521 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2522 * btrfs will cleanup all FS resources first and write sb then.
2523 */
2524 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2525 ret = btrfs_commit_super(root);
2526 if (ret)
2527 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2528 }
2529
2530 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2531 ret = btrfs_error_commit_super(root);
2532 if (ret)
2533 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2534 }
2535
2536 kthread_stop(root->fs_info->transaction_kthread);
2537 kthread_stop(root->fs_info->cleaner_kthread);
2538
2539 fs_info->closing = 2;
2540 smp_mb();
2541
2542 if (fs_info->delalloc_bytes) {
2543 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2544 (unsigned long long)fs_info->delalloc_bytes);
2545 }
2546 if (fs_info->total_ref_cache_size) {
2547 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2548 (unsigned long long)fs_info->total_ref_cache_size);
2549 }
2550
2551 free_extent_buffer(fs_info->extent_root->node);
2552 free_extent_buffer(fs_info->extent_root->commit_root);
2553 free_extent_buffer(fs_info->tree_root->node);
2554 free_extent_buffer(fs_info->tree_root->commit_root);
2555 free_extent_buffer(root->fs_info->chunk_root->node);
2556 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2557 free_extent_buffer(root->fs_info->dev_root->node);
2558 free_extent_buffer(root->fs_info->dev_root->commit_root);
2559 free_extent_buffer(root->fs_info->csum_root->node);
2560 free_extent_buffer(root->fs_info->csum_root->commit_root);
2561
2562 btrfs_free_block_groups(root->fs_info);
2563
2564 del_fs_roots(fs_info);
2565
2566 iput(fs_info->btree_inode);
2567 kfree(fs_info->delayed_root);
2568
2569 btrfs_stop_workers(&fs_info->generic_worker);
2570 btrfs_stop_workers(&fs_info->fixup_workers);
2571 btrfs_stop_workers(&fs_info->delalloc_workers);
2572 btrfs_stop_workers(&fs_info->workers);
2573 btrfs_stop_workers(&fs_info->endio_workers);
2574 btrfs_stop_workers(&fs_info->endio_meta_workers);
2575 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2576 btrfs_stop_workers(&fs_info->endio_write_workers);
2577 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2578 btrfs_stop_workers(&fs_info->submit_workers);
2579 btrfs_stop_workers(&fs_info->delayed_workers);
2580
2581 btrfs_close_devices(fs_info->fs_devices);
2582 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2583
2584 bdi_destroy(&fs_info->bdi);
2585 cleanup_srcu_struct(&fs_info->subvol_srcu);
2586
2587 kfree(fs_info->extent_root);
2588 kfree(fs_info->tree_root);
2589 kfree(fs_info->chunk_root);
2590 kfree(fs_info->dev_root);
2591 kfree(fs_info->csum_root);
2592 kfree(fs_info);
2593
2594 return 0;
2595 }
2596
2597 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2598 {
2599 int ret;
2600 struct inode *btree_inode = buf->first_page->mapping->host;
2601
2602 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2603 NULL);
2604 if (!ret)
2605 return ret;
2606
2607 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2608 parent_transid);
2609 return !ret;
2610 }
2611
2612 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2613 {
2614 struct inode *btree_inode = buf->first_page->mapping->host;
2615 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2616 buf);
2617 }
2618
2619 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2620 {
2621 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2622 u64 transid = btrfs_header_generation(buf);
2623 struct inode *btree_inode = root->fs_info->btree_inode;
2624 int was_dirty;
2625
2626 btrfs_assert_tree_locked(buf);
2627 if (transid != root->fs_info->generation) {
2628 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2629 "found %llu running %llu\n",
2630 (unsigned long long)buf->start,
2631 (unsigned long long)transid,
2632 (unsigned long long)root->fs_info->generation);
2633 WARN_ON(1);
2634 }
2635 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2636 buf);
2637 if (!was_dirty) {
2638 spin_lock(&root->fs_info->delalloc_lock);
2639 root->fs_info->dirty_metadata_bytes += buf->len;
2640 spin_unlock(&root->fs_info->delalloc_lock);
2641 }
2642 }
2643
2644 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2645 {
2646 /*
2647 * looks as though older kernels can get into trouble with
2648 * this code, they end up stuck in balance_dirty_pages forever
2649 */
2650 u64 num_dirty;
2651 unsigned long thresh = 32 * 1024 * 1024;
2652
2653 if (current->flags & PF_MEMALLOC)
2654 return;
2655
2656 btrfs_balance_delayed_items(root);
2657
2658 num_dirty = root->fs_info->dirty_metadata_bytes;
2659
2660 if (num_dirty > thresh) {
2661 balance_dirty_pages_ratelimited_nr(
2662 root->fs_info->btree_inode->i_mapping, 1);
2663 }
2664 return;
2665 }
2666
2667 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2668 {
2669 /*
2670 * looks as though older kernels can get into trouble with
2671 * this code, they end up stuck in balance_dirty_pages forever
2672 */
2673 u64 num_dirty;
2674 unsigned long thresh = 32 * 1024 * 1024;
2675
2676 if (current->flags & PF_MEMALLOC)
2677 return;
2678
2679 num_dirty = root->fs_info->dirty_metadata_bytes;
2680
2681 if (num_dirty > thresh) {
2682 balance_dirty_pages_ratelimited_nr(
2683 root->fs_info->btree_inode->i_mapping, 1);
2684 }
2685 return;
2686 }
2687
2688 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2689 {
2690 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2691 int ret;
2692 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2693 if (ret == 0)
2694 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2695 return ret;
2696 }
2697
2698 int btree_lock_page_hook(struct page *page)
2699 {
2700 struct inode *inode = page->mapping->host;
2701 struct btrfs_root *root = BTRFS_I(inode)->root;
2702 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2703 struct extent_buffer *eb;
2704 unsigned long len;
2705 u64 bytenr = page_offset(page);
2706
2707 if (page->private == EXTENT_PAGE_PRIVATE)
2708 goto out;
2709
2710 len = page->private >> 2;
2711 eb = find_extent_buffer(io_tree, bytenr, len);
2712 if (!eb)
2713 goto out;
2714
2715 btrfs_tree_lock(eb);
2716 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2717
2718 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2719 spin_lock(&root->fs_info->delalloc_lock);
2720 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2721 root->fs_info->dirty_metadata_bytes -= eb->len;
2722 else
2723 WARN_ON(1);
2724 spin_unlock(&root->fs_info->delalloc_lock);
2725 }
2726
2727 btrfs_tree_unlock(eb);
2728 free_extent_buffer(eb);
2729 out:
2730 lock_page(page);
2731 return 0;
2732 }
2733
2734 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2735 int read_only)
2736 {
2737 if (read_only)
2738 return;
2739
2740 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2741 printk(KERN_WARNING "warning: mount fs with errors, "
2742 "running btrfsck is recommended\n");
2743 }
2744
2745 int btrfs_error_commit_super(struct btrfs_root *root)
2746 {
2747 int ret;
2748
2749 mutex_lock(&root->fs_info->cleaner_mutex);
2750 btrfs_run_delayed_iputs(root);
2751 mutex_unlock(&root->fs_info->cleaner_mutex);
2752
2753 down_write(&root->fs_info->cleanup_work_sem);
2754 up_write(&root->fs_info->cleanup_work_sem);
2755
2756 /* cleanup FS via transaction */
2757 btrfs_cleanup_transaction(root);
2758
2759 ret = write_ctree_super(NULL, root, 0);
2760
2761 return ret;
2762 }
2763
2764 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2765 {
2766 struct btrfs_inode *btrfs_inode;
2767 struct list_head splice;
2768
2769 INIT_LIST_HEAD(&splice);
2770
2771 mutex_lock(&root->fs_info->ordered_operations_mutex);
2772 spin_lock(&root->fs_info->ordered_extent_lock);
2773
2774 list_splice_init(&root->fs_info->ordered_operations, &splice);
2775 while (!list_empty(&splice)) {
2776 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2777 ordered_operations);
2778
2779 list_del_init(&btrfs_inode->ordered_operations);
2780
2781 btrfs_invalidate_inodes(btrfs_inode->root);
2782 }
2783
2784 spin_unlock(&root->fs_info->ordered_extent_lock);
2785 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2786
2787 return 0;
2788 }
2789
2790 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2791 {
2792 struct list_head splice;
2793 struct btrfs_ordered_extent *ordered;
2794 struct inode *inode;
2795
2796 INIT_LIST_HEAD(&splice);
2797
2798 spin_lock(&root->fs_info->ordered_extent_lock);
2799
2800 list_splice_init(&root->fs_info->ordered_extents, &splice);
2801 while (!list_empty(&splice)) {
2802 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2803 root_extent_list);
2804
2805 list_del_init(&ordered->root_extent_list);
2806 atomic_inc(&ordered->refs);
2807
2808 /* the inode may be getting freed (in sys_unlink path). */
2809 inode = igrab(ordered->inode);
2810
2811 spin_unlock(&root->fs_info->ordered_extent_lock);
2812 if (inode)
2813 iput(inode);
2814
2815 atomic_set(&ordered->refs, 1);
2816 btrfs_put_ordered_extent(ordered);
2817
2818 spin_lock(&root->fs_info->ordered_extent_lock);
2819 }
2820
2821 spin_unlock(&root->fs_info->ordered_extent_lock);
2822
2823 return 0;
2824 }
2825
2826 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2827 struct btrfs_root *root)
2828 {
2829 struct rb_node *node;
2830 struct btrfs_delayed_ref_root *delayed_refs;
2831 struct btrfs_delayed_ref_node *ref;
2832 int ret = 0;
2833
2834 delayed_refs = &trans->delayed_refs;
2835
2836 spin_lock(&delayed_refs->lock);
2837 if (delayed_refs->num_entries == 0) {
2838 spin_unlock(&delayed_refs->lock);
2839 printk(KERN_INFO "delayed_refs has NO entry\n");
2840 return ret;
2841 }
2842
2843 node = rb_first(&delayed_refs->root);
2844 while (node) {
2845 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2846 node = rb_next(node);
2847
2848 ref->in_tree = 0;
2849 rb_erase(&ref->rb_node, &delayed_refs->root);
2850 delayed_refs->num_entries--;
2851
2852 atomic_set(&ref->refs, 1);
2853 if (btrfs_delayed_ref_is_head(ref)) {
2854 struct btrfs_delayed_ref_head *head;
2855
2856 head = btrfs_delayed_node_to_head(ref);
2857 mutex_lock(&head->mutex);
2858 kfree(head->extent_op);
2859 delayed_refs->num_heads--;
2860 if (list_empty(&head->cluster))
2861 delayed_refs->num_heads_ready--;
2862 list_del_init(&head->cluster);
2863 mutex_unlock(&head->mutex);
2864 }
2865
2866 spin_unlock(&delayed_refs->lock);
2867 btrfs_put_delayed_ref(ref);
2868
2869 cond_resched();
2870 spin_lock(&delayed_refs->lock);
2871 }
2872
2873 spin_unlock(&delayed_refs->lock);
2874
2875 return ret;
2876 }
2877
2878 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2879 {
2880 struct btrfs_pending_snapshot *snapshot;
2881 struct list_head splice;
2882
2883 INIT_LIST_HEAD(&splice);
2884
2885 list_splice_init(&t->pending_snapshots, &splice);
2886
2887 while (!list_empty(&splice)) {
2888 snapshot = list_entry(splice.next,
2889 struct btrfs_pending_snapshot,
2890 list);
2891
2892 list_del_init(&snapshot->list);
2893
2894 kfree(snapshot);
2895 }
2896
2897 return 0;
2898 }
2899
2900 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2901 {
2902 struct btrfs_inode *btrfs_inode;
2903 struct list_head splice;
2904
2905 INIT_LIST_HEAD(&splice);
2906
2907 spin_lock(&root->fs_info->delalloc_lock);
2908 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2909
2910 while (!list_empty(&splice)) {
2911 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2912 delalloc_inodes);
2913
2914 list_del_init(&btrfs_inode->delalloc_inodes);
2915
2916 btrfs_invalidate_inodes(btrfs_inode->root);
2917 }
2918
2919 spin_unlock(&root->fs_info->delalloc_lock);
2920
2921 return 0;
2922 }
2923
2924 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2925 struct extent_io_tree *dirty_pages,
2926 int mark)
2927 {
2928 int ret;
2929 struct page *page;
2930 struct inode *btree_inode = root->fs_info->btree_inode;
2931 struct extent_buffer *eb;
2932 u64 start = 0;
2933 u64 end;
2934 u64 offset;
2935 unsigned long index;
2936
2937 while (1) {
2938 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2939 mark);
2940 if (ret)
2941 break;
2942
2943 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2944 while (start <= end) {
2945 index = start >> PAGE_CACHE_SHIFT;
2946 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2947 page = find_get_page(btree_inode->i_mapping, index);
2948 if (!page)
2949 continue;
2950 offset = page_offset(page);
2951
2952 spin_lock(&dirty_pages->buffer_lock);
2953 eb = radix_tree_lookup(
2954 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2955 offset >> PAGE_CACHE_SHIFT);
2956 spin_unlock(&dirty_pages->buffer_lock);
2957 if (eb) {
2958 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
2959 &eb->bflags);
2960 atomic_set(&eb->refs, 1);
2961 }
2962 if (PageWriteback(page))
2963 end_page_writeback(page);
2964
2965 lock_page(page);
2966 if (PageDirty(page)) {
2967 clear_page_dirty_for_io(page);
2968 spin_lock_irq(&page->mapping->tree_lock);
2969 radix_tree_tag_clear(&page->mapping->page_tree,
2970 page_index(page),
2971 PAGECACHE_TAG_DIRTY);
2972 spin_unlock_irq(&page->mapping->tree_lock);
2973 }
2974
2975 page->mapping->a_ops->invalidatepage(page, 0);
2976 unlock_page(page);
2977 }
2978 }
2979
2980 return ret;
2981 }
2982
2983 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
2984 struct extent_io_tree *pinned_extents)
2985 {
2986 struct extent_io_tree *unpin;
2987 u64 start;
2988 u64 end;
2989 int ret;
2990
2991 unpin = pinned_extents;
2992 while (1) {
2993 ret = find_first_extent_bit(unpin, 0, &start, &end,
2994 EXTENT_DIRTY);
2995 if (ret)
2996 break;
2997
2998 /* opt_discard */
2999 if (btrfs_test_opt(root, DISCARD))
3000 ret = btrfs_error_discard_extent(root, start,
3001 end + 1 - start,
3002 NULL);
3003
3004 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3005 btrfs_error_unpin_extent_range(root, start, end);
3006 cond_resched();
3007 }
3008
3009 return 0;
3010 }
3011
3012 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3013 {
3014 struct btrfs_transaction *t;
3015 LIST_HEAD(list);
3016
3017 WARN_ON(1);
3018
3019 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3020
3021 spin_lock(&root->fs_info->trans_lock);
3022 list_splice_init(&root->fs_info->trans_list, &list);
3023 root->fs_info->trans_no_join = 1;
3024 spin_unlock(&root->fs_info->trans_lock);
3025
3026 while (!list_empty(&list)) {
3027 t = list_entry(list.next, struct btrfs_transaction, list);
3028 if (!t)
3029 break;
3030
3031 btrfs_destroy_ordered_operations(root);
3032
3033 btrfs_destroy_ordered_extents(root);
3034
3035 btrfs_destroy_delayed_refs(t, root);
3036
3037 btrfs_block_rsv_release(root,
3038 &root->fs_info->trans_block_rsv,
3039 t->dirty_pages.dirty_bytes);
3040
3041 /* FIXME: cleanup wait for commit */
3042 t->in_commit = 1;
3043 t->blocked = 1;
3044 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3045 wake_up(&root->fs_info->transaction_blocked_wait);
3046
3047 t->blocked = 0;
3048 if (waitqueue_active(&root->fs_info->transaction_wait))
3049 wake_up(&root->fs_info->transaction_wait);
3050
3051 t->commit_done = 1;
3052 if (waitqueue_active(&t->commit_wait))
3053 wake_up(&t->commit_wait);
3054
3055 btrfs_destroy_pending_snapshots(t);
3056
3057 btrfs_destroy_delalloc_inodes(root);
3058
3059 spin_lock(&root->fs_info->trans_lock);
3060 root->fs_info->running_transaction = NULL;
3061 spin_unlock(&root->fs_info->trans_lock);
3062
3063 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3064 EXTENT_DIRTY);
3065
3066 btrfs_destroy_pinned_extent(root,
3067 root->fs_info->pinned_extents);
3068
3069 atomic_set(&t->use_count, 0);
3070 list_del_init(&t->list);
3071 memset(t, 0, sizeof(*t));
3072 kmem_cache_free(btrfs_transaction_cachep, t);
3073 }
3074
3075 spin_lock(&root->fs_info->trans_lock);
3076 root->fs_info->trans_no_join = 0;
3077 spin_unlock(&root->fs_info->trans_lock);
3078 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3079
3080 return 0;
3081 }
3082
3083 static struct extent_io_ops btree_extent_io_ops = {
3084 .write_cache_pages_lock_hook = btree_lock_page_hook,
3085 .readpage_end_io_hook = btree_readpage_end_io_hook,
3086 .submit_bio_hook = btree_submit_bio_hook,
3087 /* note we're sharing with inode.c for the merge bio hook */
3088 .merge_bio_hook = btrfs_merge_bio_hook,
3089 };
Linux 2.6 ibm-acpi/thinkpad-acpi driver git tree