?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: fix deadlock on async thread startup
[linux-2.6/kvm.git] / fs / btrfs / disk-io.c
blob9903f042765d9fb158ecf3ddf02c5a0033d6606b
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 "compat.h"
31 #include "ctree.h"
32 #include "disk-io.h"
33 #include "transaction.h"
34 #include "btrfs_inode.h"
35 #include "volumes.h"
36 #include "print-tree.h"
37 #include "async-thread.h"
38 #include "locking.h"
39 #include "tree-log.h"
40 #include "free-space-cache.h"
41
42 static struct extent_io_ops btree_extent_io_ops;
43 static void end_workqueue_fn(struct btrfs_work *work);
44 static void free_fs_root(struct btrfs_root *root);
45
46 static atomic_t btrfs_bdi_num = ATOMIC_INIT(0);
47
48 /*
49 * end_io_wq structs are used to do processing in task context when an IO is
50 * complete. This is used during reads to verify checksums, and it is used
51 * by writes to insert metadata for new file extents after IO is complete.
52 */
53 struct end_io_wq {
54 struct bio *bio;
55 bio_end_io_t *end_io;
56 void *private;
57 struct btrfs_fs_info *info;
58 int error;
59 int metadata;
60 struct list_head list;
61 struct btrfs_work work;
62 };
63
64 /*
65 * async submit bios are used to offload expensive checksumming
66 * onto the worker threads. They checksum file and metadata bios
67 * just before they are sent down the IO stack.
68 */
69 struct async_submit_bio {
70 struct inode *inode;
71 struct bio *bio;
72 struct list_head list;
73 extent_submit_bio_hook_t *submit_bio_start;
74 extent_submit_bio_hook_t *submit_bio_done;
75 int rw;
76 int mirror_num;
77 unsigned long bio_flags;
78 struct btrfs_work work;
79 };
80
81 /* These are used to set the lockdep class on the extent buffer locks.
82 * The class is set by the readpage_end_io_hook after the buffer has
83 * passed csum validation but before the pages are unlocked.
84 *
85 * The lockdep class is also set by btrfs_init_new_buffer on freshly
86 * allocated blocks.
87 *
88 * The class is based on the level in the tree block, which allows lockdep
89 * to know that lower nodes nest inside the locks of higher nodes.
90 *
91 * We also add a check to make sure the highest level of the tree is
92 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
93 * code needs update as well.
94 */
95 #ifdef CONFIG_DEBUG_LOCK_ALLOC
96 # if BTRFS_MAX_LEVEL != 8
97 # error
98 # endif
99 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
100 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
101 /* leaf */
102 "btrfs-extent-00",
103 "btrfs-extent-01",
104 "btrfs-extent-02",
105 "btrfs-extent-03",
106 "btrfs-extent-04",
107 "btrfs-extent-05",
108 "btrfs-extent-06",
109 "btrfs-extent-07",
110 /* highest possible level */
111 "btrfs-extent-08",
112 };
113 #endif
114
115 /*
116 * extents on the btree inode are pretty simple, there's one extent
117 * that covers the entire device
118 */
119 static struct extent_map *btree_get_extent(struct inode *inode,
120 struct page *page, size_t page_offset, u64 start, u64 len,
121 int create)
122 {
123 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
124 struct extent_map *em;
125 int ret;
126
127 read_lock(&em_tree->lock);
128 em = lookup_extent_mapping(em_tree, start, len);
129 if (em) {
130 em->bdev =
131 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
132 read_unlock(&em_tree->lock);
133 goto out;
134 }
135 read_unlock(&em_tree->lock);
136
137 em = alloc_extent_map(GFP_NOFS);
138 if (!em) {
139 em = ERR_PTR(-ENOMEM);
140 goto out;
141 }
142 em->start = 0;
143 em->len = (u64)-1;
144 em->block_len = (u64)-1;
145 em->block_start = 0;
146 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
147
148 write_lock(&em_tree->lock);
149 ret = add_extent_mapping(em_tree, em);
150 if (ret == -EEXIST) {
151 u64 failed_start = em->start;
152 u64 failed_len = em->len;
153
154 free_extent_map(em);
155 em = lookup_extent_mapping(em_tree, start, len);
156 if (em) {
157 ret = 0;
158 } else {
159 em = lookup_extent_mapping(em_tree, failed_start,
160 failed_len);
161 ret = -EIO;
162 }
163 } else if (ret) {
164 free_extent_map(em);
165 em = NULL;
166 }
167 write_unlock(&em_tree->lock);
168
169 if (ret)
170 em = ERR_PTR(ret);
171 out:
172 return em;
173 }
174
175 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
176 {
177 return crc32c(seed, data, len);
178 }
179
180 void btrfs_csum_final(u32 crc, char *result)
181 {
182 *(__le32 *)result = ~cpu_to_le32(crc);
183 }
184
185 /*
186 * compute the csum for a btree block, and either verify it or write it
187 * into the csum field of the block.
188 */
189 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
190 int verify)
191 {
192 u16 csum_size =
193 btrfs_super_csum_size(&root->fs_info->super_copy);
194 char *result = NULL;
195 unsigned long len;
196 unsigned long cur_len;
197 unsigned long offset = BTRFS_CSUM_SIZE;
198 char *map_token = NULL;
199 char *kaddr;
200 unsigned long map_start;
201 unsigned long map_len;
202 int err;
203 u32 crc = ~(u32)0;
204 unsigned long inline_result;
205
206 len = buf->len - offset;
207 while (len > 0) {
208 err = map_private_extent_buffer(buf, offset, 32,
209 &map_token, &kaddr,
210 &map_start, &map_len, KM_USER0);
211 if (err)
212 return 1;
213 cur_len = min(len, map_len - (offset - map_start));
214 crc = btrfs_csum_data(root, kaddr + offset - map_start,
215 crc, cur_len);
216 len -= cur_len;
217 offset += cur_len;
218 unmap_extent_buffer(buf, map_token, KM_USER0);
219 }
220 if (csum_size > sizeof(inline_result)) {
221 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
222 if (!result)
223 return 1;
224 } else {
225 result = (char *)&inline_result;
226 }
227
228 btrfs_csum_final(crc, result);
229
230 if (verify) {
231 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
232 u32 val;
233 u32 found = 0;
234 memcpy(&found, result, csum_size);
235
236 read_extent_buffer(buf, &val, 0, csum_size);
237 if (printk_ratelimit()) {
238 printk(KERN_INFO "btrfs: %s checksum verify "
239 "failed on %llu wanted %X found %X "
240 "level %d\n",
241 root->fs_info->sb->s_id,
242 (unsigned long long)buf->start, val, found,
243 btrfs_header_level(buf));
244 }
245 if (result != (char *)&inline_result)
246 kfree(result);
247 return 1;
248 }
249 } else {
250 write_extent_buffer(buf, result, 0, csum_size);
251 }
252 if (result != (char *)&inline_result)
253 kfree(result);
254 return 0;
255 }
256
257 /*
258 * we can't consider a given block up to date unless the transid of the
259 * block matches the transid in the parent node's pointer. This is how we
260 * detect blocks that either didn't get written at all or got written
261 * in the wrong place.
262 */
263 static int verify_parent_transid(struct extent_io_tree *io_tree,
264 struct extent_buffer *eb, u64 parent_transid)
265 {
266 int ret;
267
268 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
269 return 0;
270
271 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
272 if (extent_buffer_uptodate(io_tree, eb) &&
273 btrfs_header_generation(eb) == parent_transid) {
274 ret = 0;
275 goto out;
276 }
277 if (printk_ratelimit()) {
278 printk("parent transid verify failed on %llu wanted %llu "
279 "found %llu\n",
280 (unsigned long long)eb->start,
281 (unsigned long long)parent_transid,
282 (unsigned long long)btrfs_header_generation(eb));
283 }
284 ret = 1;
285 clear_extent_buffer_uptodate(io_tree, eb);
286 out:
287 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
288 GFP_NOFS);
289 return ret;
290 }
291
292 /*
293 * helper to read a given tree block, doing retries as required when
294 * the checksums don't match and we have alternate mirrors to try.
295 */
296 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
297 struct extent_buffer *eb,
298 u64 start, u64 parent_transid)
299 {
300 struct extent_io_tree *io_tree;
301 int ret;
302 int num_copies = 0;
303 int mirror_num = 0;
304
305 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
306 while (1) {
307 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
308 btree_get_extent, mirror_num);
309 if (!ret &&
310 !verify_parent_transid(io_tree, eb, parent_transid))
311 return ret;
312
313 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
314 eb->start, eb->len);
315 if (num_copies == 1)
316 return ret;
317
318 mirror_num++;
319 if (mirror_num > num_copies)
320 return ret;
321 }
322 return -EIO;
323 }
324
325 /*
326 * checksum a dirty tree block before IO. This has extra checks to make sure
327 * we only fill in the checksum field in the first page of a multi-page block
328 */
329
330 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
331 {
332 struct extent_io_tree *tree;
333 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
334 u64 found_start;
335 int found_level;
336 unsigned long len;
337 struct extent_buffer *eb;
338 int ret;
339
340 tree = &BTRFS_I(page->mapping->host)->io_tree;
341
342 if (page->private == EXTENT_PAGE_PRIVATE)
343 goto out;
344 if (!page->private)
345 goto out;
346 len = page->private >> 2;
347 WARN_ON(len == 0);
348
349 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
350 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
351 btrfs_header_generation(eb));
352 BUG_ON(ret);
353 found_start = btrfs_header_bytenr(eb);
354 if (found_start != start) {
355 WARN_ON(1);
356 goto err;
357 }
358 if (eb->first_page != page) {
359 WARN_ON(1);
360 goto err;
361 }
362 if (!PageUptodate(page)) {
363 WARN_ON(1);
364 goto err;
365 }
366 found_level = btrfs_header_level(eb);
367
368 csum_tree_block(root, eb, 0);
369 err:
370 free_extent_buffer(eb);
371 out:
372 return 0;
373 }
374
375 static int check_tree_block_fsid(struct btrfs_root *root,
376 struct extent_buffer *eb)
377 {
378 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
379 u8 fsid[BTRFS_UUID_SIZE];
380 int ret = 1;
381
382 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
383 BTRFS_FSID_SIZE);
384 while (fs_devices) {
385 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
386 ret = 0;
387 break;
388 }
389 fs_devices = fs_devices->seed;
390 }
391 return ret;
392 }
393
394 #ifdef CONFIG_DEBUG_LOCK_ALLOC
395 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
396 {
397 lockdep_set_class_and_name(&eb->lock,
398 &btrfs_eb_class[level],
399 btrfs_eb_name[level]);
400 }
401 #endif
402
403 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
404 struct extent_state *state)
405 {
406 struct extent_io_tree *tree;
407 u64 found_start;
408 int found_level;
409 unsigned long len;
410 struct extent_buffer *eb;
411 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
412 int ret = 0;
413
414 tree = &BTRFS_I(page->mapping->host)->io_tree;
415 if (page->private == EXTENT_PAGE_PRIVATE)
416 goto out;
417 if (!page->private)
418 goto out;
419
420 len = page->private >> 2;
421 WARN_ON(len == 0);
422
423 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
424
425 found_start = btrfs_header_bytenr(eb);
426 if (found_start != start) {
427 if (printk_ratelimit()) {
428 printk(KERN_INFO "btrfs bad tree block start "
429 "%llu %llu\n",
430 (unsigned long long)found_start,
431 (unsigned long long)eb->start);
432 }
433 ret = -EIO;
434 goto err;
435 }
436 if (eb->first_page != page) {
437 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
438 eb->first_page->index, page->index);
439 WARN_ON(1);
440 ret = -EIO;
441 goto err;
442 }
443 if (check_tree_block_fsid(root, eb)) {
444 if (printk_ratelimit()) {
445 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
446 (unsigned long long)eb->start);
447 }
448 ret = -EIO;
449 goto err;
450 }
451 found_level = btrfs_header_level(eb);
452
453 btrfs_set_buffer_lockdep_class(eb, found_level);
454
455 ret = csum_tree_block(root, eb, 1);
456 if (ret)
457 ret = -EIO;
458
459 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
460 end = eb->start + end - 1;
461 err:
462 free_extent_buffer(eb);
463 out:
464 return ret;
465 }
466
467 static void end_workqueue_bio(struct bio *bio, int err)
468 {
469 struct end_io_wq *end_io_wq = bio->bi_private;
470 struct btrfs_fs_info *fs_info;
471
472 fs_info = end_io_wq->info;
473 end_io_wq->error = err;
474 end_io_wq->work.func = end_workqueue_fn;
475 end_io_wq->work.flags = 0;
476
477 if (bio->bi_rw & (1 << BIO_RW)) {
478 if (end_io_wq->metadata)
479 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
480 &end_io_wq->work);
481 else
482 btrfs_queue_worker(&fs_info->endio_write_workers,
483 &end_io_wq->work);
484 } else {
485 if (end_io_wq->metadata)
486 btrfs_queue_worker(&fs_info->endio_meta_workers,
487 &end_io_wq->work);
488 else
489 btrfs_queue_worker(&fs_info->endio_workers,
490 &end_io_wq->work);
491 }
492 }
493
494 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
495 int metadata)
496 {
497 struct end_io_wq *end_io_wq;
498 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
499 if (!end_io_wq)
500 return -ENOMEM;
501
502 end_io_wq->private = bio->bi_private;
503 end_io_wq->end_io = bio->bi_end_io;
504 end_io_wq->info = info;
505 end_io_wq->error = 0;
506 end_io_wq->bio = bio;
507 end_io_wq->metadata = metadata;
508
509 bio->bi_private = end_io_wq;
510 bio->bi_end_io = end_workqueue_bio;
511 return 0;
512 }
513
514 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
515 {
516 unsigned long limit = min_t(unsigned long,
517 info->workers.max_workers,
518 info->fs_devices->open_devices);
519 return 256 * limit;
520 }
521
522 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
523 {
524 return atomic_read(&info->nr_async_bios) >
525 btrfs_async_submit_limit(info);
526 }
527
528 static void run_one_async_start(struct btrfs_work *work)
529 {
530 struct btrfs_fs_info *fs_info;
531 struct async_submit_bio *async;
532
533 async = container_of(work, struct async_submit_bio, work);
534 fs_info = BTRFS_I(async->inode)->root->fs_info;
535 async->submit_bio_start(async->inode, async->rw, async->bio,
536 async->mirror_num, async->bio_flags);
537 }
538
539 static void run_one_async_done(struct btrfs_work *work)
540 {
541 struct btrfs_fs_info *fs_info;
542 struct async_submit_bio *async;
543 int limit;
544
545 async = container_of(work, struct async_submit_bio, work);
546 fs_info = BTRFS_I(async->inode)->root->fs_info;
547
548 limit = btrfs_async_submit_limit(fs_info);
549 limit = limit * 2 / 3;
550
551 atomic_dec(&fs_info->nr_async_submits);
552
553 if (atomic_read(&fs_info->nr_async_submits) < limit &&
554 waitqueue_active(&fs_info->async_submit_wait))
555 wake_up(&fs_info->async_submit_wait);
556
557 async->submit_bio_done(async->inode, async->rw, async->bio,
558 async->mirror_num, async->bio_flags);
559 }
560
561 static void run_one_async_free(struct btrfs_work *work)
562 {
563 struct async_submit_bio *async;
564
565 async = container_of(work, struct async_submit_bio, work);
566 kfree(async);
567 }
568
569 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
570 int rw, struct bio *bio, int mirror_num,
571 unsigned long bio_flags,
572 extent_submit_bio_hook_t *submit_bio_start,
573 extent_submit_bio_hook_t *submit_bio_done)
574 {
575 struct async_submit_bio *async;
576
577 async = kmalloc(sizeof(*async), GFP_NOFS);
578 if (!async)
579 return -ENOMEM;
580
581 async->inode = inode;
582 async->rw = rw;
583 async->bio = bio;
584 async->mirror_num = mirror_num;
585 async->submit_bio_start = submit_bio_start;
586 async->submit_bio_done = submit_bio_done;
587
588 async->work.func = run_one_async_start;
589 async->work.ordered_func = run_one_async_done;
590 async->work.ordered_free = run_one_async_free;
591
592 async->work.flags = 0;
593 async->bio_flags = bio_flags;
594
595 atomic_inc(&fs_info->nr_async_submits);
596
597 if (rw & (1 << BIO_RW_SYNCIO))
598 btrfs_set_work_high_prio(&async->work);
599
600 btrfs_queue_worker(&fs_info->workers, &async->work);
601
602 while (atomic_read(&fs_info->async_submit_draining) &&
603 atomic_read(&fs_info->nr_async_submits)) {
604 wait_event(fs_info->async_submit_wait,
605 (atomic_read(&fs_info->nr_async_submits) == 0));
606 }
607
608 return 0;
609 }
610
611 static int btree_csum_one_bio(struct bio *bio)
612 {
613 struct bio_vec *bvec = bio->bi_io_vec;
614 int bio_index = 0;
615 struct btrfs_root *root;
616
617 WARN_ON(bio->bi_vcnt <= 0);
618 while (bio_index < bio->bi_vcnt) {
619 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
620 csum_dirty_buffer(root, bvec->bv_page);
621 bio_index++;
622 bvec++;
623 }
624 return 0;
625 }
626
627 static int __btree_submit_bio_start(struct inode *inode, int rw,
628 struct bio *bio, int mirror_num,
629 unsigned long bio_flags)
630 {
631 /*
632 * when we're called for a write, we're already in the async
633 * submission context. Just jump into btrfs_map_bio
634 */
635 btree_csum_one_bio(bio);
636 return 0;
637 }
638
639 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
640 int mirror_num, unsigned long bio_flags)
641 {
642 /*
643 * when we're called for a write, we're already in the async
644 * submission context. Just jump into btrfs_map_bio
645 */
646 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
647 }
648
649 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
650 int mirror_num, unsigned long bio_flags)
651 {
652 int ret;
653
654 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
655 bio, 1);
656 BUG_ON(ret);
657
658 if (!(rw & (1 << BIO_RW))) {
659 /*
660 * called for a read, do the setup so that checksum validation
661 * can happen in the async kernel threads
662 */
663 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
664 mirror_num, 0);
665 }
666
667 /*
668 * kthread helpers are used to submit writes so that checksumming
669 * can happen in parallel across all CPUs
670 */
671 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
672 inode, rw, bio, mirror_num, 0,
673 __btree_submit_bio_start,
674 __btree_submit_bio_done);
675 }
676
677 static int btree_writepage(struct page *page, struct writeback_control *wbc)
678 {
679 struct extent_io_tree *tree;
680 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
681 struct extent_buffer *eb;
682 int was_dirty;
683
684 tree = &BTRFS_I(page->mapping->host)->io_tree;
685 if (!(current->flags & PF_MEMALLOC)) {
686 return extent_write_full_page(tree, page,
687 btree_get_extent, wbc);
688 }
689
690 redirty_page_for_writepage(wbc, page);
691 eb = btrfs_find_tree_block(root, page_offset(page),
692 PAGE_CACHE_SIZE);
693 WARN_ON(!eb);
694
695 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
696 if (!was_dirty) {
697 spin_lock(&root->fs_info->delalloc_lock);
698 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
699 spin_unlock(&root->fs_info->delalloc_lock);
700 }
701 free_extent_buffer(eb);
702
703 unlock_page(page);
704 return 0;
705 }
706
707 static int btree_writepages(struct address_space *mapping,
708 struct writeback_control *wbc)
709 {
710 struct extent_io_tree *tree;
711 tree = &BTRFS_I(mapping->host)->io_tree;
712 if (wbc->sync_mode == WB_SYNC_NONE) {
713 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
714 u64 num_dirty;
715 unsigned long thresh = 32 * 1024 * 1024;
716
717 if (wbc->for_kupdate)
718 return 0;
719
720 /* this is a bit racy, but that's ok */
721 num_dirty = root->fs_info->dirty_metadata_bytes;
722 if (num_dirty < thresh)
723 return 0;
724 }
725 return extent_writepages(tree, mapping, btree_get_extent, wbc);
726 }
727
728 static int btree_readpage(struct file *file, struct page *page)
729 {
730 struct extent_io_tree *tree;
731 tree = &BTRFS_I(page->mapping->host)->io_tree;
732 return extent_read_full_page(tree, page, btree_get_extent);
733 }
734
735 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
736 {
737 struct extent_io_tree *tree;
738 struct extent_map_tree *map;
739 int ret;
740
741 if (PageWriteback(page) || PageDirty(page))
742 return 0;
743
744 tree = &BTRFS_I(page->mapping->host)->io_tree;
745 map = &BTRFS_I(page->mapping->host)->extent_tree;
746
747 ret = try_release_extent_state(map, tree, page, gfp_flags);
748 if (!ret)
749 return 0;
750
751 ret = try_release_extent_buffer(tree, page);
752 if (ret == 1) {
753 ClearPagePrivate(page);
754 set_page_private(page, 0);
755 page_cache_release(page);
756 }
757
758 return ret;
759 }
760
761 static void btree_invalidatepage(struct page *page, unsigned long offset)
762 {
763 struct extent_io_tree *tree;
764 tree = &BTRFS_I(page->mapping->host)->io_tree;
765 extent_invalidatepage(tree, page, offset);
766 btree_releasepage(page, GFP_NOFS);
767 if (PagePrivate(page)) {
768 printk(KERN_WARNING "btrfs warning page private not zero "
769 "on page %llu\n", (unsigned long long)page_offset(page));
770 ClearPagePrivate(page);
771 set_page_private(page, 0);
772 page_cache_release(page);
773 }
774 }
775
776 static struct address_space_operations btree_aops = {
777 .readpage = btree_readpage,
778 .writepage = btree_writepage,
779 .writepages = btree_writepages,
780 .releasepage = btree_releasepage,
781 .invalidatepage = btree_invalidatepage,
782 .sync_page = block_sync_page,
783 };
784
785 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
786 u64 parent_transid)
787 {
788 struct extent_buffer *buf = NULL;
789 struct inode *btree_inode = root->fs_info->btree_inode;
790 int ret = 0;
791
792 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
793 if (!buf)
794 return 0;
795 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
796 buf, 0, 0, btree_get_extent, 0);
797 free_extent_buffer(buf);
798 return ret;
799 }
800
801 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
802 u64 bytenr, u32 blocksize)
803 {
804 struct inode *btree_inode = root->fs_info->btree_inode;
805 struct extent_buffer *eb;
806 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
807 bytenr, blocksize, GFP_NOFS);
808 return eb;
809 }
810
811 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
812 u64 bytenr, u32 blocksize)
813 {
814 struct inode *btree_inode = root->fs_info->btree_inode;
815 struct extent_buffer *eb;
816
817 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
818 bytenr, blocksize, NULL, GFP_NOFS);
819 return eb;
820 }
821
822
823 int btrfs_write_tree_block(struct extent_buffer *buf)
824 {
825 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
826 buf->start + buf->len - 1, WB_SYNC_ALL);
827 }
828
829 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
830 {
831 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
832 buf->start >> PAGE_CACHE_SHIFT,
833 (buf->start + buf->len - 1) >>
834 PAGE_CACHE_SHIFT);
835 }
836
837 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
838 u32 blocksize, u64 parent_transid)
839 {
840 struct extent_buffer *buf = NULL;
841 struct inode *btree_inode = root->fs_info->btree_inode;
842 struct extent_io_tree *io_tree;
843 int ret;
844
845 io_tree = &BTRFS_I(btree_inode)->io_tree;
846
847 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
848 if (!buf)
849 return NULL;
850
851 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
852
853 if (ret == 0)
854 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
855 return buf;
856
857 }
858
859 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
860 struct extent_buffer *buf)
861 {
862 struct inode *btree_inode = root->fs_info->btree_inode;
863 if (btrfs_header_generation(buf) ==
864 root->fs_info->running_transaction->transid) {
865 btrfs_assert_tree_locked(buf);
866
867 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
868 spin_lock(&root->fs_info->delalloc_lock);
869 if (root->fs_info->dirty_metadata_bytes >= buf->len)
870 root->fs_info->dirty_metadata_bytes -= buf->len;
871 else
872 WARN_ON(1);
873 spin_unlock(&root->fs_info->delalloc_lock);
874 }
875
876 /* ugh, clear_extent_buffer_dirty needs to lock the page */
877 btrfs_set_lock_blocking(buf);
878 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
879 buf);
880 }
881 return 0;
882 }
883
884 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
885 u32 stripesize, struct btrfs_root *root,
886 struct btrfs_fs_info *fs_info,
887 u64 objectid)
888 {
889 root->node = NULL;
890 root->commit_root = NULL;
891 root->sectorsize = sectorsize;
892 root->nodesize = nodesize;
893 root->leafsize = leafsize;
894 root->stripesize = stripesize;
895 root->ref_cows = 0;
896 root->track_dirty = 0;
897
898 root->fs_info = fs_info;
899 root->objectid = objectid;
900 root->last_trans = 0;
901 root->highest_objectid = 0;
902 root->name = NULL;
903 root->in_sysfs = 0;
904 root->inode_tree.rb_node = NULL;
905
906 INIT_LIST_HEAD(&root->dirty_list);
907 INIT_LIST_HEAD(&root->orphan_list);
908 INIT_LIST_HEAD(&root->root_list);
909 spin_lock_init(&root->node_lock);
910 spin_lock_init(&root->list_lock);
911 spin_lock_init(&root->inode_lock);
912 mutex_init(&root->objectid_mutex);
913 mutex_init(&root->log_mutex);
914 init_waitqueue_head(&root->log_writer_wait);
915 init_waitqueue_head(&root->log_commit_wait[0]);
916 init_waitqueue_head(&root->log_commit_wait[1]);
917 atomic_set(&root->log_commit[0], 0);
918 atomic_set(&root->log_commit[1], 0);
919 atomic_set(&root->log_writers, 0);
920 root->log_batch = 0;
921 root->log_transid = 0;
922 extent_io_tree_init(&root->dirty_log_pages,
923 fs_info->btree_inode->i_mapping, GFP_NOFS);
924
925 memset(&root->root_key, 0, sizeof(root->root_key));
926 memset(&root->root_item, 0, sizeof(root->root_item));
927 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
928 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
929 root->defrag_trans_start = fs_info->generation;
930 init_completion(&root->kobj_unregister);
931 root->defrag_running = 0;
932 root->defrag_level = 0;
933 root->root_key.objectid = objectid;
934 root->anon_super.s_root = NULL;
935 root->anon_super.s_dev = 0;
936 INIT_LIST_HEAD(&root->anon_super.s_list);
937 INIT_LIST_HEAD(&root->anon_super.s_instances);
938 init_rwsem(&root->anon_super.s_umount);
939
940 return 0;
941 }
942
943 static int find_and_setup_root(struct btrfs_root *tree_root,
944 struct btrfs_fs_info *fs_info,
945 u64 objectid,
946 struct btrfs_root *root)
947 {
948 int ret;
949 u32 blocksize;
950 u64 generation;
951
952 __setup_root(tree_root->nodesize, tree_root->leafsize,
953 tree_root->sectorsize, tree_root->stripesize,
954 root, fs_info, objectid);
955 ret = btrfs_find_last_root(tree_root, objectid,
956 &root->root_item, &root->root_key);
957 if (ret > 0)
958 return -ENOENT;
959 BUG_ON(ret);
960
961 generation = btrfs_root_generation(&root->root_item);
962 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
963 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
964 blocksize, generation);
965 BUG_ON(!root->node);
966 root->commit_root = btrfs_root_node(root);
967 return 0;
968 }
969
970 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
971 struct btrfs_fs_info *fs_info)
972 {
973 struct extent_buffer *eb;
974 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
975 u64 start = 0;
976 u64 end = 0;
977 int ret;
978
979 if (!log_root_tree)
980 return 0;
981
982 while (1) {
983 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
984 0, &start, &end, EXTENT_DIRTY);
985 if (ret)
986 break;
987
988 clear_extent_dirty(&log_root_tree->dirty_log_pages,
989 start, end, GFP_NOFS);
990 }
991 eb = fs_info->log_root_tree->node;
992
993 WARN_ON(btrfs_header_level(eb) != 0);
994 WARN_ON(btrfs_header_nritems(eb) != 0);
995
996 ret = btrfs_free_reserved_extent(fs_info->tree_root,
997 eb->start, eb->len);
998 BUG_ON(ret);
999
1000 free_extent_buffer(eb);
1001 kfree(fs_info->log_root_tree);
1002 fs_info->log_root_tree = NULL;
1003 return 0;
1004 }
1005
1006 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1007 struct btrfs_fs_info *fs_info)
1008 {
1009 struct btrfs_root *root;
1010 struct btrfs_root *tree_root = fs_info->tree_root;
1011 struct extent_buffer *leaf;
1012
1013 root = kzalloc(sizeof(*root), GFP_NOFS);
1014 if (!root)
1015 return ERR_PTR(-ENOMEM);
1016
1017 __setup_root(tree_root->nodesize, tree_root->leafsize,
1018 tree_root->sectorsize, tree_root->stripesize,
1019 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1020
1021 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1022 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1023 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1024 /*
1025 * log trees do not get reference counted because they go away
1026 * before a real commit is actually done. They do store pointers
1027 * to file data extents, and those reference counts still get
1028 * updated (along with back refs to the log tree).
1029 */
1030 root->ref_cows = 0;
1031
1032 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1033 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1034 if (IS_ERR(leaf)) {
1035 kfree(root);
1036 return ERR_CAST(leaf);
1037 }
1038
1039 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1040 btrfs_set_header_bytenr(leaf, leaf->start);
1041 btrfs_set_header_generation(leaf, trans->transid);
1042 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1043 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1044 root->node = leaf;
1045
1046 write_extent_buffer(root->node, root->fs_info->fsid,
1047 (unsigned long)btrfs_header_fsid(root->node),
1048 BTRFS_FSID_SIZE);
1049 btrfs_mark_buffer_dirty(root->node);
1050 btrfs_tree_unlock(root->node);
1051 return root;
1052 }
1053
1054 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1055 struct btrfs_fs_info *fs_info)
1056 {
1057 struct btrfs_root *log_root;
1058
1059 log_root = alloc_log_tree(trans, fs_info);
1060 if (IS_ERR(log_root))
1061 return PTR_ERR(log_root);
1062 WARN_ON(fs_info->log_root_tree);
1063 fs_info->log_root_tree = log_root;
1064 return 0;
1065 }
1066
1067 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1068 struct btrfs_root *root)
1069 {
1070 struct btrfs_root *log_root;
1071 struct btrfs_inode_item *inode_item;
1072
1073 log_root = alloc_log_tree(trans, root->fs_info);
1074 if (IS_ERR(log_root))
1075 return PTR_ERR(log_root);
1076
1077 log_root->last_trans = trans->transid;
1078 log_root->root_key.offset = root->root_key.objectid;
1079
1080 inode_item = &log_root->root_item.inode;
1081 inode_item->generation = cpu_to_le64(1);
1082 inode_item->size = cpu_to_le64(3);
1083 inode_item->nlink = cpu_to_le32(1);
1084 inode_item->nbytes = cpu_to_le64(root->leafsize);
1085 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1086
1087 btrfs_set_root_node(&log_root->root_item, log_root->node);
1088
1089 WARN_ON(root->log_root);
1090 root->log_root = log_root;
1091 root->log_transid = 0;
1092 return 0;
1093 }
1094
1095 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1096 struct btrfs_key *location)
1097 {
1098 struct btrfs_root *root;
1099 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1100 struct btrfs_path *path;
1101 struct extent_buffer *l;
1102 u64 generation;
1103 u32 blocksize;
1104 int ret = 0;
1105
1106 root = kzalloc(sizeof(*root), GFP_NOFS);
1107 if (!root)
1108 return ERR_PTR(-ENOMEM);
1109 if (location->offset == (u64)-1) {
1110 ret = find_and_setup_root(tree_root, fs_info,
1111 location->objectid, root);
1112 if (ret) {
1113 kfree(root);
1114 return ERR_PTR(ret);
1115 }
1116 goto out;
1117 }
1118
1119 __setup_root(tree_root->nodesize, tree_root->leafsize,
1120 tree_root->sectorsize, tree_root->stripesize,
1121 root, fs_info, location->objectid);
1122
1123 path = btrfs_alloc_path();
1124 BUG_ON(!path);
1125 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1126 if (ret == 0) {
1127 l = path->nodes[0];
1128 read_extent_buffer(l, &root->root_item,
1129 btrfs_item_ptr_offset(l, path->slots[0]),
1130 sizeof(root->root_item));
1131 memcpy(&root->root_key, location, sizeof(*location));
1132 }
1133 btrfs_free_path(path);
1134 if (ret) {
1135 if (ret > 0)
1136 ret = -ENOENT;
1137 return ERR_PTR(ret);
1138 }
1139
1140 generation = btrfs_root_generation(&root->root_item);
1141 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1142 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1143 blocksize, generation);
1144 root->commit_root = btrfs_root_node(root);
1145 BUG_ON(!root->node);
1146 out:
1147 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1148 root->ref_cows = 1;
1149
1150 return root;
1151 }
1152
1153 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1154 u64 root_objectid)
1155 {
1156 struct btrfs_root *root;
1157
1158 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1159 return fs_info->tree_root;
1160 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1161 return fs_info->extent_root;
1162
1163 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1164 (unsigned long)root_objectid);
1165 return root;
1166 }
1167
1168 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1169 struct btrfs_key *location)
1170 {
1171 struct btrfs_root *root;
1172 int ret;
1173
1174 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1175 return fs_info->tree_root;
1176 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1177 return fs_info->extent_root;
1178 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1179 return fs_info->chunk_root;
1180 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1181 return fs_info->dev_root;
1182 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1183 return fs_info->csum_root;
1184 again:
1185 spin_lock(&fs_info->fs_roots_radix_lock);
1186 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1187 (unsigned long)location->objectid);
1188 spin_unlock(&fs_info->fs_roots_radix_lock);
1189 if (root)
1190 return root;
1191
1192 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1193 if (ret == 0)
1194 ret = -ENOENT;
1195 if (ret < 0)
1196 return ERR_PTR(ret);
1197
1198 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1199 if (IS_ERR(root))
1200 return root;
1201
1202 WARN_ON(btrfs_root_refs(&root->root_item) == 0);
1203 set_anon_super(&root->anon_super, NULL);
1204
1205 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1206 if (ret)
1207 goto fail;
1208
1209 spin_lock(&fs_info->fs_roots_radix_lock);
1210 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1211 (unsigned long)root->root_key.objectid,
1212 root);
1213 if (ret == 0)
1214 root->in_radix = 1;
1215 spin_unlock(&fs_info->fs_roots_radix_lock);
1216 radix_tree_preload_end();
1217 if (ret) {
1218 if (ret == -EEXIST) {
1219 free_fs_root(root);
1220 goto again;
1221 }
1222 goto fail;
1223 }
1224
1225 ret = btrfs_find_dead_roots(fs_info->tree_root,
1226 root->root_key.objectid);
1227 WARN_ON(ret);
1228
1229 if (!(fs_info->sb->s_flags & MS_RDONLY))
1230 btrfs_orphan_cleanup(root);
1231
1232 return root;
1233 fail:
1234 free_fs_root(root);
1235 return ERR_PTR(ret);
1236 }
1237
1238 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1239 struct btrfs_key *location,
1240 const char *name, int namelen)
1241 {
1242 return btrfs_read_fs_root_no_name(fs_info, location);
1243 #if 0
1244 struct btrfs_root *root;
1245 int ret;
1246
1247 root = btrfs_read_fs_root_no_name(fs_info, location);
1248 if (!root)
1249 return NULL;
1250
1251 if (root->in_sysfs)
1252 return root;
1253
1254 ret = btrfs_set_root_name(root, name, namelen);
1255 if (ret) {
1256 free_extent_buffer(root->node);
1257 kfree(root);
1258 return ERR_PTR(ret);
1259 }
1260
1261 ret = btrfs_sysfs_add_root(root);
1262 if (ret) {
1263 free_extent_buffer(root->node);
1264 kfree(root->name);
1265 kfree(root);
1266 return ERR_PTR(ret);
1267 }
1268 root->in_sysfs = 1;
1269 return root;
1270 #endif
1271 }
1272
1273 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1274 {
1275 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1276 int ret = 0;
1277 struct btrfs_device *device;
1278 struct backing_dev_info *bdi;
1279
1280 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1281 if (!device->bdev)
1282 continue;
1283 bdi = blk_get_backing_dev_info(device->bdev);
1284 if (bdi && bdi_congested(bdi, bdi_bits)) {
1285 ret = 1;
1286 break;
1287 }
1288 }
1289 return ret;
1290 }
1291
1292 /*
1293 * this unplugs every device on the box, and it is only used when page
1294 * is null
1295 */
1296 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1297 {
1298 struct btrfs_device *device;
1299 struct btrfs_fs_info *info;
1300
1301 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1302 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1303 if (!device->bdev)
1304 continue;
1305
1306 bdi = blk_get_backing_dev_info(device->bdev);
1307 if (bdi->unplug_io_fn)
1308 bdi->unplug_io_fn(bdi, page);
1309 }
1310 }
1311
1312 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1313 {
1314 struct inode *inode;
1315 struct extent_map_tree *em_tree;
1316 struct extent_map *em;
1317 struct address_space *mapping;
1318 u64 offset;
1319
1320 /* the generic O_DIRECT read code does this */
1321 if (1 || !page) {
1322 __unplug_io_fn(bdi, page);
1323 return;
1324 }
1325
1326 /*
1327 * page->mapping may change at any time. Get a consistent copy
1328 * and use that for everything below
1329 */
1330 smp_mb();
1331 mapping = page->mapping;
1332 if (!mapping)
1333 return;
1334
1335 inode = mapping->host;
1336
1337 /*
1338 * don't do the expensive searching for a small number of
1339 * devices
1340 */
1341 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1342 __unplug_io_fn(bdi, page);
1343 return;
1344 }
1345
1346 offset = page_offset(page);
1347
1348 em_tree = &BTRFS_I(inode)->extent_tree;
1349 read_lock(&em_tree->lock);
1350 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1351 read_unlock(&em_tree->lock);
1352 if (!em) {
1353 __unplug_io_fn(bdi, page);
1354 return;
1355 }
1356
1357 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1358 free_extent_map(em);
1359 __unplug_io_fn(bdi, page);
1360 return;
1361 }
1362 offset = offset - em->start;
1363 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1364 em->block_start + offset, page);
1365 free_extent_map(em);
1366 }
1367
1368 /*
1369 * If this fails, caller must call bdi_destroy() to get rid of the
1370 * bdi again.
1371 */
1372 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1373 {
1374 int err;
1375
1376 bdi->capabilities = BDI_CAP_MAP_COPY;
1377 err = bdi_init(bdi);
1378 if (err)
1379 return err;
1380
1381 err = bdi_register(bdi, NULL, "btrfs-%d",
1382 atomic_inc_return(&btrfs_bdi_num));
1383 if (err) {
1384 bdi_destroy(bdi);
1385 return err;
1386 }
1387
1388 bdi->ra_pages = default_backing_dev_info.ra_pages;
1389 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1390 bdi->unplug_io_data = info;
1391 bdi->congested_fn = btrfs_congested_fn;
1392 bdi->congested_data = info;
1393 return 0;
1394 }
1395
1396 static int bio_ready_for_csum(struct bio *bio)
1397 {
1398 u64 length = 0;
1399 u64 buf_len = 0;
1400 u64 start = 0;
1401 struct page *page;
1402 struct extent_io_tree *io_tree = NULL;
1403 struct btrfs_fs_info *info = NULL;
1404 struct bio_vec *bvec;
1405 int i;
1406 int ret;
1407
1408 bio_for_each_segment(bvec, bio, i) {
1409 page = bvec->bv_page;
1410 if (page->private == EXTENT_PAGE_PRIVATE) {
1411 length += bvec->bv_len;
1412 continue;
1413 }
1414 if (!page->private) {
1415 length += bvec->bv_len;
1416 continue;
1417 }
1418 length = bvec->bv_len;
1419 buf_len = page->private >> 2;
1420 start = page_offset(page) + bvec->bv_offset;
1421 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1422 info = BTRFS_I(page->mapping->host)->root->fs_info;
1423 }
1424 /* are we fully contained in this bio? */
1425 if (buf_len <= length)
1426 return 1;
1427
1428 ret = extent_range_uptodate(io_tree, start + length,
1429 start + buf_len - 1);
1430 return ret;
1431 }
1432
1433 /*
1434 * called by the kthread helper functions to finally call the bio end_io
1435 * functions. This is where read checksum verification actually happens
1436 */
1437 static void end_workqueue_fn(struct btrfs_work *work)
1438 {
1439 struct bio *bio;
1440 struct end_io_wq *end_io_wq;
1441 struct btrfs_fs_info *fs_info;
1442 int error;
1443
1444 end_io_wq = container_of(work, struct end_io_wq, work);
1445 bio = end_io_wq->bio;
1446 fs_info = end_io_wq->info;
1447
1448 /* metadata bio reads are special because the whole tree block must
1449 * be checksummed at once. This makes sure the entire block is in
1450 * ram and up to date before trying to verify things. For
1451 * blocksize <= pagesize, it is basically a noop
1452 */
1453 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1454 !bio_ready_for_csum(bio)) {
1455 btrfs_queue_worker(&fs_info->endio_meta_workers,
1456 &end_io_wq->work);
1457 return;
1458 }
1459 error = end_io_wq->error;
1460 bio->bi_private = end_io_wq->private;
1461 bio->bi_end_io = end_io_wq->end_io;
1462 kfree(end_io_wq);
1463 bio_endio(bio, error);
1464 }
1465
1466 static int cleaner_kthread(void *arg)
1467 {
1468 struct btrfs_root *root = arg;
1469
1470 do {
1471 smp_mb();
1472 if (root->fs_info->closing)
1473 break;
1474
1475 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1476
1477 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1478 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1479 btrfs_clean_old_snapshots(root);
1480 mutex_unlock(&root->fs_info->cleaner_mutex);
1481 }
1482
1483 if (freezing(current)) {
1484 refrigerator();
1485 } else {
1486 smp_mb();
1487 if (root->fs_info->closing)
1488 break;
1489 set_current_state(TASK_INTERRUPTIBLE);
1490 schedule();
1491 __set_current_state(TASK_RUNNING);
1492 }
1493 } while (!kthread_should_stop());
1494 return 0;
1495 }
1496
1497 static int transaction_kthread(void *arg)
1498 {
1499 struct btrfs_root *root = arg;
1500 struct btrfs_trans_handle *trans;
1501 struct btrfs_transaction *cur;
1502 unsigned long now;
1503 unsigned long delay;
1504 int ret;
1505
1506 do {
1507 smp_mb();
1508 if (root->fs_info->closing)
1509 break;
1510
1511 delay = HZ * 30;
1512 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1513 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1514
1515 mutex_lock(&root->fs_info->trans_mutex);
1516 cur = root->fs_info->running_transaction;
1517 if (!cur) {
1518 mutex_unlock(&root->fs_info->trans_mutex);
1519 goto sleep;
1520 }
1521
1522 now = get_seconds();
1523 if (now < cur->start_time || now - cur->start_time < 30) {
1524 mutex_unlock(&root->fs_info->trans_mutex);
1525 delay = HZ * 5;
1526 goto sleep;
1527 }
1528 mutex_unlock(&root->fs_info->trans_mutex);
1529 trans = btrfs_start_transaction(root, 1);
1530 ret = btrfs_commit_transaction(trans, root);
1531
1532 sleep:
1533 wake_up_process(root->fs_info->cleaner_kthread);
1534 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1535
1536 if (freezing(current)) {
1537 refrigerator();
1538 } else {
1539 if (root->fs_info->closing)
1540 break;
1541 set_current_state(TASK_INTERRUPTIBLE);
1542 schedule_timeout(delay);
1543 __set_current_state(TASK_RUNNING);
1544 }
1545 } while (!kthread_should_stop());
1546 return 0;
1547 }
1548
1549 struct btrfs_root *open_ctree(struct super_block *sb,
1550 struct btrfs_fs_devices *fs_devices,
1551 char *options)
1552 {
1553 u32 sectorsize;
1554 u32 nodesize;
1555 u32 leafsize;
1556 u32 blocksize;
1557 u32 stripesize;
1558 u64 generation;
1559 u64 features;
1560 struct btrfs_key location;
1561 struct buffer_head *bh;
1562 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1563 GFP_NOFS);
1564 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1565 GFP_NOFS);
1566 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1567 GFP_NOFS);
1568 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1569 GFP_NOFS);
1570 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1571 GFP_NOFS);
1572 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1573 GFP_NOFS);
1574 struct btrfs_root *log_tree_root;
1575
1576 int ret;
1577 int err = -EINVAL;
1578
1579 struct btrfs_super_block *disk_super;
1580
1581 if (!extent_root || !tree_root || !fs_info ||
1582 !chunk_root || !dev_root || !csum_root) {
1583 err = -ENOMEM;
1584 goto fail;
1585 }
1586
1587 ret = init_srcu_struct(&fs_info->subvol_srcu);
1588 if (ret) {
1589 err = ret;
1590 goto fail;
1591 }
1592
1593 ret = setup_bdi(fs_info, &fs_info->bdi);
1594 if (ret) {
1595 err = ret;
1596 goto fail_srcu;
1597 }
1598
1599 fs_info->btree_inode = new_inode(sb);
1600 if (!fs_info->btree_inode) {
1601 err = -ENOMEM;
1602 goto fail_bdi;
1603 }
1604
1605 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1606 INIT_LIST_HEAD(&fs_info->trans_list);
1607 INIT_LIST_HEAD(&fs_info->dead_roots);
1608 INIT_LIST_HEAD(&fs_info->hashers);
1609 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1610 INIT_LIST_HEAD(&fs_info->ordered_operations);
1611 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1612 spin_lock_init(&fs_info->delalloc_lock);
1613 spin_lock_init(&fs_info->new_trans_lock);
1614 spin_lock_init(&fs_info->ref_cache_lock);
1615 spin_lock_init(&fs_info->fs_roots_radix_lock);
1616
1617 init_completion(&fs_info->kobj_unregister);
1618 fs_info->tree_root = tree_root;
1619 fs_info->extent_root = extent_root;
1620 fs_info->csum_root = csum_root;
1621 fs_info->chunk_root = chunk_root;
1622 fs_info->dev_root = dev_root;
1623 fs_info->fs_devices = fs_devices;
1624 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1625 INIT_LIST_HEAD(&fs_info->space_info);
1626 btrfs_mapping_init(&fs_info->mapping_tree);
1627 atomic_set(&fs_info->nr_async_submits, 0);
1628 atomic_set(&fs_info->async_delalloc_pages, 0);
1629 atomic_set(&fs_info->async_submit_draining, 0);
1630 atomic_set(&fs_info->nr_async_bios, 0);
1631 fs_info->sb = sb;
1632 fs_info->max_extent = (u64)-1;
1633 fs_info->max_inline = 8192 * 1024;
1634 fs_info->metadata_ratio = 0;
1635
1636 fs_info->thread_pool_size = min_t(unsigned long,
1637 num_online_cpus() + 2, 8);
1638
1639 INIT_LIST_HEAD(&fs_info->ordered_extents);
1640 spin_lock_init(&fs_info->ordered_extent_lock);
1641
1642 sb->s_blocksize = 4096;
1643 sb->s_blocksize_bits = blksize_bits(4096);
1644
1645 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1646 fs_info->btree_inode->i_nlink = 1;
1647 /*
1648 * we set the i_size on the btree inode to the max possible int.
1649 * the real end of the address space is determined by all of
1650 * the devices in the system
1651 */
1652 fs_info->btree_inode->i_size = OFFSET_MAX;
1653 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1654 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1655
1656 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1657 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1658 fs_info->btree_inode->i_mapping,
1659 GFP_NOFS);
1660 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1661 GFP_NOFS);
1662
1663 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1664
1665 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1666 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1667 sizeof(struct btrfs_key));
1668 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1669 insert_inode_hash(fs_info->btree_inode);
1670
1671 spin_lock_init(&fs_info->block_group_cache_lock);
1672 fs_info->block_group_cache_tree.rb_node = NULL;
1673
1674 extent_io_tree_init(&fs_info->freed_extents[0],
1675 fs_info->btree_inode->i_mapping, GFP_NOFS);
1676 extent_io_tree_init(&fs_info->freed_extents[1],
1677 fs_info->btree_inode->i_mapping, GFP_NOFS);
1678 fs_info->pinned_extents = &fs_info->freed_extents[0];
1679 fs_info->do_barriers = 1;
1680
1681
1682 mutex_init(&fs_info->trans_mutex);
1683 mutex_init(&fs_info->ordered_operations_mutex);
1684 mutex_init(&fs_info->tree_log_mutex);
1685 mutex_init(&fs_info->chunk_mutex);
1686 mutex_init(&fs_info->transaction_kthread_mutex);
1687 mutex_init(&fs_info->cleaner_mutex);
1688 mutex_init(&fs_info->volume_mutex);
1689 init_rwsem(&fs_info->extent_commit_sem);
1690 init_rwsem(&fs_info->subvol_sem);
1691
1692 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1693 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1694
1695 init_waitqueue_head(&fs_info->transaction_throttle);
1696 init_waitqueue_head(&fs_info->transaction_wait);
1697 init_waitqueue_head(&fs_info->async_submit_wait);
1698
1699 __setup_root(4096, 4096, 4096, 4096, tree_root,
1700 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1701
1702
1703 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1704 if (!bh)
1705 goto fail_iput;
1706
1707 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1708 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1709 sizeof(fs_info->super_for_commit));
1710 brelse(bh);
1711
1712 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1713
1714 disk_super = &fs_info->super_copy;
1715 if (!btrfs_super_root(disk_super))
1716 goto fail_iput;
1717
1718 ret = btrfs_parse_options(tree_root, options);
1719 if (ret) {
1720 err = ret;
1721 goto fail_iput;
1722 }
1723
1724 features = btrfs_super_incompat_flags(disk_super) &
1725 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1726 if (features) {
1727 printk(KERN_ERR "BTRFS: couldn't mount because of "
1728 "unsupported optional features (%Lx).\n",
1729 (unsigned long long)features);
1730 err = -EINVAL;
1731 goto fail_iput;
1732 }
1733
1734 features = btrfs_super_incompat_flags(disk_super);
1735 if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1736 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1737 btrfs_set_super_incompat_flags(disk_super, features);
1738 }
1739
1740 features = btrfs_super_compat_ro_flags(disk_super) &
1741 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1742 if (!(sb->s_flags & MS_RDONLY) && features) {
1743 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1744 "unsupported option features (%Lx).\n",
1745 (unsigned long long)features);
1746 err = -EINVAL;
1747 goto fail_iput;
1748 }
1749
1750 btrfs_init_workers(&fs_info->generic_worker,
1751 "genwork", 1, NULL);
1752
1753 btrfs_init_workers(&fs_info->workers, "worker",
1754 fs_info->thread_pool_size,
1755 &fs_info->generic_worker);
1756
1757 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1758 fs_info->thread_pool_size,
1759 &fs_info->generic_worker);
1760
1761 btrfs_init_workers(&fs_info->submit_workers, "submit",
1762 min_t(u64, fs_devices->num_devices,
1763 fs_info->thread_pool_size),
1764 &fs_info->generic_worker);
1765
1766 /* a higher idle thresh on the submit workers makes it much more
1767 * likely that bios will be send down in a sane order to the
1768 * devices
1769 */
1770 fs_info->submit_workers.idle_thresh = 64;
1771
1772 fs_info->workers.idle_thresh = 16;
1773 fs_info->workers.ordered = 1;
1774
1775 fs_info->delalloc_workers.idle_thresh = 2;
1776 fs_info->delalloc_workers.ordered = 1;
1777
1778 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1779 &fs_info->generic_worker);
1780 btrfs_init_workers(&fs_info->endio_workers, "endio",
1781 fs_info->thread_pool_size,
1782 &fs_info->generic_worker);
1783 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1784 fs_info->thread_pool_size,
1785 &fs_info->generic_worker);
1786 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1787 "endio-meta-write", fs_info->thread_pool_size,
1788 &fs_info->generic_worker);
1789 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1790 fs_info->thread_pool_size,
1791 &fs_info->generic_worker);
1792
1793 /*
1794 * endios are largely parallel and should have a very
1795 * low idle thresh
1796 */
1797 fs_info->endio_workers.idle_thresh = 4;
1798 fs_info->endio_meta_workers.idle_thresh = 4;
1799
1800 fs_info->endio_write_workers.idle_thresh = 2;
1801 fs_info->endio_meta_write_workers.idle_thresh = 2;
1802
1803 btrfs_start_workers(&fs_info->workers, 1);
1804 btrfs_start_workers(&fs_info->generic_worker, 1);
1805 btrfs_start_workers(&fs_info->submit_workers, 1);
1806 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1807 btrfs_start_workers(&fs_info->fixup_workers, 1);
1808 btrfs_start_workers(&fs_info->endio_workers, 1);
1809 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1810 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1811 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1812
1813 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1814 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1815 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1816
1817 nodesize = btrfs_super_nodesize(disk_super);
1818 leafsize = btrfs_super_leafsize(disk_super);
1819 sectorsize = btrfs_super_sectorsize(disk_super);
1820 stripesize = btrfs_super_stripesize(disk_super);
1821 tree_root->nodesize = nodesize;
1822 tree_root->leafsize = leafsize;
1823 tree_root->sectorsize = sectorsize;
1824 tree_root->stripesize = stripesize;
1825
1826 sb->s_blocksize = sectorsize;
1827 sb->s_blocksize_bits = blksize_bits(sectorsize);
1828
1829 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1830 sizeof(disk_super->magic))) {
1831 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1832 goto fail_sb_buffer;
1833 }
1834
1835 mutex_lock(&fs_info->chunk_mutex);
1836 ret = btrfs_read_sys_array(tree_root);
1837 mutex_unlock(&fs_info->chunk_mutex);
1838 if (ret) {
1839 printk(KERN_WARNING "btrfs: failed to read the system "
1840 "array on %s\n", sb->s_id);
1841 goto fail_sb_buffer;
1842 }
1843
1844 blocksize = btrfs_level_size(tree_root,
1845 btrfs_super_chunk_root_level(disk_super));
1846 generation = btrfs_super_chunk_root_generation(disk_super);
1847
1848 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1849 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1850
1851 chunk_root->node = read_tree_block(chunk_root,
1852 btrfs_super_chunk_root(disk_super),
1853 blocksize, generation);
1854 BUG_ON(!chunk_root->node);
1855 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1856 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1857 sb->s_id);
1858 goto fail_chunk_root;
1859 }
1860 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1861 chunk_root->commit_root = btrfs_root_node(chunk_root);
1862
1863 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1864 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1865 BTRFS_UUID_SIZE);
1866
1867 mutex_lock(&fs_info->chunk_mutex);
1868 ret = btrfs_read_chunk_tree(chunk_root);
1869 mutex_unlock(&fs_info->chunk_mutex);
1870 if (ret) {
1871 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1872 sb->s_id);
1873 goto fail_chunk_root;
1874 }
1875
1876 btrfs_close_extra_devices(fs_devices);
1877
1878 blocksize = btrfs_level_size(tree_root,
1879 btrfs_super_root_level(disk_super));
1880 generation = btrfs_super_generation(disk_super);
1881
1882 tree_root->node = read_tree_block(tree_root,
1883 btrfs_super_root(disk_super),
1884 blocksize, generation);
1885 if (!tree_root->node)
1886 goto fail_chunk_root;
1887 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1888 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1889 sb->s_id);
1890 goto fail_tree_root;
1891 }
1892 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1893 tree_root->commit_root = btrfs_root_node(tree_root);
1894
1895 ret = find_and_setup_root(tree_root, fs_info,
1896 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1897 if (ret)
1898 goto fail_tree_root;
1899 extent_root->track_dirty = 1;
1900
1901 ret = find_and_setup_root(tree_root, fs_info,
1902 BTRFS_DEV_TREE_OBJECTID, dev_root);
1903 if (ret)
1904 goto fail_extent_root;
1905 dev_root->track_dirty = 1;
1906
1907 ret = find_and_setup_root(tree_root, fs_info,
1908 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1909 if (ret)
1910 goto fail_dev_root;
1911
1912 csum_root->track_dirty = 1;
1913
1914 btrfs_read_block_groups(extent_root);
1915
1916 fs_info->generation = generation;
1917 fs_info->last_trans_committed = generation;
1918 fs_info->data_alloc_profile = (u64)-1;
1919 fs_info->metadata_alloc_profile = (u64)-1;
1920 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1921 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1922 "btrfs-cleaner");
1923 if (IS_ERR(fs_info->cleaner_kthread))
1924 goto fail_csum_root;
1925
1926 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1927 tree_root,
1928 "btrfs-transaction");
1929 if (IS_ERR(fs_info->transaction_kthread))
1930 goto fail_cleaner;
1931
1932 if (!btrfs_test_opt(tree_root, SSD) &&
1933 !btrfs_test_opt(tree_root, NOSSD) &&
1934 !fs_info->fs_devices->rotating) {
1935 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1936 "mode\n");
1937 btrfs_set_opt(fs_info->mount_opt, SSD);
1938 }
1939
1940 if (btrfs_super_log_root(disk_super) != 0) {
1941 u64 bytenr = btrfs_super_log_root(disk_super);
1942
1943 if (fs_devices->rw_devices == 0) {
1944 printk(KERN_WARNING "Btrfs log replay required "
1945 "on RO media\n");
1946 err = -EIO;
1947 goto fail_trans_kthread;
1948 }
1949 blocksize =
1950 btrfs_level_size(tree_root,
1951 btrfs_super_log_root_level(disk_super));
1952
1953 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1954 GFP_NOFS);
1955
1956 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1957 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1958
1959 log_tree_root->node = read_tree_block(tree_root, bytenr,
1960 blocksize,
1961 generation + 1);
1962 ret = btrfs_recover_log_trees(log_tree_root);
1963 BUG_ON(ret);
1964
1965 if (sb->s_flags & MS_RDONLY) {
1966 ret = btrfs_commit_super(tree_root);
1967 BUG_ON(ret);
1968 }
1969 }
1970
1971 ret = btrfs_find_orphan_roots(tree_root);
1972 BUG_ON(ret);
1973
1974 if (!(sb->s_flags & MS_RDONLY)) {
1975 ret = btrfs_recover_relocation(tree_root);
1976 BUG_ON(ret);
1977 }
1978
1979 location.objectid = BTRFS_FS_TREE_OBJECTID;
1980 location.type = BTRFS_ROOT_ITEM_KEY;
1981 location.offset = (u64)-1;
1982
1983 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1984 if (!fs_info->fs_root)
1985 goto fail_trans_kthread;
1986
1987 return tree_root;
1988
1989 fail_trans_kthread:
1990 kthread_stop(fs_info->transaction_kthread);
1991 fail_cleaner:
1992 kthread_stop(fs_info->cleaner_kthread);
1993
1994 /*
1995 * make sure we're done with the btree inode before we stop our
1996 * kthreads
1997 */
1998 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1999 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2000
2001 fail_csum_root:
2002 free_extent_buffer(csum_root->node);
2003 free_extent_buffer(csum_root->commit_root);
2004 fail_dev_root:
2005 free_extent_buffer(dev_root->node);
2006 free_extent_buffer(dev_root->commit_root);
2007 fail_extent_root:
2008 free_extent_buffer(extent_root->node);
2009 free_extent_buffer(extent_root->commit_root);
2010 fail_tree_root:
2011 free_extent_buffer(tree_root->node);
2012 free_extent_buffer(tree_root->commit_root);
2013 fail_chunk_root:
2014 free_extent_buffer(chunk_root->node);
2015 free_extent_buffer(chunk_root->commit_root);
2016 fail_sb_buffer:
2017 btrfs_stop_workers(&fs_info->generic_worker);
2018 btrfs_stop_workers(&fs_info->fixup_workers);
2019 btrfs_stop_workers(&fs_info->delalloc_workers);
2020 btrfs_stop_workers(&fs_info->workers);
2021 btrfs_stop_workers(&fs_info->endio_workers);
2022 btrfs_stop_workers(&fs_info->endio_meta_workers);
2023 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2024 btrfs_stop_workers(&fs_info->endio_write_workers);
2025 btrfs_stop_workers(&fs_info->submit_workers);
2026 fail_iput:
2027 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2028 iput(fs_info->btree_inode);
2029
2030 btrfs_close_devices(fs_info->fs_devices);
2031 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2032 fail_bdi:
2033 bdi_destroy(&fs_info->bdi);
2034 fail_srcu:
2035 cleanup_srcu_struct(&fs_info->subvol_srcu);
2036 fail:
2037 kfree(extent_root);
2038 kfree(tree_root);
2039 kfree(fs_info);
2040 kfree(chunk_root);
2041 kfree(dev_root);
2042 kfree(csum_root);
2043 return ERR_PTR(err);
2044 }
2045
2046 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2047 {
2048 char b[BDEVNAME_SIZE];
2049
2050 if (uptodate) {
2051 set_buffer_uptodate(bh);
2052 } else {
2053 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2054 printk(KERN_WARNING "lost page write due to "
2055 "I/O error on %s\n",
2056 bdevname(bh->b_bdev, b));
2057 }
2058 /* note, we dont' set_buffer_write_io_error because we have
2059 * our own ways of dealing with the IO errors
2060 */
2061 clear_buffer_uptodate(bh);
2062 }
2063 unlock_buffer(bh);
2064 put_bh(bh);
2065 }
2066
2067 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2068 {
2069 struct buffer_head *bh;
2070 struct buffer_head *latest = NULL;
2071 struct btrfs_super_block *super;
2072 int i;
2073 u64 transid = 0;
2074 u64 bytenr;
2075
2076 /* we would like to check all the supers, but that would make
2077 * a btrfs mount succeed after a mkfs from a different FS.
2078 * So, we need to add a special mount option to scan for
2079 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2080 */
2081 for (i = 0; i < 1; i++) {
2082 bytenr = btrfs_sb_offset(i);
2083 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2084 break;
2085 bh = __bread(bdev, bytenr / 4096, 4096);
2086 if (!bh)
2087 continue;
2088
2089 super = (struct btrfs_super_block *)bh->b_data;
2090 if (btrfs_super_bytenr(super) != bytenr ||
2091 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2092 sizeof(super->magic))) {
2093 brelse(bh);
2094 continue;
2095 }
2096
2097 if (!latest || btrfs_super_generation(super) > transid) {
2098 brelse(latest);
2099 latest = bh;
2100 transid = btrfs_super_generation(super);
2101 } else {
2102 brelse(bh);
2103 }
2104 }
2105 return latest;
2106 }
2107
2108 /*
2109 * this should be called twice, once with wait == 0 and
2110 * once with wait == 1. When wait == 0 is done, all the buffer heads
2111 * we write are pinned.
2112 *
2113 * They are released when wait == 1 is done.
2114 * max_mirrors must be the same for both runs, and it indicates how
2115 * many supers on this one device should be written.
2116 *
2117 * max_mirrors == 0 means to write them all.
2118 */
2119 static int write_dev_supers(struct btrfs_device *device,
2120 struct btrfs_super_block *sb,
2121 int do_barriers, int wait, int max_mirrors)
2122 {
2123 struct buffer_head *bh;
2124 int i;
2125 int ret;
2126 int errors = 0;
2127 u32 crc;
2128 u64 bytenr;
2129 int last_barrier = 0;
2130
2131 if (max_mirrors == 0)
2132 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2133
2134 /* make sure only the last submit_bh does a barrier */
2135 if (do_barriers) {
2136 for (i = 0; i < max_mirrors; i++) {
2137 bytenr = btrfs_sb_offset(i);
2138 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2139 device->total_bytes)
2140 break;
2141 last_barrier = i;
2142 }
2143 }
2144
2145 for (i = 0; i < max_mirrors; i++) {
2146 bytenr = btrfs_sb_offset(i);
2147 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2148 break;
2149
2150 if (wait) {
2151 bh = __find_get_block(device->bdev, bytenr / 4096,
2152 BTRFS_SUPER_INFO_SIZE);
2153 BUG_ON(!bh);
2154 wait_on_buffer(bh);
2155 if (!buffer_uptodate(bh))
2156 errors++;
2157
2158 /* drop our reference */
2159 brelse(bh);
2160
2161 /* drop the reference from the wait == 0 run */
2162 brelse(bh);
2163 continue;
2164 } else {
2165 btrfs_set_super_bytenr(sb, bytenr);
2166
2167 crc = ~(u32)0;
2168 crc = btrfs_csum_data(NULL, (char *)sb +
2169 BTRFS_CSUM_SIZE, crc,
2170 BTRFS_SUPER_INFO_SIZE -
2171 BTRFS_CSUM_SIZE);
2172 btrfs_csum_final(crc, sb->csum);
2173
2174 /*
2175 * one reference for us, and we leave it for the
2176 * caller
2177 */
2178 bh = __getblk(device->bdev, bytenr / 4096,
2179 BTRFS_SUPER_INFO_SIZE);
2180 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2181
2182 /* one reference for submit_bh */
2183 get_bh(bh);
2184
2185 set_buffer_uptodate(bh);
2186 lock_buffer(bh);
2187 bh->b_end_io = btrfs_end_buffer_write_sync;
2188 }
2189
2190 if (i == last_barrier && do_barriers && device->barriers) {
2191 ret = submit_bh(WRITE_BARRIER, bh);
2192 if (ret == -EOPNOTSUPP) {
2193 printk("btrfs: disabling barriers on dev %s\n",
2194 device->name);
2195 set_buffer_uptodate(bh);
2196 device->barriers = 0;
2197 /* one reference for submit_bh */
2198 get_bh(bh);
2199 lock_buffer(bh);
2200 ret = submit_bh(WRITE_SYNC, bh);
2201 }
2202 } else {
2203 ret = submit_bh(WRITE_SYNC, bh);
2204 }
2205
2206 if (ret)
2207 errors++;
2208 }
2209 return errors < i ? 0 : -1;
2210 }
2211
2212 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2213 {
2214 struct list_head *head;
2215 struct btrfs_device *dev;
2216 struct btrfs_super_block *sb;
2217 struct btrfs_dev_item *dev_item;
2218 int ret;
2219 int do_barriers;
2220 int max_errors;
2221 int total_errors = 0;
2222 u64 flags;
2223
2224 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2225 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2226
2227 sb = &root->fs_info->super_for_commit;
2228 dev_item = &sb->dev_item;
2229
2230 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2231 head = &root->fs_info->fs_devices->devices;
2232 list_for_each_entry(dev, head, dev_list) {
2233 if (!dev->bdev) {
2234 total_errors++;
2235 continue;
2236 }
2237 if (!dev->in_fs_metadata || !dev->writeable)
2238 continue;
2239
2240 btrfs_set_stack_device_generation(dev_item, 0);
2241 btrfs_set_stack_device_type(dev_item, dev->type);
2242 btrfs_set_stack_device_id(dev_item, dev->devid);
2243 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2244 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2245 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2246 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2247 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2248 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2249 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2250
2251 flags = btrfs_super_flags(sb);
2252 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2253
2254 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2255 if (ret)
2256 total_errors++;
2257 }
2258 if (total_errors > max_errors) {
2259 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2260 total_errors);
2261 BUG();
2262 }
2263
2264 total_errors = 0;
2265 list_for_each_entry(dev, head, dev_list) {
2266 if (!dev->bdev)
2267 continue;
2268 if (!dev->in_fs_metadata || !dev->writeable)
2269 continue;
2270
2271 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2272 if (ret)
2273 total_errors++;
2274 }
2275 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2276 if (total_errors > max_errors) {
2277 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2278 total_errors);
2279 BUG();
2280 }
2281 return 0;
2282 }
2283
2284 int write_ctree_super(struct btrfs_trans_handle *trans,
2285 struct btrfs_root *root, int max_mirrors)
2286 {
2287 int ret;
2288
2289 ret = write_all_supers(root, max_mirrors);
2290 return ret;
2291 }
2292
2293 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2294 {
2295 spin_lock(&fs_info->fs_roots_radix_lock);
2296 radix_tree_delete(&fs_info->fs_roots_radix,
2297 (unsigned long)root->root_key.objectid);
2298 spin_unlock(&fs_info->fs_roots_radix_lock);
2299
2300 if (btrfs_root_refs(&root->root_item) == 0)
2301 synchronize_srcu(&fs_info->subvol_srcu);
2302
2303 free_fs_root(root);
2304 return 0;
2305 }
2306
2307 static void free_fs_root(struct btrfs_root *root)
2308 {
2309 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2310 if (root->anon_super.s_dev) {
2311 down_write(&root->anon_super.s_umount);
2312 kill_anon_super(&root->anon_super);
2313 }
2314 free_extent_buffer(root->node);
2315 free_extent_buffer(root->commit_root);
2316 kfree(root->name);
2317 kfree(root);
2318 }
2319
2320 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2321 {
2322 int ret;
2323 struct btrfs_root *gang[8];
2324 int i;
2325
2326 while (!list_empty(&fs_info->dead_roots)) {
2327 gang[0] = list_entry(fs_info->dead_roots.next,
2328 struct btrfs_root, root_list);
2329 list_del(&gang[0]->root_list);
2330
2331 if (gang[0]->in_radix) {
2332 btrfs_free_fs_root(fs_info, gang[0]);
2333 } else {
2334 free_extent_buffer(gang[0]->node);
2335 free_extent_buffer(gang[0]->commit_root);
2336 kfree(gang[0]);
2337 }
2338 }
2339
2340 while (1) {
2341 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2342 (void **)gang, 0,
2343 ARRAY_SIZE(gang));
2344 if (!ret)
2345 break;
2346 for (i = 0; i < ret; i++)
2347 btrfs_free_fs_root(fs_info, gang[i]);
2348 }
2349 return 0;
2350 }
2351
2352 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2353 {
2354 u64 root_objectid = 0;
2355 struct btrfs_root *gang[8];
2356 int i;
2357 int ret;
2358
2359 while (1) {
2360 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2361 (void **)gang, root_objectid,
2362 ARRAY_SIZE(gang));
2363 if (!ret)
2364 break;
2365
2366 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2367 for (i = 0; i < ret; i++) {
2368 root_objectid = gang[i]->root_key.objectid;
2369 btrfs_orphan_cleanup(gang[i]);
2370 }
2371 root_objectid++;
2372 }
2373 return 0;
2374 }
2375
2376 int btrfs_commit_super(struct btrfs_root *root)
2377 {
2378 struct btrfs_trans_handle *trans;
2379 int ret;
2380
2381 mutex_lock(&root->fs_info->cleaner_mutex);
2382 btrfs_clean_old_snapshots(root);
2383 mutex_unlock(&root->fs_info->cleaner_mutex);
2384 trans = btrfs_start_transaction(root, 1);
2385 ret = btrfs_commit_transaction(trans, root);
2386 BUG_ON(ret);
2387 /* run commit again to drop the original snapshot */
2388 trans = btrfs_start_transaction(root, 1);
2389 btrfs_commit_transaction(trans, root);
2390 ret = btrfs_write_and_wait_transaction(NULL, root);
2391 BUG_ON(ret);
2392
2393 ret = write_ctree_super(NULL, root, 0);
2394 return ret;
2395 }
2396
2397 int close_ctree(struct btrfs_root *root)
2398 {
2399 struct btrfs_fs_info *fs_info = root->fs_info;
2400 int ret;
2401
2402 fs_info->closing = 1;
2403 smp_mb();
2404
2405 kthread_stop(root->fs_info->transaction_kthread);
2406 kthread_stop(root->fs_info->cleaner_kthread);
2407
2408 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2409 ret = btrfs_commit_super(root);
2410 if (ret)
2411 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2412 }
2413
2414 fs_info->closing = 2;
2415 smp_mb();
2416
2417 if (fs_info->delalloc_bytes) {
2418 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2419 (unsigned long long)fs_info->delalloc_bytes);
2420 }
2421 if (fs_info->total_ref_cache_size) {
2422 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2423 (unsigned long long)fs_info->total_ref_cache_size);
2424 }
2425
2426 free_extent_buffer(fs_info->extent_root->node);
2427 free_extent_buffer(fs_info->extent_root->commit_root);
2428 free_extent_buffer(fs_info->tree_root->node);
2429 free_extent_buffer(fs_info->tree_root->commit_root);
2430 free_extent_buffer(root->fs_info->chunk_root->node);
2431 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2432 free_extent_buffer(root->fs_info->dev_root->node);
2433 free_extent_buffer(root->fs_info->dev_root->commit_root);
2434 free_extent_buffer(root->fs_info->csum_root->node);
2435 free_extent_buffer(root->fs_info->csum_root->commit_root);
2436
2437 btrfs_free_block_groups(root->fs_info);
2438
2439 del_fs_roots(fs_info);
2440
2441 iput(fs_info->btree_inode);
2442
2443 btrfs_stop_workers(&fs_info->generic_worker);
2444 btrfs_stop_workers(&fs_info->fixup_workers);
2445 btrfs_stop_workers(&fs_info->delalloc_workers);
2446 btrfs_stop_workers(&fs_info->workers);
2447 btrfs_stop_workers(&fs_info->endio_workers);
2448 btrfs_stop_workers(&fs_info->endio_meta_workers);
2449 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2450 btrfs_stop_workers(&fs_info->endio_write_workers);
2451 btrfs_stop_workers(&fs_info->submit_workers);
2452
2453 btrfs_close_devices(fs_info->fs_devices);
2454 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2455
2456 bdi_destroy(&fs_info->bdi);
2457 cleanup_srcu_struct(&fs_info->subvol_srcu);
2458
2459 kfree(fs_info->extent_root);
2460 kfree(fs_info->tree_root);
2461 kfree(fs_info->chunk_root);
2462 kfree(fs_info->dev_root);
2463 kfree(fs_info->csum_root);
2464 return 0;
2465 }
2466
2467 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2468 {
2469 int ret;
2470 struct inode *btree_inode = buf->first_page->mapping->host;
2471
2472 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2473 if (!ret)
2474 return ret;
2475
2476 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2477 parent_transid);
2478 return !ret;
2479 }
2480
2481 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2482 {
2483 struct inode *btree_inode = buf->first_page->mapping->host;
2484 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2485 buf);
2486 }
2487
2488 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2489 {
2490 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2491 u64 transid = btrfs_header_generation(buf);
2492 struct inode *btree_inode = root->fs_info->btree_inode;
2493 int was_dirty;
2494
2495 btrfs_assert_tree_locked(buf);
2496 if (transid != root->fs_info->generation) {
2497 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2498 "found %llu running %llu\n",
2499 (unsigned long long)buf->start,
2500 (unsigned long long)transid,
2501 (unsigned long long)root->fs_info->generation);
2502 WARN_ON(1);
2503 }
2504 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2505 buf);
2506 if (!was_dirty) {
2507 spin_lock(&root->fs_info->delalloc_lock);
2508 root->fs_info->dirty_metadata_bytes += buf->len;
2509 spin_unlock(&root->fs_info->delalloc_lock);
2510 }
2511 }
2512
2513 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2514 {
2515 /*
2516 * looks as though older kernels can get into trouble with
2517 * this code, they end up stuck in balance_dirty_pages forever
2518 */
2519 u64 num_dirty;
2520 unsigned long thresh = 32 * 1024 * 1024;
2521
2522 if (current->flags & PF_MEMALLOC)
2523 return;
2524
2525 num_dirty = root->fs_info->dirty_metadata_bytes;
2526
2527 if (num_dirty > thresh) {
2528 balance_dirty_pages_ratelimited_nr(
2529 root->fs_info->btree_inode->i_mapping, 1);
2530 }
2531 return;
2532 }
2533
2534 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2535 {
2536 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2537 int ret;
2538 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2539 if (ret == 0)
2540 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2541 return ret;
2542 }
2543
2544 int btree_lock_page_hook(struct page *page)
2545 {
2546 struct inode *inode = page->mapping->host;
2547 struct btrfs_root *root = BTRFS_I(inode)->root;
2548 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2549 struct extent_buffer *eb;
2550 unsigned long len;
2551 u64 bytenr = page_offset(page);
2552
2553 if (page->private == EXTENT_PAGE_PRIVATE)
2554 goto out;
2555
2556 len = page->private >> 2;
2557 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2558 if (!eb)
2559 goto out;
2560
2561 btrfs_tree_lock(eb);
2562 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2563
2564 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2565 spin_lock(&root->fs_info->delalloc_lock);
2566 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2567 root->fs_info->dirty_metadata_bytes -= eb->len;
2568 else
2569 WARN_ON(1);
2570 spin_unlock(&root->fs_info->delalloc_lock);
2571 }
2572
2573 btrfs_tree_unlock(eb);
2574 free_extent_buffer(eb);
2575 out:
2576 lock_page(page);
2577 return 0;
2578 }
2579
2580 static struct extent_io_ops btree_extent_io_ops = {
2581 .write_cache_pages_lock_hook = btree_lock_page_hook,
2582 .readpage_end_io_hook = btree_readpage_end_io_hook,
2583 .submit_bio_hook = btree_submit_bio_hook,
2584 /* note we're sharing with inode.c for the merge bio hook */
2585 .merge_bio_hook = btrfs_merge_bio_hook,
2586 };
kernel-based virtual machine