?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Btrfs: Wait for kernel threads to make progress during async submission
[linux-2.6/kvm.git] / fs / btrfs / disk-io.c
blob1aed1f4616b623a9e66e75510095a25239f7da12
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/version.h>
20 #include <linux/fs.h>
21 #include <linux/blkdev.h>
22 #include <linux/scatterlist.h>
23 #include <linux/swap.h>
24 #include <linux/radix-tree.h>
25 #include <linux/writeback.h>
26 #include <linux/buffer_head.h> // for block_sync_page
27 #include <linux/workqueue.h>
28 #include <linux/kthread.h>
29 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,20)
30 # include <linux/freezer.h>
31 #else
32 # include <linux/sched.h>
33 #endif
34 #include "crc32c.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 "ref-cache.h"
44
45 #if 0
46 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
47 {
48 if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
49 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
50 (unsigned long long)extent_buffer_blocknr(buf),
51 (unsigned long long)btrfs_header_blocknr(buf));
52 return 1;
53 }
54 return 0;
55 }
56 #endif
57
58 static struct extent_io_ops btree_extent_io_ops;
59 static void end_workqueue_fn(struct btrfs_work *work);
60
61 struct end_io_wq {
62 struct bio *bio;
63 bio_end_io_t *end_io;
64 void *private;
65 struct btrfs_fs_info *info;
66 int error;
67 int metadata;
68 struct list_head list;
69 struct btrfs_work work;
70 };
71
72 struct async_submit_bio {
73 struct inode *inode;
74 struct bio *bio;
75 struct list_head list;
76 extent_submit_bio_hook_t *submit_bio_hook;
77 int rw;
78 int mirror_num;
79 struct btrfs_work work;
80 };
81
82 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
83 size_t page_offset, u64 start, u64 len,
84 int create)
85 {
86 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
87 struct extent_map *em;
88 int ret;
89
90 spin_lock(&em_tree->lock);
91 em = lookup_extent_mapping(em_tree, start, len);
92 if (em) {
93 em->bdev =
94 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
95 spin_unlock(&em_tree->lock);
96 goto out;
97 }
98 spin_unlock(&em_tree->lock);
99
100 em = alloc_extent_map(GFP_NOFS);
101 if (!em) {
102 em = ERR_PTR(-ENOMEM);
103 goto out;
104 }
105 em->start = 0;
106 em->len = (u64)-1;
107 em->block_start = 0;
108 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
109
110 spin_lock(&em_tree->lock);
111 ret = add_extent_mapping(em_tree, em);
112 if (ret == -EEXIST) {
113 u64 failed_start = em->start;
114 u64 failed_len = em->len;
115
116 printk("failed to insert %Lu %Lu -> %Lu into tree\n",
117 em->start, em->len, em->block_start);
118 free_extent_map(em);
119 em = lookup_extent_mapping(em_tree, start, len);
120 if (em) {
121 printk("after failing, found %Lu %Lu %Lu\n",
122 em->start, em->len, em->block_start);
123 ret = 0;
124 } else {
125 em = lookup_extent_mapping(em_tree, failed_start,
126 failed_len);
127 if (em) {
128 printk("double failure lookup gives us "
129 "%Lu %Lu -> %Lu\n", em->start,
130 em->len, em->block_start);
131 free_extent_map(em);
132 }
133 ret = -EIO;
134 }
135 } else if (ret) {
136 free_extent_map(em);
137 em = NULL;
138 }
139 spin_unlock(&em_tree->lock);
140
141 if (ret)
142 em = ERR_PTR(ret);
143 out:
144 return em;
145 }
146
147 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
148 {
149 return btrfs_crc32c(seed, data, len);
150 }
151
152 void btrfs_csum_final(u32 crc, char *result)
153 {
154 *(__le32 *)result = ~cpu_to_le32(crc);
155 }
156
157 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
158 int verify)
159 {
160 char result[BTRFS_CRC32_SIZE];
161 unsigned long len;
162 unsigned long cur_len;
163 unsigned long offset = BTRFS_CSUM_SIZE;
164 char *map_token = NULL;
165 char *kaddr;
166 unsigned long map_start;
167 unsigned long map_len;
168 int err;
169 u32 crc = ~(u32)0;
170
171 len = buf->len - offset;
172 while(len > 0) {
173 err = map_private_extent_buffer(buf, offset, 32,
174 &map_token, &kaddr,
175 &map_start, &map_len, KM_USER0);
176 if (err) {
177 printk("failed to map extent buffer! %lu\n",
178 offset);
179 return 1;
180 }
181 cur_len = min(len, map_len - (offset - map_start));
182 crc = btrfs_csum_data(root, kaddr + offset - map_start,
183 crc, cur_len);
184 len -= cur_len;
185 offset += cur_len;
186 unmap_extent_buffer(buf, map_token, KM_USER0);
187 }
188 btrfs_csum_final(crc, result);
189
190 if (verify) {
191 /* FIXME, this is not good */
192 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
193 u32 val;
194 u32 found = 0;
195 memcpy(&found, result, BTRFS_CRC32_SIZE);
196
197 read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
198 printk("btrfs: %s checksum verify failed on %llu "
199 "wanted %X found %X level %d\n",
200 root->fs_info->sb->s_id,
201 buf->start, val, found, btrfs_header_level(buf));
202 return 1;
203 }
204 } else {
205 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
206 }
207 return 0;
208 }
209
210 static int verify_parent_transid(struct extent_io_tree *io_tree,
211 struct extent_buffer *eb, u64 parent_transid)
212 {
213 int ret;
214
215 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
216 return 0;
217
218 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
219 if (extent_buffer_uptodate(io_tree, eb) &&
220 btrfs_header_generation(eb) == parent_transid) {
221 ret = 0;
222 goto out;
223 }
224 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
225 (unsigned long long)eb->start,
226 (unsigned long long)parent_transid,
227 (unsigned long long)btrfs_header_generation(eb));
228 ret = 1;
229 clear_extent_buffer_uptodate(io_tree, eb);
230 out:
231 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
232 GFP_NOFS);
233 return ret;
234
235 }
236
237 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
238 struct extent_buffer *eb,
239 u64 start, u64 parent_transid)
240 {
241 struct extent_io_tree *io_tree;
242 int ret;
243 int num_copies = 0;
244 int mirror_num = 0;
245
246 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
247 while (1) {
248 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
249 btree_get_extent, mirror_num);
250 if (!ret &&
251 !verify_parent_transid(io_tree, eb, parent_transid))
252 return ret;
253
254 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
255 eb->start, eb->len);
256 if (num_copies == 1)
257 return ret;
258
259 mirror_num++;
260 if (mirror_num > num_copies)
261 return ret;
262 }
263 return -EIO;
264 }
265
266 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
267 {
268 struct extent_io_tree *tree;
269 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
270 u64 found_start;
271 int found_level;
272 unsigned long len;
273 struct extent_buffer *eb;
274 int ret;
275
276 tree = &BTRFS_I(page->mapping->host)->io_tree;
277
278 if (page->private == EXTENT_PAGE_PRIVATE)
279 goto out;
280 if (!page->private)
281 goto out;
282 len = page->private >> 2;
283 if (len == 0) {
284 WARN_ON(1);
285 }
286 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
287 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
288 btrfs_header_generation(eb));
289 BUG_ON(ret);
290 found_start = btrfs_header_bytenr(eb);
291 if (found_start != start) {
292 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
293 start, found_start, len);
294 WARN_ON(1);
295 goto err;
296 }
297 if (eb->first_page != page) {
298 printk("bad first page %lu %lu\n", eb->first_page->index,
299 page->index);
300 WARN_ON(1);
301 goto err;
302 }
303 if (!PageUptodate(page)) {
304 printk("csum not up to date page %lu\n", page->index);
305 WARN_ON(1);
306 goto err;
307 }
308 found_level = btrfs_header_level(eb);
309 spin_lock(&root->fs_info->hash_lock);
310 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
311 spin_unlock(&root->fs_info->hash_lock);
312 csum_tree_block(root, eb, 0);
313 err:
314 free_extent_buffer(eb);
315 out:
316 return 0;
317 }
318
319 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
320 {
321 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
322
323 csum_dirty_buffer(root, page);
324 return 0;
325 }
326
327 int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
328 struct extent_state *state)
329 {
330 struct extent_io_tree *tree;
331 u64 found_start;
332 int found_level;
333 unsigned long len;
334 struct extent_buffer *eb;
335 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
336 int ret = 0;
337
338 tree = &BTRFS_I(page->mapping->host)->io_tree;
339 if (page->private == EXTENT_PAGE_PRIVATE)
340 goto out;
341 if (!page->private)
342 goto out;
343 len = page->private >> 2;
344 if (len == 0) {
345 WARN_ON(1);
346 }
347 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
348
349 found_start = btrfs_header_bytenr(eb);
350 if (found_start != start) {
351 ret = -EIO;
352 goto err;
353 }
354 if (eb->first_page != page) {
355 printk("bad first page %lu %lu\n", eb->first_page->index,
356 page->index);
357 WARN_ON(1);
358 ret = -EIO;
359 goto err;
360 }
361 if (memcmp_extent_buffer(eb, root->fs_info->fsid,
362 (unsigned long)btrfs_header_fsid(eb),
363 BTRFS_FSID_SIZE)) {
364 printk("bad fsid on block %Lu\n", eb->start);
365 ret = -EIO;
366 goto err;
367 }
368 found_level = btrfs_header_level(eb);
369
370 ret = csum_tree_block(root, eb, 1);
371 if (ret)
372 ret = -EIO;
373
374 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
375 end = eb->start + end - 1;
376 err:
377 free_extent_buffer(eb);
378 out:
379 return ret;
380 }
381
382 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
383 static void end_workqueue_bio(struct bio *bio, int err)
384 #else
385 static int end_workqueue_bio(struct bio *bio,
386 unsigned int bytes_done, int err)
387 #endif
388 {
389 struct end_io_wq *end_io_wq = bio->bi_private;
390 struct btrfs_fs_info *fs_info;
391
392 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
393 if (bio->bi_size)
394 return 1;
395 #endif
396
397 fs_info = end_io_wq->info;
398 end_io_wq->error = err;
399 end_io_wq->work.func = end_workqueue_fn;
400 end_io_wq->work.flags = 0;
401 if (bio->bi_rw & (1 << BIO_RW))
402 btrfs_queue_worker(&fs_info->endio_write_workers,
403 &end_io_wq->work);
404 else
405 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
406
407 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
408 return 0;
409 #endif
410 }
411
412 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
413 int metadata)
414 {
415 struct end_io_wq *end_io_wq;
416 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
417 if (!end_io_wq)
418 return -ENOMEM;
419
420 end_io_wq->private = bio->bi_private;
421 end_io_wq->end_io = bio->bi_end_io;
422 end_io_wq->info = info;
423 end_io_wq->error = 0;
424 end_io_wq->bio = bio;
425 end_io_wq->metadata = metadata;
426
427 bio->bi_private = end_io_wq;
428 bio->bi_end_io = end_workqueue_bio;
429 return 0;
430 }
431
432 static unsigned long async_submit_limit(struct btrfs_fs_info *info)
433 {
434 unsigned long limit = min_t(unsigned long,
435 info->workers.max_workers,
436 info->fs_devices->open_devices);
437 return 256 * limit;
438 }
439
440 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
441 {
442 return atomic_read(&info->nr_async_bios) > async_submit_limit(info);
443 }
444
445 static void run_one_async_submit(struct btrfs_work *work)
446 {
447 struct btrfs_fs_info *fs_info;
448 struct async_submit_bio *async;
449 int limit;
450
451 async = container_of(work, struct async_submit_bio, work);
452 fs_info = BTRFS_I(async->inode)->root->fs_info;
453
454 limit = async_submit_limit(fs_info);
455 limit = limit * 2 / 3;
456
457 atomic_dec(&fs_info->nr_async_submits);
458
459 if (atomic_read(&fs_info->nr_async_submits) < limit)
460 wake_up(&fs_info->async_submit_wait);
461
462 async->submit_bio_hook(async->inode, async->rw, async->bio,
463 async->mirror_num);
464 kfree(async);
465 }
466
467 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
468 int rw, struct bio *bio, int mirror_num,
469 extent_submit_bio_hook_t *submit_bio_hook)
470 {
471 struct async_submit_bio *async;
472 int limit = async_submit_limit(fs_info);
473
474 async = kmalloc(sizeof(*async), GFP_NOFS);
475 if (!async)
476 return -ENOMEM;
477
478 async->inode = inode;
479 async->rw = rw;
480 async->bio = bio;
481 async->mirror_num = mirror_num;
482 async->submit_bio_hook = submit_bio_hook;
483 async->work.func = run_one_async_submit;
484 async->work.flags = 0;
485 atomic_inc(&fs_info->nr_async_submits);
486 btrfs_queue_worker(&fs_info->workers, &async->work);
487
488 wait_event_timeout(fs_info->async_submit_wait,
489 (atomic_read(&fs_info->nr_async_submits) < limit),
490 HZ/10);
491 return 0;
492 }
493
494 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
495 int mirror_num)
496 {
497 struct btrfs_root *root = BTRFS_I(inode)->root;
498 u64 offset;
499 int ret;
500
501 offset = bio->bi_sector << 9;
502
503 /*
504 * when we're called for a write, we're already in the async
505 * submission context. Just jump into btrfs_map_bio
506 */
507 if (rw & (1 << BIO_RW)) {
508 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
509 mirror_num, 1);
510 }
511
512 /*
513 * called for a read, do the setup so that checksum validation
514 * can happen in the async kernel threads
515 */
516 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
517 BUG_ON(ret);
518
519 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
520 }
521
522 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
523 int mirror_num)
524 {
525 /*
526 * kthread helpers are used to submit writes so that checksumming
527 * can happen in parallel across all CPUs
528 */
529 if (!(rw & (1 << BIO_RW))) {
530 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
531 }
532 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
533 inode, rw, bio, mirror_num,
534 __btree_submit_bio_hook);
535 }
536
537 static int btree_writepage(struct page *page, struct writeback_control *wbc)
538 {
539 struct extent_io_tree *tree;
540 tree = &BTRFS_I(page->mapping->host)->io_tree;
541
542 if (current->flags & PF_MEMALLOC) {
543 redirty_page_for_writepage(wbc, page);
544 unlock_page(page);
545 return 0;
546 }
547 return extent_write_full_page(tree, page, btree_get_extent, wbc);
548 }
549
550 static int btree_writepages(struct address_space *mapping,
551 struct writeback_control *wbc)
552 {
553 struct extent_io_tree *tree;
554 tree = &BTRFS_I(mapping->host)->io_tree;
555 if (wbc->sync_mode == WB_SYNC_NONE) {
556 u64 num_dirty;
557 u64 start = 0;
558 unsigned long thresh = 8 * 1024 * 1024;
559
560 if (wbc->for_kupdate)
561 return 0;
562
563 num_dirty = count_range_bits(tree, &start, (u64)-1,
564 thresh, EXTENT_DIRTY);
565 if (num_dirty < thresh) {
566 return 0;
567 }
568 }
569 return extent_writepages(tree, mapping, btree_get_extent, wbc);
570 }
571
572 int btree_readpage(struct file *file, struct page *page)
573 {
574 struct extent_io_tree *tree;
575 tree = &BTRFS_I(page->mapping->host)->io_tree;
576 return extent_read_full_page(tree, page, btree_get_extent);
577 }
578
579 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
580 {
581 struct extent_io_tree *tree;
582 struct extent_map_tree *map;
583 int ret;
584
585 tree = &BTRFS_I(page->mapping->host)->io_tree;
586 map = &BTRFS_I(page->mapping->host)->extent_tree;
587
588 ret = try_release_extent_state(map, tree, page, gfp_flags);
589 if (!ret) {
590 return 0;
591 }
592
593 ret = try_release_extent_buffer(tree, page);
594 if (ret == 1) {
595 ClearPagePrivate(page);
596 set_page_private(page, 0);
597 page_cache_release(page);
598 }
599
600 return ret;
601 }
602
603 static void btree_invalidatepage(struct page *page, unsigned long offset)
604 {
605 struct extent_io_tree *tree;
606 tree = &BTRFS_I(page->mapping->host)->io_tree;
607 extent_invalidatepage(tree, page, offset);
608 btree_releasepage(page, GFP_NOFS);
609 if (PagePrivate(page)) {
610 printk("warning page private not zero on page %Lu\n",
611 page_offset(page));
612 ClearPagePrivate(page);
613 set_page_private(page, 0);
614 page_cache_release(page);
615 }
616 }
617
618 #if 0
619 static int btree_writepage(struct page *page, struct writeback_control *wbc)
620 {
621 struct buffer_head *bh;
622 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
623 struct buffer_head *head;
624 if (!page_has_buffers(page)) {
625 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
626 (1 << BH_Dirty)|(1 << BH_Uptodate));
627 }
628 head = page_buffers(page);
629 bh = head;
630 do {
631 if (buffer_dirty(bh))
632 csum_tree_block(root, bh, 0);
633 bh = bh->b_this_page;
634 } while (bh != head);
635 return block_write_full_page(page, btree_get_block, wbc);
636 }
637 #endif
638
639 static struct address_space_operations btree_aops = {
640 .readpage = btree_readpage,
641 .writepage = btree_writepage,
642 .writepages = btree_writepages,
643 .releasepage = btree_releasepage,
644 .invalidatepage = btree_invalidatepage,
645 .sync_page = block_sync_page,
646 };
647
648 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
649 u64 parent_transid)
650 {
651 struct extent_buffer *buf = NULL;
652 struct inode *btree_inode = root->fs_info->btree_inode;
653 int ret = 0;
654
655 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
656 if (!buf)
657 return 0;
658 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
659 buf, 0, 0, btree_get_extent, 0);
660 free_extent_buffer(buf);
661 return ret;
662 }
663
664 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
665 u64 bytenr, u32 blocksize)
666 {
667 struct inode *btree_inode = root->fs_info->btree_inode;
668 struct extent_buffer *eb;
669 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
670 bytenr, blocksize, GFP_NOFS);
671 return eb;
672 }
673
674 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
675 u64 bytenr, u32 blocksize)
676 {
677 struct inode *btree_inode = root->fs_info->btree_inode;
678 struct extent_buffer *eb;
679
680 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
681 bytenr, blocksize, NULL, GFP_NOFS);
682 return eb;
683 }
684
685
686 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
687 u32 blocksize, u64 parent_transid)
688 {
689 struct extent_buffer *buf = NULL;
690 struct inode *btree_inode = root->fs_info->btree_inode;
691 struct extent_io_tree *io_tree;
692 int ret;
693
694 io_tree = &BTRFS_I(btree_inode)->io_tree;
695
696 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
697 if (!buf)
698 return NULL;
699
700 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
701
702 if (ret == 0) {
703 buf->flags |= EXTENT_UPTODATE;
704 }
705 return buf;
706
707 }
708
709 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
710 struct extent_buffer *buf)
711 {
712 struct inode *btree_inode = root->fs_info->btree_inode;
713 if (btrfs_header_generation(buf) ==
714 root->fs_info->running_transaction->transid) {
715 WARN_ON(!btrfs_tree_locked(buf));
716 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
717 buf);
718 }
719 return 0;
720 }
721
722 int wait_on_tree_block_writeback(struct btrfs_root *root,
723 struct extent_buffer *buf)
724 {
725 struct inode *btree_inode = root->fs_info->btree_inode;
726 wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
727 buf);
728 return 0;
729 }
730
731 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
732 u32 stripesize, struct btrfs_root *root,
733 struct btrfs_fs_info *fs_info,
734 u64 objectid)
735 {
736 root->node = NULL;
737 root->inode = NULL;
738 root->commit_root = NULL;
739 root->ref_tree = NULL;
740 root->sectorsize = sectorsize;
741 root->nodesize = nodesize;
742 root->leafsize = leafsize;
743 root->stripesize = stripesize;
744 root->ref_cows = 0;
745 root->track_dirty = 0;
746
747 root->fs_info = fs_info;
748 root->objectid = objectid;
749 root->last_trans = 0;
750 root->highest_inode = 0;
751 root->last_inode_alloc = 0;
752 root->name = NULL;
753 root->in_sysfs = 0;
754
755 INIT_LIST_HEAD(&root->dirty_list);
756 INIT_LIST_HEAD(&root->orphan_list);
757 INIT_LIST_HEAD(&root->dead_list);
758 spin_lock_init(&root->node_lock);
759 spin_lock_init(&root->list_lock);
760 mutex_init(&root->objectid_mutex);
761
762 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
763 root->ref_tree = &root->ref_tree_struct;
764
765 memset(&root->root_key, 0, sizeof(root->root_key));
766 memset(&root->root_item, 0, sizeof(root->root_item));
767 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
768 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
769 root->defrag_trans_start = fs_info->generation;
770 init_completion(&root->kobj_unregister);
771 root->defrag_running = 0;
772 root->defrag_level = 0;
773 root->root_key.objectid = objectid;
774 return 0;
775 }
776
777 static int find_and_setup_root(struct btrfs_root *tree_root,
778 struct btrfs_fs_info *fs_info,
779 u64 objectid,
780 struct btrfs_root *root)
781 {
782 int ret;
783 u32 blocksize;
784
785 __setup_root(tree_root->nodesize, tree_root->leafsize,
786 tree_root->sectorsize, tree_root->stripesize,
787 root, fs_info, objectid);
788 ret = btrfs_find_last_root(tree_root, objectid,
789 &root->root_item, &root->root_key);
790 BUG_ON(ret);
791
792 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
793 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
794 blocksize, 0);
795 BUG_ON(!root->node);
796 return 0;
797 }
798
799 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
800 struct btrfs_key *location)
801 {
802 struct btrfs_root *root;
803 struct btrfs_root *tree_root = fs_info->tree_root;
804 struct btrfs_path *path;
805 struct extent_buffer *l;
806 u64 highest_inode;
807 u32 blocksize;
808 int ret = 0;
809
810 root = kzalloc(sizeof(*root), GFP_NOFS);
811 if (!root)
812 return ERR_PTR(-ENOMEM);
813 if (location->offset == (u64)-1) {
814 ret = find_and_setup_root(tree_root, fs_info,
815 location->objectid, root);
816 if (ret) {
817 kfree(root);
818 return ERR_PTR(ret);
819 }
820 goto insert;
821 }
822
823 __setup_root(tree_root->nodesize, tree_root->leafsize,
824 tree_root->sectorsize, tree_root->stripesize,
825 root, fs_info, location->objectid);
826
827 path = btrfs_alloc_path();
828 BUG_ON(!path);
829 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
830 if (ret != 0) {
831 if (ret > 0)
832 ret = -ENOENT;
833 goto out;
834 }
835 l = path->nodes[0];
836 read_extent_buffer(l, &root->root_item,
837 btrfs_item_ptr_offset(l, path->slots[0]),
838 sizeof(root->root_item));
839 memcpy(&root->root_key, location, sizeof(*location));
840 ret = 0;
841 out:
842 btrfs_release_path(root, path);
843 btrfs_free_path(path);
844 if (ret) {
845 kfree(root);
846 return ERR_PTR(ret);
847 }
848 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
849 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
850 blocksize, 0);
851 BUG_ON(!root->node);
852 insert:
853 root->ref_cows = 1;
854 ret = btrfs_find_highest_inode(root, &highest_inode);
855 if (ret == 0) {
856 root->highest_inode = highest_inode;
857 root->last_inode_alloc = highest_inode;
858 }
859 return root;
860 }
861
862 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
863 u64 root_objectid)
864 {
865 struct btrfs_root *root;
866
867 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
868 return fs_info->tree_root;
869 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
870 return fs_info->extent_root;
871
872 root = radix_tree_lookup(&fs_info->fs_roots_radix,
873 (unsigned long)root_objectid);
874 return root;
875 }
876
877 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
878 struct btrfs_key *location)
879 {
880 struct btrfs_root *root;
881 int ret;
882
883 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
884 return fs_info->tree_root;
885 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
886 return fs_info->extent_root;
887 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
888 return fs_info->chunk_root;
889 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
890 return fs_info->dev_root;
891
892 root = radix_tree_lookup(&fs_info->fs_roots_radix,
893 (unsigned long)location->objectid);
894 if (root)
895 return root;
896
897 root = btrfs_read_fs_root_no_radix(fs_info, location);
898 if (IS_ERR(root))
899 return root;
900 ret = radix_tree_insert(&fs_info->fs_roots_radix,
901 (unsigned long)root->root_key.objectid,
902 root);
903 if (ret) {
904 free_extent_buffer(root->node);
905 kfree(root);
906 return ERR_PTR(ret);
907 }
908 ret = btrfs_find_dead_roots(fs_info->tree_root,
909 root->root_key.objectid, root);
910 BUG_ON(ret);
911
912 return root;
913 }
914
915 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
916 struct btrfs_key *location,
917 const char *name, int namelen)
918 {
919 struct btrfs_root *root;
920 int ret;
921
922 root = btrfs_read_fs_root_no_name(fs_info, location);
923 if (!root)
924 return NULL;
925
926 if (root->in_sysfs)
927 return root;
928
929 ret = btrfs_set_root_name(root, name, namelen);
930 if (ret) {
931 free_extent_buffer(root->node);
932 kfree(root);
933 return ERR_PTR(ret);
934 }
935
936 ret = btrfs_sysfs_add_root(root);
937 if (ret) {
938 free_extent_buffer(root->node);
939 kfree(root->name);
940 kfree(root);
941 return ERR_PTR(ret);
942 }
943 root->in_sysfs = 1;
944 return root;
945 }
946 #if 0
947 static int add_hasher(struct btrfs_fs_info *info, char *type) {
948 struct btrfs_hasher *hasher;
949
950 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
951 if (!hasher)
952 return -ENOMEM;
953 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
954 if (!hasher->hash_tfm) {
955 kfree(hasher);
956 return -EINVAL;
957 }
958 spin_lock(&info->hash_lock);
959 list_add(&hasher->list, &info->hashers);
960 spin_unlock(&info->hash_lock);
961 return 0;
962 }
963 #endif
964
965 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
966 {
967 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
968 int ret = 0;
969 struct list_head *cur;
970 struct btrfs_device *device;
971 struct backing_dev_info *bdi;
972
973 if ((bdi_bits & (1 << BDI_write_congested)) &&
974 btrfs_congested_async(info, 0))
975 return 1;
976
977 list_for_each(cur, &info->fs_devices->devices) {
978 device = list_entry(cur, struct btrfs_device, dev_list);
979 if (!device->bdev)
980 continue;
981 bdi = blk_get_backing_dev_info(device->bdev);
982 if (bdi && bdi_congested(bdi, bdi_bits)) {
983 ret = 1;
984 break;
985 }
986 }
987 return ret;
988 }
989
990 /*
991 * this unplugs every device on the box, and it is only used when page
992 * is null
993 */
994 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
995 {
996 struct list_head *cur;
997 struct btrfs_device *device;
998 struct btrfs_fs_info *info;
999
1000 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1001 list_for_each(cur, &info->fs_devices->devices) {
1002 device = list_entry(cur, struct btrfs_device, dev_list);
1003 bdi = blk_get_backing_dev_info(device->bdev);
1004 if (bdi->unplug_io_fn) {
1005 bdi->unplug_io_fn(bdi, page);
1006 }
1007 }
1008 }
1009
1010 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1011 {
1012 struct inode *inode;
1013 struct extent_map_tree *em_tree;
1014 struct extent_map *em;
1015 struct address_space *mapping;
1016 u64 offset;
1017
1018 /* the generic O_DIRECT read code does this */
1019 if (!page) {
1020 __unplug_io_fn(bdi, page);
1021 return;
1022 }
1023
1024 /*
1025 * page->mapping may change at any time. Get a consistent copy
1026 * and use that for everything below
1027 */
1028 smp_mb();
1029 mapping = page->mapping;
1030 if (!mapping)
1031 return;
1032
1033 inode = mapping->host;
1034 offset = page_offset(page);
1035
1036 em_tree = &BTRFS_I(inode)->extent_tree;
1037 spin_lock(&em_tree->lock);
1038 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1039 spin_unlock(&em_tree->lock);
1040 if (!em) {
1041 __unplug_io_fn(bdi, page);
1042 return;
1043 }
1044
1045 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1046 free_extent_map(em);
1047 __unplug_io_fn(bdi, page);
1048 return;
1049 }
1050 offset = offset - em->start;
1051 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1052 em->block_start + offset, page);
1053 free_extent_map(em);
1054 }
1055
1056 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1057 {
1058 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1059 bdi_init(bdi);
1060 #endif
1061 bdi->ra_pages = default_backing_dev_info.ra_pages;
1062 bdi->state = 0;
1063 bdi->capabilities = default_backing_dev_info.capabilities;
1064 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1065 bdi->unplug_io_data = info;
1066 bdi->congested_fn = btrfs_congested_fn;
1067 bdi->congested_data = info;
1068 return 0;
1069 }
1070
1071 static int bio_ready_for_csum(struct bio *bio)
1072 {
1073 u64 length = 0;
1074 u64 buf_len = 0;
1075 u64 start = 0;
1076 struct page *page;
1077 struct extent_io_tree *io_tree = NULL;
1078 struct btrfs_fs_info *info = NULL;
1079 struct bio_vec *bvec;
1080 int i;
1081 int ret;
1082
1083 bio_for_each_segment(bvec, bio, i) {
1084 page = bvec->bv_page;
1085 if (page->private == EXTENT_PAGE_PRIVATE) {
1086 length += bvec->bv_len;
1087 continue;
1088 }
1089 if (!page->private) {
1090 length += bvec->bv_len;
1091 continue;
1092 }
1093 length = bvec->bv_len;
1094 buf_len = page->private >> 2;
1095 start = page_offset(page) + bvec->bv_offset;
1096 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1097 info = BTRFS_I(page->mapping->host)->root->fs_info;
1098 }
1099 /* are we fully contained in this bio? */
1100 if (buf_len <= length)
1101 return 1;
1102
1103 ret = extent_range_uptodate(io_tree, start + length,
1104 start + buf_len - 1);
1105 if (ret == 1)
1106 return ret;
1107 return ret;
1108 }
1109
1110 /*
1111 * called by the kthread helper functions to finally call the bio end_io
1112 * functions. This is where read checksum verification actually happens
1113 */
1114 static void end_workqueue_fn(struct btrfs_work *work)
1115 {
1116 struct bio *bio;
1117 struct end_io_wq *end_io_wq;
1118 struct btrfs_fs_info *fs_info;
1119 int error;
1120
1121 end_io_wq = container_of(work, struct end_io_wq, work);
1122 bio = end_io_wq->bio;
1123 fs_info = end_io_wq->info;
1124
1125 /* metadata bios are special because the whole tree block must
1126 * be checksummed at once. This makes sure the entire block is in
1127 * ram and up to date before trying to verify things. For
1128 * blocksize <= pagesize, it is basically a noop
1129 */
1130 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1131 btrfs_queue_worker(&fs_info->endio_workers,
1132 &end_io_wq->work);
1133 return;
1134 }
1135 error = end_io_wq->error;
1136 bio->bi_private = end_io_wq->private;
1137 bio->bi_end_io = end_io_wq->end_io;
1138 kfree(end_io_wq);
1139 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1140 bio_endio(bio, bio->bi_size, error);
1141 #else
1142 bio_endio(bio, error);
1143 #endif
1144 }
1145
1146 static int cleaner_kthread(void *arg)
1147 {
1148 struct btrfs_root *root = arg;
1149
1150 do {
1151 smp_mb();
1152 if (root->fs_info->closing)
1153 break;
1154
1155 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1156 mutex_lock(&root->fs_info->cleaner_mutex);
1157 btrfs_clean_old_snapshots(root);
1158 mutex_unlock(&root->fs_info->cleaner_mutex);
1159
1160 if (freezing(current)) {
1161 refrigerator();
1162 } else {
1163 smp_mb();
1164 if (root->fs_info->closing)
1165 break;
1166 set_current_state(TASK_INTERRUPTIBLE);
1167 schedule();
1168 __set_current_state(TASK_RUNNING);
1169 }
1170 } while (!kthread_should_stop());
1171 return 0;
1172 }
1173
1174 static int transaction_kthread(void *arg)
1175 {
1176 struct btrfs_root *root = arg;
1177 struct btrfs_trans_handle *trans;
1178 struct btrfs_transaction *cur;
1179 unsigned long now;
1180 unsigned long delay;
1181 int ret;
1182
1183 do {
1184 smp_mb();
1185 if (root->fs_info->closing)
1186 break;
1187
1188 delay = HZ * 30;
1189 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1190 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1191
1192 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1193 printk("btrfs: total reference cache size %Lu\n",
1194 root->fs_info->total_ref_cache_size);
1195 }
1196
1197 mutex_lock(&root->fs_info->trans_mutex);
1198 cur = root->fs_info->running_transaction;
1199 if (!cur) {
1200 mutex_unlock(&root->fs_info->trans_mutex);
1201 goto sleep;
1202 }
1203
1204 now = get_seconds();
1205 if (now < cur->start_time || now - cur->start_time < 30) {
1206 mutex_unlock(&root->fs_info->trans_mutex);
1207 delay = HZ * 5;
1208 goto sleep;
1209 }
1210 mutex_unlock(&root->fs_info->trans_mutex);
1211 trans = btrfs_start_transaction(root, 1);
1212 ret = btrfs_commit_transaction(trans, root);
1213 sleep:
1214 wake_up_process(root->fs_info->cleaner_kthread);
1215 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1216
1217 if (freezing(current)) {
1218 refrigerator();
1219 } else {
1220 if (root->fs_info->closing)
1221 break;
1222 set_current_state(TASK_INTERRUPTIBLE);
1223 schedule_timeout(delay);
1224 __set_current_state(TASK_RUNNING);
1225 }
1226 } while (!kthread_should_stop());
1227 return 0;
1228 }
1229
1230 struct btrfs_root *open_ctree(struct super_block *sb,
1231 struct btrfs_fs_devices *fs_devices,
1232 char *options)
1233 {
1234 u32 sectorsize;
1235 u32 nodesize;
1236 u32 leafsize;
1237 u32 blocksize;
1238 u32 stripesize;
1239 struct buffer_head *bh;
1240 struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1241 GFP_NOFS);
1242 struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1243 GFP_NOFS);
1244 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1245 GFP_NOFS);
1246 struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1247 GFP_NOFS);
1248 struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1249 GFP_NOFS);
1250 int ret;
1251 int err = -EINVAL;
1252
1253 struct btrfs_super_block *disk_super;
1254
1255 if (!extent_root || !tree_root || !fs_info) {
1256 err = -ENOMEM;
1257 goto fail;
1258 }
1259 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1260 INIT_LIST_HEAD(&fs_info->trans_list);
1261 INIT_LIST_HEAD(&fs_info->dead_roots);
1262 INIT_LIST_HEAD(&fs_info->hashers);
1263 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1264 spin_lock_init(&fs_info->hash_lock);
1265 spin_lock_init(&fs_info->delalloc_lock);
1266 spin_lock_init(&fs_info->new_trans_lock);
1267 spin_lock_init(&fs_info->ref_cache_lock);
1268
1269 init_completion(&fs_info->kobj_unregister);
1270 fs_info->tree_root = tree_root;
1271 fs_info->extent_root = extent_root;
1272 fs_info->chunk_root = chunk_root;
1273 fs_info->dev_root = dev_root;
1274 fs_info->fs_devices = fs_devices;
1275 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1276 INIT_LIST_HEAD(&fs_info->space_info);
1277 btrfs_mapping_init(&fs_info->mapping_tree);
1278 atomic_set(&fs_info->nr_async_submits, 0);
1279 atomic_set(&fs_info->nr_async_bios, 0);
1280 atomic_set(&fs_info->throttles, 0);
1281 atomic_set(&fs_info->throttle_gen, 0);
1282 fs_info->sb = sb;
1283 fs_info->max_extent = (u64)-1;
1284 fs_info->max_inline = 8192 * 1024;
1285 setup_bdi(fs_info, &fs_info->bdi);
1286 fs_info->btree_inode = new_inode(sb);
1287 fs_info->btree_inode->i_ino = 1;
1288 fs_info->btree_inode->i_nlink = 1;
1289 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1290
1291 INIT_LIST_HEAD(&fs_info->ordered_extents);
1292 spin_lock_init(&fs_info->ordered_extent_lock);
1293
1294 sb->s_blocksize = 4096;
1295 sb->s_blocksize_bits = blksize_bits(4096);
1296
1297 /*
1298 * we set the i_size on the btree inode to the max possible int.
1299 * the real end of the address space is determined by all of
1300 * the devices in the system
1301 */
1302 fs_info->btree_inode->i_size = OFFSET_MAX;
1303 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1304 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1305
1306 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1307 fs_info->btree_inode->i_mapping,
1308 GFP_NOFS);
1309 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1310 GFP_NOFS);
1311
1312 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1313
1314 extent_io_tree_init(&fs_info->free_space_cache,
1315 fs_info->btree_inode->i_mapping, GFP_NOFS);
1316 extent_io_tree_init(&fs_info->block_group_cache,
1317 fs_info->btree_inode->i_mapping, GFP_NOFS);
1318 extent_io_tree_init(&fs_info->pinned_extents,
1319 fs_info->btree_inode->i_mapping, GFP_NOFS);
1320 extent_io_tree_init(&fs_info->pending_del,
1321 fs_info->btree_inode->i_mapping, GFP_NOFS);
1322 extent_io_tree_init(&fs_info->extent_ins,
1323 fs_info->btree_inode->i_mapping, GFP_NOFS);
1324 fs_info->do_barriers = 1;
1325
1326 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1327 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1328 sizeof(struct btrfs_key));
1329 insert_inode_hash(fs_info->btree_inode);
1330 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1331
1332 mutex_init(&fs_info->trans_mutex);
1333 mutex_init(&fs_info->drop_mutex);
1334 mutex_init(&fs_info->alloc_mutex);
1335 mutex_init(&fs_info->chunk_mutex);
1336 mutex_init(&fs_info->transaction_kthread_mutex);
1337 mutex_init(&fs_info->cleaner_mutex);
1338 mutex_init(&fs_info->volume_mutex);
1339 init_waitqueue_head(&fs_info->transaction_throttle);
1340 init_waitqueue_head(&fs_info->transaction_wait);
1341 init_waitqueue_head(&fs_info->async_submit_wait);
1342
1343 #if 0
1344 ret = add_hasher(fs_info, "crc32c");
1345 if (ret) {
1346 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1347 err = -ENOMEM;
1348 goto fail_iput;
1349 }
1350 #endif
1351 __setup_root(4096, 4096, 4096, 4096, tree_root,
1352 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1353
1354
1355 bh = __bread(fs_devices->latest_bdev,
1356 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1357 if (!bh)
1358 goto fail_iput;
1359
1360 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1361 brelse(bh);
1362
1363 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1364
1365 disk_super = &fs_info->super_copy;
1366 if (!btrfs_super_root(disk_super))
1367 goto fail_sb_buffer;
1368
1369 err = btrfs_parse_options(tree_root, options);
1370 if (err)
1371 goto fail_sb_buffer;
1372
1373 /*
1374 * we need to start all the end_io workers up front because the
1375 * queue work function gets called at interrupt time, and so it
1376 * cannot dynamically grow.
1377 */
1378 btrfs_init_workers(&fs_info->workers, "worker",
1379 fs_info->thread_pool_size);
1380 btrfs_init_workers(&fs_info->submit_workers, "submit",
1381 min_t(u64, fs_devices->num_devices,
1382 fs_info->thread_pool_size));
1383
1384 /* a higher idle thresh on the submit workers makes it much more
1385 * likely that bios will be send down in a sane order to the
1386 * devices
1387 */
1388 fs_info->submit_workers.idle_thresh = 64;
1389 fs_info->workers.idle_thresh = 32;
1390
1391 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1392 btrfs_init_workers(&fs_info->endio_workers, "endio",
1393 fs_info->thread_pool_size);
1394 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1395 fs_info->thread_pool_size);
1396
1397 /*
1398 * endios are largely parallel and should have a very
1399 * low idle thresh
1400 */
1401 fs_info->endio_workers.idle_thresh = 4;
1402 fs_info->endio_write_workers.idle_thresh = 4;
1403
1404 btrfs_start_workers(&fs_info->workers, 1);
1405 btrfs_start_workers(&fs_info->submit_workers, 1);
1406 btrfs_start_workers(&fs_info->fixup_workers, 1);
1407 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1408 btrfs_start_workers(&fs_info->endio_write_workers,
1409 fs_info->thread_pool_size);
1410
1411 err = -EINVAL;
1412 if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1413 printk("Btrfs: wanted %llu devices, but found %llu\n",
1414 (unsigned long long)btrfs_super_num_devices(disk_super),
1415 (unsigned long long)fs_devices->open_devices);
1416 if (btrfs_test_opt(tree_root, DEGRADED))
1417 printk("continuing in degraded mode\n");
1418 else {
1419 goto fail_sb_buffer;
1420 }
1421 }
1422
1423 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1424
1425 nodesize = btrfs_super_nodesize(disk_super);
1426 leafsize = btrfs_super_leafsize(disk_super);
1427 sectorsize = btrfs_super_sectorsize(disk_super);
1428 stripesize = btrfs_super_stripesize(disk_super);
1429 tree_root->nodesize = nodesize;
1430 tree_root->leafsize = leafsize;
1431 tree_root->sectorsize = sectorsize;
1432 tree_root->stripesize = stripesize;
1433
1434 sb->s_blocksize = sectorsize;
1435 sb->s_blocksize_bits = blksize_bits(sectorsize);
1436
1437 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1438 sizeof(disk_super->magic))) {
1439 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1440 goto fail_sb_buffer;
1441 }
1442
1443 mutex_lock(&fs_info->chunk_mutex);
1444 ret = btrfs_read_sys_array(tree_root);
1445 mutex_unlock(&fs_info->chunk_mutex);
1446 if (ret) {
1447 printk("btrfs: failed to read the system array on %s\n",
1448 sb->s_id);
1449 goto fail_sys_array;
1450 }
1451
1452 blocksize = btrfs_level_size(tree_root,
1453 btrfs_super_chunk_root_level(disk_super));
1454
1455 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1456 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1457
1458 chunk_root->node = read_tree_block(chunk_root,
1459 btrfs_super_chunk_root(disk_super),
1460 blocksize, 0);
1461 BUG_ON(!chunk_root->node);
1462
1463 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1464 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1465 BTRFS_UUID_SIZE);
1466
1467 mutex_lock(&fs_info->chunk_mutex);
1468 ret = btrfs_read_chunk_tree(chunk_root);
1469 mutex_unlock(&fs_info->chunk_mutex);
1470 BUG_ON(ret);
1471
1472 btrfs_close_extra_devices(fs_devices);
1473
1474 blocksize = btrfs_level_size(tree_root,
1475 btrfs_super_root_level(disk_super));
1476
1477
1478 tree_root->node = read_tree_block(tree_root,
1479 btrfs_super_root(disk_super),
1480 blocksize, 0);
1481 if (!tree_root->node)
1482 goto fail_sb_buffer;
1483
1484
1485 ret = find_and_setup_root(tree_root, fs_info,
1486 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1487 if (ret)
1488 goto fail_tree_root;
1489 extent_root->track_dirty = 1;
1490
1491 ret = find_and_setup_root(tree_root, fs_info,
1492 BTRFS_DEV_TREE_OBJECTID, dev_root);
1493 dev_root->track_dirty = 1;
1494
1495 if (ret)
1496 goto fail_extent_root;
1497
1498 btrfs_read_block_groups(extent_root);
1499
1500 fs_info->generation = btrfs_super_generation(disk_super) + 1;
1501 fs_info->data_alloc_profile = (u64)-1;
1502 fs_info->metadata_alloc_profile = (u64)-1;
1503 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1504 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1505 "btrfs-cleaner");
1506 if (!fs_info->cleaner_kthread)
1507 goto fail_extent_root;
1508
1509 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1510 tree_root,
1511 "btrfs-transaction");
1512 if (!fs_info->transaction_kthread)
1513 goto fail_cleaner;
1514
1515
1516 return tree_root;
1517
1518 fail_cleaner:
1519 kthread_stop(fs_info->cleaner_kthread);
1520 fail_extent_root:
1521 free_extent_buffer(extent_root->node);
1522 fail_tree_root:
1523 free_extent_buffer(tree_root->node);
1524 fail_sys_array:
1525 fail_sb_buffer:
1526 btrfs_stop_workers(&fs_info->fixup_workers);
1527 btrfs_stop_workers(&fs_info->workers);
1528 btrfs_stop_workers(&fs_info->endio_workers);
1529 btrfs_stop_workers(&fs_info->endio_write_workers);
1530 btrfs_stop_workers(&fs_info->submit_workers);
1531 fail_iput:
1532 iput(fs_info->btree_inode);
1533 fail:
1534 btrfs_close_devices(fs_info->fs_devices);
1535 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1536
1537 kfree(extent_root);
1538 kfree(tree_root);
1539 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1540 bdi_destroy(&fs_info->bdi);
1541 #endif
1542 kfree(fs_info);
1543 return ERR_PTR(err);
1544 }
1545
1546 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1547 {
1548 char b[BDEVNAME_SIZE];
1549
1550 if (uptodate) {
1551 set_buffer_uptodate(bh);
1552 } else {
1553 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1554 printk(KERN_WARNING "lost page write due to "
1555 "I/O error on %s\n",
1556 bdevname(bh->b_bdev, b));
1557 }
1558 /* note, we dont' set_buffer_write_io_error because we have
1559 * our own ways of dealing with the IO errors
1560 */
1561 clear_buffer_uptodate(bh);
1562 }
1563 unlock_buffer(bh);
1564 put_bh(bh);
1565 }
1566
1567 int write_all_supers(struct btrfs_root *root)
1568 {
1569 struct list_head *cur;
1570 struct list_head *head = &root->fs_info->fs_devices->devices;
1571 struct btrfs_device *dev;
1572 struct btrfs_super_block *sb;
1573 struct btrfs_dev_item *dev_item;
1574 struct buffer_head *bh;
1575 int ret;
1576 int do_barriers;
1577 int max_errors;
1578 int total_errors = 0;
1579 u32 crc;
1580 u64 flags;
1581
1582 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1583 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1584
1585 sb = &root->fs_info->super_for_commit;
1586 dev_item = &sb->dev_item;
1587 list_for_each(cur, head) {
1588 dev = list_entry(cur, struct btrfs_device, dev_list);
1589 if (!dev->bdev) {
1590 total_errors++;
1591 continue;
1592 }
1593 if (!dev->in_fs_metadata)
1594 continue;
1595
1596 btrfs_set_stack_device_type(dev_item, dev->type);
1597 btrfs_set_stack_device_id(dev_item, dev->devid);
1598 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1599 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1600 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1601 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1602 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1603 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1604 flags = btrfs_super_flags(sb);
1605 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1606
1607
1608 crc = ~(u32)0;
1609 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1610 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1611 btrfs_csum_final(crc, sb->csum);
1612
1613 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1614 BTRFS_SUPER_INFO_SIZE);
1615
1616 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1617 dev->pending_io = bh;
1618
1619 get_bh(bh);
1620 set_buffer_uptodate(bh);
1621 lock_buffer(bh);
1622 bh->b_end_io = btrfs_end_buffer_write_sync;
1623
1624 if (do_barriers && dev->barriers) {
1625 ret = submit_bh(WRITE_BARRIER, bh);
1626 if (ret == -EOPNOTSUPP) {
1627 printk("btrfs: disabling barriers on dev %s\n",
1628 dev->name);
1629 set_buffer_uptodate(bh);
1630 dev->barriers = 0;
1631 get_bh(bh);
1632 lock_buffer(bh);
1633 ret = submit_bh(WRITE, bh);
1634 }
1635 } else {
1636 ret = submit_bh(WRITE, bh);
1637 }
1638 if (ret)
1639 total_errors++;
1640 }
1641 if (total_errors > max_errors) {
1642 printk("btrfs: %d errors while writing supers\n", total_errors);
1643 BUG();
1644 }
1645 total_errors = 0;
1646
1647 list_for_each(cur, head) {
1648 dev = list_entry(cur, struct btrfs_device, dev_list);
1649 if (!dev->bdev)
1650 continue;
1651 if (!dev->in_fs_metadata)
1652 continue;
1653
1654 BUG_ON(!dev->pending_io);
1655 bh = dev->pending_io;
1656 wait_on_buffer(bh);
1657 if (!buffer_uptodate(dev->pending_io)) {
1658 if (do_barriers && dev->barriers) {
1659 printk("btrfs: disabling barriers on dev %s\n",
1660 dev->name);
1661 set_buffer_uptodate(bh);
1662 get_bh(bh);
1663 lock_buffer(bh);
1664 dev->barriers = 0;
1665 ret = submit_bh(WRITE, bh);
1666 BUG_ON(ret);
1667 wait_on_buffer(bh);
1668 if (!buffer_uptodate(bh))
1669 total_errors++;
1670 } else {
1671 total_errors++;
1672 }
1673
1674 }
1675 dev->pending_io = NULL;
1676 brelse(bh);
1677 }
1678 if (total_errors > max_errors) {
1679 printk("btrfs: %d errors while writing supers\n", total_errors);
1680 BUG();
1681 }
1682 return 0;
1683 }
1684
1685 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1686 *root)
1687 {
1688 int ret;
1689
1690 ret = write_all_supers(root);
1691 return ret;
1692 }
1693
1694 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1695 {
1696 radix_tree_delete(&fs_info->fs_roots_radix,
1697 (unsigned long)root->root_key.objectid);
1698 if (root->in_sysfs)
1699 btrfs_sysfs_del_root(root);
1700 if (root->inode)
1701 iput(root->inode);
1702 if (root->node)
1703 free_extent_buffer(root->node);
1704 if (root->commit_root)
1705 free_extent_buffer(root->commit_root);
1706 if (root->name)
1707 kfree(root->name);
1708 kfree(root);
1709 return 0;
1710 }
1711
1712 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1713 {
1714 int ret;
1715 struct btrfs_root *gang[8];
1716 int i;
1717
1718 while(1) {
1719 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1720 (void **)gang, 0,
1721 ARRAY_SIZE(gang));
1722 if (!ret)
1723 break;
1724 for (i = 0; i < ret; i++)
1725 btrfs_free_fs_root(fs_info, gang[i]);
1726 }
1727 return 0;
1728 }
1729
1730 int close_ctree(struct btrfs_root *root)
1731 {
1732 int ret;
1733 struct btrfs_trans_handle *trans;
1734 struct btrfs_fs_info *fs_info = root->fs_info;
1735
1736 fs_info->closing = 1;
1737 smp_mb();
1738
1739 kthread_stop(root->fs_info->transaction_kthread);
1740 kthread_stop(root->fs_info->cleaner_kthread);
1741
1742 btrfs_clean_old_snapshots(root);
1743 trans = btrfs_start_transaction(root, 1);
1744 ret = btrfs_commit_transaction(trans, root);
1745 /* run commit again to drop the original snapshot */
1746 trans = btrfs_start_transaction(root, 1);
1747 btrfs_commit_transaction(trans, root);
1748 ret = btrfs_write_and_wait_transaction(NULL, root);
1749 BUG_ON(ret);
1750
1751 write_ctree_super(NULL, root);
1752
1753 if (fs_info->delalloc_bytes) {
1754 printk("btrfs: at unmount delalloc count %Lu\n",
1755 fs_info->delalloc_bytes);
1756 }
1757 if (fs_info->total_ref_cache_size) {
1758 printk("btrfs: at umount reference cache size %Lu\n",
1759 fs_info->total_ref_cache_size);
1760 }
1761
1762 if (fs_info->extent_root->node)
1763 free_extent_buffer(fs_info->extent_root->node);
1764
1765 if (fs_info->tree_root->node)
1766 free_extent_buffer(fs_info->tree_root->node);
1767
1768 if (root->fs_info->chunk_root->node);
1769 free_extent_buffer(root->fs_info->chunk_root->node);
1770
1771 if (root->fs_info->dev_root->node);
1772 free_extent_buffer(root->fs_info->dev_root->node);
1773
1774 btrfs_free_block_groups(root->fs_info);
1775 fs_info->closing = 2;
1776 del_fs_roots(fs_info);
1777
1778 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1779
1780 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1781
1782 btrfs_stop_workers(&fs_info->fixup_workers);
1783 btrfs_stop_workers(&fs_info->workers);
1784 btrfs_stop_workers(&fs_info->endio_workers);
1785 btrfs_stop_workers(&fs_info->endio_write_workers);
1786 btrfs_stop_workers(&fs_info->submit_workers);
1787
1788 iput(fs_info->btree_inode);
1789 #if 0
1790 while(!list_empty(&fs_info->hashers)) {
1791 struct btrfs_hasher *hasher;
1792 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1793 hashers);
1794 list_del(&hasher->hashers);
1795 crypto_free_hash(&fs_info->hash_tfm);
1796 kfree(hasher);
1797 }
1798 #endif
1799 btrfs_close_devices(fs_info->fs_devices);
1800 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1801
1802 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1803 bdi_destroy(&fs_info->bdi);
1804 #endif
1805
1806 kfree(fs_info->extent_root);
1807 kfree(fs_info->tree_root);
1808 kfree(fs_info->chunk_root);
1809 kfree(fs_info->dev_root);
1810 return 0;
1811 }
1812
1813 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1814 {
1815 int ret;
1816 struct inode *btree_inode = buf->first_page->mapping->host;
1817
1818 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1819 if (!ret)
1820 return ret;
1821
1822 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1823 parent_transid);
1824 return !ret;
1825 }
1826
1827 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1828 {
1829 struct inode *btree_inode = buf->first_page->mapping->host;
1830 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1831 buf);
1832 }
1833
1834 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1835 {
1836 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1837 u64 transid = btrfs_header_generation(buf);
1838 struct inode *btree_inode = root->fs_info->btree_inode;
1839
1840 WARN_ON(!btrfs_tree_locked(buf));
1841 if (transid != root->fs_info->generation) {
1842 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1843 (unsigned long long)buf->start,
1844 transid, root->fs_info->generation);
1845 WARN_ON(1);
1846 }
1847 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1848 }
1849
1850 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1851 {
1852 /*
1853 * looks as though older kernels can get into trouble with
1854 * this code, they end up stuck in balance_dirty_pages forever
1855 */
1856 struct extent_io_tree *tree;
1857 u64 num_dirty;
1858 u64 start = 0;
1859 unsigned long thresh = 12 * 1024 * 1024;
1860 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1861
1862 if (current_is_pdflush())
1863 return;
1864
1865 num_dirty = count_range_bits(tree, &start, (u64)-1,
1866 thresh, EXTENT_DIRTY);
1867 if (num_dirty > thresh) {
1868 balance_dirty_pages_ratelimited_nr(
1869 root->fs_info->btree_inode->i_mapping, 1);
1870 }
1871 return;
1872 }
1873
1874 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1875 {
1876 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1877 int ret;
1878 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1879 if (ret == 0) {
1880 buf->flags |= EXTENT_UPTODATE;
1881 }
1882 return ret;
1883 }
1884
1885 static struct extent_io_ops btree_extent_io_ops = {
1886 .writepage_io_hook = btree_writepage_io_hook,
1887 .readpage_end_io_hook = btree_readpage_end_io_hook,
1888 .submit_bio_hook = btree_submit_bio_hook,
1889 /* note we're sharing with inode.c for the merge bio hook */
1890 .merge_bio_hook = btrfs_merge_bio_hook,
1891 };
kernel-based virtual machine