projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: More throttle tuning
[deliverable/linux.git] / fs / btrfs / disk-io.c
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 void run_one_async_submit(struct btrfs_work *work)
433 {
434 struct btrfs_fs_info *fs_info;
435 struct async_submit_bio *async;
436
437 async = container_of(work, struct async_submit_bio, work);
438 fs_info = BTRFS_I(async->inode)->root->fs_info;
439 atomic_dec(&fs_info->nr_async_submits);
440 async->submit_bio_hook(async->inode, async->rw, async->bio,
441 async->mirror_num);
442 kfree(async);
443 }
444
445 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
446 int rw, struct bio *bio, int mirror_num,
447 extent_submit_bio_hook_t *submit_bio_hook)
448 {
449 struct async_submit_bio *async;
450
451 async = kmalloc(sizeof(*async), GFP_NOFS);
452 if (!async)
453 return -ENOMEM;
454
455 async->inode = inode;
456 async->rw = rw;
457 async->bio = bio;
458 async->mirror_num = mirror_num;
459 async->submit_bio_hook = submit_bio_hook;
460 async->work.func = run_one_async_submit;
461 async->work.flags = 0;
462 atomic_inc(&fs_info->nr_async_submits);
463 btrfs_queue_worker(&fs_info->workers, &async->work);
464 return 0;
465 }
466
467 static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
468 int mirror_num)
469 {
470 struct btrfs_root *root = BTRFS_I(inode)->root;
471 u64 offset;
472 int ret;
473
474 offset = bio->bi_sector << 9;
475
476 /*
477 * when we're called for a write, we're already in the async
478 * submission context. Just jump ingo btrfs_map_bio
479 */
480 if (rw & (1 << BIO_RW)) {
481 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
482 mirror_num, 0);
483 }
484
485 /*
486 * called for a read, do the setup so that checksum validation
487 * can happen in the async kernel threads
488 */
489 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
490 BUG_ON(ret);
491
492 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
493 }
494
495 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
496 int mirror_num)
497 {
498 /*
499 * kthread helpers are used to submit writes so that checksumming
500 * can happen in parallel across all CPUs
501 */
502 if (!(rw & (1 << BIO_RW))) {
503 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
504 }
505 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
506 inode, rw, bio, mirror_num,
507 __btree_submit_bio_hook);
508 }
509
510 static int btree_writepage(struct page *page, struct writeback_control *wbc)
511 {
512 struct extent_io_tree *tree;
513 tree = &BTRFS_I(page->mapping->host)->io_tree;
514 return extent_write_full_page(tree, page, btree_get_extent, wbc);
515 }
516
517 static int btree_writepages(struct address_space *mapping,
518 struct writeback_control *wbc)
519 {
520 struct extent_io_tree *tree;
521 tree = &BTRFS_I(mapping->host)->io_tree;
522 if (wbc->sync_mode == WB_SYNC_NONE) {
523 u64 num_dirty;
524 u64 start = 0;
525 unsigned long thresh = 96 * 1024 * 1024;
526
527 if (wbc->for_kupdate)
528 return 0;
529
530 if (current_is_pdflush()) {
531 thresh = 96 * 1024 * 1024;
532 } else {
533 thresh = 8 * 1024 * 1024;
534 }
535 num_dirty = count_range_bits(tree, &start, (u64)-1,
536 thresh, EXTENT_DIRTY);
537 if (num_dirty < thresh) {
538 return 0;
539 }
540 }
541 return extent_writepages(tree, mapping, btree_get_extent, wbc);
542 }
543
544 int btree_readpage(struct file *file, struct page *page)
545 {
546 struct extent_io_tree *tree;
547 tree = &BTRFS_I(page->mapping->host)->io_tree;
548 return extent_read_full_page(tree, page, btree_get_extent);
549 }
550
551 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
552 {
553 struct extent_io_tree *tree;
554 struct extent_map_tree *map;
555 int ret;
556
557 tree = &BTRFS_I(page->mapping->host)->io_tree;
558 map = &BTRFS_I(page->mapping->host)->extent_tree;
559
560 ret = try_release_extent_state(map, tree, page, gfp_flags);
561 if (!ret) {
562 return 0;
563 }
564
565 ret = try_release_extent_buffer(tree, page);
566 if (ret == 1) {
567 ClearPagePrivate(page);
568 set_page_private(page, 0);
569 page_cache_release(page);
570 }
571
572 return ret;
573 }
574
575 static void btree_invalidatepage(struct page *page, unsigned long offset)
576 {
577 struct extent_io_tree *tree;
578 tree = &BTRFS_I(page->mapping->host)->io_tree;
579 extent_invalidatepage(tree, page, offset);
580 btree_releasepage(page, GFP_NOFS);
581 if (PagePrivate(page)) {
582 printk("warning page private not zero on page %Lu\n",
583 page_offset(page));
584 ClearPagePrivate(page);
585 set_page_private(page, 0);
586 page_cache_release(page);
587 }
588 }
589
590 #if 0
591 static int btree_writepage(struct page *page, struct writeback_control *wbc)
592 {
593 struct buffer_head *bh;
594 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
595 struct buffer_head *head;
596 if (!page_has_buffers(page)) {
597 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
598 (1 << BH_Dirty)|(1 << BH_Uptodate));
599 }
600 head = page_buffers(page);
601 bh = head;
602 do {
603 if (buffer_dirty(bh))
604 csum_tree_block(root, bh, 0);
605 bh = bh->b_this_page;
606 } while (bh != head);
607 return block_write_full_page(page, btree_get_block, wbc);
608 }
609 #endif
610
611 static struct address_space_operations btree_aops = {
612 .readpage = btree_readpage,
613 .writepage = btree_writepage,
614 .writepages = btree_writepages,
615 .releasepage = btree_releasepage,
616 .invalidatepage = btree_invalidatepage,
617 .sync_page = block_sync_page,
618 };
619
620 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
621 u64 parent_transid)
622 {
623 struct extent_buffer *buf = NULL;
624 struct inode *btree_inode = root->fs_info->btree_inode;
625 int ret = 0;
626
627 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
628 if (!buf)
629 return 0;
630 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
631 buf, 0, 0, btree_get_extent, 0);
632 free_extent_buffer(buf);
633 return ret;
634 }
635
636 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
637 u64 bytenr, u32 blocksize)
638 {
639 struct inode *btree_inode = root->fs_info->btree_inode;
640 struct extent_buffer *eb;
641 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
642 bytenr, blocksize, GFP_NOFS);
643 return eb;
644 }
645
646 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
647 u64 bytenr, u32 blocksize)
648 {
649 struct inode *btree_inode = root->fs_info->btree_inode;
650 struct extent_buffer *eb;
651
652 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
653 bytenr, blocksize, NULL, GFP_NOFS);
654 return eb;
655 }
656
657
658 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
659 u32 blocksize, u64 parent_transid)
660 {
661 struct extent_buffer *buf = NULL;
662 struct inode *btree_inode = root->fs_info->btree_inode;
663 struct extent_io_tree *io_tree;
664 int ret;
665
666 io_tree = &BTRFS_I(btree_inode)->io_tree;
667
668 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
669 if (!buf)
670 return NULL;
671
672 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
673
674 if (ret == 0) {
675 buf->flags |= EXTENT_UPTODATE;
676 }
677 return buf;
678
679 }
680
681 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
682 struct extent_buffer *buf)
683 {
684 struct inode *btree_inode = root->fs_info->btree_inode;
685 if (btrfs_header_generation(buf) ==
686 root->fs_info->running_transaction->transid) {
687 WARN_ON(!btrfs_tree_locked(buf));
688 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
689 buf);
690 }
691 return 0;
692 }
693
694 int wait_on_tree_block_writeback(struct btrfs_root *root,
695 struct extent_buffer *buf)
696 {
697 struct inode *btree_inode = root->fs_info->btree_inode;
698 wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
699 buf);
700 return 0;
701 }
702
703 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
704 u32 stripesize, struct btrfs_root *root,
705 struct btrfs_fs_info *fs_info,
706 u64 objectid)
707 {
708 root->node = NULL;
709 root->inode = NULL;
710 root->commit_root = NULL;
711 root->ref_tree = NULL;
712 root->sectorsize = sectorsize;
713 root->nodesize = nodesize;
714 root->leafsize = leafsize;
715 root->stripesize = stripesize;
716 root->ref_cows = 0;
717 root->track_dirty = 0;
718
719 root->fs_info = fs_info;
720 root->objectid = objectid;
721 root->last_trans = 0;
722 root->highest_inode = 0;
723 root->last_inode_alloc = 0;
724 root->name = NULL;
725 root->in_sysfs = 0;
726
727 INIT_LIST_HEAD(&root->dirty_list);
728 INIT_LIST_HEAD(&root->orphan_list);
729 INIT_LIST_HEAD(&root->dead_list);
730 spin_lock_init(&root->node_lock);
731 spin_lock_init(&root->list_lock);
732 mutex_init(&root->objectid_mutex);
733
734 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
735 root->ref_tree = &root->ref_tree_struct;
736
737 memset(&root->root_key, 0, sizeof(root->root_key));
738 memset(&root->root_item, 0, sizeof(root->root_item));
739 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
740 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
741 root->defrag_trans_start = fs_info->generation;
742 init_completion(&root->kobj_unregister);
743 root->defrag_running = 0;
744 root->defrag_level = 0;
745 root->root_key.objectid = objectid;
746 return 0;
747 }
748
749 static int find_and_setup_root(struct btrfs_root *tree_root,
750 struct btrfs_fs_info *fs_info,
751 u64 objectid,
752 struct btrfs_root *root)
753 {
754 int ret;
755 u32 blocksize;
756
757 __setup_root(tree_root->nodesize, tree_root->leafsize,
758 tree_root->sectorsize, tree_root->stripesize,
759 root, fs_info, objectid);
760 ret = btrfs_find_last_root(tree_root, objectid,
761 &root->root_item, &root->root_key);
762 BUG_ON(ret);
763
764 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
765 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
766 blocksize, 0);
767 BUG_ON(!root->node);
768 return 0;
769 }
770
771 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
772 struct btrfs_key *location)
773 {
774 struct btrfs_root *root;
775 struct btrfs_root *tree_root = fs_info->tree_root;
776 struct btrfs_path *path;
777 struct extent_buffer *l;
778 u64 highest_inode;
779 u32 blocksize;
780 int ret = 0;
781
782 root = kzalloc(sizeof(*root), GFP_NOFS);
783 if (!root)
784 return ERR_PTR(-ENOMEM);
785 if (location->offset == (u64)-1) {
786 ret = find_and_setup_root(tree_root, fs_info,
787 location->objectid, root);
788 if (ret) {
789 kfree(root);
790 return ERR_PTR(ret);
791 }
792 goto insert;
793 }
794
795 __setup_root(tree_root->nodesize, tree_root->leafsize,
796 tree_root->sectorsize, tree_root->stripesize,
797 root, fs_info, location->objectid);
798
799 path = btrfs_alloc_path();
800 BUG_ON(!path);
801 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
802 if (ret != 0) {
803 if (ret > 0)
804 ret = -ENOENT;
805 goto out;
806 }
807 l = path->nodes[0];
808 read_extent_buffer(l, &root->root_item,
809 btrfs_item_ptr_offset(l, path->slots[0]),
810 sizeof(root->root_item));
811 memcpy(&root->root_key, location, sizeof(*location));
812 ret = 0;
813 out:
814 btrfs_release_path(root, path);
815 btrfs_free_path(path);
816 if (ret) {
817 kfree(root);
818 return ERR_PTR(ret);
819 }
820 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
821 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
822 blocksize, 0);
823 BUG_ON(!root->node);
824 insert:
825 root->ref_cows = 1;
826 ret = btrfs_find_highest_inode(root, &highest_inode);
827 if (ret == 0) {
828 root->highest_inode = highest_inode;
829 root->last_inode_alloc = highest_inode;
830 }
831 return root;
832 }
833
834 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
835 u64 root_objectid)
836 {
837 struct btrfs_root *root;
838
839 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
840 return fs_info->tree_root;
841 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
842 return fs_info->extent_root;
843
844 root = radix_tree_lookup(&fs_info->fs_roots_radix,
845 (unsigned long)root_objectid);
846 return root;
847 }
848
849 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
850 struct btrfs_key *location)
851 {
852 struct btrfs_root *root;
853 int ret;
854
855 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
856 return fs_info->tree_root;
857 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
858 return fs_info->extent_root;
859 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
860 return fs_info->chunk_root;
861 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
862 return fs_info->dev_root;
863
864 root = radix_tree_lookup(&fs_info->fs_roots_radix,
865 (unsigned long)location->objectid);
866 if (root)
867 return root;
868
869 root = btrfs_read_fs_root_no_radix(fs_info, location);
870 if (IS_ERR(root))
871 return root;
872 ret = radix_tree_insert(&fs_info->fs_roots_radix,
873 (unsigned long)root->root_key.objectid,
874 root);
875 if (ret) {
876 free_extent_buffer(root->node);
877 kfree(root);
878 return ERR_PTR(ret);
879 }
880 ret = btrfs_find_dead_roots(fs_info->tree_root,
881 root->root_key.objectid, root);
882 BUG_ON(ret);
883
884 return root;
885 }
886
887 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
888 struct btrfs_key *location,
889 const char *name, int namelen)
890 {
891 struct btrfs_root *root;
892 int ret;
893
894 root = btrfs_read_fs_root_no_name(fs_info, location);
895 if (!root)
896 return NULL;
897
898 if (root->in_sysfs)
899 return root;
900
901 ret = btrfs_set_root_name(root, name, namelen);
902 if (ret) {
903 free_extent_buffer(root->node);
904 kfree(root);
905 return ERR_PTR(ret);
906 }
907
908 ret = btrfs_sysfs_add_root(root);
909 if (ret) {
910 free_extent_buffer(root->node);
911 kfree(root->name);
912 kfree(root);
913 return ERR_PTR(ret);
914 }
915 root->in_sysfs = 1;
916 return root;
917 }
918 #if 0
919 static int add_hasher(struct btrfs_fs_info *info, char *type) {
920 struct btrfs_hasher *hasher;
921
922 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
923 if (!hasher)
924 return -ENOMEM;
925 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
926 if (!hasher->hash_tfm) {
927 kfree(hasher);
928 return -EINVAL;
929 }
930 spin_lock(&info->hash_lock);
931 list_add(&hasher->list, &info->hashers);
932 spin_unlock(&info->hash_lock);
933 return 0;
934 }
935 #endif
936
937 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
938 {
939 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
940 int ret = 0;
941 int limit = 256 * info->fs_devices->open_devices;
942 struct list_head *cur;
943 struct btrfs_device *device;
944 struct backing_dev_info *bdi;
945
946 if ((bdi_bits & (1 << BDI_write_congested)) &&
947 atomic_read(&info->nr_async_submits) > limit) {
948 return 1;
949 }
950
951 list_for_each(cur, &info->fs_devices->devices) {
952 device = list_entry(cur, struct btrfs_device, dev_list);
953 if (!device->bdev)
954 continue;
955 bdi = blk_get_backing_dev_info(device->bdev);
956 if (bdi && bdi_congested(bdi, bdi_bits)) {
957 ret = 1;
958 break;
959 }
960 }
961 return ret;
962 }
963
964 /*
965 * this unplugs every device on the box, and it is only used when page
966 * is null
967 */
968 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
969 {
970 struct list_head *cur;
971 struct btrfs_device *device;
972 struct btrfs_fs_info *info;
973
974 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
975 list_for_each(cur, &info->fs_devices->devices) {
976 device = list_entry(cur, struct btrfs_device, dev_list);
977 bdi = blk_get_backing_dev_info(device->bdev);
978 if (bdi->unplug_io_fn) {
979 bdi->unplug_io_fn(bdi, page);
980 }
981 }
982 }
983
984 void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
985 {
986 struct inode *inode;
987 struct extent_map_tree *em_tree;
988 struct extent_map *em;
989 struct address_space *mapping;
990 u64 offset;
991
992 /* the generic O_DIRECT read code does this */
993 if (!page) {
994 __unplug_io_fn(bdi, page);
995 return;
996 }
997
998 /*
999 * page->mapping may change at any time. Get a consistent copy
1000 * and use that for everything below
1001 */
1002 smp_mb();
1003 mapping = page->mapping;
1004 if (!mapping)
1005 return;
1006
1007 inode = mapping->host;
1008 offset = page_offset(page);
1009
1010 em_tree = &BTRFS_I(inode)->extent_tree;
1011 spin_lock(&em_tree->lock);
1012 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1013 spin_unlock(&em_tree->lock);
1014 if (!em) {
1015 __unplug_io_fn(bdi, page);
1016 return;
1017 }
1018
1019 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1020 free_extent_map(em);
1021 __unplug_io_fn(bdi, page);
1022 return;
1023 }
1024 offset = offset - em->start;
1025 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1026 em->block_start + offset, page);
1027 free_extent_map(em);
1028 }
1029
1030 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1031 {
1032 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1033 bdi_init(bdi);
1034 #endif
1035 bdi->ra_pages = default_backing_dev_info.ra_pages;
1036 bdi->state = 0;
1037 bdi->capabilities = default_backing_dev_info.capabilities;
1038 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1039 bdi->unplug_io_data = info;
1040 bdi->congested_fn = btrfs_congested_fn;
1041 bdi->congested_data = info;
1042 return 0;
1043 }
1044
1045 static int bio_ready_for_csum(struct bio *bio)
1046 {
1047 u64 length = 0;
1048 u64 buf_len = 0;
1049 u64 start = 0;
1050 struct page *page;
1051 struct extent_io_tree *io_tree = NULL;
1052 struct btrfs_fs_info *info = NULL;
1053 struct bio_vec *bvec;
1054 int i;
1055 int ret;
1056
1057 bio_for_each_segment(bvec, bio, i) {
1058 page = bvec->bv_page;
1059 if (page->private == EXTENT_PAGE_PRIVATE) {
1060 length += bvec->bv_len;
1061 continue;
1062 }
1063 if (!page->private) {
1064 length += bvec->bv_len;
1065 continue;
1066 }
1067 length = bvec->bv_len;
1068 buf_len = page->private >> 2;
1069 start = page_offset(page) + bvec->bv_offset;
1070 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1071 info = BTRFS_I(page->mapping->host)->root->fs_info;
1072 }
1073 /* are we fully contained in this bio? */
1074 if (buf_len <= length)
1075 return 1;
1076
1077 ret = extent_range_uptodate(io_tree, start + length,
1078 start + buf_len - 1);
1079 if (ret == 1)
1080 return ret;
1081 return ret;
1082 }
1083
1084 /*
1085 * called by the kthread helper functions to finally call the bio end_io
1086 * functions. This is where read checksum verification actually happens
1087 */
1088 static void end_workqueue_fn(struct btrfs_work *work)
1089 {
1090 struct bio *bio;
1091 struct end_io_wq *end_io_wq;
1092 struct btrfs_fs_info *fs_info;
1093 int error;
1094
1095 end_io_wq = container_of(work, struct end_io_wq, work);
1096 bio = end_io_wq->bio;
1097 fs_info = end_io_wq->info;
1098
1099 /* metadata bios are special because the whole tree block must
1100 * be checksummed at once. This makes sure the entire block is in
1101 * ram and up to date before trying to verify things. For
1102 * blocksize <= pagesize, it is basically a noop
1103 */
1104 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1105 btrfs_queue_worker(&fs_info->endio_workers,
1106 &end_io_wq->work);
1107 return;
1108 }
1109 error = end_io_wq->error;
1110 bio->bi_private = end_io_wq->private;
1111 bio->bi_end_io = end_io_wq->end_io;
1112 kfree(end_io_wq);
1113 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1114 bio_endio(bio, bio->bi_size, error);
1115 #else
1116 bio_endio(bio, error);
1117 #endif
1118 }
1119
1120 static int cleaner_kthread(void *arg)
1121 {
1122 struct btrfs_root *root = arg;
1123
1124 do {
1125 smp_mb();
1126 if (root->fs_info->closing)
1127 break;
1128
1129 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1130 mutex_lock(&root->fs_info->cleaner_mutex);
1131 btrfs_clean_old_snapshots(root);
1132 mutex_unlock(&root->fs_info->cleaner_mutex);
1133
1134 if (freezing(current)) {
1135 refrigerator();
1136 } else {
1137 smp_mb();
1138 if (root->fs_info->closing)
1139 break;
1140 set_current_state(TASK_INTERRUPTIBLE);
1141 schedule();
1142 __set_current_state(TASK_RUNNING);
1143 }
1144 } while (!kthread_should_stop());
1145 return 0;
1146 }
1147
1148 static int transaction_kthread(void *arg)
1149 {
1150 struct btrfs_root *root = arg;
1151 struct btrfs_trans_handle *trans;
1152 struct btrfs_transaction *cur;
1153 unsigned long now;
1154 unsigned long delay;
1155 int ret;
1156
1157 do {
1158 smp_mb();
1159 if (root->fs_info->closing)
1160 break;
1161
1162 delay = HZ * 30;
1163 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1164 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1165
1166 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1167 printk("btrfs: total reference cache size %Lu\n",
1168 root->fs_info->total_ref_cache_size);
1169 }
1170
1171 mutex_lock(&root->fs_info->trans_mutex);
1172 cur = root->fs_info->running_transaction;
1173 if (!cur) {
1174 mutex_unlock(&root->fs_info->trans_mutex);
1175 goto sleep;
1176 }
1177
1178 now = get_seconds();
1179 if (now < cur->start_time || now - cur->start_time < 30) {
1180 mutex_unlock(&root->fs_info->trans_mutex);
1181 delay = HZ * 5;
1182 goto sleep;
1183 }
1184 mutex_unlock(&root->fs_info->trans_mutex);
1185 trans = btrfs_start_transaction(root, 1);
1186 ret = btrfs_commit_transaction(trans, root);
1187 sleep:
1188 wake_up_process(root->fs_info->cleaner_kthread);
1189 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1190
1191 if (freezing(current)) {
1192 refrigerator();
1193 } else {
1194 if (root->fs_info->closing)
1195 break;
1196 set_current_state(TASK_INTERRUPTIBLE);
1197 schedule_timeout(delay);
1198 __set_current_state(TASK_RUNNING);
1199 }
1200 } while (!kthread_should_stop());
1201 return 0;
1202 }
1203
1204 struct btrfs_root *open_ctree(struct super_block *sb,
1205 struct btrfs_fs_devices *fs_devices,
1206 char *options)
1207 {
1208 u32 sectorsize;
1209 u32 nodesize;
1210 u32 leafsize;
1211 u32 blocksize;
1212 u32 stripesize;
1213 struct buffer_head *bh;
1214 struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1215 GFP_NOFS);
1216 struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1217 GFP_NOFS);
1218 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1219 GFP_NOFS);
1220 struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1221 GFP_NOFS);
1222 struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1223 GFP_NOFS);
1224 int ret;
1225 int err = -EINVAL;
1226
1227 struct btrfs_super_block *disk_super;
1228
1229 if (!extent_root || !tree_root || !fs_info) {
1230 err = -ENOMEM;
1231 goto fail;
1232 }
1233 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1234 INIT_LIST_HEAD(&fs_info->trans_list);
1235 INIT_LIST_HEAD(&fs_info->dead_roots);
1236 INIT_LIST_HEAD(&fs_info->hashers);
1237 spin_lock_init(&fs_info->hash_lock);
1238 spin_lock_init(&fs_info->delalloc_lock);
1239 spin_lock_init(&fs_info->new_trans_lock);
1240 spin_lock_init(&fs_info->ref_cache_lock);
1241
1242 init_completion(&fs_info->kobj_unregister);
1243 fs_info->tree_root = tree_root;
1244 fs_info->extent_root = extent_root;
1245 fs_info->chunk_root = chunk_root;
1246 fs_info->dev_root = dev_root;
1247 fs_info->fs_devices = fs_devices;
1248 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1249 INIT_LIST_HEAD(&fs_info->space_info);
1250 btrfs_mapping_init(&fs_info->mapping_tree);
1251 atomic_set(&fs_info->nr_async_submits, 0);
1252 atomic_set(&fs_info->throttles, 0);
1253 atomic_set(&fs_info->throttle_gen, 0);
1254 fs_info->sb = sb;
1255 fs_info->max_extent = (u64)-1;
1256 fs_info->max_inline = 8192 * 1024;
1257 setup_bdi(fs_info, &fs_info->bdi);
1258 fs_info->btree_inode = new_inode(sb);
1259 fs_info->btree_inode->i_ino = 1;
1260 fs_info->btree_inode->i_nlink = 1;
1261 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1262
1263 INIT_LIST_HEAD(&fs_info->ordered_extents);
1264 spin_lock_init(&fs_info->ordered_extent_lock);
1265
1266 sb->s_blocksize = 4096;
1267 sb->s_blocksize_bits = blksize_bits(4096);
1268
1269 /*
1270 * we set the i_size on the btree inode to the max possible int.
1271 * the real end of the address space is determined by all of
1272 * the devices in the system
1273 */
1274 fs_info->btree_inode->i_size = OFFSET_MAX;
1275 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1276 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1277
1278 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1279 fs_info->btree_inode->i_mapping,
1280 GFP_NOFS);
1281 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1282 GFP_NOFS);
1283
1284 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1285
1286 extent_io_tree_init(&fs_info->free_space_cache,
1287 fs_info->btree_inode->i_mapping, GFP_NOFS);
1288 extent_io_tree_init(&fs_info->block_group_cache,
1289 fs_info->btree_inode->i_mapping, GFP_NOFS);
1290 extent_io_tree_init(&fs_info->pinned_extents,
1291 fs_info->btree_inode->i_mapping, GFP_NOFS);
1292 extent_io_tree_init(&fs_info->pending_del,
1293 fs_info->btree_inode->i_mapping, GFP_NOFS);
1294 extent_io_tree_init(&fs_info->extent_ins,
1295 fs_info->btree_inode->i_mapping, GFP_NOFS);
1296 fs_info->do_barriers = 1;
1297
1298 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1299 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1300 sizeof(struct btrfs_key));
1301 insert_inode_hash(fs_info->btree_inode);
1302 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1303
1304 mutex_init(&fs_info->trans_mutex);
1305 mutex_init(&fs_info->drop_mutex);
1306 mutex_init(&fs_info->alloc_mutex);
1307 mutex_init(&fs_info->chunk_mutex);
1308 mutex_init(&fs_info->transaction_kthread_mutex);
1309 mutex_init(&fs_info->cleaner_mutex);
1310 mutex_init(&fs_info->volume_mutex);
1311 init_waitqueue_head(&fs_info->transaction_throttle);
1312 init_waitqueue_head(&fs_info->transaction_wait);
1313
1314 #if 0
1315 ret = add_hasher(fs_info, "crc32c");
1316 if (ret) {
1317 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1318 err = -ENOMEM;
1319 goto fail_iput;
1320 }
1321 #endif
1322 __setup_root(4096, 4096, 4096, 4096, tree_root,
1323 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1324
1325
1326 bh = __bread(fs_devices->latest_bdev,
1327 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1328 if (!bh)
1329 goto fail_iput;
1330
1331 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1332 brelse(bh);
1333
1334 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1335
1336 disk_super = &fs_info->super_copy;
1337 if (!btrfs_super_root(disk_super))
1338 goto fail_sb_buffer;
1339
1340 err = btrfs_parse_options(tree_root, options);
1341 if (err)
1342 goto fail_sb_buffer;
1343
1344 /*
1345 * we need to start all the end_io workers up front because the
1346 * queue work function gets called at interrupt time, and so it
1347 * cannot dynamically grow.
1348 */
1349 btrfs_init_workers(&fs_info->workers, fs_info->thread_pool_size);
1350 btrfs_init_workers(&fs_info->submit_workers, fs_info->thread_pool_size);
1351
1352 /* a higher idle thresh on the submit workers makes it much more
1353 * likely that bios will be send down in a sane order to the
1354 * devices
1355 */
1356 fs_info->submit_workers.idle_thresh = 64;
1357
1358 btrfs_init_workers(&fs_info->fixup_workers, 1);
1359 btrfs_init_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1360 btrfs_init_workers(&fs_info->endio_write_workers,
1361 fs_info->thread_pool_size);
1362
1363 /*
1364 * endios are largely parallel and should have a very
1365 * low idle thresh
1366 */
1367 fs_info->endio_workers.idle_thresh = 4;
1368 fs_info->endio_write_workers.idle_thresh = 4;
1369
1370 btrfs_start_workers(&fs_info->workers, 1);
1371 btrfs_start_workers(&fs_info->submit_workers, 1);
1372 btrfs_start_workers(&fs_info->fixup_workers, 1);
1373 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1374 btrfs_start_workers(&fs_info->endio_write_workers,
1375 fs_info->thread_pool_size);
1376
1377 err = -EINVAL;
1378 if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1379 printk("Btrfs: wanted %llu devices, but found %llu\n",
1380 (unsigned long long)btrfs_super_num_devices(disk_super),
1381 (unsigned long long)fs_devices->open_devices);
1382 if (btrfs_test_opt(tree_root, DEGRADED))
1383 printk("continuing in degraded mode\n");
1384 else {
1385 goto fail_sb_buffer;
1386 }
1387 }
1388
1389 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1390
1391 nodesize = btrfs_super_nodesize(disk_super);
1392 leafsize = btrfs_super_leafsize(disk_super);
1393 sectorsize = btrfs_super_sectorsize(disk_super);
1394 stripesize = btrfs_super_stripesize(disk_super);
1395 tree_root->nodesize = nodesize;
1396 tree_root->leafsize = leafsize;
1397 tree_root->sectorsize = sectorsize;
1398 tree_root->stripesize = stripesize;
1399
1400 sb->s_blocksize = sectorsize;
1401 sb->s_blocksize_bits = blksize_bits(sectorsize);
1402
1403 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1404 sizeof(disk_super->magic))) {
1405 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1406 goto fail_sb_buffer;
1407 }
1408
1409 mutex_lock(&fs_info->chunk_mutex);
1410 ret = btrfs_read_sys_array(tree_root);
1411 mutex_unlock(&fs_info->chunk_mutex);
1412 if (ret) {
1413 printk("btrfs: failed to read the system array on %s\n",
1414 sb->s_id);
1415 goto fail_sys_array;
1416 }
1417
1418 blocksize = btrfs_level_size(tree_root,
1419 btrfs_super_chunk_root_level(disk_super));
1420
1421 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1422 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1423
1424 chunk_root->node = read_tree_block(chunk_root,
1425 btrfs_super_chunk_root(disk_super),
1426 blocksize, 0);
1427 BUG_ON(!chunk_root->node);
1428
1429 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1430 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1431 BTRFS_UUID_SIZE);
1432
1433 mutex_lock(&fs_info->chunk_mutex);
1434 ret = btrfs_read_chunk_tree(chunk_root);
1435 mutex_unlock(&fs_info->chunk_mutex);
1436 BUG_ON(ret);
1437
1438 btrfs_close_extra_devices(fs_devices);
1439
1440 blocksize = btrfs_level_size(tree_root,
1441 btrfs_super_root_level(disk_super));
1442
1443
1444 tree_root->node = read_tree_block(tree_root,
1445 btrfs_super_root(disk_super),
1446 blocksize, 0);
1447 if (!tree_root->node)
1448 goto fail_sb_buffer;
1449
1450
1451 ret = find_and_setup_root(tree_root, fs_info,
1452 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1453 if (ret)
1454 goto fail_tree_root;
1455 extent_root->track_dirty = 1;
1456
1457 ret = find_and_setup_root(tree_root, fs_info,
1458 BTRFS_DEV_TREE_OBJECTID, dev_root);
1459 dev_root->track_dirty = 1;
1460
1461 if (ret)
1462 goto fail_extent_root;
1463
1464 btrfs_read_block_groups(extent_root);
1465
1466 fs_info->generation = btrfs_super_generation(disk_super) + 1;
1467 fs_info->data_alloc_profile = (u64)-1;
1468 fs_info->metadata_alloc_profile = (u64)-1;
1469 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1470 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1471 "btrfs-cleaner");
1472 if (!fs_info->cleaner_kthread)
1473 goto fail_extent_root;
1474
1475 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1476 tree_root,
1477 "btrfs-transaction");
1478 if (!fs_info->transaction_kthread)
1479 goto fail_cleaner;
1480
1481
1482 return tree_root;
1483
1484 fail_cleaner:
1485 kthread_stop(fs_info->cleaner_kthread);
1486 fail_extent_root:
1487 free_extent_buffer(extent_root->node);
1488 fail_tree_root:
1489 free_extent_buffer(tree_root->node);
1490 fail_sys_array:
1491 fail_sb_buffer:
1492 btrfs_stop_workers(&fs_info->fixup_workers);
1493 btrfs_stop_workers(&fs_info->workers);
1494 btrfs_stop_workers(&fs_info->endio_workers);
1495 btrfs_stop_workers(&fs_info->endio_write_workers);
1496 btrfs_stop_workers(&fs_info->submit_workers);
1497 fail_iput:
1498 iput(fs_info->btree_inode);
1499 fail:
1500 btrfs_close_devices(fs_info->fs_devices);
1501 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1502
1503 kfree(extent_root);
1504 kfree(tree_root);
1505 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1506 bdi_destroy(&fs_info->bdi);
1507 #endif
1508 kfree(fs_info);
1509 return ERR_PTR(err);
1510 }
1511
1512 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1513 {
1514 char b[BDEVNAME_SIZE];
1515
1516 if (uptodate) {
1517 set_buffer_uptodate(bh);
1518 } else {
1519 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1520 printk(KERN_WARNING "lost page write due to "
1521 "I/O error on %s\n",
1522 bdevname(bh->b_bdev, b));
1523 }
1524 /* note, we dont' set_buffer_write_io_error because we have
1525 * our own ways of dealing with the IO errors
1526 */
1527 clear_buffer_uptodate(bh);
1528 }
1529 unlock_buffer(bh);
1530 put_bh(bh);
1531 }
1532
1533 int write_all_supers(struct btrfs_root *root)
1534 {
1535 struct list_head *cur;
1536 struct list_head *head = &root->fs_info->fs_devices->devices;
1537 struct btrfs_device *dev;
1538 struct btrfs_super_block *sb;
1539 struct btrfs_dev_item *dev_item;
1540 struct buffer_head *bh;
1541 int ret;
1542 int do_barriers;
1543 int max_errors;
1544 int total_errors = 0;
1545 u32 crc;
1546 u64 flags;
1547
1548 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1549 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1550
1551 sb = &root->fs_info->super_for_commit;
1552 dev_item = &sb->dev_item;
1553 list_for_each(cur, head) {
1554 dev = list_entry(cur, struct btrfs_device, dev_list);
1555 if (!dev->bdev) {
1556 total_errors++;
1557 continue;
1558 }
1559 if (!dev->in_fs_metadata)
1560 continue;
1561
1562 btrfs_set_stack_device_type(dev_item, dev->type);
1563 btrfs_set_stack_device_id(dev_item, dev->devid);
1564 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1565 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1566 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1567 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1568 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1569 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1570 flags = btrfs_super_flags(sb);
1571 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1572
1573
1574 crc = ~(u32)0;
1575 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1576 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1577 btrfs_csum_final(crc, sb->csum);
1578
1579 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1580 BTRFS_SUPER_INFO_SIZE);
1581
1582 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1583 dev->pending_io = bh;
1584
1585 get_bh(bh);
1586 set_buffer_uptodate(bh);
1587 lock_buffer(bh);
1588 bh->b_end_io = btrfs_end_buffer_write_sync;
1589
1590 if (do_barriers && dev->barriers) {
1591 ret = submit_bh(WRITE_BARRIER, bh);
1592 if (ret == -EOPNOTSUPP) {
1593 printk("btrfs: disabling barriers on dev %s\n",
1594 dev->name);
1595 set_buffer_uptodate(bh);
1596 dev->barriers = 0;
1597 get_bh(bh);
1598 lock_buffer(bh);
1599 ret = submit_bh(WRITE, bh);
1600 }
1601 } else {
1602 ret = submit_bh(WRITE, bh);
1603 }
1604 if (ret)
1605 total_errors++;
1606 }
1607 if (total_errors > max_errors) {
1608 printk("btrfs: %d errors while writing supers\n", total_errors);
1609 BUG();
1610 }
1611 total_errors = 0;
1612
1613 list_for_each(cur, head) {
1614 dev = list_entry(cur, struct btrfs_device, dev_list);
1615 if (!dev->bdev)
1616 continue;
1617 if (!dev->in_fs_metadata)
1618 continue;
1619
1620 BUG_ON(!dev->pending_io);
1621 bh = dev->pending_io;
1622 wait_on_buffer(bh);
1623 if (!buffer_uptodate(dev->pending_io)) {
1624 if (do_barriers && dev->barriers) {
1625 printk("btrfs: disabling barriers on dev %s\n",
1626 dev->name);
1627 set_buffer_uptodate(bh);
1628 get_bh(bh);
1629 lock_buffer(bh);
1630 dev->barriers = 0;
1631 ret = submit_bh(WRITE, bh);
1632 BUG_ON(ret);
1633 wait_on_buffer(bh);
1634 if (!buffer_uptodate(bh))
1635 total_errors++;
1636 } else {
1637 total_errors++;
1638 }
1639
1640 }
1641 dev->pending_io = NULL;
1642 brelse(bh);
1643 }
1644 if (total_errors > max_errors) {
1645 printk("btrfs: %d errors while writing supers\n", total_errors);
1646 BUG();
1647 }
1648 return 0;
1649 }
1650
1651 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1652 *root)
1653 {
1654 int ret;
1655
1656 ret = write_all_supers(root);
1657 return ret;
1658 }
1659
1660 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1661 {
1662 radix_tree_delete(&fs_info->fs_roots_radix,
1663 (unsigned long)root->root_key.objectid);
1664 if (root->in_sysfs)
1665 btrfs_sysfs_del_root(root);
1666 if (root->inode)
1667 iput(root->inode);
1668 if (root->node)
1669 free_extent_buffer(root->node);
1670 if (root->commit_root)
1671 free_extent_buffer(root->commit_root);
1672 if (root->name)
1673 kfree(root->name);
1674 kfree(root);
1675 return 0;
1676 }
1677
1678 static int del_fs_roots(struct btrfs_fs_info *fs_info)
1679 {
1680 int ret;
1681 struct btrfs_root *gang[8];
1682 int i;
1683
1684 while(1) {
1685 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1686 (void **)gang, 0,
1687 ARRAY_SIZE(gang));
1688 if (!ret)
1689 break;
1690 for (i = 0; i < ret; i++)
1691 btrfs_free_fs_root(fs_info, gang[i]);
1692 }
1693 return 0;
1694 }
1695
1696 int close_ctree(struct btrfs_root *root)
1697 {
1698 int ret;
1699 struct btrfs_trans_handle *trans;
1700 struct btrfs_fs_info *fs_info = root->fs_info;
1701
1702 fs_info->closing = 1;
1703 smp_mb();
1704
1705 kthread_stop(root->fs_info->transaction_kthread);
1706 kthread_stop(root->fs_info->cleaner_kthread);
1707
1708 btrfs_clean_old_snapshots(root);
1709 trans = btrfs_start_transaction(root, 1);
1710 ret = btrfs_commit_transaction(trans, root);
1711 /* run commit again to drop the original snapshot */
1712 trans = btrfs_start_transaction(root, 1);
1713 btrfs_commit_transaction(trans, root);
1714 ret = btrfs_write_and_wait_transaction(NULL, root);
1715 BUG_ON(ret);
1716
1717 write_ctree_super(NULL, root);
1718
1719 if (fs_info->delalloc_bytes) {
1720 printk("btrfs: at unmount delalloc count %Lu\n",
1721 fs_info->delalloc_bytes);
1722 }
1723 if (fs_info->total_ref_cache_size) {
1724 printk("btrfs: at umount reference cache size %Lu\n",
1725 fs_info->total_ref_cache_size);
1726 }
1727
1728 if (fs_info->extent_root->node)
1729 free_extent_buffer(fs_info->extent_root->node);
1730
1731 if (fs_info->tree_root->node)
1732 free_extent_buffer(fs_info->tree_root->node);
1733
1734 if (root->fs_info->chunk_root->node);
1735 free_extent_buffer(root->fs_info->chunk_root->node);
1736
1737 if (root->fs_info->dev_root->node);
1738 free_extent_buffer(root->fs_info->dev_root->node);
1739
1740 btrfs_free_block_groups(root->fs_info);
1741 del_fs_roots(fs_info);
1742
1743 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1744
1745 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1746
1747 btrfs_stop_workers(&fs_info->fixup_workers);
1748 btrfs_stop_workers(&fs_info->workers);
1749 btrfs_stop_workers(&fs_info->endio_workers);
1750 btrfs_stop_workers(&fs_info->endio_write_workers);
1751 btrfs_stop_workers(&fs_info->submit_workers);
1752
1753 iput(fs_info->btree_inode);
1754 #if 0
1755 while(!list_empty(&fs_info->hashers)) {
1756 struct btrfs_hasher *hasher;
1757 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1758 hashers);
1759 list_del(&hasher->hashers);
1760 crypto_free_hash(&fs_info->hash_tfm);
1761 kfree(hasher);
1762 }
1763 #endif
1764 btrfs_close_devices(fs_info->fs_devices);
1765 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1766
1767 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1768 bdi_destroy(&fs_info->bdi);
1769 #endif
1770
1771 kfree(fs_info->extent_root);
1772 kfree(fs_info->tree_root);
1773 kfree(fs_info->chunk_root);
1774 kfree(fs_info->dev_root);
1775 return 0;
1776 }
1777
1778 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1779 {
1780 int ret;
1781 struct inode *btree_inode = buf->first_page->mapping->host;
1782
1783 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1784 if (!ret)
1785 return ret;
1786
1787 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1788 parent_transid);
1789 return !ret;
1790 }
1791
1792 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1793 {
1794 struct inode *btree_inode = buf->first_page->mapping->host;
1795 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1796 buf);
1797 }
1798
1799 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1800 {
1801 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1802 u64 transid = btrfs_header_generation(buf);
1803 struct inode *btree_inode = root->fs_info->btree_inode;
1804
1805 WARN_ON(!btrfs_tree_locked(buf));
1806 if (transid != root->fs_info->generation) {
1807 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1808 (unsigned long long)buf->start,
1809 transid, root->fs_info->generation);
1810 WARN_ON(1);
1811 }
1812 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1813 }
1814
1815 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1816 {
1817 /*
1818 * looks as though older kernels can get into trouble with
1819 * this code, they end up stuck in balance_dirty_pages forever
1820 */
1821 struct extent_io_tree *tree;
1822 u64 num_dirty;
1823 u64 start = 0;
1824 unsigned long thresh = 16 * 1024 * 1024;
1825 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1826
1827 if (current_is_pdflush())
1828 return;
1829
1830 num_dirty = count_range_bits(tree, &start, (u64)-1,
1831 thresh, EXTENT_DIRTY);
1832 if (num_dirty > thresh) {
1833 balance_dirty_pages_ratelimited_nr(
1834 root->fs_info->btree_inode->i_mapping, 1);
1835 }
1836 return;
1837 }
1838
1839 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1840 {
1841 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1842 int ret;
1843 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1844 if (ret == 0) {
1845 buf->flags |= EXTENT_UPTODATE;
1846 }
1847 return ret;
1848 }
1849
1850 static struct extent_io_ops btree_extent_io_ops = {
1851 .writepage_io_hook = btree_writepage_io_hook,
1852 .readpage_end_io_hook = btree_readpage_end_io_hook,
1853 .submit_bio_hook = btree_submit_bio_hook,
1854 /* note we're sharing with inode.c for the merge bio hook */
1855 .merge_bio_hook = btrfs_merge_bio_hook,
1856 };
Linux kernel - Deliverables
This page took 0.069805 seconds and 6 git commands to generate.