projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: Add debugging checks to track down corrupted metadata
[linux-2.6-block.git] / fs / btrfs / disk-io.c
... / ...
CommitLineData
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
46static 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
58static struct extent_io_ops btree_extent_io_ops;
59static void end_workqueue_fn(struct btrfs_work *work);
60
61struct 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
72struct 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
82struct 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);
143out:
144 return em;
145}
146
147u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
148{
149 return btrfs_crc32c(seed, data, len);
150}
151
152void btrfs_csum_final(u32 crc, char *result)
153{
154 *(__le32 *)result = ~cpu_to_le32(crc);
155}
156
157static 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
210static 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);
230out:
231 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
232 GFP_NOFS);
233 return ret;
234
235}
236
237static 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;
253printk("read extent buffer pages failed with ret %d mirror no %d\n", ret, mirror_num);
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
266int 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);
313err:
314 free_extent_buffer(eb);
315out:
316 return 0;
317}
318
319static 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
327int 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 printk("bad tree block start %llu %llu\n",
352 (unsigned long long)found_start,
353 (unsigned long long)eb->start);
354 ret = -EIO;
355 goto err;
356 }
357 if (eb->first_page != page) {
358 printk("bad first page %lu %lu\n", eb->first_page->index,
359 page->index);
360 WARN_ON(1);
361 ret = -EIO;
362 goto err;
363 }
364 if (memcmp_extent_buffer(eb, root->fs_info->fsid,
365 (unsigned long)btrfs_header_fsid(eb),
366 BTRFS_FSID_SIZE)) {
367 printk("bad fsid on block %Lu\n", eb->start);
368 ret = -EIO;
369 goto err;
370 }
371 found_level = btrfs_header_level(eb);
372
373 ret = csum_tree_block(root, eb, 1);
374 if (ret)
375 ret = -EIO;
376
377 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
378 end = eb->start + end - 1;
379err:
380 free_extent_buffer(eb);
381out:
382 return ret;
383}
384
385#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
386static void end_workqueue_bio(struct bio *bio, int err)
387#else
388static int end_workqueue_bio(struct bio *bio,
389 unsigned int bytes_done, int err)
390#endif
391{
392 struct end_io_wq *end_io_wq = bio->bi_private;
393 struct btrfs_fs_info *fs_info;
394
395#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
396 if (bio->bi_size)
397 return 1;
398#endif
399
400 fs_info = end_io_wq->info;
401 end_io_wq->error = err;
402 end_io_wq->work.func = end_workqueue_fn;
403 end_io_wq->work.flags = 0;
404 if (bio->bi_rw & (1 << BIO_RW))
405 btrfs_queue_worker(&fs_info->endio_write_workers,
406 &end_io_wq->work);
407 else
408 btrfs_queue_worker(&fs_info->endio_workers, &end_io_wq->work);
409
410#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
411 return 0;
412#endif
413}
414
415int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
416 int metadata)
417{
418 struct end_io_wq *end_io_wq;
419 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
420 if (!end_io_wq)
421 return -ENOMEM;
422
423 end_io_wq->private = bio->bi_private;
424 end_io_wq->end_io = bio->bi_end_io;
425 end_io_wq->info = info;
426 end_io_wq->error = 0;
427 end_io_wq->bio = bio;
428 end_io_wq->metadata = metadata;
429
430 bio->bi_private = end_io_wq;
431 bio->bi_end_io = end_workqueue_bio;
432 return 0;
433}
434
435unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
436{
437 unsigned long limit = min_t(unsigned long,
438 info->workers.max_workers,
439 info->fs_devices->open_devices);
440 return 256 * limit;
441}
442
443int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
444{
445 return atomic_read(&info->nr_async_bios) >
446 btrfs_async_submit_limit(info);
447}
448
449static void run_one_async_submit(struct btrfs_work *work)
450{
451 struct btrfs_fs_info *fs_info;
452 struct async_submit_bio *async;
453 int limit;
454
455 async = container_of(work, struct async_submit_bio, work);
456 fs_info = BTRFS_I(async->inode)->root->fs_info;
457
458 limit = btrfs_async_submit_limit(fs_info);
459 limit = limit * 2 / 3;
460
461 atomic_dec(&fs_info->nr_async_submits);
462
463 if (atomic_read(&fs_info->nr_async_submits) < limit &&
464 waitqueue_active(&fs_info->async_submit_wait))
465 wake_up(&fs_info->async_submit_wait);
466
467 async->submit_bio_hook(async->inode, async->rw, async->bio,
468 async->mirror_num);
469 kfree(async);
470}
471
472int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
473 int rw, struct bio *bio, int mirror_num,
474 extent_submit_bio_hook_t *submit_bio_hook)
475{
476 struct async_submit_bio *async;
477 int limit = btrfs_async_submit_limit(fs_info);
478
479 async = kmalloc(sizeof(*async), GFP_NOFS);
480 if (!async)
481 return -ENOMEM;
482
483 async->inode = inode;
484 async->rw = rw;
485 async->bio = bio;
486 async->mirror_num = mirror_num;
487 async->submit_bio_hook = submit_bio_hook;
488 async->work.func = run_one_async_submit;
489 async->work.flags = 0;
490 atomic_inc(&fs_info->nr_async_submits);
491 btrfs_queue_worker(&fs_info->workers, &async->work);
492
493 if (atomic_read(&fs_info->nr_async_submits) > limit) {
494 wait_event_timeout(fs_info->async_submit_wait,
495 (atomic_read(&fs_info->nr_async_submits) < limit),
496 HZ/10);
497
498 wait_event_timeout(fs_info->async_submit_wait,
499 (atomic_read(&fs_info->nr_async_bios) < limit),
500 HZ/10);
501 }
502 return 0;
503}
504
505static int __btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
506 int mirror_num)
507{
508 struct btrfs_root *root = BTRFS_I(inode)->root;
509 u64 offset;
510 int ret;
511
512 offset = bio->bi_sector << 9;
513
514 /*
515 * when we're called for a write, we're already in the async
516 * submission context. Just jump into btrfs_map_bio
517 */
518 if (rw & (1 << BIO_RW)) {
519 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
520 mirror_num, 1);
521 }
522
523 /*
524 * called for a read, do the setup so that checksum validation
525 * can happen in the async kernel threads
526 */
527 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 1);
528 BUG_ON(ret);
529
530 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
531}
532
533static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
534 int mirror_num)
535{
536 /*
537 * kthread helpers are used to submit writes so that checksumming
538 * can happen in parallel across all CPUs
539 */
540 if (!(rw & (1 << BIO_RW))) {
541 return __btree_submit_bio_hook(inode, rw, bio, mirror_num);
542 }
543 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
544 inode, rw, bio, mirror_num,
545 __btree_submit_bio_hook);
546}
547
548static int btree_writepage(struct page *page, struct writeback_control *wbc)
549{
550 struct extent_io_tree *tree;
551 tree = &BTRFS_I(page->mapping->host)->io_tree;
552
553 if (current->flags & PF_MEMALLOC) {
554 redirty_page_for_writepage(wbc, page);
555 unlock_page(page);
556 return 0;
557 }
558 return extent_write_full_page(tree, page, btree_get_extent, wbc);
559}
560
561static int btree_writepages(struct address_space *mapping,
562 struct writeback_control *wbc)
563{
564 struct extent_io_tree *tree;
565 tree = &BTRFS_I(mapping->host)->io_tree;
566 if (wbc->sync_mode == WB_SYNC_NONE) {
567 u64 num_dirty;
568 u64 start = 0;
569 unsigned long thresh = 8 * 1024 * 1024;
570
571 if (wbc->for_kupdate)
572 return 0;
573
574 num_dirty = count_range_bits(tree, &start, (u64)-1,
575 thresh, EXTENT_DIRTY);
576 if (num_dirty < thresh) {
577 return 0;
578 }
579 }
580 return extent_writepages(tree, mapping, btree_get_extent, wbc);
581}
582
583int btree_readpage(struct file *file, struct page *page)
584{
585 struct extent_io_tree *tree;
586 tree = &BTRFS_I(page->mapping->host)->io_tree;
587 return extent_read_full_page(tree, page, btree_get_extent);
588}
589
590static int btree_releasepage(struct page *page, gfp_t gfp_flags)
591{
592 struct extent_io_tree *tree;
593 struct extent_map_tree *map;
594 int ret;
595
596 tree = &BTRFS_I(page->mapping->host)->io_tree;
597 map = &BTRFS_I(page->mapping->host)->extent_tree;
598
599 ret = try_release_extent_state(map, tree, page, gfp_flags);
600 if (!ret) {
601 return 0;
602 }
603
604 ret = try_release_extent_buffer(tree, page);
605 if (ret == 1) {
606 ClearPagePrivate(page);
607 set_page_private(page, 0);
608 page_cache_release(page);
609 }
610
611 return ret;
612}
613
614static void btree_invalidatepage(struct page *page, unsigned long offset)
615{
616 struct extent_io_tree *tree;
617 tree = &BTRFS_I(page->mapping->host)->io_tree;
618 extent_invalidatepage(tree, page, offset);
619 btree_releasepage(page, GFP_NOFS);
620 if (PagePrivate(page)) {
621 printk("warning page private not zero on page %Lu\n",
622 page_offset(page));
623 ClearPagePrivate(page);
624 set_page_private(page, 0);
625 page_cache_release(page);
626 }
627}
628
629#if 0
630static int btree_writepage(struct page *page, struct writeback_control *wbc)
631{
632 struct buffer_head *bh;
633 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
634 struct buffer_head *head;
635 if (!page_has_buffers(page)) {
636 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
637 (1 << BH_Dirty)|(1 << BH_Uptodate));
638 }
639 head = page_buffers(page);
640 bh = head;
641 do {
642 if (buffer_dirty(bh))
643 csum_tree_block(root, bh, 0);
644 bh = bh->b_this_page;
645 } while (bh != head);
646 return block_write_full_page(page, btree_get_block, wbc);
647}
648#endif
649
650static struct address_space_operations btree_aops = {
651 .readpage = btree_readpage,
652 .writepage = btree_writepage,
653 .writepages = btree_writepages,
654 .releasepage = btree_releasepage,
655 .invalidatepage = btree_invalidatepage,
656 .sync_page = block_sync_page,
657};
658
659int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
660 u64 parent_transid)
661{
662 struct extent_buffer *buf = NULL;
663 struct inode *btree_inode = root->fs_info->btree_inode;
664 int ret = 0;
665
666 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
667 if (!buf)
668 return 0;
669 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
670 buf, 0, 0, btree_get_extent, 0);
671 free_extent_buffer(buf);
672 return ret;
673}
674
675struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
676 u64 bytenr, u32 blocksize)
677{
678 struct inode *btree_inode = root->fs_info->btree_inode;
679 struct extent_buffer *eb;
680 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
681 bytenr, blocksize, GFP_NOFS);
682 return eb;
683}
684
685struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
686 u64 bytenr, u32 blocksize)
687{
688 struct inode *btree_inode = root->fs_info->btree_inode;
689 struct extent_buffer *eb;
690
691 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
692 bytenr, blocksize, NULL, GFP_NOFS);
693 return eb;
694}
695
696
697struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
698 u32 blocksize, u64 parent_transid)
699{
700 struct extent_buffer *buf = NULL;
701 struct inode *btree_inode = root->fs_info->btree_inode;
702 struct extent_io_tree *io_tree;
703 int ret;
704
705 io_tree = &BTRFS_I(btree_inode)->io_tree;
706
707 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
708 if (!buf)
709 return NULL;
710
711 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
712
713 if (ret == 0) {
714 buf->flags |= EXTENT_UPTODATE;
715 } else {
716 WARN_ON(1);
717 }
718 return buf;
719
720}
721
722int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
723 struct extent_buffer *buf)
724{
725 struct inode *btree_inode = root->fs_info->btree_inode;
726 if (btrfs_header_generation(buf) ==
727 root->fs_info->running_transaction->transid) {
728 WARN_ON(!btrfs_tree_locked(buf));
729 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
730 buf);
731 }
732 return 0;
733}
734
735int wait_on_tree_block_writeback(struct btrfs_root *root,
736 struct extent_buffer *buf)
737{
738 struct inode *btree_inode = root->fs_info->btree_inode;
739 wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->io_tree,
740 buf);
741 return 0;
742}
743
744static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
745 u32 stripesize, struct btrfs_root *root,
746 struct btrfs_fs_info *fs_info,
747 u64 objectid)
748{
749 root->node = NULL;
750 root->inode = NULL;
751 root->commit_root = NULL;
752 root->ref_tree = NULL;
753 root->sectorsize = sectorsize;
754 root->nodesize = nodesize;
755 root->leafsize = leafsize;
756 root->stripesize = stripesize;
757 root->ref_cows = 0;
758 root->track_dirty = 0;
759
760 root->fs_info = fs_info;
761 root->objectid = objectid;
762 root->last_trans = 0;
763 root->highest_inode = 0;
764 root->last_inode_alloc = 0;
765 root->name = NULL;
766 root->in_sysfs = 0;
767
768 INIT_LIST_HEAD(&root->dirty_list);
769 INIT_LIST_HEAD(&root->orphan_list);
770 INIT_LIST_HEAD(&root->dead_list);
771 spin_lock_init(&root->node_lock);
772 spin_lock_init(&root->list_lock);
773 mutex_init(&root->objectid_mutex);
774
775 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
776 root->ref_tree = &root->ref_tree_struct;
777
778 memset(&root->root_key, 0, sizeof(root->root_key));
779 memset(&root->root_item, 0, sizeof(root->root_item));
780 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
781 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
782 root->defrag_trans_start = fs_info->generation;
783 init_completion(&root->kobj_unregister);
784 root->defrag_running = 0;
785 root->defrag_level = 0;
786 root->root_key.objectid = objectid;
787 return 0;
788}
789
790static int find_and_setup_root(struct btrfs_root *tree_root,
791 struct btrfs_fs_info *fs_info,
792 u64 objectid,
793 struct btrfs_root *root)
794{
795 int ret;
796 u32 blocksize;
797
798 __setup_root(tree_root->nodesize, tree_root->leafsize,
799 tree_root->sectorsize, tree_root->stripesize,
800 root, fs_info, objectid);
801 ret = btrfs_find_last_root(tree_root, objectid,
802 &root->root_item, &root->root_key);
803 BUG_ON(ret);
804
805 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
806 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
807 blocksize, 0);
808 BUG_ON(!root->node);
809 return 0;
810}
811
812struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
813 struct btrfs_key *location)
814{
815 struct btrfs_root *root;
816 struct btrfs_root *tree_root = fs_info->tree_root;
817 struct btrfs_path *path;
818 struct extent_buffer *l;
819 u64 highest_inode;
820 u32 blocksize;
821 int ret = 0;
822
823 root = kzalloc(sizeof(*root), GFP_NOFS);
824 if (!root)
825 return ERR_PTR(-ENOMEM);
826 if (location->offset == (u64)-1) {
827 ret = find_and_setup_root(tree_root, fs_info,
828 location->objectid, root);
829 if (ret) {
830 kfree(root);
831 return ERR_PTR(ret);
832 }
833 goto insert;
834 }
835
836 __setup_root(tree_root->nodesize, tree_root->leafsize,
837 tree_root->sectorsize, tree_root->stripesize,
838 root, fs_info, location->objectid);
839
840 path = btrfs_alloc_path();
841 BUG_ON(!path);
842 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
843 if (ret != 0) {
844 if (ret > 0)
845 ret = -ENOENT;
846 goto out;
847 }
848 l = path->nodes[0];
849 read_extent_buffer(l, &root->root_item,
850 btrfs_item_ptr_offset(l, path->slots[0]),
851 sizeof(root->root_item));
852 memcpy(&root->root_key, location, sizeof(*location));
853 ret = 0;
854out:
855 btrfs_release_path(root, path);
856 btrfs_free_path(path);
857 if (ret) {
858 kfree(root);
859 return ERR_PTR(ret);
860 }
861 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
862 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
863 blocksize, 0);
864 BUG_ON(!root->node);
865insert:
866 root->ref_cows = 1;
867 ret = btrfs_find_highest_inode(root, &highest_inode);
868 if (ret == 0) {
869 root->highest_inode = highest_inode;
870 root->last_inode_alloc = highest_inode;
871 }
872 return root;
873}
874
875struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
876 u64 root_objectid)
877{
878 struct btrfs_root *root;
879
880 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
881 return fs_info->tree_root;
882 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
883 return fs_info->extent_root;
884
885 root = radix_tree_lookup(&fs_info->fs_roots_radix,
886 (unsigned long)root_objectid);
887 return root;
888}
889
890struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
891 struct btrfs_key *location)
892{
893 struct btrfs_root *root;
894 int ret;
895
896 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
897 return fs_info->tree_root;
898 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
899 return fs_info->extent_root;
900 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
901 return fs_info->chunk_root;
902 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
903 return fs_info->dev_root;
904
905 root = radix_tree_lookup(&fs_info->fs_roots_radix,
906 (unsigned long)location->objectid);
907 if (root)
908 return root;
909
910 root = btrfs_read_fs_root_no_radix(fs_info, location);
911 if (IS_ERR(root))
912 return root;
913 ret = radix_tree_insert(&fs_info->fs_roots_radix,
914 (unsigned long)root->root_key.objectid,
915 root);
916 if (ret) {
917 free_extent_buffer(root->node);
918 kfree(root);
919 return ERR_PTR(ret);
920 }
921 ret = btrfs_find_dead_roots(fs_info->tree_root,
922 root->root_key.objectid, root);
923 BUG_ON(ret);
924
925 return root;
926}
927
928struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
929 struct btrfs_key *location,
930 const char *name, int namelen)
931{
932 struct btrfs_root *root;
933 int ret;
934
935 root = btrfs_read_fs_root_no_name(fs_info, location);
936 if (!root)
937 return NULL;
938
939 if (root->in_sysfs)
940 return root;
941
942 ret = btrfs_set_root_name(root, name, namelen);
943 if (ret) {
944 free_extent_buffer(root->node);
945 kfree(root);
946 return ERR_PTR(ret);
947 }
948
949 ret = btrfs_sysfs_add_root(root);
950 if (ret) {
951 free_extent_buffer(root->node);
952 kfree(root->name);
953 kfree(root);
954 return ERR_PTR(ret);
955 }
956 root->in_sysfs = 1;
957 return root;
958}
959#if 0
960static int add_hasher(struct btrfs_fs_info *info, char *type) {
961 struct btrfs_hasher *hasher;
962
963 hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
964 if (!hasher)
965 return -ENOMEM;
966 hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
967 if (!hasher->hash_tfm) {
968 kfree(hasher);
969 return -EINVAL;
970 }
971 spin_lock(&info->hash_lock);
972 list_add(&hasher->list, &info->hashers);
973 spin_unlock(&info->hash_lock);
974 return 0;
975}
976#endif
977
978static int btrfs_congested_fn(void *congested_data, int bdi_bits)
979{
980 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
981 int ret = 0;
982 struct list_head *cur;
983 struct btrfs_device *device;
984 struct backing_dev_info *bdi;
985
986 if ((bdi_bits & (1 << BDI_write_congested)) &&
987 btrfs_congested_async(info, 0))
988 return 1;
989
990 list_for_each(cur, &info->fs_devices->devices) {
991 device = list_entry(cur, struct btrfs_device, dev_list);
992 if (!device->bdev)
993 continue;
994 bdi = blk_get_backing_dev_info(device->bdev);
995 if (bdi && bdi_congested(bdi, bdi_bits)) {
996 ret = 1;
997 break;
998 }
999 }
1000 return ret;
1001}
1002
1003/*
1004 * this unplugs every device on the box, and it is only used when page
1005 * is null
1006 */
1007static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1008{
1009 struct list_head *cur;
1010 struct btrfs_device *device;
1011 struct btrfs_fs_info *info;
1012
1013 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1014 list_for_each(cur, &info->fs_devices->devices) {
1015 device = list_entry(cur, struct btrfs_device, dev_list);
1016 bdi = blk_get_backing_dev_info(device->bdev);
1017 if (bdi->unplug_io_fn) {
1018 bdi->unplug_io_fn(bdi, page);
1019 }
1020 }
1021}
1022
1023void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1024{
1025 struct inode *inode;
1026 struct extent_map_tree *em_tree;
1027 struct extent_map *em;
1028 struct address_space *mapping;
1029 u64 offset;
1030
1031 /* the generic O_DIRECT read code does this */
1032 if (!page) {
1033 __unplug_io_fn(bdi, page);
1034 return;
1035 }
1036
1037 /*
1038 * page->mapping may change at any time. Get a consistent copy
1039 * and use that for everything below
1040 */
1041 smp_mb();
1042 mapping = page->mapping;
1043 if (!mapping)
1044 return;
1045
1046 inode = mapping->host;
1047 offset = page_offset(page);
1048
1049 em_tree = &BTRFS_I(inode)->extent_tree;
1050 spin_lock(&em_tree->lock);
1051 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1052 spin_unlock(&em_tree->lock);
1053 if (!em) {
1054 __unplug_io_fn(bdi, page);
1055 return;
1056 }
1057
1058 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1059 free_extent_map(em);
1060 __unplug_io_fn(bdi, page);
1061 return;
1062 }
1063 offset = offset - em->start;
1064 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1065 em->block_start + offset, page);
1066 free_extent_map(em);
1067}
1068
1069static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1070{
1071#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1072 bdi_init(bdi);
1073#endif
1074 bdi->ra_pages = default_backing_dev_info.ra_pages;
1075 bdi->state = 0;
1076 bdi->capabilities = default_backing_dev_info.capabilities;
1077 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1078 bdi->unplug_io_data = info;
1079 bdi->congested_fn = btrfs_congested_fn;
1080 bdi->congested_data = info;
1081 return 0;
1082}
1083
1084static int bio_ready_for_csum(struct bio *bio)
1085{
1086 u64 length = 0;
1087 u64 buf_len = 0;
1088 u64 start = 0;
1089 struct page *page;
1090 struct extent_io_tree *io_tree = NULL;
1091 struct btrfs_fs_info *info = NULL;
1092 struct bio_vec *bvec;
1093 int i;
1094 int ret;
1095
1096 bio_for_each_segment(bvec, bio, i) {
1097 page = bvec->bv_page;
1098 if (page->private == EXTENT_PAGE_PRIVATE) {
1099 length += bvec->bv_len;
1100 continue;
1101 }
1102 if (!page->private) {
1103 length += bvec->bv_len;
1104 continue;
1105 }
1106 length = bvec->bv_len;
1107 buf_len = page->private >> 2;
1108 start = page_offset(page) + bvec->bv_offset;
1109 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1110 info = BTRFS_I(page->mapping->host)->root->fs_info;
1111 }
1112 /* are we fully contained in this bio? */
1113 if (buf_len <= length)
1114 return 1;
1115
1116 ret = extent_range_uptodate(io_tree, start + length,
1117 start + buf_len - 1);
1118 if (ret == 1)
1119 return ret;
1120 return ret;
1121}
1122
1123/*
1124 * called by the kthread helper functions to finally call the bio end_io
1125 * functions. This is where read checksum verification actually happens
1126 */
1127static void end_workqueue_fn(struct btrfs_work *work)
1128{
1129 struct bio *bio;
1130 struct end_io_wq *end_io_wq;
1131 struct btrfs_fs_info *fs_info;
1132 int error;
1133
1134 end_io_wq = container_of(work, struct end_io_wq, work);
1135 bio = end_io_wq->bio;
1136 fs_info = end_io_wq->info;
1137
1138 /* metadata bios are special because the whole tree block must
1139 * be checksummed at once. This makes sure the entire block is in
1140 * ram and up to date before trying to verify things. For
1141 * blocksize <= pagesize, it is basically a noop
1142 */
1143 if (end_io_wq->metadata && !bio_ready_for_csum(bio)) {
1144 btrfs_queue_worker(&fs_info->endio_workers,
1145 &end_io_wq->work);
1146 return;
1147 }
1148 error = end_io_wq->error;
1149 bio->bi_private = end_io_wq->private;
1150 bio->bi_end_io = end_io_wq->end_io;
1151 kfree(end_io_wq);
1152#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23)
1153 bio_endio(bio, bio->bi_size, error);
1154#else
1155 bio_endio(bio, error);
1156#endif
1157}
1158
1159static int cleaner_kthread(void *arg)
1160{
1161 struct btrfs_root *root = arg;
1162
1163 do {
1164 smp_mb();
1165 if (root->fs_info->closing)
1166 break;
1167
1168 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1169 mutex_lock(&root->fs_info->cleaner_mutex);
1170 btrfs_clean_old_snapshots(root);
1171 mutex_unlock(&root->fs_info->cleaner_mutex);
1172
1173 if (freezing(current)) {
1174 refrigerator();
1175 } else {
1176 smp_mb();
1177 if (root->fs_info->closing)
1178 break;
1179 set_current_state(TASK_INTERRUPTIBLE);
1180 schedule();
1181 __set_current_state(TASK_RUNNING);
1182 }
1183 } while (!kthread_should_stop());
1184 return 0;
1185}
1186
1187static int transaction_kthread(void *arg)
1188{
1189 struct btrfs_root *root = arg;
1190 struct btrfs_trans_handle *trans;
1191 struct btrfs_transaction *cur;
1192 unsigned long now;
1193 unsigned long delay;
1194 int ret;
1195
1196 do {
1197 smp_mb();
1198 if (root->fs_info->closing)
1199 break;
1200
1201 delay = HZ * 30;
1202 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1203 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1204
1205 if (root->fs_info->total_ref_cache_size > 20 * 1024 * 1024) {
1206 printk("btrfs: total reference cache size %Lu\n",
1207 root->fs_info->total_ref_cache_size);
1208 }
1209
1210 mutex_lock(&root->fs_info->trans_mutex);
1211 cur = root->fs_info->running_transaction;
1212 if (!cur) {
1213 mutex_unlock(&root->fs_info->trans_mutex);
1214 goto sleep;
1215 }
1216
1217 now = get_seconds();
1218 if (now < cur->start_time || now - cur->start_time < 30) {
1219 mutex_unlock(&root->fs_info->trans_mutex);
1220 delay = HZ * 5;
1221 goto sleep;
1222 }
1223 mutex_unlock(&root->fs_info->trans_mutex);
1224 trans = btrfs_start_transaction(root, 1);
1225 ret = btrfs_commit_transaction(trans, root);
1226sleep:
1227 wake_up_process(root->fs_info->cleaner_kthread);
1228 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1229
1230 if (freezing(current)) {
1231 refrigerator();
1232 } else {
1233 if (root->fs_info->closing)
1234 break;
1235 set_current_state(TASK_INTERRUPTIBLE);
1236 schedule_timeout(delay);
1237 __set_current_state(TASK_RUNNING);
1238 }
1239 } while (!kthread_should_stop());
1240 return 0;
1241}
1242
1243struct btrfs_root *open_ctree(struct super_block *sb,
1244 struct btrfs_fs_devices *fs_devices,
1245 char *options)
1246{
1247 u32 sectorsize;
1248 u32 nodesize;
1249 u32 leafsize;
1250 u32 blocksize;
1251 u32 stripesize;
1252 struct buffer_head *bh;
1253 struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
1254 GFP_NOFS);
1255 struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
1256 GFP_NOFS);
1257 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1258 GFP_NOFS);
1259 struct btrfs_root *chunk_root = kmalloc(sizeof(struct btrfs_root),
1260 GFP_NOFS);
1261 struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
1262 GFP_NOFS);
1263 int ret;
1264 int err = -EINVAL;
1265
1266 struct btrfs_super_block *disk_super;
1267
1268 if (!extent_root || !tree_root || !fs_info) {
1269 err = -ENOMEM;
1270 goto fail;
1271 }
1272 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1273 INIT_LIST_HEAD(&fs_info->trans_list);
1274 INIT_LIST_HEAD(&fs_info->dead_roots);
1275 INIT_LIST_HEAD(&fs_info->hashers);
1276 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1277 spin_lock_init(&fs_info->hash_lock);
1278 spin_lock_init(&fs_info->delalloc_lock);
1279 spin_lock_init(&fs_info->new_trans_lock);
1280 spin_lock_init(&fs_info->ref_cache_lock);
1281
1282 init_completion(&fs_info->kobj_unregister);
1283 fs_info->tree_root = tree_root;
1284 fs_info->extent_root = extent_root;
1285 fs_info->chunk_root = chunk_root;
1286 fs_info->dev_root = dev_root;
1287 fs_info->fs_devices = fs_devices;
1288 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1289 INIT_LIST_HEAD(&fs_info->space_info);
1290 btrfs_mapping_init(&fs_info->mapping_tree);
1291 atomic_set(&fs_info->nr_async_submits, 0);
1292 atomic_set(&fs_info->nr_async_bios, 0);
1293 atomic_set(&fs_info->throttles, 0);
1294 atomic_set(&fs_info->throttle_gen, 0);
1295 fs_info->sb = sb;
1296 fs_info->max_extent = (u64)-1;
1297 fs_info->max_inline = 8192 * 1024;
1298 setup_bdi(fs_info, &fs_info->bdi);
1299 fs_info->btree_inode = new_inode(sb);
1300 fs_info->btree_inode->i_ino = 1;
1301 fs_info->btree_inode->i_nlink = 1;
1302 fs_info->thread_pool_size = min(num_online_cpus() + 2, 8);
1303
1304 INIT_LIST_HEAD(&fs_info->ordered_extents);
1305 spin_lock_init(&fs_info->ordered_extent_lock);
1306
1307 sb->s_blocksize = 4096;
1308 sb->s_blocksize_bits = blksize_bits(4096);
1309
1310 /*
1311 * we set the i_size on the btree inode to the max possible int.
1312 * the real end of the address space is determined by all of
1313 * the devices in the system
1314 */
1315 fs_info->btree_inode->i_size = OFFSET_MAX;
1316 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1317 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1318
1319 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1320 fs_info->btree_inode->i_mapping,
1321 GFP_NOFS);
1322 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1323 GFP_NOFS);
1324
1325 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1326
1327 extent_io_tree_init(&fs_info->free_space_cache,
1328 fs_info->btree_inode->i_mapping, GFP_NOFS);
1329 extent_io_tree_init(&fs_info->block_group_cache,
1330 fs_info->btree_inode->i_mapping, GFP_NOFS);
1331 extent_io_tree_init(&fs_info->pinned_extents,
1332 fs_info->btree_inode->i_mapping, GFP_NOFS);
1333 extent_io_tree_init(&fs_info->pending_del,
1334 fs_info->btree_inode->i_mapping, GFP_NOFS);
1335 extent_io_tree_init(&fs_info->extent_ins,
1336 fs_info->btree_inode->i_mapping, GFP_NOFS);
1337 fs_info->do_barriers = 1;
1338
1339 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1340 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1341 sizeof(struct btrfs_key));
1342 insert_inode_hash(fs_info->btree_inode);
1343 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1344
1345 mutex_init(&fs_info->trans_mutex);
1346 mutex_init(&fs_info->drop_mutex);
1347 mutex_init(&fs_info->alloc_mutex);
1348 mutex_init(&fs_info->chunk_mutex);
1349 mutex_init(&fs_info->transaction_kthread_mutex);
1350 mutex_init(&fs_info->cleaner_mutex);
1351 mutex_init(&fs_info->volume_mutex);
1352 init_waitqueue_head(&fs_info->transaction_throttle);
1353 init_waitqueue_head(&fs_info->transaction_wait);
1354 init_waitqueue_head(&fs_info->async_submit_wait);
1355
1356#if 0
1357 ret = add_hasher(fs_info, "crc32c");
1358 if (ret) {
1359 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
1360 err = -ENOMEM;
1361 goto fail_iput;
1362 }
1363#endif
1364 __setup_root(4096, 4096, 4096, 4096, tree_root,
1365 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1366
1367
1368 bh = __bread(fs_devices->latest_bdev,
1369 BTRFS_SUPER_INFO_OFFSET / 4096, 4096);
1370 if (!bh)
1371 goto fail_iput;
1372
1373 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1374 brelse(bh);
1375
1376 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1377
1378 disk_super = &fs_info->super_copy;
1379 if (!btrfs_super_root(disk_super))
1380 goto fail_sb_buffer;
1381
1382 err = btrfs_parse_options(tree_root, options);
1383 if (err)
1384 goto fail_sb_buffer;
1385
1386 /*
1387 * we need to start all the end_io workers up front because the
1388 * queue work function gets called at interrupt time, and so it
1389 * cannot dynamically grow.
1390 */
1391 btrfs_init_workers(&fs_info->workers, "worker",
1392 fs_info->thread_pool_size);
1393 btrfs_init_workers(&fs_info->submit_workers, "submit",
1394 min_t(u64, fs_devices->num_devices,
1395 fs_info->thread_pool_size));
1396
1397 /* a higher idle thresh on the submit workers makes it much more
1398 * likely that bios will be send down in a sane order to the
1399 * devices
1400 */
1401 fs_info->submit_workers.idle_thresh = 64;
1402
1403 /* fs_info->workers is responsible for checksumming file data
1404 * blocks and metadata. Using a larger idle thresh allows each
1405 * worker thread to operate on things in roughly the order they
1406 * were sent by the writeback daemons, improving overall locality
1407 * of the IO going down the pipe.
1408 */
1409 fs_info->workers.idle_thresh = 128;
1410
1411 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1412 btrfs_init_workers(&fs_info->endio_workers, "endio",
1413 fs_info->thread_pool_size);
1414 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1415 fs_info->thread_pool_size);
1416
1417 /*
1418 * endios are largely parallel and should have a very
1419 * low idle thresh
1420 */
1421 fs_info->endio_workers.idle_thresh = 4;
1422 fs_info->endio_write_workers.idle_thresh = 4;
1423
1424 btrfs_start_workers(&fs_info->workers, 1);
1425 btrfs_start_workers(&fs_info->submit_workers, 1);
1426 btrfs_start_workers(&fs_info->fixup_workers, 1);
1427 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1428 btrfs_start_workers(&fs_info->endio_write_workers,
1429 fs_info->thread_pool_size);
1430
1431 err = -EINVAL;
1432 if (btrfs_super_num_devices(disk_super) > fs_devices->open_devices) {
1433 printk("Btrfs: wanted %llu devices, but found %llu\n",
1434 (unsigned long long)btrfs_super_num_devices(disk_super),
1435 (unsigned long long)fs_devices->open_devices);
1436 if (btrfs_test_opt(tree_root, DEGRADED))
1437 printk("continuing in degraded mode\n");
1438 else {
1439 goto fail_sb_buffer;
1440 }
1441 }
1442
1443 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1444
1445 nodesize = btrfs_super_nodesize(disk_super);
1446 leafsize = btrfs_super_leafsize(disk_super);
1447 sectorsize = btrfs_super_sectorsize(disk_super);
1448 stripesize = btrfs_super_stripesize(disk_super);
1449 tree_root->nodesize = nodesize;
1450 tree_root->leafsize = leafsize;
1451 tree_root->sectorsize = sectorsize;
1452 tree_root->stripesize = stripesize;
1453
1454 sb->s_blocksize = sectorsize;
1455 sb->s_blocksize_bits = blksize_bits(sectorsize);
1456
1457 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1458 sizeof(disk_super->magic))) {
1459 printk("btrfs: valid FS not found on %s\n", sb->s_id);
1460 goto fail_sb_buffer;
1461 }
1462
1463 mutex_lock(&fs_info->chunk_mutex);
1464 ret = btrfs_read_sys_array(tree_root);
1465 mutex_unlock(&fs_info->chunk_mutex);
1466 if (ret) {
1467 printk("btrfs: failed to read the system array on %s\n",
1468 sb->s_id);
1469 goto fail_sys_array;
1470 }
1471
1472 blocksize = btrfs_level_size(tree_root,
1473 btrfs_super_chunk_root_level(disk_super));
1474
1475 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1476 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1477
1478 chunk_root->node = read_tree_block(chunk_root,
1479 btrfs_super_chunk_root(disk_super),
1480 blocksize, 0);
1481 BUG_ON(!chunk_root->node);
1482
1483 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1484 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1485 BTRFS_UUID_SIZE);
1486
1487 mutex_lock(&fs_info->chunk_mutex);
1488 ret = btrfs_read_chunk_tree(chunk_root);
1489 mutex_unlock(&fs_info->chunk_mutex);
1490 BUG_ON(ret);
1491
1492 btrfs_close_extra_devices(fs_devices);
1493
1494 blocksize = btrfs_level_size(tree_root,
1495 btrfs_super_root_level(disk_super));
1496
1497
1498 tree_root->node = read_tree_block(tree_root,
1499 btrfs_super_root(disk_super),
1500 blocksize, 0);
1501 if (!tree_root->node)
1502 goto fail_sb_buffer;
1503
1504
1505 ret = find_and_setup_root(tree_root, fs_info,
1506 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1507 if (ret)
1508 goto fail_tree_root;
1509 extent_root->track_dirty = 1;
1510
1511 ret = find_and_setup_root(tree_root, fs_info,
1512 BTRFS_DEV_TREE_OBJECTID, dev_root);
1513 dev_root->track_dirty = 1;
1514
1515 if (ret)
1516 goto fail_extent_root;
1517
1518 btrfs_read_block_groups(extent_root);
1519
1520 fs_info->generation = btrfs_super_generation(disk_super) + 1;
1521 fs_info->data_alloc_profile = (u64)-1;
1522 fs_info->metadata_alloc_profile = (u64)-1;
1523 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1524 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1525 "btrfs-cleaner");
1526 if (!fs_info->cleaner_kthread)
1527 goto fail_extent_root;
1528
1529 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1530 tree_root,
1531 "btrfs-transaction");
1532 if (!fs_info->transaction_kthread)
1533 goto fail_cleaner;
1534
1535
1536 return tree_root;
1537
1538fail_cleaner:
1539 kthread_stop(fs_info->cleaner_kthread);
1540fail_extent_root:
1541 free_extent_buffer(extent_root->node);
1542fail_tree_root:
1543 free_extent_buffer(tree_root->node);
1544fail_sys_array:
1545fail_sb_buffer:
1546 btrfs_stop_workers(&fs_info->fixup_workers);
1547 btrfs_stop_workers(&fs_info->workers);
1548 btrfs_stop_workers(&fs_info->endio_workers);
1549 btrfs_stop_workers(&fs_info->endio_write_workers);
1550 btrfs_stop_workers(&fs_info->submit_workers);
1551fail_iput:
1552 iput(fs_info->btree_inode);
1553fail:
1554 btrfs_close_devices(fs_info->fs_devices);
1555 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1556
1557 kfree(extent_root);
1558 kfree(tree_root);
1559#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1560 bdi_destroy(&fs_info->bdi);
1561#endif
1562 kfree(fs_info);
1563 return ERR_PTR(err);
1564}
1565
1566static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1567{
1568 char b[BDEVNAME_SIZE];
1569
1570 if (uptodate) {
1571 set_buffer_uptodate(bh);
1572 } else {
1573 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1574 printk(KERN_WARNING "lost page write due to "
1575 "I/O error on %s\n",
1576 bdevname(bh->b_bdev, b));
1577 }
1578 /* note, we dont' set_buffer_write_io_error because we have
1579 * our own ways of dealing with the IO errors
1580 */
1581 clear_buffer_uptodate(bh);
1582 }
1583 unlock_buffer(bh);
1584 put_bh(bh);
1585}
1586
1587int write_all_supers(struct btrfs_root *root)
1588{
1589 struct list_head *cur;
1590 struct list_head *head = &root->fs_info->fs_devices->devices;
1591 struct btrfs_device *dev;
1592 struct btrfs_super_block *sb;
1593 struct btrfs_dev_item *dev_item;
1594 struct buffer_head *bh;
1595 int ret;
1596 int do_barriers;
1597 int max_errors;
1598 int total_errors = 0;
1599 u32 crc;
1600 u64 flags;
1601
1602 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
1603 do_barriers = !btrfs_test_opt(root, NOBARRIER);
1604
1605 sb = &root->fs_info->super_for_commit;
1606 dev_item = &sb->dev_item;
1607 list_for_each(cur, head) {
1608 dev = list_entry(cur, struct btrfs_device, dev_list);
1609 if (!dev->bdev) {
1610 total_errors++;
1611 continue;
1612 }
1613 if (!dev->in_fs_metadata)
1614 continue;
1615
1616 btrfs_set_stack_device_type(dev_item, dev->type);
1617 btrfs_set_stack_device_id(dev_item, dev->devid);
1618 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
1619 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
1620 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
1621 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
1622 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
1623 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
1624 flags = btrfs_super_flags(sb);
1625 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
1626
1627
1628 crc = ~(u32)0;
1629 crc = btrfs_csum_data(root, (char *)sb + BTRFS_CSUM_SIZE, crc,
1630 BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
1631 btrfs_csum_final(crc, sb->csum);
1632
1633 bh = __getblk(dev->bdev, BTRFS_SUPER_INFO_OFFSET / 4096,
1634 BTRFS_SUPER_INFO_SIZE);
1635
1636 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
1637 dev->pending_io = bh;
1638
1639 get_bh(bh);
1640 set_buffer_uptodate(bh);
1641 lock_buffer(bh);
1642 bh->b_end_io = btrfs_end_buffer_write_sync;
1643
1644 if (do_barriers && dev->barriers) {
1645 ret = submit_bh(WRITE_BARRIER, bh);
1646 if (ret == -EOPNOTSUPP) {
1647 printk("btrfs: disabling barriers on dev %s\n",
1648 dev->name);
1649 set_buffer_uptodate(bh);
1650 dev->barriers = 0;
1651 get_bh(bh);
1652 lock_buffer(bh);
1653 ret = submit_bh(WRITE, bh);
1654 }
1655 } else {
1656 ret = submit_bh(WRITE, bh);
1657 }
1658 if (ret)
1659 total_errors++;
1660 }
1661 if (total_errors > max_errors) {
1662 printk("btrfs: %d errors while writing supers\n", total_errors);
1663 BUG();
1664 }
1665 total_errors = 0;
1666
1667 list_for_each(cur, head) {
1668 dev = list_entry(cur, struct btrfs_device, dev_list);
1669 if (!dev->bdev)
1670 continue;
1671 if (!dev->in_fs_metadata)
1672 continue;
1673
1674 BUG_ON(!dev->pending_io);
1675 bh = dev->pending_io;
1676 wait_on_buffer(bh);
1677 if (!buffer_uptodate(dev->pending_io)) {
1678 if (do_barriers && dev->barriers) {
1679 printk("btrfs: disabling barriers on dev %s\n",
1680 dev->name);
1681 set_buffer_uptodate(bh);
1682 get_bh(bh);
1683 lock_buffer(bh);
1684 dev->barriers = 0;
1685 ret = submit_bh(WRITE, bh);
1686 BUG_ON(ret);
1687 wait_on_buffer(bh);
1688 if (!buffer_uptodate(bh))
1689 total_errors++;
1690 } else {
1691 total_errors++;
1692 }
1693
1694 }
1695 dev->pending_io = NULL;
1696 brelse(bh);
1697 }
1698 if (total_errors > max_errors) {
1699 printk("btrfs: %d errors while writing supers\n", total_errors);
1700 BUG();
1701 }
1702 return 0;
1703}
1704
1705int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
1706 *root)
1707{
1708 int ret;
1709
1710 ret = write_all_supers(root);
1711 return ret;
1712}
1713
1714int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
1715{
1716 radix_tree_delete(&fs_info->fs_roots_radix,
1717 (unsigned long)root->root_key.objectid);
1718 if (root->in_sysfs)
1719 btrfs_sysfs_del_root(root);
1720 if (root->inode)
1721 iput(root->inode);
1722 if (root->node)
1723 free_extent_buffer(root->node);
1724 if (root->commit_root)
1725 free_extent_buffer(root->commit_root);
1726 if (root->name)
1727 kfree(root->name);
1728 kfree(root);
1729 return 0;
1730}
1731
1732static int del_fs_roots(struct btrfs_fs_info *fs_info)
1733{
1734 int ret;
1735 struct btrfs_root *gang[8];
1736 int i;
1737
1738 while(1) {
1739 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
1740 (void **)gang, 0,
1741 ARRAY_SIZE(gang));
1742 if (!ret)
1743 break;
1744 for (i = 0; i < ret; i++)
1745 btrfs_free_fs_root(fs_info, gang[i]);
1746 }
1747 return 0;
1748}
1749
1750int close_ctree(struct btrfs_root *root)
1751{
1752 int ret;
1753 struct btrfs_trans_handle *trans;
1754 struct btrfs_fs_info *fs_info = root->fs_info;
1755
1756 fs_info->closing = 1;
1757 smp_mb();
1758
1759 kthread_stop(root->fs_info->transaction_kthread);
1760 kthread_stop(root->fs_info->cleaner_kthread);
1761
1762 btrfs_clean_old_snapshots(root);
1763 trans = btrfs_start_transaction(root, 1);
1764 ret = btrfs_commit_transaction(trans, root);
1765 /* run commit again to drop the original snapshot */
1766 trans = btrfs_start_transaction(root, 1);
1767 btrfs_commit_transaction(trans, root);
1768 ret = btrfs_write_and_wait_transaction(NULL, root);
1769 BUG_ON(ret);
1770
1771 write_ctree_super(NULL, root);
1772
1773 if (fs_info->delalloc_bytes) {
1774 printk("btrfs: at unmount delalloc count %Lu\n",
1775 fs_info->delalloc_bytes);
1776 }
1777 if (fs_info->total_ref_cache_size) {
1778 printk("btrfs: at umount reference cache size %Lu\n",
1779 fs_info->total_ref_cache_size);
1780 }
1781
1782 if (fs_info->extent_root->node)
1783 free_extent_buffer(fs_info->extent_root->node);
1784
1785 if (fs_info->tree_root->node)
1786 free_extent_buffer(fs_info->tree_root->node);
1787
1788 if (root->fs_info->chunk_root->node);
1789 free_extent_buffer(root->fs_info->chunk_root->node);
1790
1791 if (root->fs_info->dev_root->node);
1792 free_extent_buffer(root->fs_info->dev_root->node);
1793
1794 btrfs_free_block_groups(root->fs_info);
1795 fs_info->closing = 2;
1796 del_fs_roots(fs_info);
1797
1798 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1799
1800 truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
1801
1802 btrfs_stop_workers(&fs_info->fixup_workers);
1803 btrfs_stop_workers(&fs_info->workers);
1804 btrfs_stop_workers(&fs_info->endio_workers);
1805 btrfs_stop_workers(&fs_info->endio_write_workers);
1806 btrfs_stop_workers(&fs_info->submit_workers);
1807
1808 iput(fs_info->btree_inode);
1809#if 0
1810 while(!list_empty(&fs_info->hashers)) {
1811 struct btrfs_hasher *hasher;
1812 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
1813 hashers);
1814 list_del(&hasher->hashers);
1815 crypto_free_hash(&fs_info->hash_tfm);
1816 kfree(hasher);
1817 }
1818#endif
1819 btrfs_close_devices(fs_info->fs_devices);
1820 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1821
1822#if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
1823 bdi_destroy(&fs_info->bdi);
1824#endif
1825
1826 kfree(fs_info->extent_root);
1827 kfree(fs_info->tree_root);
1828 kfree(fs_info->chunk_root);
1829 kfree(fs_info->dev_root);
1830 return 0;
1831}
1832
1833int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
1834{
1835 int ret;
1836 struct inode *btree_inode = buf->first_page->mapping->host;
1837
1838 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
1839 if (!ret)
1840 return ret;
1841
1842 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
1843 parent_transid);
1844 return !ret;
1845}
1846
1847int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
1848{
1849 struct inode *btree_inode = buf->first_page->mapping->host;
1850 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
1851 buf);
1852}
1853
1854void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
1855{
1856 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1857 u64 transid = btrfs_header_generation(buf);
1858 struct inode *btree_inode = root->fs_info->btree_inode;
1859
1860 WARN_ON(!btrfs_tree_locked(buf));
1861 if (transid != root->fs_info->generation) {
1862 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
1863 (unsigned long long)buf->start,
1864 transid, root->fs_info->generation);
1865 WARN_ON(1);
1866 }
1867 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
1868}
1869
1870void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
1871{
1872 /*
1873 * looks as though older kernels can get into trouble with
1874 * this code, they end up stuck in balance_dirty_pages forever
1875 */
1876 struct extent_io_tree *tree;
1877 u64 num_dirty;
1878 u64 start = 0;
1879 unsigned long thresh = 96 * 1024 * 1024;
1880 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
1881
1882 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
1883 return;
1884
1885 num_dirty = count_range_bits(tree, &start, (u64)-1,
1886 thresh, EXTENT_DIRTY);
1887 if (num_dirty > thresh) {
1888 balance_dirty_pages_ratelimited_nr(
1889 root->fs_info->btree_inode->i_mapping, 1);
1890 }
1891 return;
1892}
1893
1894int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
1895{
1896 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
1897 int ret;
1898 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1899 if (ret == 0) {
1900 buf->flags |= EXTENT_UPTODATE;
1901 }
1902 return ret;
1903}
1904
1905static struct extent_io_ops btree_extent_io_ops = {
1906 .writepage_io_hook = btree_writepage_io_hook,
1907 .readpage_end_io_hook = btree_readpage_end_io_hook,
1908 .submit_bio_hook = btree_submit_bio_hook,
1909 /* note we're sharing with inode.c for the merge bio hook */
1910 .merge_bio_hook = btrfs_merge_bio_hook,
1911};
Linux 6.x block layer and io_uring tree(s)