projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: Only let very young transactions grow during commit
[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/fs.h>
20#include <linux/blkdev.h>
21#include <linux/scatterlist.h>
22#include <linux/swap.h>
23#include <linux/radix-tree.h>
24#include <linux/writeback.h>
25#include <linux/buffer_head.h>
26#include <linux/workqueue.h>
27#include <linux/kthread.h>
28#include <linux/freezer.h>
29#include "compat.h"
30#include "crc32c.h"
31#include "ctree.h"
32#include "disk-io.h"
33#include "transaction.h"
34#include "btrfs_inode.h"
35#include "volumes.h"
36#include "print-tree.h"
37#include "async-thread.h"
38#include "locking.h"
39#include "ref-cache.h"
40#include "tree-log.h"
41
42static struct extent_io_ops btree_extent_io_ops;
43static void end_workqueue_fn(struct btrfs_work *work);
44
45/*
46 * end_io_wq structs are used to do processing in task context when an IO is
47 * complete. This is used during reads to verify checksums, and it is used
48 * by writes to insert metadata for new file extents after IO is complete.
49 */
50struct end_io_wq {
51 struct bio *bio;
52 bio_end_io_t *end_io;
53 void *private;
54 struct btrfs_fs_info *info;
55 int error;
56 int metadata;
57 struct list_head list;
58 struct btrfs_work work;
59};
60
61/*
62 * async submit bios are used to offload expensive checksumming
63 * onto the worker threads. They checksum file and metadata bios
64 * just before they are sent down the IO stack.
65 */
66struct async_submit_bio {
67 struct inode *inode;
68 struct bio *bio;
69 struct list_head list;
70 extent_submit_bio_hook_t *submit_bio_start;
71 extent_submit_bio_hook_t *submit_bio_done;
72 int rw;
73 int mirror_num;
74 unsigned long bio_flags;
75 struct btrfs_work work;
76};
77
78/* These are used to set the lockdep class on the extent buffer locks.
79 * The class is set by the readpage_end_io_hook after the buffer has
80 * passed csum validation but before the pages are unlocked.
81 *
82 * The lockdep class is also set by btrfs_init_new_buffer on freshly
83 * allocated blocks.
84 *
85 * The class is based on the level in the tree block, which allows lockdep
86 * to know that lower nodes nest inside the locks of higher nodes.
87 *
88 * We also add a check to make sure the highest level of the tree is
89 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
90 * code needs update as well.
91 */
92#ifdef CONFIG_DEBUG_LOCK_ALLOC
93# if BTRFS_MAX_LEVEL != 8
94# error
95# endif
96static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
97static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
98 /* leaf */
99 "btrfs-extent-00",
100 "btrfs-extent-01",
101 "btrfs-extent-02",
102 "btrfs-extent-03",
103 "btrfs-extent-04",
104 "btrfs-extent-05",
105 "btrfs-extent-06",
106 "btrfs-extent-07",
107 /* highest possible level */
108 "btrfs-extent-08",
109};
110#endif
111
112/*
113 * extents on the btree inode are pretty simple, there's one extent
114 * that covers the entire device
115 */
116static struct extent_map *btree_get_extent(struct inode *inode,
117 struct page *page, size_t page_offset, u64 start, u64 len,
118 int create)
119{
120 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
121 struct extent_map *em;
122 int ret;
123
124 spin_lock(&em_tree->lock);
125 em = lookup_extent_mapping(em_tree, start, len);
126 if (em) {
127 em->bdev =
128 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
129 spin_unlock(&em_tree->lock);
130 goto out;
131 }
132 spin_unlock(&em_tree->lock);
133
134 em = alloc_extent_map(GFP_NOFS);
135 if (!em) {
136 em = ERR_PTR(-ENOMEM);
137 goto out;
138 }
139 em->start = 0;
140 em->len = (u64)-1;
141 em->block_len = (u64)-1;
142 em->block_start = 0;
143 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
144
145 spin_lock(&em_tree->lock);
146 ret = add_extent_mapping(em_tree, em);
147 if (ret == -EEXIST) {
148 u64 failed_start = em->start;
149 u64 failed_len = em->len;
150
151 free_extent_map(em);
152 em = lookup_extent_mapping(em_tree, start, len);
153 if (em) {
154 ret = 0;
155 } else {
156 em = lookup_extent_mapping(em_tree, failed_start,
157 failed_len);
158 ret = -EIO;
159 }
160 } else if (ret) {
161 free_extent_map(em);
162 em = NULL;
163 }
164 spin_unlock(&em_tree->lock);
165
166 if (ret)
167 em = ERR_PTR(ret);
168out:
169 return em;
170}
171
172u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
173{
174 return btrfs_crc32c(seed, data, len);
175}
176
177void btrfs_csum_final(u32 crc, char *result)
178{
179 *(__le32 *)result = ~cpu_to_le32(crc);
180}
181
182/*
183 * compute the csum for a btree block, and either verify it or write it
184 * into the csum field of the block.
185 */
186static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
187 int verify)
188{
189 u16 csum_size =
190 btrfs_super_csum_size(&root->fs_info->super_copy);
191 char *result = NULL;
192 unsigned long len;
193 unsigned long cur_len;
194 unsigned long offset = BTRFS_CSUM_SIZE;
195 char *map_token = NULL;
196 char *kaddr;
197 unsigned long map_start;
198 unsigned long map_len;
199 int err;
200 u32 crc = ~(u32)0;
201 unsigned long inline_result;
202
203 len = buf->len - offset;
204 while (len > 0) {
205 err = map_private_extent_buffer(buf, offset, 32,
206 &map_token, &kaddr,
207 &map_start, &map_len, KM_USER0);
208 if (err)
209 return 1;
210 cur_len = min(len, map_len - (offset - map_start));
211 crc = btrfs_csum_data(root, kaddr + offset - map_start,
212 crc, cur_len);
213 len -= cur_len;
214 offset += cur_len;
215 unmap_extent_buffer(buf, map_token, KM_USER0);
216 }
217 if (csum_size > sizeof(inline_result)) {
218 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
219 if (!result)
220 return 1;
221 } else {
222 result = (char *)&inline_result;
223 }
224
225 btrfs_csum_final(crc, result);
226
227 if (verify) {
228 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
229 u32 val;
230 u32 found = 0;
231 memcpy(&found, result, csum_size);
232
233 read_extent_buffer(buf, &val, 0, csum_size);
234 printk(KERN_INFO "btrfs: %s checksum verify failed "
235 "on %llu wanted %X found %X level %d\n",
236 root->fs_info->sb->s_id,
237 buf->start, val, found, btrfs_header_level(buf));
238 if (result != (char *)&inline_result)
239 kfree(result);
240 return 1;
241 }
242 } else {
243 write_extent_buffer(buf, result, 0, csum_size);
244 }
245 if (result != (char *)&inline_result)
246 kfree(result);
247 return 0;
248}
249
250/*
251 * we can't consider a given block up to date unless the transid of the
252 * block matches the transid in the parent node's pointer. This is how we
253 * detect blocks that either didn't get written at all or got written
254 * in the wrong place.
255 */
256static int verify_parent_transid(struct extent_io_tree *io_tree,
257 struct extent_buffer *eb, u64 parent_transid)
258{
259 int ret;
260
261 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
262 return 0;
263
264 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
265 if (extent_buffer_uptodate(io_tree, eb) &&
266 btrfs_header_generation(eb) == parent_transid) {
267 ret = 0;
268 goto out;
269 }
270 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
271 (unsigned long long)eb->start,
272 (unsigned long long)parent_transid,
273 (unsigned long long)btrfs_header_generation(eb));
274 ret = 1;
275 clear_extent_buffer_uptodate(io_tree, eb);
276out:
277 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
278 GFP_NOFS);
279 return ret;
280}
281
282/*
283 * helper to read a given tree block, doing retries as required when
284 * the checksums don't match and we have alternate mirrors to try.
285 */
286static int btree_read_extent_buffer_pages(struct btrfs_root *root,
287 struct extent_buffer *eb,
288 u64 start, u64 parent_transid)
289{
290 struct extent_io_tree *io_tree;
291 int ret;
292 int num_copies = 0;
293 int mirror_num = 0;
294
295 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
296 while (1) {
297 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
298 btree_get_extent, mirror_num);
299 if (!ret &&
300 !verify_parent_transid(io_tree, eb, parent_transid))
301 return ret;
302
303 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
304 eb->start, eb->len);
305 if (num_copies == 1)
306 return ret;
307
308 mirror_num++;
309 if (mirror_num > num_copies)
310 return ret;
311 }
312 return -EIO;
313}
314
315/*
316 * checksum a dirty tree block before IO. This has extra checks to make sure
317 * we only fill in the checksum field in the first page of a multi-page block
318 */
319
320static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
321{
322 struct extent_io_tree *tree;
323 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
324 u64 found_start;
325 int found_level;
326 unsigned long len;
327 struct extent_buffer *eb;
328 int ret;
329
330 tree = &BTRFS_I(page->mapping->host)->io_tree;
331
332 if (page->private == EXTENT_PAGE_PRIVATE)
333 goto out;
334 if (!page->private)
335 goto out;
336 len = page->private >> 2;
337 WARN_ON(len == 0);
338
339 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
340 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
341 btrfs_header_generation(eb));
342 BUG_ON(ret);
343 found_start = btrfs_header_bytenr(eb);
344 if (found_start != start) {
345 WARN_ON(1);
346 goto err;
347 }
348 if (eb->first_page != page) {
349 WARN_ON(1);
350 goto err;
351 }
352 if (!PageUptodate(page)) {
353 WARN_ON(1);
354 goto err;
355 }
356 found_level = btrfs_header_level(eb);
357
358 csum_tree_block(root, eb, 0);
359err:
360 free_extent_buffer(eb);
361out:
362 return 0;
363}
364
365static int check_tree_block_fsid(struct btrfs_root *root,
366 struct extent_buffer *eb)
367{
368 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
369 u8 fsid[BTRFS_UUID_SIZE];
370 int ret = 1;
371
372 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
373 BTRFS_FSID_SIZE);
374 while (fs_devices) {
375 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
376 ret = 0;
377 break;
378 }
379 fs_devices = fs_devices->seed;
380 }
381 return ret;
382}
383
384#ifdef CONFIG_DEBUG_LOCK_ALLOC
385void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
386{
387 lockdep_set_class_and_name(&eb->lock,
388 &btrfs_eb_class[level],
389 btrfs_eb_name[level]);
390}
391#endif
392
393static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
394 struct extent_state *state)
395{
396 struct extent_io_tree *tree;
397 u64 found_start;
398 int found_level;
399 unsigned long len;
400 struct extent_buffer *eb;
401 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
402 int ret = 0;
403
404 tree = &BTRFS_I(page->mapping->host)->io_tree;
405 if (page->private == EXTENT_PAGE_PRIVATE)
406 goto out;
407 if (!page->private)
408 goto out;
409
410 len = page->private >> 2;
411 WARN_ON(len == 0);
412
413 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
414
415 found_start = btrfs_header_bytenr(eb);
416 if (found_start != start) {
417 printk(KERN_INFO "btrfs bad tree block start %llu %llu\n",
418 (unsigned long long)found_start,
419 (unsigned long long)eb->start);
420 ret = -EIO;
421 goto err;
422 }
423 if (eb->first_page != page) {
424 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
425 eb->first_page->index, page->index);
426 WARN_ON(1);
427 ret = -EIO;
428 goto err;
429 }
430 if (check_tree_block_fsid(root, eb)) {
431 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
432 (unsigned long long)eb->start);
433 ret = -EIO;
434 goto err;
435 }
436 found_level = btrfs_header_level(eb);
437
438 btrfs_set_buffer_lockdep_class(eb, found_level);
439
440 ret = csum_tree_block(root, eb, 1);
441 if (ret)
442 ret = -EIO;
443
444 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
445 end = eb->start + end - 1;
446err:
447 free_extent_buffer(eb);
448out:
449 return ret;
450}
451
452static void end_workqueue_bio(struct bio *bio, int err)
453{
454 struct end_io_wq *end_io_wq = bio->bi_private;
455 struct btrfs_fs_info *fs_info;
456
457 fs_info = end_io_wq->info;
458 end_io_wq->error = err;
459 end_io_wq->work.func = end_workqueue_fn;
460 end_io_wq->work.flags = 0;
461
462 if (bio->bi_rw & (1 << BIO_RW)) {
463 if (end_io_wq->metadata)
464 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
465 &end_io_wq->work);
466 else
467 btrfs_queue_worker(&fs_info->endio_write_workers,
468 &end_io_wq->work);
469 } else {
470 if (end_io_wq->metadata)
471 btrfs_queue_worker(&fs_info->endio_meta_workers,
472 &end_io_wq->work);
473 else
474 btrfs_queue_worker(&fs_info->endio_workers,
475 &end_io_wq->work);
476 }
477}
478
479int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
480 int metadata)
481{
482 struct end_io_wq *end_io_wq;
483 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
484 if (!end_io_wq)
485 return -ENOMEM;
486
487 end_io_wq->private = bio->bi_private;
488 end_io_wq->end_io = bio->bi_end_io;
489 end_io_wq->info = info;
490 end_io_wq->error = 0;
491 end_io_wq->bio = bio;
492 end_io_wq->metadata = metadata;
493
494 bio->bi_private = end_io_wq;
495 bio->bi_end_io = end_workqueue_bio;
496 return 0;
497}
498
499unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
500{
501 unsigned long limit = min_t(unsigned long,
502 info->workers.max_workers,
503 info->fs_devices->open_devices);
504 return 256 * limit;
505}
506
507int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
508{
509 return atomic_read(&info->nr_async_bios) >
510 btrfs_async_submit_limit(info);
511}
512
513static void run_one_async_start(struct btrfs_work *work)
514{
515 struct btrfs_fs_info *fs_info;
516 struct async_submit_bio *async;
517
518 async = container_of(work, struct async_submit_bio, work);
519 fs_info = BTRFS_I(async->inode)->root->fs_info;
520 async->submit_bio_start(async->inode, async->rw, async->bio,
521 async->mirror_num, async->bio_flags);
522}
523
524static void run_one_async_done(struct btrfs_work *work)
525{
526 struct btrfs_fs_info *fs_info;
527 struct async_submit_bio *async;
528 int limit;
529
530 async = container_of(work, struct async_submit_bio, work);
531 fs_info = BTRFS_I(async->inode)->root->fs_info;
532
533 limit = btrfs_async_submit_limit(fs_info);
534 limit = limit * 2 / 3;
535
536 atomic_dec(&fs_info->nr_async_submits);
537
538 if (atomic_read(&fs_info->nr_async_submits) < limit &&
539 waitqueue_active(&fs_info->async_submit_wait))
540 wake_up(&fs_info->async_submit_wait);
541
542 async->submit_bio_done(async->inode, async->rw, async->bio,
543 async->mirror_num, async->bio_flags);
544}
545
546static void run_one_async_free(struct btrfs_work *work)
547{
548 struct async_submit_bio *async;
549
550 async = container_of(work, struct async_submit_bio, work);
551 kfree(async);
552}
553
554int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
555 int rw, struct bio *bio, int mirror_num,
556 unsigned long bio_flags,
557 extent_submit_bio_hook_t *submit_bio_start,
558 extent_submit_bio_hook_t *submit_bio_done)
559{
560 struct async_submit_bio *async;
561
562 async = kmalloc(sizeof(*async), GFP_NOFS);
563 if (!async)
564 return -ENOMEM;
565
566 async->inode = inode;
567 async->rw = rw;
568 async->bio = bio;
569 async->mirror_num = mirror_num;
570 async->submit_bio_start = submit_bio_start;
571 async->submit_bio_done = submit_bio_done;
572
573 async->work.func = run_one_async_start;
574 async->work.ordered_func = run_one_async_done;
575 async->work.ordered_free = run_one_async_free;
576
577 async->work.flags = 0;
578 async->bio_flags = bio_flags;
579
580 atomic_inc(&fs_info->nr_async_submits);
581 btrfs_queue_worker(&fs_info->workers, &async->work);
582#if 0
583 int limit = btrfs_async_submit_limit(fs_info);
584 if (atomic_read(&fs_info->nr_async_submits) > limit) {
585 wait_event_timeout(fs_info->async_submit_wait,
586 (atomic_read(&fs_info->nr_async_submits) < limit),
587 HZ/10);
588
589 wait_event_timeout(fs_info->async_submit_wait,
590 (atomic_read(&fs_info->nr_async_bios) < limit),
591 HZ/10);
592 }
593#endif
594 while (atomic_read(&fs_info->async_submit_draining) &&
595 atomic_read(&fs_info->nr_async_submits)) {
596 wait_event(fs_info->async_submit_wait,
597 (atomic_read(&fs_info->nr_async_submits) == 0));
598 }
599
600 return 0;
601}
602
603static int btree_csum_one_bio(struct bio *bio)
604{
605 struct bio_vec *bvec = bio->bi_io_vec;
606 int bio_index = 0;
607 struct btrfs_root *root;
608
609 WARN_ON(bio->bi_vcnt <= 0);
610 while (bio_index < bio->bi_vcnt) {
611 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
612 csum_dirty_buffer(root, bvec->bv_page);
613 bio_index++;
614 bvec++;
615 }
616 return 0;
617}
618
619static int __btree_submit_bio_start(struct inode *inode, int rw,
620 struct bio *bio, int mirror_num,
621 unsigned long bio_flags)
622{
623 /*
624 * when we're called for a write, we're already in the async
625 * submission context. Just jump into btrfs_map_bio
626 */
627 btree_csum_one_bio(bio);
628 return 0;
629}
630
631static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
632 int mirror_num, unsigned long bio_flags)
633{
634 /*
635 * when we're called for a write, we're already in the async
636 * submission context. Just jump into btrfs_map_bio
637 */
638 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
639}
640
641static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
642 int mirror_num, unsigned long bio_flags)
643{
644 int ret;
645
646 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
647 bio, 1);
648 BUG_ON(ret);
649
650 if (!(rw & (1 << BIO_RW))) {
651 /*
652 * called for a read, do the setup so that checksum validation
653 * can happen in the async kernel threads
654 */
655 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
656 mirror_num, 0);
657 }
658 /*
659 * kthread helpers are used to submit writes so that checksumming
660 * can happen in parallel across all CPUs
661 */
662 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
663 inode, rw, bio, mirror_num, 0,
664 __btree_submit_bio_start,
665 __btree_submit_bio_done);
666}
667
668static int btree_writepage(struct page *page, struct writeback_control *wbc)
669{
670 struct extent_io_tree *tree;
671 tree = &BTRFS_I(page->mapping->host)->io_tree;
672
673 if (current->flags & PF_MEMALLOC) {
674 redirty_page_for_writepage(wbc, page);
675 unlock_page(page);
676 return 0;
677 }
678 return extent_write_full_page(tree, page, btree_get_extent, wbc);
679}
680
681static int btree_writepages(struct address_space *mapping,
682 struct writeback_control *wbc)
683{
684 struct extent_io_tree *tree;
685 tree = &BTRFS_I(mapping->host)->io_tree;
686 if (wbc->sync_mode == WB_SYNC_NONE) {
687 u64 num_dirty;
688 u64 start = 0;
689 unsigned long thresh = 32 * 1024 * 1024;
690
691 if (wbc->for_kupdate)
692 return 0;
693
694 num_dirty = count_range_bits(tree, &start, (u64)-1,
695 thresh, EXTENT_DIRTY);
696 if (num_dirty < thresh)
697 return 0;
698 }
699 return extent_writepages(tree, mapping, btree_get_extent, wbc);
700}
701
702static int btree_readpage(struct file *file, struct page *page)
703{
704 struct extent_io_tree *tree;
705 tree = &BTRFS_I(page->mapping->host)->io_tree;
706 return extent_read_full_page(tree, page, btree_get_extent);
707}
708
709static int btree_releasepage(struct page *page, gfp_t gfp_flags)
710{
711 struct extent_io_tree *tree;
712 struct extent_map_tree *map;
713 int ret;
714
715 if (PageWriteback(page) || PageDirty(page))
716 return 0;
717
718 tree = &BTRFS_I(page->mapping->host)->io_tree;
719 map = &BTRFS_I(page->mapping->host)->extent_tree;
720
721 ret = try_release_extent_state(map, tree, page, gfp_flags);
722 if (!ret)
723 return 0;
724
725 ret = try_release_extent_buffer(tree, page);
726 if (ret == 1) {
727 ClearPagePrivate(page);
728 set_page_private(page, 0);
729 page_cache_release(page);
730 }
731
732 return ret;
733}
734
735static void btree_invalidatepage(struct page *page, unsigned long offset)
736{
737 struct extent_io_tree *tree;
738 tree = &BTRFS_I(page->mapping->host)->io_tree;
739 extent_invalidatepage(tree, page, offset);
740 btree_releasepage(page, GFP_NOFS);
741 if (PagePrivate(page)) {
742 printk(KERN_WARNING "btrfs warning page private not zero "
743 "on page %llu\n", (unsigned long long)page_offset(page));
744 ClearPagePrivate(page);
745 set_page_private(page, 0);
746 page_cache_release(page);
747 }
748}
749
750#if 0
751static int btree_writepage(struct page *page, struct writeback_control *wbc)
752{
753 struct buffer_head *bh;
754 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
755 struct buffer_head *head;
756 if (!page_has_buffers(page)) {
757 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
758 (1 << BH_Dirty)|(1 << BH_Uptodate));
759 }
760 head = page_buffers(page);
761 bh = head;
762 do {
763 if (buffer_dirty(bh))
764 csum_tree_block(root, bh, 0);
765 bh = bh->b_this_page;
766 } while (bh != head);
767 return block_write_full_page(page, btree_get_block, wbc);
768}
769#endif
770
771static struct address_space_operations btree_aops = {
772 .readpage = btree_readpage,
773 .writepage = btree_writepage,
774 .writepages = btree_writepages,
775 .releasepage = btree_releasepage,
776 .invalidatepage = btree_invalidatepage,
777 .sync_page = block_sync_page,
778};
779
780int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
781 u64 parent_transid)
782{
783 struct extent_buffer *buf = NULL;
784 struct inode *btree_inode = root->fs_info->btree_inode;
785 int ret = 0;
786
787 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
788 if (!buf)
789 return 0;
790 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
791 buf, 0, 0, btree_get_extent, 0);
792 free_extent_buffer(buf);
793 return ret;
794}
795
796struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
797 u64 bytenr, u32 blocksize)
798{
799 struct inode *btree_inode = root->fs_info->btree_inode;
800 struct extent_buffer *eb;
801 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
802 bytenr, blocksize, GFP_NOFS);
803 return eb;
804}
805
806struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
807 u64 bytenr, u32 blocksize)
808{
809 struct inode *btree_inode = root->fs_info->btree_inode;
810 struct extent_buffer *eb;
811
812 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
813 bytenr, blocksize, NULL, GFP_NOFS);
814 return eb;
815}
816
817
818int btrfs_write_tree_block(struct extent_buffer *buf)
819{
820 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
821 buf->start + buf->len - 1, WB_SYNC_ALL);
822}
823
824int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
825{
826 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
827 buf->start, buf->start + buf->len - 1);
828}
829
830struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
831 u32 blocksize, u64 parent_transid)
832{
833 struct extent_buffer *buf = NULL;
834 struct inode *btree_inode = root->fs_info->btree_inode;
835 struct extent_io_tree *io_tree;
836 int ret;
837
838 io_tree = &BTRFS_I(btree_inode)->io_tree;
839
840 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
841 if (!buf)
842 return NULL;
843
844 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
845
846 if (ret == 0)
847 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
848 else
849 WARN_ON(1);
850 return buf;
851
852}
853
854int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
855 struct extent_buffer *buf)
856{
857 struct inode *btree_inode = root->fs_info->btree_inode;
858 if (btrfs_header_generation(buf) ==
859 root->fs_info->running_transaction->transid) {
860 btrfs_assert_tree_locked(buf);
861
862 /* ugh, clear_extent_buffer_dirty can be expensive */
863 btrfs_set_lock_blocking(buf);
864
865 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
866 buf);
867 }
868 return 0;
869}
870
871static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
872 u32 stripesize, struct btrfs_root *root,
873 struct btrfs_fs_info *fs_info,
874 u64 objectid)
875{
876 root->node = NULL;
877 root->commit_root = NULL;
878 root->ref_tree = NULL;
879 root->sectorsize = sectorsize;
880 root->nodesize = nodesize;
881 root->leafsize = leafsize;
882 root->stripesize = stripesize;
883 root->ref_cows = 0;
884 root->track_dirty = 0;
885
886 root->fs_info = fs_info;
887 root->objectid = objectid;
888 root->last_trans = 0;
889 root->highest_inode = 0;
890 root->last_inode_alloc = 0;
891 root->name = NULL;
892 root->in_sysfs = 0;
893
894 INIT_LIST_HEAD(&root->dirty_list);
895 INIT_LIST_HEAD(&root->orphan_list);
896 INIT_LIST_HEAD(&root->dead_list);
897 spin_lock_init(&root->node_lock);
898 spin_lock_init(&root->list_lock);
899 mutex_init(&root->objectid_mutex);
900 mutex_init(&root->log_mutex);
901 init_waitqueue_head(&root->log_writer_wait);
902 init_waitqueue_head(&root->log_commit_wait[0]);
903 init_waitqueue_head(&root->log_commit_wait[1]);
904 atomic_set(&root->log_commit[0], 0);
905 atomic_set(&root->log_commit[1], 0);
906 atomic_set(&root->log_writers, 0);
907 root->log_batch = 0;
908 root->log_transid = 0;
909 extent_io_tree_init(&root->dirty_log_pages,
910 fs_info->btree_inode->i_mapping, GFP_NOFS);
911
912 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
913 root->ref_tree = &root->ref_tree_struct;
914
915 memset(&root->root_key, 0, sizeof(root->root_key));
916 memset(&root->root_item, 0, sizeof(root->root_item));
917 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
918 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
919 root->defrag_trans_start = fs_info->generation;
920 init_completion(&root->kobj_unregister);
921 root->defrag_running = 0;
922 root->defrag_level = 0;
923 root->root_key.objectid = objectid;
924 root->anon_super.s_root = NULL;
925 root->anon_super.s_dev = 0;
926 INIT_LIST_HEAD(&root->anon_super.s_list);
927 INIT_LIST_HEAD(&root->anon_super.s_instances);
928 init_rwsem(&root->anon_super.s_umount);
929
930 return 0;
931}
932
933static int find_and_setup_root(struct btrfs_root *tree_root,
934 struct btrfs_fs_info *fs_info,
935 u64 objectid,
936 struct btrfs_root *root)
937{
938 int ret;
939 u32 blocksize;
940 u64 generation;
941
942 __setup_root(tree_root->nodesize, tree_root->leafsize,
943 tree_root->sectorsize, tree_root->stripesize,
944 root, fs_info, objectid);
945 ret = btrfs_find_last_root(tree_root, objectid,
946 &root->root_item, &root->root_key);
947 BUG_ON(ret);
948
949 generation = btrfs_root_generation(&root->root_item);
950 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
951 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
952 blocksize, generation);
953 BUG_ON(!root->node);
954 return 0;
955}
956
957int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
958 struct btrfs_fs_info *fs_info)
959{
960 struct extent_buffer *eb;
961 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
962 u64 start = 0;
963 u64 end = 0;
964 int ret;
965
966 if (!log_root_tree)
967 return 0;
968
969 while (1) {
970 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
971 0, &start, &end, EXTENT_DIRTY);
972 if (ret)
973 break;
974
975 clear_extent_dirty(&log_root_tree->dirty_log_pages,
976 start, end, GFP_NOFS);
977 }
978 eb = fs_info->log_root_tree->node;
979
980 WARN_ON(btrfs_header_level(eb) != 0);
981 WARN_ON(btrfs_header_nritems(eb) != 0);
982
983 ret = btrfs_free_reserved_extent(fs_info->tree_root,
984 eb->start, eb->len);
985 BUG_ON(ret);
986
987 free_extent_buffer(eb);
988 kfree(fs_info->log_root_tree);
989 fs_info->log_root_tree = NULL;
990 return 0;
991}
992
993static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
994 struct btrfs_fs_info *fs_info)
995{
996 struct btrfs_root *root;
997 struct btrfs_root *tree_root = fs_info->tree_root;
998 struct extent_buffer *leaf;
999
1000 root = kzalloc(sizeof(*root), GFP_NOFS);
1001 if (!root)
1002 return ERR_PTR(-ENOMEM);
1003
1004 __setup_root(tree_root->nodesize, tree_root->leafsize,
1005 tree_root->sectorsize, tree_root->stripesize,
1006 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1007
1008 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1009 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1010 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1011 /*
1012 * log trees do not get reference counted because they go away
1013 * before a real commit is actually done. They do store pointers
1014 * to file data extents, and those reference counts still get
1015 * updated (along with back refs to the log tree).
1016 */
1017 root->ref_cows = 0;
1018
1019 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1020 0, BTRFS_TREE_LOG_OBJECTID,
1021 trans->transid, 0, 0, 0);
1022 if (IS_ERR(leaf)) {
1023 kfree(root);
1024 return ERR_CAST(leaf);
1025 }
1026
1027 root->node = leaf;
1028 btrfs_set_header_nritems(root->node, 0);
1029 btrfs_set_header_level(root->node, 0);
1030 btrfs_set_header_bytenr(root->node, root->node->start);
1031 btrfs_set_header_generation(root->node, trans->transid);
1032 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
1033
1034 write_extent_buffer(root->node, root->fs_info->fsid,
1035 (unsigned long)btrfs_header_fsid(root->node),
1036 BTRFS_FSID_SIZE);
1037 btrfs_mark_buffer_dirty(root->node);
1038 btrfs_tree_unlock(root->node);
1039 return root;
1040}
1041
1042int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1043 struct btrfs_fs_info *fs_info)
1044{
1045 struct btrfs_root *log_root;
1046
1047 log_root = alloc_log_tree(trans, fs_info);
1048 if (IS_ERR(log_root))
1049 return PTR_ERR(log_root);
1050 WARN_ON(fs_info->log_root_tree);
1051 fs_info->log_root_tree = log_root;
1052 return 0;
1053}
1054
1055int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1056 struct btrfs_root *root)
1057{
1058 struct btrfs_root *log_root;
1059 struct btrfs_inode_item *inode_item;
1060
1061 log_root = alloc_log_tree(trans, root->fs_info);
1062 if (IS_ERR(log_root))
1063 return PTR_ERR(log_root);
1064
1065 log_root->last_trans = trans->transid;
1066 log_root->root_key.offset = root->root_key.objectid;
1067
1068 inode_item = &log_root->root_item.inode;
1069 inode_item->generation = cpu_to_le64(1);
1070 inode_item->size = cpu_to_le64(3);
1071 inode_item->nlink = cpu_to_le32(1);
1072 inode_item->nbytes = cpu_to_le64(root->leafsize);
1073 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1074
1075 btrfs_set_root_bytenr(&log_root->root_item, log_root->node->start);
1076 btrfs_set_root_generation(&log_root->root_item, trans->transid);
1077
1078 WARN_ON(root->log_root);
1079 root->log_root = log_root;
1080 root->log_transid = 0;
1081 return 0;
1082}
1083
1084struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1085 struct btrfs_key *location)
1086{
1087 struct btrfs_root *root;
1088 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1089 struct btrfs_path *path;
1090 struct extent_buffer *l;
1091 u64 highest_inode;
1092 u64 generation;
1093 u32 blocksize;
1094 int ret = 0;
1095
1096 root = kzalloc(sizeof(*root), GFP_NOFS);
1097 if (!root)
1098 return ERR_PTR(-ENOMEM);
1099 if (location->offset == (u64)-1) {
1100 ret = find_and_setup_root(tree_root, fs_info,
1101 location->objectid, root);
1102 if (ret) {
1103 kfree(root);
1104 return ERR_PTR(ret);
1105 }
1106 goto insert;
1107 }
1108
1109 __setup_root(tree_root->nodesize, tree_root->leafsize,
1110 tree_root->sectorsize, tree_root->stripesize,
1111 root, fs_info, location->objectid);
1112
1113 path = btrfs_alloc_path();
1114 BUG_ON(!path);
1115 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1116 if (ret != 0) {
1117 if (ret > 0)
1118 ret = -ENOENT;
1119 goto out;
1120 }
1121 l = path->nodes[0];
1122 read_extent_buffer(l, &root->root_item,
1123 btrfs_item_ptr_offset(l, path->slots[0]),
1124 sizeof(root->root_item));
1125 memcpy(&root->root_key, location, sizeof(*location));
1126 ret = 0;
1127out:
1128 btrfs_release_path(root, path);
1129 btrfs_free_path(path);
1130 if (ret) {
1131 kfree(root);
1132 return ERR_PTR(ret);
1133 }
1134 generation = btrfs_root_generation(&root->root_item);
1135 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1136 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1137 blocksize, generation);
1138 BUG_ON(!root->node);
1139insert:
1140 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1141 root->ref_cows = 1;
1142 ret = btrfs_find_highest_inode(root, &highest_inode);
1143 if (ret == 0) {
1144 root->highest_inode = highest_inode;
1145 root->last_inode_alloc = highest_inode;
1146 }
1147 }
1148 return root;
1149}
1150
1151struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1152 u64 root_objectid)
1153{
1154 struct btrfs_root *root;
1155
1156 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1157 return fs_info->tree_root;
1158 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1159 return fs_info->extent_root;
1160
1161 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1162 (unsigned long)root_objectid);
1163 return root;
1164}
1165
1166struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1167 struct btrfs_key *location)
1168{
1169 struct btrfs_root *root;
1170 int ret;
1171
1172 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1173 return fs_info->tree_root;
1174 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1175 return fs_info->extent_root;
1176 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1177 return fs_info->chunk_root;
1178 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1179 return fs_info->dev_root;
1180 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1181 return fs_info->csum_root;
1182
1183 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1184 (unsigned long)location->objectid);
1185 if (root)
1186 return root;
1187
1188 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1189 if (IS_ERR(root))
1190 return root;
1191
1192 set_anon_super(&root->anon_super, NULL);
1193
1194 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1195 (unsigned long)root->root_key.objectid,
1196 root);
1197 if (ret) {
1198 free_extent_buffer(root->node);
1199 kfree(root);
1200 return ERR_PTR(ret);
1201 }
1202 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1203 ret = btrfs_find_dead_roots(fs_info->tree_root,
1204 root->root_key.objectid, root);
1205 BUG_ON(ret);
1206 btrfs_orphan_cleanup(root);
1207 }
1208 return root;
1209}
1210
1211struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1212 struct btrfs_key *location,
1213 const char *name, int namelen)
1214{
1215 struct btrfs_root *root;
1216 int ret;
1217
1218 root = btrfs_read_fs_root_no_name(fs_info, location);
1219 if (!root)
1220 return NULL;
1221
1222 if (root->in_sysfs)
1223 return root;
1224
1225 ret = btrfs_set_root_name(root, name, namelen);
1226 if (ret) {
1227 free_extent_buffer(root->node);
1228 kfree(root);
1229 return ERR_PTR(ret);
1230 }
1231#if 0
1232 ret = btrfs_sysfs_add_root(root);
1233 if (ret) {
1234 free_extent_buffer(root->node);
1235 kfree(root->name);
1236 kfree(root);
1237 return ERR_PTR(ret);
1238 }
1239#endif
1240 root->in_sysfs = 1;
1241 return root;
1242}
1243
1244static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1245{
1246 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1247 int ret = 0;
1248 struct btrfs_device *device;
1249 struct backing_dev_info *bdi;
1250#if 0
1251 if ((bdi_bits & (1 << BDI_write_congested)) &&
1252 btrfs_congested_async(info, 0))
1253 return 1;
1254#endif
1255 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1256 if (!device->bdev)
1257 continue;
1258 bdi = blk_get_backing_dev_info(device->bdev);
1259 if (bdi && bdi_congested(bdi, bdi_bits)) {
1260 ret = 1;
1261 break;
1262 }
1263 }
1264 return ret;
1265}
1266
1267/*
1268 * this unplugs every device on the box, and it is only used when page
1269 * is null
1270 */
1271static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1272{
1273 struct btrfs_device *device;
1274 struct btrfs_fs_info *info;
1275
1276 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1277 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1278 if (!device->bdev)
1279 continue;
1280
1281 bdi = blk_get_backing_dev_info(device->bdev);
1282 if (bdi->unplug_io_fn)
1283 bdi->unplug_io_fn(bdi, page);
1284 }
1285}
1286
1287static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1288{
1289 struct inode *inode;
1290 struct extent_map_tree *em_tree;
1291 struct extent_map *em;
1292 struct address_space *mapping;
1293 u64 offset;
1294
1295 /* the generic O_DIRECT read code does this */
1296 if (1 || !page) {
1297 __unplug_io_fn(bdi, page);
1298 return;
1299 }
1300
1301 /*
1302 * page->mapping may change at any time. Get a consistent copy
1303 * and use that for everything below
1304 */
1305 smp_mb();
1306 mapping = page->mapping;
1307 if (!mapping)
1308 return;
1309
1310 inode = mapping->host;
1311
1312 /*
1313 * don't do the expensive searching for a small number of
1314 * devices
1315 */
1316 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1317 __unplug_io_fn(bdi, page);
1318 return;
1319 }
1320
1321 offset = page_offset(page);
1322
1323 em_tree = &BTRFS_I(inode)->extent_tree;
1324 spin_lock(&em_tree->lock);
1325 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1326 spin_unlock(&em_tree->lock);
1327 if (!em) {
1328 __unplug_io_fn(bdi, page);
1329 return;
1330 }
1331
1332 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1333 free_extent_map(em);
1334 __unplug_io_fn(bdi, page);
1335 return;
1336 }
1337 offset = offset - em->start;
1338 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1339 em->block_start + offset, page);
1340 free_extent_map(em);
1341}
1342
1343static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1344{
1345 bdi_init(bdi);
1346 bdi->ra_pages = default_backing_dev_info.ra_pages;
1347 bdi->state = 0;
1348 bdi->capabilities = default_backing_dev_info.capabilities;
1349 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1350 bdi->unplug_io_data = info;
1351 bdi->congested_fn = btrfs_congested_fn;
1352 bdi->congested_data = info;
1353 return 0;
1354}
1355
1356static int bio_ready_for_csum(struct bio *bio)
1357{
1358 u64 length = 0;
1359 u64 buf_len = 0;
1360 u64 start = 0;
1361 struct page *page;
1362 struct extent_io_tree *io_tree = NULL;
1363 struct btrfs_fs_info *info = NULL;
1364 struct bio_vec *bvec;
1365 int i;
1366 int ret;
1367
1368 bio_for_each_segment(bvec, bio, i) {
1369 page = bvec->bv_page;
1370 if (page->private == EXTENT_PAGE_PRIVATE) {
1371 length += bvec->bv_len;
1372 continue;
1373 }
1374 if (!page->private) {
1375 length += bvec->bv_len;
1376 continue;
1377 }
1378 length = bvec->bv_len;
1379 buf_len = page->private >> 2;
1380 start = page_offset(page) + bvec->bv_offset;
1381 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1382 info = BTRFS_I(page->mapping->host)->root->fs_info;
1383 }
1384 /* are we fully contained in this bio? */
1385 if (buf_len <= length)
1386 return 1;
1387
1388 ret = extent_range_uptodate(io_tree, start + length,
1389 start + buf_len - 1);
1390 if (ret == 1)
1391 return ret;
1392 return ret;
1393}
1394
1395/*
1396 * called by the kthread helper functions to finally call the bio end_io
1397 * functions. This is where read checksum verification actually happens
1398 */
1399static void end_workqueue_fn(struct btrfs_work *work)
1400{
1401 struct bio *bio;
1402 struct end_io_wq *end_io_wq;
1403 struct btrfs_fs_info *fs_info;
1404 int error;
1405
1406 end_io_wq = container_of(work, struct end_io_wq, work);
1407 bio = end_io_wq->bio;
1408 fs_info = end_io_wq->info;
1409
1410 /* metadata bio reads are special because the whole tree block must
1411 * be checksummed at once. This makes sure the entire block is in
1412 * ram and up to date before trying to verify things. For
1413 * blocksize <= pagesize, it is basically a noop
1414 */
1415 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1416 !bio_ready_for_csum(bio)) {
1417 btrfs_queue_worker(&fs_info->endio_meta_workers,
1418 &end_io_wq->work);
1419 return;
1420 }
1421 error = end_io_wq->error;
1422 bio->bi_private = end_io_wq->private;
1423 bio->bi_end_io = end_io_wq->end_io;
1424 kfree(end_io_wq);
1425 bio_endio(bio, error);
1426}
1427
1428static int cleaner_kthread(void *arg)
1429{
1430 struct btrfs_root *root = arg;
1431
1432 do {
1433 smp_mb();
1434 if (root->fs_info->closing)
1435 break;
1436
1437 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1438 mutex_lock(&root->fs_info->cleaner_mutex);
1439 btrfs_clean_old_snapshots(root);
1440 mutex_unlock(&root->fs_info->cleaner_mutex);
1441
1442 if (freezing(current)) {
1443 refrigerator();
1444 } else {
1445 smp_mb();
1446 if (root->fs_info->closing)
1447 break;
1448 set_current_state(TASK_INTERRUPTIBLE);
1449 schedule();
1450 __set_current_state(TASK_RUNNING);
1451 }
1452 } while (!kthread_should_stop());
1453 return 0;
1454}
1455
1456static int transaction_kthread(void *arg)
1457{
1458 struct btrfs_root *root = arg;
1459 struct btrfs_trans_handle *trans;
1460 struct btrfs_transaction *cur;
1461 unsigned long now;
1462 unsigned long delay;
1463 int ret;
1464
1465 do {
1466 smp_mb();
1467 if (root->fs_info->closing)
1468 break;
1469
1470 delay = HZ * 30;
1471 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1472 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1473
1474 mutex_lock(&root->fs_info->trans_mutex);
1475 cur = root->fs_info->running_transaction;
1476 if (!cur) {
1477 mutex_unlock(&root->fs_info->trans_mutex);
1478 goto sleep;
1479 }
1480
1481 now = get_seconds();
1482 if (now < cur->start_time || now - cur->start_time < 30) {
1483 mutex_unlock(&root->fs_info->trans_mutex);
1484 delay = HZ * 5;
1485 goto sleep;
1486 }
1487 mutex_unlock(&root->fs_info->trans_mutex);
1488 trans = btrfs_start_transaction(root, 1);
1489 ret = btrfs_commit_transaction(trans, root);
1490
1491sleep:
1492 wake_up_process(root->fs_info->cleaner_kthread);
1493 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1494
1495 if (freezing(current)) {
1496 refrigerator();
1497 } else {
1498 if (root->fs_info->closing)
1499 break;
1500 set_current_state(TASK_INTERRUPTIBLE);
1501 schedule_timeout(delay);
1502 __set_current_state(TASK_RUNNING);
1503 }
1504 } while (!kthread_should_stop());
1505 return 0;
1506}
1507
1508struct btrfs_root *open_ctree(struct super_block *sb,
1509 struct btrfs_fs_devices *fs_devices,
1510 char *options)
1511{
1512 u32 sectorsize;
1513 u32 nodesize;
1514 u32 leafsize;
1515 u32 blocksize;
1516 u32 stripesize;
1517 u64 generation;
1518 u64 features;
1519 struct btrfs_key location;
1520 struct buffer_head *bh;
1521 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1522 GFP_NOFS);
1523 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1524 GFP_NOFS);
1525 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1526 GFP_NOFS);
1527 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1528 GFP_NOFS);
1529 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1530 GFP_NOFS);
1531 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1532 GFP_NOFS);
1533 struct btrfs_root *log_tree_root;
1534
1535 int ret;
1536 int err = -EINVAL;
1537
1538 struct btrfs_super_block *disk_super;
1539
1540 if (!extent_root || !tree_root || !fs_info ||
1541 !chunk_root || !dev_root || !csum_root) {
1542 err = -ENOMEM;
1543 goto fail;
1544 }
1545 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1546 INIT_LIST_HEAD(&fs_info->trans_list);
1547 INIT_LIST_HEAD(&fs_info->dead_roots);
1548 INIT_LIST_HEAD(&fs_info->hashers);
1549 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1550 spin_lock_init(&fs_info->delalloc_lock);
1551 spin_lock_init(&fs_info->new_trans_lock);
1552 spin_lock_init(&fs_info->ref_cache_lock);
1553
1554 init_completion(&fs_info->kobj_unregister);
1555 fs_info->tree_root = tree_root;
1556 fs_info->extent_root = extent_root;
1557 fs_info->csum_root = csum_root;
1558 fs_info->chunk_root = chunk_root;
1559 fs_info->dev_root = dev_root;
1560 fs_info->fs_devices = fs_devices;
1561 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1562 INIT_LIST_HEAD(&fs_info->space_info);
1563 btrfs_mapping_init(&fs_info->mapping_tree);
1564 atomic_set(&fs_info->nr_async_submits, 0);
1565 atomic_set(&fs_info->async_delalloc_pages, 0);
1566 atomic_set(&fs_info->async_submit_draining, 0);
1567 atomic_set(&fs_info->nr_async_bios, 0);
1568 atomic_set(&fs_info->throttles, 0);
1569 atomic_set(&fs_info->throttle_gen, 0);
1570 fs_info->sb = sb;
1571 fs_info->max_extent = (u64)-1;
1572 fs_info->max_inline = 8192 * 1024;
1573 setup_bdi(fs_info, &fs_info->bdi);
1574 fs_info->btree_inode = new_inode(sb);
1575 fs_info->btree_inode->i_ino = 1;
1576 fs_info->btree_inode->i_nlink = 1;
1577
1578 fs_info->thread_pool_size = min_t(unsigned long,
1579 num_online_cpus() + 2, 8);
1580
1581 INIT_LIST_HEAD(&fs_info->ordered_extents);
1582 spin_lock_init(&fs_info->ordered_extent_lock);
1583
1584 sb->s_blocksize = 4096;
1585 sb->s_blocksize_bits = blksize_bits(4096);
1586
1587 /*
1588 * we set the i_size on the btree inode to the max possible int.
1589 * the real end of the address space is determined by all of
1590 * the devices in the system
1591 */
1592 fs_info->btree_inode->i_size = OFFSET_MAX;
1593 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1594 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1595
1596 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1597 fs_info->btree_inode->i_mapping,
1598 GFP_NOFS);
1599 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1600 GFP_NOFS);
1601
1602 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1603
1604 spin_lock_init(&fs_info->block_group_cache_lock);
1605 fs_info->block_group_cache_tree.rb_node = NULL;
1606
1607 extent_io_tree_init(&fs_info->pinned_extents,
1608 fs_info->btree_inode->i_mapping, GFP_NOFS);
1609 fs_info->do_barriers = 1;
1610
1611 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1612 btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1613 btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1614
1615 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1616 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1617 sizeof(struct btrfs_key));
1618 insert_inode_hash(fs_info->btree_inode);
1619
1620 mutex_init(&fs_info->trans_mutex);
1621 mutex_init(&fs_info->tree_log_mutex);
1622 mutex_init(&fs_info->drop_mutex);
1623 mutex_init(&fs_info->pinned_mutex);
1624 mutex_init(&fs_info->chunk_mutex);
1625 mutex_init(&fs_info->transaction_kthread_mutex);
1626 mutex_init(&fs_info->cleaner_mutex);
1627 mutex_init(&fs_info->volume_mutex);
1628 mutex_init(&fs_info->tree_reloc_mutex);
1629 init_waitqueue_head(&fs_info->transaction_throttle);
1630 init_waitqueue_head(&fs_info->transaction_wait);
1631 init_waitqueue_head(&fs_info->async_submit_wait);
1632
1633 __setup_root(4096, 4096, 4096, 4096, tree_root,
1634 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1635
1636
1637 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1638 if (!bh)
1639 goto fail_iput;
1640
1641 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1642 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1643 sizeof(fs_info->super_for_commit));
1644 brelse(bh);
1645
1646 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1647
1648 disk_super = &fs_info->super_copy;
1649 if (!btrfs_super_root(disk_super))
1650 goto fail_iput;
1651
1652 ret = btrfs_parse_options(tree_root, options);
1653 if (ret) {
1654 err = ret;
1655 goto fail_iput;
1656 }
1657
1658 features = btrfs_super_incompat_flags(disk_super) &
1659 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1660 if (features) {
1661 printk(KERN_ERR "BTRFS: couldn't mount because of "
1662 "unsupported optional features (%Lx).\n",
1663 features);
1664 err = -EINVAL;
1665 goto fail_iput;
1666 }
1667
1668 features = btrfs_super_compat_ro_flags(disk_super) &
1669 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1670 if (!(sb->s_flags & MS_RDONLY) && features) {
1671 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1672 "unsupported option features (%Lx).\n",
1673 features);
1674 err = -EINVAL;
1675 goto fail_iput;
1676 }
1677
1678 /*
1679 * we need to start all the end_io workers up front because the
1680 * queue work function gets called at interrupt time, and so it
1681 * cannot dynamically grow.
1682 */
1683 btrfs_init_workers(&fs_info->workers, "worker",
1684 fs_info->thread_pool_size);
1685
1686 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1687 fs_info->thread_pool_size);
1688
1689 btrfs_init_workers(&fs_info->submit_workers, "submit",
1690 min_t(u64, fs_devices->num_devices,
1691 fs_info->thread_pool_size));
1692
1693 /* a higher idle thresh on the submit workers makes it much more
1694 * likely that bios will be send down in a sane order to the
1695 * devices
1696 */
1697 fs_info->submit_workers.idle_thresh = 64;
1698
1699 fs_info->workers.idle_thresh = 16;
1700 fs_info->workers.ordered = 1;
1701
1702 fs_info->delalloc_workers.idle_thresh = 2;
1703 fs_info->delalloc_workers.ordered = 1;
1704
1705 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1706 btrfs_init_workers(&fs_info->endio_workers, "endio",
1707 fs_info->thread_pool_size);
1708 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1709 fs_info->thread_pool_size);
1710 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1711 "endio-meta-write", fs_info->thread_pool_size);
1712 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1713 fs_info->thread_pool_size);
1714
1715 /*
1716 * endios are largely parallel and should have a very
1717 * low idle thresh
1718 */
1719 fs_info->endio_workers.idle_thresh = 4;
1720 fs_info->endio_meta_workers.idle_thresh = 4;
1721
1722 fs_info->endio_write_workers.idle_thresh = 64;
1723 fs_info->endio_meta_write_workers.idle_thresh = 64;
1724
1725 btrfs_start_workers(&fs_info->workers, 1);
1726 btrfs_start_workers(&fs_info->submit_workers, 1);
1727 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1728 btrfs_start_workers(&fs_info->fixup_workers, 1);
1729 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1730 btrfs_start_workers(&fs_info->endio_meta_workers,
1731 fs_info->thread_pool_size);
1732 btrfs_start_workers(&fs_info->endio_meta_write_workers,
1733 fs_info->thread_pool_size);
1734 btrfs_start_workers(&fs_info->endio_write_workers,
1735 fs_info->thread_pool_size);
1736
1737 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1738 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1739 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1740
1741 nodesize = btrfs_super_nodesize(disk_super);
1742 leafsize = btrfs_super_leafsize(disk_super);
1743 sectorsize = btrfs_super_sectorsize(disk_super);
1744 stripesize = btrfs_super_stripesize(disk_super);
1745 tree_root->nodesize = nodesize;
1746 tree_root->leafsize = leafsize;
1747 tree_root->sectorsize = sectorsize;
1748 tree_root->stripesize = stripesize;
1749
1750 sb->s_blocksize = sectorsize;
1751 sb->s_blocksize_bits = blksize_bits(sectorsize);
1752
1753 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1754 sizeof(disk_super->magic))) {
1755 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1756 goto fail_sb_buffer;
1757 }
1758
1759 mutex_lock(&fs_info->chunk_mutex);
1760 ret = btrfs_read_sys_array(tree_root);
1761 mutex_unlock(&fs_info->chunk_mutex);
1762 if (ret) {
1763 printk(KERN_WARNING "btrfs: failed to read the system "
1764 "array on %s\n", sb->s_id);
1765 goto fail_sys_array;
1766 }
1767
1768 blocksize = btrfs_level_size(tree_root,
1769 btrfs_super_chunk_root_level(disk_super));
1770 generation = btrfs_super_chunk_root_generation(disk_super);
1771
1772 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1773 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1774
1775 chunk_root->node = read_tree_block(chunk_root,
1776 btrfs_super_chunk_root(disk_super),
1777 blocksize, generation);
1778 BUG_ON(!chunk_root->node);
1779
1780 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1781 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1782 BTRFS_UUID_SIZE);
1783
1784 mutex_lock(&fs_info->chunk_mutex);
1785 ret = btrfs_read_chunk_tree(chunk_root);
1786 mutex_unlock(&fs_info->chunk_mutex);
1787 if (ret) {
1788 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1789 sb->s_id);
1790 goto fail_chunk_root;
1791 }
1792
1793 btrfs_close_extra_devices(fs_devices);
1794
1795 blocksize = btrfs_level_size(tree_root,
1796 btrfs_super_root_level(disk_super));
1797 generation = btrfs_super_generation(disk_super);
1798
1799 tree_root->node = read_tree_block(tree_root,
1800 btrfs_super_root(disk_super),
1801 blocksize, generation);
1802 if (!tree_root->node)
1803 goto fail_chunk_root;
1804
1805
1806 ret = find_and_setup_root(tree_root, fs_info,
1807 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1808 if (ret)
1809 goto fail_tree_root;
1810 extent_root->track_dirty = 1;
1811
1812 ret = find_and_setup_root(tree_root, fs_info,
1813 BTRFS_DEV_TREE_OBJECTID, dev_root);
1814 dev_root->track_dirty = 1;
1815 if (ret)
1816 goto fail_extent_root;
1817
1818 ret = find_and_setup_root(tree_root, fs_info,
1819 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1820 if (ret)
1821 goto fail_extent_root;
1822
1823 csum_root->track_dirty = 1;
1824
1825 btrfs_read_block_groups(extent_root);
1826
1827 fs_info->generation = generation;
1828 fs_info->last_trans_committed = generation;
1829 fs_info->data_alloc_profile = (u64)-1;
1830 fs_info->metadata_alloc_profile = (u64)-1;
1831 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1832 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1833 "btrfs-cleaner");
1834 if (IS_ERR(fs_info->cleaner_kthread))
1835 goto fail_csum_root;
1836
1837 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1838 tree_root,
1839 "btrfs-transaction");
1840 if (IS_ERR(fs_info->transaction_kthread))
1841 goto fail_cleaner;
1842
1843 if (btrfs_super_log_root(disk_super) != 0) {
1844 u64 bytenr = btrfs_super_log_root(disk_super);
1845
1846 if (fs_devices->rw_devices == 0) {
1847 printk(KERN_WARNING "Btrfs log replay required "
1848 "on RO media\n");
1849 err = -EIO;
1850 goto fail_trans_kthread;
1851 }
1852 blocksize =
1853 btrfs_level_size(tree_root,
1854 btrfs_super_log_root_level(disk_super));
1855
1856 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1857 GFP_NOFS);
1858
1859 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1860 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1861
1862 log_tree_root->node = read_tree_block(tree_root, bytenr,
1863 blocksize,
1864 generation + 1);
1865 ret = btrfs_recover_log_trees(log_tree_root);
1866 BUG_ON(ret);
1867
1868 if (sb->s_flags & MS_RDONLY) {
1869 ret = btrfs_commit_super(tree_root);
1870 BUG_ON(ret);
1871 }
1872 }
1873
1874 if (!(sb->s_flags & MS_RDONLY)) {
1875 ret = btrfs_cleanup_reloc_trees(tree_root);
1876 BUG_ON(ret);
1877 }
1878
1879 location.objectid = BTRFS_FS_TREE_OBJECTID;
1880 location.type = BTRFS_ROOT_ITEM_KEY;
1881 location.offset = (u64)-1;
1882
1883 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1884 if (!fs_info->fs_root)
1885 goto fail_trans_kthread;
1886 return tree_root;
1887
1888fail_trans_kthread:
1889 kthread_stop(fs_info->transaction_kthread);
1890fail_cleaner:
1891 kthread_stop(fs_info->cleaner_kthread);
1892
1893 /*
1894 * make sure we're done with the btree inode before we stop our
1895 * kthreads
1896 */
1897 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1898 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1899
1900fail_csum_root:
1901 free_extent_buffer(csum_root->node);
1902fail_extent_root:
1903 free_extent_buffer(extent_root->node);
1904fail_tree_root:
1905 free_extent_buffer(tree_root->node);
1906fail_chunk_root:
1907 free_extent_buffer(chunk_root->node);
1908fail_sys_array:
1909 free_extent_buffer(dev_root->node);
1910fail_sb_buffer:
1911 btrfs_stop_workers(&fs_info->fixup_workers);
1912 btrfs_stop_workers(&fs_info->delalloc_workers);
1913 btrfs_stop_workers(&fs_info->workers);
1914 btrfs_stop_workers(&fs_info->endio_workers);
1915 btrfs_stop_workers(&fs_info->endio_meta_workers);
1916 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1917 btrfs_stop_workers(&fs_info->endio_write_workers);
1918 btrfs_stop_workers(&fs_info->submit_workers);
1919fail_iput:
1920 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1921 iput(fs_info->btree_inode);
1922
1923 btrfs_close_devices(fs_info->fs_devices);
1924 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1925 bdi_destroy(&fs_info->bdi);
1926
1927fail:
1928 kfree(extent_root);
1929 kfree(tree_root);
1930 kfree(fs_info);
1931 kfree(chunk_root);
1932 kfree(dev_root);
1933 kfree(csum_root);
1934 return ERR_PTR(err);
1935}
1936
1937static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1938{
1939 char b[BDEVNAME_SIZE];
1940
1941 if (uptodate) {
1942 set_buffer_uptodate(bh);
1943 } else {
1944 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1945 printk(KERN_WARNING "lost page write due to "
1946 "I/O error on %s\n",
1947 bdevname(bh->b_bdev, b));
1948 }
1949 /* note, we dont' set_buffer_write_io_error because we have
1950 * our own ways of dealing with the IO errors
1951 */
1952 clear_buffer_uptodate(bh);
1953 }
1954 unlock_buffer(bh);
1955 put_bh(bh);
1956}
1957
1958struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1959{
1960 struct buffer_head *bh;
1961 struct buffer_head *latest = NULL;
1962 struct btrfs_super_block *super;
1963 int i;
1964 u64 transid = 0;
1965 u64 bytenr;
1966
1967 /* we would like to check all the supers, but that would make
1968 * a btrfs mount succeed after a mkfs from a different FS.
1969 * So, we need to add a special mount option to scan for
1970 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1971 */
1972 for (i = 0; i < 1; i++) {
1973 bytenr = btrfs_sb_offset(i);
1974 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
1975 break;
1976 bh = __bread(bdev, bytenr / 4096, 4096);
1977 if (!bh)
1978 continue;
1979
1980 super = (struct btrfs_super_block *)bh->b_data;
1981 if (btrfs_super_bytenr(super) != bytenr ||
1982 strncmp((char *)(&super->magic), BTRFS_MAGIC,
1983 sizeof(super->magic))) {
1984 brelse(bh);
1985 continue;
1986 }
1987
1988 if (!latest || btrfs_super_generation(super) > transid) {
1989 brelse(latest);
1990 latest = bh;
1991 transid = btrfs_super_generation(super);
1992 } else {
1993 brelse(bh);
1994 }
1995 }
1996 return latest;
1997}
1998
1999static int write_dev_supers(struct btrfs_device *device,
2000 struct btrfs_super_block *sb,
2001 int do_barriers, int wait, int max_mirrors)
2002{
2003 struct buffer_head *bh;
2004 int i;
2005 int ret;
2006 int errors = 0;
2007 u32 crc;
2008 u64 bytenr;
2009 int last_barrier = 0;
2010
2011 if (max_mirrors == 0)
2012 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2013
2014 /* make sure only the last submit_bh does a barrier */
2015 if (do_barriers) {
2016 for (i = 0; i < max_mirrors; i++) {
2017 bytenr = btrfs_sb_offset(i);
2018 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2019 device->total_bytes)
2020 break;
2021 last_barrier = i;
2022 }
2023 }
2024
2025 for (i = 0; i < max_mirrors; i++) {
2026 bytenr = btrfs_sb_offset(i);
2027 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2028 break;
2029
2030 if (wait) {
2031 bh = __find_get_block(device->bdev, bytenr / 4096,
2032 BTRFS_SUPER_INFO_SIZE);
2033 BUG_ON(!bh);
2034 brelse(bh);
2035 wait_on_buffer(bh);
2036 if (buffer_uptodate(bh)) {
2037 brelse(bh);
2038 continue;
2039 }
2040 } else {
2041 btrfs_set_super_bytenr(sb, bytenr);
2042
2043 crc = ~(u32)0;
2044 crc = btrfs_csum_data(NULL, (char *)sb +
2045 BTRFS_CSUM_SIZE, crc,
2046 BTRFS_SUPER_INFO_SIZE -
2047 BTRFS_CSUM_SIZE);
2048 btrfs_csum_final(crc, sb->csum);
2049
2050 bh = __getblk(device->bdev, bytenr / 4096,
2051 BTRFS_SUPER_INFO_SIZE);
2052 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2053
2054 set_buffer_uptodate(bh);
2055 get_bh(bh);
2056 lock_buffer(bh);
2057 bh->b_end_io = btrfs_end_buffer_write_sync;
2058 }
2059
2060 if (i == last_barrier && do_barriers && device->barriers) {
2061 ret = submit_bh(WRITE_BARRIER, bh);
2062 if (ret == -EOPNOTSUPP) {
2063 printk("btrfs: disabling barriers on dev %s\n",
2064 device->name);
2065 set_buffer_uptodate(bh);
2066 device->barriers = 0;
2067 get_bh(bh);
2068 lock_buffer(bh);
2069 ret = submit_bh(WRITE, bh);
2070 }
2071 } else {
2072 ret = submit_bh(WRITE, bh);
2073 }
2074
2075 if (!ret && wait) {
2076 wait_on_buffer(bh);
2077 if (!buffer_uptodate(bh))
2078 errors++;
2079 } else if (ret) {
2080 errors++;
2081 }
2082 if (wait)
2083 brelse(bh);
2084 }
2085 return errors < i ? 0 : -1;
2086}
2087
2088int write_all_supers(struct btrfs_root *root, int max_mirrors)
2089{
2090 struct list_head *head = &root->fs_info->fs_devices->devices;
2091 struct btrfs_device *dev;
2092 struct btrfs_super_block *sb;
2093 struct btrfs_dev_item *dev_item;
2094 int ret;
2095 int do_barriers;
2096 int max_errors;
2097 int total_errors = 0;
2098 u64 flags;
2099
2100 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2101 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2102
2103 sb = &root->fs_info->super_for_commit;
2104 dev_item = &sb->dev_item;
2105 list_for_each_entry(dev, head, dev_list) {
2106 if (!dev->bdev) {
2107 total_errors++;
2108 continue;
2109 }
2110 if (!dev->in_fs_metadata || !dev->writeable)
2111 continue;
2112
2113 btrfs_set_stack_device_generation(dev_item, 0);
2114 btrfs_set_stack_device_type(dev_item, dev->type);
2115 btrfs_set_stack_device_id(dev_item, dev->devid);
2116 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2117 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2118 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2119 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2120 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2121 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2122 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2123
2124 flags = btrfs_super_flags(sb);
2125 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2126
2127 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2128 if (ret)
2129 total_errors++;
2130 }
2131 if (total_errors > max_errors) {
2132 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2133 total_errors);
2134 BUG();
2135 }
2136
2137 total_errors = 0;
2138 list_for_each_entry(dev, head, dev_list) {
2139 if (!dev->bdev)
2140 continue;
2141 if (!dev->in_fs_metadata || !dev->writeable)
2142 continue;
2143
2144 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2145 if (ret)
2146 total_errors++;
2147 }
2148 if (total_errors > max_errors) {
2149 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2150 total_errors);
2151 BUG();
2152 }
2153 return 0;
2154}
2155
2156int write_ctree_super(struct btrfs_trans_handle *trans,
2157 struct btrfs_root *root, int max_mirrors)
2158{
2159 int ret;
2160
2161 ret = write_all_supers(root, max_mirrors);
2162 return ret;
2163}
2164
2165int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2166{
2167 radix_tree_delete(&fs_info->fs_roots_radix,
2168 (unsigned long)root->root_key.objectid);
2169 if (root->anon_super.s_dev) {
2170 down_write(&root->anon_super.s_umount);
2171 kill_anon_super(&root->anon_super);
2172 }
2173 if (root->node)
2174 free_extent_buffer(root->node);
2175 if (root->commit_root)
2176 free_extent_buffer(root->commit_root);
2177 kfree(root->name);
2178 kfree(root);
2179 return 0;
2180}
2181
2182static int del_fs_roots(struct btrfs_fs_info *fs_info)
2183{
2184 int ret;
2185 struct btrfs_root *gang[8];
2186 int i;
2187
2188 while (1) {
2189 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2190 (void **)gang, 0,
2191 ARRAY_SIZE(gang));
2192 if (!ret)
2193 break;
2194 for (i = 0; i < ret; i++)
2195 btrfs_free_fs_root(fs_info, gang[i]);
2196 }
2197 return 0;
2198}
2199
2200int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2201{
2202 u64 root_objectid = 0;
2203 struct btrfs_root *gang[8];
2204 int i;
2205 int ret;
2206
2207 while (1) {
2208 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2209 (void **)gang, root_objectid,
2210 ARRAY_SIZE(gang));
2211 if (!ret)
2212 break;
2213 for (i = 0; i < ret; i++) {
2214 root_objectid = gang[i]->root_key.objectid;
2215 ret = btrfs_find_dead_roots(fs_info->tree_root,
2216 root_objectid, gang[i]);
2217 BUG_ON(ret);
2218 btrfs_orphan_cleanup(gang[i]);
2219 }
2220 root_objectid++;
2221 }
2222 return 0;
2223}
2224
2225int btrfs_commit_super(struct btrfs_root *root)
2226{
2227 struct btrfs_trans_handle *trans;
2228 int ret;
2229
2230 mutex_lock(&root->fs_info->cleaner_mutex);
2231 btrfs_clean_old_snapshots(root);
2232 mutex_unlock(&root->fs_info->cleaner_mutex);
2233 trans = btrfs_start_transaction(root, 1);
2234 ret = btrfs_commit_transaction(trans, root);
2235 BUG_ON(ret);
2236 /* run commit again to drop the original snapshot */
2237 trans = btrfs_start_transaction(root, 1);
2238 btrfs_commit_transaction(trans, root);
2239 ret = btrfs_write_and_wait_transaction(NULL, root);
2240 BUG_ON(ret);
2241
2242 ret = write_ctree_super(NULL, root, 0);
2243 return ret;
2244}
2245
2246int close_ctree(struct btrfs_root *root)
2247{
2248 struct btrfs_fs_info *fs_info = root->fs_info;
2249 int ret;
2250
2251 fs_info->closing = 1;
2252 smp_mb();
2253
2254 kthread_stop(root->fs_info->transaction_kthread);
2255 kthread_stop(root->fs_info->cleaner_kthread);
2256
2257 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2258 ret = btrfs_commit_super(root);
2259 if (ret)
2260 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2261 }
2262
2263 if (fs_info->delalloc_bytes) {
2264 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2265 fs_info->delalloc_bytes);
2266 }
2267 if (fs_info->total_ref_cache_size) {
2268 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2269 (unsigned long long)fs_info->total_ref_cache_size);
2270 }
2271
2272 if (fs_info->extent_root->node)
2273 free_extent_buffer(fs_info->extent_root->node);
2274
2275 if (fs_info->tree_root->node)
2276 free_extent_buffer(fs_info->tree_root->node);
2277
2278 if (root->fs_info->chunk_root->node)
2279 free_extent_buffer(root->fs_info->chunk_root->node);
2280
2281 if (root->fs_info->dev_root->node)
2282 free_extent_buffer(root->fs_info->dev_root->node);
2283
2284 if (root->fs_info->csum_root->node)
2285 free_extent_buffer(root->fs_info->csum_root->node);
2286
2287 btrfs_free_block_groups(root->fs_info);
2288
2289 del_fs_roots(fs_info);
2290
2291 iput(fs_info->btree_inode);
2292
2293 btrfs_stop_workers(&fs_info->fixup_workers);
2294 btrfs_stop_workers(&fs_info->delalloc_workers);
2295 btrfs_stop_workers(&fs_info->workers);
2296 btrfs_stop_workers(&fs_info->endio_workers);
2297 btrfs_stop_workers(&fs_info->endio_meta_workers);
2298 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2299 btrfs_stop_workers(&fs_info->endio_write_workers);
2300 btrfs_stop_workers(&fs_info->submit_workers);
2301
2302#if 0
2303 while (!list_empty(&fs_info->hashers)) {
2304 struct btrfs_hasher *hasher;
2305 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2306 hashers);
2307 list_del(&hasher->hashers);
2308 crypto_free_hash(&fs_info->hash_tfm);
2309 kfree(hasher);
2310 }
2311#endif
2312 btrfs_close_devices(fs_info->fs_devices);
2313 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2314
2315 bdi_destroy(&fs_info->bdi);
2316
2317 kfree(fs_info->extent_root);
2318 kfree(fs_info->tree_root);
2319 kfree(fs_info->chunk_root);
2320 kfree(fs_info->dev_root);
2321 kfree(fs_info->csum_root);
2322 return 0;
2323}
2324
2325int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2326{
2327 int ret;
2328 struct inode *btree_inode = buf->first_page->mapping->host;
2329
2330 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2331 if (!ret)
2332 return ret;
2333
2334 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2335 parent_transid);
2336 return !ret;
2337}
2338
2339int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2340{
2341 struct inode *btree_inode = buf->first_page->mapping->host;
2342 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2343 buf);
2344}
2345
2346void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2347{
2348 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2349 u64 transid = btrfs_header_generation(buf);
2350 struct inode *btree_inode = root->fs_info->btree_inode;
2351
2352 btrfs_set_lock_blocking(buf);
2353
2354 btrfs_assert_tree_locked(buf);
2355 if (transid != root->fs_info->generation) {
2356 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2357 "found %llu running %llu\n",
2358 (unsigned long long)buf->start,
2359 (unsigned long long)transid,
2360 (unsigned long long)root->fs_info->generation);
2361 WARN_ON(1);
2362 }
2363 set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree, buf);
2364}
2365
2366void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2367{
2368 /*
2369 * looks as though older kernels can get into trouble with
2370 * this code, they end up stuck in balance_dirty_pages forever
2371 */
2372 struct extent_io_tree *tree;
2373 u64 num_dirty;
2374 u64 start = 0;
2375 unsigned long thresh = 32 * 1024 * 1024;
2376 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2377
2378 if (current_is_pdflush() || current->flags & PF_MEMALLOC)
2379 return;
2380
2381 num_dirty = count_range_bits(tree, &start, (u64)-1,
2382 thresh, EXTENT_DIRTY);
2383 if (num_dirty > thresh) {
2384 balance_dirty_pages_ratelimited_nr(
2385 root->fs_info->btree_inode->i_mapping, 1);
2386 }
2387 return;
2388}
2389
2390int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2391{
2392 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2393 int ret;
2394 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2395 if (ret == 0)
2396 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2397 return ret;
2398}
2399
2400int btree_lock_page_hook(struct page *page)
2401{
2402 struct inode *inode = page->mapping->host;
2403 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2404 struct extent_buffer *eb;
2405 unsigned long len;
2406 u64 bytenr = page_offset(page);
2407
2408 if (page->private == EXTENT_PAGE_PRIVATE)
2409 goto out;
2410
2411 len = page->private >> 2;
2412 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2413 if (!eb)
2414 goto out;
2415
2416 btrfs_tree_lock(eb);
2417 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2418 btrfs_tree_unlock(eb);
2419 free_extent_buffer(eb);
2420out:
2421 lock_page(page);
2422 return 0;
2423}
2424
2425static struct extent_io_ops btree_extent_io_ops = {
2426 .write_cache_pages_lock_hook = btree_lock_page_hook,
2427 .readpage_end_io_hook = btree_readpage_end_io_hook,
2428 .submit_bio_hook = btree_submit_bio_hook,
2429 /* note we're sharing with inode.c for the merge bio hook */
2430 .merge_bio_hook = btrfs_merge_bio_hook,
2431};
Linux 6.x block layer and io_uring tree(s)