projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Btrfs: change how subvolumes are organized
[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 <linux/crc32c.h>
30#include "compat.h"
31#include "ctree.h"
32#include "disk-io.h"
33#include "transaction.h"
34#include "btrfs_inode.h"
35#include "volumes.h"
36#include "print-tree.h"
37#include "async-thread.h"
38#include "locking.h"
39#include "tree-log.h"
40#include "free-space-cache.h"
41
42static struct extent_io_ops btree_extent_io_ops;
43static void end_workqueue_fn(struct btrfs_work *work);
44static void free_fs_root(struct btrfs_root *root);
45
46static atomic_t btrfs_bdi_num = ATOMIC_INIT(0);
47
48/*
49 * end_io_wq structs are used to do processing in task context when an IO is
50 * complete. This is used during reads to verify checksums, and it is used
51 * by writes to insert metadata for new file extents after IO is complete.
52 */
53struct end_io_wq {
54 struct bio *bio;
55 bio_end_io_t *end_io;
56 void *private;
57 struct btrfs_fs_info *info;
58 int error;
59 int metadata;
60 struct list_head list;
61 struct btrfs_work work;
62};
63
64/*
65 * async submit bios are used to offload expensive checksumming
66 * onto the worker threads. They checksum file and metadata bios
67 * just before they are sent down the IO stack.
68 */
69struct async_submit_bio {
70 struct inode *inode;
71 struct bio *bio;
72 struct list_head list;
73 extent_submit_bio_hook_t *submit_bio_start;
74 extent_submit_bio_hook_t *submit_bio_done;
75 int rw;
76 int mirror_num;
77 unsigned long bio_flags;
78 struct btrfs_work work;
79};
80
81/* These are used to set the lockdep class on the extent buffer locks.
82 * The class is set by the readpage_end_io_hook after the buffer has
83 * passed csum validation but before the pages are unlocked.
84 *
85 * The lockdep class is also set by btrfs_init_new_buffer on freshly
86 * allocated blocks.
87 *
88 * The class is based on the level in the tree block, which allows lockdep
89 * to know that lower nodes nest inside the locks of higher nodes.
90 *
91 * We also add a check to make sure the highest level of the tree is
92 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
93 * code needs update as well.
94 */
95#ifdef CONFIG_DEBUG_LOCK_ALLOC
96# if BTRFS_MAX_LEVEL != 8
97# error
98# endif
99static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
100static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
101 /* leaf */
102 "btrfs-extent-00",
103 "btrfs-extent-01",
104 "btrfs-extent-02",
105 "btrfs-extent-03",
106 "btrfs-extent-04",
107 "btrfs-extent-05",
108 "btrfs-extent-06",
109 "btrfs-extent-07",
110 /* highest possible level */
111 "btrfs-extent-08",
112};
113#endif
114
115/*
116 * extents on the btree inode are pretty simple, there's one extent
117 * that covers the entire device
118 */
119static struct extent_map *btree_get_extent(struct inode *inode,
120 struct page *page, size_t page_offset, u64 start, u64 len,
121 int create)
122{
123 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
124 struct extent_map *em;
125 int ret;
126
127 read_lock(&em_tree->lock);
128 em = lookup_extent_mapping(em_tree, start, len);
129 if (em) {
130 em->bdev =
131 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
132 read_unlock(&em_tree->lock);
133 goto out;
134 }
135 read_unlock(&em_tree->lock);
136
137 em = alloc_extent_map(GFP_NOFS);
138 if (!em) {
139 em = ERR_PTR(-ENOMEM);
140 goto out;
141 }
142 em->start = 0;
143 em->len = (u64)-1;
144 em->block_len = (u64)-1;
145 em->block_start = 0;
146 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
147
148 write_lock(&em_tree->lock);
149 ret = add_extent_mapping(em_tree, em);
150 if (ret == -EEXIST) {
151 u64 failed_start = em->start;
152 u64 failed_len = em->len;
153
154 free_extent_map(em);
155 em = lookup_extent_mapping(em_tree, start, len);
156 if (em) {
157 ret = 0;
158 } else {
159 em = lookup_extent_mapping(em_tree, failed_start,
160 failed_len);
161 ret = -EIO;
162 }
163 } else if (ret) {
164 free_extent_map(em);
165 em = NULL;
166 }
167 write_unlock(&em_tree->lock);
168
169 if (ret)
170 em = ERR_PTR(ret);
171out:
172 return em;
173}
174
175u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
176{
177 return crc32c(seed, data, len);
178}
179
180void btrfs_csum_final(u32 crc, char *result)
181{
182 *(__le32 *)result = ~cpu_to_le32(crc);
183}
184
185/*
186 * compute the csum for a btree block, and either verify it or write it
187 * into the csum field of the block.
188 */
189static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
190 int verify)
191{
192 u16 csum_size =
193 btrfs_super_csum_size(&root->fs_info->super_copy);
194 char *result = NULL;
195 unsigned long len;
196 unsigned long cur_len;
197 unsigned long offset = BTRFS_CSUM_SIZE;
198 char *map_token = NULL;
199 char *kaddr;
200 unsigned long map_start;
201 unsigned long map_len;
202 int err;
203 u32 crc = ~(u32)0;
204 unsigned long inline_result;
205
206 len = buf->len - offset;
207 while (len > 0) {
208 err = map_private_extent_buffer(buf, offset, 32,
209 &map_token, &kaddr,
210 &map_start, &map_len, KM_USER0);
211 if (err)
212 return 1;
213 cur_len = min(len, map_len - (offset - map_start));
214 crc = btrfs_csum_data(root, kaddr + offset - map_start,
215 crc, cur_len);
216 len -= cur_len;
217 offset += cur_len;
218 unmap_extent_buffer(buf, map_token, KM_USER0);
219 }
220 if (csum_size > sizeof(inline_result)) {
221 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
222 if (!result)
223 return 1;
224 } else {
225 result = (char *)&inline_result;
226 }
227
228 btrfs_csum_final(crc, result);
229
230 if (verify) {
231 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
232 u32 val;
233 u32 found = 0;
234 memcpy(&found, result, csum_size);
235
236 read_extent_buffer(buf, &val, 0, csum_size);
237 if (printk_ratelimit()) {
238 printk(KERN_INFO "btrfs: %s checksum verify "
239 "failed on %llu wanted %X found %X "
240 "level %d\n",
241 root->fs_info->sb->s_id,
242 (unsigned long long)buf->start, val, found,
243 btrfs_header_level(buf));
244 }
245 if (result != (char *)&inline_result)
246 kfree(result);
247 return 1;
248 }
249 } else {
250 write_extent_buffer(buf, result, 0, csum_size);
251 }
252 if (result != (char *)&inline_result)
253 kfree(result);
254 return 0;
255}
256
257/*
258 * we can't consider a given block up to date unless the transid of the
259 * block matches the transid in the parent node's pointer. This is how we
260 * detect blocks that either didn't get written at all or got written
261 * in the wrong place.
262 */
263static int verify_parent_transid(struct extent_io_tree *io_tree,
264 struct extent_buffer *eb, u64 parent_transid)
265{
266 int ret;
267
268 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
269 return 0;
270
271 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
272 if (extent_buffer_uptodate(io_tree, eb) &&
273 btrfs_header_generation(eb) == parent_transid) {
274 ret = 0;
275 goto out;
276 }
277 if (printk_ratelimit()) {
278 printk("parent transid verify failed on %llu wanted %llu "
279 "found %llu\n",
280 (unsigned long long)eb->start,
281 (unsigned long long)parent_transid,
282 (unsigned long long)btrfs_header_generation(eb));
283 }
284 ret = 1;
285 clear_extent_buffer_uptodate(io_tree, eb);
286out:
287 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
288 GFP_NOFS);
289 return ret;
290}
291
292/*
293 * helper to read a given tree block, doing retries as required when
294 * the checksums don't match and we have alternate mirrors to try.
295 */
296static int btree_read_extent_buffer_pages(struct btrfs_root *root,
297 struct extent_buffer *eb,
298 u64 start, u64 parent_transid)
299{
300 struct extent_io_tree *io_tree;
301 int ret;
302 int num_copies = 0;
303 int mirror_num = 0;
304
305 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
306 while (1) {
307 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
308 btree_get_extent, mirror_num);
309 if (!ret &&
310 !verify_parent_transid(io_tree, eb, parent_transid))
311 return ret;
312
313 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
314 eb->start, eb->len);
315 if (num_copies == 1)
316 return ret;
317
318 mirror_num++;
319 if (mirror_num > num_copies)
320 return ret;
321 }
322 return -EIO;
323}
324
325/*
326 * checksum a dirty tree block before IO. This has extra checks to make sure
327 * we only fill in the checksum field in the first page of a multi-page block
328 */
329
330static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
331{
332 struct extent_io_tree *tree;
333 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
334 u64 found_start;
335 int found_level;
336 unsigned long len;
337 struct extent_buffer *eb;
338 int ret;
339
340 tree = &BTRFS_I(page->mapping->host)->io_tree;
341
342 if (page->private == EXTENT_PAGE_PRIVATE)
343 goto out;
344 if (!page->private)
345 goto out;
346 len = page->private >> 2;
347 WARN_ON(len == 0);
348
349 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
350 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
351 btrfs_header_generation(eb));
352 BUG_ON(ret);
353 found_start = btrfs_header_bytenr(eb);
354 if (found_start != start) {
355 WARN_ON(1);
356 goto err;
357 }
358 if (eb->first_page != page) {
359 WARN_ON(1);
360 goto err;
361 }
362 if (!PageUptodate(page)) {
363 WARN_ON(1);
364 goto err;
365 }
366 found_level = btrfs_header_level(eb);
367
368 csum_tree_block(root, eb, 0);
369err:
370 free_extent_buffer(eb);
371out:
372 return 0;
373}
374
375static int check_tree_block_fsid(struct btrfs_root *root,
376 struct extent_buffer *eb)
377{
378 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
379 u8 fsid[BTRFS_UUID_SIZE];
380 int ret = 1;
381
382 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
383 BTRFS_FSID_SIZE);
384 while (fs_devices) {
385 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
386 ret = 0;
387 break;
388 }
389 fs_devices = fs_devices->seed;
390 }
391 return ret;
392}
393
394#ifdef CONFIG_DEBUG_LOCK_ALLOC
395void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
396{
397 lockdep_set_class_and_name(&eb->lock,
398 &btrfs_eb_class[level],
399 btrfs_eb_name[level]);
400}
401#endif
402
403static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
404 struct extent_state *state)
405{
406 struct extent_io_tree *tree;
407 u64 found_start;
408 int found_level;
409 unsigned long len;
410 struct extent_buffer *eb;
411 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
412 int ret = 0;
413
414 tree = &BTRFS_I(page->mapping->host)->io_tree;
415 if (page->private == EXTENT_PAGE_PRIVATE)
416 goto out;
417 if (!page->private)
418 goto out;
419
420 len = page->private >> 2;
421 WARN_ON(len == 0);
422
423 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
424
425 found_start = btrfs_header_bytenr(eb);
426 if (found_start != start) {
427 if (printk_ratelimit()) {
428 printk(KERN_INFO "btrfs bad tree block start "
429 "%llu %llu\n",
430 (unsigned long long)found_start,
431 (unsigned long long)eb->start);
432 }
433 ret = -EIO;
434 goto err;
435 }
436 if (eb->first_page != page) {
437 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
438 eb->first_page->index, page->index);
439 WARN_ON(1);
440 ret = -EIO;
441 goto err;
442 }
443 if (check_tree_block_fsid(root, eb)) {
444 if (printk_ratelimit()) {
445 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
446 (unsigned long long)eb->start);
447 }
448 ret = -EIO;
449 goto err;
450 }
451 found_level = btrfs_header_level(eb);
452
453 btrfs_set_buffer_lockdep_class(eb, found_level);
454
455 ret = csum_tree_block(root, eb, 1);
456 if (ret)
457 ret = -EIO;
458
459 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
460 end = eb->start + end - 1;
461err:
462 free_extent_buffer(eb);
463out:
464 return ret;
465}
466
467static void end_workqueue_bio(struct bio *bio, int err)
468{
469 struct end_io_wq *end_io_wq = bio->bi_private;
470 struct btrfs_fs_info *fs_info;
471
472 fs_info = end_io_wq->info;
473 end_io_wq->error = err;
474 end_io_wq->work.func = end_workqueue_fn;
475 end_io_wq->work.flags = 0;
476
477 if (bio->bi_rw & (1 << BIO_RW)) {
478 if (end_io_wq->metadata)
479 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
480 &end_io_wq->work);
481 else
482 btrfs_queue_worker(&fs_info->endio_write_workers,
483 &end_io_wq->work);
484 } else {
485 if (end_io_wq->metadata)
486 btrfs_queue_worker(&fs_info->endio_meta_workers,
487 &end_io_wq->work);
488 else
489 btrfs_queue_worker(&fs_info->endio_workers,
490 &end_io_wq->work);
491 }
492}
493
494int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
495 int metadata)
496{
497 struct end_io_wq *end_io_wq;
498 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
499 if (!end_io_wq)
500 return -ENOMEM;
501
502 end_io_wq->private = bio->bi_private;
503 end_io_wq->end_io = bio->bi_end_io;
504 end_io_wq->info = info;
505 end_io_wq->error = 0;
506 end_io_wq->bio = bio;
507 end_io_wq->metadata = metadata;
508
509 bio->bi_private = end_io_wq;
510 bio->bi_end_io = end_workqueue_bio;
511 return 0;
512}
513
514unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
515{
516 unsigned long limit = min_t(unsigned long,
517 info->workers.max_workers,
518 info->fs_devices->open_devices);
519 return 256 * limit;
520}
521
522int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
523{
524 return atomic_read(&info->nr_async_bios) >
525 btrfs_async_submit_limit(info);
526}
527
528static void run_one_async_start(struct btrfs_work *work)
529{
530 struct btrfs_fs_info *fs_info;
531 struct async_submit_bio *async;
532
533 async = container_of(work, struct async_submit_bio, work);
534 fs_info = BTRFS_I(async->inode)->root->fs_info;
535 async->submit_bio_start(async->inode, async->rw, async->bio,
536 async->mirror_num, async->bio_flags);
537}
538
539static void run_one_async_done(struct btrfs_work *work)
540{
541 struct btrfs_fs_info *fs_info;
542 struct async_submit_bio *async;
543 int limit;
544
545 async = container_of(work, struct async_submit_bio, work);
546 fs_info = BTRFS_I(async->inode)->root->fs_info;
547
548 limit = btrfs_async_submit_limit(fs_info);
549 limit = limit * 2 / 3;
550
551 atomic_dec(&fs_info->nr_async_submits);
552
553 if (atomic_read(&fs_info->nr_async_submits) < limit &&
554 waitqueue_active(&fs_info->async_submit_wait))
555 wake_up(&fs_info->async_submit_wait);
556
557 async->submit_bio_done(async->inode, async->rw, async->bio,
558 async->mirror_num, async->bio_flags);
559}
560
561static void run_one_async_free(struct btrfs_work *work)
562{
563 struct async_submit_bio *async;
564
565 async = container_of(work, struct async_submit_bio, work);
566 kfree(async);
567}
568
569int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
570 int rw, struct bio *bio, int mirror_num,
571 unsigned long bio_flags,
572 extent_submit_bio_hook_t *submit_bio_start,
573 extent_submit_bio_hook_t *submit_bio_done)
574{
575 struct async_submit_bio *async;
576
577 async = kmalloc(sizeof(*async), GFP_NOFS);
578 if (!async)
579 return -ENOMEM;
580
581 async->inode = inode;
582 async->rw = rw;
583 async->bio = bio;
584 async->mirror_num = mirror_num;
585 async->submit_bio_start = submit_bio_start;
586 async->submit_bio_done = submit_bio_done;
587
588 async->work.func = run_one_async_start;
589 async->work.ordered_func = run_one_async_done;
590 async->work.ordered_free = run_one_async_free;
591
592 async->work.flags = 0;
593 async->bio_flags = bio_flags;
594
595 atomic_inc(&fs_info->nr_async_submits);
596
597 if (rw & (1 << BIO_RW_SYNCIO))
598 btrfs_set_work_high_prio(&async->work);
599
600 btrfs_queue_worker(&fs_info->workers, &async->work);
601
602 while (atomic_read(&fs_info->async_submit_draining) &&
603 atomic_read(&fs_info->nr_async_submits)) {
604 wait_event(fs_info->async_submit_wait,
605 (atomic_read(&fs_info->nr_async_submits) == 0));
606 }
607
608 return 0;
609}
610
611static int btree_csum_one_bio(struct bio *bio)
612{
613 struct bio_vec *bvec = bio->bi_io_vec;
614 int bio_index = 0;
615 struct btrfs_root *root;
616
617 WARN_ON(bio->bi_vcnt <= 0);
618 while (bio_index < bio->bi_vcnt) {
619 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
620 csum_dirty_buffer(root, bvec->bv_page);
621 bio_index++;
622 bvec++;
623 }
624 return 0;
625}
626
627static int __btree_submit_bio_start(struct inode *inode, int rw,
628 struct bio *bio, int mirror_num,
629 unsigned long bio_flags)
630{
631 /*
632 * when we're called for a write, we're already in the async
633 * submission context. Just jump into btrfs_map_bio
634 */
635 btree_csum_one_bio(bio);
636 return 0;
637}
638
639static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
640 int mirror_num, unsigned long bio_flags)
641{
642 /*
643 * when we're called for a write, we're already in the async
644 * submission context. Just jump into btrfs_map_bio
645 */
646 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
647}
648
649static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
650 int mirror_num, unsigned long bio_flags)
651{
652 int ret;
653
654 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
655 bio, 1);
656 BUG_ON(ret);
657
658 if (!(rw & (1 << BIO_RW))) {
659 /*
660 * called for a read, do the setup so that checksum validation
661 * can happen in the async kernel threads
662 */
663 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
664 mirror_num, 0);
665 }
666
667 /*
668 * kthread helpers are used to submit writes so that checksumming
669 * can happen in parallel across all CPUs
670 */
671 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
672 inode, rw, bio, mirror_num, 0,
673 __btree_submit_bio_start,
674 __btree_submit_bio_done);
675}
676
677static int btree_writepage(struct page *page, struct writeback_control *wbc)
678{
679 struct extent_io_tree *tree;
680 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
681 struct extent_buffer *eb;
682 int was_dirty;
683
684 tree = &BTRFS_I(page->mapping->host)->io_tree;
685 if (!(current->flags & PF_MEMALLOC)) {
686 return extent_write_full_page(tree, page,
687 btree_get_extent, wbc);
688 }
689
690 redirty_page_for_writepage(wbc, page);
691 eb = btrfs_find_tree_block(root, page_offset(page),
692 PAGE_CACHE_SIZE);
693 WARN_ON(!eb);
694
695 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
696 if (!was_dirty) {
697 spin_lock(&root->fs_info->delalloc_lock);
698 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
699 spin_unlock(&root->fs_info->delalloc_lock);
700 }
701 free_extent_buffer(eb);
702
703 unlock_page(page);
704 return 0;
705}
706
707static int btree_writepages(struct address_space *mapping,
708 struct writeback_control *wbc)
709{
710 struct extent_io_tree *tree;
711 tree = &BTRFS_I(mapping->host)->io_tree;
712 if (wbc->sync_mode == WB_SYNC_NONE) {
713 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
714 u64 num_dirty;
715 unsigned long thresh = 32 * 1024 * 1024;
716
717 if (wbc->for_kupdate)
718 return 0;
719
720 /* this is a bit racy, but that's ok */
721 num_dirty = root->fs_info->dirty_metadata_bytes;
722 if (num_dirty < thresh)
723 return 0;
724 }
725 return extent_writepages(tree, mapping, btree_get_extent, wbc);
726}
727
728static int btree_readpage(struct file *file, struct page *page)
729{
730 struct extent_io_tree *tree;
731 tree = &BTRFS_I(page->mapping->host)->io_tree;
732 return extent_read_full_page(tree, page, btree_get_extent);
733}
734
735static int btree_releasepage(struct page *page, gfp_t gfp_flags)
736{
737 struct extent_io_tree *tree;
738 struct extent_map_tree *map;
739 int ret;
740
741 if (PageWriteback(page) || PageDirty(page))
742 return 0;
743
744 tree = &BTRFS_I(page->mapping->host)->io_tree;
745 map = &BTRFS_I(page->mapping->host)->extent_tree;
746
747 ret = try_release_extent_state(map, tree, page, gfp_flags);
748 if (!ret)
749 return 0;
750
751 ret = try_release_extent_buffer(tree, page);
752 if (ret == 1) {
753 ClearPagePrivate(page);
754 set_page_private(page, 0);
755 page_cache_release(page);
756 }
757
758 return ret;
759}
760
761static void btree_invalidatepage(struct page *page, unsigned long offset)
762{
763 struct extent_io_tree *tree;
764 tree = &BTRFS_I(page->mapping->host)->io_tree;
765 extent_invalidatepage(tree, page, offset);
766 btree_releasepage(page, GFP_NOFS);
767 if (PagePrivate(page)) {
768 printk(KERN_WARNING "btrfs warning page private not zero "
769 "on page %llu\n", (unsigned long long)page_offset(page));
770 ClearPagePrivate(page);
771 set_page_private(page, 0);
772 page_cache_release(page);
773 }
774}
775
776static struct address_space_operations btree_aops = {
777 .readpage = btree_readpage,
778 .writepage = btree_writepage,
779 .writepages = btree_writepages,
780 .releasepage = btree_releasepage,
781 .invalidatepage = btree_invalidatepage,
782 .sync_page = block_sync_page,
783};
784
785int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
786 u64 parent_transid)
787{
788 struct extent_buffer *buf = NULL;
789 struct inode *btree_inode = root->fs_info->btree_inode;
790 int ret = 0;
791
792 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
793 if (!buf)
794 return 0;
795 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
796 buf, 0, 0, btree_get_extent, 0);
797 free_extent_buffer(buf);
798 return ret;
799}
800
801struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
802 u64 bytenr, u32 blocksize)
803{
804 struct inode *btree_inode = root->fs_info->btree_inode;
805 struct extent_buffer *eb;
806 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
807 bytenr, blocksize, GFP_NOFS);
808 return eb;
809}
810
811struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
812 u64 bytenr, u32 blocksize)
813{
814 struct inode *btree_inode = root->fs_info->btree_inode;
815 struct extent_buffer *eb;
816
817 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
818 bytenr, blocksize, NULL, GFP_NOFS);
819 return eb;
820}
821
822
823int btrfs_write_tree_block(struct extent_buffer *buf)
824{
825 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
826 buf->start + buf->len - 1, WB_SYNC_ALL);
827}
828
829int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
830{
831 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
832 buf->start, buf->start + buf->len - 1);
833}
834
835struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
836 u32 blocksize, u64 parent_transid)
837{
838 struct extent_buffer *buf = NULL;
839 struct inode *btree_inode = root->fs_info->btree_inode;
840 struct extent_io_tree *io_tree;
841 int ret;
842
843 io_tree = &BTRFS_I(btree_inode)->io_tree;
844
845 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
846 if (!buf)
847 return NULL;
848
849 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
850
851 if (ret == 0)
852 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
853 return buf;
854
855}
856
857int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
858 struct extent_buffer *buf)
859{
860 struct inode *btree_inode = root->fs_info->btree_inode;
861 if (btrfs_header_generation(buf) ==
862 root->fs_info->running_transaction->transid) {
863 btrfs_assert_tree_locked(buf);
864
865 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
866 spin_lock(&root->fs_info->delalloc_lock);
867 if (root->fs_info->dirty_metadata_bytes >= buf->len)
868 root->fs_info->dirty_metadata_bytes -= buf->len;
869 else
870 WARN_ON(1);
871 spin_unlock(&root->fs_info->delalloc_lock);
872 }
873
874 /* ugh, clear_extent_buffer_dirty needs to lock the page */
875 btrfs_set_lock_blocking(buf);
876 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
877 buf);
878 }
879 return 0;
880}
881
882static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
883 u32 stripesize, struct btrfs_root *root,
884 struct btrfs_fs_info *fs_info,
885 u64 objectid)
886{
887 root->node = NULL;
888 root->commit_root = NULL;
889 root->sectorsize = sectorsize;
890 root->nodesize = nodesize;
891 root->leafsize = leafsize;
892 root->stripesize = stripesize;
893 root->ref_cows = 0;
894 root->track_dirty = 0;
895
896 root->fs_info = fs_info;
897 root->objectid = objectid;
898 root->last_trans = 0;
899 root->highest_objectid = 0;
900 root->name = NULL;
901 root->in_sysfs = 0;
902 root->inode_tree.rb_node = NULL;
903
904 INIT_LIST_HEAD(&root->dirty_list);
905 INIT_LIST_HEAD(&root->orphan_list);
906 INIT_LIST_HEAD(&root->root_list);
907 spin_lock_init(&root->node_lock);
908 spin_lock_init(&root->list_lock);
909 spin_lock_init(&root->inode_lock);
910 mutex_init(&root->objectid_mutex);
911 mutex_init(&root->log_mutex);
912 init_waitqueue_head(&root->log_writer_wait);
913 init_waitqueue_head(&root->log_commit_wait[0]);
914 init_waitqueue_head(&root->log_commit_wait[1]);
915 atomic_set(&root->log_commit[0], 0);
916 atomic_set(&root->log_commit[1], 0);
917 atomic_set(&root->log_writers, 0);
918 root->log_batch = 0;
919 root->log_transid = 0;
920 extent_io_tree_init(&root->dirty_log_pages,
921 fs_info->btree_inode->i_mapping, GFP_NOFS);
922
923 memset(&root->root_key, 0, sizeof(root->root_key));
924 memset(&root->root_item, 0, sizeof(root->root_item));
925 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
926 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
927 root->defrag_trans_start = fs_info->generation;
928 init_completion(&root->kobj_unregister);
929 root->defrag_running = 0;
930 root->defrag_level = 0;
931 root->root_key.objectid = objectid;
932 root->anon_super.s_root = NULL;
933 root->anon_super.s_dev = 0;
934 INIT_LIST_HEAD(&root->anon_super.s_list);
935 INIT_LIST_HEAD(&root->anon_super.s_instances);
936 init_rwsem(&root->anon_super.s_umount);
937
938 return 0;
939}
940
941static int find_and_setup_root(struct btrfs_root *tree_root,
942 struct btrfs_fs_info *fs_info,
943 u64 objectid,
944 struct btrfs_root *root)
945{
946 int ret;
947 u32 blocksize;
948 u64 generation;
949
950 __setup_root(tree_root->nodesize, tree_root->leafsize,
951 tree_root->sectorsize, tree_root->stripesize,
952 root, fs_info, objectid);
953 ret = btrfs_find_last_root(tree_root, objectid,
954 &root->root_item, &root->root_key);
955 if (ret > 0)
956 return -ENOENT;
957 BUG_ON(ret);
958
959 generation = btrfs_root_generation(&root->root_item);
960 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
961 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
962 blocksize, generation);
963 BUG_ON(!root->node);
964 root->commit_root = btrfs_root_node(root);
965 return 0;
966}
967
968int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
969 struct btrfs_fs_info *fs_info)
970{
971 struct extent_buffer *eb;
972 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
973 u64 start = 0;
974 u64 end = 0;
975 int ret;
976
977 if (!log_root_tree)
978 return 0;
979
980 while (1) {
981 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
982 0, &start, &end, EXTENT_DIRTY);
983 if (ret)
984 break;
985
986 clear_extent_dirty(&log_root_tree->dirty_log_pages,
987 start, end, GFP_NOFS);
988 }
989 eb = fs_info->log_root_tree->node;
990
991 WARN_ON(btrfs_header_level(eb) != 0);
992 WARN_ON(btrfs_header_nritems(eb) != 0);
993
994 ret = btrfs_free_reserved_extent(fs_info->tree_root,
995 eb->start, eb->len);
996 BUG_ON(ret);
997
998 free_extent_buffer(eb);
999 kfree(fs_info->log_root_tree);
1000 fs_info->log_root_tree = NULL;
1001 return 0;
1002}
1003
1004static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1005 struct btrfs_fs_info *fs_info)
1006{
1007 struct btrfs_root *root;
1008 struct btrfs_root *tree_root = fs_info->tree_root;
1009 struct extent_buffer *leaf;
1010
1011 root = kzalloc(sizeof(*root), GFP_NOFS);
1012 if (!root)
1013 return ERR_PTR(-ENOMEM);
1014
1015 __setup_root(tree_root->nodesize, tree_root->leafsize,
1016 tree_root->sectorsize, tree_root->stripesize,
1017 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1018
1019 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1020 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1021 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1022 /*
1023 * log trees do not get reference counted because they go away
1024 * before a real commit is actually done. They do store pointers
1025 * to file data extents, and those reference counts still get
1026 * updated (along with back refs to the log tree).
1027 */
1028 root->ref_cows = 0;
1029
1030 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1031 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1032 if (IS_ERR(leaf)) {
1033 kfree(root);
1034 return ERR_CAST(leaf);
1035 }
1036
1037 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1038 btrfs_set_header_bytenr(leaf, leaf->start);
1039 btrfs_set_header_generation(leaf, trans->transid);
1040 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1041 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1042 root->node = leaf;
1043
1044 write_extent_buffer(root->node, root->fs_info->fsid,
1045 (unsigned long)btrfs_header_fsid(root->node),
1046 BTRFS_FSID_SIZE);
1047 btrfs_mark_buffer_dirty(root->node);
1048 btrfs_tree_unlock(root->node);
1049 return root;
1050}
1051
1052int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1053 struct btrfs_fs_info *fs_info)
1054{
1055 struct btrfs_root *log_root;
1056
1057 log_root = alloc_log_tree(trans, fs_info);
1058 if (IS_ERR(log_root))
1059 return PTR_ERR(log_root);
1060 WARN_ON(fs_info->log_root_tree);
1061 fs_info->log_root_tree = log_root;
1062 return 0;
1063}
1064
1065int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1066 struct btrfs_root *root)
1067{
1068 struct btrfs_root *log_root;
1069 struct btrfs_inode_item *inode_item;
1070
1071 log_root = alloc_log_tree(trans, root->fs_info);
1072 if (IS_ERR(log_root))
1073 return PTR_ERR(log_root);
1074
1075 log_root->last_trans = trans->transid;
1076 log_root->root_key.offset = root->root_key.objectid;
1077
1078 inode_item = &log_root->root_item.inode;
1079 inode_item->generation = cpu_to_le64(1);
1080 inode_item->size = cpu_to_le64(3);
1081 inode_item->nlink = cpu_to_le32(1);
1082 inode_item->nbytes = cpu_to_le64(root->leafsize);
1083 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1084
1085 btrfs_set_root_node(&log_root->root_item, log_root->node);
1086
1087 WARN_ON(root->log_root);
1088 root->log_root = log_root;
1089 root->log_transid = 0;
1090 return 0;
1091}
1092
1093struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1094 struct btrfs_key *location)
1095{
1096 struct btrfs_root *root;
1097 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1098 struct btrfs_path *path;
1099 struct extent_buffer *l;
1100 u64 generation;
1101 u32 blocksize;
1102 int ret = 0;
1103
1104 root = kzalloc(sizeof(*root), GFP_NOFS);
1105 if (!root)
1106 return ERR_PTR(-ENOMEM);
1107 if (location->offset == (u64)-1) {
1108 ret = find_and_setup_root(tree_root, fs_info,
1109 location->objectid, root);
1110 if (ret) {
1111 kfree(root);
1112 return ERR_PTR(ret);
1113 }
1114 goto out;
1115 }
1116
1117 __setup_root(tree_root->nodesize, tree_root->leafsize,
1118 tree_root->sectorsize, tree_root->stripesize,
1119 root, fs_info, location->objectid);
1120
1121 path = btrfs_alloc_path();
1122 BUG_ON(!path);
1123 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1124 if (ret == 0) {
1125 l = path->nodes[0];
1126 read_extent_buffer(l, &root->root_item,
1127 btrfs_item_ptr_offset(l, path->slots[0]),
1128 sizeof(root->root_item));
1129 memcpy(&root->root_key, location, sizeof(*location));
1130 }
1131 btrfs_free_path(path);
1132 if (ret) {
1133 if (ret > 0)
1134 ret = -ENOENT;
1135 return ERR_PTR(ret);
1136 }
1137
1138 generation = btrfs_root_generation(&root->root_item);
1139 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1140 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1141 blocksize, generation);
1142 root->commit_root = btrfs_root_node(root);
1143 BUG_ON(!root->node);
1144out:
1145 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1146 root->ref_cows = 1;
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;
1182again:
1183 spin_lock(&fs_info->fs_roots_radix_lock);
1184 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1185 (unsigned long)location->objectid);
1186 spin_unlock(&fs_info->fs_roots_radix_lock);
1187 if (root)
1188 return root;
1189
1190 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1191 if (ret == 0)
1192 ret = -ENOENT;
1193 if (ret < 0)
1194 return ERR_PTR(ret);
1195
1196 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1197 if (IS_ERR(root))
1198 return root;
1199
1200 WARN_ON(btrfs_root_refs(&root->root_item) == 0);
1201 set_anon_super(&root->anon_super, NULL);
1202
1203 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1204 if (ret)
1205 goto fail;
1206
1207 spin_lock(&fs_info->fs_roots_radix_lock);
1208 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1209 (unsigned long)root->root_key.objectid,
1210 root);
1211 if (ret == 0)
1212 root->in_radix = 1;
1213 spin_unlock(&fs_info->fs_roots_radix_lock);
1214 radix_tree_preload_end();
1215 if (ret) {
1216 if (ret == -EEXIST) {
1217 free_fs_root(root);
1218 goto again;
1219 }
1220 goto fail;
1221 }
1222
1223 ret = btrfs_find_dead_roots(fs_info->tree_root,
1224 root->root_key.objectid);
1225 WARN_ON(ret);
1226
1227 if (!(fs_info->sb->s_flags & MS_RDONLY))
1228 btrfs_orphan_cleanup(root);
1229
1230 return root;
1231fail:
1232 free_fs_root(root);
1233 return ERR_PTR(ret);
1234}
1235
1236struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1237 struct btrfs_key *location,
1238 const char *name, int namelen)
1239{
1240 return btrfs_read_fs_root_no_name(fs_info, location);
1241#if 0
1242 struct btrfs_root *root;
1243 int ret;
1244
1245 root = btrfs_read_fs_root_no_name(fs_info, location);
1246 if (!root)
1247 return NULL;
1248
1249 if (root->in_sysfs)
1250 return root;
1251
1252 ret = btrfs_set_root_name(root, name, namelen);
1253 if (ret) {
1254 free_extent_buffer(root->node);
1255 kfree(root);
1256 return ERR_PTR(ret);
1257 }
1258
1259 ret = btrfs_sysfs_add_root(root);
1260 if (ret) {
1261 free_extent_buffer(root->node);
1262 kfree(root->name);
1263 kfree(root);
1264 return ERR_PTR(ret);
1265 }
1266 root->in_sysfs = 1;
1267 return root;
1268#endif
1269}
1270
1271static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1272{
1273 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1274 int ret = 0;
1275 struct btrfs_device *device;
1276 struct backing_dev_info *bdi;
1277
1278 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1279 if (!device->bdev)
1280 continue;
1281 bdi = blk_get_backing_dev_info(device->bdev);
1282 if (bdi && bdi_congested(bdi, bdi_bits)) {
1283 ret = 1;
1284 break;
1285 }
1286 }
1287 return ret;
1288}
1289
1290/*
1291 * this unplugs every device on the box, and it is only used when page
1292 * is null
1293 */
1294static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1295{
1296 struct btrfs_device *device;
1297 struct btrfs_fs_info *info;
1298
1299 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1300 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1301 if (!device->bdev)
1302 continue;
1303
1304 bdi = blk_get_backing_dev_info(device->bdev);
1305 if (bdi->unplug_io_fn)
1306 bdi->unplug_io_fn(bdi, page);
1307 }
1308}
1309
1310static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1311{
1312 struct inode *inode;
1313 struct extent_map_tree *em_tree;
1314 struct extent_map *em;
1315 struct address_space *mapping;
1316 u64 offset;
1317
1318 /* the generic O_DIRECT read code does this */
1319 if (1 || !page) {
1320 __unplug_io_fn(bdi, page);
1321 return;
1322 }
1323
1324 /*
1325 * page->mapping may change at any time. Get a consistent copy
1326 * and use that for everything below
1327 */
1328 smp_mb();
1329 mapping = page->mapping;
1330 if (!mapping)
1331 return;
1332
1333 inode = mapping->host;
1334
1335 /*
1336 * don't do the expensive searching for a small number of
1337 * devices
1338 */
1339 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1340 __unplug_io_fn(bdi, page);
1341 return;
1342 }
1343
1344 offset = page_offset(page);
1345
1346 em_tree = &BTRFS_I(inode)->extent_tree;
1347 read_lock(&em_tree->lock);
1348 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1349 read_unlock(&em_tree->lock);
1350 if (!em) {
1351 __unplug_io_fn(bdi, page);
1352 return;
1353 }
1354
1355 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1356 free_extent_map(em);
1357 __unplug_io_fn(bdi, page);
1358 return;
1359 }
1360 offset = offset - em->start;
1361 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1362 em->block_start + offset, page);
1363 free_extent_map(em);
1364}
1365
1366/*
1367 * If this fails, caller must call bdi_destroy() to get rid of the
1368 * bdi again.
1369 */
1370static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1371{
1372 int err;
1373
1374 bdi->capabilities = BDI_CAP_MAP_COPY;
1375 err = bdi_init(bdi);
1376 if (err)
1377 return err;
1378
1379 err = bdi_register(bdi, NULL, "btrfs-%d",
1380 atomic_inc_return(&btrfs_bdi_num));
1381 if (err)
1382 return err;
1383
1384 bdi->ra_pages = default_backing_dev_info.ra_pages;
1385 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1386 bdi->unplug_io_data = info;
1387 bdi->congested_fn = btrfs_congested_fn;
1388 bdi->congested_data = info;
1389 return 0;
1390}
1391
1392static int bio_ready_for_csum(struct bio *bio)
1393{
1394 u64 length = 0;
1395 u64 buf_len = 0;
1396 u64 start = 0;
1397 struct page *page;
1398 struct extent_io_tree *io_tree = NULL;
1399 struct btrfs_fs_info *info = NULL;
1400 struct bio_vec *bvec;
1401 int i;
1402 int ret;
1403
1404 bio_for_each_segment(bvec, bio, i) {
1405 page = bvec->bv_page;
1406 if (page->private == EXTENT_PAGE_PRIVATE) {
1407 length += bvec->bv_len;
1408 continue;
1409 }
1410 if (!page->private) {
1411 length += bvec->bv_len;
1412 continue;
1413 }
1414 length = bvec->bv_len;
1415 buf_len = page->private >> 2;
1416 start = page_offset(page) + bvec->bv_offset;
1417 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1418 info = BTRFS_I(page->mapping->host)->root->fs_info;
1419 }
1420 /* are we fully contained in this bio? */
1421 if (buf_len <= length)
1422 return 1;
1423
1424 ret = extent_range_uptodate(io_tree, start + length,
1425 start + buf_len - 1);
1426 return ret;
1427}
1428
1429/*
1430 * called by the kthread helper functions to finally call the bio end_io
1431 * functions. This is where read checksum verification actually happens
1432 */
1433static void end_workqueue_fn(struct btrfs_work *work)
1434{
1435 struct bio *bio;
1436 struct end_io_wq *end_io_wq;
1437 struct btrfs_fs_info *fs_info;
1438 int error;
1439
1440 end_io_wq = container_of(work, struct end_io_wq, work);
1441 bio = end_io_wq->bio;
1442 fs_info = end_io_wq->info;
1443
1444 /* metadata bio reads are special because the whole tree block must
1445 * be checksummed at once. This makes sure the entire block is in
1446 * ram and up to date before trying to verify things. For
1447 * blocksize <= pagesize, it is basically a noop
1448 */
1449 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1450 !bio_ready_for_csum(bio)) {
1451 btrfs_queue_worker(&fs_info->endio_meta_workers,
1452 &end_io_wq->work);
1453 return;
1454 }
1455 error = end_io_wq->error;
1456 bio->bi_private = end_io_wq->private;
1457 bio->bi_end_io = end_io_wq->end_io;
1458 kfree(end_io_wq);
1459 bio_endio(bio, error);
1460}
1461
1462static int cleaner_kthread(void *arg)
1463{
1464 struct btrfs_root *root = arg;
1465
1466 do {
1467 smp_mb();
1468 if (root->fs_info->closing)
1469 break;
1470
1471 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1472 mutex_lock(&root->fs_info->cleaner_mutex);
1473 btrfs_clean_old_snapshots(root);
1474 mutex_unlock(&root->fs_info->cleaner_mutex);
1475
1476 if (freezing(current)) {
1477 refrigerator();
1478 } else {
1479 smp_mb();
1480 if (root->fs_info->closing)
1481 break;
1482 set_current_state(TASK_INTERRUPTIBLE);
1483 schedule();
1484 __set_current_state(TASK_RUNNING);
1485 }
1486 } while (!kthread_should_stop());
1487 return 0;
1488}
1489
1490static int transaction_kthread(void *arg)
1491{
1492 struct btrfs_root *root = arg;
1493 struct btrfs_trans_handle *trans;
1494 struct btrfs_transaction *cur;
1495 unsigned long now;
1496 unsigned long delay;
1497 int ret;
1498
1499 do {
1500 smp_mb();
1501 if (root->fs_info->closing)
1502 break;
1503
1504 delay = HZ * 30;
1505 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1506 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1507
1508 mutex_lock(&root->fs_info->trans_mutex);
1509 cur = root->fs_info->running_transaction;
1510 if (!cur) {
1511 mutex_unlock(&root->fs_info->trans_mutex);
1512 goto sleep;
1513 }
1514
1515 now = get_seconds();
1516 if (now < cur->start_time || now - cur->start_time < 30) {
1517 mutex_unlock(&root->fs_info->trans_mutex);
1518 delay = HZ * 5;
1519 goto sleep;
1520 }
1521 mutex_unlock(&root->fs_info->trans_mutex);
1522 trans = btrfs_start_transaction(root, 1);
1523 ret = btrfs_commit_transaction(trans, root);
1524
1525sleep:
1526 wake_up_process(root->fs_info->cleaner_kthread);
1527 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1528
1529 if (freezing(current)) {
1530 refrigerator();
1531 } else {
1532 if (root->fs_info->closing)
1533 break;
1534 set_current_state(TASK_INTERRUPTIBLE);
1535 schedule_timeout(delay);
1536 __set_current_state(TASK_RUNNING);
1537 }
1538 } while (!kthread_should_stop());
1539 return 0;
1540}
1541
1542struct btrfs_root *open_ctree(struct super_block *sb,
1543 struct btrfs_fs_devices *fs_devices,
1544 char *options)
1545{
1546 u32 sectorsize;
1547 u32 nodesize;
1548 u32 leafsize;
1549 u32 blocksize;
1550 u32 stripesize;
1551 u64 generation;
1552 u64 features;
1553 struct btrfs_key location;
1554 struct buffer_head *bh;
1555 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1556 GFP_NOFS);
1557 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1558 GFP_NOFS);
1559 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1560 GFP_NOFS);
1561 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1562 GFP_NOFS);
1563 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1564 GFP_NOFS);
1565 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1566 GFP_NOFS);
1567 struct btrfs_root *log_tree_root;
1568
1569 int ret;
1570 int err = -EINVAL;
1571
1572 struct btrfs_super_block *disk_super;
1573
1574 if (!extent_root || !tree_root || !fs_info ||
1575 !chunk_root || !dev_root || !csum_root) {
1576 err = -ENOMEM;
1577 goto fail;
1578 }
1579 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1580 INIT_LIST_HEAD(&fs_info->trans_list);
1581 INIT_LIST_HEAD(&fs_info->dead_roots);
1582 INIT_LIST_HEAD(&fs_info->hashers);
1583 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1584 INIT_LIST_HEAD(&fs_info->ordered_operations);
1585 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1586 spin_lock_init(&fs_info->delalloc_lock);
1587 spin_lock_init(&fs_info->new_trans_lock);
1588 spin_lock_init(&fs_info->ref_cache_lock);
1589
1590 init_completion(&fs_info->kobj_unregister);
1591 fs_info->tree_root = tree_root;
1592 fs_info->extent_root = extent_root;
1593 fs_info->csum_root = csum_root;
1594 fs_info->chunk_root = chunk_root;
1595 fs_info->dev_root = dev_root;
1596 fs_info->fs_devices = fs_devices;
1597 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1598 INIT_LIST_HEAD(&fs_info->space_info);
1599 btrfs_mapping_init(&fs_info->mapping_tree);
1600 atomic_set(&fs_info->nr_async_submits, 0);
1601 atomic_set(&fs_info->async_delalloc_pages, 0);
1602 atomic_set(&fs_info->async_submit_draining, 0);
1603 atomic_set(&fs_info->nr_async_bios, 0);
1604 fs_info->sb = sb;
1605 fs_info->max_extent = (u64)-1;
1606 fs_info->max_inline = 8192 * 1024;
1607 if (setup_bdi(fs_info, &fs_info->bdi))
1608 goto fail_bdi;
1609 fs_info->btree_inode = new_inode(sb);
1610 fs_info->btree_inode->i_ino = 1;
1611 fs_info->btree_inode->i_nlink = 1;
1612 fs_info->metadata_ratio = 8;
1613
1614 fs_info->thread_pool_size = min_t(unsigned long,
1615 num_online_cpus() + 2, 8);
1616
1617 INIT_LIST_HEAD(&fs_info->ordered_extents);
1618 spin_lock_init(&fs_info->ordered_extent_lock);
1619
1620 sb->s_blocksize = 4096;
1621 sb->s_blocksize_bits = blksize_bits(4096);
1622
1623 /*
1624 * we set the i_size on the btree inode to the max possible int.
1625 * the real end of the address space is determined by all of
1626 * the devices in the system
1627 */
1628 fs_info->btree_inode->i_size = OFFSET_MAX;
1629 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1630 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1631
1632 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1633 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1634 fs_info->btree_inode->i_mapping,
1635 GFP_NOFS);
1636 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1637 GFP_NOFS);
1638
1639 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1640
1641 spin_lock_init(&fs_info->block_group_cache_lock);
1642 fs_info->block_group_cache_tree.rb_node = NULL;
1643
1644 extent_io_tree_init(&fs_info->freed_extents[0],
1645 fs_info->btree_inode->i_mapping, GFP_NOFS);
1646 extent_io_tree_init(&fs_info->freed_extents[1],
1647 fs_info->btree_inode->i_mapping, GFP_NOFS);
1648 fs_info->pinned_extents = &fs_info->freed_extents[0];
1649 fs_info->do_barriers = 1;
1650
1651 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1652 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1653 sizeof(struct btrfs_key));
1654 insert_inode_hash(fs_info->btree_inode);
1655
1656 mutex_init(&fs_info->trans_mutex);
1657 mutex_init(&fs_info->ordered_operations_mutex);
1658 mutex_init(&fs_info->tree_log_mutex);
1659 mutex_init(&fs_info->drop_mutex);
1660 mutex_init(&fs_info->chunk_mutex);
1661 mutex_init(&fs_info->transaction_kthread_mutex);
1662 mutex_init(&fs_info->cleaner_mutex);
1663 mutex_init(&fs_info->volume_mutex);
1664 mutex_init(&fs_info->tree_reloc_mutex);
1665 init_rwsem(&fs_info->extent_commit_sem);
1666
1667 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1668 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1669
1670 init_waitqueue_head(&fs_info->transaction_throttle);
1671 init_waitqueue_head(&fs_info->transaction_wait);
1672 init_waitqueue_head(&fs_info->async_submit_wait);
1673
1674 __setup_root(4096, 4096, 4096, 4096, tree_root,
1675 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1676
1677
1678 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1679 if (!bh)
1680 goto fail_iput;
1681
1682 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1683 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1684 sizeof(fs_info->super_for_commit));
1685 brelse(bh);
1686
1687 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1688
1689 disk_super = &fs_info->super_copy;
1690 if (!btrfs_super_root(disk_super))
1691 goto fail_iput;
1692
1693 ret = btrfs_parse_options(tree_root, options);
1694 if (ret) {
1695 err = ret;
1696 goto fail_iput;
1697 }
1698
1699 features = btrfs_super_incompat_flags(disk_super) &
1700 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1701 if (features) {
1702 printk(KERN_ERR "BTRFS: couldn't mount because of "
1703 "unsupported optional features (%Lx).\n",
1704 (unsigned long long)features);
1705 err = -EINVAL;
1706 goto fail_iput;
1707 }
1708
1709 features = btrfs_super_incompat_flags(disk_super);
1710 if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1711 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1712 btrfs_set_super_incompat_flags(disk_super, features);
1713 }
1714
1715 features = btrfs_super_compat_ro_flags(disk_super) &
1716 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1717 if (!(sb->s_flags & MS_RDONLY) && features) {
1718 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1719 "unsupported option features (%Lx).\n",
1720 (unsigned long long)features);
1721 err = -EINVAL;
1722 goto fail_iput;
1723 }
1724printk("thread pool is %d\n", fs_info->thread_pool_size);
1725 /*
1726 * we need to start all the end_io workers up front because the
1727 * queue work function gets called at interrupt time, and so it
1728 * cannot dynamically grow.
1729 */
1730 btrfs_init_workers(&fs_info->workers, "worker",
1731 fs_info->thread_pool_size);
1732
1733 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1734 fs_info->thread_pool_size);
1735
1736 btrfs_init_workers(&fs_info->submit_workers, "submit",
1737 min_t(u64, fs_devices->num_devices,
1738 fs_info->thread_pool_size));
1739
1740 /* a higher idle thresh on the submit workers makes it much more
1741 * likely that bios will be send down in a sane order to the
1742 * devices
1743 */
1744 fs_info->submit_workers.idle_thresh = 64;
1745
1746 fs_info->workers.idle_thresh = 16;
1747 fs_info->workers.ordered = 1;
1748
1749 fs_info->delalloc_workers.idle_thresh = 2;
1750 fs_info->delalloc_workers.ordered = 1;
1751
1752 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1753 btrfs_init_workers(&fs_info->endio_workers, "endio",
1754 fs_info->thread_pool_size);
1755 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1756 fs_info->thread_pool_size);
1757 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1758 "endio-meta-write", fs_info->thread_pool_size);
1759 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1760 fs_info->thread_pool_size);
1761
1762 /*
1763 * endios are largely parallel and should have a very
1764 * low idle thresh
1765 */
1766 fs_info->endio_workers.idle_thresh = 4;
1767 fs_info->endio_meta_workers.idle_thresh = 4;
1768
1769 fs_info->endio_write_workers.idle_thresh = 2;
1770 fs_info->endio_meta_write_workers.idle_thresh = 2;
1771
1772 fs_info->endio_workers.atomic_worker_start = 1;
1773 fs_info->endio_meta_workers.atomic_worker_start = 1;
1774 fs_info->endio_write_workers.atomic_worker_start = 1;
1775 fs_info->endio_meta_write_workers.atomic_worker_start = 1;
1776
1777 btrfs_start_workers(&fs_info->workers, 1);
1778 btrfs_start_workers(&fs_info->submit_workers, 1);
1779 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1780 btrfs_start_workers(&fs_info->fixup_workers, 1);
1781 btrfs_start_workers(&fs_info->endio_workers, 1);
1782 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1783 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1784 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1785
1786 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1787 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1788 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1789
1790 nodesize = btrfs_super_nodesize(disk_super);
1791 leafsize = btrfs_super_leafsize(disk_super);
1792 sectorsize = btrfs_super_sectorsize(disk_super);
1793 stripesize = btrfs_super_stripesize(disk_super);
1794 tree_root->nodesize = nodesize;
1795 tree_root->leafsize = leafsize;
1796 tree_root->sectorsize = sectorsize;
1797 tree_root->stripesize = stripesize;
1798
1799 sb->s_blocksize = sectorsize;
1800 sb->s_blocksize_bits = blksize_bits(sectorsize);
1801
1802 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1803 sizeof(disk_super->magic))) {
1804 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1805 goto fail_sb_buffer;
1806 }
1807
1808 mutex_lock(&fs_info->chunk_mutex);
1809 ret = btrfs_read_sys_array(tree_root);
1810 mutex_unlock(&fs_info->chunk_mutex);
1811 if (ret) {
1812 printk(KERN_WARNING "btrfs: failed to read the system "
1813 "array on %s\n", sb->s_id);
1814 goto fail_sb_buffer;
1815 }
1816
1817 blocksize = btrfs_level_size(tree_root,
1818 btrfs_super_chunk_root_level(disk_super));
1819 generation = btrfs_super_chunk_root_generation(disk_super);
1820
1821 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1822 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1823
1824 chunk_root->node = read_tree_block(chunk_root,
1825 btrfs_super_chunk_root(disk_super),
1826 blocksize, generation);
1827 BUG_ON(!chunk_root->node);
1828 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1829 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1830 sb->s_id);
1831 goto fail_chunk_root;
1832 }
1833 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1834 chunk_root->commit_root = btrfs_root_node(chunk_root);
1835
1836 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1837 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1838 BTRFS_UUID_SIZE);
1839
1840 mutex_lock(&fs_info->chunk_mutex);
1841 ret = btrfs_read_chunk_tree(chunk_root);
1842 mutex_unlock(&fs_info->chunk_mutex);
1843 if (ret) {
1844 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1845 sb->s_id);
1846 goto fail_chunk_root;
1847 }
1848
1849 btrfs_close_extra_devices(fs_devices);
1850
1851 blocksize = btrfs_level_size(tree_root,
1852 btrfs_super_root_level(disk_super));
1853 generation = btrfs_super_generation(disk_super);
1854
1855 tree_root->node = read_tree_block(tree_root,
1856 btrfs_super_root(disk_super),
1857 blocksize, generation);
1858 if (!tree_root->node)
1859 goto fail_chunk_root;
1860 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1861 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1862 sb->s_id);
1863 goto fail_tree_root;
1864 }
1865 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1866 tree_root->commit_root = btrfs_root_node(tree_root);
1867
1868 ret = find_and_setup_root(tree_root, fs_info,
1869 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1870 if (ret)
1871 goto fail_tree_root;
1872 extent_root->track_dirty = 1;
1873
1874 ret = find_and_setup_root(tree_root, fs_info,
1875 BTRFS_DEV_TREE_OBJECTID, dev_root);
1876 if (ret)
1877 goto fail_extent_root;
1878 dev_root->track_dirty = 1;
1879
1880 ret = find_and_setup_root(tree_root, fs_info,
1881 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1882 if (ret)
1883 goto fail_dev_root;
1884
1885 csum_root->track_dirty = 1;
1886
1887 btrfs_read_block_groups(extent_root);
1888
1889 fs_info->generation = generation;
1890 fs_info->last_trans_committed = generation;
1891 fs_info->data_alloc_profile = (u64)-1;
1892 fs_info->metadata_alloc_profile = (u64)-1;
1893 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1894 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1895 "btrfs-cleaner");
1896 if (IS_ERR(fs_info->cleaner_kthread))
1897 goto fail_csum_root;
1898
1899 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1900 tree_root,
1901 "btrfs-transaction");
1902 if (IS_ERR(fs_info->transaction_kthread))
1903 goto fail_cleaner;
1904
1905 if (!btrfs_test_opt(tree_root, SSD) &&
1906 !btrfs_test_opt(tree_root, NOSSD) &&
1907 !fs_info->fs_devices->rotating) {
1908 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1909 "mode\n");
1910 btrfs_set_opt(fs_info->mount_opt, SSD);
1911 }
1912
1913 if (btrfs_super_log_root(disk_super) != 0) {
1914 u64 bytenr = btrfs_super_log_root(disk_super);
1915
1916 if (fs_devices->rw_devices == 0) {
1917 printk(KERN_WARNING "Btrfs log replay required "
1918 "on RO media\n");
1919 err = -EIO;
1920 goto fail_trans_kthread;
1921 }
1922 blocksize =
1923 btrfs_level_size(tree_root,
1924 btrfs_super_log_root_level(disk_super));
1925
1926 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1927 GFP_NOFS);
1928
1929 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1930 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1931
1932 log_tree_root->node = read_tree_block(tree_root, bytenr,
1933 blocksize,
1934 generation + 1);
1935 ret = btrfs_recover_log_trees(log_tree_root);
1936 BUG_ON(ret);
1937
1938 if (sb->s_flags & MS_RDONLY) {
1939 ret = btrfs_commit_super(tree_root);
1940 BUG_ON(ret);
1941 }
1942 }
1943
1944 if (!(sb->s_flags & MS_RDONLY)) {
1945 ret = btrfs_recover_relocation(tree_root);
1946 BUG_ON(ret);
1947 }
1948
1949 location.objectid = BTRFS_FS_TREE_OBJECTID;
1950 location.type = BTRFS_ROOT_ITEM_KEY;
1951 location.offset = (u64)-1;
1952
1953 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1954 if (!fs_info->fs_root)
1955 goto fail_trans_kthread;
1956
1957 return tree_root;
1958
1959fail_trans_kthread:
1960 kthread_stop(fs_info->transaction_kthread);
1961fail_cleaner:
1962 kthread_stop(fs_info->cleaner_kthread);
1963
1964 /*
1965 * make sure we're done with the btree inode before we stop our
1966 * kthreads
1967 */
1968 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1969 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1970
1971fail_csum_root:
1972 free_extent_buffer(csum_root->node);
1973 free_extent_buffer(csum_root->commit_root);
1974fail_dev_root:
1975 free_extent_buffer(dev_root->node);
1976 free_extent_buffer(dev_root->commit_root);
1977fail_extent_root:
1978 free_extent_buffer(extent_root->node);
1979 free_extent_buffer(extent_root->commit_root);
1980fail_tree_root:
1981 free_extent_buffer(tree_root->node);
1982 free_extent_buffer(tree_root->commit_root);
1983fail_chunk_root:
1984 free_extent_buffer(chunk_root->node);
1985 free_extent_buffer(chunk_root->commit_root);
1986fail_sb_buffer:
1987 btrfs_stop_workers(&fs_info->fixup_workers);
1988 btrfs_stop_workers(&fs_info->delalloc_workers);
1989 btrfs_stop_workers(&fs_info->workers);
1990 btrfs_stop_workers(&fs_info->endio_workers);
1991 btrfs_stop_workers(&fs_info->endio_meta_workers);
1992 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1993 btrfs_stop_workers(&fs_info->endio_write_workers);
1994 btrfs_stop_workers(&fs_info->submit_workers);
1995fail_iput:
1996 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1997 iput(fs_info->btree_inode);
1998
1999 btrfs_close_devices(fs_info->fs_devices);
2000 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2001fail_bdi:
2002 bdi_destroy(&fs_info->bdi);
2003fail:
2004 kfree(extent_root);
2005 kfree(tree_root);
2006 kfree(fs_info);
2007 kfree(chunk_root);
2008 kfree(dev_root);
2009 kfree(csum_root);
2010 return ERR_PTR(err);
2011}
2012
2013static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2014{
2015 char b[BDEVNAME_SIZE];
2016
2017 if (uptodate) {
2018 set_buffer_uptodate(bh);
2019 } else {
2020 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2021 printk(KERN_WARNING "lost page write due to "
2022 "I/O error on %s\n",
2023 bdevname(bh->b_bdev, b));
2024 }
2025 /* note, we dont' set_buffer_write_io_error because we have
2026 * our own ways of dealing with the IO errors
2027 */
2028 clear_buffer_uptodate(bh);
2029 }
2030 unlock_buffer(bh);
2031 put_bh(bh);
2032}
2033
2034struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2035{
2036 struct buffer_head *bh;
2037 struct buffer_head *latest = NULL;
2038 struct btrfs_super_block *super;
2039 int i;
2040 u64 transid = 0;
2041 u64 bytenr;
2042
2043 /* we would like to check all the supers, but that would make
2044 * a btrfs mount succeed after a mkfs from a different FS.
2045 * So, we need to add a special mount option to scan for
2046 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2047 */
2048 for (i = 0; i < 1; i++) {
2049 bytenr = btrfs_sb_offset(i);
2050 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2051 break;
2052 bh = __bread(bdev, bytenr / 4096, 4096);
2053 if (!bh)
2054 continue;
2055
2056 super = (struct btrfs_super_block *)bh->b_data;
2057 if (btrfs_super_bytenr(super) != bytenr ||
2058 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2059 sizeof(super->magic))) {
2060 brelse(bh);
2061 continue;
2062 }
2063
2064 if (!latest || btrfs_super_generation(super) > transid) {
2065 brelse(latest);
2066 latest = bh;
2067 transid = btrfs_super_generation(super);
2068 } else {
2069 brelse(bh);
2070 }
2071 }
2072 return latest;
2073}
2074
2075/*
2076 * this should be called twice, once with wait == 0 and
2077 * once with wait == 1. When wait == 0 is done, all the buffer heads
2078 * we write are pinned.
2079 *
2080 * They are released when wait == 1 is done.
2081 * max_mirrors must be the same for both runs, and it indicates how
2082 * many supers on this one device should be written.
2083 *
2084 * max_mirrors == 0 means to write them all.
2085 */
2086static int write_dev_supers(struct btrfs_device *device,
2087 struct btrfs_super_block *sb,
2088 int do_barriers, int wait, int max_mirrors)
2089{
2090 struct buffer_head *bh;
2091 int i;
2092 int ret;
2093 int errors = 0;
2094 u32 crc;
2095 u64 bytenr;
2096 int last_barrier = 0;
2097
2098 if (max_mirrors == 0)
2099 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2100
2101 /* make sure only the last submit_bh does a barrier */
2102 if (do_barriers) {
2103 for (i = 0; i < max_mirrors; i++) {
2104 bytenr = btrfs_sb_offset(i);
2105 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2106 device->total_bytes)
2107 break;
2108 last_barrier = i;
2109 }
2110 }
2111
2112 for (i = 0; i < max_mirrors; i++) {
2113 bytenr = btrfs_sb_offset(i);
2114 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2115 break;
2116
2117 if (wait) {
2118 bh = __find_get_block(device->bdev, bytenr / 4096,
2119 BTRFS_SUPER_INFO_SIZE);
2120 BUG_ON(!bh);
2121 wait_on_buffer(bh);
2122 if (!buffer_uptodate(bh))
2123 errors++;
2124
2125 /* drop our reference */
2126 brelse(bh);
2127
2128 /* drop the reference from the wait == 0 run */
2129 brelse(bh);
2130 continue;
2131 } else {
2132 btrfs_set_super_bytenr(sb, bytenr);
2133
2134 crc = ~(u32)0;
2135 crc = btrfs_csum_data(NULL, (char *)sb +
2136 BTRFS_CSUM_SIZE, crc,
2137 BTRFS_SUPER_INFO_SIZE -
2138 BTRFS_CSUM_SIZE);
2139 btrfs_csum_final(crc, sb->csum);
2140
2141 /*
2142 * one reference for us, and we leave it for the
2143 * caller
2144 */
2145 bh = __getblk(device->bdev, bytenr / 4096,
2146 BTRFS_SUPER_INFO_SIZE);
2147 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2148
2149 /* one reference for submit_bh */
2150 get_bh(bh);
2151
2152 set_buffer_uptodate(bh);
2153 lock_buffer(bh);
2154 bh->b_end_io = btrfs_end_buffer_write_sync;
2155 }
2156
2157 if (i == last_barrier && do_barriers && device->barriers) {
2158 ret = submit_bh(WRITE_BARRIER, bh);
2159 if (ret == -EOPNOTSUPP) {
2160 printk("btrfs: disabling barriers on dev %s\n",
2161 device->name);
2162 set_buffer_uptodate(bh);
2163 device->barriers = 0;
2164 /* one reference for submit_bh */
2165 get_bh(bh);
2166 lock_buffer(bh);
2167 ret = submit_bh(WRITE_SYNC, bh);
2168 }
2169 } else {
2170 ret = submit_bh(WRITE_SYNC, bh);
2171 }
2172
2173 if (ret)
2174 errors++;
2175 }
2176 return errors < i ? 0 : -1;
2177}
2178
2179int write_all_supers(struct btrfs_root *root, int max_mirrors)
2180{
2181 struct list_head *head;
2182 struct btrfs_device *dev;
2183 struct btrfs_super_block *sb;
2184 struct btrfs_dev_item *dev_item;
2185 int ret;
2186 int do_barriers;
2187 int max_errors;
2188 int total_errors = 0;
2189 u64 flags;
2190
2191 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2192 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2193
2194 sb = &root->fs_info->super_for_commit;
2195 dev_item = &sb->dev_item;
2196
2197 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2198 head = &root->fs_info->fs_devices->devices;
2199 list_for_each_entry(dev, head, dev_list) {
2200 if (!dev->bdev) {
2201 total_errors++;
2202 continue;
2203 }
2204 if (!dev->in_fs_metadata || !dev->writeable)
2205 continue;
2206
2207 btrfs_set_stack_device_generation(dev_item, 0);
2208 btrfs_set_stack_device_type(dev_item, dev->type);
2209 btrfs_set_stack_device_id(dev_item, dev->devid);
2210 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2211 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2212 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2213 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2214 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2215 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2216 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2217
2218 flags = btrfs_super_flags(sb);
2219 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2220
2221 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2222 if (ret)
2223 total_errors++;
2224 }
2225 if (total_errors > max_errors) {
2226 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2227 total_errors);
2228 BUG();
2229 }
2230
2231 total_errors = 0;
2232 list_for_each_entry(dev, head, dev_list) {
2233 if (!dev->bdev)
2234 continue;
2235 if (!dev->in_fs_metadata || !dev->writeable)
2236 continue;
2237
2238 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2239 if (ret)
2240 total_errors++;
2241 }
2242 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2243 if (total_errors > max_errors) {
2244 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2245 total_errors);
2246 BUG();
2247 }
2248 return 0;
2249}
2250
2251int write_ctree_super(struct btrfs_trans_handle *trans,
2252 struct btrfs_root *root, int max_mirrors)
2253{
2254 int ret;
2255
2256 ret = write_all_supers(root, max_mirrors);
2257 return ret;
2258}
2259
2260int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2261{
2262 spin_lock(&fs_info->fs_roots_radix_lock);
2263 radix_tree_delete(&fs_info->fs_roots_radix,
2264 (unsigned long)root->root_key.objectid);
2265 spin_unlock(&fs_info->fs_roots_radix_lock);
2266 free_fs_root(root);
2267 return 0;
2268}
2269
2270static void free_fs_root(struct btrfs_root *root)
2271{
2272 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2273 if (root->anon_super.s_dev) {
2274 down_write(&root->anon_super.s_umount);
2275 kill_anon_super(&root->anon_super);
2276 }
2277 free_extent_buffer(root->node);
2278 free_extent_buffer(root->commit_root);
2279 kfree(root->name);
2280 kfree(root);
2281}
2282
2283static int del_fs_roots(struct btrfs_fs_info *fs_info)
2284{
2285 int ret;
2286 struct btrfs_root *gang[8];
2287 int i;
2288
2289 while (1) {
2290 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2291 (void **)gang, 0,
2292 ARRAY_SIZE(gang));
2293 if (!ret)
2294 break;
2295 for (i = 0; i < ret; i++)
2296 btrfs_free_fs_root(fs_info, gang[i]);
2297 }
2298 return 0;
2299}
2300
2301int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2302{
2303 u64 root_objectid = 0;
2304 struct btrfs_root *gang[8];
2305 int i;
2306 int ret;
2307
2308 while (1) {
2309 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2310 (void **)gang, root_objectid,
2311 ARRAY_SIZE(gang));
2312 if (!ret)
2313 break;
2314
2315 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2316 for (i = 0; i < ret; i++) {
2317 root_objectid = gang[i]->root_key.objectid;
2318 ret = btrfs_find_dead_roots(fs_info->tree_root,
2319 root_objectid);
2320 BUG_ON(ret);
2321 btrfs_orphan_cleanup(gang[i]);
2322 }
2323 root_objectid++;
2324 }
2325 return 0;
2326}
2327
2328int btrfs_commit_super(struct btrfs_root *root)
2329{
2330 struct btrfs_trans_handle *trans;
2331 int ret;
2332
2333 mutex_lock(&root->fs_info->cleaner_mutex);
2334 btrfs_clean_old_snapshots(root);
2335 mutex_unlock(&root->fs_info->cleaner_mutex);
2336 trans = btrfs_start_transaction(root, 1);
2337 ret = btrfs_commit_transaction(trans, root);
2338 BUG_ON(ret);
2339 /* run commit again to drop the original snapshot */
2340 trans = btrfs_start_transaction(root, 1);
2341 btrfs_commit_transaction(trans, root);
2342 ret = btrfs_write_and_wait_transaction(NULL, root);
2343 BUG_ON(ret);
2344
2345 ret = write_ctree_super(NULL, root, 0);
2346 return ret;
2347}
2348
2349int close_ctree(struct btrfs_root *root)
2350{
2351 struct btrfs_fs_info *fs_info = root->fs_info;
2352 int ret;
2353
2354 fs_info->closing = 1;
2355 smp_mb();
2356
2357 kthread_stop(root->fs_info->transaction_kthread);
2358 kthread_stop(root->fs_info->cleaner_kthread);
2359
2360 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2361 ret = btrfs_commit_super(root);
2362 if (ret)
2363 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2364 }
2365
2366 fs_info->closing = 2;
2367 smp_mb();
2368
2369 if (fs_info->delalloc_bytes) {
2370 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2371 (unsigned long long)fs_info->delalloc_bytes);
2372 }
2373 if (fs_info->total_ref_cache_size) {
2374 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2375 (unsigned long long)fs_info->total_ref_cache_size);
2376 }
2377
2378 free_extent_buffer(fs_info->extent_root->node);
2379 free_extent_buffer(fs_info->extent_root->commit_root);
2380 free_extent_buffer(fs_info->tree_root->node);
2381 free_extent_buffer(fs_info->tree_root->commit_root);
2382 free_extent_buffer(root->fs_info->chunk_root->node);
2383 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2384 free_extent_buffer(root->fs_info->dev_root->node);
2385 free_extent_buffer(root->fs_info->dev_root->commit_root);
2386 free_extent_buffer(root->fs_info->csum_root->node);
2387 free_extent_buffer(root->fs_info->csum_root->commit_root);
2388
2389 btrfs_free_block_groups(root->fs_info);
2390
2391 del_fs_roots(fs_info);
2392
2393 iput(fs_info->btree_inode);
2394
2395 btrfs_stop_workers(&fs_info->fixup_workers);
2396 btrfs_stop_workers(&fs_info->delalloc_workers);
2397 btrfs_stop_workers(&fs_info->workers);
2398 btrfs_stop_workers(&fs_info->endio_workers);
2399 btrfs_stop_workers(&fs_info->endio_meta_workers);
2400 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2401 btrfs_stop_workers(&fs_info->endio_write_workers);
2402 btrfs_stop_workers(&fs_info->submit_workers);
2403
2404 btrfs_close_devices(fs_info->fs_devices);
2405 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2406
2407 bdi_destroy(&fs_info->bdi);
2408
2409 kfree(fs_info->extent_root);
2410 kfree(fs_info->tree_root);
2411 kfree(fs_info->chunk_root);
2412 kfree(fs_info->dev_root);
2413 kfree(fs_info->csum_root);
2414 return 0;
2415}
2416
2417int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2418{
2419 int ret;
2420 struct inode *btree_inode = buf->first_page->mapping->host;
2421
2422 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2423 if (!ret)
2424 return ret;
2425
2426 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2427 parent_transid);
2428 return !ret;
2429}
2430
2431int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2432{
2433 struct inode *btree_inode = buf->first_page->mapping->host;
2434 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2435 buf);
2436}
2437
2438void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2439{
2440 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2441 u64 transid = btrfs_header_generation(buf);
2442 struct inode *btree_inode = root->fs_info->btree_inode;
2443 int was_dirty;
2444
2445 btrfs_assert_tree_locked(buf);
2446 if (transid != root->fs_info->generation) {
2447 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2448 "found %llu running %llu\n",
2449 (unsigned long long)buf->start,
2450 (unsigned long long)transid,
2451 (unsigned long long)root->fs_info->generation);
2452 WARN_ON(1);
2453 }
2454 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2455 buf);
2456 if (!was_dirty) {
2457 spin_lock(&root->fs_info->delalloc_lock);
2458 root->fs_info->dirty_metadata_bytes += buf->len;
2459 spin_unlock(&root->fs_info->delalloc_lock);
2460 }
2461}
2462
2463void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2464{
2465 /*
2466 * looks as though older kernels can get into trouble with
2467 * this code, they end up stuck in balance_dirty_pages forever
2468 */
2469 u64 num_dirty;
2470 unsigned long thresh = 32 * 1024 * 1024;
2471
2472 if (current->flags & PF_MEMALLOC)
2473 return;
2474
2475 num_dirty = root->fs_info->dirty_metadata_bytes;
2476
2477 if (num_dirty > thresh) {
2478 balance_dirty_pages_ratelimited_nr(
2479 root->fs_info->btree_inode->i_mapping, 1);
2480 }
2481 return;
2482}
2483
2484int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2485{
2486 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2487 int ret;
2488 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2489 if (ret == 0)
2490 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2491 return ret;
2492}
2493
2494int btree_lock_page_hook(struct page *page)
2495{
2496 struct inode *inode = page->mapping->host;
2497 struct btrfs_root *root = BTRFS_I(inode)->root;
2498 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2499 struct extent_buffer *eb;
2500 unsigned long len;
2501 u64 bytenr = page_offset(page);
2502
2503 if (page->private == EXTENT_PAGE_PRIVATE)
2504 goto out;
2505
2506 len = page->private >> 2;
2507 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2508 if (!eb)
2509 goto out;
2510
2511 btrfs_tree_lock(eb);
2512 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2513
2514 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2515 spin_lock(&root->fs_info->delalloc_lock);
2516 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2517 root->fs_info->dirty_metadata_bytes -= eb->len;
2518 else
2519 WARN_ON(1);
2520 spin_unlock(&root->fs_info->delalloc_lock);
2521 }
2522
2523 btrfs_tree_unlock(eb);
2524 free_extent_buffer(eb);
2525out:
2526 lock_page(page);
2527 return 0;
2528}
2529
2530static struct extent_io_ops btree_extent_io_ops = {
2531 .write_cache_pages_lock_hook = btree_lock_page_hook,
2532 .readpage_end_io_hook = btree_readpage_end_io_hook,
2533 .submit_bio_hook = btree_submit_bio_hook,
2534 /* note we're sharing with inode.c for the merge bio hook */
2535 .merge_bio_hook = btrfs_merge_bio_hook,
2536};
Linux 6.x block layer and io_uring tree(s)