]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blob - fs/btrfs/disk-io.c
projects / mirror_ubuntu-eoan-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: fix OOPS of empty filesystem after balance
[mirror_ubuntu-eoan-kernel.git] / fs / btrfs / disk-io.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/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 <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <asm/unaligned.h>
33 #include "compat.h"
34 #include "ctree.h"
35 #include "disk-io.h"
36 #include "transaction.h"
37 #include "btrfs_inode.h"
38 #include "volumes.h"
39 #include "print-tree.h"
40 #include "async-thread.h"
41 #include "locking.h"
42 #include "tree-log.h"
43 #include "free-space-cache.h"
44
45 static struct extent_io_ops btree_extent_io_ops;
46 static void end_workqueue_fn(struct btrfs_work *work);
47 static void free_fs_root(struct btrfs_root *root);
48 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
49 int read_only);
50 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
51 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
52 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
53 struct btrfs_root *root);
54 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
55 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
56 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
57 struct extent_io_tree *dirty_pages,
58 int mark);
59 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
60 struct extent_io_tree *pinned_extents);
61 static int btrfs_cleanup_transaction(struct btrfs_root *root);
62
63 /*
64 * end_io_wq structs are used to do processing in task context when an IO is
65 * complete. This is used during reads to verify checksums, and it is used
66 * by writes to insert metadata for new file extents after IO is complete.
67 */
68 struct end_io_wq {
69 struct bio *bio;
70 bio_end_io_t *end_io;
71 void *private;
72 struct btrfs_fs_info *info;
73 int error;
74 int metadata;
75 struct list_head list;
76 struct btrfs_work work;
77 };
78
79 /*
80 * async submit bios are used to offload expensive checksumming
81 * onto the worker threads. They checksum file and metadata bios
82 * just before they are sent down the IO stack.
83 */
84 struct async_submit_bio {
85 struct inode *inode;
86 struct bio *bio;
87 struct list_head list;
88 extent_submit_bio_hook_t *submit_bio_start;
89 extent_submit_bio_hook_t *submit_bio_done;
90 int rw;
91 int mirror_num;
92 unsigned long bio_flags;
93 /*
94 * bio_offset is optional, can be used if the pages in the bio
95 * can't tell us where in the file the bio should go
96 */
97 u64 bio_offset;
98 struct btrfs_work work;
99 };
100
101 /* These are used to set the lockdep class on the extent buffer locks.
102 * The class is set by the readpage_end_io_hook after the buffer has
103 * passed csum validation but before the pages are unlocked.
104 *
105 * The lockdep class is also set by btrfs_init_new_buffer on freshly
106 * allocated blocks.
107 *
108 * The class is based on the level in the tree block, which allows lockdep
109 * to know that lower nodes nest inside the locks of higher nodes.
110 *
111 * We also add a check to make sure the highest level of the tree is
112 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
113 * code needs update as well.
114 */
115 #ifdef CONFIG_DEBUG_LOCK_ALLOC
116 # if BTRFS_MAX_LEVEL != 8
117 # error
118 # endif
119 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
120 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
121 /* leaf */
122 "btrfs-extent-00",
123 "btrfs-extent-01",
124 "btrfs-extent-02",
125 "btrfs-extent-03",
126 "btrfs-extent-04",
127 "btrfs-extent-05",
128 "btrfs-extent-06",
129 "btrfs-extent-07",
130 /* highest possible level */
131 "btrfs-extent-08",
132 };
133 #endif
134
135 /*
136 * extents on the btree inode are pretty simple, there's one extent
137 * that covers the entire device
138 */
139 static struct extent_map *btree_get_extent(struct inode *inode,
140 struct page *page, size_t page_offset, u64 start, u64 len,
141 int create)
142 {
143 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
144 struct extent_map *em;
145 int ret;
146
147 read_lock(&em_tree->lock);
148 em = lookup_extent_mapping(em_tree, start, len);
149 if (em) {
150 em->bdev =
151 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
152 read_unlock(&em_tree->lock);
153 goto out;
154 }
155 read_unlock(&em_tree->lock);
156
157 em = alloc_extent_map(GFP_NOFS);
158 if (!em) {
159 em = ERR_PTR(-ENOMEM);
160 goto out;
161 }
162 em->start = 0;
163 em->len = (u64)-1;
164 em->block_len = (u64)-1;
165 em->block_start = 0;
166 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
167
168 write_lock(&em_tree->lock);
169 ret = add_extent_mapping(em_tree, em);
170 if (ret == -EEXIST) {
171 u64 failed_start = em->start;
172 u64 failed_len = em->len;
173
174 free_extent_map(em);
175 em = lookup_extent_mapping(em_tree, start, len);
176 if (em) {
177 ret = 0;
178 } else {
179 em = lookup_extent_mapping(em_tree, failed_start,
180 failed_len);
181 ret = -EIO;
182 }
183 } else if (ret) {
184 free_extent_map(em);
185 em = NULL;
186 }
187 write_unlock(&em_tree->lock);
188
189 if (ret)
190 em = ERR_PTR(ret);
191 out:
192 return em;
193 }
194
195 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
196 {
197 return crc32c(seed, data, len);
198 }
199
200 void btrfs_csum_final(u32 crc, char *result)
201 {
202 put_unaligned_le32(~crc, result);
203 }
204
205 /*
206 * compute the csum for a btree block, and either verify it or write it
207 * into the csum field of the block.
208 */
209 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
210 int verify)
211 {
212 u16 csum_size =
213 btrfs_super_csum_size(&root->fs_info->super_copy);
214 char *result = NULL;
215 unsigned long len;
216 unsigned long cur_len;
217 unsigned long offset = BTRFS_CSUM_SIZE;
218 char *map_token = NULL;
219 char *kaddr;
220 unsigned long map_start;
221 unsigned long map_len;
222 int err;
223 u32 crc = ~(u32)0;
224 unsigned long inline_result;
225
226 len = buf->len - offset;
227 while (len > 0) {
228 err = map_private_extent_buffer(buf, offset, 32,
229 &map_token, &kaddr,
230 &map_start, &map_len, KM_USER0);
231 if (err)
232 return 1;
233 cur_len = min(len, map_len - (offset - map_start));
234 crc = btrfs_csum_data(root, kaddr + offset - map_start,
235 crc, cur_len);
236 len -= cur_len;
237 offset += cur_len;
238 unmap_extent_buffer(buf, map_token, KM_USER0);
239 }
240 if (csum_size > sizeof(inline_result)) {
241 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
242 if (!result)
243 return 1;
244 } else {
245 result = (char *)&inline_result;
246 }
247
248 btrfs_csum_final(crc, result);
249
250 if (verify) {
251 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
252 u32 val;
253 u32 found = 0;
254 memcpy(&found, result, csum_size);
255
256 read_extent_buffer(buf, &val, 0, csum_size);
257 if (printk_ratelimit()) {
258 printk(KERN_INFO "btrfs: %s checksum verify "
259 "failed on %llu wanted %X found %X "
260 "level %d\n",
261 root->fs_info->sb->s_id,
262 (unsigned long long)buf->start, val, found,
263 btrfs_header_level(buf));
264 }
265 if (result != (char *)&inline_result)
266 kfree(result);
267 return 1;
268 }
269 } else {
270 write_extent_buffer(buf, result, 0, csum_size);
271 }
272 if (result != (char *)&inline_result)
273 kfree(result);
274 return 0;
275 }
276
277 /*
278 * we can't consider a given block up to date unless the transid of the
279 * block matches the transid in the parent node's pointer. This is how we
280 * detect blocks that either didn't get written at all or got written
281 * in the wrong place.
282 */
283 static int verify_parent_transid(struct extent_io_tree *io_tree,
284 struct extent_buffer *eb, u64 parent_transid)
285 {
286 struct extent_state *cached_state = NULL;
287 int ret;
288
289 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
290 return 0;
291
292 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
293 0, &cached_state, GFP_NOFS);
294 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
295 btrfs_header_generation(eb) == parent_transid) {
296 ret = 0;
297 goto out;
298 }
299 if (printk_ratelimit()) {
300 printk("parent transid verify failed on %llu wanted %llu "
301 "found %llu\n",
302 (unsigned long long)eb->start,
303 (unsigned long long)parent_transid,
304 (unsigned long long)btrfs_header_generation(eb));
305 }
306 ret = 1;
307 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
308 out:
309 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
310 &cached_state, GFP_NOFS);
311 return ret;
312 }
313
314 /*
315 * helper to read a given tree block, doing retries as required when
316 * the checksums don't match and we have alternate mirrors to try.
317 */
318 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
319 struct extent_buffer *eb,
320 u64 start, u64 parent_transid)
321 {
322 struct extent_io_tree *io_tree;
323 int ret;
324 int num_copies = 0;
325 int mirror_num = 0;
326
327 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
328 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
329 while (1) {
330 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
331 btree_get_extent, mirror_num);
332 if (!ret &&
333 !verify_parent_transid(io_tree, eb, parent_transid))
334 return ret;
335
336 /*
337 * This buffer's crc is fine, but its contents are corrupted, so
338 * there is no reason to read the other copies, they won't be
339 * any less wrong.
340 */
341 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
342 return ret;
343
344 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
345 eb->start, eb->len);
346 if (num_copies == 1)
347 return ret;
348
349 mirror_num++;
350 if (mirror_num > num_copies)
351 return ret;
352 }
353 return -EIO;
354 }
355
356 /*
357 * checksum a dirty tree block before IO. This has extra checks to make sure
358 * we only fill in the checksum field in the first page of a multi-page block
359 */
360
361 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
362 {
363 struct extent_io_tree *tree;
364 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
365 u64 found_start;
366 unsigned long len;
367 struct extent_buffer *eb;
368 int ret;
369
370 tree = &BTRFS_I(page->mapping->host)->io_tree;
371
372 if (page->private == EXTENT_PAGE_PRIVATE) {
373 WARN_ON(1);
374 goto out;
375 }
376 if (!page->private) {
377 WARN_ON(1);
378 goto out;
379 }
380 len = page->private >> 2;
381 WARN_ON(len == 0);
382
383 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
384 if (eb == NULL) {
385 WARN_ON(1);
386 goto out;
387 }
388 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
389 btrfs_header_generation(eb));
390 BUG_ON(ret);
391 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
392
393 found_start = btrfs_header_bytenr(eb);
394 if (found_start != start) {
395 WARN_ON(1);
396 goto err;
397 }
398 if (eb->first_page != page) {
399 WARN_ON(1);
400 goto err;
401 }
402 if (!PageUptodate(page)) {
403 WARN_ON(1);
404 goto err;
405 }
406 csum_tree_block(root, eb, 0);
407 err:
408 free_extent_buffer(eb);
409 out:
410 return 0;
411 }
412
413 static int check_tree_block_fsid(struct btrfs_root *root,
414 struct extent_buffer *eb)
415 {
416 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
417 u8 fsid[BTRFS_UUID_SIZE];
418 int ret = 1;
419
420 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
421 BTRFS_FSID_SIZE);
422 while (fs_devices) {
423 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
424 ret = 0;
425 break;
426 }
427 fs_devices = fs_devices->seed;
428 }
429 return ret;
430 }
431
432 #define CORRUPT(reason, eb, root, slot) \
433 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
434 "root=%llu, slot=%d\n", reason, \
435 (unsigned long long)btrfs_header_bytenr(eb), \
436 (unsigned long long)root->objectid, slot)
437
438 static noinline int check_leaf(struct btrfs_root *root,
439 struct extent_buffer *leaf)
440 {
441 struct btrfs_key key;
442 struct btrfs_key leaf_key;
443 u32 nritems = btrfs_header_nritems(leaf);
444 int slot;
445
446 if (nritems == 0)
447 return 0;
448
449 /* Check the 0 item */
450 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
451 BTRFS_LEAF_DATA_SIZE(root)) {
452 CORRUPT("invalid item offset size pair", leaf, root, 0);
453 return -EIO;
454 }
455
456 /*
457 * Check to make sure each items keys are in the correct order and their
458 * offsets make sense. We only have to loop through nritems-1 because
459 * we check the current slot against the next slot, which verifies the
460 * next slot's offset+size makes sense and that the current's slot
461 * offset is correct.
462 */
463 for (slot = 0; slot < nritems - 1; slot++) {
464 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
465 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
466
467 /* Make sure the keys are in the right order */
468 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
469 CORRUPT("bad key order", leaf, root, slot);
470 return -EIO;
471 }
472
473 /*
474 * Make sure the offset and ends are right, remember that the
475 * item data starts at the end of the leaf and grows towards the
476 * front.
477 */
478 if (btrfs_item_offset_nr(leaf, slot) !=
479 btrfs_item_end_nr(leaf, slot + 1)) {
480 CORRUPT("slot offset bad", leaf, root, slot);
481 return -EIO;
482 }
483
484 /*
485 * Check to make sure that we don't point outside of the leaf,
486 * just incase all the items are consistent to eachother, but
487 * all point outside of the leaf.
488 */
489 if (btrfs_item_end_nr(leaf, slot) >
490 BTRFS_LEAF_DATA_SIZE(root)) {
491 CORRUPT("slot end outside of leaf", leaf, root, slot);
492 return -EIO;
493 }
494 }
495
496 return 0;
497 }
498
499 #ifdef CONFIG_DEBUG_LOCK_ALLOC
500 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
501 {
502 lockdep_set_class_and_name(&eb->lock,
503 &btrfs_eb_class[level],
504 btrfs_eb_name[level]);
505 }
506 #endif
507
508 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
509 struct extent_state *state)
510 {
511 struct extent_io_tree *tree;
512 u64 found_start;
513 int found_level;
514 unsigned long len;
515 struct extent_buffer *eb;
516 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
517 int ret = 0;
518
519 tree = &BTRFS_I(page->mapping->host)->io_tree;
520 if (page->private == EXTENT_PAGE_PRIVATE)
521 goto out;
522 if (!page->private)
523 goto out;
524
525 len = page->private >> 2;
526 WARN_ON(len == 0);
527
528 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
529 if (eb == NULL) {
530 ret = -EIO;
531 goto out;
532 }
533
534 found_start = btrfs_header_bytenr(eb);
535 if (found_start != start) {
536 if (printk_ratelimit()) {
537 printk(KERN_INFO "btrfs bad tree block start "
538 "%llu %llu\n",
539 (unsigned long long)found_start,
540 (unsigned long long)eb->start);
541 }
542 ret = -EIO;
543 goto err;
544 }
545 if (eb->first_page != page) {
546 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
547 eb->first_page->index, page->index);
548 WARN_ON(1);
549 ret = -EIO;
550 goto err;
551 }
552 if (check_tree_block_fsid(root, eb)) {
553 if (printk_ratelimit()) {
554 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
555 (unsigned long long)eb->start);
556 }
557 ret = -EIO;
558 goto err;
559 }
560 found_level = btrfs_header_level(eb);
561
562 btrfs_set_buffer_lockdep_class(eb, found_level);
563
564 ret = csum_tree_block(root, eb, 1);
565 if (ret) {
566 ret = -EIO;
567 goto err;
568 }
569
570 /*
571 * If this is a leaf block and it is corrupt, set the corrupt bit so
572 * that we don't try and read the other copies of this block, just
573 * return -EIO.
574 */
575 if (found_level == 0 && check_leaf(root, eb)) {
576 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
577 ret = -EIO;
578 }
579
580 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
581 end = eb->start + end - 1;
582 err:
583 free_extent_buffer(eb);
584 out:
585 return ret;
586 }
587
588 static void end_workqueue_bio(struct bio *bio, int err)
589 {
590 struct end_io_wq *end_io_wq = bio->bi_private;
591 struct btrfs_fs_info *fs_info;
592
593 fs_info = end_io_wq->info;
594 end_io_wq->error = err;
595 end_io_wq->work.func = end_workqueue_fn;
596 end_io_wq->work.flags = 0;
597
598 if (bio->bi_rw & REQ_WRITE) {
599 if (end_io_wq->metadata == 1)
600 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
601 &end_io_wq->work);
602 else if (end_io_wq->metadata == 2)
603 btrfs_queue_worker(&fs_info->endio_freespace_worker,
604 &end_io_wq->work);
605 else
606 btrfs_queue_worker(&fs_info->endio_write_workers,
607 &end_io_wq->work);
608 } else {
609 if (end_io_wq->metadata)
610 btrfs_queue_worker(&fs_info->endio_meta_workers,
611 &end_io_wq->work);
612 else
613 btrfs_queue_worker(&fs_info->endio_workers,
614 &end_io_wq->work);
615 }
616 }
617
618 /*
619 * For the metadata arg you want
620 *
621 * 0 - if data
622 * 1 - if normal metadta
623 * 2 - if writing to the free space cache area
624 */
625 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
626 int metadata)
627 {
628 struct end_io_wq *end_io_wq;
629 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
630 if (!end_io_wq)
631 return -ENOMEM;
632
633 end_io_wq->private = bio->bi_private;
634 end_io_wq->end_io = bio->bi_end_io;
635 end_io_wq->info = info;
636 end_io_wq->error = 0;
637 end_io_wq->bio = bio;
638 end_io_wq->metadata = metadata;
639
640 bio->bi_private = end_io_wq;
641 bio->bi_end_io = end_workqueue_bio;
642 return 0;
643 }
644
645 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
646 {
647 unsigned long limit = min_t(unsigned long,
648 info->workers.max_workers,
649 info->fs_devices->open_devices);
650 return 256 * limit;
651 }
652
653 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
654 {
655 return atomic_read(&info->nr_async_bios) >
656 btrfs_async_submit_limit(info);
657 }
658
659 static void run_one_async_start(struct btrfs_work *work)
660 {
661 struct async_submit_bio *async;
662
663 async = container_of(work, struct async_submit_bio, work);
664 async->submit_bio_start(async->inode, async->rw, async->bio,
665 async->mirror_num, async->bio_flags,
666 async->bio_offset);
667 }
668
669 static void run_one_async_done(struct btrfs_work *work)
670 {
671 struct btrfs_fs_info *fs_info;
672 struct async_submit_bio *async;
673 int limit;
674
675 async = container_of(work, struct async_submit_bio, work);
676 fs_info = BTRFS_I(async->inode)->root->fs_info;
677
678 limit = btrfs_async_submit_limit(fs_info);
679 limit = limit * 2 / 3;
680
681 atomic_dec(&fs_info->nr_async_submits);
682
683 if (atomic_read(&fs_info->nr_async_submits) < limit &&
684 waitqueue_active(&fs_info->async_submit_wait))
685 wake_up(&fs_info->async_submit_wait);
686
687 async->submit_bio_done(async->inode, async->rw, async->bio,
688 async->mirror_num, async->bio_flags,
689 async->bio_offset);
690 }
691
692 static void run_one_async_free(struct btrfs_work *work)
693 {
694 struct async_submit_bio *async;
695
696 async = container_of(work, struct async_submit_bio, work);
697 kfree(async);
698 }
699
700 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
701 int rw, struct bio *bio, int mirror_num,
702 unsigned long bio_flags,
703 u64 bio_offset,
704 extent_submit_bio_hook_t *submit_bio_start,
705 extent_submit_bio_hook_t *submit_bio_done)
706 {
707 struct async_submit_bio *async;
708
709 async = kmalloc(sizeof(*async), GFP_NOFS);
710 if (!async)
711 return -ENOMEM;
712
713 async->inode = inode;
714 async->rw = rw;
715 async->bio = bio;
716 async->mirror_num = mirror_num;
717 async->submit_bio_start = submit_bio_start;
718 async->submit_bio_done = submit_bio_done;
719
720 async->work.func = run_one_async_start;
721 async->work.ordered_func = run_one_async_done;
722 async->work.ordered_free = run_one_async_free;
723
724 async->work.flags = 0;
725 async->bio_flags = bio_flags;
726 async->bio_offset = bio_offset;
727
728 atomic_inc(&fs_info->nr_async_submits);
729
730 if (rw & REQ_SYNC)
731 btrfs_set_work_high_prio(&async->work);
732
733 btrfs_queue_worker(&fs_info->workers, &async->work);
734
735 while (atomic_read(&fs_info->async_submit_draining) &&
736 atomic_read(&fs_info->nr_async_submits)) {
737 wait_event(fs_info->async_submit_wait,
738 (atomic_read(&fs_info->nr_async_submits) == 0));
739 }
740
741 return 0;
742 }
743
744 static int btree_csum_one_bio(struct bio *bio)
745 {
746 struct bio_vec *bvec = bio->bi_io_vec;
747 int bio_index = 0;
748 struct btrfs_root *root;
749
750 WARN_ON(bio->bi_vcnt <= 0);
751 while (bio_index < bio->bi_vcnt) {
752 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
753 csum_dirty_buffer(root, bvec->bv_page);
754 bio_index++;
755 bvec++;
756 }
757 return 0;
758 }
759
760 static int __btree_submit_bio_start(struct inode *inode, int rw,
761 struct bio *bio, int mirror_num,
762 unsigned long bio_flags,
763 u64 bio_offset)
764 {
765 /*
766 * when we're called for a write, we're already in the async
767 * submission context. Just jump into btrfs_map_bio
768 */
769 btree_csum_one_bio(bio);
770 return 0;
771 }
772
773 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
774 int mirror_num, unsigned long bio_flags,
775 u64 bio_offset)
776 {
777 /*
778 * when we're called for a write, we're already in the async
779 * submission context. Just jump into btrfs_map_bio
780 */
781 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
782 }
783
784 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
785 int mirror_num, unsigned long bio_flags,
786 u64 bio_offset)
787 {
788 int ret;
789
790 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
791 bio, 1);
792 BUG_ON(ret);
793
794 if (!(rw & REQ_WRITE)) {
795 /*
796 * called for a read, do the setup so that checksum validation
797 * can happen in the async kernel threads
798 */
799 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
800 mirror_num, 0);
801 }
802
803 /*
804 * kthread helpers are used to submit writes so that checksumming
805 * can happen in parallel across all CPUs
806 */
807 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
808 inode, rw, bio, mirror_num, 0,
809 bio_offset,
810 __btree_submit_bio_start,
811 __btree_submit_bio_done);
812 }
813
814 #ifdef CONFIG_MIGRATION
815 static int btree_migratepage(struct address_space *mapping,
816 struct page *newpage, struct page *page)
817 {
818 /*
819 * we can't safely write a btree page from here,
820 * we haven't done the locking hook
821 */
822 if (PageDirty(page))
823 return -EAGAIN;
824 /*
825 * Buffers may be managed in a filesystem specific way.
826 * We must have no buffers or drop them.
827 */
828 if (page_has_private(page) &&
829 !try_to_release_page(page, GFP_KERNEL))
830 return -EAGAIN;
831 return migrate_page(mapping, newpage, page);
832 }
833 #endif
834
835 static int btree_writepage(struct page *page, struct writeback_control *wbc)
836 {
837 struct extent_io_tree *tree;
838 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
839 struct extent_buffer *eb;
840 int was_dirty;
841
842 tree = &BTRFS_I(page->mapping->host)->io_tree;
843 if (!(current->flags & PF_MEMALLOC)) {
844 return extent_write_full_page(tree, page,
845 btree_get_extent, wbc);
846 }
847
848 redirty_page_for_writepage(wbc, page);
849 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
850 WARN_ON(!eb);
851
852 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
853 if (!was_dirty) {
854 spin_lock(&root->fs_info->delalloc_lock);
855 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
856 spin_unlock(&root->fs_info->delalloc_lock);
857 }
858 free_extent_buffer(eb);
859
860 unlock_page(page);
861 return 0;
862 }
863
864 static int btree_writepages(struct address_space *mapping,
865 struct writeback_control *wbc)
866 {
867 struct extent_io_tree *tree;
868 tree = &BTRFS_I(mapping->host)->io_tree;
869 if (wbc->sync_mode == WB_SYNC_NONE) {
870 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
871 u64 num_dirty;
872 unsigned long thresh = 32 * 1024 * 1024;
873
874 if (wbc->for_kupdate)
875 return 0;
876
877 /* this is a bit racy, but that's ok */
878 num_dirty = root->fs_info->dirty_metadata_bytes;
879 if (num_dirty < thresh)
880 return 0;
881 }
882 return extent_writepages(tree, mapping, btree_get_extent, wbc);
883 }
884
885 static int btree_readpage(struct file *file, struct page *page)
886 {
887 struct extent_io_tree *tree;
888 tree = &BTRFS_I(page->mapping->host)->io_tree;
889 return extent_read_full_page(tree, page, btree_get_extent);
890 }
891
892 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
893 {
894 struct extent_io_tree *tree;
895 struct extent_map_tree *map;
896 int ret;
897
898 if (PageWriteback(page) || PageDirty(page))
899 return 0;
900
901 tree = &BTRFS_I(page->mapping->host)->io_tree;
902 map = &BTRFS_I(page->mapping->host)->extent_tree;
903
904 ret = try_release_extent_state(map, tree, page, gfp_flags);
905 if (!ret)
906 return 0;
907
908 ret = try_release_extent_buffer(tree, page);
909 if (ret == 1) {
910 ClearPagePrivate(page);
911 set_page_private(page, 0);
912 page_cache_release(page);
913 }
914
915 return ret;
916 }
917
918 static void btree_invalidatepage(struct page *page, unsigned long offset)
919 {
920 struct extent_io_tree *tree;
921 tree = &BTRFS_I(page->mapping->host)->io_tree;
922 extent_invalidatepage(tree, page, offset);
923 btree_releasepage(page, GFP_NOFS);
924 if (PagePrivate(page)) {
925 printk(KERN_WARNING "btrfs warning page private not zero "
926 "on page %llu\n", (unsigned long long)page_offset(page));
927 ClearPagePrivate(page);
928 set_page_private(page, 0);
929 page_cache_release(page);
930 }
931 }
932
933 static const struct address_space_operations btree_aops = {
934 .readpage = btree_readpage,
935 .writepage = btree_writepage,
936 .writepages = btree_writepages,
937 .releasepage = btree_releasepage,
938 .invalidatepage = btree_invalidatepage,
939 .sync_page = block_sync_page,
940 #ifdef CONFIG_MIGRATION
941 .migratepage = btree_migratepage,
942 #endif
943 };
944
945 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
946 u64 parent_transid)
947 {
948 struct extent_buffer *buf = NULL;
949 struct inode *btree_inode = root->fs_info->btree_inode;
950 int ret = 0;
951
952 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
953 if (!buf)
954 return 0;
955 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
956 buf, 0, 0, btree_get_extent, 0);
957 free_extent_buffer(buf);
958 return ret;
959 }
960
961 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
962 u64 bytenr, u32 blocksize)
963 {
964 struct inode *btree_inode = root->fs_info->btree_inode;
965 struct extent_buffer *eb;
966 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
967 bytenr, blocksize, GFP_NOFS);
968 return eb;
969 }
970
971 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
972 u64 bytenr, u32 blocksize)
973 {
974 struct inode *btree_inode = root->fs_info->btree_inode;
975 struct extent_buffer *eb;
976
977 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
978 bytenr, blocksize, NULL, GFP_NOFS);
979 return eb;
980 }
981
982
983 int btrfs_write_tree_block(struct extent_buffer *buf)
984 {
985 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
986 buf->start + buf->len - 1);
987 }
988
989 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
990 {
991 return filemap_fdatawait_range(buf->first_page->mapping,
992 buf->start, buf->start + buf->len - 1);
993 }
994
995 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
996 u32 blocksize, u64 parent_transid)
997 {
998 struct extent_buffer *buf = NULL;
999 int ret;
1000
1001 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1002 if (!buf)
1003 return NULL;
1004
1005 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1006
1007 if (ret == 0)
1008 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1009 return buf;
1010
1011 }
1012
1013 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1014 struct extent_buffer *buf)
1015 {
1016 struct inode *btree_inode = root->fs_info->btree_inode;
1017 if (btrfs_header_generation(buf) ==
1018 root->fs_info->running_transaction->transid) {
1019 btrfs_assert_tree_locked(buf);
1020
1021 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1022 spin_lock(&root->fs_info->delalloc_lock);
1023 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1024 root->fs_info->dirty_metadata_bytes -= buf->len;
1025 else
1026 WARN_ON(1);
1027 spin_unlock(&root->fs_info->delalloc_lock);
1028 }
1029
1030 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1031 btrfs_set_lock_blocking(buf);
1032 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1033 buf);
1034 }
1035 return 0;
1036 }
1037
1038 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1039 u32 stripesize, struct btrfs_root *root,
1040 struct btrfs_fs_info *fs_info,
1041 u64 objectid)
1042 {
1043 root->node = NULL;
1044 root->commit_root = NULL;
1045 root->sectorsize = sectorsize;
1046 root->nodesize = nodesize;
1047 root->leafsize = leafsize;
1048 root->stripesize = stripesize;
1049 root->ref_cows = 0;
1050 root->track_dirty = 0;
1051 root->in_radix = 0;
1052 root->orphan_item_inserted = 0;
1053 root->orphan_cleanup_state = 0;
1054
1055 root->fs_info = fs_info;
1056 root->objectid = objectid;
1057 root->last_trans = 0;
1058 root->highest_objectid = 0;
1059 root->name = NULL;
1060 root->in_sysfs = 0;
1061 root->inode_tree = RB_ROOT;
1062 root->block_rsv = NULL;
1063 root->orphan_block_rsv = NULL;
1064
1065 INIT_LIST_HEAD(&root->dirty_list);
1066 INIT_LIST_HEAD(&root->orphan_list);
1067 INIT_LIST_HEAD(&root->root_list);
1068 spin_lock_init(&root->node_lock);
1069 spin_lock_init(&root->orphan_lock);
1070 spin_lock_init(&root->inode_lock);
1071 spin_lock_init(&root->accounting_lock);
1072 mutex_init(&root->objectid_mutex);
1073 mutex_init(&root->log_mutex);
1074 init_waitqueue_head(&root->log_writer_wait);
1075 init_waitqueue_head(&root->log_commit_wait[0]);
1076 init_waitqueue_head(&root->log_commit_wait[1]);
1077 atomic_set(&root->log_commit[0], 0);
1078 atomic_set(&root->log_commit[1], 0);
1079 atomic_set(&root->log_writers, 0);
1080 root->log_batch = 0;
1081 root->log_transid = 0;
1082 root->last_log_commit = 0;
1083 extent_io_tree_init(&root->dirty_log_pages,
1084 fs_info->btree_inode->i_mapping, GFP_NOFS);
1085
1086 memset(&root->root_key, 0, sizeof(root->root_key));
1087 memset(&root->root_item, 0, sizeof(root->root_item));
1088 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1089 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1090 root->defrag_trans_start = fs_info->generation;
1091 init_completion(&root->kobj_unregister);
1092 root->defrag_running = 0;
1093 root->root_key.objectid = objectid;
1094 root->anon_super.s_root = NULL;
1095 root->anon_super.s_dev = 0;
1096 INIT_LIST_HEAD(&root->anon_super.s_list);
1097 INIT_LIST_HEAD(&root->anon_super.s_instances);
1098 init_rwsem(&root->anon_super.s_umount);
1099
1100 return 0;
1101 }
1102
1103 static int find_and_setup_root(struct btrfs_root *tree_root,
1104 struct btrfs_fs_info *fs_info,
1105 u64 objectid,
1106 struct btrfs_root *root)
1107 {
1108 int ret;
1109 u32 blocksize;
1110 u64 generation;
1111
1112 __setup_root(tree_root->nodesize, tree_root->leafsize,
1113 tree_root->sectorsize, tree_root->stripesize,
1114 root, fs_info, objectid);
1115 ret = btrfs_find_last_root(tree_root, objectid,
1116 &root->root_item, &root->root_key);
1117 if (ret > 0)
1118 return -ENOENT;
1119 BUG_ON(ret);
1120
1121 generation = btrfs_root_generation(&root->root_item);
1122 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1123 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1124 blocksize, generation);
1125 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1126 free_extent_buffer(root->node);
1127 return -EIO;
1128 }
1129 root->commit_root = btrfs_root_node(root);
1130 return 0;
1131 }
1132
1133 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1134 struct btrfs_fs_info *fs_info)
1135 {
1136 struct btrfs_root *root;
1137 struct btrfs_root *tree_root = fs_info->tree_root;
1138 struct extent_buffer *leaf;
1139
1140 root = kzalloc(sizeof(*root), GFP_NOFS);
1141 if (!root)
1142 return ERR_PTR(-ENOMEM);
1143
1144 __setup_root(tree_root->nodesize, tree_root->leafsize,
1145 tree_root->sectorsize, tree_root->stripesize,
1146 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1147
1148 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1149 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1150 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1151 /*
1152 * log trees do not get reference counted because they go away
1153 * before a real commit is actually done. They do store pointers
1154 * to file data extents, and those reference counts still get
1155 * updated (along with back refs to the log tree).
1156 */
1157 root->ref_cows = 0;
1158
1159 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1160 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1161 if (IS_ERR(leaf)) {
1162 kfree(root);
1163 return ERR_CAST(leaf);
1164 }
1165
1166 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1167 btrfs_set_header_bytenr(leaf, leaf->start);
1168 btrfs_set_header_generation(leaf, trans->transid);
1169 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1170 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1171 root->node = leaf;
1172
1173 write_extent_buffer(root->node, root->fs_info->fsid,
1174 (unsigned long)btrfs_header_fsid(root->node),
1175 BTRFS_FSID_SIZE);
1176 btrfs_mark_buffer_dirty(root->node);
1177 btrfs_tree_unlock(root->node);
1178 return root;
1179 }
1180
1181 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1182 struct btrfs_fs_info *fs_info)
1183 {
1184 struct btrfs_root *log_root;
1185
1186 log_root = alloc_log_tree(trans, fs_info);
1187 if (IS_ERR(log_root))
1188 return PTR_ERR(log_root);
1189 WARN_ON(fs_info->log_root_tree);
1190 fs_info->log_root_tree = log_root;
1191 return 0;
1192 }
1193
1194 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1195 struct btrfs_root *root)
1196 {
1197 struct btrfs_root *log_root;
1198 struct btrfs_inode_item *inode_item;
1199
1200 log_root = alloc_log_tree(trans, root->fs_info);
1201 if (IS_ERR(log_root))
1202 return PTR_ERR(log_root);
1203
1204 log_root->last_trans = trans->transid;
1205 log_root->root_key.offset = root->root_key.objectid;
1206
1207 inode_item = &log_root->root_item.inode;
1208 inode_item->generation = cpu_to_le64(1);
1209 inode_item->size = cpu_to_le64(3);
1210 inode_item->nlink = cpu_to_le32(1);
1211 inode_item->nbytes = cpu_to_le64(root->leafsize);
1212 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1213
1214 btrfs_set_root_node(&log_root->root_item, log_root->node);
1215
1216 WARN_ON(root->log_root);
1217 root->log_root = log_root;
1218 root->log_transid = 0;
1219 root->last_log_commit = 0;
1220 return 0;
1221 }
1222
1223 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1224 struct btrfs_key *location)
1225 {
1226 struct btrfs_root *root;
1227 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1228 struct btrfs_path *path;
1229 struct extent_buffer *l;
1230 u64 generation;
1231 u32 blocksize;
1232 int ret = 0;
1233
1234 root = kzalloc(sizeof(*root), GFP_NOFS);
1235 if (!root)
1236 return ERR_PTR(-ENOMEM);
1237 if (location->offset == (u64)-1) {
1238 ret = find_and_setup_root(tree_root, fs_info,
1239 location->objectid, root);
1240 if (ret) {
1241 kfree(root);
1242 return ERR_PTR(ret);
1243 }
1244 goto out;
1245 }
1246
1247 __setup_root(tree_root->nodesize, tree_root->leafsize,
1248 tree_root->sectorsize, tree_root->stripesize,
1249 root, fs_info, location->objectid);
1250
1251 path = btrfs_alloc_path();
1252 if (!path) {
1253 kfree(root);
1254 return ERR_PTR(-ENOMEM);
1255 }
1256 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1257 if (ret == 0) {
1258 l = path->nodes[0];
1259 read_extent_buffer(l, &root->root_item,
1260 btrfs_item_ptr_offset(l, path->slots[0]),
1261 sizeof(root->root_item));
1262 memcpy(&root->root_key, location, sizeof(*location));
1263 }
1264 btrfs_free_path(path);
1265 if (ret) {
1266 kfree(root);
1267 if (ret > 0)
1268 ret = -ENOENT;
1269 return ERR_PTR(ret);
1270 }
1271
1272 generation = btrfs_root_generation(&root->root_item);
1273 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1274 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1275 blocksize, generation);
1276 root->commit_root = btrfs_root_node(root);
1277 BUG_ON(!root->node);
1278 out:
1279 if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1280 root->ref_cows = 1;
1281
1282 return root;
1283 }
1284
1285 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1286 u64 root_objectid)
1287 {
1288 struct btrfs_root *root;
1289
1290 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1291 return fs_info->tree_root;
1292 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1293 return fs_info->extent_root;
1294
1295 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1296 (unsigned long)root_objectid);
1297 return root;
1298 }
1299
1300 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1301 struct btrfs_key *location)
1302 {
1303 struct btrfs_root *root;
1304 int ret;
1305
1306 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1307 return fs_info->tree_root;
1308 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1309 return fs_info->extent_root;
1310 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1311 return fs_info->chunk_root;
1312 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1313 return fs_info->dev_root;
1314 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1315 return fs_info->csum_root;
1316 again:
1317 spin_lock(&fs_info->fs_roots_radix_lock);
1318 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1319 (unsigned long)location->objectid);
1320 spin_unlock(&fs_info->fs_roots_radix_lock);
1321 if (root)
1322 return root;
1323
1324 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1325 if (IS_ERR(root))
1326 return root;
1327
1328 set_anon_super(&root->anon_super, NULL);
1329
1330 if (btrfs_root_refs(&root->root_item) == 0) {
1331 ret = -ENOENT;
1332 goto fail;
1333 }
1334
1335 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1336 if (ret < 0)
1337 goto fail;
1338 if (ret == 0)
1339 root->orphan_item_inserted = 1;
1340
1341 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1342 if (ret)
1343 goto fail;
1344
1345 spin_lock(&fs_info->fs_roots_radix_lock);
1346 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1347 (unsigned long)root->root_key.objectid,
1348 root);
1349 if (ret == 0)
1350 root->in_radix = 1;
1351
1352 spin_unlock(&fs_info->fs_roots_radix_lock);
1353 radix_tree_preload_end();
1354 if (ret) {
1355 if (ret == -EEXIST) {
1356 free_fs_root(root);
1357 goto again;
1358 }
1359 goto fail;
1360 }
1361
1362 ret = btrfs_find_dead_roots(fs_info->tree_root,
1363 root->root_key.objectid);
1364 WARN_ON(ret);
1365 return root;
1366 fail:
1367 free_fs_root(root);
1368 return ERR_PTR(ret);
1369 }
1370
1371 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1372 struct btrfs_key *location,
1373 const char *name, int namelen)
1374 {
1375 return btrfs_read_fs_root_no_name(fs_info, location);
1376 #if 0
1377 struct btrfs_root *root;
1378 int ret;
1379
1380 root = btrfs_read_fs_root_no_name(fs_info, location);
1381 if (!root)
1382 return NULL;
1383
1384 if (root->in_sysfs)
1385 return root;
1386
1387 ret = btrfs_set_root_name(root, name, namelen);
1388 if (ret) {
1389 free_extent_buffer(root->node);
1390 kfree(root);
1391 return ERR_PTR(ret);
1392 }
1393
1394 ret = btrfs_sysfs_add_root(root);
1395 if (ret) {
1396 free_extent_buffer(root->node);
1397 kfree(root->name);
1398 kfree(root);
1399 return ERR_PTR(ret);
1400 }
1401 root->in_sysfs = 1;
1402 return root;
1403 #endif
1404 }
1405
1406 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1407 {
1408 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1409 int ret = 0;
1410 struct btrfs_device *device;
1411 struct backing_dev_info *bdi;
1412
1413 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1414 if (!device->bdev)
1415 continue;
1416 bdi = blk_get_backing_dev_info(device->bdev);
1417 if (bdi && bdi_congested(bdi, bdi_bits)) {
1418 ret = 1;
1419 break;
1420 }
1421 }
1422 return ret;
1423 }
1424
1425 /*
1426 * this unplugs every device on the box, and it is only used when page
1427 * is null
1428 */
1429 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1430 {
1431 struct btrfs_device *device;
1432 struct btrfs_fs_info *info;
1433
1434 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1435 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1436 if (!device->bdev)
1437 continue;
1438
1439 bdi = blk_get_backing_dev_info(device->bdev);
1440 if (bdi->unplug_io_fn)
1441 bdi->unplug_io_fn(bdi, page);
1442 }
1443 }
1444
1445 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1446 {
1447 struct inode *inode;
1448 struct extent_map_tree *em_tree;
1449 struct extent_map *em;
1450 struct address_space *mapping;
1451 u64 offset;
1452
1453 /* the generic O_DIRECT read code does this */
1454 if (1 || !page) {
1455 __unplug_io_fn(bdi, page);
1456 return;
1457 }
1458
1459 /*
1460 * page->mapping may change at any time. Get a consistent copy
1461 * and use that for everything below
1462 */
1463 smp_mb();
1464 mapping = page->mapping;
1465 if (!mapping)
1466 return;
1467
1468 inode = mapping->host;
1469
1470 /*
1471 * don't do the expensive searching for a small number of
1472 * devices
1473 */
1474 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1475 __unplug_io_fn(bdi, page);
1476 return;
1477 }
1478
1479 offset = page_offset(page);
1480
1481 em_tree = &BTRFS_I(inode)->extent_tree;
1482 read_lock(&em_tree->lock);
1483 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1484 read_unlock(&em_tree->lock);
1485 if (!em) {
1486 __unplug_io_fn(bdi, page);
1487 return;
1488 }
1489
1490 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1491 free_extent_map(em);
1492 __unplug_io_fn(bdi, page);
1493 return;
1494 }
1495 offset = offset - em->start;
1496 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1497 em->block_start + offset, page);
1498 free_extent_map(em);
1499 }
1500
1501 /*
1502 * If this fails, caller must call bdi_destroy() to get rid of the
1503 * bdi again.
1504 */
1505 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1506 {
1507 int err;
1508
1509 bdi->capabilities = BDI_CAP_MAP_COPY;
1510 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1511 if (err)
1512 return err;
1513
1514 bdi->ra_pages = default_backing_dev_info.ra_pages;
1515 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1516 bdi->unplug_io_data = info;
1517 bdi->congested_fn = btrfs_congested_fn;
1518 bdi->congested_data = info;
1519 return 0;
1520 }
1521
1522 static int bio_ready_for_csum(struct bio *bio)
1523 {
1524 u64 length = 0;
1525 u64 buf_len = 0;
1526 u64 start = 0;
1527 struct page *page;
1528 struct extent_io_tree *io_tree = NULL;
1529 struct bio_vec *bvec;
1530 int i;
1531 int ret;
1532
1533 bio_for_each_segment(bvec, bio, i) {
1534 page = bvec->bv_page;
1535 if (page->private == EXTENT_PAGE_PRIVATE) {
1536 length += bvec->bv_len;
1537 continue;
1538 }
1539 if (!page->private) {
1540 length += bvec->bv_len;
1541 continue;
1542 }
1543 length = bvec->bv_len;
1544 buf_len = page->private >> 2;
1545 start = page_offset(page) + bvec->bv_offset;
1546 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1547 }
1548 /* are we fully contained in this bio? */
1549 if (buf_len <= length)
1550 return 1;
1551
1552 ret = extent_range_uptodate(io_tree, start + length,
1553 start + buf_len - 1);
1554 return ret;
1555 }
1556
1557 /*
1558 * called by the kthread helper functions to finally call the bio end_io
1559 * functions. This is where read checksum verification actually happens
1560 */
1561 static void end_workqueue_fn(struct btrfs_work *work)
1562 {
1563 struct bio *bio;
1564 struct end_io_wq *end_io_wq;
1565 struct btrfs_fs_info *fs_info;
1566 int error;
1567
1568 end_io_wq = container_of(work, struct end_io_wq, work);
1569 bio = end_io_wq->bio;
1570 fs_info = end_io_wq->info;
1571
1572 /* metadata bio reads are special because the whole tree block must
1573 * be checksummed at once. This makes sure the entire block is in
1574 * ram and up to date before trying to verify things. For
1575 * blocksize <= pagesize, it is basically a noop
1576 */
1577 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1578 !bio_ready_for_csum(bio)) {
1579 btrfs_queue_worker(&fs_info->endio_meta_workers,
1580 &end_io_wq->work);
1581 return;
1582 }
1583 error = end_io_wq->error;
1584 bio->bi_private = end_io_wq->private;
1585 bio->bi_end_io = end_io_wq->end_io;
1586 kfree(end_io_wq);
1587 bio_endio(bio, error);
1588 }
1589
1590 static int cleaner_kthread(void *arg)
1591 {
1592 struct btrfs_root *root = arg;
1593
1594 do {
1595 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1596
1597 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1598 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1599 btrfs_run_delayed_iputs(root);
1600 btrfs_clean_old_snapshots(root);
1601 mutex_unlock(&root->fs_info->cleaner_mutex);
1602 }
1603
1604 if (freezing(current)) {
1605 refrigerator();
1606 } else {
1607 set_current_state(TASK_INTERRUPTIBLE);
1608 if (!kthread_should_stop())
1609 schedule();
1610 __set_current_state(TASK_RUNNING);
1611 }
1612 } while (!kthread_should_stop());
1613 return 0;
1614 }
1615
1616 static int transaction_kthread(void *arg)
1617 {
1618 struct btrfs_root *root = arg;
1619 struct btrfs_trans_handle *trans;
1620 struct btrfs_transaction *cur;
1621 u64 transid;
1622 unsigned long now;
1623 unsigned long delay;
1624 int ret;
1625
1626 do {
1627 delay = HZ * 30;
1628 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1629 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1630
1631 spin_lock(&root->fs_info->new_trans_lock);
1632 cur = root->fs_info->running_transaction;
1633 if (!cur) {
1634 spin_unlock(&root->fs_info->new_trans_lock);
1635 goto sleep;
1636 }
1637
1638 now = get_seconds();
1639 if (!cur->blocked &&
1640 (now < cur->start_time || now - cur->start_time < 30)) {
1641 spin_unlock(&root->fs_info->new_trans_lock);
1642 delay = HZ * 5;
1643 goto sleep;
1644 }
1645 transid = cur->transid;
1646 spin_unlock(&root->fs_info->new_trans_lock);
1647
1648 trans = btrfs_join_transaction(root, 1);
1649 BUG_ON(IS_ERR(trans));
1650 if (transid == trans->transid) {
1651 ret = btrfs_commit_transaction(trans, root);
1652 BUG_ON(ret);
1653 } else {
1654 btrfs_end_transaction(trans, root);
1655 }
1656 sleep:
1657 wake_up_process(root->fs_info->cleaner_kthread);
1658 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1659
1660 if (freezing(current)) {
1661 refrigerator();
1662 } else {
1663 set_current_state(TASK_INTERRUPTIBLE);
1664 if (!kthread_should_stop() &&
1665 !btrfs_transaction_blocked(root->fs_info))
1666 schedule_timeout(delay);
1667 __set_current_state(TASK_RUNNING);
1668 }
1669 } while (!kthread_should_stop());
1670 return 0;
1671 }
1672
1673 struct btrfs_root *open_ctree(struct super_block *sb,
1674 struct btrfs_fs_devices *fs_devices,
1675 char *options)
1676 {
1677 u32 sectorsize;
1678 u32 nodesize;
1679 u32 leafsize;
1680 u32 blocksize;
1681 u32 stripesize;
1682 u64 generation;
1683 u64 features;
1684 struct btrfs_key location;
1685 struct buffer_head *bh;
1686 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1687 GFP_NOFS);
1688 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1689 GFP_NOFS);
1690 struct btrfs_root *tree_root = btrfs_sb(sb);
1691 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1692 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1693 GFP_NOFS);
1694 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1695 GFP_NOFS);
1696 struct btrfs_root *log_tree_root;
1697
1698 int ret;
1699 int err = -EINVAL;
1700
1701 struct btrfs_super_block *disk_super;
1702
1703 if (!extent_root || !tree_root || !fs_info ||
1704 !chunk_root || !dev_root || !csum_root) {
1705 err = -ENOMEM;
1706 goto fail;
1707 }
1708
1709 ret = init_srcu_struct(&fs_info->subvol_srcu);
1710 if (ret) {
1711 err = ret;
1712 goto fail;
1713 }
1714
1715 ret = setup_bdi(fs_info, &fs_info->bdi);
1716 if (ret) {
1717 err = ret;
1718 goto fail_srcu;
1719 }
1720
1721 fs_info->btree_inode = new_inode(sb);
1722 if (!fs_info->btree_inode) {
1723 err = -ENOMEM;
1724 goto fail_bdi;
1725 }
1726
1727 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1728 INIT_LIST_HEAD(&fs_info->trans_list);
1729 INIT_LIST_HEAD(&fs_info->dead_roots);
1730 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1731 INIT_LIST_HEAD(&fs_info->hashers);
1732 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1733 INIT_LIST_HEAD(&fs_info->ordered_operations);
1734 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1735 spin_lock_init(&fs_info->delalloc_lock);
1736 spin_lock_init(&fs_info->new_trans_lock);
1737 spin_lock_init(&fs_info->ref_cache_lock);
1738 spin_lock_init(&fs_info->fs_roots_radix_lock);
1739 spin_lock_init(&fs_info->delayed_iput_lock);
1740
1741 init_completion(&fs_info->kobj_unregister);
1742 fs_info->tree_root = tree_root;
1743 fs_info->extent_root = extent_root;
1744 fs_info->csum_root = csum_root;
1745 fs_info->chunk_root = chunk_root;
1746 fs_info->dev_root = dev_root;
1747 fs_info->fs_devices = fs_devices;
1748 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1749 INIT_LIST_HEAD(&fs_info->space_info);
1750 btrfs_mapping_init(&fs_info->mapping_tree);
1751 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1752 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1753 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1754 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1755 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1756 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1757 mutex_init(&fs_info->durable_block_rsv_mutex);
1758 atomic_set(&fs_info->nr_async_submits, 0);
1759 atomic_set(&fs_info->async_delalloc_pages, 0);
1760 atomic_set(&fs_info->async_submit_draining, 0);
1761 atomic_set(&fs_info->nr_async_bios, 0);
1762 fs_info->sb = sb;
1763 fs_info->max_inline = 8192 * 1024;
1764 fs_info->metadata_ratio = 0;
1765
1766 fs_info->thread_pool_size = min_t(unsigned long,
1767 num_online_cpus() + 2, 8);
1768
1769 INIT_LIST_HEAD(&fs_info->ordered_extents);
1770 spin_lock_init(&fs_info->ordered_extent_lock);
1771
1772 sb->s_blocksize = 4096;
1773 sb->s_blocksize_bits = blksize_bits(4096);
1774 sb->s_bdi = &fs_info->bdi;
1775
1776 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1777 fs_info->btree_inode->i_nlink = 1;
1778 /*
1779 * we set the i_size on the btree inode to the max possible int.
1780 * the real end of the address space is determined by all of
1781 * the devices in the system
1782 */
1783 fs_info->btree_inode->i_size = OFFSET_MAX;
1784 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1785 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1786
1787 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1788 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1789 fs_info->btree_inode->i_mapping,
1790 GFP_NOFS);
1791 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1792 GFP_NOFS);
1793
1794 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1795
1796 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1797 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1798 sizeof(struct btrfs_key));
1799 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1800 insert_inode_hash(fs_info->btree_inode);
1801
1802 spin_lock_init(&fs_info->block_group_cache_lock);
1803 fs_info->block_group_cache_tree = RB_ROOT;
1804
1805 extent_io_tree_init(&fs_info->freed_extents[0],
1806 fs_info->btree_inode->i_mapping, GFP_NOFS);
1807 extent_io_tree_init(&fs_info->freed_extents[1],
1808 fs_info->btree_inode->i_mapping, GFP_NOFS);
1809 fs_info->pinned_extents = &fs_info->freed_extents[0];
1810 fs_info->do_barriers = 1;
1811
1812
1813 mutex_init(&fs_info->trans_mutex);
1814 mutex_init(&fs_info->ordered_operations_mutex);
1815 mutex_init(&fs_info->tree_log_mutex);
1816 mutex_init(&fs_info->chunk_mutex);
1817 mutex_init(&fs_info->transaction_kthread_mutex);
1818 mutex_init(&fs_info->cleaner_mutex);
1819 mutex_init(&fs_info->volume_mutex);
1820 init_rwsem(&fs_info->extent_commit_sem);
1821 init_rwsem(&fs_info->cleanup_work_sem);
1822 init_rwsem(&fs_info->subvol_sem);
1823
1824 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1825 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1826
1827 init_waitqueue_head(&fs_info->transaction_throttle);
1828 init_waitqueue_head(&fs_info->transaction_wait);
1829 init_waitqueue_head(&fs_info->transaction_blocked_wait);
1830 init_waitqueue_head(&fs_info->async_submit_wait);
1831
1832 __setup_root(4096, 4096, 4096, 4096, tree_root,
1833 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1834
1835 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1836 if (!bh) {
1837 err = -EINVAL;
1838 goto fail_iput;
1839 }
1840
1841 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1842 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1843 sizeof(fs_info->super_for_commit));
1844 brelse(bh);
1845
1846 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1847
1848 disk_super = &fs_info->super_copy;
1849 if (!btrfs_super_root(disk_super))
1850 goto fail_iput;
1851
1852 /* check FS state, whether FS is broken. */
1853 fs_info->fs_state |= btrfs_super_flags(disk_super);
1854
1855 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1856
1857 /*
1858 * In the long term, we'll store the compression type in the super
1859 * block, and it'll be used for per file compression control.
1860 */
1861 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1862
1863 ret = btrfs_parse_options(tree_root, options);
1864 if (ret) {
1865 err = ret;
1866 goto fail_iput;
1867 }
1868
1869 features = btrfs_super_incompat_flags(disk_super) &
1870 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1871 if (features) {
1872 printk(KERN_ERR "BTRFS: couldn't mount because of "
1873 "unsupported optional features (%Lx).\n",
1874 (unsigned long long)features);
1875 err = -EINVAL;
1876 goto fail_iput;
1877 }
1878
1879 features = btrfs_super_incompat_flags(disk_super);
1880 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1881 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1882 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1883 btrfs_set_super_incompat_flags(disk_super, features);
1884
1885 features = btrfs_super_compat_ro_flags(disk_super) &
1886 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1887 if (!(sb->s_flags & MS_RDONLY) && features) {
1888 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1889 "unsupported option features (%Lx).\n",
1890 (unsigned long long)features);
1891 err = -EINVAL;
1892 goto fail_iput;
1893 }
1894
1895 btrfs_init_workers(&fs_info->generic_worker,
1896 "genwork", 1, NULL);
1897
1898 btrfs_init_workers(&fs_info->workers, "worker",
1899 fs_info->thread_pool_size,
1900 &fs_info->generic_worker);
1901
1902 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1903 fs_info->thread_pool_size,
1904 &fs_info->generic_worker);
1905
1906 btrfs_init_workers(&fs_info->submit_workers, "submit",
1907 min_t(u64, fs_devices->num_devices,
1908 fs_info->thread_pool_size),
1909 &fs_info->generic_worker);
1910
1911 /* a higher idle thresh on the submit workers makes it much more
1912 * likely that bios will be send down in a sane order to the
1913 * devices
1914 */
1915 fs_info->submit_workers.idle_thresh = 64;
1916
1917 fs_info->workers.idle_thresh = 16;
1918 fs_info->workers.ordered = 1;
1919
1920 fs_info->delalloc_workers.idle_thresh = 2;
1921 fs_info->delalloc_workers.ordered = 1;
1922
1923 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1924 &fs_info->generic_worker);
1925 btrfs_init_workers(&fs_info->endio_workers, "endio",
1926 fs_info->thread_pool_size,
1927 &fs_info->generic_worker);
1928 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1929 fs_info->thread_pool_size,
1930 &fs_info->generic_worker);
1931 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1932 "endio-meta-write", fs_info->thread_pool_size,
1933 &fs_info->generic_worker);
1934 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1935 fs_info->thread_pool_size,
1936 &fs_info->generic_worker);
1937 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1938 1, &fs_info->generic_worker);
1939
1940 /*
1941 * endios are largely parallel and should have a very
1942 * low idle thresh
1943 */
1944 fs_info->endio_workers.idle_thresh = 4;
1945 fs_info->endio_meta_workers.idle_thresh = 4;
1946
1947 fs_info->endio_write_workers.idle_thresh = 2;
1948 fs_info->endio_meta_write_workers.idle_thresh = 2;
1949
1950 btrfs_start_workers(&fs_info->workers, 1);
1951 btrfs_start_workers(&fs_info->generic_worker, 1);
1952 btrfs_start_workers(&fs_info->submit_workers, 1);
1953 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1954 btrfs_start_workers(&fs_info->fixup_workers, 1);
1955 btrfs_start_workers(&fs_info->endio_workers, 1);
1956 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1957 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1958 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1959 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1960
1961 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1962 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1963 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1964
1965 nodesize = btrfs_super_nodesize(disk_super);
1966 leafsize = btrfs_super_leafsize(disk_super);
1967 sectorsize = btrfs_super_sectorsize(disk_super);
1968 stripesize = btrfs_super_stripesize(disk_super);
1969 tree_root->nodesize = nodesize;
1970 tree_root->leafsize = leafsize;
1971 tree_root->sectorsize = sectorsize;
1972 tree_root->stripesize = stripesize;
1973
1974 sb->s_blocksize = sectorsize;
1975 sb->s_blocksize_bits = blksize_bits(sectorsize);
1976
1977 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1978 sizeof(disk_super->magic))) {
1979 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1980 goto fail_sb_buffer;
1981 }
1982
1983 mutex_lock(&fs_info->chunk_mutex);
1984 ret = btrfs_read_sys_array(tree_root);
1985 mutex_unlock(&fs_info->chunk_mutex);
1986 if (ret) {
1987 printk(KERN_WARNING "btrfs: failed to read the system "
1988 "array on %s\n", sb->s_id);
1989 goto fail_sb_buffer;
1990 }
1991
1992 blocksize = btrfs_level_size(tree_root,
1993 btrfs_super_chunk_root_level(disk_super));
1994 generation = btrfs_super_chunk_root_generation(disk_super);
1995
1996 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1997 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1998
1999 chunk_root->node = read_tree_block(chunk_root,
2000 btrfs_super_chunk_root(disk_super),
2001 blocksize, generation);
2002 BUG_ON(!chunk_root->node);
2003 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2004 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2005 sb->s_id);
2006 goto fail_chunk_root;
2007 }
2008 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2009 chunk_root->commit_root = btrfs_root_node(chunk_root);
2010
2011 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2012 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2013 BTRFS_UUID_SIZE);
2014
2015 mutex_lock(&fs_info->chunk_mutex);
2016 ret = btrfs_read_chunk_tree(chunk_root);
2017 mutex_unlock(&fs_info->chunk_mutex);
2018 if (ret) {
2019 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2020 sb->s_id);
2021 goto fail_chunk_root;
2022 }
2023
2024 btrfs_close_extra_devices(fs_devices);
2025
2026 blocksize = btrfs_level_size(tree_root,
2027 btrfs_super_root_level(disk_super));
2028 generation = btrfs_super_generation(disk_super);
2029
2030 tree_root->node = read_tree_block(tree_root,
2031 btrfs_super_root(disk_super),
2032 blocksize, generation);
2033 if (!tree_root->node)
2034 goto fail_chunk_root;
2035 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2036 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2037 sb->s_id);
2038 goto fail_tree_root;
2039 }
2040 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2041 tree_root->commit_root = btrfs_root_node(tree_root);
2042
2043 ret = find_and_setup_root(tree_root, fs_info,
2044 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2045 if (ret)
2046 goto fail_tree_root;
2047 extent_root->track_dirty = 1;
2048
2049 ret = find_and_setup_root(tree_root, fs_info,
2050 BTRFS_DEV_TREE_OBJECTID, dev_root);
2051 if (ret)
2052 goto fail_extent_root;
2053 dev_root->track_dirty = 1;
2054
2055 ret = find_and_setup_root(tree_root, fs_info,
2056 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2057 if (ret)
2058 goto fail_dev_root;
2059
2060 csum_root->track_dirty = 1;
2061
2062 fs_info->generation = generation;
2063 fs_info->last_trans_committed = generation;
2064 fs_info->data_alloc_profile = (u64)-1;
2065 fs_info->metadata_alloc_profile = (u64)-1;
2066 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2067
2068 ret = btrfs_init_space_info(fs_info);
2069 if (ret) {
2070 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2071 goto fail_block_groups;
2072 }
2073
2074 ret = btrfs_read_block_groups(extent_root);
2075 if (ret) {
2076 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2077 goto fail_block_groups;
2078 }
2079
2080 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2081 "btrfs-cleaner");
2082 if (IS_ERR(fs_info->cleaner_kthread))
2083 goto fail_block_groups;
2084
2085 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2086 tree_root,
2087 "btrfs-transaction");
2088 if (IS_ERR(fs_info->transaction_kthread))
2089 goto fail_cleaner;
2090
2091 if (!btrfs_test_opt(tree_root, SSD) &&
2092 !btrfs_test_opt(tree_root, NOSSD) &&
2093 !fs_info->fs_devices->rotating) {
2094 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2095 "mode\n");
2096 btrfs_set_opt(fs_info->mount_opt, SSD);
2097 }
2098
2099 /* do not make disk changes in broken FS */
2100 if (btrfs_super_log_root(disk_super) != 0 &&
2101 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2102 u64 bytenr = btrfs_super_log_root(disk_super);
2103
2104 if (fs_devices->rw_devices == 0) {
2105 printk(KERN_WARNING "Btrfs log replay required "
2106 "on RO media\n");
2107 err = -EIO;
2108 goto fail_trans_kthread;
2109 }
2110 blocksize =
2111 btrfs_level_size(tree_root,
2112 btrfs_super_log_root_level(disk_super));
2113
2114 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2115 if (!log_tree_root) {
2116 err = -ENOMEM;
2117 goto fail_trans_kthread;
2118 }
2119
2120 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2121 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2122
2123 log_tree_root->node = read_tree_block(tree_root, bytenr,
2124 blocksize,
2125 generation + 1);
2126 ret = btrfs_recover_log_trees(log_tree_root);
2127 BUG_ON(ret);
2128
2129 if (sb->s_flags & MS_RDONLY) {
2130 ret = btrfs_commit_super(tree_root);
2131 BUG_ON(ret);
2132 }
2133 }
2134
2135 ret = btrfs_find_orphan_roots(tree_root);
2136 BUG_ON(ret);
2137
2138 if (!(sb->s_flags & MS_RDONLY)) {
2139 ret = btrfs_cleanup_fs_roots(fs_info);
2140 BUG_ON(ret);
2141
2142 ret = btrfs_recover_relocation(tree_root);
2143 if (ret < 0) {
2144 printk(KERN_WARNING
2145 "btrfs: failed to recover relocation\n");
2146 err = -EINVAL;
2147 goto fail_trans_kthread;
2148 }
2149 }
2150
2151 location.objectid = BTRFS_FS_TREE_OBJECTID;
2152 location.type = BTRFS_ROOT_ITEM_KEY;
2153 location.offset = (u64)-1;
2154
2155 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2156 if (!fs_info->fs_root)
2157 goto fail_trans_kthread;
2158 if (IS_ERR(fs_info->fs_root)) {
2159 err = PTR_ERR(fs_info->fs_root);
2160 goto fail_trans_kthread;
2161 }
2162
2163 if (!(sb->s_flags & MS_RDONLY)) {
2164 down_read(&fs_info->cleanup_work_sem);
2165 err = btrfs_orphan_cleanup(fs_info->fs_root);
2166 if (!err)
2167 err = btrfs_orphan_cleanup(fs_info->tree_root);
2168 up_read(&fs_info->cleanup_work_sem);
2169 if (err) {
2170 close_ctree(tree_root);
2171 return ERR_PTR(err);
2172 }
2173 }
2174
2175 return tree_root;
2176
2177 fail_trans_kthread:
2178 kthread_stop(fs_info->transaction_kthread);
2179 fail_cleaner:
2180 kthread_stop(fs_info->cleaner_kthread);
2181
2182 /*
2183 * make sure we're done with the btree inode before we stop our
2184 * kthreads
2185 */
2186 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2187 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2188
2189 fail_block_groups:
2190 btrfs_free_block_groups(fs_info);
2191 free_extent_buffer(csum_root->node);
2192 free_extent_buffer(csum_root->commit_root);
2193 fail_dev_root:
2194 free_extent_buffer(dev_root->node);
2195 free_extent_buffer(dev_root->commit_root);
2196 fail_extent_root:
2197 free_extent_buffer(extent_root->node);
2198 free_extent_buffer(extent_root->commit_root);
2199 fail_tree_root:
2200 free_extent_buffer(tree_root->node);
2201 free_extent_buffer(tree_root->commit_root);
2202 fail_chunk_root:
2203 free_extent_buffer(chunk_root->node);
2204 free_extent_buffer(chunk_root->commit_root);
2205 fail_sb_buffer:
2206 btrfs_stop_workers(&fs_info->generic_worker);
2207 btrfs_stop_workers(&fs_info->fixup_workers);
2208 btrfs_stop_workers(&fs_info->delalloc_workers);
2209 btrfs_stop_workers(&fs_info->workers);
2210 btrfs_stop_workers(&fs_info->endio_workers);
2211 btrfs_stop_workers(&fs_info->endio_meta_workers);
2212 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2213 btrfs_stop_workers(&fs_info->endio_write_workers);
2214 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2215 btrfs_stop_workers(&fs_info->submit_workers);
2216 fail_iput:
2217 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2218 iput(fs_info->btree_inode);
2219
2220 btrfs_close_devices(fs_info->fs_devices);
2221 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2222 fail_bdi:
2223 bdi_destroy(&fs_info->bdi);
2224 fail_srcu:
2225 cleanup_srcu_struct(&fs_info->subvol_srcu);
2226 fail:
2227 kfree(extent_root);
2228 kfree(tree_root);
2229 kfree(fs_info);
2230 kfree(chunk_root);
2231 kfree(dev_root);
2232 kfree(csum_root);
2233 return ERR_PTR(err);
2234 }
2235
2236 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2237 {
2238 char b[BDEVNAME_SIZE];
2239
2240 if (uptodate) {
2241 set_buffer_uptodate(bh);
2242 } else {
2243 if (printk_ratelimit()) {
2244 printk(KERN_WARNING "lost page write due to "
2245 "I/O error on %s\n",
2246 bdevname(bh->b_bdev, b));
2247 }
2248 /* note, we dont' set_buffer_write_io_error because we have
2249 * our own ways of dealing with the IO errors
2250 */
2251 clear_buffer_uptodate(bh);
2252 }
2253 unlock_buffer(bh);
2254 put_bh(bh);
2255 }
2256
2257 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2258 {
2259 struct buffer_head *bh;
2260 struct buffer_head *latest = NULL;
2261 struct btrfs_super_block *super;
2262 int i;
2263 u64 transid = 0;
2264 u64 bytenr;
2265
2266 /* we would like to check all the supers, but that would make
2267 * a btrfs mount succeed after a mkfs from a different FS.
2268 * So, we need to add a special mount option to scan for
2269 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2270 */
2271 for (i = 0; i < 1; i++) {
2272 bytenr = btrfs_sb_offset(i);
2273 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2274 break;
2275 bh = __bread(bdev, bytenr / 4096, 4096);
2276 if (!bh)
2277 continue;
2278
2279 super = (struct btrfs_super_block *)bh->b_data;
2280 if (btrfs_super_bytenr(super) != bytenr ||
2281 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2282 sizeof(super->magic))) {
2283 brelse(bh);
2284 continue;
2285 }
2286
2287 if (!latest || btrfs_super_generation(super) > transid) {
2288 brelse(latest);
2289 latest = bh;
2290 transid = btrfs_super_generation(super);
2291 } else {
2292 brelse(bh);
2293 }
2294 }
2295 return latest;
2296 }
2297
2298 /*
2299 * this should be called twice, once with wait == 0 and
2300 * once with wait == 1. When wait == 0 is done, all the buffer heads
2301 * we write are pinned.
2302 *
2303 * They are released when wait == 1 is done.
2304 * max_mirrors must be the same for both runs, and it indicates how
2305 * many supers on this one device should be written.
2306 *
2307 * max_mirrors == 0 means to write them all.
2308 */
2309 static int write_dev_supers(struct btrfs_device *device,
2310 struct btrfs_super_block *sb,
2311 int do_barriers, int wait, int max_mirrors)
2312 {
2313 struct buffer_head *bh;
2314 int i;
2315 int ret;
2316 int errors = 0;
2317 u32 crc;
2318 u64 bytenr;
2319 int last_barrier = 0;
2320
2321 if (max_mirrors == 0)
2322 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2323
2324 /* make sure only the last submit_bh does a barrier */
2325 if (do_barriers) {
2326 for (i = 0; i < max_mirrors; i++) {
2327 bytenr = btrfs_sb_offset(i);
2328 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2329 device->total_bytes)
2330 break;
2331 last_barrier = i;
2332 }
2333 }
2334
2335 for (i = 0; i < max_mirrors; i++) {
2336 bytenr = btrfs_sb_offset(i);
2337 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2338 break;
2339
2340 if (wait) {
2341 bh = __find_get_block(device->bdev, bytenr / 4096,
2342 BTRFS_SUPER_INFO_SIZE);
2343 BUG_ON(!bh);
2344 wait_on_buffer(bh);
2345 if (!buffer_uptodate(bh))
2346 errors++;
2347
2348 /* drop our reference */
2349 brelse(bh);
2350
2351 /* drop the reference from the wait == 0 run */
2352 brelse(bh);
2353 continue;
2354 } else {
2355 btrfs_set_super_bytenr(sb, bytenr);
2356
2357 crc = ~(u32)0;
2358 crc = btrfs_csum_data(NULL, (char *)sb +
2359 BTRFS_CSUM_SIZE, crc,
2360 BTRFS_SUPER_INFO_SIZE -
2361 BTRFS_CSUM_SIZE);
2362 btrfs_csum_final(crc, sb->csum);
2363
2364 /*
2365 * one reference for us, and we leave it for the
2366 * caller
2367 */
2368 bh = __getblk(device->bdev, bytenr / 4096,
2369 BTRFS_SUPER_INFO_SIZE);
2370 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2371
2372 /* one reference for submit_bh */
2373 get_bh(bh);
2374
2375 set_buffer_uptodate(bh);
2376 lock_buffer(bh);
2377 bh->b_end_io = btrfs_end_buffer_write_sync;
2378 }
2379
2380 if (i == last_barrier && do_barriers)
2381 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2382 else
2383 ret = submit_bh(WRITE_SYNC, bh);
2384
2385 if (ret)
2386 errors++;
2387 }
2388 return errors < i ? 0 : -1;
2389 }
2390
2391 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2392 {
2393 struct list_head *head;
2394 struct btrfs_device *dev;
2395 struct btrfs_super_block *sb;
2396 struct btrfs_dev_item *dev_item;
2397 int ret;
2398 int do_barriers;
2399 int max_errors;
2400 int total_errors = 0;
2401 u64 flags;
2402
2403 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2404 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2405
2406 sb = &root->fs_info->super_for_commit;
2407 dev_item = &sb->dev_item;
2408
2409 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2410 head = &root->fs_info->fs_devices->devices;
2411 list_for_each_entry(dev, head, dev_list) {
2412 if (!dev->bdev) {
2413 total_errors++;
2414 continue;
2415 }
2416 if (!dev->in_fs_metadata || !dev->writeable)
2417 continue;
2418
2419 btrfs_set_stack_device_generation(dev_item, 0);
2420 btrfs_set_stack_device_type(dev_item, dev->type);
2421 btrfs_set_stack_device_id(dev_item, dev->devid);
2422 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2423 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2424 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2425 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2426 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2427 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2428 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2429
2430 flags = btrfs_super_flags(sb);
2431 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2432
2433 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2434 if (ret)
2435 total_errors++;
2436 }
2437 if (total_errors > max_errors) {
2438 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2439 total_errors);
2440 BUG();
2441 }
2442
2443 total_errors = 0;
2444 list_for_each_entry(dev, head, dev_list) {
2445 if (!dev->bdev)
2446 continue;
2447 if (!dev->in_fs_metadata || !dev->writeable)
2448 continue;
2449
2450 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2451 if (ret)
2452 total_errors++;
2453 }
2454 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2455 if (total_errors > max_errors) {
2456 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2457 total_errors);
2458 BUG();
2459 }
2460 return 0;
2461 }
2462
2463 int write_ctree_super(struct btrfs_trans_handle *trans,
2464 struct btrfs_root *root, int max_mirrors)
2465 {
2466 int ret;
2467
2468 ret = write_all_supers(root, max_mirrors);
2469 return ret;
2470 }
2471
2472 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2473 {
2474 spin_lock(&fs_info->fs_roots_radix_lock);
2475 radix_tree_delete(&fs_info->fs_roots_radix,
2476 (unsigned long)root->root_key.objectid);
2477 spin_unlock(&fs_info->fs_roots_radix_lock);
2478
2479 if (btrfs_root_refs(&root->root_item) == 0)
2480 synchronize_srcu(&fs_info->subvol_srcu);
2481
2482 free_fs_root(root);
2483 return 0;
2484 }
2485
2486 static void free_fs_root(struct btrfs_root *root)
2487 {
2488 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2489 if (root->anon_super.s_dev) {
2490 down_write(&root->anon_super.s_umount);
2491 kill_anon_super(&root->anon_super);
2492 }
2493 free_extent_buffer(root->node);
2494 free_extent_buffer(root->commit_root);
2495 kfree(root->name);
2496 kfree(root);
2497 }
2498
2499 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2500 {
2501 int ret;
2502 struct btrfs_root *gang[8];
2503 int i;
2504
2505 while (!list_empty(&fs_info->dead_roots)) {
2506 gang[0] = list_entry(fs_info->dead_roots.next,
2507 struct btrfs_root, root_list);
2508 list_del(&gang[0]->root_list);
2509
2510 if (gang[0]->in_radix) {
2511 btrfs_free_fs_root(fs_info, gang[0]);
2512 } else {
2513 free_extent_buffer(gang[0]->node);
2514 free_extent_buffer(gang[0]->commit_root);
2515 kfree(gang[0]);
2516 }
2517 }
2518
2519 while (1) {
2520 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2521 (void **)gang, 0,
2522 ARRAY_SIZE(gang));
2523 if (!ret)
2524 break;
2525 for (i = 0; i < ret; i++)
2526 btrfs_free_fs_root(fs_info, gang[i]);
2527 }
2528 return 0;
2529 }
2530
2531 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2532 {
2533 u64 root_objectid = 0;
2534 struct btrfs_root *gang[8];
2535 int i;
2536 int ret;
2537
2538 while (1) {
2539 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2540 (void **)gang, root_objectid,
2541 ARRAY_SIZE(gang));
2542 if (!ret)
2543 break;
2544
2545 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2546 for (i = 0; i < ret; i++) {
2547 int err;
2548
2549 root_objectid = gang[i]->root_key.objectid;
2550 err = btrfs_orphan_cleanup(gang[i]);
2551 if (err)
2552 return err;
2553 }
2554 root_objectid++;
2555 }
2556 return 0;
2557 }
2558
2559 int btrfs_commit_super(struct btrfs_root *root)
2560 {
2561 struct btrfs_trans_handle *trans;
2562 int ret;
2563
2564 mutex_lock(&root->fs_info->cleaner_mutex);
2565 btrfs_run_delayed_iputs(root);
2566 btrfs_clean_old_snapshots(root);
2567 mutex_unlock(&root->fs_info->cleaner_mutex);
2568
2569 /* wait until ongoing cleanup work done */
2570 down_write(&root->fs_info->cleanup_work_sem);
2571 up_write(&root->fs_info->cleanup_work_sem);
2572
2573 trans = btrfs_join_transaction(root, 1);
2574 if (IS_ERR(trans))
2575 return PTR_ERR(trans);
2576 ret = btrfs_commit_transaction(trans, root);
2577 BUG_ON(ret);
2578 /* run commit again to drop the original snapshot */
2579 trans = btrfs_join_transaction(root, 1);
2580 if (IS_ERR(trans))
2581 return PTR_ERR(trans);
2582 btrfs_commit_transaction(trans, root);
2583 ret = btrfs_write_and_wait_transaction(NULL, root);
2584 BUG_ON(ret);
2585
2586 ret = write_ctree_super(NULL, root, 0);
2587 return ret;
2588 }
2589
2590 int close_ctree(struct btrfs_root *root)
2591 {
2592 struct btrfs_fs_info *fs_info = root->fs_info;
2593 int ret;
2594
2595 fs_info->closing = 1;
2596 smp_mb();
2597
2598 btrfs_put_block_group_cache(fs_info);
2599
2600 /*
2601 * Here come 2 situations when btrfs is broken to flip readonly:
2602 *
2603 * 1. when btrfs flips readonly somewhere else before
2604 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2605 * and btrfs will skip to write sb directly to keep
2606 * ERROR state on disk.
2607 *
2608 * 2. when btrfs flips readonly just in btrfs_commit_super,
2609 * and in such case, btrfs cannnot write sb via btrfs_commit_super,
2610 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2611 * btrfs will cleanup all FS resources first and write sb then.
2612 */
2613 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2614 ret = btrfs_commit_super(root);
2615 if (ret)
2616 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2617 }
2618
2619 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2620 ret = btrfs_error_commit_super(root);
2621 if (ret)
2622 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2623 }
2624
2625 kthread_stop(root->fs_info->transaction_kthread);
2626 kthread_stop(root->fs_info->cleaner_kthread);
2627
2628 fs_info->closing = 2;
2629 smp_mb();
2630
2631 if (fs_info->delalloc_bytes) {
2632 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2633 (unsigned long long)fs_info->delalloc_bytes);
2634 }
2635 if (fs_info->total_ref_cache_size) {
2636 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2637 (unsigned long long)fs_info->total_ref_cache_size);
2638 }
2639
2640 free_extent_buffer(fs_info->extent_root->node);
2641 free_extent_buffer(fs_info->extent_root->commit_root);
2642 free_extent_buffer(fs_info->tree_root->node);
2643 free_extent_buffer(fs_info->tree_root->commit_root);
2644 free_extent_buffer(root->fs_info->chunk_root->node);
2645 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2646 free_extent_buffer(root->fs_info->dev_root->node);
2647 free_extent_buffer(root->fs_info->dev_root->commit_root);
2648 free_extent_buffer(root->fs_info->csum_root->node);
2649 free_extent_buffer(root->fs_info->csum_root->commit_root);
2650
2651 btrfs_free_block_groups(root->fs_info);
2652
2653 del_fs_roots(fs_info);
2654
2655 iput(fs_info->btree_inode);
2656
2657 btrfs_stop_workers(&fs_info->generic_worker);
2658 btrfs_stop_workers(&fs_info->fixup_workers);
2659 btrfs_stop_workers(&fs_info->delalloc_workers);
2660 btrfs_stop_workers(&fs_info->workers);
2661 btrfs_stop_workers(&fs_info->endio_workers);
2662 btrfs_stop_workers(&fs_info->endio_meta_workers);
2663 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2664 btrfs_stop_workers(&fs_info->endio_write_workers);
2665 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2666 btrfs_stop_workers(&fs_info->submit_workers);
2667
2668 btrfs_close_devices(fs_info->fs_devices);
2669 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2670
2671 bdi_destroy(&fs_info->bdi);
2672 cleanup_srcu_struct(&fs_info->subvol_srcu);
2673
2674 kfree(fs_info->extent_root);
2675 kfree(fs_info->tree_root);
2676 kfree(fs_info->chunk_root);
2677 kfree(fs_info->dev_root);
2678 kfree(fs_info->csum_root);
2679 kfree(fs_info);
2680
2681 return 0;
2682 }
2683
2684 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2685 {
2686 int ret;
2687 struct inode *btree_inode = buf->first_page->mapping->host;
2688
2689 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2690 NULL);
2691 if (!ret)
2692 return ret;
2693
2694 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2695 parent_transid);
2696 return !ret;
2697 }
2698
2699 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2700 {
2701 struct inode *btree_inode = buf->first_page->mapping->host;
2702 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2703 buf);
2704 }
2705
2706 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2707 {
2708 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2709 u64 transid = btrfs_header_generation(buf);
2710 struct inode *btree_inode = root->fs_info->btree_inode;
2711 int was_dirty;
2712
2713 btrfs_assert_tree_locked(buf);
2714 if (transid != root->fs_info->generation) {
2715 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2716 "found %llu running %llu\n",
2717 (unsigned long long)buf->start,
2718 (unsigned long long)transid,
2719 (unsigned long long)root->fs_info->generation);
2720 WARN_ON(1);
2721 }
2722 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2723 buf);
2724 if (!was_dirty) {
2725 spin_lock(&root->fs_info->delalloc_lock);
2726 root->fs_info->dirty_metadata_bytes += buf->len;
2727 spin_unlock(&root->fs_info->delalloc_lock);
2728 }
2729 }
2730
2731 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2732 {
2733 /*
2734 * looks as though older kernels can get into trouble with
2735 * this code, they end up stuck in balance_dirty_pages forever
2736 */
2737 u64 num_dirty;
2738 unsigned long thresh = 32 * 1024 * 1024;
2739
2740 if (current->flags & PF_MEMALLOC)
2741 return;
2742
2743 num_dirty = root->fs_info->dirty_metadata_bytes;
2744
2745 if (num_dirty > thresh) {
2746 balance_dirty_pages_ratelimited_nr(
2747 root->fs_info->btree_inode->i_mapping, 1);
2748 }
2749 return;
2750 }
2751
2752 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2753 {
2754 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2755 int ret;
2756 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2757 if (ret == 0)
2758 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2759 return ret;
2760 }
2761
2762 int btree_lock_page_hook(struct page *page)
2763 {
2764 struct inode *inode = page->mapping->host;
2765 struct btrfs_root *root = BTRFS_I(inode)->root;
2766 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2767 struct extent_buffer *eb;
2768 unsigned long len;
2769 u64 bytenr = page_offset(page);
2770
2771 if (page->private == EXTENT_PAGE_PRIVATE)
2772 goto out;
2773
2774 len = page->private >> 2;
2775 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2776 if (!eb)
2777 goto out;
2778
2779 btrfs_tree_lock(eb);
2780 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2781
2782 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2783 spin_lock(&root->fs_info->delalloc_lock);
2784 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2785 root->fs_info->dirty_metadata_bytes -= eb->len;
2786 else
2787 WARN_ON(1);
2788 spin_unlock(&root->fs_info->delalloc_lock);
2789 }
2790
2791 btrfs_tree_unlock(eb);
2792 free_extent_buffer(eb);
2793 out:
2794 lock_page(page);
2795 return 0;
2796 }
2797
2798 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2799 int read_only)
2800 {
2801 if (read_only)
2802 return;
2803
2804 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2805 printk(KERN_WARNING "warning: mount fs with errors, "
2806 "running btrfsck is recommended\n");
2807 }
2808
2809 int btrfs_error_commit_super(struct btrfs_root *root)
2810 {
2811 int ret;
2812
2813 mutex_lock(&root->fs_info->cleaner_mutex);
2814 btrfs_run_delayed_iputs(root);
2815 mutex_unlock(&root->fs_info->cleaner_mutex);
2816
2817 down_write(&root->fs_info->cleanup_work_sem);
2818 up_write(&root->fs_info->cleanup_work_sem);
2819
2820 /* cleanup FS via transaction */
2821 btrfs_cleanup_transaction(root);
2822
2823 ret = write_ctree_super(NULL, root, 0);
2824
2825 return ret;
2826 }
2827
2828 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2829 {
2830 struct btrfs_inode *btrfs_inode;
2831 struct list_head splice;
2832
2833 INIT_LIST_HEAD(&splice);
2834
2835 mutex_lock(&root->fs_info->ordered_operations_mutex);
2836 spin_lock(&root->fs_info->ordered_extent_lock);
2837
2838 list_splice_init(&root->fs_info->ordered_operations, &splice);
2839 while (!list_empty(&splice)) {
2840 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2841 ordered_operations);
2842
2843 list_del_init(&btrfs_inode->ordered_operations);
2844
2845 btrfs_invalidate_inodes(btrfs_inode->root);
2846 }
2847
2848 spin_unlock(&root->fs_info->ordered_extent_lock);
2849 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2850
2851 return 0;
2852 }
2853
2854 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2855 {
2856 struct list_head splice;
2857 struct btrfs_ordered_extent *ordered;
2858 struct inode *inode;
2859
2860 INIT_LIST_HEAD(&splice);
2861
2862 spin_lock(&root->fs_info->ordered_extent_lock);
2863
2864 list_splice_init(&root->fs_info->ordered_extents, &splice);
2865 while (!list_empty(&splice)) {
2866 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2867 root_extent_list);
2868
2869 list_del_init(&ordered->root_extent_list);
2870 atomic_inc(&ordered->refs);
2871
2872 /* the inode may be getting freed (in sys_unlink path). */
2873 inode = igrab(ordered->inode);
2874
2875 spin_unlock(&root->fs_info->ordered_extent_lock);
2876 if (inode)
2877 iput(inode);
2878
2879 atomic_set(&ordered->refs, 1);
2880 btrfs_put_ordered_extent(ordered);
2881
2882 spin_lock(&root->fs_info->ordered_extent_lock);
2883 }
2884
2885 spin_unlock(&root->fs_info->ordered_extent_lock);
2886
2887 return 0;
2888 }
2889
2890 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2891 struct btrfs_root *root)
2892 {
2893 struct rb_node *node;
2894 struct btrfs_delayed_ref_root *delayed_refs;
2895 struct btrfs_delayed_ref_node *ref;
2896 int ret = 0;
2897
2898 delayed_refs = &trans->delayed_refs;
2899
2900 spin_lock(&delayed_refs->lock);
2901 if (delayed_refs->num_entries == 0) {
2902 printk(KERN_INFO "delayed_refs has NO entry\n");
2903 return ret;
2904 }
2905
2906 node = rb_first(&delayed_refs->root);
2907 while (node) {
2908 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2909 node = rb_next(node);
2910
2911 ref->in_tree = 0;
2912 rb_erase(&ref->rb_node, &delayed_refs->root);
2913 delayed_refs->num_entries--;
2914
2915 atomic_set(&ref->refs, 1);
2916 if (btrfs_delayed_ref_is_head(ref)) {
2917 struct btrfs_delayed_ref_head *head;
2918
2919 head = btrfs_delayed_node_to_head(ref);
2920 mutex_lock(&head->mutex);
2921 kfree(head->extent_op);
2922 delayed_refs->num_heads--;
2923 if (list_empty(&head->cluster))
2924 delayed_refs->num_heads_ready--;
2925 list_del_init(&head->cluster);
2926 mutex_unlock(&head->mutex);
2927 }
2928
2929 spin_unlock(&delayed_refs->lock);
2930 btrfs_put_delayed_ref(ref);
2931
2932 cond_resched();
2933 spin_lock(&delayed_refs->lock);
2934 }
2935
2936 spin_unlock(&delayed_refs->lock);
2937
2938 return ret;
2939 }
2940
2941 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2942 {
2943 struct btrfs_pending_snapshot *snapshot;
2944 struct list_head splice;
2945
2946 INIT_LIST_HEAD(&splice);
2947
2948 list_splice_init(&t->pending_snapshots, &splice);
2949
2950 while (!list_empty(&splice)) {
2951 snapshot = list_entry(splice.next,
2952 struct btrfs_pending_snapshot,
2953 list);
2954
2955 list_del_init(&snapshot->list);
2956
2957 kfree(snapshot);
2958 }
2959
2960 return 0;
2961 }
2962
2963 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2964 {
2965 struct btrfs_inode *btrfs_inode;
2966 struct list_head splice;
2967
2968 INIT_LIST_HEAD(&splice);
2969
2970 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2971
2972 spin_lock(&root->fs_info->delalloc_lock);
2973
2974 while (!list_empty(&splice)) {
2975 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2976 delalloc_inodes);
2977
2978 list_del_init(&btrfs_inode->delalloc_inodes);
2979
2980 btrfs_invalidate_inodes(btrfs_inode->root);
2981 }
2982
2983 spin_unlock(&root->fs_info->delalloc_lock);
2984
2985 return 0;
2986 }
2987
2988 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2989 struct extent_io_tree *dirty_pages,
2990 int mark)
2991 {
2992 int ret;
2993 struct page *page;
2994 struct inode *btree_inode = root->fs_info->btree_inode;
2995 struct extent_buffer *eb;
2996 u64 start = 0;
2997 u64 end;
2998 u64 offset;
2999 unsigned long index;
3000
3001 while (1) {
3002 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3003 mark);
3004 if (ret)
3005 break;
3006
3007 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3008 while (start <= end) {
3009 index = start >> PAGE_CACHE_SHIFT;
3010 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3011 page = find_get_page(btree_inode->i_mapping, index);
3012 if (!page)
3013 continue;
3014 offset = page_offset(page);
3015
3016 spin_lock(&dirty_pages->buffer_lock);
3017 eb = radix_tree_lookup(
3018 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3019 offset >> PAGE_CACHE_SHIFT);
3020 spin_unlock(&dirty_pages->buffer_lock);
3021 if (eb) {
3022 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3023 &eb->bflags);
3024 atomic_set(&eb->refs, 1);
3025 }
3026 if (PageWriteback(page))
3027 end_page_writeback(page);
3028
3029 lock_page(page);
3030 if (PageDirty(page)) {
3031 clear_page_dirty_for_io(page);
3032 spin_lock_irq(&page->mapping->tree_lock);
3033 radix_tree_tag_clear(&page->mapping->page_tree,
3034 page_index(page),
3035 PAGECACHE_TAG_DIRTY);
3036 spin_unlock_irq(&page->mapping->tree_lock);
3037 }
3038
3039 page->mapping->a_ops->invalidatepage(page, 0);
3040 unlock_page(page);
3041 }
3042 }
3043
3044 return ret;
3045 }
3046
3047 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3048 struct extent_io_tree *pinned_extents)
3049 {
3050 struct extent_io_tree *unpin;
3051 u64 start;
3052 u64 end;
3053 int ret;
3054
3055 unpin = pinned_extents;
3056 while (1) {
3057 ret = find_first_extent_bit(unpin, 0, &start, &end,
3058 EXTENT_DIRTY);
3059 if (ret)
3060 break;
3061
3062 /* opt_discard */
3063 if (btrfs_test_opt(root, DISCARD))
3064 ret = btrfs_error_discard_extent(root, start,
3065 end + 1 - start,
3066 NULL);
3067
3068 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3069 btrfs_error_unpin_extent_range(root, start, end);
3070 cond_resched();
3071 }
3072
3073 return 0;
3074 }
3075
3076 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3077 {
3078 struct btrfs_transaction *t;
3079 LIST_HEAD(list);
3080
3081 WARN_ON(1);
3082
3083 mutex_lock(&root->fs_info->trans_mutex);
3084 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3085
3086 list_splice_init(&root->fs_info->trans_list, &list);
3087 while (!list_empty(&list)) {
3088 t = list_entry(list.next, struct btrfs_transaction, list);
3089 if (!t)
3090 break;
3091
3092 btrfs_destroy_ordered_operations(root);
3093
3094 btrfs_destroy_ordered_extents(root);
3095
3096 btrfs_destroy_delayed_refs(t, root);
3097
3098 btrfs_block_rsv_release(root,
3099 &root->fs_info->trans_block_rsv,
3100 t->dirty_pages.dirty_bytes);
3101
3102 /* FIXME: cleanup wait for commit */
3103 t->in_commit = 1;
3104 t->blocked = 1;
3105 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3106 wake_up(&root->fs_info->transaction_blocked_wait);
3107
3108 t->blocked = 0;
3109 if (waitqueue_active(&root->fs_info->transaction_wait))
3110 wake_up(&root->fs_info->transaction_wait);
3111 mutex_unlock(&root->fs_info->trans_mutex);
3112
3113 mutex_lock(&root->fs_info->trans_mutex);
3114 t->commit_done = 1;
3115 if (waitqueue_active(&t->commit_wait))
3116 wake_up(&t->commit_wait);
3117 mutex_unlock(&root->fs_info->trans_mutex);
3118
3119 mutex_lock(&root->fs_info->trans_mutex);
3120
3121 btrfs_destroy_pending_snapshots(t);
3122
3123 btrfs_destroy_delalloc_inodes(root);
3124
3125 spin_lock(&root->fs_info->new_trans_lock);
3126 root->fs_info->running_transaction = NULL;
3127 spin_unlock(&root->fs_info->new_trans_lock);
3128
3129 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3130 EXTENT_DIRTY);
3131
3132 btrfs_destroy_pinned_extent(root,
3133 root->fs_info->pinned_extents);
3134
3135 t->use_count = 0;
3136 list_del_init(&t->list);
3137 memset(t, 0, sizeof(*t));
3138 kmem_cache_free(btrfs_transaction_cachep, t);
3139 }
3140
3141 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3142 mutex_unlock(&root->fs_info->trans_mutex);
3143
3144 return 0;
3145 }
3146
3147 static struct extent_io_ops btree_extent_io_ops = {
3148 .write_cache_pages_lock_hook = btree_lock_page_hook,
3149 .readpage_end_io_hook = btree_readpage_end_io_hook,
3150 .submit_bio_hook = btree_submit_bio_hook,
3151 /* note we're sharing with inode.c for the merge bio hook */
3152 .merge_bio_hook = btrfs_merge_bio_hook,
3153 };
Ubuntu Eoan Linux kernel mirror