Btrfs: stop the readahead threads on failed mount
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / btrfs / disk-io.c
blobe532892431f46b1d6dbfbb8f0d846dcf6256f42b
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
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.
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.
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.
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 <linux/ratelimit.h>
33 #include <asm/unaligned.h>
34 #include "compat.h"
35 #include "ctree.h"
36 #include "disk-io.h"
37 #include "transaction.h"
38 #include "btrfs_inode.h"
39 #include "volumes.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
42 #include "locking.h"
43 #include "tree-log.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
47 static struct extent_io_ops btree_extent_io_ops;
48 static void end_workqueue_fn(struct btrfs_work *work);
49 static void free_fs_root(struct btrfs_root *root);
50 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
51 int read_only);
52 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
53 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
55 struct btrfs_root *root);
56 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
57 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
59 struct extent_io_tree *dirty_pages,
60 int mark);
61 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
62 struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_root *root);
66 * end_io_wq structs are used to do processing in task context when an IO is
67 * complete. This is used during reads to verify checksums, and it is used
68 * by writes to insert metadata for new file extents after IO is complete.
70 struct end_io_wq {
71 struct bio *bio;
72 bio_end_io_t *end_io;
73 void *private;
74 struct btrfs_fs_info *info;
75 int error;
76 int metadata;
77 struct list_head list;
78 struct btrfs_work work;
82 * async submit bios are used to offload expensive checksumming
83 * onto the worker threads. They checksum file and metadata bios
84 * just before they are sent down the IO stack.
86 struct async_submit_bio {
87 struct inode *inode;
88 struct bio *bio;
89 struct list_head list;
90 extent_submit_bio_hook_t *submit_bio_start;
91 extent_submit_bio_hook_t *submit_bio_done;
92 int rw;
93 int mirror_num;
94 unsigned long bio_flags;
96 * bio_offset is optional, can be used if the pages in the bio
97 * can't tell us where in the file the bio should go
99 u64 bio_offset;
100 struct btrfs_work work;
104 * Lockdep class keys for extent_buffer->lock's in this root. For a given
105 * eb, the lockdep key is determined by the btrfs_root it belongs to and
106 * the level the eb occupies in the tree.
108 * Different roots are used for different purposes and may nest inside each
109 * other and they require separate keysets. As lockdep keys should be
110 * static, assign keysets according to the purpose of the root as indicated
111 * by btrfs_root->objectid. This ensures that all special purpose roots
112 * have separate keysets.
114 * Lock-nesting across peer nodes is always done with the immediate parent
115 * node locked thus preventing deadlock. As lockdep doesn't know this, use
116 * subclass to avoid triggering lockdep warning in such cases.
118 * The key is set by the readpage_end_io_hook after the buffer has passed
119 * csum validation but before the pages are unlocked. It is also set by
120 * btrfs_init_new_buffer on freshly allocated blocks.
122 * We also add a check to make sure the highest level of the tree is the
123 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
124 * needs update as well.
126 #ifdef CONFIG_DEBUG_LOCK_ALLOC
127 # if BTRFS_MAX_LEVEL != 8
128 # error
129 # endif
131 static struct btrfs_lockdep_keyset {
132 u64 id; /* root objectid */
133 const char *name_stem; /* lock name stem */
134 char names[BTRFS_MAX_LEVEL + 1][20];
135 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
136 } btrfs_lockdep_keysets[] = {
137 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
138 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
139 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
140 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
141 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
142 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
143 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
144 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
145 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
146 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
147 { .id = 0, .name_stem = "tree" },
150 void __init btrfs_init_lockdep(void)
152 int i, j;
154 /* initialize lockdep class names */
155 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
156 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
158 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
159 snprintf(ks->names[j], sizeof(ks->names[j]),
160 "btrfs-%s-%02d", ks->name_stem, j);
164 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
165 int level)
167 struct btrfs_lockdep_keyset *ks;
169 BUG_ON(level >= ARRAY_SIZE(ks->keys));
171 /* find the matching keyset, id 0 is the default entry */
172 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
173 if (ks->id == objectid)
174 break;
176 lockdep_set_class_and_name(&eb->lock,
177 &ks->keys[level], ks->names[level]);
180 #endif
183 * extents on the btree inode are pretty simple, there's one extent
184 * that covers the entire device
186 static struct extent_map *btree_get_extent(struct inode *inode,
187 struct page *page, size_t pg_offset, u64 start, u64 len,
188 int create)
190 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
191 struct extent_map *em;
192 int ret;
194 read_lock(&em_tree->lock);
195 em = lookup_extent_mapping(em_tree, start, len);
196 if (em) {
197 em->bdev =
198 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
199 read_unlock(&em_tree->lock);
200 goto out;
202 read_unlock(&em_tree->lock);
204 em = alloc_extent_map();
205 if (!em) {
206 em = ERR_PTR(-ENOMEM);
207 goto out;
209 em->start = 0;
210 em->len = (u64)-1;
211 em->block_len = (u64)-1;
212 em->block_start = 0;
213 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
215 write_lock(&em_tree->lock);
216 ret = add_extent_mapping(em_tree, em);
217 if (ret == -EEXIST) {
218 u64 failed_start = em->start;
219 u64 failed_len = em->len;
221 free_extent_map(em);
222 em = lookup_extent_mapping(em_tree, start, len);
223 if (em) {
224 ret = 0;
225 } else {
226 em = lookup_extent_mapping(em_tree, failed_start,
227 failed_len);
228 ret = -EIO;
230 } else if (ret) {
231 free_extent_map(em);
232 em = NULL;
234 write_unlock(&em_tree->lock);
236 if (ret)
237 em = ERR_PTR(ret);
238 out:
239 return em;
242 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
244 return crc32c(seed, data, len);
247 void btrfs_csum_final(u32 crc, char *result)
249 put_unaligned_le32(~crc, result);
253 * compute the csum for a btree block, and either verify it or write it
254 * into the csum field of the block.
256 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
257 int verify)
259 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
260 char *result = NULL;
261 unsigned long len;
262 unsigned long cur_len;
263 unsigned long offset = BTRFS_CSUM_SIZE;
264 char *kaddr;
265 unsigned long map_start;
266 unsigned long map_len;
267 int err;
268 u32 crc = ~(u32)0;
269 unsigned long inline_result;
271 len = buf->len - offset;
272 while (len > 0) {
273 err = map_private_extent_buffer(buf, offset, 32,
274 &kaddr, &map_start, &map_len);
275 if (err)
276 return 1;
277 cur_len = min(len, map_len - (offset - map_start));
278 crc = btrfs_csum_data(root, kaddr + offset - map_start,
279 crc, cur_len);
280 len -= cur_len;
281 offset += cur_len;
283 if (csum_size > sizeof(inline_result)) {
284 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
285 if (!result)
286 return 1;
287 } else {
288 result = (char *)&inline_result;
291 btrfs_csum_final(crc, result);
293 if (verify) {
294 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
295 u32 val;
296 u32 found = 0;
297 memcpy(&found, result, csum_size);
299 read_extent_buffer(buf, &val, 0, csum_size);
300 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
301 "failed on %llu wanted %X found %X "
302 "level %d\n",
303 root->fs_info->sb->s_id,
304 (unsigned long long)buf->start, val, found,
305 btrfs_header_level(buf));
306 if (result != (char *)&inline_result)
307 kfree(result);
308 return 1;
310 } else {
311 write_extent_buffer(buf, result, 0, csum_size);
313 if (result != (char *)&inline_result)
314 kfree(result);
315 return 0;
319 * we can't consider a given block up to date unless the transid of the
320 * block matches the transid in the parent node's pointer. This is how we
321 * detect blocks that either didn't get written at all or got written
322 * in the wrong place.
324 static int verify_parent_transid(struct extent_io_tree *io_tree,
325 struct extent_buffer *eb, u64 parent_transid)
327 struct extent_state *cached_state = NULL;
328 int ret;
330 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
331 return 0;
333 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
334 0, &cached_state, GFP_NOFS);
335 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
336 btrfs_header_generation(eb) == parent_transid) {
337 ret = 0;
338 goto out;
340 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
341 "found %llu\n",
342 (unsigned long long)eb->start,
343 (unsigned long long)parent_transid,
344 (unsigned long long)btrfs_header_generation(eb));
345 ret = 1;
346 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
347 out:
348 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
349 &cached_state, GFP_NOFS);
350 return ret;
354 * helper to read a given tree block, doing retries as required when
355 * the checksums don't match and we have alternate mirrors to try.
357 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
358 struct extent_buffer *eb,
359 u64 start, u64 parent_transid)
361 struct extent_io_tree *io_tree;
362 int ret;
363 int num_copies = 0;
364 int mirror_num = 0;
366 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
367 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
368 while (1) {
369 ret = read_extent_buffer_pages(io_tree, eb, start,
370 WAIT_COMPLETE,
371 btree_get_extent, mirror_num);
372 if (!ret &&
373 !verify_parent_transid(io_tree, eb, parent_transid))
374 return ret;
377 * This buffer's crc is fine, but its contents are corrupted, so
378 * there is no reason to read the other copies, they won't be
379 * any less wrong.
381 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
382 return ret;
384 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
385 eb->start, eb->len);
386 if (num_copies == 1)
387 return ret;
389 mirror_num++;
390 if (mirror_num > num_copies)
391 return ret;
393 return -EIO;
397 * checksum a dirty tree block before IO. This has extra checks to make sure
398 * we only fill in the checksum field in the first page of a multi-page block
401 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
403 struct extent_io_tree *tree;
404 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
405 u64 found_start;
406 unsigned long len;
407 struct extent_buffer *eb;
408 int ret;
410 tree = &BTRFS_I(page->mapping->host)->io_tree;
412 if (page->private == EXTENT_PAGE_PRIVATE) {
413 WARN_ON(1);
414 goto out;
416 if (!page->private) {
417 WARN_ON(1);
418 goto out;
420 len = page->private >> 2;
421 WARN_ON(len == 0);
423 eb = alloc_extent_buffer(tree, start, len, page);
424 if (eb == NULL) {
425 WARN_ON(1);
426 goto out;
428 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
429 btrfs_header_generation(eb));
430 BUG_ON(ret);
431 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
433 found_start = btrfs_header_bytenr(eb);
434 if (found_start != start) {
435 WARN_ON(1);
436 goto err;
438 if (eb->first_page != page) {
439 WARN_ON(1);
440 goto err;
442 if (!PageUptodate(page)) {
443 WARN_ON(1);
444 goto err;
446 csum_tree_block(root, eb, 0);
447 err:
448 free_extent_buffer(eb);
449 out:
450 return 0;
453 static int check_tree_block_fsid(struct btrfs_root *root,
454 struct extent_buffer *eb)
456 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
457 u8 fsid[BTRFS_UUID_SIZE];
458 int ret = 1;
460 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
461 BTRFS_FSID_SIZE);
462 while (fs_devices) {
463 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
464 ret = 0;
465 break;
467 fs_devices = fs_devices->seed;
469 return ret;
472 #define CORRUPT(reason, eb, root, slot) \
473 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
474 "root=%llu, slot=%d\n", reason, \
475 (unsigned long long)btrfs_header_bytenr(eb), \
476 (unsigned long long)root->objectid, slot)
478 static noinline int check_leaf(struct btrfs_root *root,
479 struct extent_buffer *leaf)
481 struct btrfs_key key;
482 struct btrfs_key leaf_key;
483 u32 nritems = btrfs_header_nritems(leaf);
484 int slot;
486 if (nritems == 0)
487 return 0;
489 /* Check the 0 item */
490 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
491 BTRFS_LEAF_DATA_SIZE(root)) {
492 CORRUPT("invalid item offset size pair", leaf, root, 0);
493 return -EIO;
497 * Check to make sure each items keys are in the correct order and their
498 * offsets make sense. We only have to loop through nritems-1 because
499 * we check the current slot against the next slot, which verifies the
500 * next slot's offset+size makes sense and that the current's slot
501 * offset is correct.
503 for (slot = 0; slot < nritems - 1; slot++) {
504 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
505 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
507 /* Make sure the keys are in the right order */
508 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
509 CORRUPT("bad key order", leaf, root, slot);
510 return -EIO;
514 * Make sure the offset and ends are right, remember that the
515 * item data starts at the end of the leaf and grows towards the
516 * front.
518 if (btrfs_item_offset_nr(leaf, slot) !=
519 btrfs_item_end_nr(leaf, slot + 1)) {
520 CORRUPT("slot offset bad", leaf, root, slot);
521 return -EIO;
525 * Check to make sure that we don't point outside of the leaf,
526 * just incase all the items are consistent to eachother, but
527 * all point outside of the leaf.
529 if (btrfs_item_end_nr(leaf, slot) >
530 BTRFS_LEAF_DATA_SIZE(root)) {
531 CORRUPT("slot end outside of leaf", leaf, root, slot);
532 return -EIO;
536 return 0;
539 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
540 struct extent_state *state)
542 struct extent_io_tree *tree;
543 u64 found_start;
544 int found_level;
545 unsigned long len;
546 struct extent_buffer *eb;
547 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
548 int ret = 0;
550 tree = &BTRFS_I(page->mapping->host)->io_tree;
551 if (page->private == EXTENT_PAGE_PRIVATE)
552 goto out;
553 if (!page->private)
554 goto out;
556 len = page->private >> 2;
557 WARN_ON(len == 0);
559 eb = alloc_extent_buffer(tree, start, len, page);
560 if (eb == NULL) {
561 ret = -EIO;
562 goto out;
565 found_start = btrfs_header_bytenr(eb);
566 if (found_start != start) {
567 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
568 "%llu %llu\n",
569 (unsigned long long)found_start,
570 (unsigned long long)eb->start);
571 ret = -EIO;
572 goto err;
574 if (eb->first_page != page) {
575 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
576 eb->first_page->index, page->index);
577 WARN_ON(1);
578 ret = -EIO;
579 goto err;
581 if (check_tree_block_fsid(root, eb)) {
582 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
583 (unsigned long long)eb->start);
584 ret = -EIO;
585 goto err;
587 found_level = btrfs_header_level(eb);
589 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
590 eb, found_level);
592 ret = csum_tree_block(root, eb, 1);
593 if (ret) {
594 ret = -EIO;
595 goto err;
599 * If this is a leaf block and it is corrupt, set the corrupt bit so
600 * that we don't try and read the other copies of this block, just
601 * return -EIO.
603 if (found_level == 0 && check_leaf(root, eb)) {
604 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
605 ret = -EIO;
608 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
609 end = eb->start + end - 1;
610 err:
611 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
612 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
613 btree_readahead_hook(root, eb, eb->start, ret);
616 free_extent_buffer(eb);
617 out:
618 return ret;
621 static int btree_io_failed_hook(struct bio *failed_bio,
622 struct page *page, u64 start, u64 end,
623 u64 mirror_num, struct extent_state *state)
625 struct extent_io_tree *tree;
626 unsigned long len;
627 struct extent_buffer *eb;
628 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
630 tree = &BTRFS_I(page->mapping->host)->io_tree;
631 if (page->private == EXTENT_PAGE_PRIVATE)
632 goto out;
633 if (!page->private)
634 goto out;
636 len = page->private >> 2;
637 WARN_ON(len == 0);
639 eb = alloc_extent_buffer(tree, start, len, page);
640 if (eb == NULL)
641 goto out;
643 if (test_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags)) {
644 clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags);
645 btree_readahead_hook(root, eb, eb->start, -EIO);
647 free_extent_buffer(eb);
649 out:
650 return -EIO; /* we fixed nothing */
653 static void end_workqueue_bio(struct bio *bio, int err)
655 struct end_io_wq *end_io_wq = bio->bi_private;
656 struct btrfs_fs_info *fs_info;
658 fs_info = end_io_wq->info;
659 end_io_wq->error = err;
660 end_io_wq->work.func = end_workqueue_fn;
661 end_io_wq->work.flags = 0;
663 if (bio->bi_rw & REQ_WRITE) {
664 if (end_io_wq->metadata == 1)
665 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
666 &end_io_wq->work);
667 else if (end_io_wq->metadata == 2)
668 btrfs_queue_worker(&fs_info->endio_freespace_worker,
669 &end_io_wq->work);
670 else
671 btrfs_queue_worker(&fs_info->endio_write_workers,
672 &end_io_wq->work);
673 } else {
674 if (end_io_wq->metadata)
675 btrfs_queue_worker(&fs_info->endio_meta_workers,
676 &end_io_wq->work);
677 else
678 btrfs_queue_worker(&fs_info->endio_workers,
679 &end_io_wq->work);
684 * For the metadata arg you want
686 * 0 - if data
687 * 1 - if normal metadta
688 * 2 - if writing to the free space cache area
690 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
691 int metadata)
693 struct end_io_wq *end_io_wq;
694 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
695 if (!end_io_wq)
696 return -ENOMEM;
698 end_io_wq->private = bio->bi_private;
699 end_io_wq->end_io = bio->bi_end_io;
700 end_io_wq->info = info;
701 end_io_wq->error = 0;
702 end_io_wq->bio = bio;
703 end_io_wq->metadata = metadata;
705 bio->bi_private = end_io_wq;
706 bio->bi_end_io = end_workqueue_bio;
707 return 0;
710 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
712 unsigned long limit = min_t(unsigned long,
713 info->workers.max_workers,
714 info->fs_devices->open_devices);
715 return 256 * limit;
718 static void run_one_async_start(struct btrfs_work *work)
720 struct async_submit_bio *async;
722 async = container_of(work, struct async_submit_bio, work);
723 async->submit_bio_start(async->inode, async->rw, async->bio,
724 async->mirror_num, async->bio_flags,
725 async->bio_offset);
728 static void run_one_async_done(struct btrfs_work *work)
730 struct btrfs_fs_info *fs_info;
731 struct async_submit_bio *async;
732 int limit;
734 async = container_of(work, struct async_submit_bio, work);
735 fs_info = BTRFS_I(async->inode)->root->fs_info;
737 limit = btrfs_async_submit_limit(fs_info);
738 limit = limit * 2 / 3;
740 atomic_dec(&fs_info->nr_async_submits);
742 if (atomic_read(&fs_info->nr_async_submits) < limit &&
743 waitqueue_active(&fs_info->async_submit_wait))
744 wake_up(&fs_info->async_submit_wait);
746 async->submit_bio_done(async->inode, async->rw, async->bio,
747 async->mirror_num, async->bio_flags,
748 async->bio_offset);
751 static void run_one_async_free(struct btrfs_work *work)
753 struct async_submit_bio *async;
755 async = container_of(work, struct async_submit_bio, work);
756 kfree(async);
759 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
760 int rw, struct bio *bio, int mirror_num,
761 unsigned long bio_flags,
762 u64 bio_offset,
763 extent_submit_bio_hook_t *submit_bio_start,
764 extent_submit_bio_hook_t *submit_bio_done)
766 struct async_submit_bio *async;
768 async = kmalloc(sizeof(*async), GFP_NOFS);
769 if (!async)
770 return -ENOMEM;
772 async->inode = inode;
773 async->rw = rw;
774 async->bio = bio;
775 async->mirror_num = mirror_num;
776 async->submit_bio_start = submit_bio_start;
777 async->submit_bio_done = submit_bio_done;
779 async->work.func = run_one_async_start;
780 async->work.ordered_func = run_one_async_done;
781 async->work.ordered_free = run_one_async_free;
783 async->work.flags = 0;
784 async->bio_flags = bio_flags;
785 async->bio_offset = bio_offset;
787 atomic_inc(&fs_info->nr_async_submits);
789 if (rw & REQ_SYNC)
790 btrfs_set_work_high_prio(&async->work);
792 btrfs_queue_worker(&fs_info->workers, &async->work);
794 while (atomic_read(&fs_info->async_submit_draining) &&
795 atomic_read(&fs_info->nr_async_submits)) {
796 wait_event(fs_info->async_submit_wait,
797 (atomic_read(&fs_info->nr_async_submits) == 0));
800 return 0;
803 static int btree_csum_one_bio(struct bio *bio)
805 struct bio_vec *bvec = bio->bi_io_vec;
806 int bio_index = 0;
807 struct btrfs_root *root;
809 WARN_ON(bio->bi_vcnt <= 0);
810 while (bio_index < bio->bi_vcnt) {
811 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
812 csum_dirty_buffer(root, bvec->bv_page);
813 bio_index++;
814 bvec++;
816 return 0;
819 static int __btree_submit_bio_start(struct inode *inode, int rw,
820 struct bio *bio, int mirror_num,
821 unsigned long bio_flags,
822 u64 bio_offset)
825 * when we're called for a write, we're already in the async
826 * submission context. Just jump into btrfs_map_bio
828 btree_csum_one_bio(bio);
829 return 0;
832 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
833 int mirror_num, unsigned long bio_flags,
834 u64 bio_offset)
837 * when we're called for a write, we're already in the async
838 * submission context. Just jump into btrfs_map_bio
840 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
843 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
844 int mirror_num, unsigned long bio_flags,
845 u64 bio_offset)
847 int ret;
849 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
850 bio, 1);
851 BUG_ON(ret);
853 if (!(rw & REQ_WRITE)) {
855 * called for a read, do the setup so that checksum validation
856 * can happen in the async kernel threads
858 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
859 mirror_num, 0);
863 * kthread helpers are used to submit writes so that checksumming
864 * can happen in parallel across all CPUs
866 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
867 inode, rw, bio, mirror_num, 0,
868 bio_offset,
869 __btree_submit_bio_start,
870 __btree_submit_bio_done);
873 #ifdef CONFIG_MIGRATION
874 static int btree_migratepage(struct address_space *mapping,
875 struct page *newpage, struct page *page)
878 * we can't safely write a btree page from here,
879 * we haven't done the locking hook
881 if (PageDirty(page))
882 return -EAGAIN;
884 * Buffers may be managed in a filesystem specific way.
885 * We must have no buffers or drop them.
887 if (page_has_private(page) &&
888 !try_to_release_page(page, GFP_KERNEL))
889 return -EAGAIN;
890 return migrate_page(mapping, newpage, page);
892 #endif
894 static int btree_writepage(struct page *page, struct writeback_control *wbc)
896 struct extent_io_tree *tree;
897 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
898 struct extent_buffer *eb;
899 int was_dirty;
901 tree = &BTRFS_I(page->mapping->host)->io_tree;
902 if (!(current->flags & PF_MEMALLOC)) {
903 return extent_write_full_page(tree, page,
904 btree_get_extent, wbc);
907 redirty_page_for_writepage(wbc, page);
908 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
909 WARN_ON(!eb);
911 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
912 if (!was_dirty) {
913 spin_lock(&root->fs_info->delalloc_lock);
914 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
915 spin_unlock(&root->fs_info->delalloc_lock);
917 free_extent_buffer(eb);
919 unlock_page(page);
920 return 0;
923 static int btree_writepages(struct address_space *mapping,
924 struct writeback_control *wbc)
926 struct extent_io_tree *tree;
927 tree = &BTRFS_I(mapping->host)->io_tree;
928 if (wbc->sync_mode == WB_SYNC_NONE) {
929 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
930 u64 num_dirty;
931 unsigned long thresh = 32 * 1024 * 1024;
933 if (wbc->for_kupdate)
934 return 0;
936 /* this is a bit racy, but that's ok */
937 num_dirty = root->fs_info->dirty_metadata_bytes;
938 if (num_dirty < thresh)
939 return 0;
941 return extent_writepages(tree, mapping, btree_get_extent, wbc);
944 static int btree_readpage(struct file *file, struct page *page)
946 struct extent_io_tree *tree;
947 tree = &BTRFS_I(page->mapping->host)->io_tree;
948 return extent_read_full_page(tree, page, btree_get_extent, 0);
951 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
953 struct extent_io_tree *tree;
954 struct extent_map_tree *map;
955 int ret;
957 if (PageWriteback(page) || PageDirty(page))
958 return 0;
960 tree = &BTRFS_I(page->mapping->host)->io_tree;
961 map = &BTRFS_I(page->mapping->host)->extent_tree;
963 ret = try_release_extent_state(map, tree, page, gfp_flags);
964 if (!ret)
965 return 0;
967 ret = try_release_extent_buffer(tree, page);
968 if (ret == 1) {
969 ClearPagePrivate(page);
970 set_page_private(page, 0);
971 page_cache_release(page);
974 return ret;
977 static void btree_invalidatepage(struct page *page, unsigned long offset)
979 struct extent_io_tree *tree;
980 tree = &BTRFS_I(page->mapping->host)->io_tree;
981 extent_invalidatepage(tree, page, offset);
982 btree_releasepage(page, GFP_NOFS);
983 if (PagePrivate(page)) {
984 printk(KERN_WARNING "btrfs warning page private not zero "
985 "on page %llu\n", (unsigned long long)page_offset(page));
986 ClearPagePrivate(page);
987 set_page_private(page, 0);
988 page_cache_release(page);
992 static const struct address_space_operations btree_aops = {
993 .readpage = btree_readpage,
994 .writepage = btree_writepage,
995 .writepages = btree_writepages,
996 .releasepage = btree_releasepage,
997 .invalidatepage = btree_invalidatepage,
998 #ifdef CONFIG_MIGRATION
999 .migratepage = btree_migratepage,
1000 #endif
1003 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1004 u64 parent_transid)
1006 struct extent_buffer *buf = NULL;
1007 struct inode *btree_inode = root->fs_info->btree_inode;
1008 int ret = 0;
1010 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1011 if (!buf)
1012 return 0;
1013 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1014 buf, 0, WAIT_NONE, btree_get_extent, 0);
1015 free_extent_buffer(buf);
1016 return ret;
1019 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1020 int mirror_num, struct extent_buffer **eb)
1022 struct extent_buffer *buf = NULL;
1023 struct inode *btree_inode = root->fs_info->btree_inode;
1024 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1025 int ret;
1027 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1028 if (!buf)
1029 return 0;
1031 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1033 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1034 btree_get_extent, mirror_num);
1035 if (ret) {
1036 free_extent_buffer(buf);
1037 return ret;
1040 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1041 free_extent_buffer(buf);
1042 return -EIO;
1043 } else if (extent_buffer_uptodate(io_tree, buf, NULL)) {
1044 *eb = buf;
1045 } else {
1046 free_extent_buffer(buf);
1048 return 0;
1051 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1052 u64 bytenr, u32 blocksize)
1054 struct inode *btree_inode = root->fs_info->btree_inode;
1055 struct extent_buffer *eb;
1056 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1057 bytenr, blocksize);
1058 return eb;
1061 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1062 u64 bytenr, u32 blocksize)
1064 struct inode *btree_inode = root->fs_info->btree_inode;
1065 struct extent_buffer *eb;
1067 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1068 bytenr, blocksize, NULL);
1069 return eb;
1073 int btrfs_write_tree_block(struct extent_buffer *buf)
1075 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1076 buf->start + buf->len - 1);
1079 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1081 return filemap_fdatawait_range(buf->first_page->mapping,
1082 buf->start, buf->start + buf->len - 1);
1085 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1086 u32 blocksize, u64 parent_transid)
1088 struct extent_buffer *buf = NULL;
1089 int ret;
1091 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1092 if (!buf)
1093 return NULL;
1095 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1097 if (ret == 0)
1098 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1099 return buf;
1103 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1104 struct extent_buffer *buf)
1106 struct inode *btree_inode = root->fs_info->btree_inode;
1107 if (btrfs_header_generation(buf) ==
1108 root->fs_info->running_transaction->transid) {
1109 btrfs_assert_tree_locked(buf);
1111 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1112 spin_lock(&root->fs_info->delalloc_lock);
1113 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1114 root->fs_info->dirty_metadata_bytes -= buf->len;
1115 else
1116 WARN_ON(1);
1117 spin_unlock(&root->fs_info->delalloc_lock);
1120 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1121 btrfs_set_lock_blocking(buf);
1122 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1123 buf);
1125 return 0;
1128 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1129 u32 stripesize, struct btrfs_root *root,
1130 struct btrfs_fs_info *fs_info,
1131 u64 objectid)
1133 root->node = NULL;
1134 root->commit_root = NULL;
1135 root->sectorsize = sectorsize;
1136 root->nodesize = nodesize;
1137 root->leafsize = leafsize;
1138 root->stripesize = stripesize;
1139 root->ref_cows = 0;
1140 root->track_dirty = 0;
1141 root->in_radix = 0;
1142 root->orphan_item_inserted = 0;
1143 root->orphan_cleanup_state = 0;
1145 root->fs_info = fs_info;
1146 root->objectid = objectid;
1147 root->last_trans = 0;
1148 root->highest_objectid = 0;
1149 root->name = NULL;
1150 root->inode_tree = RB_ROOT;
1151 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1152 root->block_rsv = NULL;
1153 root->orphan_block_rsv = NULL;
1155 INIT_LIST_HEAD(&root->dirty_list);
1156 INIT_LIST_HEAD(&root->orphan_list);
1157 INIT_LIST_HEAD(&root->root_list);
1158 spin_lock_init(&root->orphan_lock);
1159 spin_lock_init(&root->inode_lock);
1160 spin_lock_init(&root->accounting_lock);
1161 mutex_init(&root->objectid_mutex);
1162 mutex_init(&root->log_mutex);
1163 init_waitqueue_head(&root->log_writer_wait);
1164 init_waitqueue_head(&root->log_commit_wait[0]);
1165 init_waitqueue_head(&root->log_commit_wait[1]);
1166 atomic_set(&root->log_commit[0], 0);
1167 atomic_set(&root->log_commit[1], 0);
1168 atomic_set(&root->log_writers, 0);
1169 root->log_batch = 0;
1170 root->log_transid = 0;
1171 root->last_log_commit = 0;
1172 extent_io_tree_init(&root->dirty_log_pages,
1173 fs_info->btree_inode->i_mapping);
1175 memset(&root->root_key, 0, sizeof(root->root_key));
1176 memset(&root->root_item, 0, sizeof(root->root_item));
1177 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1178 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1179 root->defrag_trans_start = fs_info->generation;
1180 init_completion(&root->kobj_unregister);
1181 root->defrag_running = 0;
1182 root->root_key.objectid = objectid;
1183 root->anon_dev = 0;
1184 return 0;
1187 static int find_and_setup_root(struct btrfs_root *tree_root,
1188 struct btrfs_fs_info *fs_info,
1189 u64 objectid,
1190 struct btrfs_root *root)
1192 int ret;
1193 u32 blocksize;
1194 u64 generation;
1196 __setup_root(tree_root->nodesize, tree_root->leafsize,
1197 tree_root->sectorsize, tree_root->stripesize,
1198 root, fs_info, objectid);
1199 ret = btrfs_find_last_root(tree_root, objectid,
1200 &root->root_item, &root->root_key);
1201 if (ret > 0)
1202 return -ENOENT;
1203 BUG_ON(ret);
1205 generation = btrfs_root_generation(&root->root_item);
1206 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1207 root->commit_root = NULL;
1208 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1209 blocksize, generation);
1210 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1211 free_extent_buffer(root->node);
1212 root->node = NULL;
1213 return -EIO;
1215 root->commit_root = btrfs_root_node(root);
1216 return 0;
1219 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1220 struct btrfs_fs_info *fs_info)
1222 struct btrfs_root *root;
1223 struct btrfs_root *tree_root = fs_info->tree_root;
1224 struct extent_buffer *leaf;
1226 root = kzalloc(sizeof(*root), GFP_NOFS);
1227 if (!root)
1228 return ERR_PTR(-ENOMEM);
1230 __setup_root(tree_root->nodesize, tree_root->leafsize,
1231 tree_root->sectorsize, tree_root->stripesize,
1232 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1234 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1235 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1236 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1238 * log trees do not get reference counted because they go away
1239 * before a real commit is actually done. They do store pointers
1240 * to file data extents, and those reference counts still get
1241 * updated (along with back refs to the log tree).
1243 root->ref_cows = 0;
1245 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1246 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1247 if (IS_ERR(leaf)) {
1248 kfree(root);
1249 return ERR_CAST(leaf);
1252 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1253 btrfs_set_header_bytenr(leaf, leaf->start);
1254 btrfs_set_header_generation(leaf, trans->transid);
1255 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1256 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1257 root->node = leaf;
1259 write_extent_buffer(root->node, root->fs_info->fsid,
1260 (unsigned long)btrfs_header_fsid(root->node),
1261 BTRFS_FSID_SIZE);
1262 btrfs_mark_buffer_dirty(root->node);
1263 btrfs_tree_unlock(root->node);
1264 return root;
1267 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1268 struct btrfs_fs_info *fs_info)
1270 struct btrfs_root *log_root;
1272 log_root = alloc_log_tree(trans, fs_info);
1273 if (IS_ERR(log_root))
1274 return PTR_ERR(log_root);
1275 WARN_ON(fs_info->log_root_tree);
1276 fs_info->log_root_tree = log_root;
1277 return 0;
1280 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1281 struct btrfs_root *root)
1283 struct btrfs_root *log_root;
1284 struct btrfs_inode_item *inode_item;
1286 log_root = alloc_log_tree(trans, root->fs_info);
1287 if (IS_ERR(log_root))
1288 return PTR_ERR(log_root);
1290 log_root->last_trans = trans->transid;
1291 log_root->root_key.offset = root->root_key.objectid;
1293 inode_item = &log_root->root_item.inode;
1294 inode_item->generation = cpu_to_le64(1);
1295 inode_item->size = cpu_to_le64(3);
1296 inode_item->nlink = cpu_to_le32(1);
1297 inode_item->nbytes = cpu_to_le64(root->leafsize);
1298 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1300 btrfs_set_root_node(&log_root->root_item, log_root->node);
1302 WARN_ON(root->log_root);
1303 root->log_root = log_root;
1304 root->log_transid = 0;
1305 root->last_log_commit = 0;
1306 return 0;
1309 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1310 struct btrfs_key *location)
1312 struct btrfs_root *root;
1313 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1314 struct btrfs_path *path;
1315 struct extent_buffer *l;
1316 u64 generation;
1317 u32 blocksize;
1318 int ret = 0;
1320 root = kzalloc(sizeof(*root), GFP_NOFS);
1321 if (!root)
1322 return ERR_PTR(-ENOMEM);
1323 if (location->offset == (u64)-1) {
1324 ret = find_and_setup_root(tree_root, fs_info,
1325 location->objectid, root);
1326 if (ret) {
1327 kfree(root);
1328 return ERR_PTR(ret);
1330 goto out;
1333 __setup_root(tree_root->nodesize, tree_root->leafsize,
1334 tree_root->sectorsize, tree_root->stripesize,
1335 root, fs_info, location->objectid);
1337 path = btrfs_alloc_path();
1338 if (!path) {
1339 kfree(root);
1340 return ERR_PTR(-ENOMEM);
1342 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1343 if (ret == 0) {
1344 l = path->nodes[0];
1345 read_extent_buffer(l, &root->root_item,
1346 btrfs_item_ptr_offset(l, path->slots[0]),
1347 sizeof(root->root_item));
1348 memcpy(&root->root_key, location, sizeof(*location));
1350 btrfs_free_path(path);
1351 if (ret) {
1352 kfree(root);
1353 if (ret > 0)
1354 ret = -ENOENT;
1355 return ERR_PTR(ret);
1358 generation = btrfs_root_generation(&root->root_item);
1359 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1360 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1361 blocksize, generation);
1362 root->commit_root = btrfs_root_node(root);
1363 BUG_ON(!root->node);
1364 out:
1365 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1366 root->ref_cows = 1;
1367 btrfs_check_and_init_root_item(&root->root_item);
1370 return root;
1373 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1374 struct btrfs_key *location)
1376 struct btrfs_root *root;
1377 int ret;
1379 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1380 return fs_info->tree_root;
1381 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1382 return fs_info->extent_root;
1383 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1384 return fs_info->chunk_root;
1385 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1386 return fs_info->dev_root;
1387 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1388 return fs_info->csum_root;
1389 again:
1390 spin_lock(&fs_info->fs_roots_radix_lock);
1391 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1392 (unsigned long)location->objectid);
1393 spin_unlock(&fs_info->fs_roots_radix_lock);
1394 if (root)
1395 return root;
1397 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1398 if (IS_ERR(root))
1399 return root;
1401 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1402 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1403 GFP_NOFS);
1404 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1405 ret = -ENOMEM;
1406 goto fail;
1409 btrfs_init_free_ino_ctl(root);
1410 mutex_init(&root->fs_commit_mutex);
1411 spin_lock_init(&root->cache_lock);
1412 init_waitqueue_head(&root->cache_wait);
1414 ret = get_anon_bdev(&root->anon_dev);
1415 if (ret)
1416 goto fail;
1418 if (btrfs_root_refs(&root->root_item) == 0) {
1419 ret = -ENOENT;
1420 goto fail;
1423 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1424 if (ret < 0)
1425 goto fail;
1426 if (ret == 0)
1427 root->orphan_item_inserted = 1;
1429 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1430 if (ret)
1431 goto fail;
1433 spin_lock(&fs_info->fs_roots_radix_lock);
1434 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1435 (unsigned long)root->root_key.objectid,
1436 root);
1437 if (ret == 0)
1438 root->in_radix = 1;
1440 spin_unlock(&fs_info->fs_roots_radix_lock);
1441 radix_tree_preload_end();
1442 if (ret) {
1443 if (ret == -EEXIST) {
1444 free_fs_root(root);
1445 goto again;
1447 goto fail;
1450 ret = btrfs_find_dead_roots(fs_info->tree_root,
1451 root->root_key.objectid);
1452 WARN_ON(ret);
1453 return root;
1454 fail:
1455 free_fs_root(root);
1456 return ERR_PTR(ret);
1459 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1461 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1462 int ret = 0;
1463 struct btrfs_device *device;
1464 struct backing_dev_info *bdi;
1466 rcu_read_lock();
1467 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1468 if (!device->bdev)
1469 continue;
1470 bdi = blk_get_backing_dev_info(device->bdev);
1471 if (bdi && bdi_congested(bdi, bdi_bits)) {
1472 ret = 1;
1473 break;
1476 rcu_read_unlock();
1477 return ret;
1481 * If this fails, caller must call bdi_destroy() to get rid of the
1482 * bdi again.
1484 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1486 int err;
1488 bdi->capabilities = BDI_CAP_MAP_COPY;
1489 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1490 if (err)
1491 return err;
1493 bdi->ra_pages = default_backing_dev_info.ra_pages;
1494 bdi->congested_fn = btrfs_congested_fn;
1495 bdi->congested_data = info;
1496 return 0;
1499 static int bio_ready_for_csum(struct bio *bio)
1501 u64 length = 0;
1502 u64 buf_len = 0;
1503 u64 start = 0;
1504 struct page *page;
1505 struct extent_io_tree *io_tree = NULL;
1506 struct bio_vec *bvec;
1507 int i;
1508 int ret;
1510 bio_for_each_segment(bvec, bio, i) {
1511 page = bvec->bv_page;
1512 if (page->private == EXTENT_PAGE_PRIVATE) {
1513 length += bvec->bv_len;
1514 continue;
1516 if (!page->private) {
1517 length += bvec->bv_len;
1518 continue;
1520 length = bvec->bv_len;
1521 buf_len = page->private >> 2;
1522 start = page_offset(page) + bvec->bv_offset;
1523 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1525 /* are we fully contained in this bio? */
1526 if (buf_len <= length)
1527 return 1;
1529 ret = extent_range_uptodate(io_tree, start + length,
1530 start + buf_len - 1);
1531 return ret;
1535 * called by the kthread helper functions to finally call the bio end_io
1536 * functions. This is where read checksum verification actually happens
1538 static void end_workqueue_fn(struct btrfs_work *work)
1540 struct bio *bio;
1541 struct end_io_wq *end_io_wq;
1542 struct btrfs_fs_info *fs_info;
1543 int error;
1545 end_io_wq = container_of(work, struct end_io_wq, work);
1546 bio = end_io_wq->bio;
1547 fs_info = end_io_wq->info;
1549 /* metadata bio reads are special because the whole tree block must
1550 * be checksummed at once. This makes sure the entire block is in
1551 * ram and up to date before trying to verify things. For
1552 * blocksize <= pagesize, it is basically a noop
1554 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1555 !bio_ready_for_csum(bio)) {
1556 btrfs_queue_worker(&fs_info->endio_meta_workers,
1557 &end_io_wq->work);
1558 return;
1560 error = end_io_wq->error;
1561 bio->bi_private = end_io_wq->private;
1562 bio->bi_end_io = end_io_wq->end_io;
1563 kfree(end_io_wq);
1564 bio_endio(bio, error);
1567 static int cleaner_kthread(void *arg)
1569 struct btrfs_root *root = arg;
1571 do {
1572 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1574 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1575 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1576 btrfs_run_delayed_iputs(root);
1577 btrfs_clean_old_snapshots(root);
1578 mutex_unlock(&root->fs_info->cleaner_mutex);
1579 btrfs_run_defrag_inodes(root->fs_info);
1582 if (freezing(current)) {
1583 refrigerator();
1584 } else {
1585 set_current_state(TASK_INTERRUPTIBLE);
1586 if (!kthread_should_stop())
1587 schedule();
1588 __set_current_state(TASK_RUNNING);
1590 } while (!kthread_should_stop());
1591 return 0;
1594 static int transaction_kthread(void *arg)
1596 struct btrfs_root *root = arg;
1597 struct btrfs_trans_handle *trans;
1598 struct btrfs_transaction *cur;
1599 u64 transid;
1600 unsigned long now;
1601 unsigned long delay;
1602 int ret;
1604 do {
1605 delay = HZ * 30;
1606 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1607 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1609 spin_lock(&root->fs_info->trans_lock);
1610 cur = root->fs_info->running_transaction;
1611 if (!cur) {
1612 spin_unlock(&root->fs_info->trans_lock);
1613 goto sleep;
1616 now = get_seconds();
1617 if (!cur->blocked &&
1618 (now < cur->start_time || now - cur->start_time < 30)) {
1619 spin_unlock(&root->fs_info->trans_lock);
1620 delay = HZ * 5;
1621 goto sleep;
1623 transid = cur->transid;
1624 spin_unlock(&root->fs_info->trans_lock);
1626 trans = btrfs_join_transaction(root);
1627 BUG_ON(IS_ERR(trans));
1628 if (transid == trans->transid) {
1629 ret = btrfs_commit_transaction(trans, root);
1630 BUG_ON(ret);
1631 } else {
1632 btrfs_end_transaction(trans, root);
1634 sleep:
1635 wake_up_process(root->fs_info->cleaner_kthread);
1636 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1638 if (freezing(current)) {
1639 refrigerator();
1640 } else {
1641 set_current_state(TASK_INTERRUPTIBLE);
1642 if (!kthread_should_stop() &&
1643 !btrfs_transaction_blocked(root->fs_info))
1644 schedule_timeout(delay);
1645 __set_current_state(TASK_RUNNING);
1647 } while (!kthread_should_stop());
1648 return 0;
1652 * this will find the highest generation in the array of
1653 * root backups. The index of the highest array is returned,
1654 * or -1 if we can't find anything.
1656 * We check to make sure the array is valid by comparing the
1657 * generation of the latest root in the array with the generation
1658 * in the super block. If they don't match we pitch it.
1660 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1662 u64 cur;
1663 int newest_index = -1;
1664 struct btrfs_root_backup *root_backup;
1665 int i;
1667 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1668 root_backup = info->super_copy->super_roots + i;
1669 cur = btrfs_backup_tree_root_gen(root_backup);
1670 if (cur == newest_gen)
1671 newest_index = i;
1674 /* check to see if we actually wrapped around */
1675 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1676 root_backup = info->super_copy->super_roots;
1677 cur = btrfs_backup_tree_root_gen(root_backup);
1678 if (cur == newest_gen)
1679 newest_index = 0;
1681 return newest_index;
1686 * find the oldest backup so we know where to store new entries
1687 * in the backup array. This will set the backup_root_index
1688 * field in the fs_info struct
1690 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1691 u64 newest_gen)
1693 int newest_index = -1;
1695 newest_index = find_newest_super_backup(info, newest_gen);
1696 /* if there was garbage in there, just move along */
1697 if (newest_index == -1) {
1698 info->backup_root_index = 0;
1699 } else {
1700 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1705 * copy all the root pointers into the super backup array.
1706 * this will bump the backup pointer by one when it is
1707 * done
1709 static void backup_super_roots(struct btrfs_fs_info *info)
1711 int next_backup;
1712 struct btrfs_root_backup *root_backup;
1713 int last_backup;
1715 next_backup = info->backup_root_index;
1716 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1717 BTRFS_NUM_BACKUP_ROOTS;
1720 * just overwrite the last backup if we're at the same generation
1721 * this happens only at umount
1723 root_backup = info->super_for_commit->super_roots + last_backup;
1724 if (btrfs_backup_tree_root_gen(root_backup) ==
1725 btrfs_header_generation(info->tree_root->node))
1726 next_backup = last_backup;
1728 root_backup = info->super_for_commit->super_roots + next_backup;
1731 * make sure all of our padding and empty slots get zero filled
1732 * regardless of which ones we use today
1734 memset(root_backup, 0, sizeof(*root_backup));
1736 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1738 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1739 btrfs_set_backup_tree_root_gen(root_backup,
1740 btrfs_header_generation(info->tree_root->node));
1742 btrfs_set_backup_tree_root_level(root_backup,
1743 btrfs_header_level(info->tree_root->node));
1745 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1746 btrfs_set_backup_chunk_root_gen(root_backup,
1747 btrfs_header_generation(info->chunk_root->node));
1748 btrfs_set_backup_chunk_root_level(root_backup,
1749 btrfs_header_level(info->chunk_root->node));
1751 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1752 btrfs_set_backup_extent_root_gen(root_backup,
1753 btrfs_header_generation(info->extent_root->node));
1754 btrfs_set_backup_extent_root_level(root_backup,
1755 btrfs_header_level(info->extent_root->node));
1757 btrfs_set_backup_fs_root(root_backup, info->fs_root->node->start);
1758 btrfs_set_backup_fs_root_gen(root_backup,
1759 btrfs_header_generation(info->fs_root->node));
1760 btrfs_set_backup_fs_root_level(root_backup,
1761 btrfs_header_level(info->fs_root->node));
1763 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1764 btrfs_set_backup_dev_root_gen(root_backup,
1765 btrfs_header_generation(info->dev_root->node));
1766 btrfs_set_backup_dev_root_level(root_backup,
1767 btrfs_header_level(info->dev_root->node));
1769 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1770 btrfs_set_backup_csum_root_gen(root_backup,
1771 btrfs_header_generation(info->csum_root->node));
1772 btrfs_set_backup_csum_root_level(root_backup,
1773 btrfs_header_level(info->csum_root->node));
1775 btrfs_set_backup_total_bytes(root_backup,
1776 btrfs_super_total_bytes(info->super_copy));
1777 btrfs_set_backup_bytes_used(root_backup,
1778 btrfs_super_bytes_used(info->super_copy));
1779 btrfs_set_backup_num_devices(root_backup,
1780 btrfs_super_num_devices(info->super_copy));
1783 * if we don't copy this out to the super_copy, it won't get remembered
1784 * for the next commit
1786 memcpy(&info->super_copy->super_roots,
1787 &info->super_for_commit->super_roots,
1788 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1792 * this copies info out of the root backup array and back into
1793 * the in-memory super block. It is meant to help iterate through
1794 * the array, so you send it the number of backups you've already
1795 * tried and the last backup index you used.
1797 * this returns -1 when it has tried all the backups
1799 static noinline int next_root_backup(struct btrfs_fs_info *info,
1800 struct btrfs_super_block *super,
1801 int *num_backups_tried, int *backup_index)
1803 struct btrfs_root_backup *root_backup;
1804 int newest = *backup_index;
1806 if (*num_backups_tried == 0) {
1807 u64 gen = btrfs_super_generation(super);
1809 newest = find_newest_super_backup(info, gen);
1810 if (newest == -1)
1811 return -1;
1813 *backup_index = newest;
1814 *num_backups_tried = 1;
1815 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1816 /* we've tried all the backups, all done */
1817 return -1;
1818 } else {
1819 /* jump to the next oldest backup */
1820 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1821 BTRFS_NUM_BACKUP_ROOTS;
1822 *backup_index = newest;
1823 *num_backups_tried += 1;
1825 root_backup = super->super_roots + newest;
1827 btrfs_set_super_generation(super,
1828 btrfs_backup_tree_root_gen(root_backup));
1829 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1830 btrfs_set_super_root_level(super,
1831 btrfs_backup_tree_root_level(root_backup));
1832 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1835 * fixme: the total bytes and num_devices need to match or we should
1836 * need a fsck
1838 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1839 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1840 return 0;
1843 /* helper to cleanup tree roots */
1844 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1846 free_extent_buffer(info->tree_root->node);
1847 free_extent_buffer(info->tree_root->commit_root);
1848 free_extent_buffer(info->dev_root->node);
1849 free_extent_buffer(info->dev_root->commit_root);
1850 free_extent_buffer(info->extent_root->node);
1851 free_extent_buffer(info->extent_root->commit_root);
1852 free_extent_buffer(info->csum_root->node);
1853 free_extent_buffer(info->csum_root->commit_root);
1855 info->tree_root->node = NULL;
1856 info->tree_root->commit_root = NULL;
1857 info->dev_root->node = NULL;
1858 info->dev_root->commit_root = NULL;
1859 info->extent_root->node = NULL;
1860 info->extent_root->commit_root = NULL;
1861 info->csum_root->node = NULL;
1862 info->csum_root->commit_root = NULL;
1864 if (chunk_root) {
1865 free_extent_buffer(info->chunk_root->node);
1866 free_extent_buffer(info->chunk_root->commit_root);
1867 info->chunk_root->node = NULL;
1868 info->chunk_root->commit_root = NULL;
1873 struct btrfs_root *open_ctree(struct super_block *sb,
1874 struct btrfs_fs_devices *fs_devices,
1875 char *options)
1877 u32 sectorsize;
1878 u32 nodesize;
1879 u32 leafsize;
1880 u32 blocksize;
1881 u32 stripesize;
1882 u64 generation;
1883 u64 features;
1884 struct btrfs_key location;
1885 struct buffer_head *bh;
1886 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1887 GFP_NOFS);
1888 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1889 GFP_NOFS);
1890 struct btrfs_root *tree_root = btrfs_sb(sb);
1891 struct btrfs_fs_info *fs_info = NULL;
1892 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1893 GFP_NOFS);
1894 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1895 GFP_NOFS);
1896 struct btrfs_root *log_tree_root;
1898 int ret;
1899 int err = -EINVAL;
1900 int num_backups_tried = 0;
1901 int backup_index = 0;
1903 struct btrfs_super_block *disk_super;
1905 if (!extent_root || !tree_root || !tree_root->fs_info ||
1906 !chunk_root || !dev_root || !csum_root) {
1907 err = -ENOMEM;
1908 goto fail;
1910 fs_info = tree_root->fs_info;
1912 ret = init_srcu_struct(&fs_info->subvol_srcu);
1913 if (ret) {
1914 err = ret;
1915 goto fail;
1918 ret = setup_bdi(fs_info, &fs_info->bdi);
1919 if (ret) {
1920 err = ret;
1921 goto fail_srcu;
1924 fs_info->btree_inode = new_inode(sb);
1925 if (!fs_info->btree_inode) {
1926 err = -ENOMEM;
1927 goto fail_bdi;
1930 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1932 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1933 INIT_LIST_HEAD(&fs_info->trans_list);
1934 INIT_LIST_HEAD(&fs_info->dead_roots);
1935 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1936 INIT_LIST_HEAD(&fs_info->hashers);
1937 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1938 INIT_LIST_HEAD(&fs_info->ordered_operations);
1939 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1940 spin_lock_init(&fs_info->delalloc_lock);
1941 spin_lock_init(&fs_info->trans_lock);
1942 spin_lock_init(&fs_info->ref_cache_lock);
1943 spin_lock_init(&fs_info->fs_roots_radix_lock);
1944 spin_lock_init(&fs_info->delayed_iput_lock);
1945 spin_lock_init(&fs_info->defrag_inodes_lock);
1946 spin_lock_init(&fs_info->free_chunk_lock);
1947 mutex_init(&fs_info->reloc_mutex);
1949 init_completion(&fs_info->kobj_unregister);
1950 fs_info->tree_root = tree_root;
1951 fs_info->extent_root = extent_root;
1952 fs_info->csum_root = csum_root;
1953 fs_info->chunk_root = chunk_root;
1954 fs_info->dev_root = dev_root;
1955 fs_info->fs_devices = fs_devices;
1956 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1957 INIT_LIST_HEAD(&fs_info->space_info);
1958 btrfs_mapping_init(&fs_info->mapping_tree);
1959 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1960 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1961 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1962 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1963 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1964 btrfs_init_block_rsv(&fs_info->delayed_block_rsv);
1965 atomic_set(&fs_info->nr_async_submits, 0);
1966 atomic_set(&fs_info->async_delalloc_pages, 0);
1967 atomic_set(&fs_info->async_submit_draining, 0);
1968 atomic_set(&fs_info->nr_async_bios, 0);
1969 atomic_set(&fs_info->defrag_running, 0);
1970 fs_info->sb = sb;
1971 fs_info->max_inline = 8192 * 1024;
1972 fs_info->metadata_ratio = 0;
1973 fs_info->defrag_inodes = RB_ROOT;
1974 fs_info->trans_no_join = 0;
1975 fs_info->free_chunk_space = 0;
1977 /* readahead state */
1978 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
1979 spin_lock_init(&fs_info->reada_lock);
1981 fs_info->thread_pool_size = min_t(unsigned long,
1982 num_online_cpus() + 2, 8);
1984 INIT_LIST_HEAD(&fs_info->ordered_extents);
1985 spin_lock_init(&fs_info->ordered_extent_lock);
1986 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1987 GFP_NOFS);
1988 if (!fs_info->delayed_root) {
1989 err = -ENOMEM;
1990 goto fail_iput;
1992 btrfs_init_delayed_root(fs_info->delayed_root);
1994 mutex_init(&fs_info->scrub_lock);
1995 atomic_set(&fs_info->scrubs_running, 0);
1996 atomic_set(&fs_info->scrub_pause_req, 0);
1997 atomic_set(&fs_info->scrubs_paused, 0);
1998 atomic_set(&fs_info->scrub_cancel_req, 0);
1999 init_waitqueue_head(&fs_info->scrub_pause_wait);
2000 init_rwsem(&fs_info->scrub_super_lock);
2001 fs_info->scrub_workers_refcnt = 0;
2003 sb->s_blocksize = 4096;
2004 sb->s_blocksize_bits = blksize_bits(4096);
2005 sb->s_bdi = &fs_info->bdi;
2007 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2008 fs_info->btree_inode->i_nlink = 1;
2010 * we set the i_size on the btree inode to the max possible int.
2011 * the real end of the address space is determined by all of
2012 * the devices in the system
2014 fs_info->btree_inode->i_size = OFFSET_MAX;
2015 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2016 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2018 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2019 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2020 fs_info->btree_inode->i_mapping);
2021 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2023 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2025 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2026 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2027 sizeof(struct btrfs_key));
2028 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
2029 insert_inode_hash(fs_info->btree_inode);
2031 spin_lock_init(&fs_info->block_group_cache_lock);
2032 fs_info->block_group_cache_tree = RB_ROOT;
2034 extent_io_tree_init(&fs_info->freed_extents[0],
2035 fs_info->btree_inode->i_mapping);
2036 extent_io_tree_init(&fs_info->freed_extents[1],
2037 fs_info->btree_inode->i_mapping);
2038 fs_info->pinned_extents = &fs_info->freed_extents[0];
2039 fs_info->do_barriers = 1;
2042 mutex_init(&fs_info->ordered_operations_mutex);
2043 mutex_init(&fs_info->tree_log_mutex);
2044 mutex_init(&fs_info->chunk_mutex);
2045 mutex_init(&fs_info->transaction_kthread_mutex);
2046 mutex_init(&fs_info->cleaner_mutex);
2047 mutex_init(&fs_info->volume_mutex);
2048 init_rwsem(&fs_info->extent_commit_sem);
2049 init_rwsem(&fs_info->cleanup_work_sem);
2050 init_rwsem(&fs_info->subvol_sem);
2052 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2053 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2055 init_waitqueue_head(&fs_info->transaction_throttle);
2056 init_waitqueue_head(&fs_info->transaction_wait);
2057 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2058 init_waitqueue_head(&fs_info->async_submit_wait);
2060 __setup_root(4096, 4096, 4096, 4096, tree_root,
2061 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2063 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2064 if (!bh) {
2065 err = -EINVAL;
2066 goto fail_alloc;
2069 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2070 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2071 sizeof(*fs_info->super_for_commit));
2072 brelse(bh);
2074 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2076 disk_super = fs_info->super_copy;
2077 if (!btrfs_super_root(disk_super))
2078 goto fail_alloc;
2080 /* check FS state, whether FS is broken. */
2081 fs_info->fs_state |= btrfs_super_flags(disk_super);
2083 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2086 * run through our array of backup supers and setup
2087 * our ring pointer to the oldest one
2089 generation = btrfs_super_generation(disk_super);
2090 find_oldest_super_backup(fs_info, generation);
2093 * In the long term, we'll store the compression type in the super
2094 * block, and it'll be used for per file compression control.
2096 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2098 ret = btrfs_parse_options(tree_root, options);
2099 if (ret) {
2100 err = ret;
2101 goto fail_alloc;
2104 features = btrfs_super_incompat_flags(disk_super) &
2105 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2106 if (features) {
2107 printk(KERN_ERR "BTRFS: couldn't mount because of "
2108 "unsupported optional features (%Lx).\n",
2109 (unsigned long long)features);
2110 err = -EINVAL;
2111 goto fail_alloc;
2114 features = btrfs_super_incompat_flags(disk_super);
2115 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2116 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2117 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2118 btrfs_set_super_incompat_flags(disk_super, features);
2120 features = btrfs_super_compat_ro_flags(disk_super) &
2121 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2122 if (!(sb->s_flags & MS_RDONLY) && features) {
2123 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2124 "unsupported option features (%Lx).\n",
2125 (unsigned long long)features);
2126 err = -EINVAL;
2127 goto fail_alloc;
2130 btrfs_init_workers(&fs_info->generic_worker,
2131 "genwork", 1, NULL);
2133 btrfs_init_workers(&fs_info->workers, "worker",
2134 fs_info->thread_pool_size,
2135 &fs_info->generic_worker);
2137 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2138 fs_info->thread_pool_size,
2139 &fs_info->generic_worker);
2141 btrfs_init_workers(&fs_info->submit_workers, "submit",
2142 min_t(u64, fs_devices->num_devices,
2143 fs_info->thread_pool_size),
2144 &fs_info->generic_worker);
2146 btrfs_init_workers(&fs_info->caching_workers, "cache",
2147 2, &fs_info->generic_worker);
2149 /* a higher idle thresh on the submit workers makes it much more
2150 * likely that bios will be send down in a sane order to the
2151 * devices
2153 fs_info->submit_workers.idle_thresh = 64;
2155 fs_info->workers.idle_thresh = 16;
2156 fs_info->workers.ordered = 1;
2158 fs_info->delalloc_workers.idle_thresh = 2;
2159 fs_info->delalloc_workers.ordered = 1;
2161 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2162 &fs_info->generic_worker);
2163 btrfs_init_workers(&fs_info->endio_workers, "endio",
2164 fs_info->thread_pool_size,
2165 &fs_info->generic_worker);
2166 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2167 fs_info->thread_pool_size,
2168 &fs_info->generic_worker);
2169 btrfs_init_workers(&fs_info->endio_meta_write_workers,
2170 "endio-meta-write", fs_info->thread_pool_size,
2171 &fs_info->generic_worker);
2172 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2173 fs_info->thread_pool_size,
2174 &fs_info->generic_worker);
2175 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2176 1, &fs_info->generic_worker);
2177 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2178 fs_info->thread_pool_size,
2179 &fs_info->generic_worker);
2180 btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2181 fs_info->thread_pool_size,
2182 &fs_info->generic_worker);
2185 * endios are largely parallel and should have a very
2186 * low idle thresh
2188 fs_info->endio_workers.idle_thresh = 4;
2189 fs_info->endio_meta_workers.idle_thresh = 4;
2191 fs_info->endio_write_workers.idle_thresh = 2;
2192 fs_info->endio_meta_write_workers.idle_thresh = 2;
2193 fs_info->readahead_workers.idle_thresh = 2;
2195 btrfs_start_workers(&fs_info->workers, 1);
2196 btrfs_start_workers(&fs_info->generic_worker, 1);
2197 btrfs_start_workers(&fs_info->submit_workers, 1);
2198 btrfs_start_workers(&fs_info->delalloc_workers, 1);
2199 btrfs_start_workers(&fs_info->fixup_workers, 1);
2200 btrfs_start_workers(&fs_info->endio_workers, 1);
2201 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
2202 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
2203 btrfs_start_workers(&fs_info->endio_write_workers, 1);
2204 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
2205 btrfs_start_workers(&fs_info->delayed_workers, 1);
2206 btrfs_start_workers(&fs_info->caching_workers, 1);
2207 btrfs_start_workers(&fs_info->readahead_workers, 1);
2209 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2210 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2211 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2213 nodesize = btrfs_super_nodesize(disk_super);
2214 leafsize = btrfs_super_leafsize(disk_super);
2215 sectorsize = btrfs_super_sectorsize(disk_super);
2216 stripesize = btrfs_super_stripesize(disk_super);
2217 tree_root->nodesize = nodesize;
2218 tree_root->leafsize = leafsize;
2219 tree_root->sectorsize = sectorsize;
2220 tree_root->stripesize = stripesize;
2222 sb->s_blocksize = sectorsize;
2223 sb->s_blocksize_bits = blksize_bits(sectorsize);
2225 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2226 sizeof(disk_super->magic))) {
2227 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2228 goto fail_sb_buffer;
2231 mutex_lock(&fs_info->chunk_mutex);
2232 ret = btrfs_read_sys_array(tree_root);
2233 mutex_unlock(&fs_info->chunk_mutex);
2234 if (ret) {
2235 printk(KERN_WARNING "btrfs: failed to read the system "
2236 "array on %s\n", sb->s_id);
2237 goto fail_sb_buffer;
2240 blocksize = btrfs_level_size(tree_root,
2241 btrfs_super_chunk_root_level(disk_super));
2242 generation = btrfs_super_chunk_root_generation(disk_super);
2244 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2245 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2247 chunk_root->node = read_tree_block(chunk_root,
2248 btrfs_super_chunk_root(disk_super),
2249 blocksize, generation);
2250 BUG_ON(!chunk_root->node);
2251 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2252 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2253 sb->s_id);
2254 goto fail_tree_roots;
2256 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2257 chunk_root->commit_root = btrfs_root_node(chunk_root);
2259 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2260 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2261 BTRFS_UUID_SIZE);
2263 mutex_lock(&fs_info->chunk_mutex);
2264 ret = btrfs_read_chunk_tree(chunk_root);
2265 mutex_unlock(&fs_info->chunk_mutex);
2266 if (ret) {
2267 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2268 sb->s_id);
2269 goto fail_tree_roots;
2272 btrfs_close_extra_devices(fs_devices);
2274 retry_root_backup:
2275 blocksize = btrfs_level_size(tree_root,
2276 btrfs_super_root_level(disk_super));
2277 generation = btrfs_super_generation(disk_super);
2279 tree_root->node = read_tree_block(tree_root,
2280 btrfs_super_root(disk_super),
2281 blocksize, generation);
2282 if (!tree_root->node ||
2283 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2284 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2285 sb->s_id);
2287 goto recovery_tree_root;
2290 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2291 tree_root->commit_root = btrfs_root_node(tree_root);
2293 ret = find_and_setup_root(tree_root, fs_info,
2294 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2295 if (ret)
2296 goto recovery_tree_root;
2297 extent_root->track_dirty = 1;
2299 ret = find_and_setup_root(tree_root, fs_info,
2300 BTRFS_DEV_TREE_OBJECTID, dev_root);
2301 if (ret)
2302 goto recovery_tree_root;
2303 dev_root->track_dirty = 1;
2305 ret = find_and_setup_root(tree_root, fs_info,
2306 BTRFS_CSUM_TREE_OBJECTID, csum_root);
2307 if (ret)
2308 goto recovery_tree_root;
2310 csum_root->track_dirty = 1;
2312 fs_info->generation = generation;
2313 fs_info->last_trans_committed = generation;
2314 fs_info->data_alloc_profile = (u64)-1;
2315 fs_info->metadata_alloc_profile = (u64)-1;
2316 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2318 ret = btrfs_init_space_info(fs_info);
2319 if (ret) {
2320 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2321 goto fail_block_groups;
2324 ret = btrfs_read_block_groups(extent_root);
2325 if (ret) {
2326 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2327 goto fail_block_groups;
2330 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2331 "btrfs-cleaner");
2332 if (IS_ERR(fs_info->cleaner_kthread))
2333 goto fail_block_groups;
2335 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2336 tree_root,
2337 "btrfs-transaction");
2338 if (IS_ERR(fs_info->transaction_kthread))
2339 goto fail_cleaner;
2341 if (!btrfs_test_opt(tree_root, SSD) &&
2342 !btrfs_test_opt(tree_root, NOSSD) &&
2343 !fs_info->fs_devices->rotating) {
2344 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2345 "mode\n");
2346 btrfs_set_opt(fs_info->mount_opt, SSD);
2349 /* do not make disk changes in broken FS */
2350 if (btrfs_super_log_root(disk_super) != 0 &&
2351 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2352 u64 bytenr = btrfs_super_log_root(disk_super);
2354 if (fs_devices->rw_devices == 0) {
2355 printk(KERN_WARNING "Btrfs log replay required "
2356 "on RO media\n");
2357 err = -EIO;
2358 goto fail_trans_kthread;
2360 blocksize =
2361 btrfs_level_size(tree_root,
2362 btrfs_super_log_root_level(disk_super));
2364 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2365 if (!log_tree_root) {
2366 err = -ENOMEM;
2367 goto fail_trans_kthread;
2370 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2371 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2373 log_tree_root->node = read_tree_block(tree_root, bytenr,
2374 blocksize,
2375 generation + 1);
2376 ret = btrfs_recover_log_trees(log_tree_root);
2377 BUG_ON(ret);
2379 if (sb->s_flags & MS_RDONLY) {
2380 ret = btrfs_commit_super(tree_root);
2381 BUG_ON(ret);
2385 ret = btrfs_find_orphan_roots(tree_root);
2386 BUG_ON(ret);
2388 if (!(sb->s_flags & MS_RDONLY)) {
2389 ret = btrfs_cleanup_fs_roots(fs_info);
2390 BUG_ON(ret);
2392 ret = btrfs_recover_relocation(tree_root);
2393 if (ret < 0) {
2394 printk(KERN_WARNING
2395 "btrfs: failed to recover relocation\n");
2396 err = -EINVAL;
2397 goto fail_trans_kthread;
2401 location.objectid = BTRFS_FS_TREE_OBJECTID;
2402 location.type = BTRFS_ROOT_ITEM_KEY;
2403 location.offset = (u64)-1;
2405 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2406 if (!fs_info->fs_root)
2407 goto fail_trans_kthread;
2408 if (IS_ERR(fs_info->fs_root)) {
2409 err = PTR_ERR(fs_info->fs_root);
2410 goto fail_trans_kthread;
2413 if (!(sb->s_flags & MS_RDONLY)) {
2414 down_read(&fs_info->cleanup_work_sem);
2415 err = btrfs_orphan_cleanup(fs_info->fs_root);
2416 if (!err)
2417 err = btrfs_orphan_cleanup(fs_info->tree_root);
2418 up_read(&fs_info->cleanup_work_sem);
2419 if (err) {
2420 close_ctree(tree_root);
2421 return ERR_PTR(err);
2425 return tree_root;
2427 fail_trans_kthread:
2428 kthread_stop(fs_info->transaction_kthread);
2429 fail_cleaner:
2430 kthread_stop(fs_info->cleaner_kthread);
2433 * make sure we're done with the btree inode before we stop our
2434 * kthreads
2436 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2437 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2439 fail_block_groups:
2440 btrfs_free_block_groups(fs_info);
2442 fail_tree_roots:
2443 free_root_pointers(fs_info, 1);
2445 fail_sb_buffer:
2446 btrfs_stop_workers(&fs_info->generic_worker);
2447 btrfs_stop_workers(&fs_info->readahead_workers);
2448 btrfs_stop_workers(&fs_info->fixup_workers);
2449 btrfs_stop_workers(&fs_info->delalloc_workers);
2450 btrfs_stop_workers(&fs_info->workers);
2451 btrfs_stop_workers(&fs_info->endio_workers);
2452 btrfs_stop_workers(&fs_info->endio_meta_workers);
2453 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2454 btrfs_stop_workers(&fs_info->endio_write_workers);
2455 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2456 btrfs_stop_workers(&fs_info->submit_workers);
2457 btrfs_stop_workers(&fs_info->delayed_workers);
2458 btrfs_stop_workers(&fs_info->caching_workers);
2459 fail_alloc:
2460 fail_iput:
2461 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2462 iput(fs_info->btree_inode);
2464 btrfs_close_devices(fs_info->fs_devices);
2465 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2466 fail_bdi:
2467 bdi_destroy(&fs_info->bdi);
2468 fail_srcu:
2469 cleanup_srcu_struct(&fs_info->subvol_srcu);
2470 fail:
2471 free_fs_info(fs_info);
2472 return ERR_PTR(err);
2474 recovery_tree_root:
2476 if (!btrfs_test_opt(tree_root, RECOVERY))
2477 goto fail_tree_roots;
2479 free_root_pointers(fs_info, 0);
2481 /* don't use the log in recovery mode, it won't be valid */
2482 btrfs_set_super_log_root(disk_super, 0);
2484 /* we can't trust the free space cache either */
2485 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2487 ret = next_root_backup(fs_info, fs_info->super_copy,
2488 &num_backups_tried, &backup_index);
2489 if (ret == -1)
2490 goto fail_block_groups;
2491 goto retry_root_backup;
2494 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2496 char b[BDEVNAME_SIZE];
2498 if (uptodate) {
2499 set_buffer_uptodate(bh);
2500 } else {
2501 printk_ratelimited(KERN_WARNING "lost page write due to "
2502 "I/O error on %s\n",
2503 bdevname(bh->b_bdev, b));
2504 /* note, we dont' set_buffer_write_io_error because we have
2505 * our own ways of dealing with the IO errors
2507 clear_buffer_uptodate(bh);
2509 unlock_buffer(bh);
2510 put_bh(bh);
2513 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2515 struct buffer_head *bh;
2516 struct buffer_head *latest = NULL;
2517 struct btrfs_super_block *super;
2518 int i;
2519 u64 transid = 0;
2520 u64 bytenr;
2522 /* we would like to check all the supers, but that would make
2523 * a btrfs mount succeed after a mkfs from a different FS.
2524 * So, we need to add a special mount option to scan for
2525 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2527 for (i = 0; i < 1; i++) {
2528 bytenr = btrfs_sb_offset(i);
2529 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2530 break;
2531 bh = __bread(bdev, bytenr / 4096, 4096);
2532 if (!bh)
2533 continue;
2535 super = (struct btrfs_super_block *)bh->b_data;
2536 if (btrfs_super_bytenr(super) != bytenr ||
2537 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2538 sizeof(super->magic))) {
2539 brelse(bh);
2540 continue;
2543 if (!latest || btrfs_super_generation(super) > transid) {
2544 brelse(latest);
2545 latest = bh;
2546 transid = btrfs_super_generation(super);
2547 } else {
2548 brelse(bh);
2551 return latest;
2555 * this should be called twice, once with wait == 0 and
2556 * once with wait == 1. When wait == 0 is done, all the buffer heads
2557 * we write are pinned.
2559 * They are released when wait == 1 is done.
2560 * max_mirrors must be the same for both runs, and it indicates how
2561 * many supers on this one device should be written.
2563 * max_mirrors == 0 means to write them all.
2565 static int write_dev_supers(struct btrfs_device *device,
2566 struct btrfs_super_block *sb,
2567 int do_barriers, int wait, int max_mirrors)
2569 struct buffer_head *bh;
2570 int i;
2571 int ret;
2572 int errors = 0;
2573 u32 crc;
2574 u64 bytenr;
2575 int last_barrier = 0;
2577 if (max_mirrors == 0)
2578 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2580 /* make sure only the last submit_bh does a barrier */
2581 if (do_barriers) {
2582 for (i = 0; i < max_mirrors; i++) {
2583 bytenr = btrfs_sb_offset(i);
2584 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2585 device->total_bytes)
2586 break;
2587 last_barrier = i;
2591 for (i = 0; i < max_mirrors; i++) {
2592 bytenr = btrfs_sb_offset(i);
2593 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2594 break;
2596 if (wait) {
2597 bh = __find_get_block(device->bdev, bytenr / 4096,
2598 BTRFS_SUPER_INFO_SIZE);
2599 BUG_ON(!bh);
2600 wait_on_buffer(bh);
2601 if (!buffer_uptodate(bh))
2602 errors++;
2604 /* drop our reference */
2605 brelse(bh);
2607 /* drop the reference from the wait == 0 run */
2608 brelse(bh);
2609 continue;
2610 } else {
2611 btrfs_set_super_bytenr(sb, bytenr);
2613 crc = ~(u32)0;
2614 crc = btrfs_csum_data(NULL, (char *)sb +
2615 BTRFS_CSUM_SIZE, crc,
2616 BTRFS_SUPER_INFO_SIZE -
2617 BTRFS_CSUM_SIZE);
2618 btrfs_csum_final(crc, sb->csum);
2621 * one reference for us, and we leave it for the
2622 * caller
2624 bh = __getblk(device->bdev, bytenr / 4096,
2625 BTRFS_SUPER_INFO_SIZE);
2626 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2628 /* one reference for submit_bh */
2629 get_bh(bh);
2631 set_buffer_uptodate(bh);
2632 lock_buffer(bh);
2633 bh->b_end_io = btrfs_end_buffer_write_sync;
2636 if (i == last_barrier && do_barriers)
2637 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2638 else
2639 ret = submit_bh(WRITE_SYNC, bh);
2641 if (ret)
2642 errors++;
2644 return errors < i ? 0 : -1;
2647 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2649 struct list_head *head;
2650 struct btrfs_device *dev;
2651 struct btrfs_super_block *sb;
2652 struct btrfs_dev_item *dev_item;
2653 int ret;
2654 int do_barriers;
2655 int max_errors;
2656 int total_errors = 0;
2657 u64 flags;
2659 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2660 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2661 backup_super_roots(root->fs_info);
2663 sb = root->fs_info->super_for_commit;
2664 dev_item = &sb->dev_item;
2666 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2667 head = &root->fs_info->fs_devices->devices;
2668 list_for_each_entry_rcu(dev, head, dev_list) {
2669 if (!dev->bdev) {
2670 total_errors++;
2671 continue;
2673 if (!dev->in_fs_metadata || !dev->writeable)
2674 continue;
2676 btrfs_set_stack_device_generation(dev_item, 0);
2677 btrfs_set_stack_device_type(dev_item, dev->type);
2678 btrfs_set_stack_device_id(dev_item, dev->devid);
2679 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2680 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2681 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2682 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2683 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2684 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2685 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2687 flags = btrfs_super_flags(sb);
2688 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2690 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2691 if (ret)
2692 total_errors++;
2694 if (total_errors > max_errors) {
2695 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2696 total_errors);
2697 BUG();
2700 total_errors = 0;
2701 list_for_each_entry_rcu(dev, head, dev_list) {
2702 if (!dev->bdev)
2703 continue;
2704 if (!dev->in_fs_metadata || !dev->writeable)
2705 continue;
2707 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2708 if (ret)
2709 total_errors++;
2711 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2712 if (total_errors > max_errors) {
2713 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2714 total_errors);
2715 BUG();
2717 return 0;
2720 int write_ctree_super(struct btrfs_trans_handle *trans,
2721 struct btrfs_root *root, int max_mirrors)
2723 int ret;
2725 ret = write_all_supers(root, max_mirrors);
2726 return ret;
2729 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2731 spin_lock(&fs_info->fs_roots_radix_lock);
2732 radix_tree_delete(&fs_info->fs_roots_radix,
2733 (unsigned long)root->root_key.objectid);
2734 spin_unlock(&fs_info->fs_roots_radix_lock);
2736 if (btrfs_root_refs(&root->root_item) == 0)
2737 synchronize_srcu(&fs_info->subvol_srcu);
2739 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2740 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2741 free_fs_root(root);
2742 return 0;
2745 static void free_fs_root(struct btrfs_root *root)
2747 iput(root->cache_inode);
2748 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2749 if (root->anon_dev)
2750 free_anon_bdev(root->anon_dev);
2751 free_extent_buffer(root->node);
2752 free_extent_buffer(root->commit_root);
2753 kfree(root->free_ino_ctl);
2754 kfree(root->free_ino_pinned);
2755 kfree(root->name);
2756 kfree(root);
2759 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2761 int ret;
2762 struct btrfs_root *gang[8];
2763 int i;
2765 while (!list_empty(&fs_info->dead_roots)) {
2766 gang[0] = list_entry(fs_info->dead_roots.next,
2767 struct btrfs_root, root_list);
2768 list_del(&gang[0]->root_list);
2770 if (gang[0]->in_radix) {
2771 btrfs_free_fs_root(fs_info, gang[0]);
2772 } else {
2773 free_extent_buffer(gang[0]->node);
2774 free_extent_buffer(gang[0]->commit_root);
2775 kfree(gang[0]);
2779 while (1) {
2780 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2781 (void **)gang, 0,
2782 ARRAY_SIZE(gang));
2783 if (!ret)
2784 break;
2785 for (i = 0; i < ret; i++)
2786 btrfs_free_fs_root(fs_info, gang[i]);
2788 return 0;
2791 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2793 u64 root_objectid = 0;
2794 struct btrfs_root *gang[8];
2795 int i;
2796 int ret;
2798 while (1) {
2799 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2800 (void **)gang, root_objectid,
2801 ARRAY_SIZE(gang));
2802 if (!ret)
2803 break;
2805 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2806 for (i = 0; i < ret; i++) {
2807 int err;
2809 root_objectid = gang[i]->root_key.objectid;
2810 err = btrfs_orphan_cleanup(gang[i]);
2811 if (err)
2812 return err;
2814 root_objectid++;
2816 return 0;
2819 int btrfs_commit_super(struct btrfs_root *root)
2821 struct btrfs_trans_handle *trans;
2822 int ret;
2824 mutex_lock(&root->fs_info->cleaner_mutex);
2825 btrfs_run_delayed_iputs(root);
2826 btrfs_clean_old_snapshots(root);
2827 mutex_unlock(&root->fs_info->cleaner_mutex);
2829 /* wait until ongoing cleanup work done */
2830 down_write(&root->fs_info->cleanup_work_sem);
2831 up_write(&root->fs_info->cleanup_work_sem);
2833 trans = btrfs_join_transaction(root);
2834 if (IS_ERR(trans))
2835 return PTR_ERR(trans);
2836 ret = btrfs_commit_transaction(trans, root);
2837 BUG_ON(ret);
2838 /* run commit again to drop the original snapshot */
2839 trans = btrfs_join_transaction(root);
2840 if (IS_ERR(trans))
2841 return PTR_ERR(trans);
2842 btrfs_commit_transaction(trans, root);
2843 ret = btrfs_write_and_wait_transaction(NULL, root);
2844 BUG_ON(ret);
2846 ret = write_ctree_super(NULL, root, 0);
2847 return ret;
2850 int close_ctree(struct btrfs_root *root)
2852 struct btrfs_fs_info *fs_info = root->fs_info;
2853 int ret;
2855 fs_info->closing = 1;
2856 smp_mb();
2858 btrfs_scrub_cancel(root);
2860 /* wait for any defraggers to finish */
2861 wait_event(fs_info->transaction_wait,
2862 (atomic_read(&fs_info->defrag_running) == 0));
2864 /* clear out the rbtree of defraggable inodes */
2865 btrfs_run_defrag_inodes(root->fs_info);
2868 * Here come 2 situations when btrfs is broken to flip readonly:
2870 * 1. when btrfs flips readonly somewhere else before
2871 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2872 * and btrfs will skip to write sb directly to keep
2873 * ERROR state on disk.
2875 * 2. when btrfs flips readonly just in btrfs_commit_super,
2876 * and in such case, btrfs cannot write sb via btrfs_commit_super,
2877 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2878 * btrfs will cleanup all FS resources first and write sb then.
2880 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2881 ret = btrfs_commit_super(root);
2882 if (ret)
2883 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2886 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2887 ret = btrfs_error_commit_super(root);
2888 if (ret)
2889 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2892 btrfs_put_block_group_cache(fs_info);
2894 kthread_stop(root->fs_info->transaction_kthread);
2895 kthread_stop(root->fs_info->cleaner_kthread);
2897 fs_info->closing = 2;
2898 smp_mb();
2900 if (fs_info->delalloc_bytes) {
2901 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2902 (unsigned long long)fs_info->delalloc_bytes);
2904 if (fs_info->total_ref_cache_size) {
2905 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2906 (unsigned long long)fs_info->total_ref_cache_size);
2909 free_extent_buffer(fs_info->extent_root->node);
2910 free_extent_buffer(fs_info->extent_root->commit_root);
2911 free_extent_buffer(fs_info->tree_root->node);
2912 free_extent_buffer(fs_info->tree_root->commit_root);
2913 free_extent_buffer(root->fs_info->chunk_root->node);
2914 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2915 free_extent_buffer(root->fs_info->dev_root->node);
2916 free_extent_buffer(root->fs_info->dev_root->commit_root);
2917 free_extent_buffer(root->fs_info->csum_root->node);
2918 free_extent_buffer(root->fs_info->csum_root->commit_root);
2920 btrfs_free_block_groups(root->fs_info);
2922 del_fs_roots(fs_info);
2924 iput(fs_info->btree_inode);
2926 btrfs_stop_workers(&fs_info->generic_worker);
2927 btrfs_stop_workers(&fs_info->fixup_workers);
2928 btrfs_stop_workers(&fs_info->delalloc_workers);
2929 btrfs_stop_workers(&fs_info->workers);
2930 btrfs_stop_workers(&fs_info->endio_workers);
2931 btrfs_stop_workers(&fs_info->endio_meta_workers);
2932 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2933 btrfs_stop_workers(&fs_info->endio_write_workers);
2934 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2935 btrfs_stop_workers(&fs_info->submit_workers);
2936 btrfs_stop_workers(&fs_info->delayed_workers);
2937 btrfs_stop_workers(&fs_info->caching_workers);
2938 btrfs_stop_workers(&fs_info->readahead_workers);
2940 btrfs_close_devices(fs_info->fs_devices);
2941 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2943 bdi_destroy(&fs_info->bdi);
2944 cleanup_srcu_struct(&fs_info->subvol_srcu);
2946 free_fs_info(fs_info);
2948 return 0;
2951 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2953 int ret;
2954 struct inode *btree_inode = buf->first_page->mapping->host;
2956 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2957 NULL);
2958 if (!ret)
2959 return ret;
2961 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2962 parent_transid);
2963 return !ret;
2966 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2968 struct inode *btree_inode = buf->first_page->mapping->host;
2969 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2970 buf);
2973 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2975 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2976 u64 transid = btrfs_header_generation(buf);
2977 struct inode *btree_inode = root->fs_info->btree_inode;
2978 int was_dirty;
2980 btrfs_assert_tree_locked(buf);
2981 if (transid != root->fs_info->generation) {
2982 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2983 "found %llu running %llu\n",
2984 (unsigned long long)buf->start,
2985 (unsigned long long)transid,
2986 (unsigned long long)root->fs_info->generation);
2987 WARN_ON(1);
2989 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2990 buf);
2991 if (!was_dirty) {
2992 spin_lock(&root->fs_info->delalloc_lock);
2993 root->fs_info->dirty_metadata_bytes += buf->len;
2994 spin_unlock(&root->fs_info->delalloc_lock);
2998 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3001 * looks as though older kernels can get into trouble with
3002 * this code, they end up stuck in balance_dirty_pages forever
3004 u64 num_dirty;
3005 unsigned long thresh = 32 * 1024 * 1024;
3007 if (current->flags & PF_MEMALLOC)
3008 return;
3010 btrfs_balance_delayed_items(root);
3012 num_dirty = root->fs_info->dirty_metadata_bytes;
3014 if (num_dirty > thresh) {
3015 balance_dirty_pages_ratelimited_nr(
3016 root->fs_info->btree_inode->i_mapping, 1);
3018 return;
3021 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
3024 * looks as though older kernels can get into trouble with
3025 * this code, they end up stuck in balance_dirty_pages forever
3027 u64 num_dirty;
3028 unsigned long thresh = 32 * 1024 * 1024;
3030 if (current->flags & PF_MEMALLOC)
3031 return;
3033 num_dirty = root->fs_info->dirty_metadata_bytes;
3035 if (num_dirty > thresh) {
3036 balance_dirty_pages_ratelimited_nr(
3037 root->fs_info->btree_inode->i_mapping, 1);
3039 return;
3042 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3044 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
3045 int ret;
3046 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3047 if (ret == 0)
3048 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
3049 return ret;
3052 static int btree_lock_page_hook(struct page *page, void *data,
3053 void (*flush_fn)(void *))
3055 struct inode *inode = page->mapping->host;
3056 struct btrfs_root *root = BTRFS_I(inode)->root;
3057 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3058 struct extent_buffer *eb;
3059 unsigned long len;
3060 u64 bytenr = page_offset(page);
3062 if (page->private == EXTENT_PAGE_PRIVATE)
3063 goto out;
3065 len = page->private >> 2;
3066 eb = find_extent_buffer(io_tree, bytenr, len);
3067 if (!eb)
3068 goto out;
3070 if (!btrfs_try_tree_write_lock(eb)) {
3071 flush_fn(data);
3072 btrfs_tree_lock(eb);
3074 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
3076 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
3077 spin_lock(&root->fs_info->delalloc_lock);
3078 if (root->fs_info->dirty_metadata_bytes >= eb->len)
3079 root->fs_info->dirty_metadata_bytes -= eb->len;
3080 else
3081 WARN_ON(1);
3082 spin_unlock(&root->fs_info->delalloc_lock);
3085 btrfs_tree_unlock(eb);
3086 free_extent_buffer(eb);
3087 out:
3088 if (!trylock_page(page)) {
3089 flush_fn(data);
3090 lock_page(page);
3092 return 0;
3095 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3096 int read_only)
3098 if (read_only)
3099 return;
3101 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3102 printk(KERN_WARNING "warning: mount fs with errors, "
3103 "running btrfsck is recommended\n");
3106 int btrfs_error_commit_super(struct btrfs_root *root)
3108 int ret;
3110 mutex_lock(&root->fs_info->cleaner_mutex);
3111 btrfs_run_delayed_iputs(root);
3112 mutex_unlock(&root->fs_info->cleaner_mutex);
3114 down_write(&root->fs_info->cleanup_work_sem);
3115 up_write(&root->fs_info->cleanup_work_sem);
3117 /* cleanup FS via transaction */
3118 btrfs_cleanup_transaction(root);
3120 ret = write_ctree_super(NULL, root, 0);
3122 return ret;
3125 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
3127 struct btrfs_inode *btrfs_inode;
3128 struct list_head splice;
3130 INIT_LIST_HEAD(&splice);
3132 mutex_lock(&root->fs_info->ordered_operations_mutex);
3133 spin_lock(&root->fs_info->ordered_extent_lock);
3135 list_splice_init(&root->fs_info->ordered_operations, &splice);
3136 while (!list_empty(&splice)) {
3137 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3138 ordered_operations);
3140 list_del_init(&btrfs_inode->ordered_operations);
3142 btrfs_invalidate_inodes(btrfs_inode->root);
3145 spin_unlock(&root->fs_info->ordered_extent_lock);
3146 mutex_unlock(&root->fs_info->ordered_operations_mutex);
3148 return 0;
3151 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
3153 struct list_head splice;
3154 struct btrfs_ordered_extent *ordered;
3155 struct inode *inode;
3157 INIT_LIST_HEAD(&splice);
3159 spin_lock(&root->fs_info->ordered_extent_lock);
3161 list_splice_init(&root->fs_info->ordered_extents, &splice);
3162 while (!list_empty(&splice)) {
3163 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3164 root_extent_list);
3166 list_del_init(&ordered->root_extent_list);
3167 atomic_inc(&ordered->refs);
3169 /* the inode may be getting freed (in sys_unlink path). */
3170 inode = igrab(ordered->inode);
3172 spin_unlock(&root->fs_info->ordered_extent_lock);
3173 if (inode)
3174 iput(inode);
3176 atomic_set(&ordered->refs, 1);
3177 btrfs_put_ordered_extent(ordered);
3179 spin_lock(&root->fs_info->ordered_extent_lock);
3182 spin_unlock(&root->fs_info->ordered_extent_lock);
3184 return 0;
3187 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3188 struct btrfs_root *root)
3190 struct rb_node *node;
3191 struct btrfs_delayed_ref_root *delayed_refs;
3192 struct btrfs_delayed_ref_node *ref;
3193 int ret = 0;
3195 delayed_refs = &trans->delayed_refs;
3197 spin_lock(&delayed_refs->lock);
3198 if (delayed_refs->num_entries == 0) {
3199 spin_unlock(&delayed_refs->lock);
3200 printk(KERN_INFO "delayed_refs has NO entry\n");
3201 return ret;
3204 node = rb_first(&delayed_refs->root);
3205 while (node) {
3206 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3207 node = rb_next(node);
3209 ref->in_tree = 0;
3210 rb_erase(&ref->rb_node, &delayed_refs->root);
3211 delayed_refs->num_entries--;
3213 atomic_set(&ref->refs, 1);
3214 if (btrfs_delayed_ref_is_head(ref)) {
3215 struct btrfs_delayed_ref_head *head;
3217 head = btrfs_delayed_node_to_head(ref);
3218 mutex_lock(&head->mutex);
3219 kfree(head->extent_op);
3220 delayed_refs->num_heads--;
3221 if (list_empty(&head->cluster))
3222 delayed_refs->num_heads_ready--;
3223 list_del_init(&head->cluster);
3224 mutex_unlock(&head->mutex);
3227 spin_unlock(&delayed_refs->lock);
3228 btrfs_put_delayed_ref(ref);
3230 cond_resched();
3231 spin_lock(&delayed_refs->lock);
3234 spin_unlock(&delayed_refs->lock);
3236 return ret;
3239 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3241 struct btrfs_pending_snapshot *snapshot;
3242 struct list_head splice;
3244 INIT_LIST_HEAD(&splice);
3246 list_splice_init(&t->pending_snapshots, &splice);
3248 while (!list_empty(&splice)) {
3249 snapshot = list_entry(splice.next,
3250 struct btrfs_pending_snapshot,
3251 list);
3253 list_del_init(&snapshot->list);
3255 kfree(snapshot);
3258 return 0;
3261 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3263 struct btrfs_inode *btrfs_inode;
3264 struct list_head splice;
3266 INIT_LIST_HEAD(&splice);
3268 spin_lock(&root->fs_info->delalloc_lock);
3269 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3271 while (!list_empty(&splice)) {
3272 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3273 delalloc_inodes);
3275 list_del_init(&btrfs_inode->delalloc_inodes);
3277 btrfs_invalidate_inodes(btrfs_inode->root);
3280 spin_unlock(&root->fs_info->delalloc_lock);
3282 return 0;
3285 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3286 struct extent_io_tree *dirty_pages,
3287 int mark)
3289 int ret;
3290 struct page *page;
3291 struct inode *btree_inode = root->fs_info->btree_inode;
3292 struct extent_buffer *eb;
3293 u64 start = 0;
3294 u64 end;
3295 u64 offset;
3296 unsigned long index;
3298 while (1) {
3299 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3300 mark);
3301 if (ret)
3302 break;
3304 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3305 while (start <= end) {
3306 index = start >> PAGE_CACHE_SHIFT;
3307 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3308 page = find_get_page(btree_inode->i_mapping, index);
3309 if (!page)
3310 continue;
3311 offset = page_offset(page);
3313 spin_lock(&dirty_pages->buffer_lock);
3314 eb = radix_tree_lookup(
3315 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3316 offset >> PAGE_CACHE_SHIFT);
3317 spin_unlock(&dirty_pages->buffer_lock);
3318 if (eb) {
3319 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3320 &eb->bflags);
3321 atomic_set(&eb->refs, 1);
3323 if (PageWriteback(page))
3324 end_page_writeback(page);
3326 lock_page(page);
3327 if (PageDirty(page)) {
3328 clear_page_dirty_for_io(page);
3329 spin_lock_irq(&page->mapping->tree_lock);
3330 radix_tree_tag_clear(&page->mapping->page_tree,
3331 page_index(page),
3332 PAGECACHE_TAG_DIRTY);
3333 spin_unlock_irq(&page->mapping->tree_lock);
3336 page->mapping->a_ops->invalidatepage(page, 0);
3337 unlock_page(page);
3341 return ret;
3344 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3345 struct extent_io_tree *pinned_extents)
3347 struct extent_io_tree *unpin;
3348 u64 start;
3349 u64 end;
3350 int ret;
3352 unpin = pinned_extents;
3353 while (1) {
3354 ret = find_first_extent_bit(unpin, 0, &start, &end,
3355 EXTENT_DIRTY);
3356 if (ret)
3357 break;
3359 /* opt_discard */
3360 if (btrfs_test_opt(root, DISCARD))
3361 ret = btrfs_error_discard_extent(root, start,
3362 end + 1 - start,
3363 NULL);
3365 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3366 btrfs_error_unpin_extent_range(root, start, end);
3367 cond_resched();
3370 return 0;
3373 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3375 struct btrfs_transaction *t;
3376 LIST_HEAD(list);
3378 WARN_ON(1);
3380 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3382 spin_lock(&root->fs_info->trans_lock);
3383 list_splice_init(&root->fs_info->trans_list, &list);
3384 root->fs_info->trans_no_join = 1;
3385 spin_unlock(&root->fs_info->trans_lock);
3387 while (!list_empty(&list)) {
3388 t = list_entry(list.next, struct btrfs_transaction, list);
3389 if (!t)
3390 break;
3392 btrfs_destroy_ordered_operations(root);
3394 btrfs_destroy_ordered_extents(root);
3396 btrfs_destroy_delayed_refs(t, root);
3398 btrfs_block_rsv_release(root,
3399 &root->fs_info->trans_block_rsv,
3400 t->dirty_pages.dirty_bytes);
3402 /* FIXME: cleanup wait for commit */
3403 t->in_commit = 1;
3404 t->blocked = 1;
3405 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3406 wake_up(&root->fs_info->transaction_blocked_wait);
3408 t->blocked = 0;
3409 if (waitqueue_active(&root->fs_info->transaction_wait))
3410 wake_up(&root->fs_info->transaction_wait);
3412 t->commit_done = 1;
3413 if (waitqueue_active(&t->commit_wait))
3414 wake_up(&t->commit_wait);
3416 btrfs_destroy_pending_snapshots(t);
3418 btrfs_destroy_delalloc_inodes(root);
3420 spin_lock(&root->fs_info->trans_lock);
3421 root->fs_info->running_transaction = NULL;
3422 spin_unlock(&root->fs_info->trans_lock);
3424 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3425 EXTENT_DIRTY);
3427 btrfs_destroy_pinned_extent(root,
3428 root->fs_info->pinned_extents);
3430 atomic_set(&t->use_count, 0);
3431 list_del_init(&t->list);
3432 memset(t, 0, sizeof(*t));
3433 kmem_cache_free(btrfs_transaction_cachep, t);
3436 spin_lock(&root->fs_info->trans_lock);
3437 root->fs_info->trans_no_join = 0;
3438 spin_unlock(&root->fs_info->trans_lock);
3439 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3441 return 0;
3444 static struct extent_io_ops btree_extent_io_ops = {
3445 .write_cache_pages_lock_hook = btree_lock_page_hook,
3446 .readpage_end_io_hook = btree_readpage_end_io_hook,
3447 .readpage_io_failed_hook = btree_io_failed_hook,
3448 .submit_bio_hook = btree_submit_bio_hook,
3449 /* note we're sharing with inode.c for the merge bio hook */
3450 .merge_bio_hook = btrfs_merge_bio_hook,