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.
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/slab.h>
30 #include <linux/migrate.h>
31 #include <linux/ratelimit.h>
32 #include <linux/uuid.h>
33 #include <linux/semaphore.h>
34 #include <asm/unaligned.h>
38 #include "transaction.h"
39 #include "btrfs_inode.h"
41 #include "print-tree.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
54 #include <asm/cpufeature.h>
57 static struct extent_io_ops btree_extent_io_ops;
58 static void end_workqueue_fn(struct btrfs_work *work);
59 static void free_fs_root(struct btrfs_root *root);
60 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
62 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
63 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
64 struct btrfs_root *root);
65 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
66 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
67 struct extent_io_tree *dirty_pages,
69 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
70 struct extent_io_tree *pinned_extents);
71 static int btrfs_cleanup_transaction(struct btrfs_root *root);
72 static void btrfs_error_commit_super(struct btrfs_root *root);
75 * end_io_wq structs are used to do processing in task context when an IO is
76 * complete. This is used during reads to verify checksums, and it is used
77 * by writes to insert metadata for new file extents after IO is complete.
83 struct btrfs_fs_info *info;
86 struct list_head list;
87 struct btrfs_work work;
91 * async submit bios are used to offload expensive checksumming
92 * onto the worker threads. They checksum file and metadata bios
93 * just before they are sent down the IO stack.
95 struct async_submit_bio {
98 struct list_head list;
99 extent_submit_bio_hook_t *submit_bio_start;
100 extent_submit_bio_hook_t *submit_bio_done;
103 unsigned long bio_flags;
105 * bio_offset is optional, can be used if the pages in the bio
106 * can't tell us where in the file the bio should go
109 struct btrfs_work work;
114 * Lockdep class keys for extent_buffer->lock's in this root. For a given
115 * eb, the lockdep key is determined by the btrfs_root it belongs to and
116 * the level the eb occupies in the tree.
118 * Different roots are used for different purposes and may nest inside each
119 * other and they require separate keysets. As lockdep keys should be
120 * static, assign keysets according to the purpose of the root as indicated
121 * by btrfs_root->objectid. This ensures that all special purpose roots
122 * have separate keysets.
124 * Lock-nesting across peer nodes is always done with the immediate parent
125 * node locked thus preventing deadlock. As lockdep doesn't know this, use
126 * subclass to avoid triggering lockdep warning in such cases.
128 * The key is set by the readpage_end_io_hook after the buffer has passed
129 * csum validation but before the pages are unlocked. It is also set by
130 * btrfs_init_new_buffer on freshly allocated blocks.
132 * We also add a check to make sure the highest level of the tree is the
133 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
134 * needs update as well.
136 #ifdef CONFIG_DEBUG_LOCK_ALLOC
137 # if BTRFS_MAX_LEVEL != 8
141 static struct btrfs_lockdep_keyset {
142 u64 id; /* root objectid */
143 const char *name_stem; /* lock name stem */
144 char names[BTRFS_MAX_LEVEL + 1][20];
145 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
146 } btrfs_lockdep_keysets[] = {
147 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
148 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
149 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
150 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
151 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
152 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
153 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
154 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
155 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
156 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
157 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
158 { .id = 0, .name_stem = "tree" },
161 void __init btrfs_init_lockdep(void)
165 /* initialize lockdep class names */
166 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
167 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
169 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
170 snprintf(ks->names[j], sizeof(ks->names[j]),
171 "btrfs-%s-%02d", ks->name_stem, j);
175 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
178 struct btrfs_lockdep_keyset *ks;
180 BUG_ON(level >= ARRAY_SIZE(ks->keys));
182 /* find the matching keyset, id 0 is the default entry */
183 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
184 if (ks->id == objectid)
187 lockdep_set_class_and_name(&eb->lock,
188 &ks->keys[level], ks->names[level]);
194 * extents on the btree inode are pretty simple, there's one extent
195 * that covers the entire device
197 static struct extent_map *btree_get_extent(struct inode *inode,
198 struct page *page, size_t pg_offset, u64 start, u64 len,
201 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
202 struct extent_map *em;
205 read_lock(&em_tree->lock);
206 em = lookup_extent_mapping(em_tree, start, len);
209 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
210 read_unlock(&em_tree->lock);
213 read_unlock(&em_tree->lock);
215 em = alloc_extent_map();
217 em = ERR_PTR(-ENOMEM);
222 em->block_len = (u64)-1;
224 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
226 write_lock(&em_tree->lock);
227 ret = add_extent_mapping(em_tree, em, 0);
228 if (ret == -EEXIST) {
230 em = lookup_extent_mapping(em_tree, start, len);
237 write_unlock(&em_tree->lock);
243 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
245 return btrfs_crc32c(seed, data, len);
248 void btrfs_csum_final(u32 crc, char *result)
250 put_unaligned_le32(~crc, result);
254 * compute the csum for a btree block, and either verify it or write it
255 * into the csum field of the block.
257 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
260 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
263 unsigned long cur_len;
264 unsigned long offset = BTRFS_CSUM_SIZE;
266 unsigned long map_start;
267 unsigned long map_len;
270 unsigned long inline_result;
272 len = buf->len - offset;
274 err = map_private_extent_buffer(buf, offset, 32,
275 &kaddr, &map_start, &map_len);
278 cur_len = min(len, map_len - (offset - map_start));
279 crc = btrfs_csum_data(kaddr + offset - map_start,
284 if (csum_size > sizeof(inline_result)) {
285 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
289 result = (char *)&inline_result;
292 btrfs_csum_final(crc, result);
295 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
298 memcpy(&found, result, csum_size);
300 read_extent_buffer(buf, &val, 0, csum_size);
301 printk_ratelimited(KERN_INFO
302 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
304 root->fs_info->sb->s_id, buf->start,
305 val, found, btrfs_header_level(buf));
306 if (result != (char *)&inline_result)
311 write_extent_buffer(buf, result, 0, csum_size);
313 if (result != (char *)&inline_result)
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,
328 struct extent_state *cached_state = NULL;
330 bool need_lock = (current->journal_info ==
331 (void *)BTRFS_SEND_TRANS_STUB);
333 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
340 btrfs_tree_read_lock(eb);
341 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
344 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
346 if (extent_buffer_uptodate(eb) &&
347 btrfs_header_generation(eb) == parent_transid) {
351 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
353 eb->start, parent_transid, btrfs_header_generation(eb));
357 * Things reading via commit roots that don't have normal protection,
358 * like send, can have a really old block in cache that may point at a
359 * block that has been free'd and re-allocated. So don't clear uptodate
360 * if we find an eb that is under IO (dirty/writeback) because we could
361 * end up reading in the stale data and then writing it back out and
362 * making everybody very sad.
364 if (!extent_buffer_under_io(eb))
365 clear_extent_buffer_uptodate(eb);
367 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
368 &cached_state, GFP_NOFS);
370 btrfs_tree_read_unlock_blocking(eb);
375 * Return 0 if the superblock checksum type matches the checksum value of that
376 * algorithm. Pass the raw disk superblock data.
378 static int btrfs_check_super_csum(char *raw_disk_sb)
380 struct btrfs_super_block *disk_sb =
381 (struct btrfs_super_block *)raw_disk_sb;
382 u16 csum_type = btrfs_super_csum_type(disk_sb);
385 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
387 const int csum_size = sizeof(crc);
388 char result[csum_size];
391 * The super_block structure does not span the whole
392 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
393 * is filled with zeros and is included in the checkum.
395 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
396 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
397 btrfs_csum_final(crc, result);
399 if (memcmp(raw_disk_sb, result, csum_size))
402 if (ret && btrfs_super_generation(disk_sb) < 10) {
404 "BTRFS: super block crcs don't match, older mkfs detected\n");
409 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
410 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
419 * helper to read a given tree block, doing retries as required when
420 * the checksums don't match and we have alternate mirrors to try.
422 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
423 struct extent_buffer *eb,
424 u64 start, u64 parent_transid)
426 struct extent_io_tree *io_tree;
431 int failed_mirror = 0;
433 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
434 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
436 ret = read_extent_buffer_pages(io_tree, eb, start,
438 btree_get_extent, mirror_num);
440 if (!verify_parent_transid(io_tree, eb,
448 * This buffer's crc is fine, but its contents are corrupted, so
449 * there is no reason to read the other copies, they won't be
452 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
455 num_copies = btrfs_num_copies(root->fs_info,
460 if (!failed_mirror) {
462 failed_mirror = eb->read_mirror;
466 if (mirror_num == failed_mirror)
469 if (mirror_num > num_copies)
473 if (failed && !ret && failed_mirror)
474 repair_eb_io_failure(root, eb, failed_mirror);
480 * checksum a dirty tree block before IO. This has extra checks to make sure
481 * we only fill in the checksum field in the first page of a multi-page block
484 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
486 u64 start = page_offset(page);
488 struct extent_buffer *eb;
490 eb = (struct extent_buffer *)page->private;
491 if (page != eb->pages[0])
493 found_start = btrfs_header_bytenr(eb);
494 if (WARN_ON(found_start != start || !PageUptodate(page)))
496 csum_tree_block(root, eb, 0);
500 static int check_tree_block_fsid(struct btrfs_root *root,
501 struct extent_buffer *eb)
503 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
504 u8 fsid[BTRFS_UUID_SIZE];
507 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
509 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
513 fs_devices = fs_devices->seed;
518 #define CORRUPT(reason, eb, root, slot) \
519 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
520 "root=%llu, slot=%d", reason, \
521 btrfs_header_bytenr(eb), root->objectid, slot)
523 static noinline int check_leaf(struct btrfs_root *root,
524 struct extent_buffer *leaf)
526 struct btrfs_key key;
527 struct btrfs_key leaf_key;
528 u32 nritems = btrfs_header_nritems(leaf);
534 /* Check the 0 item */
535 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
536 BTRFS_LEAF_DATA_SIZE(root)) {
537 CORRUPT("invalid item offset size pair", leaf, root, 0);
542 * Check to make sure each items keys are in the correct order and their
543 * offsets make sense. We only have to loop through nritems-1 because
544 * we check the current slot against the next slot, which verifies the
545 * next slot's offset+size makes sense and that the current's slot
548 for (slot = 0; slot < nritems - 1; slot++) {
549 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
550 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
552 /* Make sure the keys are in the right order */
553 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
554 CORRUPT("bad key order", leaf, root, slot);
559 * Make sure the offset and ends are right, remember that the
560 * item data starts at the end of the leaf and grows towards the
563 if (btrfs_item_offset_nr(leaf, slot) !=
564 btrfs_item_end_nr(leaf, slot + 1)) {
565 CORRUPT("slot offset bad", leaf, root, slot);
570 * Check to make sure that we don't point outside of the leaf,
571 * just incase all the items are consistent to eachother, but
572 * all point outside of the leaf.
574 if (btrfs_item_end_nr(leaf, slot) >
575 BTRFS_LEAF_DATA_SIZE(root)) {
576 CORRUPT("slot end outside of leaf", leaf, root, slot);
584 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
585 u64 phy_offset, struct page *page,
586 u64 start, u64 end, int mirror)
590 struct extent_buffer *eb;
591 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
598 eb = (struct extent_buffer *)page->private;
600 /* the pending IO might have been the only thing that kept this buffer
601 * in memory. Make sure we have a ref for all this other checks
603 extent_buffer_get(eb);
605 reads_done = atomic_dec_and_test(&eb->io_pages);
609 eb->read_mirror = mirror;
610 if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
615 found_start = btrfs_header_bytenr(eb);
616 if (found_start != eb->start) {
617 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
619 found_start, eb->start);
623 if (check_tree_block_fsid(root, eb)) {
624 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
629 found_level = btrfs_header_level(eb);
630 if (found_level >= BTRFS_MAX_LEVEL) {
631 btrfs_info(root->fs_info, "bad tree block level %d",
632 (int)btrfs_header_level(eb));
637 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
640 ret = csum_tree_block(root, eb, 1);
647 * If this is a leaf block and it is corrupt, set the corrupt bit so
648 * that we don't try and read the other copies of this block, just
651 if (found_level == 0 && check_leaf(root, eb)) {
652 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
657 set_extent_buffer_uptodate(eb);
660 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
661 btree_readahead_hook(root, eb, eb->start, ret);
665 * our io error hook is going to dec the io pages
666 * again, we have to make sure it has something
669 atomic_inc(&eb->io_pages);
670 clear_extent_buffer_uptodate(eb);
672 free_extent_buffer(eb);
677 static int btree_io_failed_hook(struct page *page, int failed_mirror)
679 struct extent_buffer *eb;
680 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
682 eb = (struct extent_buffer *)page->private;
683 set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
684 eb->read_mirror = failed_mirror;
685 atomic_dec(&eb->io_pages);
686 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
687 btree_readahead_hook(root, eb, eb->start, -EIO);
688 return -EIO; /* we fixed nothing */
691 static void end_workqueue_bio(struct bio *bio, int err)
693 struct end_io_wq *end_io_wq = bio->bi_private;
694 struct btrfs_fs_info *fs_info;
695 struct btrfs_workqueue *wq;
696 btrfs_work_func_t func;
698 fs_info = end_io_wq->info;
699 end_io_wq->error = err;
701 if (bio->bi_rw & REQ_WRITE) {
702 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
703 wq = fs_info->endio_meta_write_workers;
704 func = btrfs_endio_meta_write_helper;
705 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
706 wq = fs_info->endio_freespace_worker;
707 func = btrfs_freespace_write_helper;
708 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
709 wq = fs_info->endio_raid56_workers;
710 func = btrfs_endio_raid56_helper;
712 wq = fs_info->endio_write_workers;
713 func = btrfs_endio_write_helper;
716 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
717 wq = fs_info->endio_raid56_workers;
718 func = btrfs_endio_raid56_helper;
719 } else if (end_io_wq->metadata) {
720 wq = fs_info->endio_meta_workers;
721 func = btrfs_endio_meta_helper;
723 wq = fs_info->endio_workers;
724 func = btrfs_endio_helper;
728 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
729 btrfs_queue_work(wq, &end_io_wq->work);
733 * For the metadata arg you want
736 * 1 - if normal metadta
737 * 2 - if writing to the free space cache area
738 * 3 - raid parity work
740 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
743 struct end_io_wq *end_io_wq;
744 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
748 end_io_wq->private = bio->bi_private;
749 end_io_wq->end_io = bio->bi_end_io;
750 end_io_wq->info = info;
751 end_io_wq->error = 0;
752 end_io_wq->bio = bio;
753 end_io_wq->metadata = metadata;
755 bio->bi_private = end_io_wq;
756 bio->bi_end_io = end_workqueue_bio;
760 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
762 unsigned long limit = min_t(unsigned long,
763 info->thread_pool_size,
764 info->fs_devices->open_devices);
768 static void run_one_async_start(struct btrfs_work *work)
770 struct async_submit_bio *async;
773 async = container_of(work, struct async_submit_bio, work);
774 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
775 async->mirror_num, async->bio_flags,
781 static void run_one_async_done(struct btrfs_work *work)
783 struct btrfs_fs_info *fs_info;
784 struct async_submit_bio *async;
787 async = container_of(work, struct async_submit_bio, work);
788 fs_info = BTRFS_I(async->inode)->root->fs_info;
790 limit = btrfs_async_submit_limit(fs_info);
791 limit = limit * 2 / 3;
793 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
794 waitqueue_active(&fs_info->async_submit_wait))
795 wake_up(&fs_info->async_submit_wait);
797 /* If an error occured we just want to clean up the bio and move on */
799 bio_endio(async->bio, async->error);
803 async->submit_bio_done(async->inode, async->rw, async->bio,
804 async->mirror_num, async->bio_flags,
808 static void run_one_async_free(struct btrfs_work *work)
810 struct async_submit_bio *async;
812 async = container_of(work, struct async_submit_bio, work);
816 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
817 int rw, struct bio *bio, int mirror_num,
818 unsigned long bio_flags,
820 extent_submit_bio_hook_t *submit_bio_start,
821 extent_submit_bio_hook_t *submit_bio_done)
823 struct async_submit_bio *async;
825 async = kmalloc(sizeof(*async), GFP_NOFS);
829 async->inode = inode;
832 async->mirror_num = mirror_num;
833 async->submit_bio_start = submit_bio_start;
834 async->submit_bio_done = submit_bio_done;
836 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
837 run_one_async_done, run_one_async_free);
839 async->bio_flags = bio_flags;
840 async->bio_offset = bio_offset;
844 atomic_inc(&fs_info->nr_async_submits);
847 btrfs_set_work_high_priority(&async->work);
849 btrfs_queue_work(fs_info->workers, &async->work);
851 while (atomic_read(&fs_info->async_submit_draining) &&
852 atomic_read(&fs_info->nr_async_submits)) {
853 wait_event(fs_info->async_submit_wait,
854 (atomic_read(&fs_info->nr_async_submits) == 0));
860 static int btree_csum_one_bio(struct bio *bio)
862 struct bio_vec *bvec;
863 struct btrfs_root *root;
866 bio_for_each_segment_all(bvec, bio, i) {
867 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
868 ret = csum_dirty_buffer(root, bvec->bv_page);
876 static int __btree_submit_bio_start(struct inode *inode, int rw,
877 struct bio *bio, int mirror_num,
878 unsigned long bio_flags,
882 * when we're called for a write, we're already in the async
883 * submission context. Just jump into btrfs_map_bio
885 return btree_csum_one_bio(bio);
888 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
889 int mirror_num, unsigned long bio_flags,
895 * when we're called for a write, we're already in the async
896 * submission context. Just jump into btrfs_map_bio
898 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
904 static int check_async_write(struct inode *inode, unsigned long bio_flags)
906 if (bio_flags & EXTENT_BIO_TREE_LOG)
915 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
916 int mirror_num, unsigned long bio_flags,
919 int async = check_async_write(inode, bio_flags);
922 if (!(rw & REQ_WRITE)) {
924 * called for a read, do the setup so that checksum validation
925 * can happen in the async kernel threads
927 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
931 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
934 ret = btree_csum_one_bio(bio);
937 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
941 * kthread helpers are used to submit writes so that
942 * checksumming can happen in parallel across all CPUs
944 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
945 inode, rw, bio, mirror_num, 0,
947 __btree_submit_bio_start,
948 __btree_submit_bio_done);
958 #ifdef CONFIG_MIGRATION
959 static int btree_migratepage(struct address_space *mapping,
960 struct page *newpage, struct page *page,
961 enum migrate_mode mode)
964 * we can't safely write a btree page from here,
965 * we haven't done the locking hook
970 * Buffers may be managed in a filesystem specific way.
971 * We must have no buffers or drop them.
973 if (page_has_private(page) &&
974 !try_to_release_page(page, GFP_KERNEL))
976 return migrate_page(mapping, newpage, page, mode);
981 static int btree_writepages(struct address_space *mapping,
982 struct writeback_control *wbc)
984 struct btrfs_fs_info *fs_info;
987 if (wbc->sync_mode == WB_SYNC_NONE) {
989 if (wbc->for_kupdate)
992 fs_info = BTRFS_I(mapping->host)->root->fs_info;
993 /* this is a bit racy, but that's ok */
994 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
995 BTRFS_DIRTY_METADATA_THRESH);
999 return btree_write_cache_pages(mapping, wbc);
1002 static int btree_readpage(struct file *file, struct page *page)
1004 struct extent_io_tree *tree;
1005 tree = &BTRFS_I(page->mapping->host)->io_tree;
1006 return extent_read_full_page(tree, page, btree_get_extent, 0);
1009 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1011 if (PageWriteback(page) || PageDirty(page))
1014 return try_release_extent_buffer(page);
1017 static void btree_invalidatepage(struct page *page, unsigned int offset,
1018 unsigned int length)
1020 struct extent_io_tree *tree;
1021 tree = &BTRFS_I(page->mapping->host)->io_tree;
1022 extent_invalidatepage(tree, page, offset);
1023 btree_releasepage(page, GFP_NOFS);
1024 if (PagePrivate(page)) {
1025 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1026 "page private not zero on page %llu",
1027 (unsigned long long)page_offset(page));
1028 ClearPagePrivate(page);
1029 set_page_private(page, 0);
1030 page_cache_release(page);
1034 static int btree_set_page_dirty(struct page *page)
1037 struct extent_buffer *eb;
1039 BUG_ON(!PagePrivate(page));
1040 eb = (struct extent_buffer *)page->private;
1042 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1043 BUG_ON(!atomic_read(&eb->refs));
1044 btrfs_assert_tree_locked(eb);
1046 return __set_page_dirty_nobuffers(page);
1049 static const struct address_space_operations btree_aops = {
1050 .readpage = btree_readpage,
1051 .writepages = btree_writepages,
1052 .releasepage = btree_releasepage,
1053 .invalidatepage = btree_invalidatepage,
1054 #ifdef CONFIG_MIGRATION
1055 .migratepage = btree_migratepage,
1057 .set_page_dirty = btree_set_page_dirty,
1060 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1063 struct extent_buffer *buf = NULL;
1064 struct inode *btree_inode = root->fs_info->btree_inode;
1067 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1070 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1071 buf, 0, WAIT_NONE, btree_get_extent, 0);
1072 free_extent_buffer(buf);
1076 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1077 int mirror_num, struct extent_buffer **eb)
1079 struct extent_buffer *buf = NULL;
1080 struct inode *btree_inode = root->fs_info->btree_inode;
1081 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1084 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1088 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1090 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1091 btree_get_extent, mirror_num);
1093 free_extent_buffer(buf);
1097 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1098 free_extent_buffer(buf);
1100 } else if (extent_buffer_uptodate(buf)) {
1103 free_extent_buffer(buf);
1108 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1109 u64 bytenr, u32 blocksize)
1111 return find_extent_buffer(root->fs_info, bytenr);
1114 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1115 u64 bytenr, u32 blocksize)
1117 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1118 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state)))
1119 return alloc_test_extent_buffer(root->fs_info, bytenr,
1122 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1126 int btrfs_write_tree_block(struct extent_buffer *buf)
1128 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1129 buf->start + buf->len - 1);
1132 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1134 return filemap_fdatawait_range(buf->pages[0]->mapping,
1135 buf->start, buf->start + buf->len - 1);
1138 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1139 u32 blocksize, u64 parent_transid)
1141 struct extent_buffer *buf = NULL;
1144 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1148 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1150 free_extent_buffer(buf);
1157 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1158 struct extent_buffer *buf)
1160 struct btrfs_fs_info *fs_info = root->fs_info;
1162 if (btrfs_header_generation(buf) ==
1163 fs_info->running_transaction->transid) {
1164 btrfs_assert_tree_locked(buf);
1166 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1167 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1169 fs_info->dirty_metadata_batch);
1170 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1171 btrfs_set_lock_blocking(buf);
1172 clear_extent_buffer_dirty(buf);
1177 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1179 struct btrfs_subvolume_writers *writers;
1182 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1184 return ERR_PTR(-ENOMEM);
1186 ret = percpu_counter_init(&writers->counter, 0);
1189 return ERR_PTR(ret);
1192 init_waitqueue_head(&writers->wait);
1197 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1199 percpu_counter_destroy(&writers->counter);
1203 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1204 u32 stripesize, struct btrfs_root *root,
1205 struct btrfs_fs_info *fs_info,
1209 root->commit_root = NULL;
1210 root->sectorsize = sectorsize;
1211 root->nodesize = nodesize;
1212 root->leafsize = leafsize;
1213 root->stripesize = stripesize;
1215 root->orphan_cleanup_state = 0;
1217 root->objectid = objectid;
1218 root->last_trans = 0;
1219 root->highest_objectid = 0;
1220 root->nr_delalloc_inodes = 0;
1221 root->nr_ordered_extents = 0;
1223 root->inode_tree = RB_ROOT;
1224 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1225 root->block_rsv = NULL;
1226 root->orphan_block_rsv = NULL;
1228 INIT_LIST_HEAD(&root->dirty_list);
1229 INIT_LIST_HEAD(&root->root_list);
1230 INIT_LIST_HEAD(&root->delalloc_inodes);
1231 INIT_LIST_HEAD(&root->delalloc_root);
1232 INIT_LIST_HEAD(&root->ordered_extents);
1233 INIT_LIST_HEAD(&root->ordered_root);
1234 INIT_LIST_HEAD(&root->logged_list[0]);
1235 INIT_LIST_HEAD(&root->logged_list[1]);
1236 spin_lock_init(&root->orphan_lock);
1237 spin_lock_init(&root->inode_lock);
1238 spin_lock_init(&root->delalloc_lock);
1239 spin_lock_init(&root->ordered_extent_lock);
1240 spin_lock_init(&root->accounting_lock);
1241 spin_lock_init(&root->log_extents_lock[0]);
1242 spin_lock_init(&root->log_extents_lock[1]);
1243 mutex_init(&root->objectid_mutex);
1244 mutex_init(&root->log_mutex);
1245 mutex_init(&root->ordered_extent_mutex);
1246 mutex_init(&root->delalloc_mutex);
1247 init_waitqueue_head(&root->log_writer_wait);
1248 init_waitqueue_head(&root->log_commit_wait[0]);
1249 init_waitqueue_head(&root->log_commit_wait[1]);
1250 INIT_LIST_HEAD(&root->log_ctxs[0]);
1251 INIT_LIST_HEAD(&root->log_ctxs[1]);
1252 atomic_set(&root->log_commit[0], 0);
1253 atomic_set(&root->log_commit[1], 0);
1254 atomic_set(&root->log_writers, 0);
1255 atomic_set(&root->log_batch, 0);
1256 atomic_set(&root->orphan_inodes, 0);
1257 atomic_set(&root->refs, 1);
1258 atomic_set(&root->will_be_snapshoted, 0);
1259 root->log_transid = 0;
1260 root->log_transid_committed = -1;
1261 root->last_log_commit = 0;
1263 extent_io_tree_init(&root->dirty_log_pages,
1264 fs_info->btree_inode->i_mapping);
1266 memset(&root->root_key, 0, sizeof(root->root_key));
1267 memset(&root->root_item, 0, sizeof(root->root_item));
1268 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1269 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1271 root->defrag_trans_start = fs_info->generation;
1273 root->defrag_trans_start = 0;
1274 init_completion(&root->kobj_unregister);
1275 root->root_key.objectid = objectid;
1278 spin_lock_init(&root->root_item_lock);
1281 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1283 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1285 root->fs_info = fs_info;
1289 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1290 /* Should only be used by the testing infrastructure */
1291 struct btrfs_root *btrfs_alloc_dummy_root(void)
1293 struct btrfs_root *root;
1295 root = btrfs_alloc_root(NULL);
1297 return ERR_PTR(-ENOMEM);
1298 __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1299 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1300 root->alloc_bytenr = 0;
1306 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1307 struct btrfs_fs_info *fs_info,
1310 struct extent_buffer *leaf;
1311 struct btrfs_root *tree_root = fs_info->tree_root;
1312 struct btrfs_root *root;
1313 struct btrfs_key key;
1317 root = btrfs_alloc_root(fs_info);
1319 return ERR_PTR(-ENOMEM);
1321 __setup_root(tree_root->nodesize, tree_root->leafsize,
1322 tree_root->sectorsize, tree_root->stripesize,
1323 root, fs_info, objectid);
1324 root->root_key.objectid = objectid;
1325 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1326 root->root_key.offset = 0;
1328 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1329 0, objectid, NULL, 0, 0, 0);
1331 ret = PTR_ERR(leaf);
1336 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1337 btrfs_set_header_bytenr(leaf, leaf->start);
1338 btrfs_set_header_generation(leaf, trans->transid);
1339 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1340 btrfs_set_header_owner(leaf, objectid);
1343 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1345 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1346 btrfs_header_chunk_tree_uuid(leaf),
1348 btrfs_mark_buffer_dirty(leaf);
1350 root->commit_root = btrfs_root_node(root);
1351 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1353 root->root_item.flags = 0;
1354 root->root_item.byte_limit = 0;
1355 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1356 btrfs_set_root_generation(&root->root_item, trans->transid);
1357 btrfs_set_root_level(&root->root_item, 0);
1358 btrfs_set_root_refs(&root->root_item, 1);
1359 btrfs_set_root_used(&root->root_item, leaf->len);
1360 btrfs_set_root_last_snapshot(&root->root_item, 0);
1361 btrfs_set_root_dirid(&root->root_item, 0);
1363 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1364 root->root_item.drop_level = 0;
1366 key.objectid = objectid;
1367 key.type = BTRFS_ROOT_ITEM_KEY;
1369 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1373 btrfs_tree_unlock(leaf);
1379 btrfs_tree_unlock(leaf);
1380 free_extent_buffer(root->commit_root);
1381 free_extent_buffer(leaf);
1385 return ERR_PTR(ret);
1388 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1389 struct btrfs_fs_info *fs_info)
1391 struct btrfs_root *root;
1392 struct btrfs_root *tree_root = fs_info->tree_root;
1393 struct extent_buffer *leaf;
1395 root = btrfs_alloc_root(fs_info);
1397 return ERR_PTR(-ENOMEM);
1399 __setup_root(tree_root->nodesize, tree_root->leafsize,
1400 tree_root->sectorsize, tree_root->stripesize,
1401 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1403 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1404 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1405 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1408 * DON'T set REF_COWS for log trees
1410 * log trees do not get reference counted because they go away
1411 * before a real commit is actually done. They do store pointers
1412 * to file data extents, and those reference counts still get
1413 * updated (along with back refs to the log tree).
1416 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1417 BTRFS_TREE_LOG_OBJECTID, NULL,
1421 return ERR_CAST(leaf);
1424 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1425 btrfs_set_header_bytenr(leaf, leaf->start);
1426 btrfs_set_header_generation(leaf, trans->transid);
1427 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1428 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1431 write_extent_buffer(root->node, root->fs_info->fsid,
1432 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1433 btrfs_mark_buffer_dirty(root->node);
1434 btrfs_tree_unlock(root->node);
1438 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1439 struct btrfs_fs_info *fs_info)
1441 struct btrfs_root *log_root;
1443 log_root = alloc_log_tree(trans, fs_info);
1444 if (IS_ERR(log_root))
1445 return PTR_ERR(log_root);
1446 WARN_ON(fs_info->log_root_tree);
1447 fs_info->log_root_tree = log_root;
1451 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1452 struct btrfs_root *root)
1454 struct btrfs_root *log_root;
1455 struct btrfs_inode_item *inode_item;
1457 log_root = alloc_log_tree(trans, root->fs_info);
1458 if (IS_ERR(log_root))
1459 return PTR_ERR(log_root);
1461 log_root->last_trans = trans->transid;
1462 log_root->root_key.offset = root->root_key.objectid;
1464 inode_item = &log_root->root_item.inode;
1465 btrfs_set_stack_inode_generation(inode_item, 1);
1466 btrfs_set_stack_inode_size(inode_item, 3);
1467 btrfs_set_stack_inode_nlink(inode_item, 1);
1468 btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1469 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1471 btrfs_set_root_node(&log_root->root_item, log_root->node);
1473 WARN_ON(root->log_root);
1474 root->log_root = log_root;
1475 root->log_transid = 0;
1476 root->log_transid_committed = -1;
1477 root->last_log_commit = 0;
1481 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1482 struct btrfs_key *key)
1484 struct btrfs_root *root;
1485 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1486 struct btrfs_path *path;
1491 path = btrfs_alloc_path();
1493 return ERR_PTR(-ENOMEM);
1495 root = btrfs_alloc_root(fs_info);
1501 __setup_root(tree_root->nodesize, tree_root->leafsize,
1502 tree_root->sectorsize, tree_root->stripesize,
1503 root, fs_info, key->objectid);
1505 ret = btrfs_find_root(tree_root, key, path,
1506 &root->root_item, &root->root_key);
1513 generation = btrfs_root_generation(&root->root_item);
1514 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1515 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1516 blocksize, generation);
1520 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1524 root->commit_root = btrfs_root_node(root);
1526 btrfs_free_path(path);
1530 free_extent_buffer(root->node);
1534 root = ERR_PTR(ret);
1538 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1539 struct btrfs_key *location)
1541 struct btrfs_root *root;
1543 root = btrfs_read_tree_root(tree_root, location);
1547 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1548 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1549 btrfs_check_and_init_root_item(&root->root_item);
1555 int btrfs_init_fs_root(struct btrfs_root *root)
1558 struct btrfs_subvolume_writers *writers;
1560 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1561 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1563 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1568 writers = btrfs_alloc_subvolume_writers();
1569 if (IS_ERR(writers)) {
1570 ret = PTR_ERR(writers);
1573 root->subv_writers = writers;
1575 btrfs_init_free_ino_ctl(root);
1576 spin_lock_init(&root->cache_lock);
1577 init_waitqueue_head(&root->cache_wait);
1579 ret = get_anon_bdev(&root->anon_dev);
1585 btrfs_free_subvolume_writers(root->subv_writers);
1587 kfree(root->free_ino_ctl);
1588 kfree(root->free_ino_pinned);
1592 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1595 struct btrfs_root *root;
1597 spin_lock(&fs_info->fs_roots_radix_lock);
1598 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1599 (unsigned long)root_id);
1600 spin_unlock(&fs_info->fs_roots_radix_lock);
1604 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1605 struct btrfs_root *root)
1609 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1613 spin_lock(&fs_info->fs_roots_radix_lock);
1614 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1615 (unsigned long)root->root_key.objectid,
1618 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1619 spin_unlock(&fs_info->fs_roots_radix_lock);
1620 radix_tree_preload_end();
1625 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1626 struct btrfs_key *location,
1629 struct btrfs_root *root;
1632 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1633 return fs_info->tree_root;
1634 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1635 return fs_info->extent_root;
1636 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1637 return fs_info->chunk_root;
1638 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1639 return fs_info->dev_root;
1640 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1641 return fs_info->csum_root;
1642 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1643 return fs_info->quota_root ? fs_info->quota_root :
1645 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1646 return fs_info->uuid_root ? fs_info->uuid_root :
1649 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1651 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1652 return ERR_PTR(-ENOENT);
1656 root = btrfs_read_fs_root(fs_info->tree_root, location);
1660 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1665 ret = btrfs_init_fs_root(root);
1669 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1670 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1674 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1676 ret = btrfs_insert_fs_root(fs_info, root);
1678 if (ret == -EEXIST) {
1687 return ERR_PTR(ret);
1690 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1692 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1694 struct btrfs_device *device;
1695 struct backing_dev_info *bdi;
1698 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1701 bdi = blk_get_backing_dev_info(device->bdev);
1702 if (bdi && bdi_congested(bdi, bdi_bits)) {
1712 * If this fails, caller must call bdi_destroy() to get rid of the
1715 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1719 bdi->capabilities = BDI_CAP_MAP_COPY;
1720 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1724 bdi->ra_pages = default_backing_dev_info.ra_pages;
1725 bdi->congested_fn = btrfs_congested_fn;
1726 bdi->congested_data = info;
1731 * called by the kthread helper functions to finally call the bio end_io
1732 * functions. This is where read checksum verification actually happens
1734 static void end_workqueue_fn(struct btrfs_work *work)
1737 struct end_io_wq *end_io_wq;
1740 end_io_wq = container_of(work, struct end_io_wq, work);
1741 bio = end_io_wq->bio;
1743 error = end_io_wq->error;
1744 bio->bi_private = end_io_wq->private;
1745 bio->bi_end_io = end_io_wq->end_io;
1747 bio_endio_nodec(bio, error);
1750 static int cleaner_kthread(void *arg)
1752 struct btrfs_root *root = arg;
1758 /* Make the cleaner go to sleep early. */
1759 if (btrfs_need_cleaner_sleep(root))
1762 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1766 * Avoid the problem that we change the status of the fs
1767 * during the above check and trylock.
1769 if (btrfs_need_cleaner_sleep(root)) {
1770 mutex_unlock(&root->fs_info->cleaner_mutex);
1774 btrfs_run_delayed_iputs(root);
1775 again = btrfs_clean_one_deleted_snapshot(root);
1776 mutex_unlock(&root->fs_info->cleaner_mutex);
1779 * The defragger has dealt with the R/O remount and umount,
1780 * needn't do anything special here.
1782 btrfs_run_defrag_inodes(root->fs_info);
1784 if (!try_to_freeze() && !again) {
1785 set_current_state(TASK_INTERRUPTIBLE);
1786 if (!kthread_should_stop())
1788 __set_current_state(TASK_RUNNING);
1790 } while (!kthread_should_stop());
1794 static int transaction_kthread(void *arg)
1796 struct btrfs_root *root = arg;
1797 struct btrfs_trans_handle *trans;
1798 struct btrfs_transaction *cur;
1801 unsigned long delay;
1805 cannot_commit = false;
1806 delay = HZ * root->fs_info->commit_interval;
1807 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1809 spin_lock(&root->fs_info->trans_lock);
1810 cur = root->fs_info->running_transaction;
1812 spin_unlock(&root->fs_info->trans_lock);
1816 now = get_seconds();
1817 if (cur->state < TRANS_STATE_BLOCKED &&
1818 (now < cur->start_time ||
1819 now - cur->start_time < root->fs_info->commit_interval)) {
1820 spin_unlock(&root->fs_info->trans_lock);
1824 transid = cur->transid;
1825 spin_unlock(&root->fs_info->trans_lock);
1827 /* If the file system is aborted, this will always fail. */
1828 trans = btrfs_attach_transaction(root);
1829 if (IS_ERR(trans)) {
1830 if (PTR_ERR(trans) != -ENOENT)
1831 cannot_commit = true;
1834 if (transid == trans->transid) {
1835 btrfs_commit_transaction(trans, root);
1837 btrfs_end_transaction(trans, root);
1840 wake_up_process(root->fs_info->cleaner_kthread);
1841 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1843 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1844 &root->fs_info->fs_state)))
1845 btrfs_cleanup_transaction(root);
1846 if (!try_to_freeze()) {
1847 set_current_state(TASK_INTERRUPTIBLE);
1848 if (!kthread_should_stop() &&
1849 (!btrfs_transaction_blocked(root->fs_info) ||
1851 schedule_timeout(delay);
1852 __set_current_state(TASK_RUNNING);
1854 } while (!kthread_should_stop());
1859 * this will find the highest generation in the array of
1860 * root backups. The index of the highest array is returned,
1861 * or -1 if we can't find anything.
1863 * We check to make sure the array is valid by comparing the
1864 * generation of the latest root in the array with the generation
1865 * in the super block. If they don't match we pitch it.
1867 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1870 int newest_index = -1;
1871 struct btrfs_root_backup *root_backup;
1874 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1875 root_backup = info->super_copy->super_roots + i;
1876 cur = btrfs_backup_tree_root_gen(root_backup);
1877 if (cur == newest_gen)
1881 /* check to see if we actually wrapped around */
1882 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1883 root_backup = info->super_copy->super_roots;
1884 cur = btrfs_backup_tree_root_gen(root_backup);
1885 if (cur == newest_gen)
1888 return newest_index;
1893 * find the oldest backup so we know where to store new entries
1894 * in the backup array. This will set the backup_root_index
1895 * field in the fs_info struct
1897 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1900 int newest_index = -1;
1902 newest_index = find_newest_super_backup(info, newest_gen);
1903 /* if there was garbage in there, just move along */
1904 if (newest_index == -1) {
1905 info->backup_root_index = 0;
1907 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1912 * copy all the root pointers into the super backup array.
1913 * this will bump the backup pointer by one when it is
1916 static void backup_super_roots(struct btrfs_fs_info *info)
1919 struct btrfs_root_backup *root_backup;
1922 next_backup = info->backup_root_index;
1923 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1924 BTRFS_NUM_BACKUP_ROOTS;
1927 * just overwrite the last backup if we're at the same generation
1928 * this happens only at umount
1930 root_backup = info->super_for_commit->super_roots + last_backup;
1931 if (btrfs_backup_tree_root_gen(root_backup) ==
1932 btrfs_header_generation(info->tree_root->node))
1933 next_backup = last_backup;
1935 root_backup = info->super_for_commit->super_roots + next_backup;
1938 * make sure all of our padding and empty slots get zero filled
1939 * regardless of which ones we use today
1941 memset(root_backup, 0, sizeof(*root_backup));
1943 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1945 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1946 btrfs_set_backup_tree_root_gen(root_backup,
1947 btrfs_header_generation(info->tree_root->node));
1949 btrfs_set_backup_tree_root_level(root_backup,
1950 btrfs_header_level(info->tree_root->node));
1952 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1953 btrfs_set_backup_chunk_root_gen(root_backup,
1954 btrfs_header_generation(info->chunk_root->node));
1955 btrfs_set_backup_chunk_root_level(root_backup,
1956 btrfs_header_level(info->chunk_root->node));
1958 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1959 btrfs_set_backup_extent_root_gen(root_backup,
1960 btrfs_header_generation(info->extent_root->node));
1961 btrfs_set_backup_extent_root_level(root_backup,
1962 btrfs_header_level(info->extent_root->node));
1965 * we might commit during log recovery, which happens before we set
1966 * the fs_root. Make sure it is valid before we fill it in.
1968 if (info->fs_root && info->fs_root->node) {
1969 btrfs_set_backup_fs_root(root_backup,
1970 info->fs_root->node->start);
1971 btrfs_set_backup_fs_root_gen(root_backup,
1972 btrfs_header_generation(info->fs_root->node));
1973 btrfs_set_backup_fs_root_level(root_backup,
1974 btrfs_header_level(info->fs_root->node));
1977 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1978 btrfs_set_backup_dev_root_gen(root_backup,
1979 btrfs_header_generation(info->dev_root->node));
1980 btrfs_set_backup_dev_root_level(root_backup,
1981 btrfs_header_level(info->dev_root->node));
1983 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1984 btrfs_set_backup_csum_root_gen(root_backup,
1985 btrfs_header_generation(info->csum_root->node));
1986 btrfs_set_backup_csum_root_level(root_backup,
1987 btrfs_header_level(info->csum_root->node));
1989 btrfs_set_backup_total_bytes(root_backup,
1990 btrfs_super_total_bytes(info->super_copy));
1991 btrfs_set_backup_bytes_used(root_backup,
1992 btrfs_super_bytes_used(info->super_copy));
1993 btrfs_set_backup_num_devices(root_backup,
1994 btrfs_super_num_devices(info->super_copy));
1997 * if we don't copy this out to the super_copy, it won't get remembered
1998 * for the next commit
2000 memcpy(&info->super_copy->super_roots,
2001 &info->super_for_commit->super_roots,
2002 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2006 * this copies info out of the root backup array and back into
2007 * the in-memory super block. It is meant to help iterate through
2008 * the array, so you send it the number of backups you've already
2009 * tried and the last backup index you used.
2011 * this returns -1 when it has tried all the backups
2013 static noinline int next_root_backup(struct btrfs_fs_info *info,
2014 struct btrfs_super_block *super,
2015 int *num_backups_tried, int *backup_index)
2017 struct btrfs_root_backup *root_backup;
2018 int newest = *backup_index;
2020 if (*num_backups_tried == 0) {
2021 u64 gen = btrfs_super_generation(super);
2023 newest = find_newest_super_backup(info, gen);
2027 *backup_index = newest;
2028 *num_backups_tried = 1;
2029 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2030 /* we've tried all the backups, all done */
2033 /* jump to the next oldest backup */
2034 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2035 BTRFS_NUM_BACKUP_ROOTS;
2036 *backup_index = newest;
2037 *num_backups_tried += 1;
2039 root_backup = super->super_roots + newest;
2041 btrfs_set_super_generation(super,
2042 btrfs_backup_tree_root_gen(root_backup));
2043 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2044 btrfs_set_super_root_level(super,
2045 btrfs_backup_tree_root_level(root_backup));
2046 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2049 * fixme: the total bytes and num_devices need to match or we should
2052 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2053 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2057 /* helper to cleanup workers */
2058 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2060 btrfs_destroy_workqueue(fs_info->fixup_workers);
2061 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2062 btrfs_destroy_workqueue(fs_info->workers);
2063 btrfs_destroy_workqueue(fs_info->endio_workers);
2064 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2065 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2066 btrfs_destroy_workqueue(fs_info->rmw_workers);
2067 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2068 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2069 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2070 btrfs_destroy_workqueue(fs_info->submit_workers);
2071 btrfs_destroy_workqueue(fs_info->delayed_workers);
2072 btrfs_destroy_workqueue(fs_info->caching_workers);
2073 btrfs_destroy_workqueue(fs_info->readahead_workers);
2074 btrfs_destroy_workqueue(fs_info->flush_workers);
2075 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2076 btrfs_destroy_workqueue(fs_info->extent_workers);
2079 static void free_root_extent_buffers(struct btrfs_root *root)
2082 free_extent_buffer(root->node);
2083 free_extent_buffer(root->commit_root);
2085 root->commit_root = NULL;
2089 /* helper to cleanup tree roots */
2090 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2092 free_root_extent_buffers(info->tree_root);
2094 free_root_extent_buffers(info->dev_root);
2095 free_root_extent_buffers(info->extent_root);
2096 free_root_extent_buffers(info->csum_root);
2097 free_root_extent_buffers(info->quota_root);
2098 free_root_extent_buffers(info->uuid_root);
2100 free_root_extent_buffers(info->chunk_root);
2103 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2106 struct btrfs_root *gang[8];
2109 while (!list_empty(&fs_info->dead_roots)) {
2110 gang[0] = list_entry(fs_info->dead_roots.next,
2111 struct btrfs_root, root_list);
2112 list_del(&gang[0]->root_list);
2114 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2115 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2117 free_extent_buffer(gang[0]->node);
2118 free_extent_buffer(gang[0]->commit_root);
2119 btrfs_put_fs_root(gang[0]);
2124 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2129 for (i = 0; i < ret; i++)
2130 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2133 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2134 btrfs_free_log_root_tree(NULL, fs_info);
2135 btrfs_destroy_pinned_extent(fs_info->tree_root,
2136 fs_info->pinned_extents);
2140 int open_ctree(struct super_block *sb,
2141 struct btrfs_fs_devices *fs_devices,
2151 struct btrfs_key location;
2152 struct buffer_head *bh;
2153 struct btrfs_super_block *disk_super;
2154 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2155 struct btrfs_root *tree_root;
2156 struct btrfs_root *extent_root;
2157 struct btrfs_root *csum_root;
2158 struct btrfs_root *chunk_root;
2159 struct btrfs_root *dev_root;
2160 struct btrfs_root *quota_root;
2161 struct btrfs_root *uuid_root;
2162 struct btrfs_root *log_tree_root;
2165 int num_backups_tried = 0;
2166 int backup_index = 0;
2168 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2169 bool create_uuid_tree;
2170 bool check_uuid_tree;
2172 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2173 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2174 if (!tree_root || !chunk_root) {
2179 ret = init_srcu_struct(&fs_info->subvol_srcu);
2185 ret = setup_bdi(fs_info, &fs_info->bdi);
2191 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2196 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2197 (1 + ilog2(nr_cpu_ids));
2199 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2202 goto fail_dirty_metadata_bytes;
2205 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2208 goto fail_delalloc_bytes;
2211 fs_info->btree_inode = new_inode(sb);
2212 if (!fs_info->btree_inode) {
2214 goto fail_bio_counter;
2217 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2219 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2220 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2221 INIT_LIST_HEAD(&fs_info->trans_list);
2222 INIT_LIST_HEAD(&fs_info->dead_roots);
2223 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2224 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2225 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2226 spin_lock_init(&fs_info->delalloc_root_lock);
2227 spin_lock_init(&fs_info->trans_lock);
2228 spin_lock_init(&fs_info->fs_roots_radix_lock);
2229 spin_lock_init(&fs_info->delayed_iput_lock);
2230 spin_lock_init(&fs_info->defrag_inodes_lock);
2231 spin_lock_init(&fs_info->free_chunk_lock);
2232 spin_lock_init(&fs_info->tree_mod_seq_lock);
2233 spin_lock_init(&fs_info->super_lock);
2234 spin_lock_init(&fs_info->qgroup_op_lock);
2235 spin_lock_init(&fs_info->buffer_lock);
2236 rwlock_init(&fs_info->tree_mod_log_lock);
2237 mutex_init(&fs_info->reloc_mutex);
2238 mutex_init(&fs_info->delalloc_root_mutex);
2239 seqlock_init(&fs_info->profiles_lock);
2241 init_completion(&fs_info->kobj_unregister);
2242 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2243 INIT_LIST_HEAD(&fs_info->space_info);
2244 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2245 btrfs_mapping_init(&fs_info->mapping_tree);
2246 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2247 BTRFS_BLOCK_RSV_GLOBAL);
2248 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2249 BTRFS_BLOCK_RSV_DELALLOC);
2250 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2251 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2252 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2253 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2254 BTRFS_BLOCK_RSV_DELOPS);
2255 atomic_set(&fs_info->nr_async_submits, 0);
2256 atomic_set(&fs_info->async_delalloc_pages, 0);
2257 atomic_set(&fs_info->async_submit_draining, 0);
2258 atomic_set(&fs_info->nr_async_bios, 0);
2259 atomic_set(&fs_info->defrag_running, 0);
2260 atomic_set(&fs_info->qgroup_op_seq, 0);
2261 atomic64_set(&fs_info->tree_mod_seq, 0);
2263 fs_info->max_inline = 8192 * 1024;
2264 fs_info->metadata_ratio = 0;
2265 fs_info->defrag_inodes = RB_ROOT;
2266 fs_info->free_chunk_space = 0;
2267 fs_info->tree_mod_log = RB_ROOT;
2268 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2269 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2270 /* readahead state */
2271 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2272 spin_lock_init(&fs_info->reada_lock);
2274 fs_info->thread_pool_size = min_t(unsigned long,
2275 num_online_cpus() + 2, 8);
2277 INIT_LIST_HEAD(&fs_info->ordered_roots);
2278 spin_lock_init(&fs_info->ordered_root_lock);
2279 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2281 if (!fs_info->delayed_root) {
2285 btrfs_init_delayed_root(fs_info->delayed_root);
2287 mutex_init(&fs_info->scrub_lock);
2288 atomic_set(&fs_info->scrubs_running, 0);
2289 atomic_set(&fs_info->scrub_pause_req, 0);
2290 atomic_set(&fs_info->scrubs_paused, 0);
2291 atomic_set(&fs_info->scrub_cancel_req, 0);
2292 init_waitqueue_head(&fs_info->replace_wait);
2293 init_waitqueue_head(&fs_info->scrub_pause_wait);
2294 fs_info->scrub_workers_refcnt = 0;
2295 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2296 fs_info->check_integrity_print_mask = 0;
2299 spin_lock_init(&fs_info->balance_lock);
2300 mutex_init(&fs_info->balance_mutex);
2301 atomic_set(&fs_info->balance_running, 0);
2302 atomic_set(&fs_info->balance_pause_req, 0);
2303 atomic_set(&fs_info->balance_cancel_req, 0);
2304 fs_info->balance_ctl = NULL;
2305 init_waitqueue_head(&fs_info->balance_wait_q);
2306 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2308 sb->s_blocksize = 4096;
2309 sb->s_blocksize_bits = blksize_bits(4096);
2310 sb->s_bdi = &fs_info->bdi;
2312 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2313 set_nlink(fs_info->btree_inode, 1);
2315 * we set the i_size on the btree inode to the max possible int.
2316 * the real end of the address space is determined by all of
2317 * the devices in the system
2319 fs_info->btree_inode->i_size = OFFSET_MAX;
2320 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2321 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2323 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2324 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2325 fs_info->btree_inode->i_mapping);
2326 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2327 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2329 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2331 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2332 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2333 sizeof(struct btrfs_key));
2334 set_bit(BTRFS_INODE_DUMMY,
2335 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2336 btrfs_insert_inode_hash(fs_info->btree_inode);
2338 spin_lock_init(&fs_info->block_group_cache_lock);
2339 fs_info->block_group_cache_tree = RB_ROOT;
2340 fs_info->first_logical_byte = (u64)-1;
2342 extent_io_tree_init(&fs_info->freed_extents[0],
2343 fs_info->btree_inode->i_mapping);
2344 extent_io_tree_init(&fs_info->freed_extents[1],
2345 fs_info->btree_inode->i_mapping);
2346 fs_info->pinned_extents = &fs_info->freed_extents[0];
2347 fs_info->do_barriers = 1;
2350 mutex_init(&fs_info->ordered_operations_mutex);
2351 mutex_init(&fs_info->ordered_extent_flush_mutex);
2352 mutex_init(&fs_info->tree_log_mutex);
2353 mutex_init(&fs_info->chunk_mutex);
2354 mutex_init(&fs_info->transaction_kthread_mutex);
2355 mutex_init(&fs_info->cleaner_mutex);
2356 mutex_init(&fs_info->volume_mutex);
2357 init_rwsem(&fs_info->commit_root_sem);
2358 init_rwsem(&fs_info->cleanup_work_sem);
2359 init_rwsem(&fs_info->subvol_sem);
2360 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2361 fs_info->dev_replace.lock_owner = 0;
2362 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2363 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2364 mutex_init(&fs_info->dev_replace.lock_management_lock);
2365 mutex_init(&fs_info->dev_replace.lock);
2367 spin_lock_init(&fs_info->qgroup_lock);
2368 mutex_init(&fs_info->qgroup_ioctl_lock);
2369 fs_info->qgroup_tree = RB_ROOT;
2370 fs_info->qgroup_op_tree = RB_ROOT;
2371 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2372 fs_info->qgroup_seq = 1;
2373 fs_info->quota_enabled = 0;
2374 fs_info->pending_quota_state = 0;
2375 fs_info->qgroup_ulist = NULL;
2376 mutex_init(&fs_info->qgroup_rescan_lock);
2378 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2379 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2381 init_waitqueue_head(&fs_info->transaction_throttle);
2382 init_waitqueue_head(&fs_info->transaction_wait);
2383 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2384 init_waitqueue_head(&fs_info->async_submit_wait);
2386 ret = btrfs_alloc_stripe_hash_table(fs_info);
2392 __setup_root(4096, 4096, 4096, 4096, tree_root,
2393 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2395 invalidate_bdev(fs_devices->latest_bdev);
2398 * Read super block and check the signature bytes only
2400 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2407 * We want to check superblock checksum, the type is stored inside.
2408 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2410 if (btrfs_check_super_csum(bh->b_data)) {
2411 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2417 * super_copy is zeroed at allocation time and we never touch the
2418 * following bytes up to INFO_SIZE, the checksum is calculated from
2419 * the whole block of INFO_SIZE
2421 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2422 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2423 sizeof(*fs_info->super_for_commit));
2426 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2428 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2430 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2435 disk_super = fs_info->super_copy;
2436 if (!btrfs_super_root(disk_super))
2439 /* check FS state, whether FS is broken. */
2440 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2441 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2444 * run through our array of backup supers and setup
2445 * our ring pointer to the oldest one
2447 generation = btrfs_super_generation(disk_super);
2448 find_oldest_super_backup(fs_info, generation);
2451 * In the long term, we'll store the compression type in the super
2452 * block, and it'll be used for per file compression control.
2454 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2456 ret = btrfs_parse_options(tree_root, options);
2462 features = btrfs_super_incompat_flags(disk_super) &
2463 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2465 printk(KERN_ERR "BTRFS: couldn't mount because of "
2466 "unsupported optional features (%Lx).\n",
2472 if (btrfs_super_leafsize(disk_super) !=
2473 btrfs_super_nodesize(disk_super)) {
2474 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2475 "blocksizes don't match. node %d leaf %d\n",
2476 btrfs_super_nodesize(disk_super),
2477 btrfs_super_leafsize(disk_super));
2481 if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2482 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2483 "blocksize (%d) was too large\n",
2484 btrfs_super_leafsize(disk_super));
2489 features = btrfs_super_incompat_flags(disk_super);
2490 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2491 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2492 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2494 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2495 printk(KERN_ERR "BTRFS: has skinny extents\n");
2498 * flag our filesystem as having big metadata blocks if
2499 * they are bigger than the page size
2501 if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2502 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2503 printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2504 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2507 nodesize = btrfs_super_nodesize(disk_super);
2508 leafsize = btrfs_super_leafsize(disk_super);
2509 sectorsize = btrfs_super_sectorsize(disk_super);
2510 stripesize = btrfs_super_stripesize(disk_super);
2511 fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2512 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2515 * mixed block groups end up with duplicate but slightly offset
2516 * extent buffers for the same range. It leads to corruptions
2518 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2519 (sectorsize != leafsize)) {
2520 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2521 "are not allowed for mixed block groups on %s\n",
2527 * Needn't use the lock because there is no other task which will
2530 btrfs_set_super_incompat_flags(disk_super, features);
2532 features = btrfs_super_compat_ro_flags(disk_super) &
2533 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2534 if (!(sb->s_flags & MS_RDONLY) && features) {
2535 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2536 "unsupported option features (%Lx).\n",
2542 max_active = fs_info->thread_pool_size;
2545 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2548 fs_info->delalloc_workers =
2549 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2551 fs_info->flush_workers =
2552 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2554 fs_info->caching_workers =
2555 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2558 * a higher idle thresh on the submit workers makes it much more
2559 * likely that bios will be send down in a sane order to the
2562 fs_info->submit_workers =
2563 btrfs_alloc_workqueue("submit", flags,
2564 min_t(u64, fs_devices->num_devices,
2567 fs_info->fixup_workers =
2568 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2571 * endios are largely parallel and should have a very
2574 fs_info->endio_workers =
2575 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2576 fs_info->endio_meta_workers =
2577 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2578 fs_info->endio_meta_write_workers =
2579 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2580 fs_info->endio_raid56_workers =
2581 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2582 fs_info->rmw_workers =
2583 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2584 fs_info->endio_write_workers =
2585 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2586 fs_info->endio_freespace_worker =
2587 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2588 fs_info->delayed_workers =
2589 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2590 fs_info->readahead_workers =
2591 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2592 fs_info->qgroup_rescan_workers =
2593 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2594 fs_info->extent_workers =
2595 btrfs_alloc_workqueue("extent-refs", flags,
2596 min_t(u64, fs_devices->num_devices,
2599 if (!(fs_info->workers && fs_info->delalloc_workers &&
2600 fs_info->submit_workers && fs_info->flush_workers &&
2601 fs_info->endio_workers && fs_info->endio_meta_workers &&
2602 fs_info->endio_meta_write_workers &&
2603 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2604 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2605 fs_info->caching_workers && fs_info->readahead_workers &&
2606 fs_info->fixup_workers && fs_info->delayed_workers &&
2607 fs_info->fixup_workers && fs_info->extent_workers &&
2608 fs_info->qgroup_rescan_workers)) {
2610 goto fail_sb_buffer;
2613 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2614 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2615 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2617 tree_root->nodesize = nodesize;
2618 tree_root->leafsize = leafsize;
2619 tree_root->sectorsize = sectorsize;
2620 tree_root->stripesize = stripesize;
2622 sb->s_blocksize = sectorsize;
2623 sb->s_blocksize_bits = blksize_bits(sectorsize);
2625 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2626 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2627 goto fail_sb_buffer;
2630 if (sectorsize != PAGE_SIZE) {
2631 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2632 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2633 goto fail_sb_buffer;
2636 mutex_lock(&fs_info->chunk_mutex);
2637 ret = btrfs_read_sys_array(tree_root);
2638 mutex_unlock(&fs_info->chunk_mutex);
2640 printk(KERN_WARNING "BTRFS: failed to read the system "
2641 "array on %s\n", sb->s_id);
2642 goto fail_sb_buffer;
2645 blocksize = btrfs_level_size(tree_root,
2646 btrfs_super_chunk_root_level(disk_super));
2647 generation = btrfs_super_chunk_root_generation(disk_super);
2649 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2650 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2652 chunk_root->node = read_tree_block(chunk_root,
2653 btrfs_super_chunk_root(disk_super),
2654 blocksize, generation);
2655 if (!chunk_root->node ||
2656 !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2657 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2659 goto fail_tree_roots;
2661 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2662 chunk_root->commit_root = btrfs_root_node(chunk_root);
2664 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2665 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2667 ret = btrfs_read_chunk_tree(chunk_root);
2669 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2671 goto fail_tree_roots;
2675 * keep the device that is marked to be the target device for the
2676 * dev_replace procedure
2678 btrfs_close_extra_devices(fs_info, fs_devices, 0);
2680 if (!fs_devices->latest_bdev) {
2681 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2683 goto fail_tree_roots;
2687 blocksize = btrfs_level_size(tree_root,
2688 btrfs_super_root_level(disk_super));
2689 generation = btrfs_super_generation(disk_super);
2691 tree_root->node = read_tree_block(tree_root,
2692 btrfs_super_root(disk_super),
2693 blocksize, generation);
2694 if (!tree_root->node ||
2695 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2696 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2699 goto recovery_tree_root;
2702 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2703 tree_root->commit_root = btrfs_root_node(tree_root);
2704 btrfs_set_root_refs(&tree_root->root_item, 1);
2706 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2707 location.type = BTRFS_ROOT_ITEM_KEY;
2708 location.offset = 0;
2710 extent_root = btrfs_read_tree_root(tree_root, &location);
2711 if (IS_ERR(extent_root)) {
2712 ret = PTR_ERR(extent_root);
2713 goto recovery_tree_root;
2715 set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2716 fs_info->extent_root = extent_root;
2718 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2719 dev_root = btrfs_read_tree_root(tree_root, &location);
2720 if (IS_ERR(dev_root)) {
2721 ret = PTR_ERR(dev_root);
2722 goto recovery_tree_root;
2724 set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2725 fs_info->dev_root = dev_root;
2726 btrfs_init_devices_late(fs_info);
2728 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2729 csum_root = btrfs_read_tree_root(tree_root, &location);
2730 if (IS_ERR(csum_root)) {
2731 ret = PTR_ERR(csum_root);
2732 goto recovery_tree_root;
2734 set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2735 fs_info->csum_root = csum_root;
2737 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2738 quota_root = btrfs_read_tree_root(tree_root, &location);
2739 if (!IS_ERR(quota_root)) {
2740 set_bit(BTRFS_ROOT_TRACK_DIRTY, "a_root->state);
2741 fs_info->quota_enabled = 1;
2742 fs_info->pending_quota_state = 1;
2743 fs_info->quota_root = quota_root;
2746 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2747 uuid_root = btrfs_read_tree_root(tree_root, &location);
2748 if (IS_ERR(uuid_root)) {
2749 ret = PTR_ERR(uuid_root);
2751 goto recovery_tree_root;
2752 create_uuid_tree = true;
2753 check_uuid_tree = false;
2755 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2756 fs_info->uuid_root = uuid_root;
2757 create_uuid_tree = false;
2759 generation != btrfs_super_uuid_tree_generation(disk_super);
2762 fs_info->generation = generation;
2763 fs_info->last_trans_committed = generation;
2765 ret = btrfs_recover_balance(fs_info);
2767 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2768 goto fail_block_groups;
2771 ret = btrfs_init_dev_stats(fs_info);
2773 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2775 goto fail_block_groups;
2778 ret = btrfs_init_dev_replace(fs_info);
2780 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2781 goto fail_block_groups;
2784 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2786 ret = btrfs_sysfs_add_one(fs_info);
2788 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2789 goto fail_block_groups;
2792 ret = btrfs_init_space_info(fs_info);
2794 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2798 ret = btrfs_read_block_groups(extent_root);
2800 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2803 fs_info->num_tolerated_disk_barrier_failures =
2804 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2805 if (fs_info->fs_devices->missing_devices >
2806 fs_info->num_tolerated_disk_barrier_failures &&
2807 !(sb->s_flags & MS_RDONLY)) {
2808 printk(KERN_WARNING "BTRFS: "
2809 "too many missing devices, writeable mount is not allowed\n");
2813 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2815 if (IS_ERR(fs_info->cleaner_kthread))
2818 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2820 "btrfs-transaction");
2821 if (IS_ERR(fs_info->transaction_kthread))
2824 if (!btrfs_test_opt(tree_root, SSD) &&
2825 !btrfs_test_opt(tree_root, NOSSD) &&
2826 !fs_info->fs_devices->rotating) {
2827 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2829 btrfs_set_opt(fs_info->mount_opt, SSD);
2832 /* Set the real inode map cache flag */
2833 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2834 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2836 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2837 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2838 ret = btrfsic_mount(tree_root, fs_devices,
2839 btrfs_test_opt(tree_root,
2840 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2842 fs_info->check_integrity_print_mask);
2844 printk(KERN_WARNING "BTRFS: failed to initialize"
2845 " integrity check module %s\n", sb->s_id);
2848 ret = btrfs_read_qgroup_config(fs_info);
2850 goto fail_trans_kthread;
2852 /* do not make disk changes in broken FS */
2853 if (btrfs_super_log_root(disk_super) != 0) {
2854 u64 bytenr = btrfs_super_log_root(disk_super);
2856 if (fs_devices->rw_devices == 0) {
2857 printk(KERN_WARNING "BTRFS: log replay required "
2863 btrfs_level_size(tree_root,
2864 btrfs_super_log_root_level(disk_super));
2866 log_tree_root = btrfs_alloc_root(fs_info);
2867 if (!log_tree_root) {
2872 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2873 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2875 log_tree_root->node = read_tree_block(tree_root, bytenr,
2878 if (!log_tree_root->node ||
2879 !extent_buffer_uptodate(log_tree_root->node)) {
2880 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2881 free_extent_buffer(log_tree_root->node);
2882 kfree(log_tree_root);
2885 /* returns with log_tree_root freed on success */
2886 ret = btrfs_recover_log_trees(log_tree_root);
2888 btrfs_error(tree_root->fs_info, ret,
2889 "Failed to recover log tree");
2890 free_extent_buffer(log_tree_root->node);
2891 kfree(log_tree_root);
2895 if (sb->s_flags & MS_RDONLY) {
2896 ret = btrfs_commit_super(tree_root);
2902 ret = btrfs_find_orphan_roots(tree_root);
2906 if (!(sb->s_flags & MS_RDONLY)) {
2907 ret = btrfs_cleanup_fs_roots(fs_info);
2911 mutex_lock(&fs_info->cleaner_mutex);
2912 ret = btrfs_recover_relocation(tree_root);
2913 mutex_unlock(&fs_info->cleaner_mutex);
2916 "BTRFS: failed to recover relocation\n");
2922 location.objectid = BTRFS_FS_TREE_OBJECTID;
2923 location.type = BTRFS_ROOT_ITEM_KEY;
2924 location.offset = 0;
2926 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2927 if (IS_ERR(fs_info->fs_root)) {
2928 err = PTR_ERR(fs_info->fs_root);
2932 if (sb->s_flags & MS_RDONLY)
2935 down_read(&fs_info->cleanup_work_sem);
2936 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2937 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2938 up_read(&fs_info->cleanup_work_sem);
2939 close_ctree(tree_root);
2942 up_read(&fs_info->cleanup_work_sem);
2944 ret = btrfs_resume_balance_async(fs_info);
2946 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2947 close_ctree(tree_root);
2951 ret = btrfs_resume_dev_replace_async(fs_info);
2953 pr_warn("BTRFS: failed to resume dev_replace\n");
2954 close_ctree(tree_root);
2958 btrfs_qgroup_rescan_resume(fs_info);
2960 if (create_uuid_tree) {
2961 pr_info("BTRFS: creating UUID tree\n");
2962 ret = btrfs_create_uuid_tree(fs_info);
2964 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2966 close_ctree(tree_root);
2969 } else if (check_uuid_tree ||
2970 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2971 pr_info("BTRFS: checking UUID tree\n");
2972 ret = btrfs_check_uuid_tree(fs_info);
2974 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2976 close_ctree(tree_root);
2980 fs_info->update_uuid_tree_gen = 1;
2986 btrfs_free_qgroup_config(fs_info);
2988 kthread_stop(fs_info->transaction_kthread);
2989 btrfs_cleanup_transaction(fs_info->tree_root);
2990 btrfs_free_fs_roots(fs_info);
2992 kthread_stop(fs_info->cleaner_kthread);
2995 * make sure we're done with the btree inode before we stop our
2998 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3001 btrfs_sysfs_remove_one(fs_info);
3004 btrfs_put_block_group_cache(fs_info);
3005 btrfs_free_block_groups(fs_info);
3008 free_root_pointers(fs_info, 1);
3009 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3012 btrfs_stop_all_workers(fs_info);
3015 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3017 iput(fs_info->btree_inode);
3019 percpu_counter_destroy(&fs_info->bio_counter);
3020 fail_delalloc_bytes:
3021 percpu_counter_destroy(&fs_info->delalloc_bytes);
3022 fail_dirty_metadata_bytes:
3023 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3025 bdi_destroy(&fs_info->bdi);
3027 cleanup_srcu_struct(&fs_info->subvol_srcu);
3029 btrfs_free_stripe_hash_table(fs_info);
3030 btrfs_close_devices(fs_info->fs_devices);
3034 if (!btrfs_test_opt(tree_root, RECOVERY))
3035 goto fail_tree_roots;
3037 free_root_pointers(fs_info, 0);
3039 /* don't use the log in recovery mode, it won't be valid */
3040 btrfs_set_super_log_root(disk_super, 0);
3042 /* we can't trust the free space cache either */
3043 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3045 ret = next_root_backup(fs_info, fs_info->super_copy,
3046 &num_backups_tried, &backup_index);
3048 goto fail_block_groups;
3049 goto retry_root_backup;
3052 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3055 set_buffer_uptodate(bh);
3057 struct btrfs_device *device = (struct btrfs_device *)
3060 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3061 "I/O error on %s\n",
3062 rcu_str_deref(device->name));
3063 /* note, we dont' set_buffer_write_io_error because we have
3064 * our own ways of dealing with the IO errors
3066 clear_buffer_uptodate(bh);
3067 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3073 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3075 struct buffer_head *bh;
3076 struct buffer_head *latest = NULL;
3077 struct btrfs_super_block *super;
3082 /* we would like to check all the supers, but that would make
3083 * a btrfs mount succeed after a mkfs from a different FS.
3084 * So, we need to add a special mount option to scan for
3085 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3087 for (i = 0; i < 1; i++) {
3088 bytenr = btrfs_sb_offset(i);
3089 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3090 i_size_read(bdev->bd_inode))
3092 bh = __bread(bdev, bytenr / 4096,
3093 BTRFS_SUPER_INFO_SIZE);
3097 super = (struct btrfs_super_block *)bh->b_data;
3098 if (btrfs_super_bytenr(super) != bytenr ||
3099 btrfs_super_magic(super) != BTRFS_MAGIC) {
3104 if (!latest || btrfs_super_generation(super) > transid) {
3107 transid = btrfs_super_generation(super);
3116 * this should be called twice, once with wait == 0 and
3117 * once with wait == 1. When wait == 0 is done, all the buffer heads
3118 * we write are pinned.
3120 * They are released when wait == 1 is done.
3121 * max_mirrors must be the same for both runs, and it indicates how
3122 * many supers on this one device should be written.
3124 * max_mirrors == 0 means to write them all.
3126 static int write_dev_supers(struct btrfs_device *device,
3127 struct btrfs_super_block *sb,
3128 int do_barriers, int wait, int max_mirrors)
3130 struct buffer_head *bh;
3137 if (max_mirrors == 0)
3138 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3140 for (i = 0; i < max_mirrors; i++) {
3141 bytenr = btrfs_sb_offset(i);
3142 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3146 bh = __find_get_block(device->bdev, bytenr / 4096,
3147 BTRFS_SUPER_INFO_SIZE);
3153 if (!buffer_uptodate(bh))
3156 /* drop our reference */
3159 /* drop the reference from the wait == 0 run */
3163 btrfs_set_super_bytenr(sb, bytenr);
3166 crc = btrfs_csum_data((char *)sb +
3167 BTRFS_CSUM_SIZE, crc,
3168 BTRFS_SUPER_INFO_SIZE -
3170 btrfs_csum_final(crc, sb->csum);
3173 * one reference for us, and we leave it for the
3176 bh = __getblk(device->bdev, bytenr / 4096,
3177 BTRFS_SUPER_INFO_SIZE);
3179 printk(KERN_ERR "BTRFS: couldn't get super "
3180 "buffer head for bytenr %Lu\n", bytenr);
3185 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3187 /* one reference for submit_bh */
3190 set_buffer_uptodate(bh);
3192 bh->b_end_io = btrfs_end_buffer_write_sync;
3193 bh->b_private = device;
3197 * we fua the first super. The others we allow
3201 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3203 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3207 return errors < i ? 0 : -1;
3211 * endio for the write_dev_flush, this will wake anyone waiting
3212 * for the barrier when it is done
3214 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3217 if (err == -EOPNOTSUPP)
3218 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3219 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3221 if (bio->bi_private)
3222 complete(bio->bi_private);
3227 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3228 * sent down. With wait == 1, it waits for the previous flush.
3230 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3233 static int write_dev_flush(struct btrfs_device *device, int wait)
3238 if (device->nobarriers)
3242 bio = device->flush_bio;
3246 wait_for_completion(&device->flush_wait);
3248 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3249 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3250 rcu_str_deref(device->name));
3251 device->nobarriers = 1;
3252 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3254 btrfs_dev_stat_inc_and_print(device,
3255 BTRFS_DEV_STAT_FLUSH_ERRS);
3258 /* drop the reference from the wait == 0 run */
3260 device->flush_bio = NULL;
3266 * one reference for us, and we leave it for the
3269 device->flush_bio = NULL;
3270 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3274 bio->bi_end_io = btrfs_end_empty_barrier;
3275 bio->bi_bdev = device->bdev;
3276 init_completion(&device->flush_wait);
3277 bio->bi_private = &device->flush_wait;
3278 device->flush_bio = bio;
3281 btrfsic_submit_bio(WRITE_FLUSH, bio);
3287 * send an empty flush down to each device in parallel,
3288 * then wait for them
3290 static int barrier_all_devices(struct btrfs_fs_info *info)
3292 struct list_head *head;
3293 struct btrfs_device *dev;
3294 int errors_send = 0;
3295 int errors_wait = 0;
3298 /* send down all the barriers */
3299 head = &info->fs_devices->devices;
3300 list_for_each_entry_rcu(dev, head, dev_list) {
3307 if (!dev->in_fs_metadata || !dev->writeable)
3310 ret = write_dev_flush(dev, 0);
3315 /* wait for all the barriers */
3316 list_for_each_entry_rcu(dev, head, dev_list) {
3323 if (!dev->in_fs_metadata || !dev->writeable)
3326 ret = write_dev_flush(dev, 1);
3330 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3331 errors_wait > info->num_tolerated_disk_barrier_failures)
3336 int btrfs_calc_num_tolerated_disk_barrier_failures(
3337 struct btrfs_fs_info *fs_info)
3339 struct btrfs_ioctl_space_info space;
3340 struct btrfs_space_info *sinfo;
3341 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3342 BTRFS_BLOCK_GROUP_SYSTEM,
3343 BTRFS_BLOCK_GROUP_METADATA,
3344 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3348 int num_tolerated_disk_barrier_failures =
3349 (int)fs_info->fs_devices->num_devices;
3351 for (i = 0; i < num_types; i++) {
3352 struct btrfs_space_info *tmp;
3356 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3357 if (tmp->flags == types[i]) {
3367 down_read(&sinfo->groups_sem);
3368 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3369 if (!list_empty(&sinfo->block_groups[c])) {
3372 btrfs_get_block_group_info(
3373 &sinfo->block_groups[c], &space);
3374 if (space.total_bytes == 0 ||
3375 space.used_bytes == 0)
3377 flags = space.flags;
3380 * 0: if dup, single or RAID0 is configured for
3381 * any of metadata, system or data, else
3382 * 1: if RAID5 is configured, or if RAID1 or
3383 * RAID10 is configured and only two mirrors
3385 * 2: if RAID6 is configured, else
3386 * num_mirrors - 1: if RAID1 or RAID10 is
3387 * configured and more than
3388 * 2 mirrors are used.
3390 if (num_tolerated_disk_barrier_failures > 0 &&
3391 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3392 BTRFS_BLOCK_GROUP_RAID0)) ||
3393 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3395 num_tolerated_disk_barrier_failures = 0;
3396 else if (num_tolerated_disk_barrier_failures > 1) {
3397 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3398 BTRFS_BLOCK_GROUP_RAID5 |
3399 BTRFS_BLOCK_GROUP_RAID10)) {
3400 num_tolerated_disk_barrier_failures = 1;
3402 BTRFS_BLOCK_GROUP_RAID6) {
3403 num_tolerated_disk_barrier_failures = 2;
3408 up_read(&sinfo->groups_sem);
3411 return num_tolerated_disk_barrier_failures;
3414 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3416 struct list_head *head;
3417 struct btrfs_device *dev;
3418 struct btrfs_super_block *sb;
3419 struct btrfs_dev_item *dev_item;
3423 int total_errors = 0;
3426 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3427 backup_super_roots(root->fs_info);
3429 sb = root->fs_info->super_for_commit;
3430 dev_item = &sb->dev_item;
3432 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3433 head = &root->fs_info->fs_devices->devices;
3434 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3437 ret = barrier_all_devices(root->fs_info);
3440 &root->fs_info->fs_devices->device_list_mutex);
3441 btrfs_error(root->fs_info, ret,
3442 "errors while submitting device barriers.");
3447 list_for_each_entry_rcu(dev, head, dev_list) {
3452 if (!dev->in_fs_metadata || !dev->writeable)
3455 btrfs_set_stack_device_generation(dev_item, 0);
3456 btrfs_set_stack_device_type(dev_item, dev->type);
3457 btrfs_set_stack_device_id(dev_item, dev->devid);
3458 btrfs_set_stack_device_total_bytes(dev_item,
3459 dev->disk_total_bytes);
3460 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3461 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3462 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3463 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3464 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3465 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3467 flags = btrfs_super_flags(sb);
3468 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3470 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3474 if (total_errors > max_errors) {
3475 btrfs_err(root->fs_info, "%d errors while writing supers",
3477 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3479 /* FUA is masked off if unsupported and can't be the reason */
3480 btrfs_error(root->fs_info, -EIO,
3481 "%d errors while writing supers", total_errors);
3486 list_for_each_entry_rcu(dev, head, dev_list) {
3489 if (!dev->in_fs_metadata || !dev->writeable)
3492 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3496 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3497 if (total_errors > max_errors) {
3498 btrfs_error(root->fs_info, -EIO,
3499 "%d errors while writing supers", total_errors);
3505 int write_ctree_super(struct btrfs_trans_handle *trans,
3506 struct btrfs_root *root, int max_mirrors)
3508 return write_all_supers(root, max_mirrors);
3511 /* Drop a fs root from the radix tree and free it. */
3512 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3513 struct btrfs_root *root)
3515 spin_lock(&fs_info->fs_roots_radix_lock);
3516 radix_tree_delete(&fs_info->fs_roots_radix,
3517 (unsigned long)root->root_key.objectid);
3518 spin_unlock(&fs_info->fs_roots_radix_lock);
3520 if (btrfs_root_refs(&root->root_item) == 0)
3521 synchronize_srcu(&fs_info->subvol_srcu);
3523 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3524 btrfs_free_log(NULL, root);
3526 if (root->free_ino_pinned)
3527 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3528 if (root->free_ino_ctl)
3529 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3533 static void free_fs_root(struct btrfs_root *root)
3535 iput(root->cache_inode);
3536 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3537 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3538 root->orphan_block_rsv = NULL;
3540 free_anon_bdev(root->anon_dev);
3541 if (root->subv_writers)
3542 btrfs_free_subvolume_writers(root->subv_writers);
3543 free_extent_buffer(root->node);
3544 free_extent_buffer(root->commit_root);
3545 kfree(root->free_ino_ctl);
3546 kfree(root->free_ino_pinned);
3548 btrfs_put_fs_root(root);
3551 void btrfs_free_fs_root(struct btrfs_root *root)
3556 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3558 u64 root_objectid = 0;
3559 struct btrfs_root *gang[8];
3562 unsigned int ret = 0;
3566 index = srcu_read_lock(&fs_info->subvol_srcu);
3567 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3568 (void **)gang, root_objectid,
3571 srcu_read_unlock(&fs_info->subvol_srcu, index);
3574 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3576 for (i = 0; i < ret; i++) {
3577 /* Avoid to grab roots in dead_roots */
3578 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3582 /* grab all the search result for later use */
3583 gang[i] = btrfs_grab_fs_root(gang[i]);
3585 srcu_read_unlock(&fs_info->subvol_srcu, index);
3587 for (i = 0; i < ret; i++) {
3590 root_objectid = gang[i]->root_key.objectid;
3591 err = btrfs_orphan_cleanup(gang[i]);
3594 btrfs_put_fs_root(gang[i]);
3599 /* release the uncleaned roots due to error */
3600 for (; i < ret; i++) {
3602 btrfs_put_fs_root(gang[i]);
3607 int btrfs_commit_super(struct btrfs_root *root)
3609 struct btrfs_trans_handle *trans;
3611 mutex_lock(&root->fs_info->cleaner_mutex);
3612 btrfs_run_delayed_iputs(root);
3613 mutex_unlock(&root->fs_info->cleaner_mutex);
3614 wake_up_process(root->fs_info->cleaner_kthread);
3616 /* wait until ongoing cleanup work done */
3617 down_write(&root->fs_info->cleanup_work_sem);
3618 up_write(&root->fs_info->cleanup_work_sem);
3620 trans = btrfs_join_transaction(root);
3622 return PTR_ERR(trans);
3623 return btrfs_commit_transaction(trans, root);
3626 int close_ctree(struct btrfs_root *root)
3628 struct btrfs_fs_info *fs_info = root->fs_info;
3631 fs_info->closing = 1;
3634 /* wait for the uuid_scan task to finish */
3635 down(&fs_info->uuid_tree_rescan_sem);
3636 /* avoid complains from lockdep et al., set sem back to initial state */
3637 up(&fs_info->uuid_tree_rescan_sem);
3639 /* pause restriper - we want to resume on mount */
3640 btrfs_pause_balance(fs_info);
3642 btrfs_dev_replace_suspend_for_unmount(fs_info);
3644 btrfs_scrub_cancel(fs_info);
3646 /* wait for any defraggers to finish */
3647 wait_event(fs_info->transaction_wait,
3648 (atomic_read(&fs_info->defrag_running) == 0));
3650 /* clear out the rbtree of defraggable inodes */
3651 btrfs_cleanup_defrag_inodes(fs_info);
3653 cancel_work_sync(&fs_info->async_reclaim_work);
3655 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3656 ret = btrfs_commit_super(root);
3658 btrfs_err(root->fs_info, "commit super ret %d", ret);
3661 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3662 btrfs_error_commit_super(root);
3664 kthread_stop(fs_info->transaction_kthread);
3665 kthread_stop(fs_info->cleaner_kthread);
3667 fs_info->closing = 2;
3670 btrfs_free_qgroup_config(root->fs_info);
3672 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3673 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3674 percpu_counter_sum(&fs_info->delalloc_bytes));
3677 btrfs_sysfs_remove_one(fs_info);
3679 btrfs_free_fs_roots(fs_info);
3681 btrfs_put_block_group_cache(fs_info);
3683 btrfs_free_block_groups(fs_info);
3686 * we must make sure there is not any read request to
3687 * submit after we stopping all workers.
3689 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3690 btrfs_stop_all_workers(fs_info);
3692 free_root_pointers(fs_info, 1);
3694 iput(fs_info->btree_inode);
3696 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3697 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3698 btrfsic_unmount(root, fs_info->fs_devices);
3701 btrfs_close_devices(fs_info->fs_devices);
3702 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3704 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3705 percpu_counter_destroy(&fs_info->delalloc_bytes);
3706 percpu_counter_destroy(&fs_info->bio_counter);
3707 bdi_destroy(&fs_info->bdi);
3708 cleanup_srcu_struct(&fs_info->subvol_srcu);
3710 btrfs_free_stripe_hash_table(fs_info);
3712 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3713 root->orphan_block_rsv = NULL;
3718 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3722 struct inode *btree_inode = buf->pages[0]->mapping->host;
3724 ret = extent_buffer_uptodate(buf);
3728 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3729 parent_transid, atomic);
3735 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3737 return set_extent_buffer_uptodate(buf);
3740 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3742 struct btrfs_root *root;
3743 u64 transid = btrfs_header_generation(buf);
3746 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3748 * This is a fast path so only do this check if we have sanity tests
3749 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3750 * outside of the sanity tests.
3752 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3755 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3756 btrfs_assert_tree_locked(buf);
3757 if (transid != root->fs_info->generation)
3758 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3759 "found %llu running %llu\n",
3760 buf->start, transid, root->fs_info->generation);
3761 was_dirty = set_extent_buffer_dirty(buf);
3763 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3765 root->fs_info->dirty_metadata_batch);
3766 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3767 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3768 btrfs_print_leaf(root, buf);
3774 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3778 * looks as though older kernels can get into trouble with
3779 * this code, they end up stuck in balance_dirty_pages forever
3783 if (current->flags & PF_MEMALLOC)
3787 btrfs_balance_delayed_items(root);
3789 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3790 BTRFS_DIRTY_METADATA_THRESH);
3792 balance_dirty_pages_ratelimited(
3793 root->fs_info->btree_inode->i_mapping);
3798 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3800 __btrfs_btree_balance_dirty(root, 1);
3803 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3805 __btrfs_btree_balance_dirty(root, 0);
3808 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3810 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3811 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3814 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3818 * Placeholder for checks
3823 static void btrfs_error_commit_super(struct btrfs_root *root)
3825 mutex_lock(&root->fs_info->cleaner_mutex);
3826 btrfs_run_delayed_iputs(root);
3827 mutex_unlock(&root->fs_info->cleaner_mutex);
3829 down_write(&root->fs_info->cleanup_work_sem);
3830 up_write(&root->fs_info->cleanup_work_sem);
3832 /* cleanup FS via transaction */
3833 btrfs_cleanup_transaction(root);
3836 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3838 struct btrfs_ordered_extent *ordered;
3840 spin_lock(&root->ordered_extent_lock);
3842 * This will just short circuit the ordered completion stuff which will
3843 * make sure the ordered extent gets properly cleaned up.
3845 list_for_each_entry(ordered, &root->ordered_extents,
3847 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3848 spin_unlock(&root->ordered_extent_lock);
3851 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3853 struct btrfs_root *root;
3854 struct list_head splice;
3856 INIT_LIST_HEAD(&splice);
3858 spin_lock(&fs_info->ordered_root_lock);
3859 list_splice_init(&fs_info->ordered_roots, &splice);
3860 while (!list_empty(&splice)) {
3861 root = list_first_entry(&splice, struct btrfs_root,
3863 list_move_tail(&root->ordered_root,
3864 &fs_info->ordered_roots);
3866 spin_unlock(&fs_info->ordered_root_lock);
3867 btrfs_destroy_ordered_extents(root);
3870 spin_lock(&fs_info->ordered_root_lock);
3872 spin_unlock(&fs_info->ordered_root_lock);
3875 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3876 struct btrfs_root *root)
3878 struct rb_node *node;
3879 struct btrfs_delayed_ref_root *delayed_refs;
3880 struct btrfs_delayed_ref_node *ref;
3883 delayed_refs = &trans->delayed_refs;
3885 spin_lock(&delayed_refs->lock);
3886 if (atomic_read(&delayed_refs->num_entries) == 0) {
3887 spin_unlock(&delayed_refs->lock);
3888 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3892 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3893 struct btrfs_delayed_ref_head *head;
3894 bool pin_bytes = false;
3896 head = rb_entry(node, struct btrfs_delayed_ref_head,
3898 if (!mutex_trylock(&head->mutex)) {
3899 atomic_inc(&head->node.refs);
3900 spin_unlock(&delayed_refs->lock);
3902 mutex_lock(&head->mutex);
3903 mutex_unlock(&head->mutex);
3904 btrfs_put_delayed_ref(&head->node);
3905 spin_lock(&delayed_refs->lock);
3908 spin_lock(&head->lock);
3909 while ((node = rb_first(&head->ref_root)) != NULL) {
3910 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3913 rb_erase(&ref->rb_node, &head->ref_root);
3914 atomic_dec(&delayed_refs->num_entries);
3915 btrfs_put_delayed_ref(ref);
3917 if (head->must_insert_reserved)
3919 btrfs_free_delayed_extent_op(head->extent_op);
3920 delayed_refs->num_heads--;
3921 if (head->processing == 0)
3922 delayed_refs->num_heads_ready--;
3923 atomic_dec(&delayed_refs->num_entries);
3924 head->node.in_tree = 0;
3925 rb_erase(&head->href_node, &delayed_refs->href_root);
3926 spin_unlock(&head->lock);
3927 spin_unlock(&delayed_refs->lock);
3928 mutex_unlock(&head->mutex);
3931 btrfs_pin_extent(root, head->node.bytenr,
3932 head->node.num_bytes, 1);
3933 btrfs_put_delayed_ref(&head->node);
3935 spin_lock(&delayed_refs->lock);
3938 spin_unlock(&delayed_refs->lock);
3943 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3945 struct btrfs_inode *btrfs_inode;
3946 struct list_head splice;
3948 INIT_LIST_HEAD(&splice);
3950 spin_lock(&root->delalloc_lock);
3951 list_splice_init(&root->delalloc_inodes, &splice);
3953 while (!list_empty(&splice)) {
3954 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3957 list_del_init(&btrfs_inode->delalloc_inodes);
3958 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3959 &btrfs_inode->runtime_flags);
3960 spin_unlock(&root->delalloc_lock);
3962 btrfs_invalidate_inodes(btrfs_inode->root);
3964 spin_lock(&root->delalloc_lock);
3967 spin_unlock(&root->delalloc_lock);
3970 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3972 struct btrfs_root *root;
3973 struct list_head splice;
3975 INIT_LIST_HEAD(&splice);
3977 spin_lock(&fs_info->delalloc_root_lock);
3978 list_splice_init(&fs_info->delalloc_roots, &splice);
3979 while (!list_empty(&splice)) {
3980 root = list_first_entry(&splice, struct btrfs_root,
3982 list_del_init(&root->delalloc_root);
3983 root = btrfs_grab_fs_root(root);
3985 spin_unlock(&fs_info->delalloc_root_lock);
3987 btrfs_destroy_delalloc_inodes(root);
3988 btrfs_put_fs_root(root);
3990 spin_lock(&fs_info->delalloc_root_lock);
3992 spin_unlock(&fs_info->delalloc_root_lock);
3995 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3996 struct extent_io_tree *dirty_pages,
4000 struct extent_buffer *eb;
4005 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4010 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4011 while (start <= end) {
4012 eb = btrfs_find_tree_block(root, start,
4014 start += root->leafsize;
4017 wait_on_extent_buffer_writeback(eb);
4019 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4021 clear_extent_buffer_dirty(eb);
4022 free_extent_buffer_stale(eb);
4029 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4030 struct extent_io_tree *pinned_extents)
4032 struct extent_io_tree *unpin;
4038 unpin = pinned_extents;
4041 ret = find_first_extent_bit(unpin, 0, &start, &end,
4042 EXTENT_DIRTY, NULL);
4047 if (btrfs_test_opt(root, DISCARD))
4048 ret = btrfs_error_discard_extent(root, start,
4052 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4053 btrfs_error_unpin_extent_range(root, start, end);
4058 if (unpin == &root->fs_info->freed_extents[0])
4059 unpin = &root->fs_info->freed_extents[1];
4061 unpin = &root->fs_info->freed_extents[0];
4069 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4070 struct btrfs_root *root)
4072 btrfs_destroy_delayed_refs(cur_trans, root);
4074 cur_trans->state = TRANS_STATE_COMMIT_START;
4075 wake_up(&root->fs_info->transaction_blocked_wait);
4077 cur_trans->state = TRANS_STATE_UNBLOCKED;
4078 wake_up(&root->fs_info->transaction_wait);
4080 btrfs_destroy_delayed_inodes(root);
4081 btrfs_assert_delayed_root_empty(root);
4083 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4085 btrfs_destroy_pinned_extent(root,
4086 root->fs_info->pinned_extents);
4088 cur_trans->state =TRANS_STATE_COMPLETED;
4089 wake_up(&cur_trans->commit_wait);
4092 memset(cur_trans, 0, sizeof(*cur_trans));
4093 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4097 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4099 struct btrfs_transaction *t;
4101 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4103 spin_lock(&root->fs_info->trans_lock);
4104 while (!list_empty(&root->fs_info->trans_list)) {
4105 t = list_first_entry(&root->fs_info->trans_list,
4106 struct btrfs_transaction, list);
4107 if (t->state >= TRANS_STATE_COMMIT_START) {
4108 atomic_inc(&t->use_count);
4109 spin_unlock(&root->fs_info->trans_lock);
4110 btrfs_wait_for_commit(root, t->transid);
4111 btrfs_put_transaction(t);
4112 spin_lock(&root->fs_info->trans_lock);
4115 if (t == root->fs_info->running_transaction) {
4116 t->state = TRANS_STATE_COMMIT_DOING;
4117 spin_unlock(&root->fs_info->trans_lock);
4119 * We wait for 0 num_writers since we don't hold a trans
4120 * handle open currently for this transaction.
4122 wait_event(t->writer_wait,
4123 atomic_read(&t->num_writers) == 0);
4125 spin_unlock(&root->fs_info->trans_lock);
4127 btrfs_cleanup_one_transaction(t, root);
4129 spin_lock(&root->fs_info->trans_lock);
4130 if (t == root->fs_info->running_transaction)
4131 root->fs_info->running_transaction = NULL;
4132 list_del_init(&t->list);
4133 spin_unlock(&root->fs_info->trans_lock);
4135 btrfs_put_transaction(t);
4136 trace_btrfs_transaction_commit(root);
4137 spin_lock(&root->fs_info->trans_lock);
4139 spin_unlock(&root->fs_info->trans_lock);
4140 btrfs_destroy_all_ordered_extents(root->fs_info);
4141 btrfs_destroy_delayed_inodes(root);
4142 btrfs_assert_delayed_root_empty(root);
4143 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4144 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4145 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4150 static struct extent_io_ops btree_extent_io_ops = {
4151 .readpage_end_io_hook = btree_readpage_end_io_hook,
4152 .readpage_io_failed_hook = btree_io_failed_hook,
4153 .submit_bio_hook = btree_submit_bio_hook,
4154 /* note we're sharing with inode.c for the merge bio hook */
4155 .merge_bio_hook = btrfs_merge_bio_hook,