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(KERN_INFO "BTRFS (device %s): parent transid verify failed on %llu wanted %llu found %llu\n",
352 eb->fs_info->sb->s_id, eb->start,
353 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 (device %s): bad tree block start "
619 eb->fs_info->sb->s_id, found_start, eb->start);
623 if (check_tree_block_fsid(root, eb)) {
624 printk_ratelimited(KERN_INFO "BTRFS (device %s): bad fsid on block %llu\n",
625 eb->fs_info->sb->s_id, eb->start);
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 (unlikely(end_io_wq->metadata ==
717 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
718 wq = fs_info->endio_repair_workers;
719 func = btrfs_endio_repair_helper;
720 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
721 wq = fs_info->endio_raid56_workers;
722 func = btrfs_endio_raid56_helper;
723 } else if (end_io_wq->metadata) {
724 wq = fs_info->endio_meta_workers;
725 func = btrfs_endio_meta_helper;
727 wq = fs_info->endio_workers;
728 func = btrfs_endio_helper;
732 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
733 btrfs_queue_work(wq, &end_io_wq->work);
737 * For the metadata arg you want
740 * 1 - if normal metadta
741 * 2 - if writing to the free space cache area
742 * 3 - raid parity work
744 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
747 struct end_io_wq *end_io_wq;
749 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
753 end_io_wq->private = bio->bi_private;
754 end_io_wq->end_io = bio->bi_end_io;
755 end_io_wq->info = info;
756 end_io_wq->error = 0;
757 end_io_wq->bio = bio;
758 end_io_wq->metadata = metadata;
760 bio->bi_private = end_io_wq;
761 bio->bi_end_io = end_workqueue_bio;
765 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
767 unsigned long limit = min_t(unsigned long,
768 info->thread_pool_size,
769 info->fs_devices->open_devices);
773 static void run_one_async_start(struct btrfs_work *work)
775 struct async_submit_bio *async;
778 async = container_of(work, struct async_submit_bio, work);
779 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
780 async->mirror_num, async->bio_flags,
786 static void run_one_async_done(struct btrfs_work *work)
788 struct btrfs_fs_info *fs_info;
789 struct async_submit_bio *async;
792 async = container_of(work, struct async_submit_bio, work);
793 fs_info = BTRFS_I(async->inode)->root->fs_info;
795 limit = btrfs_async_submit_limit(fs_info);
796 limit = limit * 2 / 3;
798 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
799 waitqueue_active(&fs_info->async_submit_wait))
800 wake_up(&fs_info->async_submit_wait);
802 /* If an error occured we just want to clean up the bio and move on */
804 bio_endio(async->bio, async->error);
808 async->submit_bio_done(async->inode, async->rw, async->bio,
809 async->mirror_num, async->bio_flags,
813 static void run_one_async_free(struct btrfs_work *work)
815 struct async_submit_bio *async;
817 async = container_of(work, struct async_submit_bio, work);
821 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
822 int rw, struct bio *bio, int mirror_num,
823 unsigned long bio_flags,
825 extent_submit_bio_hook_t *submit_bio_start,
826 extent_submit_bio_hook_t *submit_bio_done)
828 struct async_submit_bio *async;
830 async = kmalloc(sizeof(*async), GFP_NOFS);
834 async->inode = inode;
837 async->mirror_num = mirror_num;
838 async->submit_bio_start = submit_bio_start;
839 async->submit_bio_done = submit_bio_done;
841 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
842 run_one_async_done, run_one_async_free);
844 async->bio_flags = bio_flags;
845 async->bio_offset = bio_offset;
849 atomic_inc(&fs_info->nr_async_submits);
852 btrfs_set_work_high_priority(&async->work);
854 btrfs_queue_work(fs_info->workers, &async->work);
856 while (atomic_read(&fs_info->async_submit_draining) &&
857 atomic_read(&fs_info->nr_async_submits)) {
858 wait_event(fs_info->async_submit_wait,
859 (atomic_read(&fs_info->nr_async_submits) == 0));
865 static int btree_csum_one_bio(struct bio *bio)
867 struct bio_vec *bvec;
868 struct btrfs_root *root;
871 bio_for_each_segment_all(bvec, bio, i) {
872 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
873 ret = csum_dirty_buffer(root, bvec->bv_page);
881 static int __btree_submit_bio_start(struct inode *inode, int rw,
882 struct bio *bio, int mirror_num,
883 unsigned long bio_flags,
887 * when we're called for a write, we're already in the async
888 * submission context. Just jump into btrfs_map_bio
890 return btree_csum_one_bio(bio);
893 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
894 int mirror_num, unsigned long bio_flags,
900 * when we're called for a write, we're already in the async
901 * submission context. Just jump into btrfs_map_bio
903 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
909 static int check_async_write(struct inode *inode, unsigned long bio_flags)
911 if (bio_flags & EXTENT_BIO_TREE_LOG)
920 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
921 int mirror_num, unsigned long bio_flags,
924 int async = check_async_write(inode, bio_flags);
927 if (!(rw & REQ_WRITE)) {
929 * called for a read, do the setup so that checksum validation
930 * can happen in the async kernel threads
932 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
936 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
939 ret = btree_csum_one_bio(bio);
942 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
946 * kthread helpers are used to submit writes so that
947 * checksumming can happen in parallel across all CPUs
949 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
950 inode, rw, bio, mirror_num, 0,
952 __btree_submit_bio_start,
953 __btree_submit_bio_done);
963 #ifdef CONFIG_MIGRATION
964 static int btree_migratepage(struct address_space *mapping,
965 struct page *newpage, struct page *page,
966 enum migrate_mode mode)
969 * we can't safely write a btree page from here,
970 * we haven't done the locking hook
975 * Buffers may be managed in a filesystem specific way.
976 * We must have no buffers or drop them.
978 if (page_has_private(page) &&
979 !try_to_release_page(page, GFP_KERNEL))
981 return migrate_page(mapping, newpage, page, mode);
986 static int btree_writepages(struct address_space *mapping,
987 struct writeback_control *wbc)
989 struct btrfs_fs_info *fs_info;
992 if (wbc->sync_mode == WB_SYNC_NONE) {
994 if (wbc->for_kupdate)
997 fs_info = BTRFS_I(mapping->host)->root->fs_info;
998 /* this is a bit racy, but that's ok */
999 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1000 BTRFS_DIRTY_METADATA_THRESH);
1004 return btree_write_cache_pages(mapping, wbc);
1007 static int btree_readpage(struct file *file, struct page *page)
1009 struct extent_io_tree *tree;
1010 tree = &BTRFS_I(page->mapping->host)->io_tree;
1011 return extent_read_full_page(tree, page, btree_get_extent, 0);
1014 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1016 if (PageWriteback(page) || PageDirty(page))
1019 return try_release_extent_buffer(page);
1022 static void btree_invalidatepage(struct page *page, unsigned int offset,
1023 unsigned int length)
1025 struct extent_io_tree *tree;
1026 tree = &BTRFS_I(page->mapping->host)->io_tree;
1027 extent_invalidatepage(tree, page, offset);
1028 btree_releasepage(page, GFP_NOFS);
1029 if (PagePrivate(page)) {
1030 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1031 "page private not zero on page %llu",
1032 (unsigned long long)page_offset(page));
1033 ClearPagePrivate(page);
1034 set_page_private(page, 0);
1035 page_cache_release(page);
1039 static int btree_set_page_dirty(struct page *page)
1042 struct extent_buffer *eb;
1044 BUG_ON(!PagePrivate(page));
1045 eb = (struct extent_buffer *)page->private;
1047 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1048 BUG_ON(!atomic_read(&eb->refs));
1049 btrfs_assert_tree_locked(eb);
1051 return __set_page_dirty_nobuffers(page);
1054 static const struct address_space_operations btree_aops = {
1055 .readpage = btree_readpage,
1056 .writepages = btree_writepages,
1057 .releasepage = btree_releasepage,
1058 .invalidatepage = btree_invalidatepage,
1059 #ifdef CONFIG_MIGRATION
1060 .migratepage = btree_migratepage,
1062 .set_page_dirty = btree_set_page_dirty,
1065 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1068 struct extent_buffer *buf = NULL;
1069 struct inode *btree_inode = root->fs_info->btree_inode;
1072 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1075 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1076 buf, 0, WAIT_NONE, btree_get_extent, 0);
1077 free_extent_buffer(buf);
1081 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1082 int mirror_num, struct extent_buffer **eb)
1084 struct extent_buffer *buf = NULL;
1085 struct inode *btree_inode = root->fs_info->btree_inode;
1086 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1089 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1093 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1095 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1096 btree_get_extent, mirror_num);
1098 free_extent_buffer(buf);
1102 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1103 free_extent_buffer(buf);
1105 } else if (extent_buffer_uptodate(buf)) {
1108 free_extent_buffer(buf);
1113 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1114 u64 bytenr, u32 blocksize)
1116 return find_extent_buffer(root->fs_info, bytenr);
1119 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1120 u64 bytenr, u32 blocksize)
1122 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1123 if (unlikely(test_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state)))
1124 return alloc_test_extent_buffer(root->fs_info, bytenr,
1127 return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1131 int btrfs_write_tree_block(struct extent_buffer *buf)
1133 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1134 buf->start + buf->len - 1);
1137 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1139 return filemap_fdatawait_range(buf->pages[0]->mapping,
1140 buf->start, buf->start + buf->len - 1);
1143 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1144 u32 blocksize, u64 parent_transid)
1146 struct extent_buffer *buf = NULL;
1149 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1153 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1155 free_extent_buffer(buf);
1162 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1163 struct extent_buffer *buf)
1165 struct btrfs_fs_info *fs_info = root->fs_info;
1167 if (btrfs_header_generation(buf) ==
1168 fs_info->running_transaction->transid) {
1169 btrfs_assert_tree_locked(buf);
1171 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1172 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1174 fs_info->dirty_metadata_batch);
1175 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1176 btrfs_set_lock_blocking(buf);
1177 clear_extent_buffer_dirty(buf);
1182 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1184 struct btrfs_subvolume_writers *writers;
1187 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1189 return ERR_PTR(-ENOMEM);
1191 ret = percpu_counter_init(&writers->counter, 0);
1194 return ERR_PTR(ret);
1197 init_waitqueue_head(&writers->wait);
1202 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1204 percpu_counter_destroy(&writers->counter);
1208 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1209 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1213 root->commit_root = NULL;
1214 root->sectorsize = sectorsize;
1215 root->nodesize = nodesize;
1216 root->stripesize = stripesize;
1218 root->orphan_cleanup_state = 0;
1220 root->objectid = objectid;
1221 root->last_trans = 0;
1222 root->highest_objectid = 0;
1223 root->nr_delalloc_inodes = 0;
1224 root->nr_ordered_extents = 0;
1226 root->inode_tree = RB_ROOT;
1227 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1228 root->block_rsv = NULL;
1229 root->orphan_block_rsv = NULL;
1231 INIT_LIST_HEAD(&root->dirty_list);
1232 INIT_LIST_HEAD(&root->root_list);
1233 INIT_LIST_HEAD(&root->delalloc_inodes);
1234 INIT_LIST_HEAD(&root->delalloc_root);
1235 INIT_LIST_HEAD(&root->ordered_extents);
1236 INIT_LIST_HEAD(&root->ordered_root);
1237 INIT_LIST_HEAD(&root->logged_list[0]);
1238 INIT_LIST_HEAD(&root->logged_list[1]);
1239 spin_lock_init(&root->orphan_lock);
1240 spin_lock_init(&root->inode_lock);
1241 spin_lock_init(&root->delalloc_lock);
1242 spin_lock_init(&root->ordered_extent_lock);
1243 spin_lock_init(&root->accounting_lock);
1244 spin_lock_init(&root->log_extents_lock[0]);
1245 spin_lock_init(&root->log_extents_lock[1]);
1246 mutex_init(&root->objectid_mutex);
1247 mutex_init(&root->log_mutex);
1248 mutex_init(&root->ordered_extent_mutex);
1249 mutex_init(&root->delalloc_mutex);
1250 init_waitqueue_head(&root->log_writer_wait);
1251 init_waitqueue_head(&root->log_commit_wait[0]);
1252 init_waitqueue_head(&root->log_commit_wait[1]);
1253 INIT_LIST_HEAD(&root->log_ctxs[0]);
1254 INIT_LIST_HEAD(&root->log_ctxs[1]);
1255 atomic_set(&root->log_commit[0], 0);
1256 atomic_set(&root->log_commit[1], 0);
1257 atomic_set(&root->log_writers, 0);
1258 atomic_set(&root->log_batch, 0);
1259 atomic_set(&root->orphan_inodes, 0);
1260 atomic_set(&root->refs, 1);
1261 atomic_set(&root->will_be_snapshoted, 0);
1262 root->log_transid = 0;
1263 root->log_transid_committed = -1;
1264 root->last_log_commit = 0;
1266 extent_io_tree_init(&root->dirty_log_pages,
1267 fs_info->btree_inode->i_mapping);
1269 memset(&root->root_key, 0, sizeof(root->root_key));
1270 memset(&root->root_item, 0, sizeof(root->root_item));
1271 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1272 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1274 root->defrag_trans_start = fs_info->generation;
1276 root->defrag_trans_start = 0;
1277 init_completion(&root->kobj_unregister);
1278 root->root_key.objectid = objectid;
1281 spin_lock_init(&root->root_item_lock);
1284 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1286 struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1288 root->fs_info = fs_info;
1292 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1293 /* Should only be used by the testing infrastructure */
1294 struct btrfs_root *btrfs_alloc_dummy_root(void)
1296 struct btrfs_root *root;
1298 root = btrfs_alloc_root(NULL);
1300 return ERR_PTR(-ENOMEM);
1301 __setup_root(4096, 4096, 4096, root, NULL, 1);
1302 set_bit(BTRFS_ROOT_DUMMY_ROOT, &root->state);
1303 root->alloc_bytenr = 0;
1309 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1310 struct btrfs_fs_info *fs_info,
1313 struct extent_buffer *leaf;
1314 struct btrfs_root *tree_root = fs_info->tree_root;
1315 struct btrfs_root *root;
1316 struct btrfs_key key;
1320 root = btrfs_alloc_root(fs_info);
1322 return ERR_PTR(-ENOMEM);
1324 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1325 tree_root->stripesize, root, fs_info, objectid);
1326 root->root_key.objectid = objectid;
1327 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1328 root->root_key.offset = 0;
1330 leaf = btrfs_alloc_free_block(trans, root, root->nodesize,
1331 0, objectid, NULL, 0, 0, 0);
1333 ret = PTR_ERR(leaf);
1338 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1339 btrfs_set_header_bytenr(leaf, leaf->start);
1340 btrfs_set_header_generation(leaf, trans->transid);
1341 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1342 btrfs_set_header_owner(leaf, objectid);
1345 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1347 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1348 btrfs_header_chunk_tree_uuid(leaf),
1350 btrfs_mark_buffer_dirty(leaf);
1352 root->commit_root = btrfs_root_node(root);
1353 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1355 root->root_item.flags = 0;
1356 root->root_item.byte_limit = 0;
1357 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1358 btrfs_set_root_generation(&root->root_item, trans->transid);
1359 btrfs_set_root_level(&root->root_item, 0);
1360 btrfs_set_root_refs(&root->root_item, 1);
1361 btrfs_set_root_used(&root->root_item, leaf->len);
1362 btrfs_set_root_last_snapshot(&root->root_item, 0);
1363 btrfs_set_root_dirid(&root->root_item, 0);
1365 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1366 root->root_item.drop_level = 0;
1368 key.objectid = objectid;
1369 key.type = BTRFS_ROOT_ITEM_KEY;
1371 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1375 btrfs_tree_unlock(leaf);
1381 btrfs_tree_unlock(leaf);
1382 free_extent_buffer(root->commit_root);
1383 free_extent_buffer(leaf);
1387 return ERR_PTR(ret);
1390 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1391 struct btrfs_fs_info *fs_info)
1393 struct btrfs_root *root;
1394 struct btrfs_root *tree_root = fs_info->tree_root;
1395 struct extent_buffer *leaf;
1397 root = btrfs_alloc_root(fs_info);
1399 return ERR_PTR(-ENOMEM);
1401 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1402 tree_root->stripesize, root, fs_info,
1403 BTRFS_TREE_LOG_OBJECTID);
1405 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1406 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1407 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1410 * DON'T set REF_COWS for log trees
1412 * log trees do not get reference counted because they go away
1413 * before a real commit is actually done. They do store pointers
1414 * to file data extents, and those reference counts still get
1415 * updated (along with back refs to the log tree).
1418 leaf = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
1419 BTRFS_TREE_LOG_OBJECTID, NULL,
1423 return ERR_CAST(leaf);
1426 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1427 btrfs_set_header_bytenr(leaf, leaf->start);
1428 btrfs_set_header_generation(leaf, trans->transid);
1429 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1430 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1433 write_extent_buffer(root->node, root->fs_info->fsid,
1434 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1435 btrfs_mark_buffer_dirty(root->node);
1436 btrfs_tree_unlock(root->node);
1440 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1441 struct btrfs_fs_info *fs_info)
1443 struct btrfs_root *log_root;
1445 log_root = alloc_log_tree(trans, fs_info);
1446 if (IS_ERR(log_root))
1447 return PTR_ERR(log_root);
1448 WARN_ON(fs_info->log_root_tree);
1449 fs_info->log_root_tree = log_root;
1453 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1454 struct btrfs_root *root)
1456 struct btrfs_root *log_root;
1457 struct btrfs_inode_item *inode_item;
1459 log_root = alloc_log_tree(trans, root->fs_info);
1460 if (IS_ERR(log_root))
1461 return PTR_ERR(log_root);
1463 log_root->last_trans = trans->transid;
1464 log_root->root_key.offset = root->root_key.objectid;
1466 inode_item = &log_root->root_item.inode;
1467 btrfs_set_stack_inode_generation(inode_item, 1);
1468 btrfs_set_stack_inode_size(inode_item, 3);
1469 btrfs_set_stack_inode_nlink(inode_item, 1);
1470 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1471 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1473 btrfs_set_root_node(&log_root->root_item, log_root->node);
1475 WARN_ON(root->log_root);
1476 root->log_root = log_root;
1477 root->log_transid = 0;
1478 root->log_transid_committed = -1;
1479 root->last_log_commit = 0;
1483 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1484 struct btrfs_key *key)
1486 struct btrfs_root *root;
1487 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1488 struct btrfs_path *path;
1493 path = btrfs_alloc_path();
1495 return ERR_PTR(-ENOMEM);
1497 root = btrfs_alloc_root(fs_info);
1503 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1504 tree_root->stripesize, root, fs_info, key->objectid);
1506 ret = btrfs_find_root(tree_root, key, path,
1507 &root->root_item, &root->root_key);
1514 generation = btrfs_root_generation(&root->root_item);
1515 blocksize = root->nodesize;
1516 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1517 blocksize, generation);
1521 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1525 root->commit_root = btrfs_root_node(root);
1527 btrfs_free_path(path);
1531 free_extent_buffer(root->node);
1535 root = ERR_PTR(ret);
1539 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1540 struct btrfs_key *location)
1542 struct btrfs_root *root;
1544 root = btrfs_read_tree_root(tree_root, location);
1548 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1549 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1550 btrfs_check_and_init_root_item(&root->root_item);
1556 int btrfs_init_fs_root(struct btrfs_root *root)
1559 struct btrfs_subvolume_writers *writers;
1561 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1562 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1564 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1569 writers = btrfs_alloc_subvolume_writers();
1570 if (IS_ERR(writers)) {
1571 ret = PTR_ERR(writers);
1574 root->subv_writers = writers;
1576 btrfs_init_free_ino_ctl(root);
1577 spin_lock_init(&root->ino_cache_lock);
1578 init_waitqueue_head(&root->ino_cache_wait);
1580 ret = get_anon_bdev(&root->anon_dev);
1586 btrfs_free_subvolume_writers(root->subv_writers);
1588 kfree(root->free_ino_ctl);
1589 kfree(root->free_ino_pinned);
1593 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1596 struct btrfs_root *root;
1598 spin_lock(&fs_info->fs_roots_radix_lock);
1599 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1600 (unsigned long)root_id);
1601 spin_unlock(&fs_info->fs_roots_radix_lock);
1605 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1606 struct btrfs_root *root)
1610 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1614 spin_lock(&fs_info->fs_roots_radix_lock);
1615 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1616 (unsigned long)root->root_key.objectid,
1619 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1620 spin_unlock(&fs_info->fs_roots_radix_lock);
1621 radix_tree_preload_end();
1626 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1627 struct btrfs_key *location,
1630 struct btrfs_root *root;
1633 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1634 return fs_info->tree_root;
1635 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1636 return fs_info->extent_root;
1637 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1638 return fs_info->chunk_root;
1639 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1640 return fs_info->dev_root;
1641 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1642 return fs_info->csum_root;
1643 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1644 return fs_info->quota_root ? fs_info->quota_root :
1646 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1647 return fs_info->uuid_root ? fs_info->uuid_root :
1650 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1652 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1653 return ERR_PTR(-ENOENT);
1657 root = btrfs_read_fs_root(fs_info->tree_root, location);
1661 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1666 ret = btrfs_init_fs_root(root);
1670 ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1671 location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1675 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1677 ret = btrfs_insert_fs_root(fs_info, root);
1679 if (ret == -EEXIST) {
1688 return ERR_PTR(ret);
1691 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1693 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1695 struct btrfs_device *device;
1696 struct backing_dev_info *bdi;
1699 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1702 bdi = blk_get_backing_dev_info(device->bdev);
1703 if (bdi && bdi_congested(bdi, bdi_bits)) {
1712 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1716 bdi->capabilities = BDI_CAP_MAP_COPY;
1717 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1721 bdi->ra_pages = default_backing_dev_info.ra_pages;
1722 bdi->congested_fn = btrfs_congested_fn;
1723 bdi->congested_data = info;
1728 * called by the kthread helper functions to finally call the bio end_io
1729 * functions. This is where read checksum verification actually happens
1731 static void end_workqueue_fn(struct btrfs_work *work)
1734 struct end_io_wq *end_io_wq;
1737 end_io_wq = container_of(work, struct end_io_wq, work);
1738 bio = end_io_wq->bio;
1740 error = end_io_wq->error;
1741 bio->bi_private = end_io_wq->private;
1742 bio->bi_end_io = end_io_wq->end_io;
1744 bio_endio_nodec(bio, error);
1747 static int cleaner_kthread(void *arg)
1749 struct btrfs_root *root = arg;
1755 /* Make the cleaner go to sleep early. */
1756 if (btrfs_need_cleaner_sleep(root))
1759 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1763 * Avoid the problem that we change the status of the fs
1764 * during the above check and trylock.
1766 if (btrfs_need_cleaner_sleep(root)) {
1767 mutex_unlock(&root->fs_info->cleaner_mutex);
1771 btrfs_run_delayed_iputs(root);
1772 again = btrfs_clean_one_deleted_snapshot(root);
1773 mutex_unlock(&root->fs_info->cleaner_mutex);
1776 * The defragger has dealt with the R/O remount and umount,
1777 * needn't do anything special here.
1779 btrfs_run_defrag_inodes(root->fs_info);
1781 if (!try_to_freeze() && !again) {
1782 set_current_state(TASK_INTERRUPTIBLE);
1783 if (!kthread_should_stop())
1785 __set_current_state(TASK_RUNNING);
1787 } while (!kthread_should_stop());
1791 static int transaction_kthread(void *arg)
1793 struct btrfs_root *root = arg;
1794 struct btrfs_trans_handle *trans;
1795 struct btrfs_transaction *cur;
1798 unsigned long delay;
1802 cannot_commit = false;
1803 delay = HZ * root->fs_info->commit_interval;
1804 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1806 spin_lock(&root->fs_info->trans_lock);
1807 cur = root->fs_info->running_transaction;
1809 spin_unlock(&root->fs_info->trans_lock);
1813 now = get_seconds();
1814 if (cur->state < TRANS_STATE_BLOCKED &&
1815 (now < cur->start_time ||
1816 now - cur->start_time < root->fs_info->commit_interval)) {
1817 spin_unlock(&root->fs_info->trans_lock);
1821 transid = cur->transid;
1822 spin_unlock(&root->fs_info->trans_lock);
1824 /* If the file system is aborted, this will always fail. */
1825 trans = btrfs_attach_transaction(root);
1826 if (IS_ERR(trans)) {
1827 if (PTR_ERR(trans) != -ENOENT)
1828 cannot_commit = true;
1831 if (transid == trans->transid) {
1832 btrfs_commit_transaction(trans, root);
1834 btrfs_end_transaction(trans, root);
1837 wake_up_process(root->fs_info->cleaner_kthread);
1838 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1840 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1841 &root->fs_info->fs_state)))
1842 btrfs_cleanup_transaction(root);
1843 if (!try_to_freeze()) {
1844 set_current_state(TASK_INTERRUPTIBLE);
1845 if (!kthread_should_stop() &&
1846 (!btrfs_transaction_blocked(root->fs_info) ||
1848 schedule_timeout(delay);
1849 __set_current_state(TASK_RUNNING);
1851 } while (!kthread_should_stop());
1856 * this will find the highest generation in the array of
1857 * root backups. The index of the highest array is returned,
1858 * or -1 if we can't find anything.
1860 * We check to make sure the array is valid by comparing the
1861 * generation of the latest root in the array with the generation
1862 * in the super block. If they don't match we pitch it.
1864 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1867 int newest_index = -1;
1868 struct btrfs_root_backup *root_backup;
1871 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1872 root_backup = info->super_copy->super_roots + i;
1873 cur = btrfs_backup_tree_root_gen(root_backup);
1874 if (cur == newest_gen)
1878 /* check to see if we actually wrapped around */
1879 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1880 root_backup = info->super_copy->super_roots;
1881 cur = btrfs_backup_tree_root_gen(root_backup);
1882 if (cur == newest_gen)
1885 return newest_index;
1890 * find the oldest backup so we know where to store new entries
1891 * in the backup array. This will set the backup_root_index
1892 * field in the fs_info struct
1894 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1897 int newest_index = -1;
1899 newest_index = find_newest_super_backup(info, newest_gen);
1900 /* if there was garbage in there, just move along */
1901 if (newest_index == -1) {
1902 info->backup_root_index = 0;
1904 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1909 * copy all the root pointers into the super backup array.
1910 * this will bump the backup pointer by one when it is
1913 static void backup_super_roots(struct btrfs_fs_info *info)
1916 struct btrfs_root_backup *root_backup;
1919 next_backup = info->backup_root_index;
1920 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1921 BTRFS_NUM_BACKUP_ROOTS;
1924 * just overwrite the last backup if we're at the same generation
1925 * this happens only at umount
1927 root_backup = info->super_for_commit->super_roots + last_backup;
1928 if (btrfs_backup_tree_root_gen(root_backup) ==
1929 btrfs_header_generation(info->tree_root->node))
1930 next_backup = last_backup;
1932 root_backup = info->super_for_commit->super_roots + next_backup;
1935 * make sure all of our padding and empty slots get zero filled
1936 * regardless of which ones we use today
1938 memset(root_backup, 0, sizeof(*root_backup));
1940 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1942 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1943 btrfs_set_backup_tree_root_gen(root_backup,
1944 btrfs_header_generation(info->tree_root->node));
1946 btrfs_set_backup_tree_root_level(root_backup,
1947 btrfs_header_level(info->tree_root->node));
1949 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1950 btrfs_set_backup_chunk_root_gen(root_backup,
1951 btrfs_header_generation(info->chunk_root->node));
1952 btrfs_set_backup_chunk_root_level(root_backup,
1953 btrfs_header_level(info->chunk_root->node));
1955 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1956 btrfs_set_backup_extent_root_gen(root_backup,
1957 btrfs_header_generation(info->extent_root->node));
1958 btrfs_set_backup_extent_root_level(root_backup,
1959 btrfs_header_level(info->extent_root->node));
1962 * we might commit during log recovery, which happens before we set
1963 * the fs_root. Make sure it is valid before we fill it in.
1965 if (info->fs_root && info->fs_root->node) {
1966 btrfs_set_backup_fs_root(root_backup,
1967 info->fs_root->node->start);
1968 btrfs_set_backup_fs_root_gen(root_backup,
1969 btrfs_header_generation(info->fs_root->node));
1970 btrfs_set_backup_fs_root_level(root_backup,
1971 btrfs_header_level(info->fs_root->node));
1974 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1975 btrfs_set_backup_dev_root_gen(root_backup,
1976 btrfs_header_generation(info->dev_root->node));
1977 btrfs_set_backup_dev_root_level(root_backup,
1978 btrfs_header_level(info->dev_root->node));
1980 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1981 btrfs_set_backup_csum_root_gen(root_backup,
1982 btrfs_header_generation(info->csum_root->node));
1983 btrfs_set_backup_csum_root_level(root_backup,
1984 btrfs_header_level(info->csum_root->node));
1986 btrfs_set_backup_total_bytes(root_backup,
1987 btrfs_super_total_bytes(info->super_copy));
1988 btrfs_set_backup_bytes_used(root_backup,
1989 btrfs_super_bytes_used(info->super_copy));
1990 btrfs_set_backup_num_devices(root_backup,
1991 btrfs_super_num_devices(info->super_copy));
1994 * if we don't copy this out to the super_copy, it won't get remembered
1995 * for the next commit
1997 memcpy(&info->super_copy->super_roots,
1998 &info->super_for_commit->super_roots,
1999 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2003 * this copies info out of the root backup array and back into
2004 * the in-memory super block. It is meant to help iterate through
2005 * the array, so you send it the number of backups you've already
2006 * tried and the last backup index you used.
2008 * this returns -1 when it has tried all the backups
2010 static noinline int next_root_backup(struct btrfs_fs_info *info,
2011 struct btrfs_super_block *super,
2012 int *num_backups_tried, int *backup_index)
2014 struct btrfs_root_backup *root_backup;
2015 int newest = *backup_index;
2017 if (*num_backups_tried == 0) {
2018 u64 gen = btrfs_super_generation(super);
2020 newest = find_newest_super_backup(info, gen);
2024 *backup_index = newest;
2025 *num_backups_tried = 1;
2026 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2027 /* we've tried all the backups, all done */
2030 /* jump to the next oldest backup */
2031 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2032 BTRFS_NUM_BACKUP_ROOTS;
2033 *backup_index = newest;
2034 *num_backups_tried += 1;
2036 root_backup = super->super_roots + newest;
2038 btrfs_set_super_generation(super,
2039 btrfs_backup_tree_root_gen(root_backup));
2040 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2041 btrfs_set_super_root_level(super,
2042 btrfs_backup_tree_root_level(root_backup));
2043 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2046 * fixme: the total bytes and num_devices need to match or we should
2049 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2050 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2054 /* helper to cleanup workers */
2055 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2057 btrfs_destroy_workqueue(fs_info->fixup_workers);
2058 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2059 btrfs_destroy_workqueue(fs_info->workers);
2060 btrfs_destroy_workqueue(fs_info->endio_workers);
2061 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2062 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2063 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2064 btrfs_destroy_workqueue(fs_info->rmw_workers);
2065 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2066 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2067 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2068 btrfs_destroy_workqueue(fs_info->submit_workers);
2069 btrfs_destroy_workqueue(fs_info->delayed_workers);
2070 btrfs_destroy_workqueue(fs_info->caching_workers);
2071 btrfs_destroy_workqueue(fs_info->readahead_workers);
2072 btrfs_destroy_workqueue(fs_info->flush_workers);
2073 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2074 btrfs_destroy_workqueue(fs_info->extent_workers);
2077 static void free_root_extent_buffers(struct btrfs_root *root)
2080 free_extent_buffer(root->node);
2081 free_extent_buffer(root->commit_root);
2083 root->commit_root = NULL;
2087 /* helper to cleanup tree roots */
2088 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2090 free_root_extent_buffers(info->tree_root);
2092 free_root_extent_buffers(info->dev_root);
2093 free_root_extent_buffers(info->extent_root);
2094 free_root_extent_buffers(info->csum_root);
2095 free_root_extent_buffers(info->quota_root);
2096 free_root_extent_buffers(info->uuid_root);
2098 free_root_extent_buffers(info->chunk_root);
2101 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2104 struct btrfs_root *gang[8];
2107 while (!list_empty(&fs_info->dead_roots)) {
2108 gang[0] = list_entry(fs_info->dead_roots.next,
2109 struct btrfs_root, root_list);
2110 list_del(&gang[0]->root_list);
2112 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2113 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2115 free_extent_buffer(gang[0]->node);
2116 free_extent_buffer(gang[0]->commit_root);
2117 btrfs_put_fs_root(gang[0]);
2122 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2127 for (i = 0; i < ret; i++)
2128 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2131 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2132 btrfs_free_log_root_tree(NULL, fs_info);
2133 btrfs_destroy_pinned_extent(fs_info->tree_root,
2134 fs_info->pinned_extents);
2138 int open_ctree(struct super_block *sb,
2139 struct btrfs_fs_devices *fs_devices,
2148 struct btrfs_key location;
2149 struct buffer_head *bh;
2150 struct btrfs_super_block *disk_super;
2151 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2152 struct btrfs_root *tree_root;
2153 struct btrfs_root *extent_root;
2154 struct btrfs_root *csum_root;
2155 struct btrfs_root *chunk_root;
2156 struct btrfs_root *dev_root;
2157 struct btrfs_root *quota_root;
2158 struct btrfs_root *uuid_root;
2159 struct btrfs_root *log_tree_root;
2162 int num_backups_tried = 0;
2163 int backup_index = 0;
2165 int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2166 bool create_uuid_tree;
2167 bool check_uuid_tree;
2169 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2170 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2171 if (!tree_root || !chunk_root) {
2176 ret = init_srcu_struct(&fs_info->subvol_srcu);
2182 ret = setup_bdi(fs_info, &fs_info->bdi);
2188 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2193 fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2194 (1 + ilog2(nr_cpu_ids));
2196 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2199 goto fail_dirty_metadata_bytes;
2202 ret = percpu_counter_init(&fs_info->bio_counter, 0);
2205 goto fail_delalloc_bytes;
2208 fs_info->btree_inode = new_inode(sb);
2209 if (!fs_info->btree_inode) {
2211 goto fail_bio_counter;
2214 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2216 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2217 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2218 INIT_LIST_HEAD(&fs_info->trans_list);
2219 INIT_LIST_HEAD(&fs_info->dead_roots);
2220 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2221 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2222 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2223 spin_lock_init(&fs_info->delalloc_root_lock);
2224 spin_lock_init(&fs_info->trans_lock);
2225 spin_lock_init(&fs_info->fs_roots_radix_lock);
2226 spin_lock_init(&fs_info->delayed_iput_lock);
2227 spin_lock_init(&fs_info->defrag_inodes_lock);
2228 spin_lock_init(&fs_info->free_chunk_lock);
2229 spin_lock_init(&fs_info->tree_mod_seq_lock);
2230 spin_lock_init(&fs_info->super_lock);
2231 spin_lock_init(&fs_info->qgroup_op_lock);
2232 spin_lock_init(&fs_info->buffer_lock);
2233 rwlock_init(&fs_info->tree_mod_log_lock);
2234 mutex_init(&fs_info->reloc_mutex);
2235 mutex_init(&fs_info->delalloc_root_mutex);
2236 seqlock_init(&fs_info->profiles_lock);
2238 init_completion(&fs_info->kobj_unregister);
2239 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2240 INIT_LIST_HEAD(&fs_info->space_info);
2241 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2242 btrfs_mapping_init(&fs_info->mapping_tree);
2243 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2244 BTRFS_BLOCK_RSV_GLOBAL);
2245 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2246 BTRFS_BLOCK_RSV_DELALLOC);
2247 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2248 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2249 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2250 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2251 BTRFS_BLOCK_RSV_DELOPS);
2252 atomic_set(&fs_info->nr_async_submits, 0);
2253 atomic_set(&fs_info->async_delalloc_pages, 0);
2254 atomic_set(&fs_info->async_submit_draining, 0);
2255 atomic_set(&fs_info->nr_async_bios, 0);
2256 atomic_set(&fs_info->defrag_running, 0);
2257 atomic_set(&fs_info->qgroup_op_seq, 0);
2258 atomic64_set(&fs_info->tree_mod_seq, 0);
2260 fs_info->max_inline = 8192 * 1024;
2261 fs_info->metadata_ratio = 0;
2262 fs_info->defrag_inodes = RB_ROOT;
2263 fs_info->free_chunk_space = 0;
2264 fs_info->tree_mod_log = RB_ROOT;
2265 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2266 fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2267 /* readahead state */
2268 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2269 spin_lock_init(&fs_info->reada_lock);
2271 fs_info->thread_pool_size = min_t(unsigned long,
2272 num_online_cpus() + 2, 8);
2274 INIT_LIST_HEAD(&fs_info->ordered_roots);
2275 spin_lock_init(&fs_info->ordered_root_lock);
2276 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2278 if (!fs_info->delayed_root) {
2282 btrfs_init_delayed_root(fs_info->delayed_root);
2284 mutex_init(&fs_info->scrub_lock);
2285 atomic_set(&fs_info->scrubs_running, 0);
2286 atomic_set(&fs_info->scrub_pause_req, 0);
2287 atomic_set(&fs_info->scrubs_paused, 0);
2288 atomic_set(&fs_info->scrub_cancel_req, 0);
2289 init_waitqueue_head(&fs_info->replace_wait);
2290 init_waitqueue_head(&fs_info->scrub_pause_wait);
2291 fs_info->scrub_workers_refcnt = 0;
2292 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2293 fs_info->check_integrity_print_mask = 0;
2296 spin_lock_init(&fs_info->balance_lock);
2297 mutex_init(&fs_info->balance_mutex);
2298 atomic_set(&fs_info->balance_running, 0);
2299 atomic_set(&fs_info->balance_pause_req, 0);
2300 atomic_set(&fs_info->balance_cancel_req, 0);
2301 fs_info->balance_ctl = NULL;
2302 init_waitqueue_head(&fs_info->balance_wait_q);
2303 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2305 sb->s_blocksize = 4096;
2306 sb->s_blocksize_bits = blksize_bits(4096);
2307 sb->s_bdi = &fs_info->bdi;
2309 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2310 set_nlink(fs_info->btree_inode, 1);
2312 * we set the i_size on the btree inode to the max possible int.
2313 * the real end of the address space is determined by all of
2314 * the devices in the system
2316 fs_info->btree_inode->i_size = OFFSET_MAX;
2317 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2318 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2320 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2321 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2322 fs_info->btree_inode->i_mapping);
2323 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2324 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2326 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2328 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2329 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2330 sizeof(struct btrfs_key));
2331 set_bit(BTRFS_INODE_DUMMY,
2332 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2333 btrfs_insert_inode_hash(fs_info->btree_inode);
2335 spin_lock_init(&fs_info->block_group_cache_lock);
2336 fs_info->block_group_cache_tree = RB_ROOT;
2337 fs_info->first_logical_byte = (u64)-1;
2339 extent_io_tree_init(&fs_info->freed_extents[0],
2340 fs_info->btree_inode->i_mapping);
2341 extent_io_tree_init(&fs_info->freed_extents[1],
2342 fs_info->btree_inode->i_mapping);
2343 fs_info->pinned_extents = &fs_info->freed_extents[0];
2344 fs_info->do_barriers = 1;
2347 mutex_init(&fs_info->ordered_operations_mutex);
2348 mutex_init(&fs_info->ordered_extent_flush_mutex);
2349 mutex_init(&fs_info->tree_log_mutex);
2350 mutex_init(&fs_info->chunk_mutex);
2351 mutex_init(&fs_info->transaction_kthread_mutex);
2352 mutex_init(&fs_info->cleaner_mutex);
2353 mutex_init(&fs_info->volume_mutex);
2354 init_rwsem(&fs_info->commit_root_sem);
2355 init_rwsem(&fs_info->cleanup_work_sem);
2356 init_rwsem(&fs_info->subvol_sem);
2357 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2358 fs_info->dev_replace.lock_owner = 0;
2359 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2360 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2361 mutex_init(&fs_info->dev_replace.lock_management_lock);
2362 mutex_init(&fs_info->dev_replace.lock);
2364 spin_lock_init(&fs_info->qgroup_lock);
2365 mutex_init(&fs_info->qgroup_ioctl_lock);
2366 fs_info->qgroup_tree = RB_ROOT;
2367 fs_info->qgroup_op_tree = RB_ROOT;
2368 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2369 fs_info->qgroup_seq = 1;
2370 fs_info->quota_enabled = 0;
2371 fs_info->pending_quota_state = 0;
2372 fs_info->qgroup_ulist = NULL;
2373 mutex_init(&fs_info->qgroup_rescan_lock);
2375 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2376 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2378 init_waitqueue_head(&fs_info->transaction_throttle);
2379 init_waitqueue_head(&fs_info->transaction_wait);
2380 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2381 init_waitqueue_head(&fs_info->async_submit_wait);
2383 ret = btrfs_alloc_stripe_hash_table(fs_info);
2389 __setup_root(4096, 4096, 4096, tree_root,
2390 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2392 invalidate_bdev(fs_devices->latest_bdev);
2395 * Read super block and check the signature bytes only
2397 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2404 * We want to check superblock checksum, the type is stored inside.
2405 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2407 if (btrfs_check_super_csum(bh->b_data)) {
2408 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2414 * super_copy is zeroed at allocation time and we never touch the
2415 * following bytes up to INFO_SIZE, the checksum is calculated from
2416 * the whole block of INFO_SIZE
2418 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2419 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2420 sizeof(*fs_info->super_for_commit));
2423 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2425 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2427 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2432 disk_super = fs_info->super_copy;
2433 if (!btrfs_super_root(disk_super))
2436 /* check FS state, whether FS is broken. */
2437 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2438 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2441 * run through our array of backup supers and setup
2442 * our ring pointer to the oldest one
2444 generation = btrfs_super_generation(disk_super);
2445 find_oldest_super_backup(fs_info, generation);
2448 * In the long term, we'll store the compression type in the super
2449 * block, and it'll be used for per file compression control.
2451 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2453 ret = btrfs_parse_options(tree_root, options);
2459 features = btrfs_super_incompat_flags(disk_super) &
2460 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2462 printk(KERN_ERR "BTRFS: couldn't mount because of "
2463 "unsupported optional features (%Lx).\n",
2470 * Leafsize and nodesize were always equal, this is only a sanity check.
2472 if (le32_to_cpu(disk_super->__unused_leafsize) !=
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 le32_to_cpu(disk_super->__unused_leafsize));
2481 if (btrfs_super_nodesize(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_nodesize(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_nodesize(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 sectorsize = btrfs_super_sectorsize(disk_super);
2509 stripesize = btrfs_super_stripesize(disk_super);
2510 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2511 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2514 * mixed block groups end up with duplicate but slightly offset
2515 * extent buffers for the same range. It leads to corruptions
2517 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2518 (sectorsize != nodesize)) {
2519 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2520 "are not allowed for mixed block groups on %s\n",
2526 * Needn't use the lock because there is no other task which will
2529 btrfs_set_super_incompat_flags(disk_super, features);
2531 features = btrfs_super_compat_ro_flags(disk_super) &
2532 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2533 if (!(sb->s_flags & MS_RDONLY) && features) {
2534 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2535 "unsupported option features (%Lx).\n",
2541 max_active = fs_info->thread_pool_size;
2544 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2547 fs_info->delalloc_workers =
2548 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2550 fs_info->flush_workers =
2551 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2553 fs_info->caching_workers =
2554 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2557 * a higher idle thresh on the submit workers makes it much more
2558 * likely that bios will be send down in a sane order to the
2561 fs_info->submit_workers =
2562 btrfs_alloc_workqueue("submit", flags,
2563 min_t(u64, fs_devices->num_devices,
2566 fs_info->fixup_workers =
2567 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2570 * endios are largely parallel and should have a very
2573 fs_info->endio_workers =
2574 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2575 fs_info->endio_meta_workers =
2576 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2577 fs_info->endio_meta_write_workers =
2578 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2579 fs_info->endio_raid56_workers =
2580 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2581 fs_info->endio_repair_workers =
2582 btrfs_alloc_workqueue("endio-repair", flags, 1, 0);
2583 fs_info->rmw_workers =
2584 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2585 fs_info->endio_write_workers =
2586 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2587 fs_info->endio_freespace_worker =
2588 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2589 fs_info->delayed_workers =
2590 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2591 fs_info->readahead_workers =
2592 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2593 fs_info->qgroup_rescan_workers =
2594 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2595 fs_info->extent_workers =
2596 btrfs_alloc_workqueue("extent-refs", flags,
2597 min_t(u64, fs_devices->num_devices,
2600 if (!(fs_info->workers && fs_info->delalloc_workers &&
2601 fs_info->submit_workers && fs_info->flush_workers &&
2602 fs_info->endio_workers && fs_info->endio_meta_workers &&
2603 fs_info->endio_meta_write_workers &&
2604 fs_info->endio_repair_workers &&
2605 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2606 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2607 fs_info->caching_workers && fs_info->readahead_workers &&
2608 fs_info->fixup_workers && fs_info->delayed_workers &&
2609 fs_info->extent_workers &&
2610 fs_info->qgroup_rescan_workers)) {
2612 goto fail_sb_buffer;
2615 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2616 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2617 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2619 tree_root->nodesize = nodesize;
2620 tree_root->sectorsize = sectorsize;
2621 tree_root->stripesize = stripesize;
2623 sb->s_blocksize = sectorsize;
2624 sb->s_blocksize_bits = blksize_bits(sectorsize);
2626 if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2627 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2628 goto fail_sb_buffer;
2631 if (sectorsize != PAGE_SIZE) {
2632 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2633 "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2634 goto fail_sb_buffer;
2637 mutex_lock(&fs_info->chunk_mutex);
2638 ret = btrfs_read_sys_array(tree_root);
2639 mutex_unlock(&fs_info->chunk_mutex);
2641 printk(KERN_WARNING "BTRFS: failed to read the system "
2642 "array on %s\n", sb->s_id);
2643 goto fail_sb_buffer;
2646 blocksize = tree_root->nodesize;
2647 generation = btrfs_super_chunk_root_generation(disk_super);
2649 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2650 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 = tree_root->nodesize;
2688 generation = btrfs_super_generation(disk_super);
2690 tree_root->node = read_tree_block(tree_root,
2691 btrfs_super_root(disk_super),
2692 blocksize, generation);
2693 if (!tree_root->node ||
2694 !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2695 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2698 goto recovery_tree_root;
2701 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2702 tree_root->commit_root = btrfs_root_node(tree_root);
2703 btrfs_set_root_refs(&tree_root->root_item, 1);
2705 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2706 location.type = BTRFS_ROOT_ITEM_KEY;
2707 location.offset = 0;
2709 extent_root = btrfs_read_tree_root(tree_root, &location);
2710 if (IS_ERR(extent_root)) {
2711 ret = PTR_ERR(extent_root);
2712 goto recovery_tree_root;
2714 set_bit(BTRFS_ROOT_TRACK_DIRTY, &extent_root->state);
2715 fs_info->extent_root = extent_root;
2717 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2718 dev_root = btrfs_read_tree_root(tree_root, &location);
2719 if (IS_ERR(dev_root)) {
2720 ret = PTR_ERR(dev_root);
2721 goto recovery_tree_root;
2723 set_bit(BTRFS_ROOT_TRACK_DIRTY, &dev_root->state);
2724 fs_info->dev_root = dev_root;
2725 btrfs_init_devices_late(fs_info);
2727 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2728 csum_root = btrfs_read_tree_root(tree_root, &location);
2729 if (IS_ERR(csum_root)) {
2730 ret = PTR_ERR(csum_root);
2731 goto recovery_tree_root;
2733 set_bit(BTRFS_ROOT_TRACK_DIRTY, &csum_root->state);
2734 fs_info->csum_root = csum_root;
2736 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2737 quota_root = btrfs_read_tree_root(tree_root, &location);
2738 if (!IS_ERR(quota_root)) {
2739 set_bit(BTRFS_ROOT_TRACK_DIRTY, "a_root->state);
2740 fs_info->quota_enabled = 1;
2741 fs_info->pending_quota_state = 1;
2742 fs_info->quota_root = quota_root;
2745 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2746 uuid_root = btrfs_read_tree_root(tree_root, &location);
2747 if (IS_ERR(uuid_root)) {
2748 ret = PTR_ERR(uuid_root);
2750 goto recovery_tree_root;
2751 create_uuid_tree = true;
2752 check_uuid_tree = false;
2754 set_bit(BTRFS_ROOT_TRACK_DIRTY, &uuid_root->state);
2755 fs_info->uuid_root = uuid_root;
2756 create_uuid_tree = false;
2758 generation != btrfs_super_uuid_tree_generation(disk_super);
2761 fs_info->generation = generation;
2762 fs_info->last_trans_committed = generation;
2764 ret = btrfs_recover_balance(fs_info);
2766 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2767 goto fail_block_groups;
2770 ret = btrfs_init_dev_stats(fs_info);
2772 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2774 goto fail_block_groups;
2777 ret = btrfs_init_dev_replace(fs_info);
2779 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2780 goto fail_block_groups;
2783 btrfs_close_extra_devices(fs_info, fs_devices, 1);
2785 ret = btrfs_sysfs_add_one(fs_info);
2787 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2788 goto fail_block_groups;
2791 ret = btrfs_init_space_info(fs_info);
2793 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2797 ret = btrfs_read_block_groups(extent_root);
2799 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2802 fs_info->num_tolerated_disk_barrier_failures =
2803 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2804 if (fs_info->fs_devices->missing_devices >
2805 fs_info->num_tolerated_disk_barrier_failures &&
2806 !(sb->s_flags & MS_RDONLY)) {
2807 printk(KERN_WARNING "BTRFS: "
2808 "too many missing devices, writeable mount is not allowed\n");
2812 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2814 if (IS_ERR(fs_info->cleaner_kthread))
2817 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2819 "btrfs-transaction");
2820 if (IS_ERR(fs_info->transaction_kthread))
2823 if (!btrfs_test_opt(tree_root, SSD) &&
2824 !btrfs_test_opt(tree_root, NOSSD) &&
2825 !fs_info->fs_devices->rotating) {
2826 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2828 btrfs_set_opt(fs_info->mount_opt, SSD);
2831 /* Set the real inode map cache flag */
2832 if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2833 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2835 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2836 if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2837 ret = btrfsic_mount(tree_root, fs_devices,
2838 btrfs_test_opt(tree_root,
2839 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2841 fs_info->check_integrity_print_mask);
2843 printk(KERN_WARNING "BTRFS: failed to initialize"
2844 " integrity check module %s\n", sb->s_id);
2847 ret = btrfs_read_qgroup_config(fs_info);
2849 goto fail_trans_kthread;
2851 /* do not make disk changes in broken FS */
2852 if (btrfs_super_log_root(disk_super) != 0) {
2853 u64 bytenr = btrfs_super_log_root(disk_super);
2855 if (fs_devices->rw_devices == 0) {
2856 printk(KERN_WARNING "BTRFS: log replay required "
2861 blocksize = tree_root->nodesize;
2863 log_tree_root = btrfs_alloc_root(fs_info);
2864 if (!log_tree_root) {
2869 __setup_root(nodesize, sectorsize, stripesize,
2870 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2872 log_tree_root->node = read_tree_block(tree_root, bytenr,
2875 if (!log_tree_root->node ||
2876 !extent_buffer_uptodate(log_tree_root->node)) {
2877 printk(KERN_ERR "BTRFS: failed to read log tree\n");
2878 free_extent_buffer(log_tree_root->node);
2879 kfree(log_tree_root);
2882 /* returns with log_tree_root freed on success */
2883 ret = btrfs_recover_log_trees(log_tree_root);
2885 btrfs_error(tree_root->fs_info, ret,
2886 "Failed to recover log tree");
2887 free_extent_buffer(log_tree_root->node);
2888 kfree(log_tree_root);
2892 if (sb->s_flags & MS_RDONLY) {
2893 ret = btrfs_commit_super(tree_root);
2899 ret = btrfs_find_orphan_roots(tree_root);
2903 if (!(sb->s_flags & MS_RDONLY)) {
2904 ret = btrfs_cleanup_fs_roots(fs_info);
2908 mutex_lock(&fs_info->cleaner_mutex);
2909 ret = btrfs_recover_relocation(tree_root);
2910 mutex_unlock(&fs_info->cleaner_mutex);
2913 "BTRFS: failed to recover relocation\n");
2919 location.objectid = BTRFS_FS_TREE_OBJECTID;
2920 location.type = BTRFS_ROOT_ITEM_KEY;
2921 location.offset = 0;
2923 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2924 if (IS_ERR(fs_info->fs_root)) {
2925 err = PTR_ERR(fs_info->fs_root);
2929 if (sb->s_flags & MS_RDONLY)
2932 down_read(&fs_info->cleanup_work_sem);
2933 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2934 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2935 up_read(&fs_info->cleanup_work_sem);
2936 close_ctree(tree_root);
2939 up_read(&fs_info->cleanup_work_sem);
2941 ret = btrfs_resume_balance_async(fs_info);
2943 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2944 close_ctree(tree_root);
2948 ret = btrfs_resume_dev_replace_async(fs_info);
2950 pr_warn("BTRFS: failed to resume dev_replace\n");
2951 close_ctree(tree_root);
2955 btrfs_qgroup_rescan_resume(fs_info);
2957 if (create_uuid_tree) {
2958 pr_info("BTRFS: creating UUID tree\n");
2959 ret = btrfs_create_uuid_tree(fs_info);
2961 pr_warn("BTRFS: failed to create the UUID tree %d\n",
2963 close_ctree(tree_root);
2966 } else if (check_uuid_tree ||
2967 btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2968 pr_info("BTRFS: checking UUID tree\n");
2969 ret = btrfs_check_uuid_tree(fs_info);
2971 pr_warn("BTRFS: failed to check the UUID tree %d\n",
2973 close_ctree(tree_root);
2977 fs_info->update_uuid_tree_gen = 1;
2983 btrfs_free_qgroup_config(fs_info);
2985 kthread_stop(fs_info->transaction_kthread);
2986 btrfs_cleanup_transaction(fs_info->tree_root);
2987 btrfs_free_fs_roots(fs_info);
2989 kthread_stop(fs_info->cleaner_kthread);
2992 * make sure we're done with the btree inode before we stop our
2995 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2998 btrfs_sysfs_remove_one(fs_info);
3001 btrfs_put_block_group_cache(fs_info);
3002 btrfs_free_block_groups(fs_info);
3005 free_root_pointers(fs_info, 1);
3006 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3009 btrfs_stop_all_workers(fs_info);
3012 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3014 iput(fs_info->btree_inode);
3016 percpu_counter_destroy(&fs_info->bio_counter);
3017 fail_delalloc_bytes:
3018 percpu_counter_destroy(&fs_info->delalloc_bytes);
3019 fail_dirty_metadata_bytes:
3020 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3022 bdi_destroy(&fs_info->bdi);
3024 cleanup_srcu_struct(&fs_info->subvol_srcu);
3026 btrfs_free_stripe_hash_table(fs_info);
3027 btrfs_close_devices(fs_info->fs_devices);
3031 if (!btrfs_test_opt(tree_root, RECOVERY))
3032 goto fail_tree_roots;
3034 free_root_pointers(fs_info, 0);
3036 /* don't use the log in recovery mode, it won't be valid */
3037 btrfs_set_super_log_root(disk_super, 0);
3039 /* we can't trust the free space cache either */
3040 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3042 ret = next_root_backup(fs_info, fs_info->super_copy,
3043 &num_backups_tried, &backup_index);
3045 goto fail_block_groups;
3046 goto retry_root_backup;
3049 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3052 set_buffer_uptodate(bh);
3054 struct btrfs_device *device = (struct btrfs_device *)
3057 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3058 "I/O error on %s\n",
3059 rcu_str_deref(device->name));
3060 /* note, we dont' set_buffer_write_io_error because we have
3061 * our own ways of dealing with the IO errors
3063 clear_buffer_uptodate(bh);
3064 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3070 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3072 struct buffer_head *bh;
3073 struct buffer_head *latest = NULL;
3074 struct btrfs_super_block *super;
3079 /* we would like to check all the supers, but that would make
3080 * a btrfs mount succeed after a mkfs from a different FS.
3081 * So, we need to add a special mount option to scan for
3082 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3084 for (i = 0; i < 1; i++) {
3085 bytenr = btrfs_sb_offset(i);
3086 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3087 i_size_read(bdev->bd_inode))
3089 bh = __bread(bdev, bytenr / 4096,
3090 BTRFS_SUPER_INFO_SIZE);
3094 super = (struct btrfs_super_block *)bh->b_data;
3095 if (btrfs_super_bytenr(super) != bytenr ||
3096 btrfs_super_magic(super) != BTRFS_MAGIC) {
3101 if (!latest || btrfs_super_generation(super) > transid) {
3104 transid = btrfs_super_generation(super);
3113 * this should be called twice, once with wait == 0 and
3114 * once with wait == 1. When wait == 0 is done, all the buffer heads
3115 * we write are pinned.
3117 * They are released when wait == 1 is done.
3118 * max_mirrors must be the same for both runs, and it indicates how
3119 * many supers on this one device should be written.
3121 * max_mirrors == 0 means to write them all.
3123 static int write_dev_supers(struct btrfs_device *device,
3124 struct btrfs_super_block *sb,
3125 int do_barriers, int wait, int max_mirrors)
3127 struct buffer_head *bh;
3134 if (max_mirrors == 0)
3135 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3137 for (i = 0; i < max_mirrors; i++) {
3138 bytenr = btrfs_sb_offset(i);
3139 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3140 device->commit_total_bytes)
3144 bh = __find_get_block(device->bdev, bytenr / 4096,
3145 BTRFS_SUPER_INFO_SIZE);
3151 if (!buffer_uptodate(bh))
3154 /* drop our reference */
3157 /* drop the reference from the wait == 0 run */
3161 btrfs_set_super_bytenr(sb, bytenr);
3164 crc = btrfs_csum_data((char *)sb +
3165 BTRFS_CSUM_SIZE, crc,
3166 BTRFS_SUPER_INFO_SIZE -
3168 btrfs_csum_final(crc, sb->csum);
3171 * one reference for us, and we leave it for the
3174 bh = __getblk(device->bdev, bytenr / 4096,
3175 BTRFS_SUPER_INFO_SIZE);
3177 printk(KERN_ERR "BTRFS: couldn't get super "
3178 "buffer head for bytenr %Lu\n", bytenr);
3183 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3185 /* one reference for submit_bh */
3188 set_buffer_uptodate(bh);
3190 bh->b_end_io = btrfs_end_buffer_write_sync;
3191 bh->b_private = device;
3195 * we fua the first super. The others we allow
3199 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3201 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3205 return errors < i ? 0 : -1;
3209 * endio for the write_dev_flush, this will wake anyone waiting
3210 * for the barrier when it is done
3212 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3215 if (err == -EOPNOTSUPP)
3216 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3217 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3219 if (bio->bi_private)
3220 complete(bio->bi_private);
3225 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3226 * sent down. With wait == 1, it waits for the previous flush.
3228 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3231 static int write_dev_flush(struct btrfs_device *device, int wait)
3236 if (device->nobarriers)
3240 bio = device->flush_bio;
3244 wait_for_completion(&device->flush_wait);
3246 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3247 printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3248 rcu_str_deref(device->name));
3249 device->nobarriers = 1;
3250 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3252 btrfs_dev_stat_inc_and_print(device,
3253 BTRFS_DEV_STAT_FLUSH_ERRS);
3256 /* drop the reference from the wait == 0 run */
3258 device->flush_bio = NULL;
3264 * one reference for us, and we leave it for the
3267 device->flush_bio = NULL;
3268 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3272 bio->bi_end_io = btrfs_end_empty_barrier;
3273 bio->bi_bdev = device->bdev;
3274 init_completion(&device->flush_wait);
3275 bio->bi_private = &device->flush_wait;
3276 device->flush_bio = bio;
3279 btrfsic_submit_bio(WRITE_FLUSH, bio);
3285 * send an empty flush down to each device in parallel,
3286 * then wait for them
3288 static int barrier_all_devices(struct btrfs_fs_info *info)
3290 struct list_head *head;
3291 struct btrfs_device *dev;
3292 int errors_send = 0;
3293 int errors_wait = 0;
3296 /* send down all the barriers */
3297 head = &info->fs_devices->devices;
3298 list_for_each_entry_rcu(dev, head, dev_list) {
3305 if (!dev->in_fs_metadata || !dev->writeable)
3308 ret = write_dev_flush(dev, 0);
3313 /* wait for all the barriers */
3314 list_for_each_entry_rcu(dev, head, dev_list) {
3321 if (!dev->in_fs_metadata || !dev->writeable)
3324 ret = write_dev_flush(dev, 1);
3328 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3329 errors_wait > info->num_tolerated_disk_barrier_failures)
3334 int btrfs_calc_num_tolerated_disk_barrier_failures(
3335 struct btrfs_fs_info *fs_info)
3337 struct btrfs_ioctl_space_info space;
3338 struct btrfs_space_info *sinfo;
3339 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3340 BTRFS_BLOCK_GROUP_SYSTEM,
3341 BTRFS_BLOCK_GROUP_METADATA,
3342 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3346 int num_tolerated_disk_barrier_failures =
3347 (int)fs_info->fs_devices->num_devices;
3349 for (i = 0; i < num_types; i++) {
3350 struct btrfs_space_info *tmp;
3354 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3355 if (tmp->flags == types[i]) {
3365 down_read(&sinfo->groups_sem);
3366 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3367 if (!list_empty(&sinfo->block_groups[c])) {
3370 btrfs_get_block_group_info(
3371 &sinfo->block_groups[c], &space);
3372 if (space.total_bytes == 0 ||
3373 space.used_bytes == 0)
3375 flags = space.flags;
3378 * 0: if dup, single or RAID0 is configured for
3379 * any of metadata, system or data, else
3380 * 1: if RAID5 is configured, or if RAID1 or
3381 * RAID10 is configured and only two mirrors
3383 * 2: if RAID6 is configured, else
3384 * num_mirrors - 1: if RAID1 or RAID10 is
3385 * configured and more than
3386 * 2 mirrors are used.
3388 if (num_tolerated_disk_barrier_failures > 0 &&
3389 ((flags & (BTRFS_BLOCK_GROUP_DUP |
3390 BTRFS_BLOCK_GROUP_RAID0)) ||
3391 ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3393 num_tolerated_disk_barrier_failures = 0;
3394 else if (num_tolerated_disk_barrier_failures > 1) {
3395 if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3396 BTRFS_BLOCK_GROUP_RAID5 |
3397 BTRFS_BLOCK_GROUP_RAID10)) {
3398 num_tolerated_disk_barrier_failures = 1;
3400 BTRFS_BLOCK_GROUP_RAID6) {
3401 num_tolerated_disk_barrier_failures = 2;
3406 up_read(&sinfo->groups_sem);
3409 return num_tolerated_disk_barrier_failures;
3412 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3414 struct list_head *head;
3415 struct btrfs_device *dev;
3416 struct btrfs_super_block *sb;
3417 struct btrfs_dev_item *dev_item;
3421 int total_errors = 0;
3424 do_barriers = !btrfs_test_opt(root, NOBARRIER);
3425 backup_super_roots(root->fs_info);
3427 sb = root->fs_info->super_for_commit;
3428 dev_item = &sb->dev_item;
3430 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3431 head = &root->fs_info->fs_devices->devices;
3432 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3435 ret = barrier_all_devices(root->fs_info);
3438 &root->fs_info->fs_devices->device_list_mutex);
3439 btrfs_error(root->fs_info, ret,
3440 "errors while submitting device barriers.");
3445 list_for_each_entry_rcu(dev, head, dev_list) {
3450 if (!dev->in_fs_metadata || !dev->writeable)
3453 btrfs_set_stack_device_generation(dev_item, 0);
3454 btrfs_set_stack_device_type(dev_item, dev->type);
3455 btrfs_set_stack_device_id(dev_item, dev->devid);
3456 btrfs_set_stack_device_total_bytes(dev_item,
3457 dev->commit_total_bytes);
3458 btrfs_set_stack_device_bytes_used(dev_item,
3459 dev->commit_bytes_used);
3460 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3461 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3462 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3463 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3464 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3466 flags = btrfs_super_flags(sb);
3467 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3469 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3473 if (total_errors > max_errors) {
3474 btrfs_err(root->fs_info, "%d errors while writing supers",
3476 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3478 /* FUA is masked off if unsupported and can't be the reason */
3479 btrfs_error(root->fs_info, -EIO,
3480 "%d errors while writing supers", total_errors);
3485 list_for_each_entry_rcu(dev, head, dev_list) {
3488 if (!dev->in_fs_metadata || !dev->writeable)
3491 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3495 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3496 if (total_errors > max_errors) {
3497 btrfs_error(root->fs_info, -EIO,
3498 "%d errors while writing supers", total_errors);
3504 int write_ctree_super(struct btrfs_trans_handle *trans,
3505 struct btrfs_root *root, int max_mirrors)
3507 return write_all_supers(root, max_mirrors);
3510 /* Drop a fs root from the radix tree and free it. */
3511 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3512 struct btrfs_root *root)
3514 spin_lock(&fs_info->fs_roots_radix_lock);
3515 radix_tree_delete(&fs_info->fs_roots_radix,
3516 (unsigned long)root->root_key.objectid);
3517 spin_unlock(&fs_info->fs_roots_radix_lock);
3519 if (btrfs_root_refs(&root->root_item) == 0)
3520 synchronize_srcu(&fs_info->subvol_srcu);
3522 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3523 btrfs_free_log(NULL, root);
3525 if (root->free_ino_pinned)
3526 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3527 if (root->free_ino_ctl)
3528 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3532 static void free_fs_root(struct btrfs_root *root)
3534 iput(root->ino_cache_inode);
3535 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3536 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3537 root->orphan_block_rsv = NULL;
3539 free_anon_bdev(root->anon_dev);
3540 if (root->subv_writers)
3541 btrfs_free_subvolume_writers(root->subv_writers);
3542 free_extent_buffer(root->node);
3543 free_extent_buffer(root->commit_root);
3544 kfree(root->free_ino_ctl);
3545 kfree(root->free_ino_pinned);
3547 btrfs_put_fs_root(root);
3550 void btrfs_free_fs_root(struct btrfs_root *root)
3555 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3557 u64 root_objectid = 0;
3558 struct btrfs_root *gang[8];
3561 unsigned int ret = 0;
3565 index = srcu_read_lock(&fs_info->subvol_srcu);
3566 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3567 (void **)gang, root_objectid,
3570 srcu_read_unlock(&fs_info->subvol_srcu, index);
3573 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3575 for (i = 0; i < ret; i++) {
3576 /* Avoid to grab roots in dead_roots */
3577 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3581 /* grab all the search result for later use */
3582 gang[i] = btrfs_grab_fs_root(gang[i]);
3584 srcu_read_unlock(&fs_info->subvol_srcu, index);
3586 for (i = 0; i < ret; i++) {
3589 root_objectid = gang[i]->root_key.objectid;
3590 err = btrfs_orphan_cleanup(gang[i]);
3593 btrfs_put_fs_root(gang[i]);
3598 /* release the uncleaned roots due to error */
3599 for (; i < ret; i++) {
3601 btrfs_put_fs_root(gang[i]);
3606 int btrfs_commit_super(struct btrfs_root *root)
3608 struct btrfs_trans_handle *trans;
3610 mutex_lock(&root->fs_info->cleaner_mutex);
3611 btrfs_run_delayed_iputs(root);
3612 mutex_unlock(&root->fs_info->cleaner_mutex);
3613 wake_up_process(root->fs_info->cleaner_kthread);
3615 /* wait until ongoing cleanup work done */
3616 down_write(&root->fs_info->cleanup_work_sem);
3617 up_write(&root->fs_info->cleanup_work_sem);
3619 trans = btrfs_join_transaction(root);
3621 return PTR_ERR(trans);
3622 return btrfs_commit_transaction(trans, root);
3625 void close_ctree(struct btrfs_root *root)
3627 struct btrfs_fs_info *fs_info = root->fs_info;
3630 fs_info->closing = 1;
3633 /* wait for the uuid_scan task to finish */
3634 down(&fs_info->uuid_tree_rescan_sem);
3635 /* avoid complains from lockdep et al., set sem back to initial state */
3636 up(&fs_info->uuid_tree_rescan_sem);
3638 /* pause restriper - we want to resume on mount */
3639 btrfs_pause_balance(fs_info);
3641 btrfs_dev_replace_suspend_for_unmount(fs_info);
3643 btrfs_scrub_cancel(fs_info);
3645 /* wait for any defraggers to finish */
3646 wait_event(fs_info->transaction_wait,
3647 (atomic_read(&fs_info->defrag_running) == 0));
3649 /* clear out the rbtree of defraggable inodes */
3650 btrfs_cleanup_defrag_inodes(fs_info);
3652 cancel_work_sync(&fs_info->async_reclaim_work);
3654 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3655 ret = btrfs_commit_super(root);
3657 btrfs_err(root->fs_info, "commit super ret %d", ret);
3660 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3661 btrfs_error_commit_super(root);
3663 kthread_stop(fs_info->transaction_kthread);
3664 kthread_stop(fs_info->cleaner_kthread);
3666 fs_info->closing = 2;
3669 btrfs_free_qgroup_config(root->fs_info);
3671 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3672 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3673 percpu_counter_sum(&fs_info->delalloc_bytes));
3676 btrfs_sysfs_remove_one(fs_info);
3678 btrfs_free_fs_roots(fs_info);
3680 btrfs_put_block_group_cache(fs_info);
3682 btrfs_free_block_groups(fs_info);
3685 * we must make sure there is not any read request to
3686 * submit after we stopping all workers.
3688 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3689 btrfs_stop_all_workers(fs_info);
3691 free_root_pointers(fs_info, 1);
3693 iput(fs_info->btree_inode);
3695 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3696 if (btrfs_test_opt(root, CHECK_INTEGRITY))
3697 btrfsic_unmount(root, fs_info->fs_devices);
3700 btrfs_close_devices(fs_info->fs_devices);
3701 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3703 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3704 percpu_counter_destroy(&fs_info->delalloc_bytes);
3705 percpu_counter_destroy(&fs_info->bio_counter);
3706 bdi_destroy(&fs_info->bdi);
3707 cleanup_srcu_struct(&fs_info->subvol_srcu);
3709 btrfs_free_stripe_hash_table(fs_info);
3711 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3712 root->orphan_block_rsv = NULL;
3715 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3719 struct inode *btree_inode = buf->pages[0]->mapping->host;
3721 ret = extent_buffer_uptodate(buf);
3725 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3726 parent_transid, atomic);
3732 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3734 return set_extent_buffer_uptodate(buf);
3737 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3739 struct btrfs_root *root;
3740 u64 transid = btrfs_header_generation(buf);
3743 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3745 * This is a fast path so only do this check if we have sanity tests
3746 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3747 * outside of the sanity tests.
3749 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3752 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3753 btrfs_assert_tree_locked(buf);
3754 if (transid != root->fs_info->generation)
3755 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3756 "found %llu running %llu\n",
3757 buf->start, transid, root->fs_info->generation);
3758 was_dirty = set_extent_buffer_dirty(buf);
3760 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3762 root->fs_info->dirty_metadata_batch);
3763 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3764 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
3765 btrfs_print_leaf(root, buf);
3771 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3775 * looks as though older kernels can get into trouble with
3776 * this code, they end up stuck in balance_dirty_pages forever
3780 if (current->flags & PF_MEMALLOC)
3784 btrfs_balance_delayed_items(root);
3786 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3787 BTRFS_DIRTY_METADATA_THRESH);
3789 balance_dirty_pages_ratelimited(
3790 root->fs_info->btree_inode->i_mapping);
3795 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3797 __btrfs_btree_balance_dirty(root, 1);
3800 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3802 __btrfs_btree_balance_dirty(root, 0);
3805 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3807 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3808 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3811 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3815 * Placeholder for checks
3820 static void btrfs_error_commit_super(struct btrfs_root *root)
3822 mutex_lock(&root->fs_info->cleaner_mutex);
3823 btrfs_run_delayed_iputs(root);
3824 mutex_unlock(&root->fs_info->cleaner_mutex);
3826 down_write(&root->fs_info->cleanup_work_sem);
3827 up_write(&root->fs_info->cleanup_work_sem);
3829 /* cleanup FS via transaction */
3830 btrfs_cleanup_transaction(root);
3833 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3835 struct btrfs_ordered_extent *ordered;
3837 spin_lock(&root->ordered_extent_lock);
3839 * This will just short circuit the ordered completion stuff which will
3840 * make sure the ordered extent gets properly cleaned up.
3842 list_for_each_entry(ordered, &root->ordered_extents,
3844 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3845 spin_unlock(&root->ordered_extent_lock);
3848 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3850 struct btrfs_root *root;
3851 struct list_head splice;
3853 INIT_LIST_HEAD(&splice);
3855 spin_lock(&fs_info->ordered_root_lock);
3856 list_splice_init(&fs_info->ordered_roots, &splice);
3857 while (!list_empty(&splice)) {
3858 root = list_first_entry(&splice, struct btrfs_root,
3860 list_move_tail(&root->ordered_root,
3861 &fs_info->ordered_roots);
3863 spin_unlock(&fs_info->ordered_root_lock);
3864 btrfs_destroy_ordered_extents(root);
3867 spin_lock(&fs_info->ordered_root_lock);
3869 spin_unlock(&fs_info->ordered_root_lock);
3872 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3873 struct btrfs_root *root)
3875 struct rb_node *node;
3876 struct btrfs_delayed_ref_root *delayed_refs;
3877 struct btrfs_delayed_ref_node *ref;
3880 delayed_refs = &trans->delayed_refs;
3882 spin_lock(&delayed_refs->lock);
3883 if (atomic_read(&delayed_refs->num_entries) == 0) {
3884 spin_unlock(&delayed_refs->lock);
3885 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3889 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3890 struct btrfs_delayed_ref_head *head;
3891 bool pin_bytes = false;
3893 head = rb_entry(node, struct btrfs_delayed_ref_head,
3895 if (!mutex_trylock(&head->mutex)) {
3896 atomic_inc(&head->node.refs);
3897 spin_unlock(&delayed_refs->lock);
3899 mutex_lock(&head->mutex);
3900 mutex_unlock(&head->mutex);
3901 btrfs_put_delayed_ref(&head->node);
3902 spin_lock(&delayed_refs->lock);
3905 spin_lock(&head->lock);
3906 while ((node = rb_first(&head->ref_root)) != NULL) {
3907 ref = rb_entry(node, struct btrfs_delayed_ref_node,
3910 rb_erase(&ref->rb_node, &head->ref_root);
3911 atomic_dec(&delayed_refs->num_entries);
3912 btrfs_put_delayed_ref(ref);
3914 if (head->must_insert_reserved)
3916 btrfs_free_delayed_extent_op(head->extent_op);
3917 delayed_refs->num_heads--;
3918 if (head->processing == 0)
3919 delayed_refs->num_heads_ready--;
3920 atomic_dec(&delayed_refs->num_entries);
3921 head->node.in_tree = 0;
3922 rb_erase(&head->href_node, &delayed_refs->href_root);
3923 spin_unlock(&head->lock);
3924 spin_unlock(&delayed_refs->lock);
3925 mutex_unlock(&head->mutex);
3928 btrfs_pin_extent(root, head->node.bytenr,
3929 head->node.num_bytes, 1);
3930 btrfs_put_delayed_ref(&head->node);
3932 spin_lock(&delayed_refs->lock);
3935 spin_unlock(&delayed_refs->lock);
3940 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3942 struct btrfs_inode *btrfs_inode;
3943 struct list_head splice;
3945 INIT_LIST_HEAD(&splice);
3947 spin_lock(&root->delalloc_lock);
3948 list_splice_init(&root->delalloc_inodes, &splice);
3950 while (!list_empty(&splice)) {
3951 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3954 list_del_init(&btrfs_inode->delalloc_inodes);
3955 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3956 &btrfs_inode->runtime_flags);
3957 spin_unlock(&root->delalloc_lock);
3959 btrfs_invalidate_inodes(btrfs_inode->root);
3961 spin_lock(&root->delalloc_lock);
3964 spin_unlock(&root->delalloc_lock);
3967 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3969 struct btrfs_root *root;
3970 struct list_head splice;
3972 INIT_LIST_HEAD(&splice);
3974 spin_lock(&fs_info->delalloc_root_lock);
3975 list_splice_init(&fs_info->delalloc_roots, &splice);
3976 while (!list_empty(&splice)) {
3977 root = list_first_entry(&splice, struct btrfs_root,
3979 list_del_init(&root->delalloc_root);
3980 root = btrfs_grab_fs_root(root);
3982 spin_unlock(&fs_info->delalloc_root_lock);
3984 btrfs_destroy_delalloc_inodes(root);
3985 btrfs_put_fs_root(root);
3987 spin_lock(&fs_info->delalloc_root_lock);
3989 spin_unlock(&fs_info->delalloc_root_lock);
3992 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3993 struct extent_io_tree *dirty_pages,
3997 struct extent_buffer *eb;
4002 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4007 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
4008 while (start <= end) {
4009 eb = btrfs_find_tree_block(root, start,
4011 start += root->nodesize;
4014 wait_on_extent_buffer_writeback(eb);
4016 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4018 clear_extent_buffer_dirty(eb);
4019 free_extent_buffer_stale(eb);
4026 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4027 struct extent_io_tree *pinned_extents)
4029 struct extent_io_tree *unpin;
4035 unpin = pinned_extents;
4038 ret = find_first_extent_bit(unpin, 0, &start, &end,
4039 EXTENT_DIRTY, NULL);
4044 if (btrfs_test_opt(root, DISCARD))
4045 ret = btrfs_error_discard_extent(root, start,
4049 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4050 btrfs_error_unpin_extent_range(root, start, end);
4055 if (unpin == &root->fs_info->freed_extents[0])
4056 unpin = &root->fs_info->freed_extents[1];
4058 unpin = &root->fs_info->freed_extents[0];
4066 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4067 struct btrfs_root *root)
4069 btrfs_destroy_delayed_refs(cur_trans, root);
4071 cur_trans->state = TRANS_STATE_COMMIT_START;
4072 wake_up(&root->fs_info->transaction_blocked_wait);
4074 cur_trans->state = TRANS_STATE_UNBLOCKED;
4075 wake_up(&root->fs_info->transaction_wait);
4077 btrfs_destroy_delayed_inodes(root);
4078 btrfs_assert_delayed_root_empty(root);
4080 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4082 btrfs_destroy_pinned_extent(root,
4083 root->fs_info->pinned_extents);
4085 cur_trans->state =TRANS_STATE_COMPLETED;
4086 wake_up(&cur_trans->commit_wait);
4089 memset(cur_trans, 0, sizeof(*cur_trans));
4090 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4094 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4096 struct btrfs_transaction *t;
4098 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4100 spin_lock(&root->fs_info->trans_lock);
4101 while (!list_empty(&root->fs_info->trans_list)) {
4102 t = list_first_entry(&root->fs_info->trans_list,
4103 struct btrfs_transaction, list);
4104 if (t->state >= TRANS_STATE_COMMIT_START) {
4105 atomic_inc(&t->use_count);
4106 spin_unlock(&root->fs_info->trans_lock);
4107 btrfs_wait_for_commit(root, t->transid);
4108 btrfs_put_transaction(t);
4109 spin_lock(&root->fs_info->trans_lock);
4112 if (t == root->fs_info->running_transaction) {
4113 t->state = TRANS_STATE_COMMIT_DOING;
4114 spin_unlock(&root->fs_info->trans_lock);
4116 * We wait for 0 num_writers since we don't hold a trans
4117 * handle open currently for this transaction.
4119 wait_event(t->writer_wait,
4120 atomic_read(&t->num_writers) == 0);
4122 spin_unlock(&root->fs_info->trans_lock);
4124 btrfs_cleanup_one_transaction(t, root);
4126 spin_lock(&root->fs_info->trans_lock);
4127 if (t == root->fs_info->running_transaction)
4128 root->fs_info->running_transaction = NULL;
4129 list_del_init(&t->list);
4130 spin_unlock(&root->fs_info->trans_lock);
4132 btrfs_put_transaction(t);
4133 trace_btrfs_transaction_commit(root);
4134 spin_lock(&root->fs_info->trans_lock);
4136 spin_unlock(&root->fs_info->trans_lock);
4137 btrfs_destroy_all_ordered_extents(root->fs_info);
4138 btrfs_destroy_delayed_inodes(root);
4139 btrfs_assert_delayed_root_empty(root);
4140 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4141 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4142 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4147 static struct extent_io_ops btree_extent_io_ops = {
4148 .readpage_end_io_hook = btree_readpage_end_io_hook,
4149 .readpage_io_failed_hook = btree_io_failed_hook,
4150 .submit_bio_hook = btree_submit_bio_hook,
4151 /* note we're sharing with inode.c for the merge bio hook */
4152 .merge_bio_hook = btrfs_merge_bio_hook,