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/slab.h>
29 #include <linux/migrate.h>
30 #include <linux/ratelimit.h>
31 #include <linux/uuid.h>
32 #include <linux/semaphore.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
43 #include "free-space-cache.h"
44 #include "free-space-tree.h"
45 #include "inode-map.h"
46 #include "check-integrity.h"
47 #include "rcu-string.h"
48 #include "dev-replace.h"
52 #include "compression.h"
55 #include <asm/cpufeature.h>
58 #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN |\
59 BTRFS_HEADER_FLAG_RELOC |\
60 BTRFS_SUPER_FLAG_ERROR |\
61 BTRFS_SUPER_FLAG_SEEDING |\
62 BTRFS_SUPER_FLAG_METADUMP)
64 static const struct extent_io_ops btree_extent_io_ops;
65 static void end_workqueue_fn(struct btrfs_work *work);
66 static void free_fs_root(struct btrfs_root *root);
67 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
69 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
70 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
71 struct btrfs_root *root);
72 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
73 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
74 struct extent_io_tree *dirty_pages,
76 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
77 struct extent_io_tree *pinned_extents);
78 static int btrfs_cleanup_transaction(struct btrfs_root *root);
79 static void btrfs_error_commit_super(struct btrfs_root *root);
82 * btrfs_end_io_wq structs are used to do processing in task context when an IO
83 * is complete. This is used during reads to verify checksums, and it is used
84 * by writes to insert metadata for new file extents after IO is complete.
86 struct btrfs_end_io_wq {
90 struct btrfs_fs_info *info;
92 enum btrfs_wq_endio_type metadata;
93 struct list_head list;
94 struct btrfs_work work;
97 static struct kmem_cache *btrfs_end_io_wq_cache;
99 int __init btrfs_end_io_wq_init(void)
101 btrfs_end_io_wq_cache = kmem_cache_create("btrfs_end_io_wq",
102 sizeof(struct btrfs_end_io_wq),
106 if (!btrfs_end_io_wq_cache)
111 void btrfs_end_io_wq_exit(void)
113 kmem_cache_destroy(btrfs_end_io_wq_cache);
117 * async submit bios are used to offload expensive checksumming
118 * onto the worker threads. They checksum file and metadata bios
119 * just before they are sent down the IO stack.
121 struct async_submit_bio {
124 struct list_head list;
125 extent_submit_bio_hook_t *submit_bio_start;
126 extent_submit_bio_hook_t *submit_bio_done;
129 unsigned long bio_flags;
131 * bio_offset is optional, can be used if the pages in the bio
132 * can't tell us where in the file the bio should go
135 struct btrfs_work work;
140 * Lockdep class keys for extent_buffer->lock's in this root. For a given
141 * eb, the lockdep key is determined by the btrfs_root it belongs to and
142 * the level the eb occupies in the tree.
144 * Different roots are used for different purposes and may nest inside each
145 * other and they require separate keysets. As lockdep keys should be
146 * static, assign keysets according to the purpose of the root as indicated
147 * by btrfs_root->objectid. This ensures that all special purpose roots
148 * have separate keysets.
150 * Lock-nesting across peer nodes is always done with the immediate parent
151 * node locked thus preventing deadlock. As lockdep doesn't know this, use
152 * subclass to avoid triggering lockdep warning in such cases.
154 * The key is set by the readpage_end_io_hook after the buffer has passed
155 * csum validation but before the pages are unlocked. It is also set by
156 * btrfs_init_new_buffer on freshly allocated blocks.
158 * We also add a check to make sure the highest level of the tree is the
159 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
160 * needs update as well.
162 #ifdef CONFIG_DEBUG_LOCK_ALLOC
163 # if BTRFS_MAX_LEVEL != 8
167 static struct btrfs_lockdep_keyset {
168 u64 id; /* root objectid */
169 const char *name_stem; /* lock name stem */
170 char names[BTRFS_MAX_LEVEL + 1][20];
171 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
172 } btrfs_lockdep_keysets[] = {
173 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
174 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
175 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
176 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
177 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
178 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
179 { .id = BTRFS_QUOTA_TREE_OBJECTID, .name_stem = "quota" },
180 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
181 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
182 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
183 { .id = BTRFS_UUID_TREE_OBJECTID, .name_stem = "uuid" },
184 { .id = BTRFS_FREE_SPACE_TREE_OBJECTID, .name_stem = "free-space" },
185 { .id = 0, .name_stem = "tree" },
188 void __init btrfs_init_lockdep(void)
192 /* initialize lockdep class names */
193 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
194 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
196 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
197 snprintf(ks->names[j], sizeof(ks->names[j]),
198 "btrfs-%s-%02d", ks->name_stem, j);
202 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
205 struct btrfs_lockdep_keyset *ks;
207 BUG_ON(level >= ARRAY_SIZE(ks->keys));
209 /* find the matching keyset, id 0 is the default entry */
210 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
211 if (ks->id == objectid)
214 lockdep_set_class_and_name(&eb->lock,
215 &ks->keys[level], ks->names[level]);
221 * extents on the btree inode are pretty simple, there's one extent
222 * that covers the entire device
224 static struct extent_map *btree_get_extent(struct inode *inode,
225 struct page *page, size_t pg_offset, u64 start, u64 len,
228 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
229 struct extent_map *em;
232 read_lock(&em_tree->lock);
233 em = lookup_extent_mapping(em_tree, start, len);
236 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
237 read_unlock(&em_tree->lock);
240 read_unlock(&em_tree->lock);
242 em = alloc_extent_map();
244 em = ERR_PTR(-ENOMEM);
249 em->block_len = (u64)-1;
251 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
253 write_lock(&em_tree->lock);
254 ret = add_extent_mapping(em_tree, em, 0);
255 if (ret == -EEXIST) {
257 em = lookup_extent_mapping(em_tree, start, len);
264 write_unlock(&em_tree->lock);
270 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
272 return btrfs_crc32c(seed, data, len);
275 void btrfs_csum_final(u32 crc, char *result)
277 put_unaligned_le32(~crc, result);
281 * compute the csum for a btree block, and either verify it or write it
282 * into the csum field of the block.
284 static int csum_tree_block(struct btrfs_fs_info *fs_info,
285 struct extent_buffer *buf,
288 u16 csum_size = btrfs_super_csum_size(fs_info->super_copy);
291 unsigned long cur_len;
292 unsigned long offset = BTRFS_CSUM_SIZE;
294 unsigned long map_start;
295 unsigned long map_len;
298 unsigned long inline_result;
300 len = buf->len - offset;
302 err = map_private_extent_buffer(buf, offset, 32,
303 &kaddr, &map_start, &map_len);
306 cur_len = min(len, map_len - (offset - map_start));
307 crc = btrfs_csum_data(kaddr + offset - map_start,
312 if (csum_size > sizeof(inline_result)) {
313 result = kzalloc(csum_size, GFP_NOFS);
317 result = (char *)&inline_result;
320 btrfs_csum_final(crc, result);
323 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
326 memcpy(&found, result, csum_size);
328 read_extent_buffer(buf, &val, 0, csum_size);
329 btrfs_warn_rl(fs_info,
330 "%s checksum verify failed on %llu wanted %X found %X "
332 fs_info->sb->s_id, buf->start,
333 val, found, btrfs_header_level(buf));
334 if (result != (char *)&inline_result)
339 write_extent_buffer(buf, result, 0, csum_size);
341 if (result != (char *)&inline_result)
347 * we can't consider a given block up to date unless the transid of the
348 * block matches the transid in the parent node's pointer. This is how we
349 * detect blocks that either didn't get written at all or got written
350 * in the wrong place.
352 static int verify_parent_transid(struct extent_io_tree *io_tree,
353 struct extent_buffer *eb, u64 parent_transid,
356 struct extent_state *cached_state = NULL;
358 bool need_lock = (current->journal_info == BTRFS_SEND_TRANS_STUB);
360 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
367 btrfs_tree_read_lock(eb);
368 btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
371 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
373 if (extent_buffer_uptodate(eb) &&
374 btrfs_header_generation(eb) == parent_transid) {
378 btrfs_err_rl(eb->fs_info,
379 "parent transid verify failed on %llu wanted %llu found %llu",
381 parent_transid, btrfs_header_generation(eb));
385 * Things reading via commit roots that don't have normal protection,
386 * like send, can have a really old block in cache that may point at a
387 * block that has been freed and re-allocated. So don't clear uptodate
388 * if we find an eb that is under IO (dirty/writeback) because we could
389 * end up reading in the stale data and then writing it back out and
390 * making everybody very sad.
392 if (!extent_buffer_under_io(eb))
393 clear_extent_buffer_uptodate(eb);
395 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
396 &cached_state, GFP_NOFS);
398 btrfs_tree_read_unlock_blocking(eb);
403 * Return 0 if the superblock checksum type matches the checksum value of that
404 * algorithm. Pass the raw disk superblock data.
406 static int btrfs_check_super_csum(char *raw_disk_sb)
408 struct btrfs_super_block *disk_sb =
409 (struct btrfs_super_block *)raw_disk_sb;
410 u16 csum_type = btrfs_super_csum_type(disk_sb);
413 if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
415 const int csum_size = sizeof(crc);
416 char result[csum_size];
419 * The super_block structure does not span the whole
420 * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
421 * is filled with zeros and is included in the checksum.
423 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
424 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
425 btrfs_csum_final(crc, result);
427 if (memcmp(raw_disk_sb, result, csum_size))
431 if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
432 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
441 * helper to read a given tree block, doing retries as required when
442 * the checksums don't match and we have alternate mirrors to try.
444 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
445 struct extent_buffer *eb,
446 u64 start, u64 parent_transid)
448 struct extent_io_tree *io_tree;
453 int failed_mirror = 0;
455 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
456 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
458 ret = read_extent_buffer_pages(io_tree, eb, start,
460 btree_get_extent, mirror_num);
462 if (!verify_parent_transid(io_tree, eb,
470 * This buffer's crc is fine, but its contents are corrupted, so
471 * there is no reason to read the other copies, they won't be
474 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
477 num_copies = btrfs_num_copies(root->fs_info,
482 if (!failed_mirror) {
484 failed_mirror = eb->read_mirror;
488 if (mirror_num == failed_mirror)
491 if (mirror_num > num_copies)
495 if (failed && !ret && failed_mirror)
496 repair_eb_io_failure(root, eb, failed_mirror);
502 * checksum a dirty tree block before IO. This has extra checks to make sure
503 * we only fill in the checksum field in the first page of a multi-page block
506 static int csum_dirty_buffer(struct btrfs_fs_info *fs_info, struct page *page)
508 u64 start = page_offset(page);
510 struct extent_buffer *eb;
512 eb = (struct extent_buffer *)page->private;
513 if (page != eb->pages[0])
516 found_start = btrfs_header_bytenr(eb);
518 * Please do not consolidate these warnings into a single if.
519 * It is useful to know what went wrong.
521 if (WARN_ON(found_start != start))
523 if (WARN_ON(!PageUptodate(page)))
526 ASSERT(memcmp_extent_buffer(eb, fs_info->fsid,
527 btrfs_header_fsid(), BTRFS_FSID_SIZE) == 0);
529 return csum_tree_block(fs_info, eb, 0);
532 static int check_tree_block_fsid(struct btrfs_fs_info *fs_info,
533 struct extent_buffer *eb)
535 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
536 u8 fsid[BTRFS_UUID_SIZE];
539 read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
541 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
545 fs_devices = fs_devices->seed;
550 #define CORRUPT(reason, eb, root, slot) \
551 btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu," \
552 "root=%llu, slot=%d", reason, \
553 btrfs_header_bytenr(eb), root->objectid, slot)
555 static noinline int check_leaf(struct btrfs_root *root,
556 struct extent_buffer *leaf)
558 struct btrfs_key key;
559 struct btrfs_key leaf_key;
560 u32 nritems = btrfs_header_nritems(leaf);
564 struct btrfs_root *check_root;
566 key.objectid = btrfs_header_owner(leaf);
567 key.type = BTRFS_ROOT_ITEM_KEY;
568 key.offset = (u64)-1;
570 check_root = btrfs_get_fs_root(root->fs_info, &key, false);
572 * The only reason we also check NULL here is that during
573 * open_ctree() some roots has not yet been set up.
575 if (!IS_ERR_OR_NULL(check_root)) {
576 /* if leaf is the root, then it's fine */
578 btrfs_root_bytenr(&check_root->root_item)) {
579 CORRUPT("non-root leaf's nritems is 0",
587 /* Check the 0 item */
588 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
589 BTRFS_LEAF_DATA_SIZE(root)) {
590 CORRUPT("invalid item offset size pair", leaf, root, 0);
595 * Check to make sure each items keys are in the correct order and their
596 * offsets make sense. We only have to loop through nritems-1 because
597 * we check the current slot against the next slot, which verifies the
598 * next slot's offset+size makes sense and that the current's slot
601 for (slot = 0; slot < nritems - 1; slot++) {
602 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
603 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
605 /* Make sure the keys are in the right order */
606 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
607 CORRUPT("bad key order", leaf, root, slot);
612 * Make sure the offset and ends are right, remember that the
613 * item data starts at the end of the leaf and grows towards the
616 if (btrfs_item_offset_nr(leaf, slot) !=
617 btrfs_item_end_nr(leaf, slot + 1)) {
618 CORRUPT("slot offset bad", leaf, root, slot);
623 * Check to make sure that we don't point outside of the leaf,
624 * just in case all the items are consistent to each other, but
625 * all point outside of the leaf.
627 if (btrfs_item_end_nr(leaf, slot) >
628 BTRFS_LEAF_DATA_SIZE(root)) {
629 CORRUPT("slot end outside of leaf", leaf, root, slot);
637 static int check_node(struct btrfs_root *root, struct extent_buffer *node)
639 unsigned long nr = btrfs_header_nritems(node);
641 if (nr == 0 || nr > BTRFS_NODEPTRS_PER_BLOCK(root)) {
642 btrfs_crit(root->fs_info,
643 "corrupt node: block %llu root %llu nritems %lu",
644 node->start, root->objectid, nr);
650 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
651 u64 phy_offset, struct page *page,
652 u64 start, u64 end, int mirror)
656 struct extent_buffer *eb;
657 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
658 struct btrfs_fs_info *fs_info = root->fs_info;
665 eb = (struct extent_buffer *)page->private;
667 /* the pending IO might have been the only thing that kept this buffer
668 * in memory. Make sure we have a ref for all this other checks
670 extent_buffer_get(eb);
672 reads_done = atomic_dec_and_test(&eb->io_pages);
676 eb->read_mirror = mirror;
677 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
682 found_start = btrfs_header_bytenr(eb);
683 if (found_start != eb->start) {
684 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
685 found_start, eb->start);
689 if (check_tree_block_fsid(fs_info, eb)) {
690 btrfs_err_rl(fs_info, "bad fsid on block %llu",
695 found_level = btrfs_header_level(eb);
696 if (found_level >= BTRFS_MAX_LEVEL) {
697 btrfs_err(fs_info, "bad tree block level %d",
698 (int)btrfs_header_level(eb));
703 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
706 ret = csum_tree_block(fs_info, eb, 1);
711 * If this is a leaf block and it is corrupt, set the corrupt bit so
712 * that we don't try and read the other copies of this block, just
715 if (found_level == 0 && check_leaf(root, eb)) {
716 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
720 if (found_level > 0 && check_node(root, eb))
724 set_extent_buffer_uptodate(eb);
727 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
728 btree_readahead_hook(fs_info, eb, eb->start, ret);
732 * our io error hook is going to dec the io pages
733 * again, we have to make sure it has something
736 atomic_inc(&eb->io_pages);
737 clear_extent_buffer_uptodate(eb);
739 free_extent_buffer(eb);
744 static int btree_io_failed_hook(struct page *page, int failed_mirror)
746 struct extent_buffer *eb;
748 eb = (struct extent_buffer *)page->private;
749 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
750 eb->read_mirror = failed_mirror;
751 atomic_dec(&eb->io_pages);
752 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
753 btree_readahead_hook(eb->fs_info, eb, eb->start, -EIO);
754 return -EIO; /* we fixed nothing */
757 static void end_workqueue_bio(struct bio *bio)
759 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
760 struct btrfs_fs_info *fs_info;
761 struct btrfs_workqueue *wq;
762 btrfs_work_func_t func;
764 fs_info = end_io_wq->info;
765 end_io_wq->error = bio->bi_error;
767 if (bio->bi_rw & REQ_WRITE) {
768 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
769 wq = fs_info->endio_meta_write_workers;
770 func = btrfs_endio_meta_write_helper;
771 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
772 wq = fs_info->endio_freespace_worker;
773 func = btrfs_freespace_write_helper;
774 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
775 wq = fs_info->endio_raid56_workers;
776 func = btrfs_endio_raid56_helper;
778 wq = fs_info->endio_write_workers;
779 func = btrfs_endio_write_helper;
782 if (unlikely(end_io_wq->metadata ==
783 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
784 wq = fs_info->endio_repair_workers;
785 func = btrfs_endio_repair_helper;
786 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
787 wq = fs_info->endio_raid56_workers;
788 func = btrfs_endio_raid56_helper;
789 } else if (end_io_wq->metadata) {
790 wq = fs_info->endio_meta_workers;
791 func = btrfs_endio_meta_helper;
793 wq = fs_info->endio_workers;
794 func = btrfs_endio_helper;
798 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
799 btrfs_queue_work(wq, &end_io_wq->work);
802 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
803 enum btrfs_wq_endio_type metadata)
805 struct btrfs_end_io_wq *end_io_wq;
807 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
811 end_io_wq->private = bio->bi_private;
812 end_io_wq->end_io = bio->bi_end_io;
813 end_io_wq->info = info;
814 end_io_wq->error = 0;
815 end_io_wq->bio = bio;
816 end_io_wq->metadata = metadata;
818 bio->bi_private = end_io_wq;
819 bio->bi_end_io = end_workqueue_bio;
823 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
825 unsigned long limit = min_t(unsigned long,
826 info->thread_pool_size,
827 info->fs_devices->open_devices);
831 static void run_one_async_start(struct btrfs_work *work)
833 struct async_submit_bio *async;
836 async = container_of(work, struct async_submit_bio, work);
837 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
838 async->mirror_num, async->bio_flags,
844 static void run_one_async_done(struct btrfs_work *work)
846 struct btrfs_fs_info *fs_info;
847 struct async_submit_bio *async;
850 async = container_of(work, struct async_submit_bio, work);
851 fs_info = BTRFS_I(async->inode)->root->fs_info;
853 limit = btrfs_async_submit_limit(fs_info);
854 limit = limit * 2 / 3;
857 * atomic_dec_return implies a barrier for waitqueue_active
859 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
860 waitqueue_active(&fs_info->async_submit_wait))
861 wake_up(&fs_info->async_submit_wait);
863 /* If an error occurred we just want to clean up the bio and move on */
865 async->bio->bi_error = async->error;
866 bio_endio(async->bio);
870 async->submit_bio_done(async->inode, async->rw, async->bio,
871 async->mirror_num, async->bio_flags,
875 static void run_one_async_free(struct btrfs_work *work)
877 struct async_submit_bio *async;
879 async = container_of(work, struct async_submit_bio, work);
883 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
884 int rw, struct bio *bio, int mirror_num,
885 unsigned long bio_flags,
887 extent_submit_bio_hook_t *submit_bio_start,
888 extent_submit_bio_hook_t *submit_bio_done)
890 struct async_submit_bio *async;
892 async = kmalloc(sizeof(*async), GFP_NOFS);
896 async->inode = inode;
899 async->mirror_num = mirror_num;
900 async->submit_bio_start = submit_bio_start;
901 async->submit_bio_done = submit_bio_done;
903 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
904 run_one_async_done, run_one_async_free);
906 async->bio_flags = bio_flags;
907 async->bio_offset = bio_offset;
911 atomic_inc(&fs_info->nr_async_submits);
914 btrfs_set_work_high_priority(&async->work);
916 btrfs_queue_work(fs_info->workers, &async->work);
918 while (atomic_read(&fs_info->async_submit_draining) &&
919 atomic_read(&fs_info->nr_async_submits)) {
920 wait_event(fs_info->async_submit_wait,
921 (atomic_read(&fs_info->nr_async_submits) == 0));
927 static int btree_csum_one_bio(struct bio *bio)
929 struct bio_vec *bvec;
930 struct btrfs_root *root;
933 bio_for_each_segment_all(bvec, bio, i) {
934 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
935 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
943 static int __btree_submit_bio_start(struct inode *inode, int rw,
944 struct bio *bio, int mirror_num,
945 unsigned long bio_flags,
949 * when we're called for a write, we're already in the async
950 * submission context. Just jump into btrfs_map_bio
952 return btree_csum_one_bio(bio);
955 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
956 int mirror_num, unsigned long bio_flags,
962 * when we're called for a write, we're already in the async
963 * submission context. Just jump into btrfs_map_bio
965 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
973 static int check_async_write(struct inode *inode, unsigned long bio_flags)
975 if (bio_flags & EXTENT_BIO_TREE_LOG)
978 if (static_cpu_has(X86_FEATURE_XMM4_2))
984 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
985 int mirror_num, unsigned long bio_flags,
988 int async = check_async_write(inode, bio_flags);
991 if (!(rw & REQ_WRITE)) {
993 * called for a read, do the setup so that checksum validation
994 * can happen in the async kernel threads
996 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
997 bio, BTRFS_WQ_ENDIO_METADATA);
1000 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
1002 } else if (!async) {
1003 ret = btree_csum_one_bio(bio);
1006 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
1010 * kthread helpers are used to submit writes so that
1011 * checksumming can happen in parallel across all CPUs
1013 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1014 inode, rw, bio, mirror_num, 0,
1016 __btree_submit_bio_start,
1017 __btree_submit_bio_done);
1025 bio->bi_error = ret;
1030 #ifdef CONFIG_MIGRATION
1031 static int btree_migratepage(struct address_space *mapping,
1032 struct page *newpage, struct page *page,
1033 enum migrate_mode mode)
1036 * we can't safely write a btree page from here,
1037 * we haven't done the locking hook
1039 if (PageDirty(page))
1042 * Buffers may be managed in a filesystem specific way.
1043 * We must have no buffers or drop them.
1045 if (page_has_private(page) &&
1046 !try_to_release_page(page, GFP_KERNEL))
1048 return migrate_page(mapping, newpage, page, mode);
1053 static int btree_writepages(struct address_space *mapping,
1054 struct writeback_control *wbc)
1056 struct btrfs_fs_info *fs_info;
1059 if (wbc->sync_mode == WB_SYNC_NONE) {
1061 if (wbc->for_kupdate)
1064 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1065 /* this is a bit racy, but that's ok */
1066 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1067 BTRFS_DIRTY_METADATA_THRESH);
1071 return btree_write_cache_pages(mapping, wbc);
1074 static int btree_readpage(struct file *file, struct page *page)
1076 struct extent_io_tree *tree;
1077 tree = &BTRFS_I(page->mapping->host)->io_tree;
1078 return extent_read_full_page(tree, page, btree_get_extent, 0);
1081 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1083 if (PageWriteback(page) || PageDirty(page))
1086 return try_release_extent_buffer(page);
1089 static void btree_invalidatepage(struct page *page, unsigned int offset,
1090 unsigned int length)
1092 struct extent_io_tree *tree;
1093 tree = &BTRFS_I(page->mapping->host)->io_tree;
1094 extent_invalidatepage(tree, page, offset);
1095 btree_releasepage(page, GFP_NOFS);
1096 if (PagePrivate(page)) {
1097 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1098 "page private not zero on page %llu",
1099 (unsigned long long)page_offset(page));
1100 ClearPagePrivate(page);
1101 set_page_private(page, 0);
1106 static int btree_set_page_dirty(struct page *page)
1109 struct extent_buffer *eb;
1111 BUG_ON(!PagePrivate(page));
1112 eb = (struct extent_buffer *)page->private;
1114 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1115 BUG_ON(!atomic_read(&eb->refs));
1116 btrfs_assert_tree_locked(eb);
1118 return __set_page_dirty_nobuffers(page);
1121 static const struct address_space_operations btree_aops = {
1122 .readpage = btree_readpage,
1123 .writepages = btree_writepages,
1124 .releasepage = btree_releasepage,
1125 .invalidatepage = btree_invalidatepage,
1126 #ifdef CONFIG_MIGRATION
1127 .migratepage = btree_migratepage,
1129 .set_page_dirty = btree_set_page_dirty,
1132 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1134 struct extent_buffer *buf = NULL;
1135 struct inode *btree_inode = root->fs_info->btree_inode;
1137 buf = btrfs_find_create_tree_block(root, bytenr);
1140 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1141 buf, 0, WAIT_NONE, btree_get_extent, 0);
1142 free_extent_buffer(buf);
1145 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1146 int mirror_num, struct extent_buffer **eb)
1148 struct extent_buffer *buf = NULL;
1149 struct inode *btree_inode = root->fs_info->btree_inode;
1150 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1153 buf = btrfs_find_create_tree_block(root, bytenr);
1157 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1159 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1160 btree_get_extent, mirror_num);
1162 free_extent_buffer(buf);
1166 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1167 free_extent_buffer(buf);
1169 } else if (extent_buffer_uptodate(buf)) {
1172 free_extent_buffer(buf);
1177 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1180 return find_extent_buffer(fs_info, bytenr);
1183 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1186 if (btrfs_is_testing(root->fs_info))
1187 return alloc_test_extent_buffer(root->fs_info, bytenr,
1189 return alloc_extent_buffer(root->fs_info, bytenr);
1193 int btrfs_write_tree_block(struct extent_buffer *buf)
1195 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1196 buf->start + buf->len - 1);
1199 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1201 return filemap_fdatawait_range(buf->pages[0]->mapping,
1202 buf->start, buf->start + buf->len - 1);
1205 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1208 struct extent_buffer *buf = NULL;
1211 buf = btrfs_find_create_tree_block(root, bytenr);
1215 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1217 free_extent_buffer(buf);
1218 return ERR_PTR(ret);
1224 void clean_tree_block(struct btrfs_trans_handle *trans,
1225 struct btrfs_fs_info *fs_info,
1226 struct extent_buffer *buf)
1228 if (btrfs_header_generation(buf) ==
1229 fs_info->running_transaction->transid) {
1230 btrfs_assert_tree_locked(buf);
1232 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1233 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1235 fs_info->dirty_metadata_batch);
1236 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1237 btrfs_set_lock_blocking(buf);
1238 clear_extent_buffer_dirty(buf);
1243 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1245 struct btrfs_subvolume_writers *writers;
1248 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1250 return ERR_PTR(-ENOMEM);
1252 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1255 return ERR_PTR(ret);
1258 init_waitqueue_head(&writers->wait);
1263 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1265 percpu_counter_destroy(&writers->counter);
1269 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1270 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1273 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1275 root->commit_root = NULL;
1276 root->sectorsize = sectorsize;
1277 root->nodesize = nodesize;
1278 root->stripesize = stripesize;
1280 root->orphan_cleanup_state = 0;
1282 root->objectid = objectid;
1283 root->last_trans = 0;
1284 root->highest_objectid = 0;
1285 root->nr_delalloc_inodes = 0;
1286 root->nr_ordered_extents = 0;
1288 root->inode_tree = RB_ROOT;
1289 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1290 root->block_rsv = NULL;
1291 root->orphan_block_rsv = NULL;
1293 INIT_LIST_HEAD(&root->dirty_list);
1294 INIT_LIST_HEAD(&root->root_list);
1295 INIT_LIST_HEAD(&root->delalloc_inodes);
1296 INIT_LIST_HEAD(&root->delalloc_root);
1297 INIT_LIST_HEAD(&root->ordered_extents);
1298 INIT_LIST_HEAD(&root->ordered_root);
1299 INIT_LIST_HEAD(&root->logged_list[0]);
1300 INIT_LIST_HEAD(&root->logged_list[1]);
1301 spin_lock_init(&root->orphan_lock);
1302 spin_lock_init(&root->inode_lock);
1303 spin_lock_init(&root->delalloc_lock);
1304 spin_lock_init(&root->ordered_extent_lock);
1305 spin_lock_init(&root->accounting_lock);
1306 spin_lock_init(&root->log_extents_lock[0]);
1307 spin_lock_init(&root->log_extents_lock[1]);
1308 mutex_init(&root->objectid_mutex);
1309 mutex_init(&root->log_mutex);
1310 mutex_init(&root->ordered_extent_mutex);
1311 mutex_init(&root->delalloc_mutex);
1312 init_waitqueue_head(&root->log_writer_wait);
1313 init_waitqueue_head(&root->log_commit_wait[0]);
1314 init_waitqueue_head(&root->log_commit_wait[1]);
1315 INIT_LIST_HEAD(&root->log_ctxs[0]);
1316 INIT_LIST_HEAD(&root->log_ctxs[1]);
1317 atomic_set(&root->log_commit[0], 0);
1318 atomic_set(&root->log_commit[1], 0);
1319 atomic_set(&root->log_writers, 0);
1320 atomic_set(&root->log_batch, 0);
1321 atomic_set(&root->orphan_inodes, 0);
1322 atomic_set(&root->refs, 1);
1323 atomic_set(&root->will_be_snapshoted, 0);
1324 atomic_set(&root->qgroup_meta_rsv, 0);
1325 root->log_transid = 0;
1326 root->log_transid_committed = -1;
1327 root->last_log_commit = 0;
1329 extent_io_tree_init(&root->dirty_log_pages,
1330 fs_info->btree_inode->i_mapping);
1332 memset(&root->root_key, 0, sizeof(root->root_key));
1333 memset(&root->root_item, 0, sizeof(root->root_item));
1334 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1336 root->defrag_trans_start = fs_info->generation;
1338 root->defrag_trans_start = 0;
1339 root->root_key.objectid = objectid;
1342 spin_lock_init(&root->root_item_lock);
1345 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1348 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1350 root->fs_info = fs_info;
1354 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1355 /* Should only be used by the testing infrastructure */
1356 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info,
1357 u32 sectorsize, u32 nodesize)
1359 struct btrfs_root *root;
1362 return ERR_PTR(-EINVAL);
1364 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1366 return ERR_PTR(-ENOMEM);
1367 /* We don't use the stripesize in selftest, set it as sectorsize */
1368 __setup_root(nodesize, sectorsize, sectorsize, root, fs_info,
1369 BTRFS_ROOT_TREE_OBJECTID);
1370 root->alloc_bytenr = 0;
1376 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1377 struct btrfs_fs_info *fs_info,
1380 struct extent_buffer *leaf;
1381 struct btrfs_root *tree_root = fs_info->tree_root;
1382 struct btrfs_root *root;
1383 struct btrfs_key key;
1387 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1389 return ERR_PTR(-ENOMEM);
1391 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1392 tree_root->stripesize, root, fs_info, objectid);
1393 root->root_key.objectid = objectid;
1394 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1395 root->root_key.offset = 0;
1397 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1399 ret = PTR_ERR(leaf);
1404 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1405 btrfs_set_header_bytenr(leaf, leaf->start);
1406 btrfs_set_header_generation(leaf, trans->transid);
1407 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1408 btrfs_set_header_owner(leaf, objectid);
1411 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1413 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1414 btrfs_header_chunk_tree_uuid(leaf),
1416 btrfs_mark_buffer_dirty(leaf);
1418 root->commit_root = btrfs_root_node(root);
1419 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1421 root->root_item.flags = 0;
1422 root->root_item.byte_limit = 0;
1423 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1424 btrfs_set_root_generation(&root->root_item, trans->transid);
1425 btrfs_set_root_level(&root->root_item, 0);
1426 btrfs_set_root_refs(&root->root_item, 1);
1427 btrfs_set_root_used(&root->root_item, leaf->len);
1428 btrfs_set_root_last_snapshot(&root->root_item, 0);
1429 btrfs_set_root_dirid(&root->root_item, 0);
1431 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1432 root->root_item.drop_level = 0;
1434 key.objectid = objectid;
1435 key.type = BTRFS_ROOT_ITEM_KEY;
1437 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1441 btrfs_tree_unlock(leaf);
1447 btrfs_tree_unlock(leaf);
1448 free_extent_buffer(root->commit_root);
1449 free_extent_buffer(leaf);
1453 return ERR_PTR(ret);
1456 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1457 struct btrfs_fs_info *fs_info)
1459 struct btrfs_root *root;
1460 struct btrfs_root *tree_root = fs_info->tree_root;
1461 struct extent_buffer *leaf;
1463 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1465 return ERR_PTR(-ENOMEM);
1467 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1468 tree_root->stripesize, root, fs_info,
1469 BTRFS_TREE_LOG_OBJECTID);
1471 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1472 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1473 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1476 * DON'T set REF_COWS for log trees
1478 * log trees do not get reference counted because they go away
1479 * before a real commit is actually done. They do store pointers
1480 * to file data extents, and those reference counts still get
1481 * updated (along with back refs to the log tree).
1484 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1488 return ERR_CAST(leaf);
1491 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1492 btrfs_set_header_bytenr(leaf, leaf->start);
1493 btrfs_set_header_generation(leaf, trans->transid);
1494 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1495 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1498 write_extent_buffer(root->node, root->fs_info->fsid,
1499 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1500 btrfs_mark_buffer_dirty(root->node);
1501 btrfs_tree_unlock(root->node);
1505 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1506 struct btrfs_fs_info *fs_info)
1508 struct btrfs_root *log_root;
1510 log_root = alloc_log_tree(trans, fs_info);
1511 if (IS_ERR(log_root))
1512 return PTR_ERR(log_root);
1513 WARN_ON(fs_info->log_root_tree);
1514 fs_info->log_root_tree = log_root;
1518 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1519 struct btrfs_root *root)
1521 struct btrfs_root *log_root;
1522 struct btrfs_inode_item *inode_item;
1524 log_root = alloc_log_tree(trans, root->fs_info);
1525 if (IS_ERR(log_root))
1526 return PTR_ERR(log_root);
1528 log_root->last_trans = trans->transid;
1529 log_root->root_key.offset = root->root_key.objectid;
1531 inode_item = &log_root->root_item.inode;
1532 btrfs_set_stack_inode_generation(inode_item, 1);
1533 btrfs_set_stack_inode_size(inode_item, 3);
1534 btrfs_set_stack_inode_nlink(inode_item, 1);
1535 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1536 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1538 btrfs_set_root_node(&log_root->root_item, log_root->node);
1540 WARN_ON(root->log_root);
1541 root->log_root = log_root;
1542 root->log_transid = 0;
1543 root->log_transid_committed = -1;
1544 root->last_log_commit = 0;
1548 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1549 struct btrfs_key *key)
1551 struct btrfs_root *root;
1552 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1553 struct btrfs_path *path;
1557 path = btrfs_alloc_path();
1559 return ERR_PTR(-ENOMEM);
1561 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1567 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1568 tree_root->stripesize, root, fs_info, key->objectid);
1570 ret = btrfs_find_root(tree_root, key, path,
1571 &root->root_item, &root->root_key);
1578 generation = btrfs_root_generation(&root->root_item);
1579 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1581 if (IS_ERR(root->node)) {
1582 ret = PTR_ERR(root->node);
1584 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1586 free_extent_buffer(root->node);
1589 root->commit_root = btrfs_root_node(root);
1591 btrfs_free_path(path);
1597 root = ERR_PTR(ret);
1601 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1602 struct btrfs_key *location)
1604 struct btrfs_root *root;
1606 root = btrfs_read_tree_root(tree_root, location);
1610 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1611 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1612 btrfs_check_and_init_root_item(&root->root_item);
1618 int btrfs_init_fs_root(struct btrfs_root *root)
1621 struct btrfs_subvolume_writers *writers;
1623 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1624 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1626 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1631 writers = btrfs_alloc_subvolume_writers();
1632 if (IS_ERR(writers)) {
1633 ret = PTR_ERR(writers);
1636 root->subv_writers = writers;
1638 btrfs_init_free_ino_ctl(root);
1639 spin_lock_init(&root->ino_cache_lock);
1640 init_waitqueue_head(&root->ino_cache_wait);
1642 ret = get_anon_bdev(&root->anon_dev);
1646 mutex_lock(&root->objectid_mutex);
1647 ret = btrfs_find_highest_objectid(root,
1648 &root->highest_objectid);
1650 mutex_unlock(&root->objectid_mutex);
1654 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1656 mutex_unlock(&root->objectid_mutex);
1660 /* the caller is responsible to call free_fs_root */
1664 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1667 struct btrfs_root *root;
1669 spin_lock(&fs_info->fs_roots_radix_lock);
1670 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1671 (unsigned long)root_id);
1672 spin_unlock(&fs_info->fs_roots_radix_lock);
1676 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1677 struct btrfs_root *root)
1681 ret = radix_tree_preload(GFP_NOFS);
1685 spin_lock(&fs_info->fs_roots_radix_lock);
1686 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1687 (unsigned long)root->root_key.objectid,
1690 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1691 spin_unlock(&fs_info->fs_roots_radix_lock);
1692 radix_tree_preload_end();
1697 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1698 struct btrfs_key *location,
1701 struct btrfs_root *root;
1702 struct btrfs_path *path;
1703 struct btrfs_key key;
1706 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1707 return fs_info->tree_root;
1708 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1709 return fs_info->extent_root;
1710 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1711 return fs_info->chunk_root;
1712 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1713 return fs_info->dev_root;
1714 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1715 return fs_info->csum_root;
1716 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1717 return fs_info->quota_root ? fs_info->quota_root :
1719 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1720 return fs_info->uuid_root ? fs_info->uuid_root :
1722 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1723 return fs_info->free_space_root ? fs_info->free_space_root :
1726 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1728 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1729 return ERR_PTR(-ENOENT);
1733 root = btrfs_read_fs_root(fs_info->tree_root, location);
1737 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1742 ret = btrfs_init_fs_root(root);
1746 path = btrfs_alloc_path();
1751 key.objectid = BTRFS_ORPHAN_OBJECTID;
1752 key.type = BTRFS_ORPHAN_ITEM_KEY;
1753 key.offset = location->objectid;
1755 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1756 btrfs_free_path(path);
1760 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1762 ret = btrfs_insert_fs_root(fs_info, root);
1764 if (ret == -EEXIST) {
1773 return ERR_PTR(ret);
1776 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1778 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1780 struct btrfs_device *device;
1781 struct backing_dev_info *bdi;
1784 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1787 bdi = blk_get_backing_dev_info(device->bdev);
1788 if (bdi_congested(bdi, bdi_bits)) {
1797 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1801 err = bdi_setup_and_register(bdi, "btrfs");
1805 bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1806 bdi->congested_fn = btrfs_congested_fn;
1807 bdi->congested_data = info;
1808 bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1813 * called by the kthread helper functions to finally call the bio end_io
1814 * functions. This is where read checksum verification actually happens
1816 static void end_workqueue_fn(struct btrfs_work *work)
1819 struct btrfs_end_io_wq *end_io_wq;
1821 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1822 bio = end_io_wq->bio;
1824 bio->bi_error = end_io_wq->error;
1825 bio->bi_private = end_io_wq->private;
1826 bio->bi_end_io = end_io_wq->end_io;
1827 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1831 static int cleaner_kthread(void *arg)
1833 struct btrfs_root *root = arg;
1835 struct btrfs_trans_handle *trans;
1840 /* Make the cleaner go to sleep early. */
1841 if (btrfs_need_cleaner_sleep(root))
1845 * Do not do anything if we might cause open_ctree() to block
1846 * before we have finished mounting the filesystem.
1848 if (!root->fs_info->open)
1851 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1855 * Avoid the problem that we change the status of the fs
1856 * during the above check and trylock.
1858 if (btrfs_need_cleaner_sleep(root)) {
1859 mutex_unlock(&root->fs_info->cleaner_mutex);
1863 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1864 btrfs_run_delayed_iputs(root);
1865 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1867 again = btrfs_clean_one_deleted_snapshot(root);
1868 mutex_unlock(&root->fs_info->cleaner_mutex);
1871 * The defragger has dealt with the R/O remount and umount,
1872 * needn't do anything special here.
1874 btrfs_run_defrag_inodes(root->fs_info);
1877 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1878 * with relocation (btrfs_relocate_chunk) and relocation
1879 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1880 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1881 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1882 * unused block groups.
1884 btrfs_delete_unused_bgs(root->fs_info);
1887 set_current_state(TASK_INTERRUPTIBLE);
1888 if (!kthread_should_stop())
1890 __set_current_state(TASK_RUNNING);
1892 } while (!kthread_should_stop());
1895 * Transaction kthread is stopped before us and wakes us up.
1896 * However we might have started a new transaction and COWed some
1897 * tree blocks when deleting unused block groups for example. So
1898 * make sure we commit the transaction we started to have a clean
1899 * shutdown when evicting the btree inode - if it has dirty pages
1900 * when we do the final iput() on it, eviction will trigger a
1901 * writeback for it which will fail with null pointer dereferences
1902 * since work queues and other resources were already released and
1903 * destroyed by the time the iput/eviction/writeback is made.
1905 trans = btrfs_attach_transaction(root);
1906 if (IS_ERR(trans)) {
1907 if (PTR_ERR(trans) != -ENOENT)
1908 btrfs_err(root->fs_info,
1909 "cleaner transaction attach returned %ld",
1914 ret = btrfs_commit_transaction(trans, root);
1916 btrfs_err(root->fs_info,
1917 "cleaner open transaction commit returned %d",
1924 static int transaction_kthread(void *arg)
1926 struct btrfs_root *root = arg;
1927 struct btrfs_trans_handle *trans;
1928 struct btrfs_transaction *cur;
1931 unsigned long delay;
1935 cannot_commit = false;
1936 delay = HZ * root->fs_info->commit_interval;
1937 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1939 spin_lock(&root->fs_info->trans_lock);
1940 cur = root->fs_info->running_transaction;
1942 spin_unlock(&root->fs_info->trans_lock);
1946 now = get_seconds();
1947 if (cur->state < TRANS_STATE_BLOCKED &&
1948 (now < cur->start_time ||
1949 now - cur->start_time < root->fs_info->commit_interval)) {
1950 spin_unlock(&root->fs_info->trans_lock);
1954 transid = cur->transid;
1955 spin_unlock(&root->fs_info->trans_lock);
1957 /* If the file system is aborted, this will always fail. */
1958 trans = btrfs_attach_transaction(root);
1959 if (IS_ERR(trans)) {
1960 if (PTR_ERR(trans) != -ENOENT)
1961 cannot_commit = true;
1964 if (transid == trans->transid) {
1965 btrfs_commit_transaction(trans, root);
1967 btrfs_end_transaction(trans, root);
1970 wake_up_process(root->fs_info->cleaner_kthread);
1971 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1973 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1974 &root->fs_info->fs_state)))
1975 btrfs_cleanup_transaction(root);
1976 set_current_state(TASK_INTERRUPTIBLE);
1977 if (!kthread_should_stop() &&
1978 (!btrfs_transaction_blocked(root->fs_info) ||
1980 schedule_timeout(delay);
1981 __set_current_state(TASK_RUNNING);
1982 } while (!kthread_should_stop());
1987 * this will find the highest generation in the array of
1988 * root backups. The index of the highest array is returned,
1989 * or -1 if we can't find anything.
1991 * We check to make sure the array is valid by comparing the
1992 * generation of the latest root in the array with the generation
1993 * in the super block. If they don't match we pitch it.
1995 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1998 int newest_index = -1;
1999 struct btrfs_root_backup *root_backup;
2002 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
2003 root_backup = info->super_copy->super_roots + i;
2004 cur = btrfs_backup_tree_root_gen(root_backup);
2005 if (cur == newest_gen)
2009 /* check to see if we actually wrapped around */
2010 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
2011 root_backup = info->super_copy->super_roots;
2012 cur = btrfs_backup_tree_root_gen(root_backup);
2013 if (cur == newest_gen)
2016 return newest_index;
2021 * find the oldest backup so we know where to store new entries
2022 * in the backup array. This will set the backup_root_index
2023 * field in the fs_info struct
2025 static void find_oldest_super_backup(struct btrfs_fs_info *info,
2028 int newest_index = -1;
2030 newest_index = find_newest_super_backup(info, newest_gen);
2031 /* if there was garbage in there, just move along */
2032 if (newest_index == -1) {
2033 info->backup_root_index = 0;
2035 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
2040 * copy all the root pointers into the super backup array.
2041 * this will bump the backup pointer by one when it is
2044 static void backup_super_roots(struct btrfs_fs_info *info)
2047 struct btrfs_root_backup *root_backup;
2050 next_backup = info->backup_root_index;
2051 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2052 BTRFS_NUM_BACKUP_ROOTS;
2055 * just overwrite the last backup if we're at the same generation
2056 * this happens only at umount
2058 root_backup = info->super_for_commit->super_roots + last_backup;
2059 if (btrfs_backup_tree_root_gen(root_backup) ==
2060 btrfs_header_generation(info->tree_root->node))
2061 next_backup = last_backup;
2063 root_backup = info->super_for_commit->super_roots + next_backup;
2066 * make sure all of our padding and empty slots get zero filled
2067 * regardless of which ones we use today
2069 memset(root_backup, 0, sizeof(*root_backup));
2071 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2073 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2074 btrfs_set_backup_tree_root_gen(root_backup,
2075 btrfs_header_generation(info->tree_root->node));
2077 btrfs_set_backup_tree_root_level(root_backup,
2078 btrfs_header_level(info->tree_root->node));
2080 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2081 btrfs_set_backup_chunk_root_gen(root_backup,
2082 btrfs_header_generation(info->chunk_root->node));
2083 btrfs_set_backup_chunk_root_level(root_backup,
2084 btrfs_header_level(info->chunk_root->node));
2086 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2087 btrfs_set_backup_extent_root_gen(root_backup,
2088 btrfs_header_generation(info->extent_root->node));
2089 btrfs_set_backup_extent_root_level(root_backup,
2090 btrfs_header_level(info->extent_root->node));
2093 * we might commit during log recovery, which happens before we set
2094 * the fs_root. Make sure it is valid before we fill it in.
2096 if (info->fs_root && info->fs_root->node) {
2097 btrfs_set_backup_fs_root(root_backup,
2098 info->fs_root->node->start);
2099 btrfs_set_backup_fs_root_gen(root_backup,
2100 btrfs_header_generation(info->fs_root->node));
2101 btrfs_set_backup_fs_root_level(root_backup,
2102 btrfs_header_level(info->fs_root->node));
2105 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2106 btrfs_set_backup_dev_root_gen(root_backup,
2107 btrfs_header_generation(info->dev_root->node));
2108 btrfs_set_backup_dev_root_level(root_backup,
2109 btrfs_header_level(info->dev_root->node));
2111 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2112 btrfs_set_backup_csum_root_gen(root_backup,
2113 btrfs_header_generation(info->csum_root->node));
2114 btrfs_set_backup_csum_root_level(root_backup,
2115 btrfs_header_level(info->csum_root->node));
2117 btrfs_set_backup_total_bytes(root_backup,
2118 btrfs_super_total_bytes(info->super_copy));
2119 btrfs_set_backup_bytes_used(root_backup,
2120 btrfs_super_bytes_used(info->super_copy));
2121 btrfs_set_backup_num_devices(root_backup,
2122 btrfs_super_num_devices(info->super_copy));
2125 * if we don't copy this out to the super_copy, it won't get remembered
2126 * for the next commit
2128 memcpy(&info->super_copy->super_roots,
2129 &info->super_for_commit->super_roots,
2130 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2134 * this copies info out of the root backup array and back into
2135 * the in-memory super block. It is meant to help iterate through
2136 * the array, so you send it the number of backups you've already
2137 * tried and the last backup index you used.
2139 * this returns -1 when it has tried all the backups
2141 static noinline int next_root_backup(struct btrfs_fs_info *info,
2142 struct btrfs_super_block *super,
2143 int *num_backups_tried, int *backup_index)
2145 struct btrfs_root_backup *root_backup;
2146 int newest = *backup_index;
2148 if (*num_backups_tried == 0) {
2149 u64 gen = btrfs_super_generation(super);
2151 newest = find_newest_super_backup(info, gen);
2155 *backup_index = newest;
2156 *num_backups_tried = 1;
2157 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2158 /* we've tried all the backups, all done */
2161 /* jump to the next oldest backup */
2162 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2163 BTRFS_NUM_BACKUP_ROOTS;
2164 *backup_index = newest;
2165 *num_backups_tried += 1;
2167 root_backup = super->super_roots + newest;
2169 btrfs_set_super_generation(super,
2170 btrfs_backup_tree_root_gen(root_backup));
2171 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2172 btrfs_set_super_root_level(super,
2173 btrfs_backup_tree_root_level(root_backup));
2174 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2177 * fixme: the total bytes and num_devices need to match or we should
2180 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2181 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2185 /* helper to cleanup workers */
2186 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2188 btrfs_destroy_workqueue(fs_info->fixup_workers);
2189 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2190 btrfs_destroy_workqueue(fs_info->workers);
2191 btrfs_destroy_workqueue(fs_info->endio_workers);
2192 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2193 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2194 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2195 btrfs_destroy_workqueue(fs_info->rmw_workers);
2196 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2197 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2198 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2199 btrfs_destroy_workqueue(fs_info->submit_workers);
2200 btrfs_destroy_workqueue(fs_info->delayed_workers);
2201 btrfs_destroy_workqueue(fs_info->caching_workers);
2202 btrfs_destroy_workqueue(fs_info->readahead_workers);
2203 btrfs_destroy_workqueue(fs_info->flush_workers);
2204 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2205 btrfs_destroy_workqueue(fs_info->extent_workers);
2208 static void free_root_extent_buffers(struct btrfs_root *root)
2211 free_extent_buffer(root->node);
2212 free_extent_buffer(root->commit_root);
2214 root->commit_root = NULL;
2218 /* helper to cleanup tree roots */
2219 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2221 free_root_extent_buffers(info->tree_root);
2223 free_root_extent_buffers(info->dev_root);
2224 free_root_extent_buffers(info->extent_root);
2225 free_root_extent_buffers(info->csum_root);
2226 free_root_extent_buffers(info->quota_root);
2227 free_root_extent_buffers(info->uuid_root);
2229 free_root_extent_buffers(info->chunk_root);
2230 free_root_extent_buffers(info->free_space_root);
2233 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2236 struct btrfs_root *gang[8];
2239 while (!list_empty(&fs_info->dead_roots)) {
2240 gang[0] = list_entry(fs_info->dead_roots.next,
2241 struct btrfs_root, root_list);
2242 list_del(&gang[0]->root_list);
2244 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2245 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2247 free_extent_buffer(gang[0]->node);
2248 free_extent_buffer(gang[0]->commit_root);
2249 btrfs_put_fs_root(gang[0]);
2254 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2259 for (i = 0; i < ret; i++)
2260 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2263 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2264 btrfs_free_log_root_tree(NULL, fs_info);
2265 btrfs_destroy_pinned_extent(fs_info->tree_root,
2266 fs_info->pinned_extents);
2270 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2272 mutex_init(&fs_info->scrub_lock);
2273 atomic_set(&fs_info->scrubs_running, 0);
2274 atomic_set(&fs_info->scrub_pause_req, 0);
2275 atomic_set(&fs_info->scrubs_paused, 0);
2276 atomic_set(&fs_info->scrub_cancel_req, 0);
2277 init_waitqueue_head(&fs_info->scrub_pause_wait);
2278 fs_info->scrub_workers_refcnt = 0;
2281 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2283 spin_lock_init(&fs_info->balance_lock);
2284 mutex_init(&fs_info->balance_mutex);
2285 atomic_set(&fs_info->balance_running, 0);
2286 atomic_set(&fs_info->balance_pause_req, 0);
2287 atomic_set(&fs_info->balance_cancel_req, 0);
2288 fs_info->balance_ctl = NULL;
2289 init_waitqueue_head(&fs_info->balance_wait_q);
2292 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2293 struct btrfs_root *tree_root)
2295 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2296 set_nlink(fs_info->btree_inode, 1);
2298 * we set the i_size on the btree inode to the max possible int.
2299 * the real end of the address space is determined by all of
2300 * the devices in the system
2302 fs_info->btree_inode->i_size = OFFSET_MAX;
2303 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2305 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2306 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2307 fs_info->btree_inode->i_mapping);
2308 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2309 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2311 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2313 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2314 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2315 sizeof(struct btrfs_key));
2316 set_bit(BTRFS_INODE_DUMMY,
2317 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2318 btrfs_insert_inode_hash(fs_info->btree_inode);
2321 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2323 fs_info->dev_replace.lock_owner = 0;
2324 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2325 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2326 rwlock_init(&fs_info->dev_replace.lock);
2327 atomic_set(&fs_info->dev_replace.read_locks, 0);
2328 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2329 init_waitqueue_head(&fs_info->replace_wait);
2330 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2333 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2335 spin_lock_init(&fs_info->qgroup_lock);
2336 mutex_init(&fs_info->qgroup_ioctl_lock);
2337 fs_info->qgroup_tree = RB_ROOT;
2338 fs_info->qgroup_op_tree = RB_ROOT;
2339 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2340 fs_info->qgroup_seq = 1;
2341 fs_info->quota_enabled = 0;
2342 fs_info->pending_quota_state = 0;
2343 fs_info->qgroup_ulist = NULL;
2344 fs_info->qgroup_rescan_running = false;
2345 mutex_init(&fs_info->qgroup_rescan_lock);
2348 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2349 struct btrfs_fs_devices *fs_devices)
2351 int max_active = fs_info->thread_pool_size;
2352 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2355 btrfs_alloc_workqueue(fs_info, "worker",
2356 flags | WQ_HIGHPRI, max_active, 16);
2358 fs_info->delalloc_workers =
2359 btrfs_alloc_workqueue(fs_info, "delalloc",
2360 flags, max_active, 2);
2362 fs_info->flush_workers =
2363 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2364 flags, max_active, 0);
2366 fs_info->caching_workers =
2367 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2370 * a higher idle thresh on the submit workers makes it much more
2371 * likely that bios will be send down in a sane order to the
2374 fs_info->submit_workers =
2375 btrfs_alloc_workqueue(fs_info, "submit", flags,
2376 min_t(u64, fs_devices->num_devices,
2379 fs_info->fixup_workers =
2380 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2383 * endios are largely parallel and should have a very
2386 fs_info->endio_workers =
2387 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2388 fs_info->endio_meta_workers =
2389 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2391 fs_info->endio_meta_write_workers =
2392 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2394 fs_info->endio_raid56_workers =
2395 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2397 fs_info->endio_repair_workers =
2398 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2399 fs_info->rmw_workers =
2400 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2401 fs_info->endio_write_workers =
2402 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2404 fs_info->endio_freespace_worker =
2405 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2407 fs_info->delayed_workers =
2408 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2410 fs_info->readahead_workers =
2411 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2413 fs_info->qgroup_rescan_workers =
2414 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2415 fs_info->extent_workers =
2416 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2417 min_t(u64, fs_devices->num_devices,
2420 if (!(fs_info->workers && fs_info->delalloc_workers &&
2421 fs_info->submit_workers && fs_info->flush_workers &&
2422 fs_info->endio_workers && fs_info->endio_meta_workers &&
2423 fs_info->endio_meta_write_workers &&
2424 fs_info->endio_repair_workers &&
2425 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2426 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2427 fs_info->caching_workers && fs_info->readahead_workers &&
2428 fs_info->fixup_workers && fs_info->delayed_workers &&
2429 fs_info->extent_workers &&
2430 fs_info->qgroup_rescan_workers)) {
2437 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2438 struct btrfs_fs_devices *fs_devices)
2441 struct btrfs_root *tree_root = fs_info->tree_root;
2442 struct btrfs_root *log_tree_root;
2443 struct btrfs_super_block *disk_super = fs_info->super_copy;
2444 u64 bytenr = btrfs_super_log_root(disk_super);
2446 if (fs_devices->rw_devices == 0) {
2447 btrfs_warn(fs_info, "log replay required on RO media");
2451 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2455 __setup_root(tree_root->nodesize, tree_root->sectorsize,
2456 tree_root->stripesize, log_tree_root, fs_info,
2457 BTRFS_TREE_LOG_OBJECTID);
2459 log_tree_root->node = read_tree_block(tree_root, bytenr,
2460 fs_info->generation + 1);
2461 if (IS_ERR(log_tree_root->node)) {
2462 btrfs_warn(fs_info, "failed to read log tree");
2463 ret = PTR_ERR(log_tree_root->node);
2464 kfree(log_tree_root);
2466 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2467 btrfs_err(fs_info, "failed to read log tree");
2468 free_extent_buffer(log_tree_root->node);
2469 kfree(log_tree_root);
2472 /* returns with log_tree_root freed on success */
2473 ret = btrfs_recover_log_trees(log_tree_root);
2475 btrfs_handle_fs_error(tree_root->fs_info, ret,
2476 "Failed to recover log tree");
2477 free_extent_buffer(log_tree_root->node);
2478 kfree(log_tree_root);
2482 if (fs_info->sb->s_flags & MS_RDONLY) {
2483 ret = btrfs_commit_super(tree_root);
2491 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2492 struct btrfs_root *tree_root)
2494 struct btrfs_root *root;
2495 struct btrfs_key location;
2498 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2499 location.type = BTRFS_ROOT_ITEM_KEY;
2500 location.offset = 0;
2502 root = btrfs_read_tree_root(tree_root, &location);
2504 return PTR_ERR(root);
2505 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2506 fs_info->extent_root = root;
2508 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2509 root = btrfs_read_tree_root(tree_root, &location);
2511 return PTR_ERR(root);
2512 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2513 fs_info->dev_root = root;
2514 btrfs_init_devices_late(fs_info);
2516 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2517 root = btrfs_read_tree_root(tree_root, &location);
2519 return PTR_ERR(root);
2520 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2521 fs_info->csum_root = root;
2523 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2524 root = btrfs_read_tree_root(tree_root, &location);
2525 if (!IS_ERR(root)) {
2526 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2527 fs_info->quota_enabled = 1;
2528 fs_info->pending_quota_state = 1;
2529 fs_info->quota_root = root;
2532 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2533 root = btrfs_read_tree_root(tree_root, &location);
2535 ret = PTR_ERR(root);
2539 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2540 fs_info->uuid_root = root;
2543 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2544 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2545 root = btrfs_read_tree_root(tree_root, &location);
2547 return PTR_ERR(root);
2548 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2549 fs_info->free_space_root = root;
2555 int open_ctree(struct super_block *sb,
2556 struct btrfs_fs_devices *fs_devices,
2564 struct btrfs_key location;
2565 struct buffer_head *bh;
2566 struct btrfs_super_block *disk_super;
2567 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2568 struct btrfs_root *tree_root;
2569 struct btrfs_root *chunk_root;
2572 int num_backups_tried = 0;
2573 int backup_index = 0;
2576 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2577 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2578 if (!tree_root || !chunk_root) {
2583 ret = init_srcu_struct(&fs_info->subvol_srcu);
2589 ret = setup_bdi(fs_info, &fs_info->bdi);
2595 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2600 fs_info->dirty_metadata_batch = PAGE_SIZE *
2601 (1 + ilog2(nr_cpu_ids));
2603 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2606 goto fail_dirty_metadata_bytes;
2609 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2612 goto fail_delalloc_bytes;
2615 fs_info->btree_inode = new_inode(sb);
2616 if (!fs_info->btree_inode) {
2618 goto fail_bio_counter;
2621 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2623 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2624 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2625 INIT_LIST_HEAD(&fs_info->trans_list);
2626 INIT_LIST_HEAD(&fs_info->dead_roots);
2627 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2628 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2629 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2630 spin_lock_init(&fs_info->delalloc_root_lock);
2631 spin_lock_init(&fs_info->trans_lock);
2632 spin_lock_init(&fs_info->fs_roots_radix_lock);
2633 spin_lock_init(&fs_info->delayed_iput_lock);
2634 spin_lock_init(&fs_info->defrag_inodes_lock);
2635 spin_lock_init(&fs_info->free_chunk_lock);
2636 spin_lock_init(&fs_info->tree_mod_seq_lock);
2637 spin_lock_init(&fs_info->super_lock);
2638 spin_lock_init(&fs_info->qgroup_op_lock);
2639 spin_lock_init(&fs_info->buffer_lock);
2640 spin_lock_init(&fs_info->unused_bgs_lock);
2641 rwlock_init(&fs_info->tree_mod_log_lock);
2642 mutex_init(&fs_info->unused_bg_unpin_mutex);
2643 mutex_init(&fs_info->delete_unused_bgs_mutex);
2644 mutex_init(&fs_info->reloc_mutex);
2645 mutex_init(&fs_info->delalloc_root_mutex);
2646 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2647 seqlock_init(&fs_info->profiles_lock);
2649 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2650 INIT_LIST_HEAD(&fs_info->space_info);
2651 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2652 INIT_LIST_HEAD(&fs_info->unused_bgs);
2653 btrfs_mapping_init(&fs_info->mapping_tree);
2654 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2655 BTRFS_BLOCK_RSV_GLOBAL);
2656 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2657 BTRFS_BLOCK_RSV_DELALLOC);
2658 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2659 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2660 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2661 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2662 BTRFS_BLOCK_RSV_DELOPS);
2663 atomic_set(&fs_info->nr_async_submits, 0);
2664 atomic_set(&fs_info->async_delalloc_pages, 0);
2665 atomic_set(&fs_info->async_submit_draining, 0);
2666 atomic_set(&fs_info->nr_async_bios, 0);
2667 atomic_set(&fs_info->defrag_running, 0);
2668 atomic_set(&fs_info->qgroup_op_seq, 0);
2669 atomic_set(&fs_info->reada_works_cnt, 0);
2670 atomic64_set(&fs_info->tree_mod_seq, 0);
2671 fs_info->fs_frozen = 0;
2673 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2674 fs_info->metadata_ratio = 0;
2675 fs_info->defrag_inodes = RB_ROOT;
2676 fs_info->free_chunk_space = 0;
2677 fs_info->tree_mod_log = RB_ROOT;
2678 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2679 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2680 /* readahead state */
2681 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2682 spin_lock_init(&fs_info->reada_lock);
2684 fs_info->thread_pool_size = min_t(unsigned long,
2685 num_online_cpus() + 2, 8);
2687 INIT_LIST_HEAD(&fs_info->ordered_roots);
2688 spin_lock_init(&fs_info->ordered_root_lock);
2689 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2691 if (!fs_info->delayed_root) {
2695 btrfs_init_delayed_root(fs_info->delayed_root);
2697 btrfs_init_scrub(fs_info);
2698 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2699 fs_info->check_integrity_print_mask = 0;
2701 btrfs_init_balance(fs_info);
2702 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2704 sb->s_blocksize = 4096;
2705 sb->s_blocksize_bits = blksize_bits(4096);
2706 sb->s_bdi = &fs_info->bdi;
2708 btrfs_init_btree_inode(fs_info, tree_root);
2710 spin_lock_init(&fs_info->block_group_cache_lock);
2711 fs_info->block_group_cache_tree = RB_ROOT;
2712 fs_info->first_logical_byte = (u64)-1;
2714 extent_io_tree_init(&fs_info->freed_extents[0],
2715 fs_info->btree_inode->i_mapping);
2716 extent_io_tree_init(&fs_info->freed_extents[1],
2717 fs_info->btree_inode->i_mapping);
2718 fs_info->pinned_extents = &fs_info->freed_extents[0];
2719 fs_info->do_barriers = 1;
2722 mutex_init(&fs_info->ordered_operations_mutex);
2723 mutex_init(&fs_info->tree_log_mutex);
2724 mutex_init(&fs_info->chunk_mutex);
2725 mutex_init(&fs_info->transaction_kthread_mutex);
2726 mutex_init(&fs_info->cleaner_mutex);
2727 mutex_init(&fs_info->volume_mutex);
2728 mutex_init(&fs_info->ro_block_group_mutex);
2729 init_rwsem(&fs_info->commit_root_sem);
2730 init_rwsem(&fs_info->cleanup_work_sem);
2731 init_rwsem(&fs_info->subvol_sem);
2732 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2734 btrfs_init_dev_replace_locks(fs_info);
2735 btrfs_init_qgroup(fs_info);
2737 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2738 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2740 init_waitqueue_head(&fs_info->transaction_throttle);
2741 init_waitqueue_head(&fs_info->transaction_wait);
2742 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2743 init_waitqueue_head(&fs_info->async_submit_wait);
2745 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2747 ret = btrfs_alloc_stripe_hash_table(fs_info);
2753 __setup_root(4096, 4096, 4096, tree_root,
2754 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2756 invalidate_bdev(fs_devices->latest_bdev);
2759 * Read super block and check the signature bytes only
2761 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2768 * We want to check superblock checksum, the type is stored inside.
2769 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2771 if (btrfs_check_super_csum(bh->b_data)) {
2772 btrfs_err(fs_info, "superblock checksum mismatch");
2779 * super_copy is zeroed at allocation time and we never touch the
2780 * following bytes up to INFO_SIZE, the checksum is calculated from
2781 * the whole block of INFO_SIZE
2783 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2784 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2785 sizeof(*fs_info->super_for_commit));
2788 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2790 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2792 btrfs_err(fs_info, "superblock contains fatal errors");
2797 disk_super = fs_info->super_copy;
2798 if (!btrfs_super_root(disk_super))
2801 /* check FS state, whether FS is broken. */
2802 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2803 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2806 * run through our array of backup supers and setup
2807 * our ring pointer to the oldest one
2809 generation = btrfs_super_generation(disk_super);
2810 find_oldest_super_backup(fs_info, generation);
2813 * In the long term, we'll store the compression type in the super
2814 * block, and it'll be used for per file compression control.
2816 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2818 ret = btrfs_parse_options(tree_root, options, sb->s_flags);
2824 features = btrfs_super_incompat_flags(disk_super) &
2825 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2828 "cannot mount because of unsupported optional features (%llx)",
2834 features = btrfs_super_incompat_flags(disk_super);
2835 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2836 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2837 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2839 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2840 btrfs_info(fs_info, "has skinny extents");
2843 * flag our filesystem as having big metadata blocks if
2844 * they are bigger than the page size
2846 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2847 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2849 "flagging fs with big metadata feature");
2850 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2853 nodesize = btrfs_super_nodesize(disk_super);
2854 sectorsize = btrfs_super_sectorsize(disk_super);
2855 stripesize = sectorsize;
2856 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2857 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2860 * mixed block groups end up with duplicate but slightly offset
2861 * extent buffers for the same range. It leads to corruptions
2863 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2864 (sectorsize != nodesize)) {
2866 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2867 nodesize, sectorsize);
2872 * Needn't use the lock because there is no other task which will
2875 btrfs_set_super_incompat_flags(disk_super, features);
2877 features = btrfs_super_compat_ro_flags(disk_super) &
2878 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2879 if (!(sb->s_flags & MS_RDONLY) && features) {
2881 "cannot mount read-write because of unsupported optional features (%llx)",
2887 max_active = fs_info->thread_pool_size;
2889 ret = btrfs_init_workqueues(fs_info, fs_devices);
2892 goto fail_sb_buffer;
2895 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2896 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2899 tree_root->nodesize = nodesize;
2900 tree_root->sectorsize = sectorsize;
2901 tree_root->stripesize = stripesize;
2903 sb->s_blocksize = sectorsize;
2904 sb->s_blocksize_bits = blksize_bits(sectorsize);
2906 mutex_lock(&fs_info->chunk_mutex);
2907 ret = btrfs_read_sys_array(tree_root);
2908 mutex_unlock(&fs_info->chunk_mutex);
2910 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2911 goto fail_sb_buffer;
2914 generation = btrfs_super_chunk_root_generation(disk_super);
2916 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2917 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2919 chunk_root->node = read_tree_block(chunk_root,
2920 btrfs_super_chunk_root(disk_super),
2922 if (IS_ERR(chunk_root->node) ||
2923 !extent_buffer_uptodate(chunk_root->node)) {
2924 btrfs_err(fs_info, "failed to read chunk root");
2925 if (!IS_ERR(chunk_root->node))
2926 free_extent_buffer(chunk_root->node);
2927 chunk_root->node = NULL;
2928 goto fail_tree_roots;
2930 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2931 chunk_root->commit_root = btrfs_root_node(chunk_root);
2933 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2934 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2936 ret = btrfs_read_chunk_tree(chunk_root);
2938 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2939 goto fail_tree_roots;
2943 * keep the device that is marked to be the target device for the
2944 * dev_replace procedure
2946 btrfs_close_extra_devices(fs_devices, 0);
2948 if (!fs_devices->latest_bdev) {
2949 btrfs_err(fs_info, "failed to read devices");
2950 goto fail_tree_roots;
2954 generation = btrfs_super_generation(disk_super);
2956 tree_root->node = read_tree_block(tree_root,
2957 btrfs_super_root(disk_super),
2959 if (IS_ERR(tree_root->node) ||
2960 !extent_buffer_uptodate(tree_root->node)) {
2961 btrfs_warn(fs_info, "failed to read tree root");
2962 if (!IS_ERR(tree_root->node))
2963 free_extent_buffer(tree_root->node);
2964 tree_root->node = NULL;
2965 goto recovery_tree_root;
2968 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2969 tree_root->commit_root = btrfs_root_node(tree_root);
2970 btrfs_set_root_refs(&tree_root->root_item, 1);
2972 mutex_lock(&tree_root->objectid_mutex);
2973 ret = btrfs_find_highest_objectid(tree_root,
2974 &tree_root->highest_objectid);
2976 mutex_unlock(&tree_root->objectid_mutex);
2977 goto recovery_tree_root;
2980 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2982 mutex_unlock(&tree_root->objectid_mutex);
2984 ret = btrfs_read_roots(fs_info, tree_root);
2986 goto recovery_tree_root;
2988 fs_info->generation = generation;
2989 fs_info->last_trans_committed = generation;
2991 ret = btrfs_recover_balance(fs_info);
2993 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2994 goto fail_block_groups;
2997 ret = btrfs_init_dev_stats(fs_info);
2999 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
3000 goto fail_block_groups;
3003 ret = btrfs_init_dev_replace(fs_info);
3005 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
3006 goto fail_block_groups;
3009 btrfs_close_extra_devices(fs_devices, 1);
3011 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
3013 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
3015 goto fail_block_groups;
3018 ret = btrfs_sysfs_add_device(fs_devices);
3020 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
3022 goto fail_fsdev_sysfs;
3025 ret = btrfs_sysfs_add_mounted(fs_info);
3027 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
3028 goto fail_fsdev_sysfs;
3031 ret = btrfs_init_space_info(fs_info);
3033 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3037 ret = btrfs_read_block_groups(fs_info->extent_root);
3039 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3042 fs_info->num_tolerated_disk_barrier_failures =
3043 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3044 if (fs_info->fs_devices->missing_devices >
3045 fs_info->num_tolerated_disk_barrier_failures &&
3046 !(sb->s_flags & MS_RDONLY)) {
3048 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3049 fs_info->fs_devices->missing_devices,
3050 fs_info->num_tolerated_disk_barrier_failures);
3054 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3056 if (IS_ERR(fs_info->cleaner_kthread))
3059 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3061 "btrfs-transaction");
3062 if (IS_ERR(fs_info->transaction_kthread))
3065 if (!btrfs_test_opt(tree_root->fs_info, SSD) &&
3066 !btrfs_test_opt(tree_root->fs_info, NOSSD) &&
3067 !fs_info->fs_devices->rotating) {
3068 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3069 btrfs_set_opt(fs_info->mount_opt, SSD);
3073 * Mount does not set all options immediately, we can do it now and do
3074 * not have to wait for transaction commit
3076 btrfs_apply_pending_changes(fs_info);
3078 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3079 if (btrfs_test_opt(tree_root->fs_info, CHECK_INTEGRITY)) {
3080 ret = btrfsic_mount(tree_root, fs_devices,
3081 btrfs_test_opt(tree_root->fs_info,
3082 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3084 fs_info->check_integrity_print_mask);
3087 "failed to initialize integrity check module: %d",
3091 ret = btrfs_read_qgroup_config(fs_info);
3093 goto fail_trans_kthread;
3095 /* do not make disk changes in broken FS or nologreplay is given */
3096 if (btrfs_super_log_root(disk_super) != 0 &&
3097 !btrfs_test_opt(tree_root->fs_info, NOLOGREPLAY)) {
3098 ret = btrfs_replay_log(fs_info, fs_devices);
3105 ret = btrfs_find_orphan_roots(tree_root);
3109 if (!(sb->s_flags & MS_RDONLY)) {
3110 ret = btrfs_cleanup_fs_roots(fs_info);
3114 mutex_lock(&fs_info->cleaner_mutex);
3115 ret = btrfs_recover_relocation(tree_root);
3116 mutex_unlock(&fs_info->cleaner_mutex);
3118 btrfs_warn(fs_info, "failed to recover relocation: %d",
3125 location.objectid = BTRFS_FS_TREE_OBJECTID;
3126 location.type = BTRFS_ROOT_ITEM_KEY;
3127 location.offset = 0;
3129 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3130 if (IS_ERR(fs_info->fs_root)) {
3131 err = PTR_ERR(fs_info->fs_root);
3135 if (sb->s_flags & MS_RDONLY)
3138 if (btrfs_test_opt(tree_root->fs_info, FREE_SPACE_TREE) &&
3139 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3140 btrfs_info(fs_info, "creating free space tree");
3141 ret = btrfs_create_free_space_tree(fs_info);
3144 "failed to create free space tree: %d", ret);
3145 close_ctree(tree_root);
3150 down_read(&fs_info->cleanup_work_sem);
3151 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3152 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3153 up_read(&fs_info->cleanup_work_sem);
3154 close_ctree(tree_root);
3157 up_read(&fs_info->cleanup_work_sem);
3159 ret = btrfs_resume_balance_async(fs_info);
3161 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3162 close_ctree(tree_root);
3166 ret = btrfs_resume_dev_replace_async(fs_info);
3168 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3169 close_ctree(tree_root);
3173 btrfs_qgroup_rescan_resume(fs_info);
3175 if (btrfs_test_opt(tree_root->fs_info, CLEAR_CACHE) &&
3176 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3177 btrfs_info(fs_info, "clearing free space tree");
3178 ret = btrfs_clear_free_space_tree(fs_info);
3181 "failed to clear free space tree: %d", ret);
3182 close_ctree(tree_root);
3187 if (!fs_info->uuid_root) {
3188 btrfs_info(fs_info, "creating UUID tree");
3189 ret = btrfs_create_uuid_tree(fs_info);
3192 "failed to create the UUID tree: %d", ret);
3193 close_ctree(tree_root);
3196 } else if (btrfs_test_opt(tree_root->fs_info, RESCAN_UUID_TREE) ||
3197 fs_info->generation !=
3198 btrfs_super_uuid_tree_generation(disk_super)) {
3199 btrfs_info(fs_info, "checking UUID tree");
3200 ret = btrfs_check_uuid_tree(fs_info);
3203 "failed to check the UUID tree: %d", ret);
3204 close_ctree(tree_root);
3208 fs_info->update_uuid_tree_gen = 1;
3214 * backuproot only affect mount behavior, and if open_ctree succeeded,
3215 * no need to keep the flag
3217 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3222 btrfs_free_qgroup_config(fs_info);
3224 kthread_stop(fs_info->transaction_kthread);
3225 btrfs_cleanup_transaction(fs_info->tree_root);
3226 btrfs_free_fs_roots(fs_info);
3228 kthread_stop(fs_info->cleaner_kthread);
3231 * make sure we're done with the btree inode before we stop our
3234 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3237 btrfs_sysfs_remove_mounted(fs_info);
3240 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3243 btrfs_put_block_group_cache(fs_info);
3244 btrfs_free_block_groups(fs_info);
3247 free_root_pointers(fs_info, 1);
3248 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3251 btrfs_stop_all_workers(fs_info);
3254 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3256 iput(fs_info->btree_inode);
3258 percpu_counter_destroy(&fs_info->bio_counter);
3259 fail_delalloc_bytes:
3260 percpu_counter_destroy(&fs_info->delalloc_bytes);
3261 fail_dirty_metadata_bytes:
3262 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3264 bdi_destroy(&fs_info->bdi);
3266 cleanup_srcu_struct(&fs_info->subvol_srcu);
3268 btrfs_free_stripe_hash_table(fs_info);
3269 btrfs_close_devices(fs_info->fs_devices);
3273 if (!btrfs_test_opt(tree_root->fs_info, USEBACKUPROOT))
3274 goto fail_tree_roots;
3276 free_root_pointers(fs_info, 0);
3278 /* don't use the log in recovery mode, it won't be valid */
3279 btrfs_set_super_log_root(disk_super, 0);
3281 /* we can't trust the free space cache either */
3282 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3284 ret = next_root_backup(fs_info, fs_info->super_copy,
3285 &num_backups_tried, &backup_index);
3287 goto fail_block_groups;
3288 goto retry_root_backup;
3291 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3294 set_buffer_uptodate(bh);
3296 struct btrfs_device *device = (struct btrfs_device *)
3299 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3300 "lost page write due to IO error on %s",
3301 rcu_str_deref(device->name));
3302 /* note, we don't set_buffer_write_io_error because we have
3303 * our own ways of dealing with the IO errors
3305 clear_buffer_uptodate(bh);
3306 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3312 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3313 struct buffer_head **bh_ret)
3315 struct buffer_head *bh;
3316 struct btrfs_super_block *super;
3319 bytenr = btrfs_sb_offset(copy_num);
3320 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3323 bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3325 * If we fail to read from the underlying devices, as of now
3326 * the best option we have is to mark it EIO.
3331 super = (struct btrfs_super_block *)bh->b_data;
3332 if (btrfs_super_bytenr(super) != bytenr ||
3333 btrfs_super_magic(super) != BTRFS_MAGIC) {
3343 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3345 struct buffer_head *bh;
3346 struct buffer_head *latest = NULL;
3347 struct btrfs_super_block *super;
3352 /* we would like to check all the supers, but that would make
3353 * a btrfs mount succeed after a mkfs from a different FS.
3354 * So, we need to add a special mount option to scan for
3355 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3357 for (i = 0; i < 1; i++) {
3358 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3362 super = (struct btrfs_super_block *)bh->b_data;
3364 if (!latest || btrfs_super_generation(super) > transid) {
3367 transid = btrfs_super_generation(super);
3374 return ERR_PTR(ret);
3380 * this should be called twice, once with wait == 0 and
3381 * once with wait == 1. When wait == 0 is done, all the buffer heads
3382 * we write are pinned.
3384 * They are released when wait == 1 is done.
3385 * max_mirrors must be the same for both runs, and it indicates how
3386 * many supers on this one device should be written.
3388 * max_mirrors == 0 means to write them all.
3390 static int write_dev_supers(struct btrfs_device *device,
3391 struct btrfs_super_block *sb,
3392 int do_barriers, int wait, int max_mirrors)
3394 struct buffer_head *bh;
3401 if (max_mirrors == 0)
3402 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3404 for (i = 0; i < max_mirrors; i++) {
3405 bytenr = btrfs_sb_offset(i);
3406 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3407 device->commit_total_bytes)
3411 bh = __find_get_block(device->bdev, bytenr / 4096,
3412 BTRFS_SUPER_INFO_SIZE);
3418 if (!buffer_uptodate(bh))
3421 /* drop our reference */
3424 /* drop the reference from the wait == 0 run */
3428 btrfs_set_super_bytenr(sb, bytenr);
3431 crc = btrfs_csum_data((char *)sb +
3432 BTRFS_CSUM_SIZE, crc,
3433 BTRFS_SUPER_INFO_SIZE -
3435 btrfs_csum_final(crc, sb->csum);
3438 * one reference for us, and we leave it for the
3441 bh = __getblk(device->bdev, bytenr / 4096,
3442 BTRFS_SUPER_INFO_SIZE);
3444 btrfs_err(device->dev_root->fs_info,
3445 "couldn't get super buffer head for bytenr %llu",
3451 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3453 /* one reference for submit_bh */
3456 set_buffer_uptodate(bh);
3458 bh->b_end_io = btrfs_end_buffer_write_sync;
3459 bh->b_private = device;
3463 * we fua the first super. The others we allow
3467 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3469 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3473 return errors < i ? 0 : -1;
3477 * endio for the write_dev_flush, this will wake anyone waiting
3478 * for the barrier when it is done
3480 static void btrfs_end_empty_barrier(struct bio *bio)
3482 if (bio->bi_private)
3483 complete(bio->bi_private);
3488 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3489 * sent down. With wait == 1, it waits for the previous flush.
3491 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3494 static int write_dev_flush(struct btrfs_device *device, int wait)
3499 if (device->nobarriers)
3503 bio = device->flush_bio;
3507 wait_for_completion(&device->flush_wait);
3509 if (bio->bi_error) {
3510 ret = bio->bi_error;
3511 btrfs_dev_stat_inc_and_print(device,
3512 BTRFS_DEV_STAT_FLUSH_ERRS);
3515 /* drop the reference from the wait == 0 run */
3517 device->flush_bio = NULL;
3523 * one reference for us, and we leave it for the
3526 device->flush_bio = NULL;
3527 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3531 bio->bi_end_io = btrfs_end_empty_barrier;
3532 bio->bi_bdev = device->bdev;
3533 init_completion(&device->flush_wait);
3534 bio->bi_private = &device->flush_wait;
3535 device->flush_bio = bio;
3538 btrfsic_submit_bio(WRITE_FLUSH, bio);
3544 * send an empty flush down to each device in parallel,
3545 * then wait for them
3547 static int barrier_all_devices(struct btrfs_fs_info *info)
3549 struct list_head *head;
3550 struct btrfs_device *dev;
3551 int errors_send = 0;
3552 int errors_wait = 0;
3555 /* send down all the barriers */
3556 head = &info->fs_devices->devices;
3557 list_for_each_entry_rcu(dev, head, dev_list) {
3564 if (!dev->in_fs_metadata || !dev->writeable)
3567 ret = write_dev_flush(dev, 0);
3572 /* wait for all the barriers */
3573 list_for_each_entry_rcu(dev, head, dev_list) {
3580 if (!dev->in_fs_metadata || !dev->writeable)
3583 ret = write_dev_flush(dev, 1);
3587 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3588 errors_wait > info->num_tolerated_disk_barrier_failures)
3593 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3596 int min_tolerated = INT_MAX;
3598 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3599 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3600 min_tolerated = min(min_tolerated,
3601 btrfs_raid_array[BTRFS_RAID_SINGLE].
3602 tolerated_failures);
3604 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3605 if (raid_type == BTRFS_RAID_SINGLE)
3607 if (!(flags & btrfs_raid_group[raid_type]))
3609 min_tolerated = min(min_tolerated,
3610 btrfs_raid_array[raid_type].
3611 tolerated_failures);
3614 if (min_tolerated == INT_MAX) {
3615 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3619 return min_tolerated;
3622 int btrfs_calc_num_tolerated_disk_barrier_failures(
3623 struct btrfs_fs_info *fs_info)
3625 struct btrfs_ioctl_space_info space;
3626 struct btrfs_space_info *sinfo;
3627 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3628 BTRFS_BLOCK_GROUP_SYSTEM,
3629 BTRFS_BLOCK_GROUP_METADATA,
3630 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3633 int num_tolerated_disk_barrier_failures =
3634 (int)fs_info->fs_devices->num_devices;
3636 for (i = 0; i < ARRAY_SIZE(types); i++) {
3637 struct btrfs_space_info *tmp;
3641 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3642 if (tmp->flags == types[i]) {
3652 down_read(&sinfo->groups_sem);
3653 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3656 if (list_empty(&sinfo->block_groups[c]))
3659 btrfs_get_block_group_info(&sinfo->block_groups[c],
3661 if (space.total_bytes == 0 || space.used_bytes == 0)
3663 flags = space.flags;
3665 num_tolerated_disk_barrier_failures = min(
3666 num_tolerated_disk_barrier_failures,
3667 btrfs_get_num_tolerated_disk_barrier_failures(
3670 up_read(&sinfo->groups_sem);
3673 return num_tolerated_disk_barrier_failures;
3676 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3678 struct list_head *head;
3679 struct btrfs_device *dev;
3680 struct btrfs_super_block *sb;
3681 struct btrfs_dev_item *dev_item;
3685 int total_errors = 0;
3688 do_barriers = !btrfs_test_opt(root->fs_info, NOBARRIER);
3689 backup_super_roots(root->fs_info);
3691 sb = root->fs_info->super_for_commit;
3692 dev_item = &sb->dev_item;
3694 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3695 head = &root->fs_info->fs_devices->devices;
3696 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3699 ret = barrier_all_devices(root->fs_info);
3702 &root->fs_info->fs_devices->device_list_mutex);
3703 btrfs_handle_fs_error(root->fs_info, ret,
3704 "errors while submitting device barriers.");
3709 list_for_each_entry_rcu(dev, head, dev_list) {
3714 if (!dev->in_fs_metadata || !dev->writeable)
3717 btrfs_set_stack_device_generation(dev_item, 0);
3718 btrfs_set_stack_device_type(dev_item, dev->type);
3719 btrfs_set_stack_device_id(dev_item, dev->devid);
3720 btrfs_set_stack_device_total_bytes(dev_item,
3721 dev->commit_total_bytes);
3722 btrfs_set_stack_device_bytes_used(dev_item,
3723 dev->commit_bytes_used);
3724 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3725 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3726 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3727 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3728 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3730 flags = btrfs_super_flags(sb);
3731 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3733 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3737 if (total_errors > max_errors) {
3738 btrfs_err(root->fs_info, "%d errors while writing supers",
3740 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3742 /* FUA is masked off if unsupported and can't be the reason */
3743 btrfs_handle_fs_error(root->fs_info, -EIO,
3744 "%d errors while writing supers", total_errors);
3749 list_for_each_entry_rcu(dev, head, dev_list) {
3752 if (!dev->in_fs_metadata || !dev->writeable)
3755 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3759 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3760 if (total_errors > max_errors) {
3761 btrfs_handle_fs_error(root->fs_info, -EIO,
3762 "%d errors while writing supers", total_errors);
3768 int write_ctree_super(struct btrfs_trans_handle *trans,
3769 struct btrfs_root *root, int max_mirrors)
3771 return write_all_supers(root, max_mirrors);
3774 /* Drop a fs root from the radix tree and free it. */
3775 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3776 struct btrfs_root *root)
3778 spin_lock(&fs_info->fs_roots_radix_lock);
3779 radix_tree_delete(&fs_info->fs_roots_radix,
3780 (unsigned long)root->root_key.objectid);
3781 spin_unlock(&fs_info->fs_roots_radix_lock);
3783 if (btrfs_root_refs(&root->root_item) == 0)
3784 synchronize_srcu(&fs_info->subvol_srcu);
3786 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3787 btrfs_free_log(NULL, root);
3788 if (root->reloc_root) {
3789 free_extent_buffer(root->reloc_root->node);
3790 free_extent_buffer(root->reloc_root->commit_root);
3791 btrfs_put_fs_root(root->reloc_root);
3792 root->reloc_root = NULL;
3796 if (root->free_ino_pinned)
3797 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3798 if (root->free_ino_ctl)
3799 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3803 static void free_fs_root(struct btrfs_root *root)
3805 iput(root->ino_cache_inode);
3806 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3807 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3808 root->orphan_block_rsv = NULL;
3810 free_anon_bdev(root->anon_dev);
3811 if (root->subv_writers)
3812 btrfs_free_subvolume_writers(root->subv_writers);
3813 free_extent_buffer(root->node);
3814 free_extent_buffer(root->commit_root);
3815 kfree(root->free_ino_ctl);
3816 kfree(root->free_ino_pinned);
3818 btrfs_put_fs_root(root);
3821 void btrfs_free_fs_root(struct btrfs_root *root)
3826 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3828 u64 root_objectid = 0;
3829 struct btrfs_root *gang[8];
3832 unsigned int ret = 0;
3836 index = srcu_read_lock(&fs_info->subvol_srcu);
3837 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3838 (void **)gang, root_objectid,
3841 srcu_read_unlock(&fs_info->subvol_srcu, index);
3844 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3846 for (i = 0; i < ret; i++) {
3847 /* Avoid to grab roots in dead_roots */
3848 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3852 /* grab all the search result for later use */
3853 gang[i] = btrfs_grab_fs_root(gang[i]);
3855 srcu_read_unlock(&fs_info->subvol_srcu, index);
3857 for (i = 0; i < ret; i++) {
3860 root_objectid = gang[i]->root_key.objectid;
3861 err = btrfs_orphan_cleanup(gang[i]);
3864 btrfs_put_fs_root(gang[i]);
3869 /* release the uncleaned roots due to error */
3870 for (; i < ret; i++) {
3872 btrfs_put_fs_root(gang[i]);
3877 int btrfs_commit_super(struct btrfs_root *root)
3879 struct btrfs_trans_handle *trans;
3881 mutex_lock(&root->fs_info->cleaner_mutex);
3882 btrfs_run_delayed_iputs(root);
3883 mutex_unlock(&root->fs_info->cleaner_mutex);
3884 wake_up_process(root->fs_info->cleaner_kthread);
3886 /* wait until ongoing cleanup work done */
3887 down_write(&root->fs_info->cleanup_work_sem);
3888 up_write(&root->fs_info->cleanup_work_sem);
3890 trans = btrfs_join_transaction(root);
3892 return PTR_ERR(trans);
3893 return btrfs_commit_transaction(trans, root);
3896 void close_ctree(struct btrfs_root *root)
3898 struct btrfs_fs_info *fs_info = root->fs_info;
3901 fs_info->closing = 1;
3904 /* wait for the qgroup rescan worker to stop */
3905 btrfs_qgroup_wait_for_completion(fs_info, false);
3907 /* wait for the uuid_scan task to finish */
3908 down(&fs_info->uuid_tree_rescan_sem);
3909 /* avoid complains from lockdep et al., set sem back to initial state */
3910 up(&fs_info->uuid_tree_rescan_sem);
3912 /* pause restriper - we want to resume on mount */
3913 btrfs_pause_balance(fs_info);
3915 btrfs_dev_replace_suspend_for_unmount(fs_info);
3917 btrfs_scrub_cancel(fs_info);
3919 /* wait for any defraggers to finish */
3920 wait_event(fs_info->transaction_wait,
3921 (atomic_read(&fs_info->defrag_running) == 0));
3923 /* clear out the rbtree of defraggable inodes */
3924 btrfs_cleanup_defrag_inodes(fs_info);
3926 cancel_work_sync(&fs_info->async_reclaim_work);
3928 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3930 * If the cleaner thread is stopped and there are
3931 * block groups queued for removal, the deletion will be
3932 * skipped when we quit the cleaner thread.
3934 btrfs_delete_unused_bgs(root->fs_info);
3936 ret = btrfs_commit_super(root);
3938 btrfs_err(fs_info, "commit super ret %d", ret);
3941 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3942 btrfs_error_commit_super(root);
3944 kthread_stop(fs_info->transaction_kthread);
3945 kthread_stop(fs_info->cleaner_kthread);
3947 fs_info->closing = 2;
3950 btrfs_free_qgroup_config(fs_info);
3952 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3953 btrfs_info(fs_info, "at unmount delalloc count %lld",
3954 percpu_counter_sum(&fs_info->delalloc_bytes));
3957 btrfs_sysfs_remove_mounted(fs_info);
3958 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3960 btrfs_free_fs_roots(fs_info);
3962 btrfs_put_block_group_cache(fs_info);
3964 btrfs_free_block_groups(fs_info);
3967 * we must make sure there is not any read request to
3968 * submit after we stopping all workers.
3970 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3971 btrfs_stop_all_workers(fs_info);
3974 free_root_pointers(fs_info, 1);
3976 iput(fs_info->btree_inode);
3978 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3979 if (btrfs_test_opt(root->fs_info, CHECK_INTEGRITY))
3980 btrfsic_unmount(root, fs_info->fs_devices);
3983 btrfs_close_devices(fs_info->fs_devices);
3984 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3986 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3987 percpu_counter_destroy(&fs_info->delalloc_bytes);
3988 percpu_counter_destroy(&fs_info->bio_counter);
3989 bdi_destroy(&fs_info->bdi);
3990 cleanup_srcu_struct(&fs_info->subvol_srcu);
3992 btrfs_free_stripe_hash_table(fs_info);
3994 __btrfs_free_block_rsv(root->orphan_block_rsv);
3995 root->orphan_block_rsv = NULL;
3998 while (!list_empty(&fs_info->pinned_chunks)) {
3999 struct extent_map *em;
4001 em = list_first_entry(&fs_info->pinned_chunks,
4002 struct extent_map, list);
4003 list_del_init(&em->list);
4004 free_extent_map(em);
4006 unlock_chunks(root);
4009 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
4013 struct inode *btree_inode = buf->pages[0]->mapping->host;
4015 ret = extent_buffer_uptodate(buf);
4019 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
4020 parent_transid, atomic);
4026 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
4028 struct btrfs_root *root;
4029 u64 transid = btrfs_header_generation(buf);
4032 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
4034 * This is a fast path so only do this check if we have sanity tests
4035 * enabled. Normal people shouldn't be marking dummy buffers as dirty
4036 * outside of the sanity tests.
4038 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4041 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4042 btrfs_assert_tree_locked(buf);
4043 if (transid != root->fs_info->generation)
4044 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
4045 "found %llu running %llu\n",
4046 buf->start, transid, root->fs_info->generation);
4047 was_dirty = set_extent_buffer_dirty(buf);
4049 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
4051 root->fs_info->dirty_metadata_batch);
4052 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4053 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4054 btrfs_print_leaf(root, buf);
4060 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
4064 * looks as though older kernels can get into trouble with
4065 * this code, they end up stuck in balance_dirty_pages forever
4069 if (current->flags & PF_MEMALLOC)
4073 btrfs_balance_delayed_items(root);
4075 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
4076 BTRFS_DIRTY_METADATA_THRESH);
4078 balance_dirty_pages_ratelimited(
4079 root->fs_info->btree_inode->i_mapping);
4083 void btrfs_btree_balance_dirty(struct btrfs_root *root)
4085 __btrfs_btree_balance_dirty(root, 1);
4088 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4090 __btrfs_btree_balance_dirty(root, 0);
4093 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4095 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4096 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4099 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4102 struct btrfs_super_block *sb = fs_info->super_copy;
4103 u64 nodesize = btrfs_super_nodesize(sb);
4104 u64 sectorsize = btrfs_super_sectorsize(sb);
4107 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4108 printk(KERN_ERR "BTRFS: no valid FS found\n");
4111 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4112 printk(KERN_WARNING "BTRFS: unrecognized super flag: %llu\n",
4113 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4114 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4115 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4116 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4119 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4120 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4121 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4124 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4125 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4126 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4131 * Check sectorsize and nodesize first, other check will need it.
4132 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4134 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4135 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4136 printk(KERN_ERR "BTRFS: invalid sectorsize %llu\n", sectorsize);
4139 /* Only PAGE SIZE is supported yet */
4140 if (sectorsize != PAGE_SIZE) {
4141 printk(KERN_ERR "BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4142 sectorsize, PAGE_SIZE);
4145 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4146 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4147 printk(KERN_ERR "BTRFS: invalid nodesize %llu\n", nodesize);
4150 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4151 printk(KERN_ERR "BTRFS: invalid leafsize %u, should be %llu\n",
4152 le32_to_cpu(sb->__unused_leafsize),
4157 /* Root alignment check */
4158 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4159 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4160 btrfs_super_root(sb));
4163 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4164 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4165 btrfs_super_chunk_root(sb));
4168 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4169 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4170 btrfs_super_log_root(sb));
4174 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4175 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4176 fs_info->fsid, sb->dev_item.fsid);
4181 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4184 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4185 btrfs_err(fs_info, "bytes_used is too small %llu",
4186 btrfs_super_bytes_used(sb));
4189 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4190 btrfs_err(fs_info, "invalid stripesize %u",
4191 btrfs_super_stripesize(sb));
4194 if (btrfs_super_num_devices(sb) > (1UL << 31))
4195 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4196 btrfs_super_num_devices(sb));
4197 if (btrfs_super_num_devices(sb) == 0) {
4198 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4202 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4203 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4204 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4209 * Obvious sys_chunk_array corruptions, it must hold at least one key
4212 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4213 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4214 btrfs_super_sys_array_size(sb),
4215 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4218 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4219 + sizeof(struct btrfs_chunk)) {
4220 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4221 btrfs_super_sys_array_size(sb),
4222 sizeof(struct btrfs_disk_key)
4223 + sizeof(struct btrfs_chunk));
4228 * The generation is a global counter, we'll trust it more than the others
4229 * but it's still possible that it's the one that's wrong.
4231 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4233 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4234 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4235 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4236 && btrfs_super_cache_generation(sb) != (u64)-1)
4238 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4239 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4244 static void btrfs_error_commit_super(struct btrfs_root *root)
4246 mutex_lock(&root->fs_info->cleaner_mutex);
4247 btrfs_run_delayed_iputs(root);
4248 mutex_unlock(&root->fs_info->cleaner_mutex);
4250 down_write(&root->fs_info->cleanup_work_sem);
4251 up_write(&root->fs_info->cleanup_work_sem);
4253 /* cleanup FS via transaction */
4254 btrfs_cleanup_transaction(root);
4257 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4259 struct btrfs_ordered_extent *ordered;
4261 spin_lock(&root->ordered_extent_lock);
4263 * This will just short circuit the ordered completion stuff which will
4264 * make sure the ordered extent gets properly cleaned up.
4266 list_for_each_entry(ordered, &root->ordered_extents,
4268 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4269 spin_unlock(&root->ordered_extent_lock);
4272 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4274 struct btrfs_root *root;
4275 struct list_head splice;
4277 INIT_LIST_HEAD(&splice);
4279 spin_lock(&fs_info->ordered_root_lock);
4280 list_splice_init(&fs_info->ordered_roots, &splice);
4281 while (!list_empty(&splice)) {
4282 root = list_first_entry(&splice, struct btrfs_root,
4284 list_move_tail(&root->ordered_root,
4285 &fs_info->ordered_roots);
4287 spin_unlock(&fs_info->ordered_root_lock);
4288 btrfs_destroy_ordered_extents(root);
4291 spin_lock(&fs_info->ordered_root_lock);
4293 spin_unlock(&fs_info->ordered_root_lock);
4296 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4297 struct btrfs_root *root)
4299 struct rb_node *node;
4300 struct btrfs_delayed_ref_root *delayed_refs;
4301 struct btrfs_delayed_ref_node *ref;
4304 delayed_refs = &trans->delayed_refs;
4306 spin_lock(&delayed_refs->lock);
4307 if (atomic_read(&delayed_refs->num_entries) == 0) {
4308 spin_unlock(&delayed_refs->lock);
4309 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4313 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4314 struct btrfs_delayed_ref_head *head;
4315 struct btrfs_delayed_ref_node *tmp;
4316 bool pin_bytes = false;
4318 head = rb_entry(node, struct btrfs_delayed_ref_head,
4320 if (!mutex_trylock(&head->mutex)) {
4321 atomic_inc(&head->node.refs);
4322 spin_unlock(&delayed_refs->lock);
4324 mutex_lock(&head->mutex);
4325 mutex_unlock(&head->mutex);
4326 btrfs_put_delayed_ref(&head->node);
4327 spin_lock(&delayed_refs->lock);
4330 spin_lock(&head->lock);
4331 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4334 list_del(&ref->list);
4335 atomic_dec(&delayed_refs->num_entries);
4336 btrfs_put_delayed_ref(ref);
4338 if (head->must_insert_reserved)
4340 btrfs_free_delayed_extent_op(head->extent_op);
4341 delayed_refs->num_heads--;
4342 if (head->processing == 0)
4343 delayed_refs->num_heads_ready--;
4344 atomic_dec(&delayed_refs->num_entries);
4345 head->node.in_tree = 0;
4346 rb_erase(&head->href_node, &delayed_refs->href_root);
4347 spin_unlock(&head->lock);
4348 spin_unlock(&delayed_refs->lock);
4349 mutex_unlock(&head->mutex);
4352 btrfs_pin_extent(root, head->node.bytenr,
4353 head->node.num_bytes, 1);
4354 btrfs_put_delayed_ref(&head->node);
4356 spin_lock(&delayed_refs->lock);
4359 spin_unlock(&delayed_refs->lock);
4364 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4366 struct btrfs_inode *btrfs_inode;
4367 struct list_head splice;
4369 INIT_LIST_HEAD(&splice);
4371 spin_lock(&root->delalloc_lock);
4372 list_splice_init(&root->delalloc_inodes, &splice);
4374 while (!list_empty(&splice)) {
4375 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4378 list_del_init(&btrfs_inode->delalloc_inodes);
4379 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4380 &btrfs_inode->runtime_flags);
4381 spin_unlock(&root->delalloc_lock);
4383 btrfs_invalidate_inodes(btrfs_inode->root);
4385 spin_lock(&root->delalloc_lock);
4388 spin_unlock(&root->delalloc_lock);
4391 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4393 struct btrfs_root *root;
4394 struct list_head splice;
4396 INIT_LIST_HEAD(&splice);
4398 spin_lock(&fs_info->delalloc_root_lock);
4399 list_splice_init(&fs_info->delalloc_roots, &splice);
4400 while (!list_empty(&splice)) {
4401 root = list_first_entry(&splice, struct btrfs_root,
4403 list_del_init(&root->delalloc_root);
4404 root = btrfs_grab_fs_root(root);
4406 spin_unlock(&fs_info->delalloc_root_lock);
4408 btrfs_destroy_delalloc_inodes(root);
4409 btrfs_put_fs_root(root);
4411 spin_lock(&fs_info->delalloc_root_lock);
4413 spin_unlock(&fs_info->delalloc_root_lock);
4416 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4417 struct extent_io_tree *dirty_pages,
4421 struct extent_buffer *eb;
4426 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4431 clear_extent_bits(dirty_pages, start, end, mark);
4432 while (start <= end) {
4433 eb = btrfs_find_tree_block(root->fs_info, start);
4434 start += root->nodesize;
4437 wait_on_extent_buffer_writeback(eb);
4439 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4441 clear_extent_buffer_dirty(eb);
4442 free_extent_buffer_stale(eb);
4449 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4450 struct extent_io_tree *pinned_extents)
4452 struct extent_io_tree *unpin;
4458 unpin = pinned_extents;
4461 ret = find_first_extent_bit(unpin, 0, &start, &end,
4462 EXTENT_DIRTY, NULL);
4466 clear_extent_dirty(unpin, start, end);
4467 btrfs_error_unpin_extent_range(root, start, end);
4472 if (unpin == &root->fs_info->freed_extents[0])
4473 unpin = &root->fs_info->freed_extents[1];
4475 unpin = &root->fs_info->freed_extents[0];
4483 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4484 struct btrfs_root *root)
4486 btrfs_destroy_delayed_refs(cur_trans, root);
4488 cur_trans->state = TRANS_STATE_COMMIT_START;
4489 wake_up(&root->fs_info->transaction_blocked_wait);
4491 cur_trans->state = TRANS_STATE_UNBLOCKED;
4492 wake_up(&root->fs_info->transaction_wait);
4494 btrfs_destroy_delayed_inodes(root);
4495 btrfs_assert_delayed_root_empty(root);
4497 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4499 btrfs_destroy_pinned_extent(root,
4500 root->fs_info->pinned_extents);
4502 cur_trans->state =TRANS_STATE_COMPLETED;
4503 wake_up(&cur_trans->commit_wait);
4506 memset(cur_trans, 0, sizeof(*cur_trans));
4507 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4511 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4513 struct btrfs_transaction *t;
4515 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4517 spin_lock(&root->fs_info->trans_lock);
4518 while (!list_empty(&root->fs_info->trans_list)) {
4519 t = list_first_entry(&root->fs_info->trans_list,
4520 struct btrfs_transaction, list);
4521 if (t->state >= TRANS_STATE_COMMIT_START) {
4522 atomic_inc(&t->use_count);
4523 spin_unlock(&root->fs_info->trans_lock);
4524 btrfs_wait_for_commit(root, t->transid);
4525 btrfs_put_transaction(t);
4526 spin_lock(&root->fs_info->trans_lock);
4529 if (t == root->fs_info->running_transaction) {
4530 t->state = TRANS_STATE_COMMIT_DOING;
4531 spin_unlock(&root->fs_info->trans_lock);
4533 * We wait for 0 num_writers since we don't hold a trans
4534 * handle open currently for this transaction.
4536 wait_event(t->writer_wait,
4537 atomic_read(&t->num_writers) == 0);
4539 spin_unlock(&root->fs_info->trans_lock);
4541 btrfs_cleanup_one_transaction(t, root);
4543 spin_lock(&root->fs_info->trans_lock);
4544 if (t == root->fs_info->running_transaction)
4545 root->fs_info->running_transaction = NULL;
4546 list_del_init(&t->list);
4547 spin_unlock(&root->fs_info->trans_lock);
4549 btrfs_put_transaction(t);
4550 trace_btrfs_transaction_commit(root);
4551 spin_lock(&root->fs_info->trans_lock);
4553 spin_unlock(&root->fs_info->trans_lock);
4554 btrfs_destroy_all_ordered_extents(root->fs_info);
4555 btrfs_destroy_delayed_inodes(root);
4556 btrfs_assert_delayed_root_empty(root);
4557 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4558 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4559 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4564 static const struct extent_io_ops btree_extent_io_ops = {
4565 .readpage_end_io_hook = btree_readpage_end_io_hook,
4566 .readpage_io_failed_hook = btree_io_failed_hook,
4567 .submit_bio_hook = btree_submit_bio_hook,
4568 /* note we're sharing with inode.c for the merge bio hook */
4569 .merge_bio_hook = btrfs_merge_bio_hook,