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);
566 /* Check the 0 item */
567 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
568 BTRFS_LEAF_DATA_SIZE(root)) {
569 CORRUPT("invalid item offset size pair", leaf, root, 0);
574 * Check to make sure each items keys are in the correct order and their
575 * offsets make sense. We only have to loop through nritems-1 because
576 * we check the current slot against the next slot, which verifies the
577 * next slot's offset+size makes sense and that the current's slot
580 for (slot = 0; slot < nritems - 1; slot++) {
581 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
582 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
584 /* Make sure the keys are in the right order */
585 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
586 CORRUPT("bad key order", leaf, root, slot);
591 * Make sure the offset and ends are right, remember that the
592 * item data starts at the end of the leaf and grows towards the
595 if (btrfs_item_offset_nr(leaf, slot) !=
596 btrfs_item_end_nr(leaf, slot + 1)) {
597 CORRUPT("slot offset bad", leaf, root, slot);
602 * Check to make sure that we don't point outside of the leaf,
603 * just in case all the items are consistent to each other, but
604 * all point outside of the leaf.
606 if (btrfs_item_end_nr(leaf, slot) >
607 BTRFS_LEAF_DATA_SIZE(root)) {
608 CORRUPT("slot end outside of leaf", leaf, root, slot);
616 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
617 u64 phy_offset, struct page *page,
618 u64 start, u64 end, int mirror)
622 struct extent_buffer *eb;
623 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
624 struct btrfs_fs_info *fs_info = root->fs_info;
631 eb = (struct extent_buffer *)page->private;
633 /* the pending IO might have been the only thing that kept this buffer
634 * in memory. Make sure we have a ref for all this other checks
636 extent_buffer_get(eb);
638 reads_done = atomic_dec_and_test(&eb->io_pages);
642 eb->read_mirror = mirror;
643 if (test_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags)) {
648 found_start = btrfs_header_bytenr(eb);
649 if (found_start != eb->start) {
650 btrfs_err_rl(fs_info, "bad tree block start %llu %llu",
651 found_start, eb->start);
655 if (check_tree_block_fsid(fs_info, eb)) {
656 btrfs_err_rl(fs_info, "bad fsid on block %llu",
661 found_level = btrfs_header_level(eb);
662 if (found_level >= BTRFS_MAX_LEVEL) {
663 btrfs_err(fs_info, "bad tree block level %d",
664 (int)btrfs_header_level(eb));
669 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
672 ret = csum_tree_block(fs_info, eb, 1);
677 * If this is a leaf block and it is corrupt, set the corrupt bit so
678 * that we don't try and read the other copies of this block, just
681 if (found_level == 0 && check_leaf(root, eb)) {
682 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
687 set_extent_buffer_uptodate(eb);
690 test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
691 btree_readahead_hook(fs_info, eb, eb->start, ret);
695 * our io error hook is going to dec the io pages
696 * again, we have to make sure it has something
699 atomic_inc(&eb->io_pages);
700 clear_extent_buffer_uptodate(eb);
702 free_extent_buffer(eb);
707 static int btree_io_failed_hook(struct page *page, int failed_mirror)
709 struct extent_buffer *eb;
711 eb = (struct extent_buffer *)page->private;
712 set_bit(EXTENT_BUFFER_READ_ERR, &eb->bflags);
713 eb->read_mirror = failed_mirror;
714 atomic_dec(&eb->io_pages);
715 if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
716 btree_readahead_hook(eb->fs_info, eb, eb->start, -EIO);
717 return -EIO; /* we fixed nothing */
720 static void end_workqueue_bio(struct bio *bio)
722 struct btrfs_end_io_wq *end_io_wq = bio->bi_private;
723 struct btrfs_fs_info *fs_info;
724 struct btrfs_workqueue *wq;
725 btrfs_work_func_t func;
727 fs_info = end_io_wq->info;
728 end_io_wq->error = bio->bi_error;
730 if (bio->bi_rw & REQ_WRITE) {
731 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA) {
732 wq = fs_info->endio_meta_write_workers;
733 func = btrfs_endio_meta_write_helper;
734 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE) {
735 wq = fs_info->endio_freespace_worker;
736 func = btrfs_freespace_write_helper;
737 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
738 wq = fs_info->endio_raid56_workers;
739 func = btrfs_endio_raid56_helper;
741 wq = fs_info->endio_write_workers;
742 func = btrfs_endio_write_helper;
745 if (unlikely(end_io_wq->metadata ==
746 BTRFS_WQ_ENDIO_DIO_REPAIR)) {
747 wq = fs_info->endio_repair_workers;
748 func = btrfs_endio_repair_helper;
749 } else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56) {
750 wq = fs_info->endio_raid56_workers;
751 func = btrfs_endio_raid56_helper;
752 } else if (end_io_wq->metadata) {
753 wq = fs_info->endio_meta_workers;
754 func = btrfs_endio_meta_helper;
756 wq = fs_info->endio_workers;
757 func = btrfs_endio_helper;
761 btrfs_init_work(&end_io_wq->work, func, end_workqueue_fn, NULL, NULL);
762 btrfs_queue_work(wq, &end_io_wq->work);
765 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
766 enum btrfs_wq_endio_type metadata)
768 struct btrfs_end_io_wq *end_io_wq;
770 end_io_wq = kmem_cache_alloc(btrfs_end_io_wq_cache, GFP_NOFS);
774 end_io_wq->private = bio->bi_private;
775 end_io_wq->end_io = bio->bi_end_io;
776 end_io_wq->info = info;
777 end_io_wq->error = 0;
778 end_io_wq->bio = bio;
779 end_io_wq->metadata = metadata;
781 bio->bi_private = end_io_wq;
782 bio->bi_end_io = end_workqueue_bio;
786 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
788 unsigned long limit = min_t(unsigned long,
789 info->thread_pool_size,
790 info->fs_devices->open_devices);
794 static void run_one_async_start(struct btrfs_work *work)
796 struct async_submit_bio *async;
799 async = container_of(work, struct async_submit_bio, work);
800 ret = async->submit_bio_start(async->inode, async->rw, async->bio,
801 async->mirror_num, async->bio_flags,
807 static void run_one_async_done(struct btrfs_work *work)
809 struct btrfs_fs_info *fs_info;
810 struct async_submit_bio *async;
813 async = container_of(work, struct async_submit_bio, work);
814 fs_info = BTRFS_I(async->inode)->root->fs_info;
816 limit = btrfs_async_submit_limit(fs_info);
817 limit = limit * 2 / 3;
820 * atomic_dec_return implies a barrier for waitqueue_active
822 if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
823 waitqueue_active(&fs_info->async_submit_wait))
824 wake_up(&fs_info->async_submit_wait);
826 /* If an error occurred we just want to clean up the bio and move on */
828 async->bio->bi_error = async->error;
829 bio_endio(async->bio);
833 async->submit_bio_done(async->inode, async->rw, async->bio,
834 async->mirror_num, async->bio_flags,
838 static void run_one_async_free(struct btrfs_work *work)
840 struct async_submit_bio *async;
842 async = container_of(work, struct async_submit_bio, work);
846 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
847 int rw, struct bio *bio, int mirror_num,
848 unsigned long bio_flags,
850 extent_submit_bio_hook_t *submit_bio_start,
851 extent_submit_bio_hook_t *submit_bio_done)
853 struct async_submit_bio *async;
855 async = kmalloc(sizeof(*async), GFP_NOFS);
859 async->inode = inode;
862 async->mirror_num = mirror_num;
863 async->submit_bio_start = submit_bio_start;
864 async->submit_bio_done = submit_bio_done;
866 btrfs_init_work(&async->work, btrfs_worker_helper, run_one_async_start,
867 run_one_async_done, run_one_async_free);
869 async->bio_flags = bio_flags;
870 async->bio_offset = bio_offset;
874 atomic_inc(&fs_info->nr_async_submits);
877 btrfs_set_work_high_priority(&async->work);
879 btrfs_queue_work(fs_info->workers, &async->work);
881 while (atomic_read(&fs_info->async_submit_draining) &&
882 atomic_read(&fs_info->nr_async_submits)) {
883 wait_event(fs_info->async_submit_wait,
884 (atomic_read(&fs_info->nr_async_submits) == 0));
890 static int btree_csum_one_bio(struct bio *bio)
892 struct bio_vec *bvec;
893 struct btrfs_root *root;
896 bio_for_each_segment_all(bvec, bio, i) {
897 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
898 ret = csum_dirty_buffer(root->fs_info, bvec->bv_page);
906 static int __btree_submit_bio_start(struct inode *inode, int rw,
907 struct bio *bio, int mirror_num,
908 unsigned long bio_flags,
912 * when we're called for a write, we're already in the async
913 * submission context. Just jump into btrfs_map_bio
915 return btree_csum_one_bio(bio);
918 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
919 int mirror_num, unsigned long bio_flags,
925 * when we're called for a write, we're already in the async
926 * submission context. Just jump into btrfs_map_bio
928 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
936 static int check_async_write(struct inode *inode, unsigned long bio_flags)
938 if (bio_flags & EXTENT_BIO_TREE_LOG)
941 if (static_cpu_has(X86_FEATURE_XMM4_2))
947 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
948 int mirror_num, unsigned long bio_flags,
951 int async = check_async_write(inode, bio_flags);
954 if (!(rw & REQ_WRITE)) {
956 * called for a read, do the setup so that checksum validation
957 * can happen in the async kernel threads
959 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
960 bio, BTRFS_WQ_ENDIO_METADATA);
963 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
966 ret = btree_csum_one_bio(bio);
969 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
973 * kthread helpers are used to submit writes so that
974 * checksumming can happen in parallel across all CPUs
976 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
977 inode, rw, bio, mirror_num, 0,
979 __btree_submit_bio_start,
980 __btree_submit_bio_done);
993 #ifdef CONFIG_MIGRATION
994 static int btree_migratepage(struct address_space *mapping,
995 struct page *newpage, struct page *page,
996 enum migrate_mode mode)
999 * we can't safely write a btree page from here,
1000 * we haven't done the locking hook
1002 if (PageDirty(page))
1005 * Buffers may be managed in a filesystem specific way.
1006 * We must have no buffers or drop them.
1008 if (page_has_private(page) &&
1009 !try_to_release_page(page, GFP_KERNEL))
1011 return migrate_page(mapping, newpage, page, mode);
1016 static int btree_writepages(struct address_space *mapping,
1017 struct writeback_control *wbc)
1019 struct btrfs_fs_info *fs_info;
1022 if (wbc->sync_mode == WB_SYNC_NONE) {
1024 if (wbc->for_kupdate)
1027 fs_info = BTRFS_I(mapping->host)->root->fs_info;
1028 /* this is a bit racy, but that's ok */
1029 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
1030 BTRFS_DIRTY_METADATA_THRESH);
1034 return btree_write_cache_pages(mapping, wbc);
1037 static int btree_readpage(struct file *file, struct page *page)
1039 struct extent_io_tree *tree;
1040 tree = &BTRFS_I(page->mapping->host)->io_tree;
1041 return extent_read_full_page(tree, page, btree_get_extent, 0);
1044 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
1046 if (PageWriteback(page) || PageDirty(page))
1049 return try_release_extent_buffer(page);
1052 static void btree_invalidatepage(struct page *page, unsigned int offset,
1053 unsigned int length)
1055 struct extent_io_tree *tree;
1056 tree = &BTRFS_I(page->mapping->host)->io_tree;
1057 extent_invalidatepage(tree, page, offset);
1058 btree_releasepage(page, GFP_NOFS);
1059 if (PagePrivate(page)) {
1060 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1061 "page private not zero on page %llu",
1062 (unsigned long long)page_offset(page));
1063 ClearPagePrivate(page);
1064 set_page_private(page, 0);
1069 static int btree_set_page_dirty(struct page *page)
1072 struct extent_buffer *eb;
1074 BUG_ON(!PagePrivate(page));
1075 eb = (struct extent_buffer *)page->private;
1077 BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1078 BUG_ON(!atomic_read(&eb->refs));
1079 btrfs_assert_tree_locked(eb);
1081 return __set_page_dirty_nobuffers(page);
1084 static const struct address_space_operations btree_aops = {
1085 .readpage = btree_readpage,
1086 .writepages = btree_writepages,
1087 .releasepage = btree_releasepage,
1088 .invalidatepage = btree_invalidatepage,
1089 #ifdef CONFIG_MIGRATION
1090 .migratepage = btree_migratepage,
1092 .set_page_dirty = btree_set_page_dirty,
1095 void readahead_tree_block(struct btrfs_root *root, u64 bytenr)
1097 struct extent_buffer *buf = NULL;
1098 struct inode *btree_inode = root->fs_info->btree_inode;
1100 buf = btrfs_find_create_tree_block(root, bytenr);
1103 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1104 buf, 0, WAIT_NONE, btree_get_extent, 0);
1105 free_extent_buffer(buf);
1108 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr,
1109 int mirror_num, struct extent_buffer **eb)
1111 struct extent_buffer *buf = NULL;
1112 struct inode *btree_inode = root->fs_info->btree_inode;
1113 struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1116 buf = btrfs_find_create_tree_block(root, bytenr);
1120 set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1122 ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1123 btree_get_extent, mirror_num);
1125 free_extent_buffer(buf);
1129 if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1130 free_extent_buffer(buf);
1132 } else if (extent_buffer_uptodate(buf)) {
1135 free_extent_buffer(buf);
1140 struct extent_buffer *btrfs_find_tree_block(struct btrfs_fs_info *fs_info,
1143 return find_extent_buffer(fs_info, bytenr);
1146 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1149 if (btrfs_is_testing(root->fs_info))
1150 return alloc_test_extent_buffer(root->fs_info, bytenr,
1152 return alloc_extent_buffer(root->fs_info, bytenr);
1156 int btrfs_write_tree_block(struct extent_buffer *buf)
1158 return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1159 buf->start + buf->len - 1);
1162 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1164 return filemap_fdatawait_range(buf->pages[0]->mapping,
1165 buf->start, buf->start + buf->len - 1);
1168 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1171 struct extent_buffer *buf = NULL;
1174 buf = btrfs_find_create_tree_block(root, bytenr);
1178 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1180 free_extent_buffer(buf);
1181 return ERR_PTR(ret);
1187 void clean_tree_block(struct btrfs_trans_handle *trans,
1188 struct btrfs_fs_info *fs_info,
1189 struct extent_buffer *buf)
1191 if (btrfs_header_generation(buf) ==
1192 fs_info->running_transaction->transid) {
1193 btrfs_assert_tree_locked(buf);
1195 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1196 __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1198 fs_info->dirty_metadata_batch);
1199 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1200 btrfs_set_lock_blocking(buf);
1201 clear_extent_buffer_dirty(buf);
1206 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1208 struct btrfs_subvolume_writers *writers;
1211 writers = kmalloc(sizeof(*writers), GFP_NOFS);
1213 return ERR_PTR(-ENOMEM);
1215 ret = percpu_counter_init(&writers->counter, 0, GFP_KERNEL);
1218 return ERR_PTR(ret);
1221 init_waitqueue_head(&writers->wait);
1226 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1228 percpu_counter_destroy(&writers->counter);
1232 static void __setup_root(u32 nodesize, u32 sectorsize, u32 stripesize,
1233 struct btrfs_root *root, struct btrfs_fs_info *fs_info,
1236 bool dummy = test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state);
1238 root->commit_root = NULL;
1239 root->sectorsize = sectorsize;
1240 root->nodesize = nodesize;
1241 root->stripesize = stripesize;
1243 root->orphan_cleanup_state = 0;
1245 root->objectid = objectid;
1246 root->last_trans = 0;
1247 root->highest_objectid = 0;
1248 root->nr_delalloc_inodes = 0;
1249 root->nr_ordered_extents = 0;
1251 root->inode_tree = RB_ROOT;
1252 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1253 root->block_rsv = NULL;
1254 root->orphan_block_rsv = NULL;
1256 INIT_LIST_HEAD(&root->dirty_list);
1257 INIT_LIST_HEAD(&root->root_list);
1258 INIT_LIST_HEAD(&root->delalloc_inodes);
1259 INIT_LIST_HEAD(&root->delalloc_root);
1260 INIT_LIST_HEAD(&root->ordered_extents);
1261 INIT_LIST_HEAD(&root->ordered_root);
1262 INIT_LIST_HEAD(&root->logged_list[0]);
1263 INIT_LIST_HEAD(&root->logged_list[1]);
1264 spin_lock_init(&root->orphan_lock);
1265 spin_lock_init(&root->inode_lock);
1266 spin_lock_init(&root->delalloc_lock);
1267 spin_lock_init(&root->ordered_extent_lock);
1268 spin_lock_init(&root->accounting_lock);
1269 spin_lock_init(&root->log_extents_lock[0]);
1270 spin_lock_init(&root->log_extents_lock[1]);
1271 mutex_init(&root->objectid_mutex);
1272 mutex_init(&root->log_mutex);
1273 mutex_init(&root->ordered_extent_mutex);
1274 mutex_init(&root->delalloc_mutex);
1275 init_waitqueue_head(&root->log_writer_wait);
1276 init_waitqueue_head(&root->log_commit_wait[0]);
1277 init_waitqueue_head(&root->log_commit_wait[1]);
1278 INIT_LIST_HEAD(&root->log_ctxs[0]);
1279 INIT_LIST_HEAD(&root->log_ctxs[1]);
1280 atomic_set(&root->log_commit[0], 0);
1281 atomic_set(&root->log_commit[1], 0);
1282 atomic_set(&root->log_writers, 0);
1283 atomic_set(&root->log_batch, 0);
1284 atomic_set(&root->orphan_inodes, 0);
1285 atomic_set(&root->refs, 1);
1286 atomic_set(&root->will_be_snapshoted, 0);
1287 atomic_set(&root->qgroup_meta_rsv, 0);
1288 root->log_transid = 0;
1289 root->log_transid_committed = -1;
1290 root->last_log_commit = 0;
1292 extent_io_tree_init(&root->dirty_log_pages,
1293 fs_info->btree_inode->i_mapping);
1295 memset(&root->root_key, 0, sizeof(root->root_key));
1296 memset(&root->root_item, 0, sizeof(root->root_item));
1297 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1299 root->defrag_trans_start = fs_info->generation;
1301 root->defrag_trans_start = 0;
1302 root->root_key.objectid = objectid;
1305 spin_lock_init(&root->root_item_lock);
1308 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info,
1311 struct btrfs_root *root = kzalloc(sizeof(*root), flags);
1313 root->fs_info = fs_info;
1317 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1318 /* Should only be used by the testing infrastructure */
1319 struct btrfs_root *btrfs_alloc_dummy_root(struct btrfs_fs_info *fs_info,
1320 u32 sectorsize, u32 nodesize)
1322 struct btrfs_root *root;
1325 return ERR_PTR(-EINVAL);
1327 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1329 return ERR_PTR(-ENOMEM);
1330 /* We don't use the stripesize in selftest, set it as sectorsize */
1331 __setup_root(nodesize, sectorsize, sectorsize, root, fs_info,
1332 BTRFS_ROOT_TREE_OBJECTID);
1333 root->alloc_bytenr = 0;
1339 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1340 struct btrfs_fs_info *fs_info,
1343 struct extent_buffer *leaf;
1344 struct btrfs_root *tree_root = fs_info->tree_root;
1345 struct btrfs_root *root;
1346 struct btrfs_key key;
1350 root = btrfs_alloc_root(fs_info, GFP_KERNEL);
1352 return ERR_PTR(-ENOMEM);
1354 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1355 tree_root->stripesize, root, fs_info, objectid);
1356 root->root_key.objectid = objectid;
1357 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1358 root->root_key.offset = 0;
1360 leaf = btrfs_alloc_tree_block(trans, root, 0, objectid, NULL, 0, 0, 0);
1362 ret = PTR_ERR(leaf);
1367 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1368 btrfs_set_header_bytenr(leaf, leaf->start);
1369 btrfs_set_header_generation(leaf, trans->transid);
1370 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1371 btrfs_set_header_owner(leaf, objectid);
1374 write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1376 write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1377 btrfs_header_chunk_tree_uuid(leaf),
1379 btrfs_mark_buffer_dirty(leaf);
1381 root->commit_root = btrfs_root_node(root);
1382 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
1384 root->root_item.flags = 0;
1385 root->root_item.byte_limit = 0;
1386 btrfs_set_root_bytenr(&root->root_item, leaf->start);
1387 btrfs_set_root_generation(&root->root_item, trans->transid);
1388 btrfs_set_root_level(&root->root_item, 0);
1389 btrfs_set_root_refs(&root->root_item, 1);
1390 btrfs_set_root_used(&root->root_item, leaf->len);
1391 btrfs_set_root_last_snapshot(&root->root_item, 0);
1392 btrfs_set_root_dirid(&root->root_item, 0);
1394 memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1395 root->root_item.drop_level = 0;
1397 key.objectid = objectid;
1398 key.type = BTRFS_ROOT_ITEM_KEY;
1400 ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1404 btrfs_tree_unlock(leaf);
1410 btrfs_tree_unlock(leaf);
1411 free_extent_buffer(root->commit_root);
1412 free_extent_buffer(leaf);
1416 return ERR_PTR(ret);
1419 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1420 struct btrfs_fs_info *fs_info)
1422 struct btrfs_root *root;
1423 struct btrfs_root *tree_root = fs_info->tree_root;
1424 struct extent_buffer *leaf;
1426 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1428 return ERR_PTR(-ENOMEM);
1430 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1431 tree_root->stripesize, root, fs_info,
1432 BTRFS_TREE_LOG_OBJECTID);
1434 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1435 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1436 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1439 * DON'T set REF_COWS for log trees
1441 * log trees do not get reference counted because they go away
1442 * before a real commit is actually done. They do store pointers
1443 * to file data extents, and those reference counts still get
1444 * updated (along with back refs to the log tree).
1447 leaf = btrfs_alloc_tree_block(trans, root, 0, BTRFS_TREE_LOG_OBJECTID,
1451 return ERR_CAST(leaf);
1454 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1455 btrfs_set_header_bytenr(leaf, leaf->start);
1456 btrfs_set_header_generation(leaf, trans->transid);
1457 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1458 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1461 write_extent_buffer(root->node, root->fs_info->fsid,
1462 btrfs_header_fsid(), BTRFS_FSID_SIZE);
1463 btrfs_mark_buffer_dirty(root->node);
1464 btrfs_tree_unlock(root->node);
1468 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1469 struct btrfs_fs_info *fs_info)
1471 struct btrfs_root *log_root;
1473 log_root = alloc_log_tree(trans, fs_info);
1474 if (IS_ERR(log_root))
1475 return PTR_ERR(log_root);
1476 WARN_ON(fs_info->log_root_tree);
1477 fs_info->log_root_tree = log_root;
1481 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1482 struct btrfs_root *root)
1484 struct btrfs_root *log_root;
1485 struct btrfs_inode_item *inode_item;
1487 log_root = alloc_log_tree(trans, root->fs_info);
1488 if (IS_ERR(log_root))
1489 return PTR_ERR(log_root);
1491 log_root->last_trans = trans->transid;
1492 log_root->root_key.offset = root->root_key.objectid;
1494 inode_item = &log_root->root_item.inode;
1495 btrfs_set_stack_inode_generation(inode_item, 1);
1496 btrfs_set_stack_inode_size(inode_item, 3);
1497 btrfs_set_stack_inode_nlink(inode_item, 1);
1498 btrfs_set_stack_inode_nbytes(inode_item, root->nodesize);
1499 btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1501 btrfs_set_root_node(&log_root->root_item, log_root->node);
1503 WARN_ON(root->log_root);
1504 root->log_root = log_root;
1505 root->log_transid = 0;
1506 root->log_transid_committed = -1;
1507 root->last_log_commit = 0;
1511 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1512 struct btrfs_key *key)
1514 struct btrfs_root *root;
1515 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1516 struct btrfs_path *path;
1520 path = btrfs_alloc_path();
1522 return ERR_PTR(-ENOMEM);
1524 root = btrfs_alloc_root(fs_info, GFP_NOFS);
1530 __setup_root(tree_root->nodesize, tree_root->sectorsize,
1531 tree_root->stripesize, root, fs_info, key->objectid);
1533 ret = btrfs_find_root(tree_root, key, path,
1534 &root->root_item, &root->root_key);
1541 generation = btrfs_root_generation(&root->root_item);
1542 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1544 if (IS_ERR(root->node)) {
1545 ret = PTR_ERR(root->node);
1547 } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1549 free_extent_buffer(root->node);
1552 root->commit_root = btrfs_root_node(root);
1554 btrfs_free_path(path);
1560 root = ERR_PTR(ret);
1564 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1565 struct btrfs_key *location)
1567 struct btrfs_root *root;
1569 root = btrfs_read_tree_root(tree_root, location);
1573 if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1574 set_bit(BTRFS_ROOT_REF_COWS, &root->state);
1575 btrfs_check_and_init_root_item(&root->root_item);
1581 int btrfs_init_fs_root(struct btrfs_root *root)
1584 struct btrfs_subvolume_writers *writers;
1586 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1587 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1589 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1594 writers = btrfs_alloc_subvolume_writers();
1595 if (IS_ERR(writers)) {
1596 ret = PTR_ERR(writers);
1599 root->subv_writers = writers;
1601 btrfs_init_free_ino_ctl(root);
1602 spin_lock_init(&root->ino_cache_lock);
1603 init_waitqueue_head(&root->ino_cache_wait);
1605 ret = get_anon_bdev(&root->anon_dev);
1609 mutex_lock(&root->objectid_mutex);
1610 ret = btrfs_find_highest_objectid(root,
1611 &root->highest_objectid);
1613 mutex_unlock(&root->objectid_mutex);
1617 ASSERT(root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
1619 mutex_unlock(&root->objectid_mutex);
1623 /* the caller is responsible to call free_fs_root */
1627 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1630 struct btrfs_root *root;
1632 spin_lock(&fs_info->fs_roots_radix_lock);
1633 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1634 (unsigned long)root_id);
1635 spin_unlock(&fs_info->fs_roots_radix_lock);
1639 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1640 struct btrfs_root *root)
1644 ret = radix_tree_preload(GFP_NOFS);
1648 spin_lock(&fs_info->fs_roots_radix_lock);
1649 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1650 (unsigned long)root->root_key.objectid,
1653 set_bit(BTRFS_ROOT_IN_RADIX, &root->state);
1654 spin_unlock(&fs_info->fs_roots_radix_lock);
1655 radix_tree_preload_end();
1660 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1661 struct btrfs_key *location,
1664 struct btrfs_root *root;
1665 struct btrfs_path *path;
1666 struct btrfs_key key;
1669 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1670 return fs_info->tree_root;
1671 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1672 return fs_info->extent_root;
1673 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1674 return fs_info->chunk_root;
1675 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1676 return fs_info->dev_root;
1677 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1678 return fs_info->csum_root;
1679 if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1680 return fs_info->quota_root ? fs_info->quota_root :
1682 if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1683 return fs_info->uuid_root ? fs_info->uuid_root :
1685 if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
1686 return fs_info->free_space_root ? fs_info->free_space_root :
1689 root = btrfs_lookup_fs_root(fs_info, location->objectid);
1691 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1692 return ERR_PTR(-ENOENT);
1696 root = btrfs_read_fs_root(fs_info->tree_root, location);
1700 if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1705 ret = btrfs_init_fs_root(root);
1709 path = btrfs_alloc_path();
1714 key.objectid = BTRFS_ORPHAN_OBJECTID;
1715 key.type = BTRFS_ORPHAN_ITEM_KEY;
1716 key.offset = location->objectid;
1718 ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
1719 btrfs_free_path(path);
1723 set_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state);
1725 ret = btrfs_insert_fs_root(fs_info, root);
1727 if (ret == -EEXIST) {
1736 return ERR_PTR(ret);
1739 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1741 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1743 struct btrfs_device *device;
1744 struct backing_dev_info *bdi;
1747 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1750 bdi = blk_get_backing_dev_info(device->bdev);
1751 if (bdi_congested(bdi, bdi_bits)) {
1760 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1764 err = bdi_setup_and_register(bdi, "btrfs");
1768 bdi->ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_SIZE;
1769 bdi->congested_fn = btrfs_congested_fn;
1770 bdi->congested_data = info;
1771 bdi->capabilities |= BDI_CAP_CGROUP_WRITEBACK;
1776 * called by the kthread helper functions to finally call the bio end_io
1777 * functions. This is where read checksum verification actually happens
1779 static void end_workqueue_fn(struct btrfs_work *work)
1782 struct btrfs_end_io_wq *end_io_wq;
1784 end_io_wq = container_of(work, struct btrfs_end_io_wq, work);
1785 bio = end_io_wq->bio;
1787 bio->bi_error = end_io_wq->error;
1788 bio->bi_private = end_io_wq->private;
1789 bio->bi_end_io = end_io_wq->end_io;
1790 kmem_cache_free(btrfs_end_io_wq_cache, end_io_wq);
1794 static int cleaner_kthread(void *arg)
1796 struct btrfs_root *root = arg;
1798 struct btrfs_trans_handle *trans;
1803 /* Make the cleaner go to sleep early. */
1804 if (btrfs_need_cleaner_sleep(root))
1808 * Do not do anything if we might cause open_ctree() to block
1809 * before we have finished mounting the filesystem.
1811 if (!root->fs_info->open)
1814 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1818 * Avoid the problem that we change the status of the fs
1819 * during the above check and trylock.
1821 if (btrfs_need_cleaner_sleep(root)) {
1822 mutex_unlock(&root->fs_info->cleaner_mutex);
1826 mutex_lock(&root->fs_info->cleaner_delayed_iput_mutex);
1827 btrfs_run_delayed_iputs(root);
1828 mutex_unlock(&root->fs_info->cleaner_delayed_iput_mutex);
1830 again = btrfs_clean_one_deleted_snapshot(root);
1831 mutex_unlock(&root->fs_info->cleaner_mutex);
1834 * The defragger has dealt with the R/O remount and umount,
1835 * needn't do anything special here.
1837 btrfs_run_defrag_inodes(root->fs_info);
1840 * Acquires fs_info->delete_unused_bgs_mutex to avoid racing
1841 * with relocation (btrfs_relocate_chunk) and relocation
1842 * acquires fs_info->cleaner_mutex (btrfs_relocate_block_group)
1843 * after acquiring fs_info->delete_unused_bgs_mutex. So we
1844 * can't hold, nor need to, fs_info->cleaner_mutex when deleting
1845 * unused block groups.
1847 btrfs_delete_unused_bgs(root->fs_info);
1850 set_current_state(TASK_INTERRUPTIBLE);
1851 if (!kthread_should_stop())
1853 __set_current_state(TASK_RUNNING);
1855 } while (!kthread_should_stop());
1858 * Transaction kthread is stopped before us and wakes us up.
1859 * However we might have started a new transaction and COWed some
1860 * tree blocks when deleting unused block groups for example. So
1861 * make sure we commit the transaction we started to have a clean
1862 * shutdown when evicting the btree inode - if it has dirty pages
1863 * when we do the final iput() on it, eviction will trigger a
1864 * writeback for it which will fail with null pointer dereferences
1865 * since work queues and other resources were already released and
1866 * destroyed by the time the iput/eviction/writeback is made.
1868 trans = btrfs_attach_transaction(root);
1869 if (IS_ERR(trans)) {
1870 if (PTR_ERR(trans) != -ENOENT)
1871 btrfs_err(root->fs_info,
1872 "cleaner transaction attach returned %ld",
1877 ret = btrfs_commit_transaction(trans, root);
1879 btrfs_err(root->fs_info,
1880 "cleaner open transaction commit returned %d",
1887 static int transaction_kthread(void *arg)
1889 struct btrfs_root *root = arg;
1890 struct btrfs_trans_handle *trans;
1891 struct btrfs_transaction *cur;
1894 unsigned long delay;
1898 cannot_commit = false;
1899 delay = HZ * root->fs_info->commit_interval;
1900 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1902 spin_lock(&root->fs_info->trans_lock);
1903 cur = root->fs_info->running_transaction;
1905 spin_unlock(&root->fs_info->trans_lock);
1909 now = get_seconds();
1910 if (cur->state < TRANS_STATE_BLOCKED &&
1911 (now < cur->start_time ||
1912 now - cur->start_time < root->fs_info->commit_interval)) {
1913 spin_unlock(&root->fs_info->trans_lock);
1917 transid = cur->transid;
1918 spin_unlock(&root->fs_info->trans_lock);
1920 /* If the file system is aborted, this will always fail. */
1921 trans = btrfs_attach_transaction(root);
1922 if (IS_ERR(trans)) {
1923 if (PTR_ERR(trans) != -ENOENT)
1924 cannot_commit = true;
1927 if (transid == trans->transid) {
1928 btrfs_commit_transaction(trans, root);
1930 btrfs_end_transaction(trans, root);
1933 wake_up_process(root->fs_info->cleaner_kthread);
1934 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1936 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1937 &root->fs_info->fs_state)))
1938 btrfs_cleanup_transaction(root);
1939 set_current_state(TASK_INTERRUPTIBLE);
1940 if (!kthread_should_stop() &&
1941 (!btrfs_transaction_blocked(root->fs_info) ||
1943 schedule_timeout(delay);
1944 __set_current_state(TASK_RUNNING);
1945 } while (!kthread_should_stop());
1950 * this will find the highest generation in the array of
1951 * root backups. The index of the highest array is returned,
1952 * or -1 if we can't find anything.
1954 * We check to make sure the array is valid by comparing the
1955 * generation of the latest root in the array with the generation
1956 * in the super block. If they don't match we pitch it.
1958 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1961 int newest_index = -1;
1962 struct btrfs_root_backup *root_backup;
1965 for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1966 root_backup = info->super_copy->super_roots + i;
1967 cur = btrfs_backup_tree_root_gen(root_backup);
1968 if (cur == newest_gen)
1972 /* check to see if we actually wrapped around */
1973 if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1974 root_backup = info->super_copy->super_roots;
1975 cur = btrfs_backup_tree_root_gen(root_backup);
1976 if (cur == newest_gen)
1979 return newest_index;
1984 * find the oldest backup so we know where to store new entries
1985 * in the backup array. This will set the backup_root_index
1986 * field in the fs_info struct
1988 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1991 int newest_index = -1;
1993 newest_index = find_newest_super_backup(info, newest_gen);
1994 /* if there was garbage in there, just move along */
1995 if (newest_index == -1) {
1996 info->backup_root_index = 0;
1998 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
2003 * copy all the root pointers into the super backup array.
2004 * this will bump the backup pointer by one when it is
2007 static void backup_super_roots(struct btrfs_fs_info *info)
2010 struct btrfs_root_backup *root_backup;
2013 next_backup = info->backup_root_index;
2014 last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
2015 BTRFS_NUM_BACKUP_ROOTS;
2018 * just overwrite the last backup if we're at the same generation
2019 * this happens only at umount
2021 root_backup = info->super_for_commit->super_roots + last_backup;
2022 if (btrfs_backup_tree_root_gen(root_backup) ==
2023 btrfs_header_generation(info->tree_root->node))
2024 next_backup = last_backup;
2026 root_backup = info->super_for_commit->super_roots + next_backup;
2029 * make sure all of our padding and empty slots get zero filled
2030 * regardless of which ones we use today
2032 memset(root_backup, 0, sizeof(*root_backup));
2034 info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
2036 btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
2037 btrfs_set_backup_tree_root_gen(root_backup,
2038 btrfs_header_generation(info->tree_root->node));
2040 btrfs_set_backup_tree_root_level(root_backup,
2041 btrfs_header_level(info->tree_root->node));
2043 btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
2044 btrfs_set_backup_chunk_root_gen(root_backup,
2045 btrfs_header_generation(info->chunk_root->node));
2046 btrfs_set_backup_chunk_root_level(root_backup,
2047 btrfs_header_level(info->chunk_root->node));
2049 btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
2050 btrfs_set_backup_extent_root_gen(root_backup,
2051 btrfs_header_generation(info->extent_root->node));
2052 btrfs_set_backup_extent_root_level(root_backup,
2053 btrfs_header_level(info->extent_root->node));
2056 * we might commit during log recovery, which happens before we set
2057 * the fs_root. Make sure it is valid before we fill it in.
2059 if (info->fs_root && info->fs_root->node) {
2060 btrfs_set_backup_fs_root(root_backup,
2061 info->fs_root->node->start);
2062 btrfs_set_backup_fs_root_gen(root_backup,
2063 btrfs_header_generation(info->fs_root->node));
2064 btrfs_set_backup_fs_root_level(root_backup,
2065 btrfs_header_level(info->fs_root->node));
2068 btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
2069 btrfs_set_backup_dev_root_gen(root_backup,
2070 btrfs_header_generation(info->dev_root->node));
2071 btrfs_set_backup_dev_root_level(root_backup,
2072 btrfs_header_level(info->dev_root->node));
2074 btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
2075 btrfs_set_backup_csum_root_gen(root_backup,
2076 btrfs_header_generation(info->csum_root->node));
2077 btrfs_set_backup_csum_root_level(root_backup,
2078 btrfs_header_level(info->csum_root->node));
2080 btrfs_set_backup_total_bytes(root_backup,
2081 btrfs_super_total_bytes(info->super_copy));
2082 btrfs_set_backup_bytes_used(root_backup,
2083 btrfs_super_bytes_used(info->super_copy));
2084 btrfs_set_backup_num_devices(root_backup,
2085 btrfs_super_num_devices(info->super_copy));
2088 * if we don't copy this out to the super_copy, it won't get remembered
2089 * for the next commit
2091 memcpy(&info->super_copy->super_roots,
2092 &info->super_for_commit->super_roots,
2093 sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
2097 * this copies info out of the root backup array and back into
2098 * the in-memory super block. It is meant to help iterate through
2099 * the array, so you send it the number of backups you've already
2100 * tried and the last backup index you used.
2102 * this returns -1 when it has tried all the backups
2104 static noinline int next_root_backup(struct btrfs_fs_info *info,
2105 struct btrfs_super_block *super,
2106 int *num_backups_tried, int *backup_index)
2108 struct btrfs_root_backup *root_backup;
2109 int newest = *backup_index;
2111 if (*num_backups_tried == 0) {
2112 u64 gen = btrfs_super_generation(super);
2114 newest = find_newest_super_backup(info, gen);
2118 *backup_index = newest;
2119 *num_backups_tried = 1;
2120 } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2121 /* we've tried all the backups, all done */
2124 /* jump to the next oldest backup */
2125 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2126 BTRFS_NUM_BACKUP_ROOTS;
2127 *backup_index = newest;
2128 *num_backups_tried += 1;
2130 root_backup = super->super_roots + newest;
2132 btrfs_set_super_generation(super,
2133 btrfs_backup_tree_root_gen(root_backup));
2134 btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2135 btrfs_set_super_root_level(super,
2136 btrfs_backup_tree_root_level(root_backup));
2137 btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2140 * fixme: the total bytes and num_devices need to match or we should
2143 btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2144 btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2148 /* helper to cleanup workers */
2149 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2151 btrfs_destroy_workqueue(fs_info->fixup_workers);
2152 btrfs_destroy_workqueue(fs_info->delalloc_workers);
2153 btrfs_destroy_workqueue(fs_info->workers);
2154 btrfs_destroy_workqueue(fs_info->endio_workers);
2155 btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2156 btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2157 btrfs_destroy_workqueue(fs_info->endio_repair_workers);
2158 btrfs_destroy_workqueue(fs_info->rmw_workers);
2159 btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2160 btrfs_destroy_workqueue(fs_info->endio_write_workers);
2161 btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2162 btrfs_destroy_workqueue(fs_info->submit_workers);
2163 btrfs_destroy_workqueue(fs_info->delayed_workers);
2164 btrfs_destroy_workqueue(fs_info->caching_workers);
2165 btrfs_destroy_workqueue(fs_info->readahead_workers);
2166 btrfs_destroy_workqueue(fs_info->flush_workers);
2167 btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2168 btrfs_destroy_workqueue(fs_info->extent_workers);
2171 static void free_root_extent_buffers(struct btrfs_root *root)
2174 free_extent_buffer(root->node);
2175 free_extent_buffer(root->commit_root);
2177 root->commit_root = NULL;
2181 /* helper to cleanup tree roots */
2182 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2184 free_root_extent_buffers(info->tree_root);
2186 free_root_extent_buffers(info->dev_root);
2187 free_root_extent_buffers(info->extent_root);
2188 free_root_extent_buffers(info->csum_root);
2189 free_root_extent_buffers(info->quota_root);
2190 free_root_extent_buffers(info->uuid_root);
2192 free_root_extent_buffers(info->chunk_root);
2193 free_root_extent_buffers(info->free_space_root);
2196 void btrfs_free_fs_roots(struct btrfs_fs_info *fs_info)
2199 struct btrfs_root *gang[8];
2202 while (!list_empty(&fs_info->dead_roots)) {
2203 gang[0] = list_entry(fs_info->dead_roots.next,
2204 struct btrfs_root, root_list);
2205 list_del(&gang[0]->root_list);
2207 if (test_bit(BTRFS_ROOT_IN_RADIX, &gang[0]->state)) {
2208 btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2210 free_extent_buffer(gang[0]->node);
2211 free_extent_buffer(gang[0]->commit_root);
2212 btrfs_put_fs_root(gang[0]);
2217 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2222 for (i = 0; i < ret; i++)
2223 btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2226 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2227 btrfs_free_log_root_tree(NULL, fs_info);
2228 btrfs_destroy_pinned_extent(fs_info->tree_root,
2229 fs_info->pinned_extents);
2233 static void btrfs_init_scrub(struct btrfs_fs_info *fs_info)
2235 mutex_init(&fs_info->scrub_lock);
2236 atomic_set(&fs_info->scrubs_running, 0);
2237 atomic_set(&fs_info->scrub_pause_req, 0);
2238 atomic_set(&fs_info->scrubs_paused, 0);
2239 atomic_set(&fs_info->scrub_cancel_req, 0);
2240 init_waitqueue_head(&fs_info->scrub_pause_wait);
2241 fs_info->scrub_workers_refcnt = 0;
2244 static void btrfs_init_balance(struct btrfs_fs_info *fs_info)
2246 spin_lock_init(&fs_info->balance_lock);
2247 mutex_init(&fs_info->balance_mutex);
2248 atomic_set(&fs_info->balance_running, 0);
2249 atomic_set(&fs_info->balance_pause_req, 0);
2250 atomic_set(&fs_info->balance_cancel_req, 0);
2251 fs_info->balance_ctl = NULL;
2252 init_waitqueue_head(&fs_info->balance_wait_q);
2255 static void btrfs_init_btree_inode(struct btrfs_fs_info *fs_info,
2256 struct btrfs_root *tree_root)
2258 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2259 set_nlink(fs_info->btree_inode, 1);
2261 * we set the i_size on the btree inode to the max possible int.
2262 * the real end of the address space is determined by all of
2263 * the devices in the system
2265 fs_info->btree_inode->i_size = OFFSET_MAX;
2266 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2268 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2269 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2270 fs_info->btree_inode->i_mapping);
2271 BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2272 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2274 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2276 BTRFS_I(fs_info->btree_inode)->root = tree_root;
2277 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2278 sizeof(struct btrfs_key));
2279 set_bit(BTRFS_INODE_DUMMY,
2280 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2281 btrfs_insert_inode_hash(fs_info->btree_inode);
2284 static void btrfs_init_dev_replace_locks(struct btrfs_fs_info *fs_info)
2286 fs_info->dev_replace.lock_owner = 0;
2287 atomic_set(&fs_info->dev_replace.nesting_level, 0);
2288 mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2289 rwlock_init(&fs_info->dev_replace.lock);
2290 atomic_set(&fs_info->dev_replace.read_locks, 0);
2291 atomic_set(&fs_info->dev_replace.blocking_readers, 0);
2292 init_waitqueue_head(&fs_info->replace_wait);
2293 init_waitqueue_head(&fs_info->dev_replace.read_lock_wq);
2296 static void btrfs_init_qgroup(struct btrfs_fs_info *fs_info)
2298 spin_lock_init(&fs_info->qgroup_lock);
2299 mutex_init(&fs_info->qgroup_ioctl_lock);
2300 fs_info->qgroup_tree = RB_ROOT;
2301 fs_info->qgroup_op_tree = RB_ROOT;
2302 INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2303 fs_info->qgroup_seq = 1;
2304 fs_info->quota_enabled = 0;
2305 fs_info->pending_quota_state = 0;
2306 fs_info->qgroup_ulist = NULL;
2307 fs_info->qgroup_rescan_running = false;
2308 mutex_init(&fs_info->qgroup_rescan_lock);
2311 static int btrfs_init_workqueues(struct btrfs_fs_info *fs_info,
2312 struct btrfs_fs_devices *fs_devices)
2314 int max_active = fs_info->thread_pool_size;
2315 unsigned int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2318 btrfs_alloc_workqueue(fs_info, "worker",
2319 flags | WQ_HIGHPRI, max_active, 16);
2321 fs_info->delalloc_workers =
2322 btrfs_alloc_workqueue(fs_info, "delalloc",
2323 flags, max_active, 2);
2325 fs_info->flush_workers =
2326 btrfs_alloc_workqueue(fs_info, "flush_delalloc",
2327 flags, max_active, 0);
2329 fs_info->caching_workers =
2330 btrfs_alloc_workqueue(fs_info, "cache", flags, max_active, 0);
2333 * a higher idle thresh on the submit workers makes it much more
2334 * likely that bios will be send down in a sane order to the
2337 fs_info->submit_workers =
2338 btrfs_alloc_workqueue(fs_info, "submit", flags,
2339 min_t(u64, fs_devices->num_devices,
2342 fs_info->fixup_workers =
2343 btrfs_alloc_workqueue(fs_info, "fixup", flags, 1, 0);
2346 * endios are largely parallel and should have a very
2349 fs_info->endio_workers =
2350 btrfs_alloc_workqueue(fs_info, "endio", flags, max_active, 4);
2351 fs_info->endio_meta_workers =
2352 btrfs_alloc_workqueue(fs_info, "endio-meta", flags,
2354 fs_info->endio_meta_write_workers =
2355 btrfs_alloc_workqueue(fs_info, "endio-meta-write", flags,
2357 fs_info->endio_raid56_workers =
2358 btrfs_alloc_workqueue(fs_info, "endio-raid56", flags,
2360 fs_info->endio_repair_workers =
2361 btrfs_alloc_workqueue(fs_info, "endio-repair", flags, 1, 0);
2362 fs_info->rmw_workers =
2363 btrfs_alloc_workqueue(fs_info, "rmw", flags, max_active, 2);
2364 fs_info->endio_write_workers =
2365 btrfs_alloc_workqueue(fs_info, "endio-write", flags,
2367 fs_info->endio_freespace_worker =
2368 btrfs_alloc_workqueue(fs_info, "freespace-write", flags,
2370 fs_info->delayed_workers =
2371 btrfs_alloc_workqueue(fs_info, "delayed-meta", flags,
2373 fs_info->readahead_workers =
2374 btrfs_alloc_workqueue(fs_info, "readahead", flags,
2376 fs_info->qgroup_rescan_workers =
2377 btrfs_alloc_workqueue(fs_info, "qgroup-rescan", flags, 1, 0);
2378 fs_info->extent_workers =
2379 btrfs_alloc_workqueue(fs_info, "extent-refs", flags,
2380 min_t(u64, fs_devices->num_devices,
2383 if (!(fs_info->workers && fs_info->delalloc_workers &&
2384 fs_info->submit_workers && fs_info->flush_workers &&
2385 fs_info->endio_workers && fs_info->endio_meta_workers &&
2386 fs_info->endio_meta_write_workers &&
2387 fs_info->endio_repair_workers &&
2388 fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2389 fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2390 fs_info->caching_workers && fs_info->readahead_workers &&
2391 fs_info->fixup_workers && fs_info->delayed_workers &&
2392 fs_info->extent_workers &&
2393 fs_info->qgroup_rescan_workers)) {
2400 static int btrfs_replay_log(struct btrfs_fs_info *fs_info,
2401 struct btrfs_fs_devices *fs_devices)
2404 struct btrfs_root *tree_root = fs_info->tree_root;
2405 struct btrfs_root *log_tree_root;
2406 struct btrfs_super_block *disk_super = fs_info->super_copy;
2407 u64 bytenr = btrfs_super_log_root(disk_super);
2409 if (fs_devices->rw_devices == 0) {
2410 btrfs_warn(fs_info, "log replay required on RO media");
2414 log_tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2418 __setup_root(tree_root->nodesize, tree_root->sectorsize,
2419 tree_root->stripesize, log_tree_root, fs_info,
2420 BTRFS_TREE_LOG_OBJECTID);
2422 log_tree_root->node = read_tree_block(tree_root, bytenr,
2423 fs_info->generation + 1);
2424 if (IS_ERR(log_tree_root->node)) {
2425 btrfs_warn(fs_info, "failed to read log tree");
2426 ret = PTR_ERR(log_tree_root->node);
2427 kfree(log_tree_root);
2429 } else if (!extent_buffer_uptodate(log_tree_root->node)) {
2430 btrfs_err(fs_info, "failed to read log tree");
2431 free_extent_buffer(log_tree_root->node);
2432 kfree(log_tree_root);
2435 /* returns with log_tree_root freed on success */
2436 ret = btrfs_recover_log_trees(log_tree_root);
2438 btrfs_handle_fs_error(tree_root->fs_info, ret,
2439 "Failed to recover log tree");
2440 free_extent_buffer(log_tree_root->node);
2441 kfree(log_tree_root);
2445 if (fs_info->sb->s_flags & MS_RDONLY) {
2446 ret = btrfs_commit_super(tree_root);
2454 static int btrfs_read_roots(struct btrfs_fs_info *fs_info,
2455 struct btrfs_root *tree_root)
2457 struct btrfs_root *root;
2458 struct btrfs_key location;
2461 location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2462 location.type = BTRFS_ROOT_ITEM_KEY;
2463 location.offset = 0;
2465 root = btrfs_read_tree_root(tree_root, &location);
2467 return PTR_ERR(root);
2468 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2469 fs_info->extent_root = root;
2471 location.objectid = BTRFS_DEV_TREE_OBJECTID;
2472 root = btrfs_read_tree_root(tree_root, &location);
2474 return PTR_ERR(root);
2475 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2476 fs_info->dev_root = root;
2477 btrfs_init_devices_late(fs_info);
2479 location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2480 root = btrfs_read_tree_root(tree_root, &location);
2482 return PTR_ERR(root);
2483 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2484 fs_info->csum_root = root;
2486 location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2487 root = btrfs_read_tree_root(tree_root, &location);
2488 if (!IS_ERR(root)) {
2489 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2490 fs_info->quota_enabled = 1;
2491 fs_info->pending_quota_state = 1;
2492 fs_info->quota_root = root;
2495 location.objectid = BTRFS_UUID_TREE_OBJECTID;
2496 root = btrfs_read_tree_root(tree_root, &location);
2498 ret = PTR_ERR(root);
2502 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2503 fs_info->uuid_root = root;
2506 if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
2507 location.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID;
2508 root = btrfs_read_tree_root(tree_root, &location);
2510 return PTR_ERR(root);
2511 set_bit(BTRFS_ROOT_TRACK_DIRTY, &root->state);
2512 fs_info->free_space_root = root;
2518 int open_ctree(struct super_block *sb,
2519 struct btrfs_fs_devices *fs_devices,
2527 struct btrfs_key location;
2528 struct buffer_head *bh;
2529 struct btrfs_super_block *disk_super;
2530 struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2531 struct btrfs_root *tree_root;
2532 struct btrfs_root *chunk_root;
2535 int num_backups_tried = 0;
2536 int backup_index = 0;
2539 tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2540 chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info, GFP_KERNEL);
2541 if (!tree_root || !chunk_root) {
2546 ret = init_srcu_struct(&fs_info->subvol_srcu);
2552 ret = setup_bdi(fs_info, &fs_info->bdi);
2558 ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0, GFP_KERNEL);
2563 fs_info->dirty_metadata_batch = PAGE_SIZE *
2564 (1 + ilog2(nr_cpu_ids));
2566 ret = percpu_counter_init(&fs_info->delalloc_bytes, 0, GFP_KERNEL);
2569 goto fail_dirty_metadata_bytes;
2572 ret = percpu_counter_init(&fs_info->bio_counter, 0, GFP_KERNEL);
2575 goto fail_delalloc_bytes;
2578 fs_info->btree_inode = new_inode(sb);
2579 if (!fs_info->btree_inode) {
2581 goto fail_bio_counter;
2584 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2586 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2587 INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2588 INIT_LIST_HEAD(&fs_info->trans_list);
2589 INIT_LIST_HEAD(&fs_info->dead_roots);
2590 INIT_LIST_HEAD(&fs_info->delayed_iputs);
2591 INIT_LIST_HEAD(&fs_info->delalloc_roots);
2592 INIT_LIST_HEAD(&fs_info->caching_block_groups);
2593 spin_lock_init(&fs_info->delalloc_root_lock);
2594 spin_lock_init(&fs_info->trans_lock);
2595 spin_lock_init(&fs_info->fs_roots_radix_lock);
2596 spin_lock_init(&fs_info->delayed_iput_lock);
2597 spin_lock_init(&fs_info->defrag_inodes_lock);
2598 spin_lock_init(&fs_info->free_chunk_lock);
2599 spin_lock_init(&fs_info->tree_mod_seq_lock);
2600 spin_lock_init(&fs_info->super_lock);
2601 spin_lock_init(&fs_info->qgroup_op_lock);
2602 spin_lock_init(&fs_info->buffer_lock);
2603 spin_lock_init(&fs_info->unused_bgs_lock);
2604 rwlock_init(&fs_info->tree_mod_log_lock);
2605 mutex_init(&fs_info->unused_bg_unpin_mutex);
2606 mutex_init(&fs_info->delete_unused_bgs_mutex);
2607 mutex_init(&fs_info->reloc_mutex);
2608 mutex_init(&fs_info->delalloc_root_mutex);
2609 mutex_init(&fs_info->cleaner_delayed_iput_mutex);
2610 seqlock_init(&fs_info->profiles_lock);
2612 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2613 INIT_LIST_HEAD(&fs_info->space_info);
2614 INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2615 INIT_LIST_HEAD(&fs_info->unused_bgs);
2616 btrfs_mapping_init(&fs_info->mapping_tree);
2617 btrfs_init_block_rsv(&fs_info->global_block_rsv,
2618 BTRFS_BLOCK_RSV_GLOBAL);
2619 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2620 BTRFS_BLOCK_RSV_DELALLOC);
2621 btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2622 btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2623 btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2624 btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2625 BTRFS_BLOCK_RSV_DELOPS);
2626 atomic_set(&fs_info->nr_async_submits, 0);
2627 atomic_set(&fs_info->async_delalloc_pages, 0);
2628 atomic_set(&fs_info->async_submit_draining, 0);
2629 atomic_set(&fs_info->nr_async_bios, 0);
2630 atomic_set(&fs_info->defrag_running, 0);
2631 atomic_set(&fs_info->qgroup_op_seq, 0);
2632 atomic_set(&fs_info->reada_works_cnt, 0);
2633 atomic64_set(&fs_info->tree_mod_seq, 0);
2634 fs_info->fs_frozen = 0;
2636 fs_info->max_inline = BTRFS_DEFAULT_MAX_INLINE;
2637 fs_info->metadata_ratio = 0;
2638 fs_info->defrag_inodes = RB_ROOT;
2639 fs_info->free_chunk_space = 0;
2640 fs_info->tree_mod_log = RB_ROOT;
2641 fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2642 fs_info->avg_delayed_ref_runtime = NSEC_PER_SEC >> 6; /* div by 64 */
2643 /* readahead state */
2644 INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_DIRECT_RECLAIM);
2645 spin_lock_init(&fs_info->reada_lock);
2647 fs_info->thread_pool_size = min_t(unsigned long,
2648 num_online_cpus() + 2, 8);
2650 INIT_LIST_HEAD(&fs_info->ordered_roots);
2651 spin_lock_init(&fs_info->ordered_root_lock);
2652 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2654 if (!fs_info->delayed_root) {
2658 btrfs_init_delayed_root(fs_info->delayed_root);
2660 btrfs_init_scrub(fs_info);
2661 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2662 fs_info->check_integrity_print_mask = 0;
2664 btrfs_init_balance(fs_info);
2665 btrfs_init_async_reclaim_work(&fs_info->async_reclaim_work);
2667 sb->s_blocksize = 4096;
2668 sb->s_blocksize_bits = blksize_bits(4096);
2669 sb->s_bdi = &fs_info->bdi;
2671 btrfs_init_btree_inode(fs_info, tree_root);
2673 spin_lock_init(&fs_info->block_group_cache_lock);
2674 fs_info->block_group_cache_tree = RB_ROOT;
2675 fs_info->first_logical_byte = (u64)-1;
2677 extent_io_tree_init(&fs_info->freed_extents[0],
2678 fs_info->btree_inode->i_mapping);
2679 extent_io_tree_init(&fs_info->freed_extents[1],
2680 fs_info->btree_inode->i_mapping);
2681 fs_info->pinned_extents = &fs_info->freed_extents[0];
2682 fs_info->do_barriers = 1;
2685 mutex_init(&fs_info->ordered_operations_mutex);
2686 mutex_init(&fs_info->tree_log_mutex);
2687 mutex_init(&fs_info->chunk_mutex);
2688 mutex_init(&fs_info->transaction_kthread_mutex);
2689 mutex_init(&fs_info->cleaner_mutex);
2690 mutex_init(&fs_info->volume_mutex);
2691 mutex_init(&fs_info->ro_block_group_mutex);
2692 init_rwsem(&fs_info->commit_root_sem);
2693 init_rwsem(&fs_info->cleanup_work_sem);
2694 init_rwsem(&fs_info->subvol_sem);
2695 sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2697 btrfs_init_dev_replace_locks(fs_info);
2698 btrfs_init_qgroup(fs_info);
2700 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2701 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2703 init_waitqueue_head(&fs_info->transaction_throttle);
2704 init_waitqueue_head(&fs_info->transaction_wait);
2705 init_waitqueue_head(&fs_info->transaction_blocked_wait);
2706 init_waitqueue_head(&fs_info->async_submit_wait);
2708 INIT_LIST_HEAD(&fs_info->pinned_chunks);
2710 ret = btrfs_alloc_stripe_hash_table(fs_info);
2716 __setup_root(4096, 4096, 4096, tree_root,
2717 fs_info, BTRFS_ROOT_TREE_OBJECTID);
2719 invalidate_bdev(fs_devices->latest_bdev);
2722 * Read super block and check the signature bytes only
2724 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2731 * We want to check superblock checksum, the type is stored inside.
2732 * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2734 if (btrfs_check_super_csum(bh->b_data)) {
2735 btrfs_err(fs_info, "superblock checksum mismatch");
2742 * super_copy is zeroed at allocation time and we never touch the
2743 * following bytes up to INFO_SIZE, the checksum is calculated from
2744 * the whole block of INFO_SIZE
2746 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2747 memcpy(fs_info->super_for_commit, fs_info->super_copy,
2748 sizeof(*fs_info->super_for_commit));
2751 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2753 ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2755 btrfs_err(fs_info, "superblock contains fatal errors");
2760 disk_super = fs_info->super_copy;
2761 if (!btrfs_super_root(disk_super))
2764 /* check FS state, whether FS is broken. */
2765 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2766 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2769 * run through our array of backup supers and setup
2770 * our ring pointer to the oldest one
2772 generation = btrfs_super_generation(disk_super);
2773 find_oldest_super_backup(fs_info, generation);
2776 * In the long term, we'll store the compression type in the super
2777 * block, and it'll be used for per file compression control.
2779 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2781 ret = btrfs_parse_options(tree_root, options, sb->s_flags);
2787 features = btrfs_super_incompat_flags(disk_super) &
2788 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2791 "cannot mount because of unsupported optional features (%llx)",
2797 features = btrfs_super_incompat_flags(disk_super);
2798 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2799 if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2800 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2802 if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2803 btrfs_info(fs_info, "has skinny extents");
2806 * flag our filesystem as having big metadata blocks if
2807 * they are bigger than the page size
2809 if (btrfs_super_nodesize(disk_super) > PAGE_SIZE) {
2810 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2812 "flagging fs with big metadata feature");
2813 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2816 nodesize = btrfs_super_nodesize(disk_super);
2817 sectorsize = btrfs_super_sectorsize(disk_super);
2818 stripesize = sectorsize;
2819 fs_info->dirty_metadata_batch = nodesize * (1 + ilog2(nr_cpu_ids));
2820 fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2823 * mixed block groups end up with duplicate but slightly offset
2824 * extent buffers for the same range. It leads to corruptions
2826 if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2827 (sectorsize != nodesize)) {
2829 "unequal nodesize/sectorsize (%u != %u) are not allowed for mixed block groups",
2830 nodesize, sectorsize);
2835 * Needn't use the lock because there is no other task which will
2838 btrfs_set_super_incompat_flags(disk_super, features);
2840 features = btrfs_super_compat_ro_flags(disk_super) &
2841 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2842 if (!(sb->s_flags & MS_RDONLY) && features) {
2844 "cannot mount read-write because of unsupported optional features (%llx)",
2850 max_active = fs_info->thread_pool_size;
2852 ret = btrfs_init_workqueues(fs_info, fs_devices);
2855 goto fail_sb_buffer;
2858 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2859 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2862 tree_root->nodesize = nodesize;
2863 tree_root->sectorsize = sectorsize;
2864 tree_root->stripesize = stripesize;
2866 sb->s_blocksize = sectorsize;
2867 sb->s_blocksize_bits = blksize_bits(sectorsize);
2869 mutex_lock(&fs_info->chunk_mutex);
2870 ret = btrfs_read_sys_array(tree_root);
2871 mutex_unlock(&fs_info->chunk_mutex);
2873 btrfs_err(fs_info, "failed to read the system array: %d", ret);
2874 goto fail_sb_buffer;
2877 generation = btrfs_super_chunk_root_generation(disk_super);
2879 __setup_root(nodesize, sectorsize, stripesize, chunk_root,
2880 fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2882 chunk_root->node = read_tree_block(chunk_root,
2883 btrfs_super_chunk_root(disk_super),
2885 if (IS_ERR(chunk_root->node) ||
2886 !extent_buffer_uptodate(chunk_root->node)) {
2887 btrfs_err(fs_info, "failed to read chunk root");
2888 if (!IS_ERR(chunk_root->node))
2889 free_extent_buffer(chunk_root->node);
2890 chunk_root->node = NULL;
2891 goto fail_tree_roots;
2893 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2894 chunk_root->commit_root = btrfs_root_node(chunk_root);
2896 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2897 btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2899 ret = btrfs_read_chunk_tree(chunk_root);
2901 btrfs_err(fs_info, "failed to read chunk tree: %d", ret);
2902 goto fail_tree_roots;
2906 * keep the device that is marked to be the target device for the
2907 * dev_replace procedure
2909 btrfs_close_extra_devices(fs_devices, 0);
2911 if (!fs_devices->latest_bdev) {
2912 btrfs_err(fs_info, "failed to read devices");
2913 goto fail_tree_roots;
2917 generation = btrfs_super_generation(disk_super);
2919 tree_root->node = read_tree_block(tree_root,
2920 btrfs_super_root(disk_super),
2922 if (IS_ERR(tree_root->node) ||
2923 !extent_buffer_uptodate(tree_root->node)) {
2924 btrfs_warn(fs_info, "failed to read tree root");
2925 if (!IS_ERR(tree_root->node))
2926 free_extent_buffer(tree_root->node);
2927 tree_root->node = NULL;
2928 goto recovery_tree_root;
2931 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2932 tree_root->commit_root = btrfs_root_node(tree_root);
2933 btrfs_set_root_refs(&tree_root->root_item, 1);
2935 mutex_lock(&tree_root->objectid_mutex);
2936 ret = btrfs_find_highest_objectid(tree_root,
2937 &tree_root->highest_objectid);
2939 mutex_unlock(&tree_root->objectid_mutex);
2940 goto recovery_tree_root;
2943 ASSERT(tree_root->highest_objectid <= BTRFS_LAST_FREE_OBJECTID);
2945 mutex_unlock(&tree_root->objectid_mutex);
2947 ret = btrfs_read_roots(fs_info, tree_root);
2949 goto recovery_tree_root;
2951 fs_info->generation = generation;
2952 fs_info->last_trans_committed = generation;
2954 ret = btrfs_recover_balance(fs_info);
2956 btrfs_err(fs_info, "failed to recover balance: %d", ret);
2957 goto fail_block_groups;
2960 ret = btrfs_init_dev_stats(fs_info);
2962 btrfs_err(fs_info, "failed to init dev_stats: %d", ret);
2963 goto fail_block_groups;
2966 ret = btrfs_init_dev_replace(fs_info);
2968 btrfs_err(fs_info, "failed to init dev_replace: %d", ret);
2969 goto fail_block_groups;
2972 btrfs_close_extra_devices(fs_devices, 1);
2974 ret = btrfs_sysfs_add_fsid(fs_devices, NULL);
2976 btrfs_err(fs_info, "failed to init sysfs fsid interface: %d",
2978 goto fail_block_groups;
2981 ret = btrfs_sysfs_add_device(fs_devices);
2983 btrfs_err(fs_info, "failed to init sysfs device interface: %d",
2985 goto fail_fsdev_sysfs;
2988 ret = btrfs_sysfs_add_mounted(fs_info);
2990 btrfs_err(fs_info, "failed to init sysfs interface: %d", ret);
2991 goto fail_fsdev_sysfs;
2994 ret = btrfs_init_space_info(fs_info);
2996 btrfs_err(fs_info, "failed to initialize space info: %d", ret);
3000 ret = btrfs_read_block_groups(fs_info->extent_root);
3002 btrfs_err(fs_info, "failed to read block groups: %d", ret);
3005 fs_info->num_tolerated_disk_barrier_failures =
3006 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3007 if (fs_info->fs_devices->missing_devices >
3008 fs_info->num_tolerated_disk_barrier_failures &&
3009 !(sb->s_flags & MS_RDONLY)) {
3011 "missing devices (%llu) exceeds the limit (%d), writeable mount is not allowed",
3012 fs_info->fs_devices->missing_devices,
3013 fs_info->num_tolerated_disk_barrier_failures);
3017 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
3019 if (IS_ERR(fs_info->cleaner_kthread))
3022 fs_info->transaction_kthread = kthread_run(transaction_kthread,
3024 "btrfs-transaction");
3025 if (IS_ERR(fs_info->transaction_kthread))
3028 if (!btrfs_test_opt(tree_root->fs_info, SSD) &&
3029 !btrfs_test_opt(tree_root->fs_info, NOSSD) &&
3030 !fs_info->fs_devices->rotating) {
3031 btrfs_info(fs_info, "detected SSD devices, enabling SSD mode");
3032 btrfs_set_opt(fs_info->mount_opt, SSD);
3036 * Mount does not set all options immediately, we can do it now and do
3037 * not have to wait for transaction commit
3039 btrfs_apply_pending_changes(fs_info);
3041 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3042 if (btrfs_test_opt(tree_root->fs_info, CHECK_INTEGRITY)) {
3043 ret = btrfsic_mount(tree_root, fs_devices,
3044 btrfs_test_opt(tree_root->fs_info,
3045 CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
3047 fs_info->check_integrity_print_mask);
3050 "failed to initialize integrity check module: %d",
3054 ret = btrfs_read_qgroup_config(fs_info);
3056 goto fail_trans_kthread;
3058 /* do not make disk changes in broken FS or nologreplay is given */
3059 if (btrfs_super_log_root(disk_super) != 0 &&
3060 !btrfs_test_opt(tree_root->fs_info, NOLOGREPLAY)) {
3061 ret = btrfs_replay_log(fs_info, fs_devices);
3068 ret = btrfs_find_orphan_roots(tree_root);
3072 if (!(sb->s_flags & MS_RDONLY)) {
3073 ret = btrfs_cleanup_fs_roots(fs_info);
3077 mutex_lock(&fs_info->cleaner_mutex);
3078 ret = btrfs_recover_relocation(tree_root);
3079 mutex_unlock(&fs_info->cleaner_mutex);
3081 btrfs_warn(fs_info, "failed to recover relocation: %d",
3088 location.objectid = BTRFS_FS_TREE_OBJECTID;
3089 location.type = BTRFS_ROOT_ITEM_KEY;
3090 location.offset = 0;
3092 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
3093 if (IS_ERR(fs_info->fs_root)) {
3094 err = PTR_ERR(fs_info->fs_root);
3098 if (sb->s_flags & MS_RDONLY)
3101 if (btrfs_test_opt(tree_root->fs_info, FREE_SPACE_TREE) &&
3102 !btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3103 btrfs_info(fs_info, "creating free space tree");
3104 ret = btrfs_create_free_space_tree(fs_info);
3107 "failed to create free space tree: %d", ret);
3108 close_ctree(tree_root);
3113 down_read(&fs_info->cleanup_work_sem);
3114 if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
3115 (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
3116 up_read(&fs_info->cleanup_work_sem);
3117 close_ctree(tree_root);
3120 up_read(&fs_info->cleanup_work_sem);
3122 ret = btrfs_resume_balance_async(fs_info);
3124 btrfs_warn(fs_info, "failed to resume balance: %d", ret);
3125 close_ctree(tree_root);
3129 ret = btrfs_resume_dev_replace_async(fs_info);
3131 btrfs_warn(fs_info, "failed to resume device replace: %d", ret);
3132 close_ctree(tree_root);
3136 btrfs_qgroup_rescan_resume(fs_info);
3138 if (btrfs_test_opt(tree_root->fs_info, CLEAR_CACHE) &&
3139 btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) {
3140 btrfs_info(fs_info, "clearing free space tree");
3141 ret = btrfs_clear_free_space_tree(fs_info);
3144 "failed to clear free space tree: %d", ret);
3145 close_ctree(tree_root);
3150 if (!fs_info->uuid_root) {
3151 btrfs_info(fs_info, "creating UUID tree");
3152 ret = btrfs_create_uuid_tree(fs_info);
3155 "failed to create the UUID tree: %d", ret);
3156 close_ctree(tree_root);
3159 } else if (btrfs_test_opt(tree_root->fs_info, RESCAN_UUID_TREE) ||
3160 fs_info->generation !=
3161 btrfs_super_uuid_tree_generation(disk_super)) {
3162 btrfs_info(fs_info, "checking UUID tree");
3163 ret = btrfs_check_uuid_tree(fs_info);
3166 "failed to check the UUID tree: %d", ret);
3167 close_ctree(tree_root);
3171 fs_info->update_uuid_tree_gen = 1;
3177 * backuproot only affect mount behavior, and if open_ctree succeeded,
3178 * no need to keep the flag
3180 btrfs_clear_opt(fs_info->mount_opt, USEBACKUPROOT);
3185 btrfs_free_qgroup_config(fs_info);
3187 kthread_stop(fs_info->transaction_kthread);
3188 btrfs_cleanup_transaction(fs_info->tree_root);
3189 btrfs_free_fs_roots(fs_info);
3191 kthread_stop(fs_info->cleaner_kthread);
3194 * make sure we're done with the btree inode before we stop our
3197 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
3200 btrfs_sysfs_remove_mounted(fs_info);
3203 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3206 btrfs_put_block_group_cache(fs_info);
3207 btrfs_free_block_groups(fs_info);
3210 free_root_pointers(fs_info, 1);
3211 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3214 btrfs_stop_all_workers(fs_info);
3217 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3219 iput(fs_info->btree_inode);
3221 percpu_counter_destroy(&fs_info->bio_counter);
3222 fail_delalloc_bytes:
3223 percpu_counter_destroy(&fs_info->delalloc_bytes);
3224 fail_dirty_metadata_bytes:
3225 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3227 bdi_destroy(&fs_info->bdi);
3229 cleanup_srcu_struct(&fs_info->subvol_srcu);
3231 btrfs_free_stripe_hash_table(fs_info);
3232 btrfs_close_devices(fs_info->fs_devices);
3236 if (!btrfs_test_opt(tree_root->fs_info, USEBACKUPROOT))
3237 goto fail_tree_roots;
3239 free_root_pointers(fs_info, 0);
3241 /* don't use the log in recovery mode, it won't be valid */
3242 btrfs_set_super_log_root(disk_super, 0);
3244 /* we can't trust the free space cache either */
3245 btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3247 ret = next_root_backup(fs_info, fs_info->super_copy,
3248 &num_backups_tried, &backup_index);
3250 goto fail_block_groups;
3251 goto retry_root_backup;
3254 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3257 set_buffer_uptodate(bh);
3259 struct btrfs_device *device = (struct btrfs_device *)
3262 btrfs_warn_rl_in_rcu(device->dev_root->fs_info,
3263 "lost page write due to IO error on %s",
3264 rcu_str_deref(device->name));
3265 /* note, we don't set_buffer_write_io_error because we have
3266 * our own ways of dealing with the IO errors
3268 clear_buffer_uptodate(bh);
3269 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3275 int btrfs_read_dev_one_super(struct block_device *bdev, int copy_num,
3276 struct buffer_head **bh_ret)
3278 struct buffer_head *bh;
3279 struct btrfs_super_block *super;
3282 bytenr = btrfs_sb_offset(copy_num);
3283 if (bytenr + BTRFS_SUPER_INFO_SIZE >= i_size_read(bdev->bd_inode))
3286 bh = __bread(bdev, bytenr / 4096, BTRFS_SUPER_INFO_SIZE);
3288 * If we fail to read from the underlying devices, as of now
3289 * the best option we have is to mark it EIO.
3294 super = (struct btrfs_super_block *)bh->b_data;
3295 if (btrfs_super_bytenr(super) != bytenr ||
3296 btrfs_super_magic(super) != BTRFS_MAGIC) {
3306 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3308 struct buffer_head *bh;
3309 struct buffer_head *latest = NULL;
3310 struct btrfs_super_block *super;
3315 /* we would like to check all the supers, but that would make
3316 * a btrfs mount succeed after a mkfs from a different FS.
3317 * So, we need to add a special mount option to scan for
3318 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3320 for (i = 0; i < 1; i++) {
3321 ret = btrfs_read_dev_one_super(bdev, i, &bh);
3325 super = (struct btrfs_super_block *)bh->b_data;
3327 if (!latest || btrfs_super_generation(super) > transid) {
3330 transid = btrfs_super_generation(super);
3337 return ERR_PTR(ret);
3343 * this should be called twice, once with wait == 0 and
3344 * once with wait == 1. When wait == 0 is done, all the buffer heads
3345 * we write are pinned.
3347 * They are released when wait == 1 is done.
3348 * max_mirrors must be the same for both runs, and it indicates how
3349 * many supers on this one device should be written.
3351 * max_mirrors == 0 means to write them all.
3353 static int write_dev_supers(struct btrfs_device *device,
3354 struct btrfs_super_block *sb,
3355 int do_barriers, int wait, int max_mirrors)
3357 struct buffer_head *bh;
3364 if (max_mirrors == 0)
3365 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3367 for (i = 0; i < max_mirrors; i++) {
3368 bytenr = btrfs_sb_offset(i);
3369 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3370 device->commit_total_bytes)
3374 bh = __find_get_block(device->bdev, bytenr / 4096,
3375 BTRFS_SUPER_INFO_SIZE);
3381 if (!buffer_uptodate(bh))
3384 /* drop our reference */
3387 /* drop the reference from the wait == 0 run */
3391 btrfs_set_super_bytenr(sb, bytenr);
3394 crc = btrfs_csum_data((char *)sb +
3395 BTRFS_CSUM_SIZE, crc,
3396 BTRFS_SUPER_INFO_SIZE -
3398 btrfs_csum_final(crc, sb->csum);
3401 * one reference for us, and we leave it for the
3404 bh = __getblk(device->bdev, bytenr / 4096,
3405 BTRFS_SUPER_INFO_SIZE);
3407 btrfs_err(device->dev_root->fs_info,
3408 "couldn't get super buffer head for bytenr %llu",
3414 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3416 /* one reference for submit_bh */
3419 set_buffer_uptodate(bh);
3421 bh->b_end_io = btrfs_end_buffer_write_sync;
3422 bh->b_private = device;
3426 * we fua the first super. The others we allow
3430 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3432 ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3436 return errors < i ? 0 : -1;
3440 * endio for the write_dev_flush, this will wake anyone waiting
3441 * for the barrier when it is done
3443 static void btrfs_end_empty_barrier(struct bio *bio)
3445 if (bio->bi_private)
3446 complete(bio->bi_private);
3451 * trigger flushes for one the devices. If you pass wait == 0, the flushes are
3452 * sent down. With wait == 1, it waits for the previous flush.
3454 * any device where the flush fails with eopnotsupp are flagged as not-barrier
3457 static int write_dev_flush(struct btrfs_device *device, int wait)
3462 if (device->nobarriers)
3466 bio = device->flush_bio;
3470 wait_for_completion(&device->flush_wait);
3472 if (bio->bi_error) {
3473 ret = bio->bi_error;
3474 btrfs_dev_stat_inc_and_print(device,
3475 BTRFS_DEV_STAT_FLUSH_ERRS);
3478 /* drop the reference from the wait == 0 run */
3480 device->flush_bio = NULL;
3486 * one reference for us, and we leave it for the
3489 device->flush_bio = NULL;
3490 bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3494 bio->bi_end_io = btrfs_end_empty_barrier;
3495 bio->bi_bdev = device->bdev;
3496 init_completion(&device->flush_wait);
3497 bio->bi_private = &device->flush_wait;
3498 device->flush_bio = bio;
3501 btrfsic_submit_bio(WRITE_FLUSH, bio);
3507 * send an empty flush down to each device in parallel,
3508 * then wait for them
3510 static int barrier_all_devices(struct btrfs_fs_info *info)
3512 struct list_head *head;
3513 struct btrfs_device *dev;
3514 int errors_send = 0;
3515 int errors_wait = 0;
3518 /* send down all the barriers */
3519 head = &info->fs_devices->devices;
3520 list_for_each_entry_rcu(dev, head, dev_list) {
3527 if (!dev->in_fs_metadata || !dev->writeable)
3530 ret = write_dev_flush(dev, 0);
3535 /* wait for all the barriers */
3536 list_for_each_entry_rcu(dev, head, dev_list) {
3543 if (!dev->in_fs_metadata || !dev->writeable)
3546 ret = write_dev_flush(dev, 1);
3550 if (errors_send > info->num_tolerated_disk_barrier_failures ||
3551 errors_wait > info->num_tolerated_disk_barrier_failures)
3556 int btrfs_get_num_tolerated_disk_barrier_failures(u64 flags)
3559 int min_tolerated = INT_MAX;
3561 if ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 ||
3562 (flags & BTRFS_AVAIL_ALLOC_BIT_SINGLE))
3563 min_tolerated = min(min_tolerated,
3564 btrfs_raid_array[BTRFS_RAID_SINGLE].
3565 tolerated_failures);
3567 for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) {
3568 if (raid_type == BTRFS_RAID_SINGLE)
3570 if (!(flags & btrfs_raid_group[raid_type]))
3572 min_tolerated = min(min_tolerated,
3573 btrfs_raid_array[raid_type].
3574 tolerated_failures);
3577 if (min_tolerated == INT_MAX) {
3578 pr_warn("BTRFS: unknown raid flag: %llu\n", flags);
3582 return min_tolerated;
3585 int btrfs_calc_num_tolerated_disk_barrier_failures(
3586 struct btrfs_fs_info *fs_info)
3588 struct btrfs_ioctl_space_info space;
3589 struct btrfs_space_info *sinfo;
3590 u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3591 BTRFS_BLOCK_GROUP_SYSTEM,
3592 BTRFS_BLOCK_GROUP_METADATA,
3593 BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3596 int num_tolerated_disk_barrier_failures =
3597 (int)fs_info->fs_devices->num_devices;
3599 for (i = 0; i < ARRAY_SIZE(types); i++) {
3600 struct btrfs_space_info *tmp;
3604 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3605 if (tmp->flags == types[i]) {
3615 down_read(&sinfo->groups_sem);
3616 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3619 if (list_empty(&sinfo->block_groups[c]))
3622 btrfs_get_block_group_info(&sinfo->block_groups[c],
3624 if (space.total_bytes == 0 || space.used_bytes == 0)
3626 flags = space.flags;
3628 num_tolerated_disk_barrier_failures = min(
3629 num_tolerated_disk_barrier_failures,
3630 btrfs_get_num_tolerated_disk_barrier_failures(
3633 up_read(&sinfo->groups_sem);
3636 return num_tolerated_disk_barrier_failures;
3639 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3641 struct list_head *head;
3642 struct btrfs_device *dev;
3643 struct btrfs_super_block *sb;
3644 struct btrfs_dev_item *dev_item;
3648 int total_errors = 0;
3651 do_barriers = !btrfs_test_opt(root->fs_info, NOBARRIER);
3652 backup_super_roots(root->fs_info);
3654 sb = root->fs_info->super_for_commit;
3655 dev_item = &sb->dev_item;
3657 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3658 head = &root->fs_info->fs_devices->devices;
3659 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3662 ret = barrier_all_devices(root->fs_info);
3665 &root->fs_info->fs_devices->device_list_mutex);
3666 btrfs_handle_fs_error(root->fs_info, ret,
3667 "errors while submitting device barriers.");
3672 list_for_each_entry_rcu(dev, head, dev_list) {
3677 if (!dev->in_fs_metadata || !dev->writeable)
3680 btrfs_set_stack_device_generation(dev_item, 0);
3681 btrfs_set_stack_device_type(dev_item, dev->type);
3682 btrfs_set_stack_device_id(dev_item, dev->devid);
3683 btrfs_set_stack_device_total_bytes(dev_item,
3684 dev->commit_total_bytes);
3685 btrfs_set_stack_device_bytes_used(dev_item,
3686 dev->commit_bytes_used);
3687 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3688 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3689 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3690 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3691 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3693 flags = btrfs_super_flags(sb);
3694 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3696 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3700 if (total_errors > max_errors) {
3701 btrfs_err(root->fs_info, "%d errors while writing supers",
3703 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3705 /* FUA is masked off if unsupported and can't be the reason */
3706 btrfs_handle_fs_error(root->fs_info, -EIO,
3707 "%d errors while writing supers", total_errors);
3712 list_for_each_entry_rcu(dev, head, dev_list) {
3715 if (!dev->in_fs_metadata || !dev->writeable)
3718 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3722 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3723 if (total_errors > max_errors) {
3724 btrfs_handle_fs_error(root->fs_info, -EIO,
3725 "%d errors while writing supers", total_errors);
3731 int write_ctree_super(struct btrfs_trans_handle *trans,
3732 struct btrfs_root *root, int max_mirrors)
3734 return write_all_supers(root, max_mirrors);
3737 /* Drop a fs root from the radix tree and free it. */
3738 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3739 struct btrfs_root *root)
3741 spin_lock(&fs_info->fs_roots_radix_lock);
3742 radix_tree_delete(&fs_info->fs_roots_radix,
3743 (unsigned long)root->root_key.objectid);
3744 spin_unlock(&fs_info->fs_roots_radix_lock);
3746 if (btrfs_root_refs(&root->root_item) == 0)
3747 synchronize_srcu(&fs_info->subvol_srcu);
3749 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3750 btrfs_free_log(NULL, root);
3751 if (root->reloc_root) {
3752 free_extent_buffer(root->reloc_root->node);
3753 free_extent_buffer(root->reloc_root->commit_root);
3754 btrfs_put_fs_root(root->reloc_root);
3755 root->reloc_root = NULL;
3759 if (root->free_ino_pinned)
3760 __btrfs_remove_free_space_cache(root->free_ino_pinned);
3761 if (root->free_ino_ctl)
3762 __btrfs_remove_free_space_cache(root->free_ino_ctl);
3766 static void free_fs_root(struct btrfs_root *root)
3768 iput(root->ino_cache_inode);
3769 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3770 btrfs_free_block_rsv(root, root->orphan_block_rsv);
3771 root->orphan_block_rsv = NULL;
3773 free_anon_bdev(root->anon_dev);
3774 if (root->subv_writers)
3775 btrfs_free_subvolume_writers(root->subv_writers);
3776 free_extent_buffer(root->node);
3777 free_extent_buffer(root->commit_root);
3778 kfree(root->free_ino_ctl);
3779 kfree(root->free_ino_pinned);
3781 btrfs_put_fs_root(root);
3784 void btrfs_free_fs_root(struct btrfs_root *root)
3789 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3791 u64 root_objectid = 0;
3792 struct btrfs_root *gang[8];
3795 unsigned int ret = 0;
3799 index = srcu_read_lock(&fs_info->subvol_srcu);
3800 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3801 (void **)gang, root_objectid,
3804 srcu_read_unlock(&fs_info->subvol_srcu, index);
3807 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3809 for (i = 0; i < ret; i++) {
3810 /* Avoid to grab roots in dead_roots */
3811 if (btrfs_root_refs(&gang[i]->root_item) == 0) {
3815 /* grab all the search result for later use */
3816 gang[i] = btrfs_grab_fs_root(gang[i]);
3818 srcu_read_unlock(&fs_info->subvol_srcu, index);
3820 for (i = 0; i < ret; i++) {
3823 root_objectid = gang[i]->root_key.objectid;
3824 err = btrfs_orphan_cleanup(gang[i]);
3827 btrfs_put_fs_root(gang[i]);
3832 /* release the uncleaned roots due to error */
3833 for (; i < ret; i++) {
3835 btrfs_put_fs_root(gang[i]);
3840 int btrfs_commit_super(struct btrfs_root *root)
3842 struct btrfs_trans_handle *trans;
3844 mutex_lock(&root->fs_info->cleaner_mutex);
3845 btrfs_run_delayed_iputs(root);
3846 mutex_unlock(&root->fs_info->cleaner_mutex);
3847 wake_up_process(root->fs_info->cleaner_kthread);
3849 /* wait until ongoing cleanup work done */
3850 down_write(&root->fs_info->cleanup_work_sem);
3851 up_write(&root->fs_info->cleanup_work_sem);
3853 trans = btrfs_join_transaction(root);
3855 return PTR_ERR(trans);
3856 return btrfs_commit_transaction(trans, root);
3859 void close_ctree(struct btrfs_root *root)
3861 struct btrfs_fs_info *fs_info = root->fs_info;
3864 fs_info->closing = 1;
3867 /* wait for the qgroup rescan worker to stop */
3868 btrfs_qgroup_wait_for_completion(fs_info, false);
3870 /* wait for the uuid_scan task to finish */
3871 down(&fs_info->uuid_tree_rescan_sem);
3872 /* avoid complains from lockdep et al., set sem back to initial state */
3873 up(&fs_info->uuid_tree_rescan_sem);
3875 /* pause restriper - we want to resume on mount */
3876 btrfs_pause_balance(fs_info);
3878 btrfs_dev_replace_suspend_for_unmount(fs_info);
3880 btrfs_scrub_cancel(fs_info);
3882 /* wait for any defraggers to finish */
3883 wait_event(fs_info->transaction_wait,
3884 (atomic_read(&fs_info->defrag_running) == 0));
3886 /* clear out the rbtree of defraggable inodes */
3887 btrfs_cleanup_defrag_inodes(fs_info);
3889 cancel_work_sync(&fs_info->async_reclaim_work);
3891 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3893 * If the cleaner thread is stopped and there are
3894 * block groups queued for removal, the deletion will be
3895 * skipped when we quit the cleaner thread.
3897 btrfs_delete_unused_bgs(root->fs_info);
3899 ret = btrfs_commit_super(root);
3901 btrfs_err(fs_info, "commit super ret %d", ret);
3904 if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3905 btrfs_error_commit_super(root);
3907 kthread_stop(fs_info->transaction_kthread);
3908 kthread_stop(fs_info->cleaner_kthread);
3910 fs_info->closing = 2;
3913 btrfs_free_qgroup_config(fs_info);
3915 if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3916 btrfs_info(fs_info, "at unmount delalloc count %lld",
3917 percpu_counter_sum(&fs_info->delalloc_bytes));
3920 btrfs_sysfs_remove_mounted(fs_info);
3921 btrfs_sysfs_remove_fsid(fs_info->fs_devices);
3923 btrfs_free_fs_roots(fs_info);
3925 btrfs_put_block_group_cache(fs_info);
3927 btrfs_free_block_groups(fs_info);
3930 * we must make sure there is not any read request to
3931 * submit after we stopping all workers.
3933 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
3934 btrfs_stop_all_workers(fs_info);
3937 free_root_pointers(fs_info, 1);
3939 iput(fs_info->btree_inode);
3941 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3942 if (btrfs_test_opt(root->fs_info, CHECK_INTEGRITY))
3943 btrfsic_unmount(root, fs_info->fs_devices);
3946 btrfs_close_devices(fs_info->fs_devices);
3947 btrfs_mapping_tree_free(&fs_info->mapping_tree);
3949 percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3950 percpu_counter_destroy(&fs_info->delalloc_bytes);
3951 percpu_counter_destroy(&fs_info->bio_counter);
3952 bdi_destroy(&fs_info->bdi);
3953 cleanup_srcu_struct(&fs_info->subvol_srcu);
3955 btrfs_free_stripe_hash_table(fs_info);
3957 __btrfs_free_block_rsv(root->orphan_block_rsv);
3958 root->orphan_block_rsv = NULL;
3961 while (!list_empty(&fs_info->pinned_chunks)) {
3962 struct extent_map *em;
3964 em = list_first_entry(&fs_info->pinned_chunks,
3965 struct extent_map, list);
3966 list_del_init(&em->list);
3967 free_extent_map(em);
3969 unlock_chunks(root);
3972 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3976 struct inode *btree_inode = buf->pages[0]->mapping->host;
3978 ret = extent_buffer_uptodate(buf);
3982 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3983 parent_transid, atomic);
3989 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3991 struct btrfs_root *root;
3992 u64 transid = btrfs_header_generation(buf);
3995 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3997 * This is a fast path so only do this check if we have sanity tests
3998 * enabled. Normal people shouldn't be marking dummy buffers as dirty
3999 * outside of the sanity tests.
4001 if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
4004 root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4005 btrfs_assert_tree_locked(buf);
4006 if (transid != root->fs_info->generation)
4007 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
4008 "found %llu running %llu\n",
4009 buf->start, transid, root->fs_info->generation);
4010 was_dirty = set_extent_buffer_dirty(buf);
4012 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
4014 root->fs_info->dirty_metadata_batch);
4015 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
4016 if (btrfs_header_level(buf) == 0 && check_leaf(root, buf)) {
4017 btrfs_print_leaf(root, buf);
4023 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
4027 * looks as though older kernels can get into trouble with
4028 * this code, they end up stuck in balance_dirty_pages forever
4032 if (current->flags & PF_MEMALLOC)
4036 btrfs_balance_delayed_items(root);
4038 ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
4039 BTRFS_DIRTY_METADATA_THRESH);
4041 balance_dirty_pages_ratelimited(
4042 root->fs_info->btree_inode->i_mapping);
4046 void btrfs_btree_balance_dirty(struct btrfs_root *root)
4048 __btrfs_btree_balance_dirty(root, 1);
4051 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
4053 __btrfs_btree_balance_dirty(root, 0);
4056 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
4058 struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
4059 return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
4062 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
4065 struct btrfs_super_block *sb = fs_info->super_copy;
4066 u64 nodesize = btrfs_super_nodesize(sb);
4067 u64 sectorsize = btrfs_super_sectorsize(sb);
4070 if (btrfs_super_magic(sb) != BTRFS_MAGIC) {
4071 printk(KERN_ERR "BTRFS: no valid FS found\n");
4074 if (btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP)
4075 printk(KERN_WARNING "BTRFS: unrecognized super flag: %llu\n",
4076 btrfs_super_flags(sb) & ~BTRFS_SUPER_FLAG_SUPP);
4077 if (btrfs_super_root_level(sb) >= BTRFS_MAX_LEVEL) {
4078 printk(KERN_ERR "BTRFS: tree_root level too big: %d >= %d\n",
4079 btrfs_super_root_level(sb), BTRFS_MAX_LEVEL);
4082 if (btrfs_super_chunk_root_level(sb) >= BTRFS_MAX_LEVEL) {
4083 printk(KERN_ERR "BTRFS: chunk_root level too big: %d >= %d\n",
4084 btrfs_super_chunk_root_level(sb), BTRFS_MAX_LEVEL);
4087 if (btrfs_super_log_root_level(sb) >= BTRFS_MAX_LEVEL) {
4088 printk(KERN_ERR "BTRFS: log_root level too big: %d >= %d\n",
4089 btrfs_super_log_root_level(sb), BTRFS_MAX_LEVEL);
4094 * Check sectorsize and nodesize first, other check will need it.
4095 * Check all possible sectorsize(4K, 8K, 16K, 32K, 64K) here.
4097 if (!is_power_of_2(sectorsize) || sectorsize < 4096 ||
4098 sectorsize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4099 printk(KERN_ERR "BTRFS: invalid sectorsize %llu\n", sectorsize);
4102 /* Only PAGE SIZE is supported yet */
4103 if (sectorsize != PAGE_SIZE) {
4104 printk(KERN_ERR "BTRFS: sectorsize %llu not supported yet, only support %lu\n",
4105 sectorsize, PAGE_SIZE);
4108 if (!is_power_of_2(nodesize) || nodesize < sectorsize ||
4109 nodesize > BTRFS_MAX_METADATA_BLOCKSIZE) {
4110 printk(KERN_ERR "BTRFS: invalid nodesize %llu\n", nodesize);
4113 if (nodesize != le32_to_cpu(sb->__unused_leafsize)) {
4114 printk(KERN_ERR "BTRFS: invalid leafsize %u, should be %llu\n",
4115 le32_to_cpu(sb->__unused_leafsize),
4120 /* Root alignment check */
4121 if (!IS_ALIGNED(btrfs_super_root(sb), sectorsize)) {
4122 printk(KERN_WARNING "BTRFS: tree_root block unaligned: %llu\n",
4123 btrfs_super_root(sb));
4126 if (!IS_ALIGNED(btrfs_super_chunk_root(sb), sectorsize)) {
4127 printk(KERN_WARNING "BTRFS: chunk_root block unaligned: %llu\n",
4128 btrfs_super_chunk_root(sb));
4131 if (!IS_ALIGNED(btrfs_super_log_root(sb), sectorsize)) {
4132 printk(KERN_WARNING "BTRFS: log_root block unaligned: %llu\n",
4133 btrfs_super_log_root(sb));
4137 if (memcmp(fs_info->fsid, sb->dev_item.fsid, BTRFS_UUID_SIZE) != 0) {
4138 printk(KERN_ERR "BTRFS: dev_item UUID does not match fsid: %pU != %pU\n",
4139 fs_info->fsid, sb->dev_item.fsid);
4144 * Hint to catch really bogus numbers, bitflips or so, more exact checks are
4147 if (btrfs_super_bytes_used(sb) < 6 * btrfs_super_nodesize(sb)) {
4148 btrfs_err(fs_info, "bytes_used is too small %llu",
4149 btrfs_super_bytes_used(sb));
4152 if (!is_power_of_2(btrfs_super_stripesize(sb))) {
4153 btrfs_err(fs_info, "invalid stripesize %u",
4154 btrfs_super_stripesize(sb));
4157 if (btrfs_super_num_devices(sb) > (1UL << 31))
4158 printk(KERN_WARNING "BTRFS: suspicious number of devices: %llu\n",
4159 btrfs_super_num_devices(sb));
4160 if (btrfs_super_num_devices(sb) == 0) {
4161 printk(KERN_ERR "BTRFS: number of devices is 0\n");
4165 if (btrfs_super_bytenr(sb) != BTRFS_SUPER_INFO_OFFSET) {
4166 printk(KERN_ERR "BTRFS: super offset mismatch %llu != %u\n",
4167 btrfs_super_bytenr(sb), BTRFS_SUPER_INFO_OFFSET);
4172 * Obvious sys_chunk_array corruptions, it must hold at least one key
4175 if (btrfs_super_sys_array_size(sb) > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) {
4176 printk(KERN_ERR "BTRFS: system chunk array too big %u > %u\n",
4177 btrfs_super_sys_array_size(sb),
4178 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE);
4181 if (btrfs_super_sys_array_size(sb) < sizeof(struct btrfs_disk_key)
4182 + sizeof(struct btrfs_chunk)) {
4183 printk(KERN_ERR "BTRFS: system chunk array too small %u < %zu\n",
4184 btrfs_super_sys_array_size(sb),
4185 sizeof(struct btrfs_disk_key)
4186 + sizeof(struct btrfs_chunk));
4191 * The generation is a global counter, we'll trust it more than the others
4192 * but it's still possible that it's the one that's wrong.
4194 if (btrfs_super_generation(sb) < btrfs_super_chunk_root_generation(sb))
4196 "BTRFS: suspicious: generation < chunk_root_generation: %llu < %llu\n",
4197 btrfs_super_generation(sb), btrfs_super_chunk_root_generation(sb));
4198 if (btrfs_super_generation(sb) < btrfs_super_cache_generation(sb)
4199 && btrfs_super_cache_generation(sb) != (u64)-1)
4201 "BTRFS: suspicious: generation < cache_generation: %llu < %llu\n",
4202 btrfs_super_generation(sb), btrfs_super_cache_generation(sb));
4207 static void btrfs_error_commit_super(struct btrfs_root *root)
4209 mutex_lock(&root->fs_info->cleaner_mutex);
4210 btrfs_run_delayed_iputs(root);
4211 mutex_unlock(&root->fs_info->cleaner_mutex);
4213 down_write(&root->fs_info->cleanup_work_sem);
4214 up_write(&root->fs_info->cleanup_work_sem);
4216 /* cleanup FS via transaction */
4217 btrfs_cleanup_transaction(root);
4220 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
4222 struct btrfs_ordered_extent *ordered;
4224 spin_lock(&root->ordered_extent_lock);
4226 * This will just short circuit the ordered completion stuff which will
4227 * make sure the ordered extent gets properly cleaned up.
4229 list_for_each_entry(ordered, &root->ordered_extents,
4231 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
4232 spin_unlock(&root->ordered_extent_lock);
4235 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
4237 struct btrfs_root *root;
4238 struct list_head splice;
4240 INIT_LIST_HEAD(&splice);
4242 spin_lock(&fs_info->ordered_root_lock);
4243 list_splice_init(&fs_info->ordered_roots, &splice);
4244 while (!list_empty(&splice)) {
4245 root = list_first_entry(&splice, struct btrfs_root,
4247 list_move_tail(&root->ordered_root,
4248 &fs_info->ordered_roots);
4250 spin_unlock(&fs_info->ordered_root_lock);
4251 btrfs_destroy_ordered_extents(root);
4254 spin_lock(&fs_info->ordered_root_lock);
4256 spin_unlock(&fs_info->ordered_root_lock);
4259 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
4260 struct btrfs_root *root)
4262 struct rb_node *node;
4263 struct btrfs_delayed_ref_root *delayed_refs;
4264 struct btrfs_delayed_ref_node *ref;
4267 delayed_refs = &trans->delayed_refs;
4269 spin_lock(&delayed_refs->lock);
4270 if (atomic_read(&delayed_refs->num_entries) == 0) {
4271 spin_unlock(&delayed_refs->lock);
4272 btrfs_info(root->fs_info, "delayed_refs has NO entry");
4276 while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
4277 struct btrfs_delayed_ref_head *head;
4278 struct btrfs_delayed_ref_node *tmp;
4279 bool pin_bytes = false;
4281 head = rb_entry(node, struct btrfs_delayed_ref_head,
4283 if (!mutex_trylock(&head->mutex)) {
4284 atomic_inc(&head->node.refs);
4285 spin_unlock(&delayed_refs->lock);
4287 mutex_lock(&head->mutex);
4288 mutex_unlock(&head->mutex);
4289 btrfs_put_delayed_ref(&head->node);
4290 spin_lock(&delayed_refs->lock);
4293 spin_lock(&head->lock);
4294 list_for_each_entry_safe_reverse(ref, tmp, &head->ref_list,
4297 list_del(&ref->list);
4298 atomic_dec(&delayed_refs->num_entries);
4299 btrfs_put_delayed_ref(ref);
4301 if (head->must_insert_reserved)
4303 btrfs_free_delayed_extent_op(head->extent_op);
4304 delayed_refs->num_heads--;
4305 if (head->processing == 0)
4306 delayed_refs->num_heads_ready--;
4307 atomic_dec(&delayed_refs->num_entries);
4308 head->node.in_tree = 0;
4309 rb_erase(&head->href_node, &delayed_refs->href_root);
4310 spin_unlock(&head->lock);
4311 spin_unlock(&delayed_refs->lock);
4312 mutex_unlock(&head->mutex);
4315 btrfs_pin_extent(root, head->node.bytenr,
4316 head->node.num_bytes, 1);
4317 btrfs_put_delayed_ref(&head->node);
4319 spin_lock(&delayed_refs->lock);
4322 spin_unlock(&delayed_refs->lock);
4327 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
4329 struct btrfs_inode *btrfs_inode;
4330 struct list_head splice;
4332 INIT_LIST_HEAD(&splice);
4334 spin_lock(&root->delalloc_lock);
4335 list_splice_init(&root->delalloc_inodes, &splice);
4337 while (!list_empty(&splice)) {
4338 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
4341 list_del_init(&btrfs_inode->delalloc_inodes);
4342 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
4343 &btrfs_inode->runtime_flags);
4344 spin_unlock(&root->delalloc_lock);
4346 btrfs_invalidate_inodes(btrfs_inode->root);
4348 spin_lock(&root->delalloc_lock);
4351 spin_unlock(&root->delalloc_lock);
4354 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
4356 struct btrfs_root *root;
4357 struct list_head splice;
4359 INIT_LIST_HEAD(&splice);
4361 spin_lock(&fs_info->delalloc_root_lock);
4362 list_splice_init(&fs_info->delalloc_roots, &splice);
4363 while (!list_empty(&splice)) {
4364 root = list_first_entry(&splice, struct btrfs_root,
4366 list_del_init(&root->delalloc_root);
4367 root = btrfs_grab_fs_root(root);
4369 spin_unlock(&fs_info->delalloc_root_lock);
4371 btrfs_destroy_delalloc_inodes(root);
4372 btrfs_put_fs_root(root);
4374 spin_lock(&fs_info->delalloc_root_lock);
4376 spin_unlock(&fs_info->delalloc_root_lock);
4379 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
4380 struct extent_io_tree *dirty_pages,
4384 struct extent_buffer *eb;
4389 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
4394 clear_extent_bits(dirty_pages, start, end, mark);
4395 while (start <= end) {
4396 eb = btrfs_find_tree_block(root->fs_info, start);
4397 start += root->nodesize;
4400 wait_on_extent_buffer_writeback(eb);
4402 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
4404 clear_extent_buffer_dirty(eb);
4405 free_extent_buffer_stale(eb);
4412 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
4413 struct extent_io_tree *pinned_extents)
4415 struct extent_io_tree *unpin;
4421 unpin = pinned_extents;
4424 ret = find_first_extent_bit(unpin, 0, &start, &end,
4425 EXTENT_DIRTY, NULL);
4429 clear_extent_dirty(unpin, start, end);
4430 btrfs_error_unpin_extent_range(root, start, end);
4435 if (unpin == &root->fs_info->freed_extents[0])
4436 unpin = &root->fs_info->freed_extents[1];
4438 unpin = &root->fs_info->freed_extents[0];
4446 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4447 struct btrfs_root *root)
4449 btrfs_destroy_delayed_refs(cur_trans, root);
4451 cur_trans->state = TRANS_STATE_COMMIT_START;
4452 wake_up(&root->fs_info->transaction_blocked_wait);
4454 cur_trans->state = TRANS_STATE_UNBLOCKED;
4455 wake_up(&root->fs_info->transaction_wait);
4457 btrfs_destroy_delayed_inodes(root);
4458 btrfs_assert_delayed_root_empty(root);
4460 btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4462 btrfs_destroy_pinned_extent(root,
4463 root->fs_info->pinned_extents);
4465 cur_trans->state =TRANS_STATE_COMPLETED;
4466 wake_up(&cur_trans->commit_wait);
4469 memset(cur_trans, 0, sizeof(*cur_trans));
4470 kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4474 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4476 struct btrfs_transaction *t;
4478 mutex_lock(&root->fs_info->transaction_kthread_mutex);
4480 spin_lock(&root->fs_info->trans_lock);
4481 while (!list_empty(&root->fs_info->trans_list)) {
4482 t = list_first_entry(&root->fs_info->trans_list,
4483 struct btrfs_transaction, list);
4484 if (t->state >= TRANS_STATE_COMMIT_START) {
4485 atomic_inc(&t->use_count);
4486 spin_unlock(&root->fs_info->trans_lock);
4487 btrfs_wait_for_commit(root, t->transid);
4488 btrfs_put_transaction(t);
4489 spin_lock(&root->fs_info->trans_lock);
4492 if (t == root->fs_info->running_transaction) {
4493 t->state = TRANS_STATE_COMMIT_DOING;
4494 spin_unlock(&root->fs_info->trans_lock);
4496 * We wait for 0 num_writers since we don't hold a trans
4497 * handle open currently for this transaction.
4499 wait_event(t->writer_wait,
4500 atomic_read(&t->num_writers) == 0);
4502 spin_unlock(&root->fs_info->trans_lock);
4504 btrfs_cleanup_one_transaction(t, root);
4506 spin_lock(&root->fs_info->trans_lock);
4507 if (t == root->fs_info->running_transaction)
4508 root->fs_info->running_transaction = NULL;
4509 list_del_init(&t->list);
4510 spin_unlock(&root->fs_info->trans_lock);
4512 btrfs_put_transaction(t);
4513 trace_btrfs_transaction_commit(root);
4514 spin_lock(&root->fs_info->trans_lock);
4516 spin_unlock(&root->fs_info->trans_lock);
4517 btrfs_destroy_all_ordered_extents(root->fs_info);
4518 btrfs_destroy_delayed_inodes(root);
4519 btrfs_assert_delayed_root_empty(root);
4520 btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4521 btrfs_destroy_all_delalloc_inodes(root->fs_info);
4522 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4527 static const struct extent_io_ops btree_extent_io_ops = {
4528 .readpage_end_io_hook = btree_readpage_end_io_hook,
4529 .readpage_io_failed_hook = btree_io_failed_hook,
4530 .submit_bio_hook = btree_submit_bio_hook,
4531 /* note we're sharing with inode.c for the merge bio hook */
4532 .merge_bio_hook = btrfs_merge_bio_hook,