2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include <linux/crc32c.h>
30 #include <linux/slab.h>
31 #include <linux/migrate.h>
32 #include <linux/ratelimit.h>
33 #include <asm/unaligned.h>
37 #include "transaction.h"
38 #include "btrfs_inode.h"
40 #include "print-tree.h"
41 #include "async-thread.h"
44 #include "free-space-cache.h"
45 #include "inode-map.h"
47 static struct extent_io_ops btree_extent_io_ops;
48 static void end_workqueue_fn(struct btrfs_work *work);
49 static void free_fs_root(struct btrfs_root *root);
50 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
52 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
53 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
55 struct btrfs_root *root);
56 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
57 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
58 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
59 struct extent_io_tree *dirty_pages,
61 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
62 struct extent_io_tree *pinned_extents);
63 static int btrfs_cleanup_transaction(struct btrfs_root *root);
66 * end_io_wq structs are used to do processing in task context when an IO is
67 * complete. This is used during reads to verify checksums, and it is used
68 * by writes to insert metadata for new file extents after IO is complete.
74 struct btrfs_fs_info *info;
77 struct list_head list;
78 struct btrfs_work work;
82 * async submit bios are used to offload expensive checksumming
83 * onto the worker threads. They checksum file and metadata bios
84 * just before they are sent down the IO stack.
86 struct async_submit_bio {
89 struct list_head list;
90 extent_submit_bio_hook_t *submit_bio_start;
91 extent_submit_bio_hook_t *submit_bio_done;
94 unsigned long bio_flags;
96 * bio_offset is optional, can be used if the pages in the bio
97 * can't tell us where in the file the bio should go
100 struct btrfs_work work;
104 * Lockdep class keys for extent_buffer->lock's in this root. For a given
105 * eb, the lockdep key is determined by the btrfs_root it belongs to and
106 * the level the eb occupies in the tree.
108 * Different roots are used for different purposes and may nest inside each
109 * other and they require separate keysets. As lockdep keys should be
110 * static, assign keysets according to the purpose of the root as indicated
111 * by btrfs_root->objectid. This ensures that all special purpose roots
112 * have separate keysets.
114 * Lock-nesting across peer nodes is always done with the immediate parent
115 * node locked thus preventing deadlock. As lockdep doesn't know this, use
116 * subclass to avoid triggering lockdep warning in such cases.
118 * The key is set by the readpage_end_io_hook after the buffer has passed
119 * csum validation but before the pages are unlocked. It is also set by
120 * btrfs_init_new_buffer on freshly allocated blocks.
122 * We also add a check to make sure the highest level of the tree is the
123 * same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this code
124 * needs update as well.
126 #ifdef CONFIG_DEBUG_LOCK_ALLOC
127 # if BTRFS_MAX_LEVEL != 8
131 static struct btrfs_lockdep_keyset {
132 u64 id; /* root objectid */
133 const char *name_stem; /* lock name stem */
134 char names[BTRFS_MAX_LEVEL + 1][20];
135 struct lock_class_key keys[BTRFS_MAX_LEVEL + 1];
136 } btrfs_lockdep_keysets[] = {
137 { .id = BTRFS_ROOT_TREE_OBJECTID, .name_stem = "root" },
138 { .id = BTRFS_EXTENT_TREE_OBJECTID, .name_stem = "extent" },
139 { .id = BTRFS_CHUNK_TREE_OBJECTID, .name_stem = "chunk" },
140 { .id = BTRFS_DEV_TREE_OBJECTID, .name_stem = "dev" },
141 { .id = BTRFS_FS_TREE_OBJECTID, .name_stem = "fs" },
142 { .id = BTRFS_CSUM_TREE_OBJECTID, .name_stem = "csum" },
143 { .id = BTRFS_ORPHAN_OBJECTID, .name_stem = "orphan" },
144 { .id = BTRFS_TREE_LOG_OBJECTID, .name_stem = "log" },
145 { .id = BTRFS_TREE_RELOC_OBJECTID, .name_stem = "treloc" },
146 { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc" },
147 { .id = 0, .name_stem = "tree" },
150 void __init btrfs_init_lockdep(void)
154 /* initialize lockdep class names */
155 for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
156 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
158 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
159 snprintf(ks->names[j], sizeof(ks->names[j]),
160 "btrfs-%s-%02d", ks->name_stem, j);
164 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
167 struct btrfs_lockdep_keyset *ks;
169 BUG_ON(level >= ARRAY_SIZE(ks->keys));
171 /* find the matching keyset, id 0 is the default entry */
172 for (ks = btrfs_lockdep_keysets; ks->id; ks++)
173 if (ks->id == objectid)
176 lockdep_set_class_and_name(&eb->lock,
177 &ks->keys[level], ks->names[level]);
183 * extents on the btree inode are pretty simple, there's one extent
184 * that covers the entire device
186 static struct extent_map *btree_get_extent(struct inode *inode,
187 struct page *page, size_t pg_offset, u64 start, u64 len,
190 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
191 struct extent_map *em;
194 read_lock(&em_tree->lock);
195 em = lookup_extent_mapping(em_tree, start, len);
198 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
199 read_unlock(&em_tree->lock);
202 read_unlock(&em_tree->lock);
204 em = alloc_extent_map();
206 em = ERR_PTR(-ENOMEM);
211 em->block_len = (u64)-1;
213 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
215 write_lock(&em_tree->lock);
216 ret = add_extent_mapping(em_tree, em);
217 if (ret == -EEXIST) {
218 u64 failed_start = em->start;
219 u64 failed_len = em->len;
222 em = lookup_extent_mapping(em_tree, start, len);
226 em = lookup_extent_mapping(em_tree, failed_start,
234 write_unlock(&em_tree->lock);
242 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
244 return crc32c(seed, data, len);
247 void btrfs_csum_final(u32 crc, char *result)
249 put_unaligned_le32(~crc, result);
253 * compute the csum for a btree block, and either verify it or write it
254 * into the csum field of the block.
256 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
260 btrfs_super_csum_size(&root->fs_info->super_copy);
263 unsigned long cur_len;
264 unsigned long offset = BTRFS_CSUM_SIZE;
266 unsigned long map_start;
267 unsigned long map_len;
270 unsigned long inline_result;
272 len = buf->len - offset;
274 err = map_private_extent_buffer(buf, offset, 32,
275 &kaddr, &map_start, &map_len);
278 cur_len = min(len, map_len - (offset - map_start));
279 crc = btrfs_csum_data(root, kaddr + offset - map_start,
284 if (csum_size > sizeof(inline_result)) {
285 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
289 result = (char *)&inline_result;
292 btrfs_csum_final(crc, result);
295 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
298 memcpy(&found, result, csum_size);
300 read_extent_buffer(buf, &val, 0, csum_size);
301 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
302 "failed on %llu wanted %X found %X "
304 root->fs_info->sb->s_id,
305 (unsigned long long)buf->start, val, found,
306 btrfs_header_level(buf));
307 if (result != (char *)&inline_result)
312 write_extent_buffer(buf, result, 0, csum_size);
314 if (result != (char *)&inline_result)
320 * we can't consider a given block up to date unless the transid of the
321 * block matches the transid in the parent node's pointer. This is how we
322 * detect blocks that either didn't get written at all or got written
323 * in the wrong place.
325 static int verify_parent_transid(struct extent_io_tree *io_tree,
326 struct extent_buffer *eb, u64 parent_transid)
328 struct extent_state *cached_state = NULL;
331 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
334 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
335 0, &cached_state, GFP_NOFS);
336 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
337 btrfs_header_generation(eb) == parent_transid) {
341 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
343 (unsigned long long)eb->start,
344 (unsigned long long)parent_transid,
345 (unsigned long long)btrfs_header_generation(eb));
347 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
349 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
350 &cached_state, GFP_NOFS);
355 * helper to read a given tree block, doing retries as required when
356 * the checksums don't match and we have alternate mirrors to try.
358 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
359 struct extent_buffer *eb,
360 u64 start, u64 parent_transid)
362 struct extent_io_tree *io_tree;
367 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
368 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
370 ret = read_extent_buffer_pages(io_tree, eb, start,
372 btree_get_extent, mirror_num);
374 !verify_parent_transid(io_tree, eb, parent_transid))
378 * This buffer's crc is fine, but its contents are corrupted, so
379 * there is no reason to read the other copies, they won't be
382 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
385 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
391 if (mirror_num > num_copies)
398 * checksum a dirty tree block before IO. This has extra checks to make sure
399 * we only fill in the checksum field in the first page of a multi-page block
402 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
404 struct extent_io_tree *tree;
405 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
408 struct extent_buffer *eb;
411 tree = &BTRFS_I(page->mapping->host)->io_tree;
413 if (page->private == EXTENT_PAGE_PRIVATE) {
417 if (!page->private) {
421 len = page->private >> 2;
424 eb = alloc_extent_buffer(tree, start, len, page);
429 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
430 btrfs_header_generation(eb));
432 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
434 found_start = btrfs_header_bytenr(eb);
435 if (found_start != start) {
439 if (eb->first_page != page) {
443 if (!PageUptodate(page)) {
447 csum_tree_block(root, eb, 0);
449 free_extent_buffer(eb);
454 static int check_tree_block_fsid(struct btrfs_root *root,
455 struct extent_buffer *eb)
457 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
458 u8 fsid[BTRFS_UUID_SIZE];
461 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
464 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
468 fs_devices = fs_devices->seed;
473 #define CORRUPT(reason, eb, root, slot) \
474 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
475 "root=%llu, slot=%d\n", reason, \
476 (unsigned long long)btrfs_header_bytenr(eb), \
477 (unsigned long long)root->objectid, slot)
479 static noinline int check_leaf(struct btrfs_root *root,
480 struct extent_buffer *leaf)
482 struct btrfs_key key;
483 struct btrfs_key leaf_key;
484 u32 nritems = btrfs_header_nritems(leaf);
490 /* Check the 0 item */
491 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
492 BTRFS_LEAF_DATA_SIZE(root)) {
493 CORRUPT("invalid item offset size pair", leaf, root, 0);
498 * Check to make sure each items keys are in the correct order and their
499 * offsets make sense. We only have to loop through nritems-1 because
500 * we check the current slot against the next slot, which verifies the
501 * next slot's offset+size makes sense and that the current's slot
504 for (slot = 0; slot < nritems - 1; slot++) {
505 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
506 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
508 /* Make sure the keys are in the right order */
509 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
510 CORRUPT("bad key order", leaf, root, slot);
515 * Make sure the offset and ends are right, remember that the
516 * item data starts at the end of the leaf and grows towards the
519 if (btrfs_item_offset_nr(leaf, slot) !=
520 btrfs_item_end_nr(leaf, slot + 1)) {
521 CORRUPT("slot offset bad", leaf, root, slot);
526 * Check to make sure that we don't point outside of the leaf,
527 * just incase all the items are consistent to eachother, but
528 * all point outside of the leaf.
530 if (btrfs_item_end_nr(leaf, slot) >
531 BTRFS_LEAF_DATA_SIZE(root)) {
532 CORRUPT("slot end outside of leaf", leaf, root, slot);
540 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
541 struct extent_state *state)
543 struct extent_io_tree *tree;
547 struct extent_buffer *eb;
548 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
551 tree = &BTRFS_I(page->mapping->host)->io_tree;
552 if (page->private == EXTENT_PAGE_PRIVATE)
557 len = page->private >> 2;
560 eb = alloc_extent_buffer(tree, start, len, page);
566 found_start = btrfs_header_bytenr(eb);
567 if (found_start != start) {
568 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
570 (unsigned long long)found_start,
571 (unsigned long long)eb->start);
575 if (eb->first_page != page) {
576 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
577 eb->first_page->index, page->index);
582 if (check_tree_block_fsid(root, eb)) {
583 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
584 (unsigned long long)eb->start);
588 found_level = btrfs_header_level(eb);
590 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
593 ret = csum_tree_block(root, eb, 1);
600 * If this is a leaf block and it is corrupt, set the corrupt bit so
601 * that we don't try and read the other copies of this block, just
604 if (found_level == 0 && check_leaf(root, eb)) {
605 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
609 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
610 end = eb->start + end - 1;
612 free_extent_buffer(eb);
617 static void end_workqueue_bio(struct bio *bio, int err)
619 struct end_io_wq *end_io_wq = bio->bi_private;
620 struct btrfs_fs_info *fs_info;
622 fs_info = end_io_wq->info;
623 end_io_wq->error = err;
624 end_io_wq->work.func = end_workqueue_fn;
625 end_io_wq->work.flags = 0;
627 if (bio->bi_rw & REQ_WRITE) {
628 if (end_io_wq->metadata == 1)
629 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
631 else if (end_io_wq->metadata == 2)
632 btrfs_queue_worker(&fs_info->endio_freespace_worker,
635 btrfs_queue_worker(&fs_info->endio_write_workers,
638 if (end_io_wq->metadata)
639 btrfs_queue_worker(&fs_info->endio_meta_workers,
642 btrfs_queue_worker(&fs_info->endio_workers,
648 * For the metadata arg you want
651 * 1 - if normal metadta
652 * 2 - if writing to the free space cache area
654 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
657 struct end_io_wq *end_io_wq;
658 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
662 end_io_wq->private = bio->bi_private;
663 end_io_wq->end_io = bio->bi_end_io;
664 end_io_wq->info = info;
665 end_io_wq->error = 0;
666 end_io_wq->bio = bio;
667 end_io_wq->metadata = metadata;
669 bio->bi_private = end_io_wq;
670 bio->bi_end_io = end_workqueue_bio;
674 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
676 unsigned long limit = min_t(unsigned long,
677 info->workers.max_workers,
678 info->fs_devices->open_devices);
682 static void run_one_async_start(struct btrfs_work *work)
684 struct async_submit_bio *async;
686 async = container_of(work, struct async_submit_bio, work);
687 async->submit_bio_start(async->inode, async->rw, async->bio,
688 async->mirror_num, async->bio_flags,
692 static void run_one_async_done(struct btrfs_work *work)
694 struct btrfs_fs_info *fs_info;
695 struct async_submit_bio *async;
698 async = container_of(work, struct async_submit_bio, work);
699 fs_info = BTRFS_I(async->inode)->root->fs_info;
701 limit = btrfs_async_submit_limit(fs_info);
702 limit = limit * 2 / 3;
704 atomic_dec(&fs_info->nr_async_submits);
706 if (atomic_read(&fs_info->nr_async_submits) < limit &&
707 waitqueue_active(&fs_info->async_submit_wait))
708 wake_up(&fs_info->async_submit_wait);
710 async->submit_bio_done(async->inode, async->rw, async->bio,
711 async->mirror_num, async->bio_flags,
715 static void run_one_async_free(struct btrfs_work *work)
717 struct async_submit_bio *async;
719 async = container_of(work, struct async_submit_bio, work);
723 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
724 int rw, struct bio *bio, int mirror_num,
725 unsigned long bio_flags,
727 extent_submit_bio_hook_t *submit_bio_start,
728 extent_submit_bio_hook_t *submit_bio_done)
730 struct async_submit_bio *async;
732 async = kmalloc(sizeof(*async), GFP_NOFS);
736 async->inode = inode;
739 async->mirror_num = mirror_num;
740 async->submit_bio_start = submit_bio_start;
741 async->submit_bio_done = submit_bio_done;
743 async->work.func = run_one_async_start;
744 async->work.ordered_func = run_one_async_done;
745 async->work.ordered_free = run_one_async_free;
747 async->work.flags = 0;
748 async->bio_flags = bio_flags;
749 async->bio_offset = bio_offset;
751 atomic_inc(&fs_info->nr_async_submits);
754 btrfs_set_work_high_prio(&async->work);
756 btrfs_queue_worker(&fs_info->workers, &async->work);
758 while (atomic_read(&fs_info->async_submit_draining) &&
759 atomic_read(&fs_info->nr_async_submits)) {
760 wait_event(fs_info->async_submit_wait,
761 (atomic_read(&fs_info->nr_async_submits) == 0));
767 static int btree_csum_one_bio(struct bio *bio)
769 struct bio_vec *bvec = bio->bi_io_vec;
771 struct btrfs_root *root;
773 WARN_ON(bio->bi_vcnt <= 0);
774 while (bio_index < bio->bi_vcnt) {
775 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
776 csum_dirty_buffer(root, bvec->bv_page);
783 static int __btree_submit_bio_start(struct inode *inode, int rw,
784 struct bio *bio, int mirror_num,
785 unsigned long bio_flags,
789 * when we're called for a write, we're already in the async
790 * submission context. Just jump into btrfs_map_bio
792 btree_csum_one_bio(bio);
796 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
797 int mirror_num, unsigned long bio_flags,
801 * when we're called for a write, we're already in the async
802 * submission context. Just jump into btrfs_map_bio
804 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
807 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
808 int mirror_num, unsigned long bio_flags,
813 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
817 if (!(rw & REQ_WRITE)) {
819 * called for a read, do the setup so that checksum validation
820 * can happen in the async kernel threads
822 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
827 * kthread helpers are used to submit writes so that checksumming
828 * can happen in parallel across all CPUs
830 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
831 inode, rw, bio, mirror_num, 0,
833 __btree_submit_bio_start,
834 __btree_submit_bio_done);
837 #ifdef CONFIG_MIGRATION
838 static int btree_migratepage(struct address_space *mapping,
839 struct page *newpage, struct page *page)
842 * we can't safely write a btree page from here,
843 * we haven't done the locking hook
848 * Buffers may be managed in a filesystem specific way.
849 * We must have no buffers or drop them.
851 if (page_has_private(page) &&
852 !try_to_release_page(page, GFP_KERNEL))
854 return migrate_page(mapping, newpage, page);
858 static int btree_writepage(struct page *page, struct writeback_control *wbc)
860 struct extent_io_tree *tree;
861 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
862 struct extent_buffer *eb;
865 tree = &BTRFS_I(page->mapping->host)->io_tree;
866 if (!(current->flags & PF_MEMALLOC)) {
867 return extent_write_full_page(tree, page,
868 btree_get_extent, wbc);
871 redirty_page_for_writepage(wbc, page);
872 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
875 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
877 spin_lock(&root->fs_info->delalloc_lock);
878 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
879 spin_unlock(&root->fs_info->delalloc_lock);
881 free_extent_buffer(eb);
887 static int btree_writepages(struct address_space *mapping,
888 struct writeback_control *wbc)
890 struct extent_io_tree *tree;
891 tree = &BTRFS_I(mapping->host)->io_tree;
892 if (wbc->sync_mode == WB_SYNC_NONE) {
893 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
895 unsigned long thresh = 32 * 1024 * 1024;
897 if (wbc->for_kupdate)
900 /* this is a bit racy, but that's ok */
901 num_dirty = root->fs_info->dirty_metadata_bytes;
902 if (num_dirty < thresh)
905 return extent_writepages(tree, mapping, btree_get_extent, wbc);
908 static int btree_readpage(struct file *file, struct page *page)
910 struct extent_io_tree *tree;
911 tree = &BTRFS_I(page->mapping->host)->io_tree;
912 return extent_read_full_page(tree, page, btree_get_extent);
915 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
917 struct extent_io_tree *tree;
918 struct extent_map_tree *map;
921 if (PageWriteback(page) || PageDirty(page))
924 tree = &BTRFS_I(page->mapping->host)->io_tree;
925 map = &BTRFS_I(page->mapping->host)->extent_tree;
927 ret = try_release_extent_state(map, tree, page, gfp_flags);
931 ret = try_release_extent_buffer(tree, page);
933 ClearPagePrivate(page);
934 set_page_private(page, 0);
935 page_cache_release(page);
941 static void btree_invalidatepage(struct page *page, unsigned long offset)
943 struct extent_io_tree *tree;
944 tree = &BTRFS_I(page->mapping->host)->io_tree;
945 extent_invalidatepage(tree, page, offset);
946 btree_releasepage(page, GFP_NOFS);
947 if (PagePrivate(page)) {
948 printk(KERN_WARNING "btrfs warning page private not zero "
949 "on page %llu\n", (unsigned long long)page_offset(page));
950 ClearPagePrivate(page);
951 set_page_private(page, 0);
952 page_cache_release(page);
956 static const struct address_space_operations btree_aops = {
957 .readpage = btree_readpage,
958 .writepage = btree_writepage,
959 .writepages = btree_writepages,
960 .releasepage = btree_releasepage,
961 .invalidatepage = btree_invalidatepage,
962 #ifdef CONFIG_MIGRATION
963 .migratepage = btree_migratepage,
967 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
970 struct extent_buffer *buf = NULL;
971 struct inode *btree_inode = root->fs_info->btree_inode;
974 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
977 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
978 buf, 0, WAIT_NONE, btree_get_extent, 0);
979 free_extent_buffer(buf);
983 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
984 u64 bytenr, u32 blocksize)
986 struct inode *btree_inode = root->fs_info->btree_inode;
987 struct extent_buffer *eb;
988 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
993 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
994 u64 bytenr, u32 blocksize)
996 struct inode *btree_inode = root->fs_info->btree_inode;
997 struct extent_buffer *eb;
999 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1000 bytenr, blocksize, NULL);
1005 int btrfs_write_tree_block(struct extent_buffer *buf)
1007 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1008 buf->start + buf->len - 1);
1011 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1013 return filemap_fdatawait_range(buf->first_page->mapping,
1014 buf->start, buf->start + buf->len - 1);
1017 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1018 u32 blocksize, u64 parent_transid)
1020 struct extent_buffer *buf = NULL;
1023 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1027 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1030 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1035 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1036 struct extent_buffer *buf)
1038 struct inode *btree_inode = root->fs_info->btree_inode;
1039 if (btrfs_header_generation(buf) ==
1040 root->fs_info->running_transaction->transid) {
1041 btrfs_assert_tree_locked(buf);
1043 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1044 spin_lock(&root->fs_info->delalloc_lock);
1045 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1046 root->fs_info->dirty_metadata_bytes -= buf->len;
1049 spin_unlock(&root->fs_info->delalloc_lock);
1052 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1053 btrfs_set_lock_blocking(buf);
1054 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1060 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1061 u32 stripesize, struct btrfs_root *root,
1062 struct btrfs_fs_info *fs_info,
1066 root->commit_root = NULL;
1067 root->sectorsize = sectorsize;
1068 root->nodesize = nodesize;
1069 root->leafsize = leafsize;
1070 root->stripesize = stripesize;
1072 root->track_dirty = 0;
1074 root->orphan_item_inserted = 0;
1075 root->orphan_cleanup_state = 0;
1077 root->fs_info = fs_info;
1078 root->objectid = objectid;
1079 root->last_trans = 0;
1080 root->highest_objectid = 0;
1082 root->inode_tree = RB_ROOT;
1083 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1084 root->block_rsv = NULL;
1085 root->orphan_block_rsv = NULL;
1087 INIT_LIST_HEAD(&root->dirty_list);
1088 INIT_LIST_HEAD(&root->orphan_list);
1089 INIT_LIST_HEAD(&root->root_list);
1090 spin_lock_init(&root->orphan_lock);
1091 spin_lock_init(&root->inode_lock);
1092 spin_lock_init(&root->accounting_lock);
1093 mutex_init(&root->objectid_mutex);
1094 mutex_init(&root->log_mutex);
1095 init_waitqueue_head(&root->log_writer_wait);
1096 init_waitqueue_head(&root->log_commit_wait[0]);
1097 init_waitqueue_head(&root->log_commit_wait[1]);
1098 atomic_set(&root->log_commit[0], 0);
1099 atomic_set(&root->log_commit[1], 0);
1100 atomic_set(&root->log_writers, 0);
1101 root->log_batch = 0;
1102 root->log_transid = 0;
1103 root->last_log_commit = 0;
1104 extent_io_tree_init(&root->dirty_log_pages,
1105 fs_info->btree_inode->i_mapping);
1107 memset(&root->root_key, 0, sizeof(root->root_key));
1108 memset(&root->root_item, 0, sizeof(root->root_item));
1109 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1110 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1111 root->defrag_trans_start = fs_info->generation;
1112 init_completion(&root->kobj_unregister);
1113 root->defrag_running = 0;
1114 root->root_key.objectid = objectid;
1119 static int find_and_setup_root(struct btrfs_root *tree_root,
1120 struct btrfs_fs_info *fs_info,
1122 struct btrfs_root *root)
1128 __setup_root(tree_root->nodesize, tree_root->leafsize,
1129 tree_root->sectorsize, tree_root->stripesize,
1130 root, fs_info, objectid);
1131 ret = btrfs_find_last_root(tree_root, objectid,
1132 &root->root_item, &root->root_key);
1137 generation = btrfs_root_generation(&root->root_item);
1138 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1139 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1140 blocksize, generation);
1141 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1142 free_extent_buffer(root->node);
1145 root->commit_root = btrfs_root_node(root);
1149 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1150 struct btrfs_fs_info *fs_info)
1152 struct btrfs_root *root;
1153 struct btrfs_root *tree_root = fs_info->tree_root;
1154 struct extent_buffer *leaf;
1156 root = kzalloc(sizeof(*root), GFP_NOFS);
1158 return ERR_PTR(-ENOMEM);
1160 __setup_root(tree_root->nodesize, tree_root->leafsize,
1161 tree_root->sectorsize, tree_root->stripesize,
1162 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1164 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1165 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1166 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1168 * log trees do not get reference counted because they go away
1169 * before a real commit is actually done. They do store pointers
1170 * to file data extents, and those reference counts still get
1171 * updated (along with back refs to the log tree).
1175 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1176 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1179 return ERR_CAST(leaf);
1182 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1183 btrfs_set_header_bytenr(leaf, leaf->start);
1184 btrfs_set_header_generation(leaf, trans->transid);
1185 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1186 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1189 write_extent_buffer(root->node, root->fs_info->fsid,
1190 (unsigned long)btrfs_header_fsid(root->node),
1192 btrfs_mark_buffer_dirty(root->node);
1193 btrfs_tree_unlock(root->node);
1197 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1198 struct btrfs_fs_info *fs_info)
1200 struct btrfs_root *log_root;
1202 log_root = alloc_log_tree(trans, fs_info);
1203 if (IS_ERR(log_root))
1204 return PTR_ERR(log_root);
1205 WARN_ON(fs_info->log_root_tree);
1206 fs_info->log_root_tree = log_root;
1210 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1211 struct btrfs_root *root)
1213 struct btrfs_root *log_root;
1214 struct btrfs_inode_item *inode_item;
1216 log_root = alloc_log_tree(trans, root->fs_info);
1217 if (IS_ERR(log_root))
1218 return PTR_ERR(log_root);
1220 log_root->last_trans = trans->transid;
1221 log_root->root_key.offset = root->root_key.objectid;
1223 inode_item = &log_root->root_item.inode;
1224 inode_item->generation = cpu_to_le64(1);
1225 inode_item->size = cpu_to_le64(3);
1226 inode_item->nlink = cpu_to_le32(1);
1227 inode_item->nbytes = cpu_to_le64(root->leafsize);
1228 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1230 btrfs_set_root_node(&log_root->root_item, log_root->node);
1232 WARN_ON(root->log_root);
1233 root->log_root = log_root;
1234 root->log_transid = 0;
1235 root->last_log_commit = 0;
1239 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1240 struct btrfs_key *location)
1242 struct btrfs_root *root;
1243 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1244 struct btrfs_path *path;
1245 struct extent_buffer *l;
1250 root = kzalloc(sizeof(*root), GFP_NOFS);
1252 return ERR_PTR(-ENOMEM);
1253 if (location->offset == (u64)-1) {
1254 ret = find_and_setup_root(tree_root, fs_info,
1255 location->objectid, root);
1258 return ERR_PTR(ret);
1263 __setup_root(tree_root->nodesize, tree_root->leafsize,
1264 tree_root->sectorsize, tree_root->stripesize,
1265 root, fs_info, location->objectid);
1267 path = btrfs_alloc_path();
1270 return ERR_PTR(-ENOMEM);
1272 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1275 read_extent_buffer(l, &root->root_item,
1276 btrfs_item_ptr_offset(l, path->slots[0]),
1277 sizeof(root->root_item));
1278 memcpy(&root->root_key, location, sizeof(*location));
1280 btrfs_free_path(path);
1285 return ERR_PTR(ret);
1288 generation = btrfs_root_generation(&root->root_item);
1289 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1290 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1291 blocksize, generation);
1292 root->commit_root = btrfs_root_node(root);
1293 BUG_ON(!root->node);
1295 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1297 btrfs_check_and_init_root_item(&root->root_item);
1303 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1304 struct btrfs_key *location)
1306 struct btrfs_root *root;
1309 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1310 return fs_info->tree_root;
1311 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1312 return fs_info->extent_root;
1313 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1314 return fs_info->chunk_root;
1315 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1316 return fs_info->dev_root;
1317 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1318 return fs_info->csum_root;
1320 spin_lock(&fs_info->fs_roots_radix_lock);
1321 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1322 (unsigned long)location->objectid);
1323 spin_unlock(&fs_info->fs_roots_radix_lock);
1327 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1331 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1332 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1334 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1339 btrfs_init_free_ino_ctl(root);
1340 mutex_init(&root->fs_commit_mutex);
1341 spin_lock_init(&root->cache_lock);
1342 init_waitqueue_head(&root->cache_wait);
1344 ret = get_anon_bdev(&root->anon_dev);
1348 if (btrfs_root_refs(&root->root_item) == 0) {
1353 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1357 root->orphan_item_inserted = 1;
1359 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1363 spin_lock(&fs_info->fs_roots_radix_lock);
1364 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1365 (unsigned long)root->root_key.objectid,
1370 spin_unlock(&fs_info->fs_roots_radix_lock);
1371 radix_tree_preload_end();
1373 if (ret == -EEXIST) {
1380 ret = btrfs_find_dead_roots(fs_info->tree_root,
1381 root->root_key.objectid);
1386 return ERR_PTR(ret);
1389 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1391 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1393 struct btrfs_device *device;
1394 struct backing_dev_info *bdi;
1397 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1400 bdi = blk_get_backing_dev_info(device->bdev);
1401 if (bdi && bdi_congested(bdi, bdi_bits)) {
1411 * If this fails, caller must call bdi_destroy() to get rid of the
1414 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1418 bdi->capabilities = BDI_CAP_MAP_COPY;
1419 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1423 bdi->ra_pages = default_backing_dev_info.ra_pages;
1424 bdi->congested_fn = btrfs_congested_fn;
1425 bdi->congested_data = info;
1429 static int bio_ready_for_csum(struct bio *bio)
1435 struct extent_io_tree *io_tree = NULL;
1436 struct bio_vec *bvec;
1440 bio_for_each_segment(bvec, bio, i) {
1441 page = bvec->bv_page;
1442 if (page->private == EXTENT_PAGE_PRIVATE) {
1443 length += bvec->bv_len;
1446 if (!page->private) {
1447 length += bvec->bv_len;
1450 length = bvec->bv_len;
1451 buf_len = page->private >> 2;
1452 start = page_offset(page) + bvec->bv_offset;
1453 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1455 /* are we fully contained in this bio? */
1456 if (buf_len <= length)
1459 ret = extent_range_uptodate(io_tree, start + length,
1460 start + buf_len - 1);
1465 * called by the kthread helper functions to finally call the bio end_io
1466 * functions. This is where read checksum verification actually happens
1468 static void end_workqueue_fn(struct btrfs_work *work)
1471 struct end_io_wq *end_io_wq;
1472 struct btrfs_fs_info *fs_info;
1475 end_io_wq = container_of(work, struct end_io_wq, work);
1476 bio = end_io_wq->bio;
1477 fs_info = end_io_wq->info;
1479 /* metadata bio reads are special because the whole tree block must
1480 * be checksummed at once. This makes sure the entire block is in
1481 * ram and up to date before trying to verify things. For
1482 * blocksize <= pagesize, it is basically a noop
1484 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1485 !bio_ready_for_csum(bio)) {
1486 btrfs_queue_worker(&fs_info->endio_meta_workers,
1490 error = end_io_wq->error;
1491 bio->bi_private = end_io_wq->private;
1492 bio->bi_end_io = end_io_wq->end_io;
1494 bio_endio(bio, error);
1497 static int cleaner_kthread(void *arg)
1499 struct btrfs_root *root = arg;
1502 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1504 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1505 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1506 btrfs_run_delayed_iputs(root);
1507 btrfs_clean_old_snapshots(root);
1508 mutex_unlock(&root->fs_info->cleaner_mutex);
1509 btrfs_run_defrag_inodes(root->fs_info);
1512 if (freezing(current)) {
1515 set_current_state(TASK_INTERRUPTIBLE);
1516 if (!kthread_should_stop())
1518 __set_current_state(TASK_RUNNING);
1520 } while (!kthread_should_stop());
1524 static int transaction_kthread(void *arg)
1526 struct btrfs_root *root = arg;
1527 struct btrfs_trans_handle *trans;
1528 struct btrfs_transaction *cur;
1531 unsigned long delay;
1536 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1537 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1539 spin_lock(&root->fs_info->trans_lock);
1540 cur = root->fs_info->running_transaction;
1542 spin_unlock(&root->fs_info->trans_lock);
1546 now = get_seconds();
1547 if (!cur->blocked &&
1548 (now < cur->start_time || now - cur->start_time < 30)) {
1549 spin_unlock(&root->fs_info->trans_lock);
1553 transid = cur->transid;
1554 spin_unlock(&root->fs_info->trans_lock);
1556 trans = btrfs_join_transaction(root);
1557 BUG_ON(IS_ERR(trans));
1558 if (transid == trans->transid) {
1559 ret = btrfs_commit_transaction(trans, root);
1562 btrfs_end_transaction(trans, root);
1565 wake_up_process(root->fs_info->cleaner_kthread);
1566 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1568 if (freezing(current)) {
1571 set_current_state(TASK_INTERRUPTIBLE);
1572 if (!kthread_should_stop() &&
1573 !btrfs_transaction_blocked(root->fs_info))
1574 schedule_timeout(delay);
1575 __set_current_state(TASK_RUNNING);
1577 } while (!kthread_should_stop());
1581 struct btrfs_root *open_ctree(struct super_block *sb,
1582 struct btrfs_fs_devices *fs_devices,
1592 struct btrfs_key location;
1593 struct buffer_head *bh;
1594 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1596 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1598 struct btrfs_root *tree_root = btrfs_sb(sb);
1599 struct btrfs_fs_info *fs_info = NULL;
1600 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1602 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1604 struct btrfs_root *log_tree_root;
1609 struct btrfs_super_block *disk_super;
1611 if (!extent_root || !tree_root || !tree_root->fs_info ||
1612 !chunk_root || !dev_root || !csum_root) {
1616 fs_info = tree_root->fs_info;
1618 ret = init_srcu_struct(&fs_info->subvol_srcu);
1624 ret = setup_bdi(fs_info, &fs_info->bdi);
1630 fs_info->btree_inode = new_inode(sb);
1631 if (!fs_info->btree_inode) {
1636 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1638 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1639 INIT_LIST_HEAD(&fs_info->trans_list);
1640 INIT_LIST_HEAD(&fs_info->dead_roots);
1641 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1642 INIT_LIST_HEAD(&fs_info->hashers);
1643 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1644 INIT_LIST_HEAD(&fs_info->ordered_operations);
1645 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1646 spin_lock_init(&fs_info->delalloc_lock);
1647 spin_lock_init(&fs_info->trans_lock);
1648 spin_lock_init(&fs_info->ref_cache_lock);
1649 spin_lock_init(&fs_info->fs_roots_radix_lock);
1650 spin_lock_init(&fs_info->delayed_iput_lock);
1651 spin_lock_init(&fs_info->defrag_inodes_lock);
1652 mutex_init(&fs_info->reloc_mutex);
1654 init_completion(&fs_info->kobj_unregister);
1655 fs_info->tree_root = tree_root;
1656 fs_info->extent_root = extent_root;
1657 fs_info->csum_root = csum_root;
1658 fs_info->chunk_root = chunk_root;
1659 fs_info->dev_root = dev_root;
1660 fs_info->fs_devices = fs_devices;
1661 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1662 INIT_LIST_HEAD(&fs_info->space_info);
1663 btrfs_mapping_init(&fs_info->mapping_tree);
1664 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1665 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1666 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1667 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1668 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1669 INIT_LIST_HEAD(&fs_info->durable_block_rsv_list);
1670 mutex_init(&fs_info->durable_block_rsv_mutex);
1671 atomic_set(&fs_info->nr_async_submits, 0);
1672 atomic_set(&fs_info->async_delalloc_pages, 0);
1673 atomic_set(&fs_info->async_submit_draining, 0);
1674 atomic_set(&fs_info->nr_async_bios, 0);
1675 atomic_set(&fs_info->defrag_running, 0);
1677 fs_info->max_inline = 8192 * 1024;
1678 fs_info->metadata_ratio = 0;
1679 fs_info->defrag_inodes = RB_ROOT;
1680 fs_info->trans_no_join = 0;
1682 fs_info->thread_pool_size = min_t(unsigned long,
1683 num_online_cpus() + 2, 8);
1685 INIT_LIST_HEAD(&fs_info->ordered_extents);
1686 spin_lock_init(&fs_info->ordered_extent_lock);
1687 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1689 if (!fs_info->delayed_root) {
1693 btrfs_init_delayed_root(fs_info->delayed_root);
1695 mutex_init(&fs_info->scrub_lock);
1696 atomic_set(&fs_info->scrubs_running, 0);
1697 atomic_set(&fs_info->scrub_pause_req, 0);
1698 atomic_set(&fs_info->scrubs_paused, 0);
1699 atomic_set(&fs_info->scrub_cancel_req, 0);
1700 init_waitqueue_head(&fs_info->scrub_pause_wait);
1701 init_rwsem(&fs_info->scrub_super_lock);
1702 fs_info->scrub_workers_refcnt = 0;
1704 sb->s_blocksize = 4096;
1705 sb->s_blocksize_bits = blksize_bits(4096);
1706 sb->s_bdi = &fs_info->bdi;
1708 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1709 fs_info->btree_inode->i_nlink = 1;
1711 * we set the i_size on the btree inode to the max possible int.
1712 * the real end of the address space is determined by all of
1713 * the devices in the system
1715 fs_info->btree_inode->i_size = OFFSET_MAX;
1716 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1717 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1719 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1720 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1721 fs_info->btree_inode->i_mapping);
1722 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1724 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1726 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1727 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1728 sizeof(struct btrfs_key));
1729 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1730 insert_inode_hash(fs_info->btree_inode);
1732 spin_lock_init(&fs_info->block_group_cache_lock);
1733 fs_info->block_group_cache_tree = RB_ROOT;
1735 extent_io_tree_init(&fs_info->freed_extents[0],
1736 fs_info->btree_inode->i_mapping);
1737 extent_io_tree_init(&fs_info->freed_extents[1],
1738 fs_info->btree_inode->i_mapping);
1739 fs_info->pinned_extents = &fs_info->freed_extents[0];
1740 fs_info->do_barriers = 1;
1743 mutex_init(&fs_info->ordered_operations_mutex);
1744 mutex_init(&fs_info->tree_log_mutex);
1745 mutex_init(&fs_info->chunk_mutex);
1746 mutex_init(&fs_info->transaction_kthread_mutex);
1747 mutex_init(&fs_info->cleaner_mutex);
1748 mutex_init(&fs_info->volume_mutex);
1749 init_rwsem(&fs_info->extent_commit_sem);
1750 init_rwsem(&fs_info->cleanup_work_sem);
1751 init_rwsem(&fs_info->subvol_sem);
1753 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1754 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1756 init_waitqueue_head(&fs_info->transaction_throttle);
1757 init_waitqueue_head(&fs_info->transaction_wait);
1758 init_waitqueue_head(&fs_info->transaction_blocked_wait);
1759 init_waitqueue_head(&fs_info->async_submit_wait);
1761 __setup_root(4096, 4096, 4096, 4096, tree_root,
1762 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1764 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1770 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1771 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1772 sizeof(fs_info->super_for_commit));
1775 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1777 disk_super = &fs_info->super_copy;
1778 if (!btrfs_super_root(disk_super))
1781 /* check FS state, whether FS is broken. */
1782 fs_info->fs_state |= btrfs_super_flags(disk_super);
1784 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1787 * In the long term, we'll store the compression type in the super
1788 * block, and it'll be used for per file compression control.
1790 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1792 ret = btrfs_parse_options(tree_root, options);
1798 features = btrfs_super_incompat_flags(disk_super) &
1799 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1801 printk(KERN_ERR "BTRFS: couldn't mount because of "
1802 "unsupported optional features (%Lx).\n",
1803 (unsigned long long)features);
1808 features = btrfs_super_incompat_flags(disk_super);
1809 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1810 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1811 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1812 btrfs_set_super_incompat_flags(disk_super, features);
1814 features = btrfs_super_compat_ro_flags(disk_super) &
1815 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1816 if (!(sb->s_flags & MS_RDONLY) && features) {
1817 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1818 "unsupported option features (%Lx).\n",
1819 (unsigned long long)features);
1824 btrfs_init_workers(&fs_info->generic_worker,
1825 "genwork", 1, NULL);
1827 btrfs_init_workers(&fs_info->workers, "worker",
1828 fs_info->thread_pool_size,
1829 &fs_info->generic_worker);
1831 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1832 fs_info->thread_pool_size,
1833 &fs_info->generic_worker);
1835 btrfs_init_workers(&fs_info->submit_workers, "submit",
1836 min_t(u64, fs_devices->num_devices,
1837 fs_info->thread_pool_size),
1838 &fs_info->generic_worker);
1840 btrfs_init_workers(&fs_info->caching_workers, "cache",
1841 2, &fs_info->generic_worker);
1843 /* a higher idle thresh on the submit workers makes it much more
1844 * likely that bios will be send down in a sane order to the
1847 fs_info->submit_workers.idle_thresh = 64;
1849 fs_info->workers.idle_thresh = 16;
1850 fs_info->workers.ordered = 1;
1852 fs_info->delalloc_workers.idle_thresh = 2;
1853 fs_info->delalloc_workers.ordered = 1;
1855 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1856 &fs_info->generic_worker);
1857 btrfs_init_workers(&fs_info->endio_workers, "endio",
1858 fs_info->thread_pool_size,
1859 &fs_info->generic_worker);
1860 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1861 fs_info->thread_pool_size,
1862 &fs_info->generic_worker);
1863 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1864 "endio-meta-write", fs_info->thread_pool_size,
1865 &fs_info->generic_worker);
1866 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1867 fs_info->thread_pool_size,
1868 &fs_info->generic_worker);
1869 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1870 1, &fs_info->generic_worker);
1871 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
1872 fs_info->thread_pool_size,
1873 &fs_info->generic_worker);
1876 * endios are largely parallel and should have a very
1879 fs_info->endio_workers.idle_thresh = 4;
1880 fs_info->endio_meta_workers.idle_thresh = 4;
1882 fs_info->endio_write_workers.idle_thresh = 2;
1883 fs_info->endio_meta_write_workers.idle_thresh = 2;
1885 btrfs_start_workers(&fs_info->workers, 1);
1886 btrfs_start_workers(&fs_info->generic_worker, 1);
1887 btrfs_start_workers(&fs_info->submit_workers, 1);
1888 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1889 btrfs_start_workers(&fs_info->fixup_workers, 1);
1890 btrfs_start_workers(&fs_info->endio_workers, 1);
1891 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1892 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1893 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1894 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1895 btrfs_start_workers(&fs_info->delayed_workers, 1);
1896 btrfs_start_workers(&fs_info->caching_workers, 1);
1898 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1899 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1900 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1902 nodesize = btrfs_super_nodesize(disk_super);
1903 leafsize = btrfs_super_leafsize(disk_super);
1904 sectorsize = btrfs_super_sectorsize(disk_super);
1905 stripesize = btrfs_super_stripesize(disk_super);
1906 tree_root->nodesize = nodesize;
1907 tree_root->leafsize = leafsize;
1908 tree_root->sectorsize = sectorsize;
1909 tree_root->stripesize = stripesize;
1911 sb->s_blocksize = sectorsize;
1912 sb->s_blocksize_bits = blksize_bits(sectorsize);
1914 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1915 sizeof(disk_super->magic))) {
1916 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1917 goto fail_sb_buffer;
1920 mutex_lock(&fs_info->chunk_mutex);
1921 ret = btrfs_read_sys_array(tree_root);
1922 mutex_unlock(&fs_info->chunk_mutex);
1924 printk(KERN_WARNING "btrfs: failed to read the system "
1925 "array on %s\n", sb->s_id);
1926 goto fail_sb_buffer;
1929 blocksize = btrfs_level_size(tree_root,
1930 btrfs_super_chunk_root_level(disk_super));
1931 generation = btrfs_super_chunk_root_generation(disk_super);
1933 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1934 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1936 chunk_root->node = read_tree_block(chunk_root,
1937 btrfs_super_chunk_root(disk_super),
1938 blocksize, generation);
1939 BUG_ON(!chunk_root->node);
1940 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1941 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1943 goto fail_chunk_root;
1945 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1946 chunk_root->commit_root = btrfs_root_node(chunk_root);
1948 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1949 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1952 mutex_lock(&fs_info->chunk_mutex);
1953 ret = btrfs_read_chunk_tree(chunk_root);
1954 mutex_unlock(&fs_info->chunk_mutex);
1956 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1958 goto fail_chunk_root;
1961 btrfs_close_extra_devices(fs_devices);
1963 blocksize = btrfs_level_size(tree_root,
1964 btrfs_super_root_level(disk_super));
1965 generation = btrfs_super_generation(disk_super);
1967 tree_root->node = read_tree_block(tree_root,
1968 btrfs_super_root(disk_super),
1969 blocksize, generation);
1970 if (!tree_root->node)
1971 goto fail_chunk_root;
1972 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1973 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1975 goto fail_tree_root;
1977 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1978 tree_root->commit_root = btrfs_root_node(tree_root);
1980 ret = find_and_setup_root(tree_root, fs_info,
1981 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1983 goto fail_tree_root;
1984 extent_root->track_dirty = 1;
1986 ret = find_and_setup_root(tree_root, fs_info,
1987 BTRFS_DEV_TREE_OBJECTID, dev_root);
1989 goto fail_extent_root;
1990 dev_root->track_dirty = 1;
1992 ret = find_and_setup_root(tree_root, fs_info,
1993 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1997 csum_root->track_dirty = 1;
1999 fs_info->generation = generation;
2000 fs_info->last_trans_committed = generation;
2001 fs_info->data_alloc_profile = (u64)-1;
2002 fs_info->metadata_alloc_profile = (u64)-1;
2003 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2005 ret = btrfs_init_space_info(fs_info);
2007 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2008 goto fail_block_groups;
2011 ret = btrfs_read_block_groups(extent_root);
2013 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2014 goto fail_block_groups;
2017 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2019 if (IS_ERR(fs_info->cleaner_kthread))
2020 goto fail_block_groups;
2022 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2024 "btrfs-transaction");
2025 if (IS_ERR(fs_info->transaction_kthread))
2028 if (!btrfs_test_opt(tree_root, SSD) &&
2029 !btrfs_test_opt(tree_root, NOSSD) &&
2030 !fs_info->fs_devices->rotating) {
2031 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2033 btrfs_set_opt(fs_info->mount_opt, SSD);
2036 /* do not make disk changes in broken FS */
2037 if (btrfs_super_log_root(disk_super) != 0 &&
2038 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2039 u64 bytenr = btrfs_super_log_root(disk_super);
2041 if (fs_devices->rw_devices == 0) {
2042 printk(KERN_WARNING "Btrfs log replay required "
2045 goto fail_trans_kthread;
2048 btrfs_level_size(tree_root,
2049 btrfs_super_log_root_level(disk_super));
2051 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2052 if (!log_tree_root) {
2054 goto fail_trans_kthread;
2057 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2058 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2060 log_tree_root->node = read_tree_block(tree_root, bytenr,
2063 ret = btrfs_recover_log_trees(log_tree_root);
2066 if (sb->s_flags & MS_RDONLY) {
2067 ret = btrfs_commit_super(tree_root);
2072 ret = btrfs_find_orphan_roots(tree_root);
2075 if (!(sb->s_flags & MS_RDONLY)) {
2076 ret = btrfs_cleanup_fs_roots(fs_info);
2079 ret = btrfs_recover_relocation(tree_root);
2082 "btrfs: failed to recover relocation\n");
2084 goto fail_trans_kthread;
2088 location.objectid = BTRFS_FS_TREE_OBJECTID;
2089 location.type = BTRFS_ROOT_ITEM_KEY;
2090 location.offset = (u64)-1;
2092 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2093 if (!fs_info->fs_root)
2094 goto fail_trans_kthread;
2095 if (IS_ERR(fs_info->fs_root)) {
2096 err = PTR_ERR(fs_info->fs_root);
2097 goto fail_trans_kthread;
2100 if (!(sb->s_flags & MS_RDONLY)) {
2101 down_read(&fs_info->cleanup_work_sem);
2102 err = btrfs_orphan_cleanup(fs_info->fs_root);
2104 err = btrfs_orphan_cleanup(fs_info->tree_root);
2105 up_read(&fs_info->cleanup_work_sem);
2107 close_ctree(tree_root);
2108 return ERR_PTR(err);
2115 kthread_stop(fs_info->transaction_kthread);
2117 kthread_stop(fs_info->cleaner_kthread);
2120 * make sure we're done with the btree inode before we stop our
2123 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2124 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2127 btrfs_free_block_groups(fs_info);
2128 free_extent_buffer(csum_root->node);
2129 free_extent_buffer(csum_root->commit_root);
2131 free_extent_buffer(dev_root->node);
2132 free_extent_buffer(dev_root->commit_root);
2134 free_extent_buffer(extent_root->node);
2135 free_extent_buffer(extent_root->commit_root);
2137 free_extent_buffer(tree_root->node);
2138 free_extent_buffer(tree_root->commit_root);
2140 free_extent_buffer(chunk_root->node);
2141 free_extent_buffer(chunk_root->commit_root);
2143 btrfs_stop_workers(&fs_info->generic_worker);
2144 btrfs_stop_workers(&fs_info->fixup_workers);
2145 btrfs_stop_workers(&fs_info->delalloc_workers);
2146 btrfs_stop_workers(&fs_info->workers);
2147 btrfs_stop_workers(&fs_info->endio_workers);
2148 btrfs_stop_workers(&fs_info->endio_meta_workers);
2149 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2150 btrfs_stop_workers(&fs_info->endio_write_workers);
2151 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2152 btrfs_stop_workers(&fs_info->submit_workers);
2153 btrfs_stop_workers(&fs_info->delayed_workers);
2154 btrfs_stop_workers(&fs_info->caching_workers);
2156 kfree(fs_info->delayed_root);
2158 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2159 iput(fs_info->btree_inode);
2161 btrfs_close_devices(fs_info->fs_devices);
2162 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2164 bdi_destroy(&fs_info->bdi);
2166 cleanup_srcu_struct(&fs_info->subvol_srcu);
2174 return ERR_PTR(err);
2177 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2179 char b[BDEVNAME_SIZE];
2182 set_buffer_uptodate(bh);
2184 printk_ratelimited(KERN_WARNING "lost page write due to "
2185 "I/O error on %s\n",
2186 bdevname(bh->b_bdev, b));
2187 /* note, we dont' set_buffer_write_io_error because we have
2188 * our own ways of dealing with the IO errors
2190 clear_buffer_uptodate(bh);
2196 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2198 struct buffer_head *bh;
2199 struct buffer_head *latest = NULL;
2200 struct btrfs_super_block *super;
2205 /* we would like to check all the supers, but that would make
2206 * a btrfs mount succeed after a mkfs from a different FS.
2207 * So, we need to add a special mount option to scan for
2208 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2210 for (i = 0; i < 1; i++) {
2211 bytenr = btrfs_sb_offset(i);
2212 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2214 bh = __bread(bdev, bytenr / 4096, 4096);
2218 super = (struct btrfs_super_block *)bh->b_data;
2219 if (btrfs_super_bytenr(super) != bytenr ||
2220 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2221 sizeof(super->magic))) {
2226 if (!latest || btrfs_super_generation(super) > transid) {
2229 transid = btrfs_super_generation(super);
2238 * this should be called twice, once with wait == 0 and
2239 * once with wait == 1. When wait == 0 is done, all the buffer heads
2240 * we write are pinned.
2242 * They are released when wait == 1 is done.
2243 * max_mirrors must be the same for both runs, and it indicates how
2244 * many supers on this one device should be written.
2246 * max_mirrors == 0 means to write them all.
2248 static int write_dev_supers(struct btrfs_device *device,
2249 struct btrfs_super_block *sb,
2250 int do_barriers, int wait, int max_mirrors)
2252 struct buffer_head *bh;
2258 int last_barrier = 0;
2260 if (max_mirrors == 0)
2261 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2263 /* make sure only the last submit_bh does a barrier */
2265 for (i = 0; i < max_mirrors; i++) {
2266 bytenr = btrfs_sb_offset(i);
2267 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2268 device->total_bytes)
2274 for (i = 0; i < max_mirrors; i++) {
2275 bytenr = btrfs_sb_offset(i);
2276 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2280 bh = __find_get_block(device->bdev, bytenr / 4096,
2281 BTRFS_SUPER_INFO_SIZE);
2284 if (!buffer_uptodate(bh))
2287 /* drop our reference */
2290 /* drop the reference from the wait == 0 run */
2294 btrfs_set_super_bytenr(sb, bytenr);
2297 crc = btrfs_csum_data(NULL, (char *)sb +
2298 BTRFS_CSUM_SIZE, crc,
2299 BTRFS_SUPER_INFO_SIZE -
2301 btrfs_csum_final(crc, sb->csum);
2304 * one reference for us, and we leave it for the
2307 bh = __getblk(device->bdev, bytenr / 4096,
2308 BTRFS_SUPER_INFO_SIZE);
2309 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2311 /* one reference for submit_bh */
2314 set_buffer_uptodate(bh);
2316 bh->b_end_io = btrfs_end_buffer_write_sync;
2319 if (i == last_barrier && do_barriers)
2320 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2322 ret = submit_bh(WRITE_SYNC, bh);
2327 return errors < i ? 0 : -1;
2330 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2332 struct list_head *head;
2333 struct btrfs_device *dev;
2334 struct btrfs_super_block *sb;
2335 struct btrfs_dev_item *dev_item;
2339 int total_errors = 0;
2342 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2343 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2345 sb = &root->fs_info->super_for_commit;
2346 dev_item = &sb->dev_item;
2348 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2349 head = &root->fs_info->fs_devices->devices;
2350 list_for_each_entry_rcu(dev, head, dev_list) {
2355 if (!dev->in_fs_metadata || !dev->writeable)
2358 btrfs_set_stack_device_generation(dev_item, 0);
2359 btrfs_set_stack_device_type(dev_item, dev->type);
2360 btrfs_set_stack_device_id(dev_item, dev->devid);
2361 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2362 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2363 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2364 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2365 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2366 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2367 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2369 flags = btrfs_super_flags(sb);
2370 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2372 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2376 if (total_errors > max_errors) {
2377 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2383 list_for_each_entry_rcu(dev, head, dev_list) {
2386 if (!dev->in_fs_metadata || !dev->writeable)
2389 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2393 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2394 if (total_errors > max_errors) {
2395 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2402 int write_ctree_super(struct btrfs_trans_handle *trans,
2403 struct btrfs_root *root, int max_mirrors)
2407 ret = write_all_supers(root, max_mirrors);
2411 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2413 spin_lock(&fs_info->fs_roots_radix_lock);
2414 radix_tree_delete(&fs_info->fs_roots_radix,
2415 (unsigned long)root->root_key.objectid);
2416 spin_unlock(&fs_info->fs_roots_radix_lock);
2418 if (btrfs_root_refs(&root->root_item) == 0)
2419 synchronize_srcu(&fs_info->subvol_srcu);
2421 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2422 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2427 static void free_fs_root(struct btrfs_root *root)
2429 iput(root->cache_inode);
2430 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2432 free_anon_bdev(root->anon_dev);
2433 free_extent_buffer(root->node);
2434 free_extent_buffer(root->commit_root);
2435 kfree(root->free_ino_ctl);
2436 kfree(root->free_ino_pinned);
2441 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2444 struct btrfs_root *gang[8];
2447 while (!list_empty(&fs_info->dead_roots)) {
2448 gang[0] = list_entry(fs_info->dead_roots.next,
2449 struct btrfs_root, root_list);
2450 list_del(&gang[0]->root_list);
2452 if (gang[0]->in_radix) {
2453 btrfs_free_fs_root(fs_info, gang[0]);
2455 free_extent_buffer(gang[0]->node);
2456 free_extent_buffer(gang[0]->commit_root);
2462 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2467 for (i = 0; i < ret; i++)
2468 btrfs_free_fs_root(fs_info, gang[i]);
2473 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2475 u64 root_objectid = 0;
2476 struct btrfs_root *gang[8];
2481 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2482 (void **)gang, root_objectid,
2487 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2488 for (i = 0; i < ret; i++) {
2491 root_objectid = gang[i]->root_key.objectid;
2492 err = btrfs_orphan_cleanup(gang[i]);
2501 int btrfs_commit_super(struct btrfs_root *root)
2503 struct btrfs_trans_handle *trans;
2506 mutex_lock(&root->fs_info->cleaner_mutex);
2507 btrfs_run_delayed_iputs(root);
2508 btrfs_clean_old_snapshots(root);
2509 mutex_unlock(&root->fs_info->cleaner_mutex);
2511 /* wait until ongoing cleanup work done */
2512 down_write(&root->fs_info->cleanup_work_sem);
2513 up_write(&root->fs_info->cleanup_work_sem);
2515 trans = btrfs_join_transaction(root);
2517 return PTR_ERR(trans);
2518 ret = btrfs_commit_transaction(trans, root);
2520 /* run commit again to drop the original snapshot */
2521 trans = btrfs_join_transaction(root);
2523 return PTR_ERR(trans);
2524 btrfs_commit_transaction(trans, root);
2525 ret = btrfs_write_and_wait_transaction(NULL, root);
2528 ret = write_ctree_super(NULL, root, 0);
2532 int close_ctree(struct btrfs_root *root)
2534 struct btrfs_fs_info *fs_info = root->fs_info;
2537 fs_info->closing = 1;
2540 btrfs_scrub_cancel(root);
2542 /* wait for any defraggers to finish */
2543 wait_event(fs_info->transaction_wait,
2544 (atomic_read(&fs_info->defrag_running) == 0));
2546 /* clear out the rbtree of defraggable inodes */
2547 btrfs_run_defrag_inodes(root->fs_info);
2549 btrfs_put_block_group_cache(fs_info);
2552 * Here come 2 situations when btrfs is broken to flip readonly:
2554 * 1. when btrfs flips readonly somewhere else before
2555 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2556 * and btrfs will skip to write sb directly to keep
2557 * ERROR state on disk.
2559 * 2. when btrfs flips readonly just in btrfs_commit_super,
2560 * and in such case, btrfs cannot write sb via btrfs_commit_super,
2561 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2562 * btrfs will cleanup all FS resources first and write sb then.
2564 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2565 ret = btrfs_commit_super(root);
2567 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2570 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2571 ret = btrfs_error_commit_super(root);
2573 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2576 kthread_stop(root->fs_info->transaction_kthread);
2577 kthread_stop(root->fs_info->cleaner_kthread);
2579 fs_info->closing = 2;
2582 if (fs_info->delalloc_bytes) {
2583 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2584 (unsigned long long)fs_info->delalloc_bytes);
2586 if (fs_info->total_ref_cache_size) {
2587 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2588 (unsigned long long)fs_info->total_ref_cache_size);
2591 free_extent_buffer(fs_info->extent_root->node);
2592 free_extent_buffer(fs_info->extent_root->commit_root);
2593 free_extent_buffer(fs_info->tree_root->node);
2594 free_extent_buffer(fs_info->tree_root->commit_root);
2595 free_extent_buffer(root->fs_info->chunk_root->node);
2596 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2597 free_extent_buffer(root->fs_info->dev_root->node);
2598 free_extent_buffer(root->fs_info->dev_root->commit_root);
2599 free_extent_buffer(root->fs_info->csum_root->node);
2600 free_extent_buffer(root->fs_info->csum_root->commit_root);
2602 btrfs_free_block_groups(root->fs_info);
2604 del_fs_roots(fs_info);
2606 iput(fs_info->btree_inode);
2607 kfree(fs_info->delayed_root);
2609 btrfs_stop_workers(&fs_info->generic_worker);
2610 btrfs_stop_workers(&fs_info->fixup_workers);
2611 btrfs_stop_workers(&fs_info->delalloc_workers);
2612 btrfs_stop_workers(&fs_info->workers);
2613 btrfs_stop_workers(&fs_info->endio_workers);
2614 btrfs_stop_workers(&fs_info->endio_meta_workers);
2615 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2616 btrfs_stop_workers(&fs_info->endio_write_workers);
2617 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2618 btrfs_stop_workers(&fs_info->submit_workers);
2619 btrfs_stop_workers(&fs_info->delayed_workers);
2620 btrfs_stop_workers(&fs_info->caching_workers);
2622 btrfs_close_devices(fs_info->fs_devices);
2623 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2625 bdi_destroy(&fs_info->bdi);
2626 cleanup_srcu_struct(&fs_info->subvol_srcu);
2628 kfree(fs_info->extent_root);
2629 kfree(fs_info->tree_root);
2630 kfree(fs_info->chunk_root);
2631 kfree(fs_info->dev_root);
2632 kfree(fs_info->csum_root);
2638 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2641 struct inode *btree_inode = buf->first_page->mapping->host;
2643 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2648 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2653 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2655 struct inode *btree_inode = buf->first_page->mapping->host;
2656 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2660 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2662 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2663 u64 transid = btrfs_header_generation(buf);
2664 struct inode *btree_inode = root->fs_info->btree_inode;
2667 btrfs_assert_tree_locked(buf);
2668 if (transid != root->fs_info->generation) {
2669 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2670 "found %llu running %llu\n",
2671 (unsigned long long)buf->start,
2672 (unsigned long long)transid,
2673 (unsigned long long)root->fs_info->generation);
2676 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2679 spin_lock(&root->fs_info->delalloc_lock);
2680 root->fs_info->dirty_metadata_bytes += buf->len;
2681 spin_unlock(&root->fs_info->delalloc_lock);
2685 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2688 * looks as though older kernels can get into trouble with
2689 * this code, they end up stuck in balance_dirty_pages forever
2692 unsigned long thresh = 32 * 1024 * 1024;
2694 if (current->flags & PF_MEMALLOC)
2697 btrfs_balance_delayed_items(root);
2699 num_dirty = root->fs_info->dirty_metadata_bytes;
2701 if (num_dirty > thresh) {
2702 balance_dirty_pages_ratelimited_nr(
2703 root->fs_info->btree_inode->i_mapping, 1);
2708 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2711 * looks as though older kernels can get into trouble with
2712 * this code, they end up stuck in balance_dirty_pages forever
2715 unsigned long thresh = 32 * 1024 * 1024;
2717 if (current->flags & PF_MEMALLOC)
2720 num_dirty = root->fs_info->dirty_metadata_bytes;
2722 if (num_dirty > thresh) {
2723 balance_dirty_pages_ratelimited_nr(
2724 root->fs_info->btree_inode->i_mapping, 1);
2729 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2731 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2733 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2735 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2739 int btree_lock_page_hook(struct page *page)
2741 struct inode *inode = page->mapping->host;
2742 struct btrfs_root *root = BTRFS_I(inode)->root;
2743 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2744 struct extent_buffer *eb;
2746 u64 bytenr = page_offset(page);
2748 if (page->private == EXTENT_PAGE_PRIVATE)
2751 len = page->private >> 2;
2752 eb = find_extent_buffer(io_tree, bytenr, len);
2756 btrfs_tree_lock(eb);
2757 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2759 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2760 spin_lock(&root->fs_info->delalloc_lock);
2761 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2762 root->fs_info->dirty_metadata_bytes -= eb->len;
2765 spin_unlock(&root->fs_info->delalloc_lock);
2768 btrfs_tree_unlock(eb);
2769 free_extent_buffer(eb);
2775 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2781 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2782 printk(KERN_WARNING "warning: mount fs with errors, "
2783 "running btrfsck is recommended\n");
2786 int btrfs_error_commit_super(struct btrfs_root *root)
2790 mutex_lock(&root->fs_info->cleaner_mutex);
2791 btrfs_run_delayed_iputs(root);
2792 mutex_unlock(&root->fs_info->cleaner_mutex);
2794 down_write(&root->fs_info->cleanup_work_sem);
2795 up_write(&root->fs_info->cleanup_work_sem);
2797 /* cleanup FS via transaction */
2798 btrfs_cleanup_transaction(root);
2800 ret = write_ctree_super(NULL, root, 0);
2805 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2807 struct btrfs_inode *btrfs_inode;
2808 struct list_head splice;
2810 INIT_LIST_HEAD(&splice);
2812 mutex_lock(&root->fs_info->ordered_operations_mutex);
2813 spin_lock(&root->fs_info->ordered_extent_lock);
2815 list_splice_init(&root->fs_info->ordered_operations, &splice);
2816 while (!list_empty(&splice)) {
2817 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2818 ordered_operations);
2820 list_del_init(&btrfs_inode->ordered_operations);
2822 btrfs_invalidate_inodes(btrfs_inode->root);
2825 spin_unlock(&root->fs_info->ordered_extent_lock);
2826 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2831 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2833 struct list_head splice;
2834 struct btrfs_ordered_extent *ordered;
2835 struct inode *inode;
2837 INIT_LIST_HEAD(&splice);
2839 spin_lock(&root->fs_info->ordered_extent_lock);
2841 list_splice_init(&root->fs_info->ordered_extents, &splice);
2842 while (!list_empty(&splice)) {
2843 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2846 list_del_init(&ordered->root_extent_list);
2847 atomic_inc(&ordered->refs);
2849 /* the inode may be getting freed (in sys_unlink path). */
2850 inode = igrab(ordered->inode);
2852 spin_unlock(&root->fs_info->ordered_extent_lock);
2856 atomic_set(&ordered->refs, 1);
2857 btrfs_put_ordered_extent(ordered);
2859 spin_lock(&root->fs_info->ordered_extent_lock);
2862 spin_unlock(&root->fs_info->ordered_extent_lock);
2867 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2868 struct btrfs_root *root)
2870 struct rb_node *node;
2871 struct btrfs_delayed_ref_root *delayed_refs;
2872 struct btrfs_delayed_ref_node *ref;
2875 delayed_refs = &trans->delayed_refs;
2877 spin_lock(&delayed_refs->lock);
2878 if (delayed_refs->num_entries == 0) {
2879 spin_unlock(&delayed_refs->lock);
2880 printk(KERN_INFO "delayed_refs has NO entry\n");
2884 node = rb_first(&delayed_refs->root);
2886 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2887 node = rb_next(node);
2890 rb_erase(&ref->rb_node, &delayed_refs->root);
2891 delayed_refs->num_entries--;
2893 atomic_set(&ref->refs, 1);
2894 if (btrfs_delayed_ref_is_head(ref)) {
2895 struct btrfs_delayed_ref_head *head;
2897 head = btrfs_delayed_node_to_head(ref);
2898 mutex_lock(&head->mutex);
2899 kfree(head->extent_op);
2900 delayed_refs->num_heads--;
2901 if (list_empty(&head->cluster))
2902 delayed_refs->num_heads_ready--;
2903 list_del_init(&head->cluster);
2904 mutex_unlock(&head->mutex);
2907 spin_unlock(&delayed_refs->lock);
2908 btrfs_put_delayed_ref(ref);
2911 spin_lock(&delayed_refs->lock);
2914 spin_unlock(&delayed_refs->lock);
2919 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2921 struct btrfs_pending_snapshot *snapshot;
2922 struct list_head splice;
2924 INIT_LIST_HEAD(&splice);
2926 list_splice_init(&t->pending_snapshots, &splice);
2928 while (!list_empty(&splice)) {
2929 snapshot = list_entry(splice.next,
2930 struct btrfs_pending_snapshot,
2933 list_del_init(&snapshot->list);
2941 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2943 struct btrfs_inode *btrfs_inode;
2944 struct list_head splice;
2946 INIT_LIST_HEAD(&splice);
2948 spin_lock(&root->fs_info->delalloc_lock);
2949 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2951 while (!list_empty(&splice)) {
2952 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2955 list_del_init(&btrfs_inode->delalloc_inodes);
2957 btrfs_invalidate_inodes(btrfs_inode->root);
2960 spin_unlock(&root->fs_info->delalloc_lock);
2965 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2966 struct extent_io_tree *dirty_pages,
2971 struct inode *btree_inode = root->fs_info->btree_inode;
2972 struct extent_buffer *eb;
2976 unsigned long index;
2979 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2984 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2985 while (start <= end) {
2986 index = start >> PAGE_CACHE_SHIFT;
2987 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2988 page = find_get_page(btree_inode->i_mapping, index);
2991 offset = page_offset(page);
2993 spin_lock(&dirty_pages->buffer_lock);
2994 eb = radix_tree_lookup(
2995 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2996 offset >> PAGE_CACHE_SHIFT);
2997 spin_unlock(&dirty_pages->buffer_lock);
2999 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3001 atomic_set(&eb->refs, 1);
3003 if (PageWriteback(page))
3004 end_page_writeback(page);
3007 if (PageDirty(page)) {
3008 clear_page_dirty_for_io(page);
3009 spin_lock_irq(&page->mapping->tree_lock);
3010 radix_tree_tag_clear(&page->mapping->page_tree,
3012 PAGECACHE_TAG_DIRTY);
3013 spin_unlock_irq(&page->mapping->tree_lock);
3016 page->mapping->a_ops->invalidatepage(page, 0);
3024 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3025 struct extent_io_tree *pinned_extents)
3027 struct extent_io_tree *unpin;
3032 unpin = pinned_extents;
3034 ret = find_first_extent_bit(unpin, 0, &start, &end,
3040 if (btrfs_test_opt(root, DISCARD))
3041 ret = btrfs_error_discard_extent(root, start,
3045 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3046 btrfs_error_unpin_extent_range(root, start, end);
3053 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3055 struct btrfs_transaction *t;
3060 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3062 spin_lock(&root->fs_info->trans_lock);
3063 list_splice_init(&root->fs_info->trans_list, &list);
3064 root->fs_info->trans_no_join = 1;
3065 spin_unlock(&root->fs_info->trans_lock);
3067 while (!list_empty(&list)) {
3068 t = list_entry(list.next, struct btrfs_transaction, list);
3072 btrfs_destroy_ordered_operations(root);
3074 btrfs_destroy_ordered_extents(root);
3076 btrfs_destroy_delayed_refs(t, root);
3078 btrfs_block_rsv_release(root,
3079 &root->fs_info->trans_block_rsv,
3080 t->dirty_pages.dirty_bytes);
3082 /* FIXME: cleanup wait for commit */
3085 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3086 wake_up(&root->fs_info->transaction_blocked_wait);
3089 if (waitqueue_active(&root->fs_info->transaction_wait))
3090 wake_up(&root->fs_info->transaction_wait);
3093 if (waitqueue_active(&t->commit_wait))
3094 wake_up(&t->commit_wait);
3096 btrfs_destroy_pending_snapshots(t);
3098 btrfs_destroy_delalloc_inodes(root);
3100 spin_lock(&root->fs_info->trans_lock);
3101 root->fs_info->running_transaction = NULL;
3102 spin_unlock(&root->fs_info->trans_lock);
3104 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3107 btrfs_destroy_pinned_extent(root,
3108 root->fs_info->pinned_extents);
3110 atomic_set(&t->use_count, 0);
3111 list_del_init(&t->list);
3112 memset(t, 0, sizeof(*t));
3113 kmem_cache_free(btrfs_transaction_cachep, t);
3116 spin_lock(&root->fs_info->trans_lock);
3117 root->fs_info->trans_no_join = 0;
3118 spin_unlock(&root->fs_info->trans_lock);
3119 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3124 static struct extent_io_ops btree_extent_io_ops = {
3125 .write_cache_pages_lock_hook = btree_lock_page_hook,
3126 .readpage_end_io_hook = btree_readpage_end_io_hook,
3127 .submit_bio_hook = btree_submit_bio_hook,
3128 /* note we're sharing with inode.c for the merge bio hook */
3129 .merge_bio_hook = btrfs_merge_bio_hook,