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,
259 u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
262 unsigned long cur_len;
263 unsigned long offset = BTRFS_CSUM_SIZE;
265 unsigned long map_start;
266 unsigned long map_len;
269 unsigned long inline_result;
271 len = buf->len - offset;
273 err = map_private_extent_buffer(buf, offset, 32,
274 &kaddr, &map_start, &map_len);
277 cur_len = min(len, map_len - (offset - map_start));
278 crc = btrfs_csum_data(root, kaddr + offset - map_start,
283 if (csum_size > sizeof(inline_result)) {
284 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
288 result = (char *)&inline_result;
291 btrfs_csum_final(crc, result);
294 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
297 memcpy(&found, result, csum_size);
299 read_extent_buffer(buf, &val, 0, csum_size);
300 printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
301 "failed on %llu wanted %X found %X "
303 root->fs_info->sb->s_id,
304 (unsigned long long)buf->start, val, found,
305 btrfs_header_level(buf));
306 if (result != (char *)&inline_result)
311 write_extent_buffer(buf, result, 0, csum_size);
313 if (result != (char *)&inline_result)
319 * we can't consider a given block up to date unless the transid of the
320 * block matches the transid in the parent node's pointer. This is how we
321 * detect blocks that either didn't get written at all or got written
322 * in the wrong place.
324 static int verify_parent_transid(struct extent_io_tree *io_tree,
325 struct extent_buffer *eb, u64 parent_transid)
327 struct extent_state *cached_state = NULL;
330 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
333 lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
334 0, &cached_state, GFP_NOFS);
335 if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
336 btrfs_header_generation(eb) == parent_transid) {
340 printk_ratelimited("parent transid verify failed on %llu wanted %llu "
342 (unsigned long long)eb->start,
343 (unsigned long long)parent_transid,
344 (unsigned long long)btrfs_header_generation(eb));
346 clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
348 unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
349 &cached_state, GFP_NOFS);
354 * helper to read a given tree block, doing retries as required when
355 * the checksums don't match and we have alternate mirrors to try.
357 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
358 struct extent_buffer *eb,
359 u64 start, u64 parent_transid)
361 struct extent_io_tree *io_tree;
366 clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
367 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
369 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
370 btree_get_extent, mirror_num);
372 !verify_parent_transid(io_tree, eb, parent_transid))
376 * This buffer's crc is fine, but its contents are corrupted, so
377 * there is no reason to read the other copies, they won't be
380 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
383 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
389 if (mirror_num > num_copies)
396 * checksum a dirty tree block before IO. This has extra checks to make sure
397 * we only fill in the checksum field in the first page of a multi-page block
400 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
402 struct extent_io_tree *tree;
403 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
406 struct extent_buffer *eb;
409 tree = &BTRFS_I(page->mapping->host)->io_tree;
411 if (page->private == EXTENT_PAGE_PRIVATE) {
415 if (!page->private) {
419 len = page->private >> 2;
422 eb = alloc_extent_buffer(tree, start, len, page);
427 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
428 btrfs_header_generation(eb));
430 WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
432 found_start = btrfs_header_bytenr(eb);
433 if (found_start != start) {
437 if (eb->first_page != page) {
441 if (!PageUptodate(page)) {
445 csum_tree_block(root, eb, 0);
447 free_extent_buffer(eb);
452 static int check_tree_block_fsid(struct btrfs_root *root,
453 struct extent_buffer *eb)
455 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
456 u8 fsid[BTRFS_UUID_SIZE];
459 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
462 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
466 fs_devices = fs_devices->seed;
471 #define CORRUPT(reason, eb, root, slot) \
472 printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
473 "root=%llu, slot=%d\n", reason, \
474 (unsigned long long)btrfs_header_bytenr(eb), \
475 (unsigned long long)root->objectid, slot)
477 static noinline int check_leaf(struct btrfs_root *root,
478 struct extent_buffer *leaf)
480 struct btrfs_key key;
481 struct btrfs_key leaf_key;
482 u32 nritems = btrfs_header_nritems(leaf);
488 /* Check the 0 item */
489 if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
490 BTRFS_LEAF_DATA_SIZE(root)) {
491 CORRUPT("invalid item offset size pair", leaf, root, 0);
496 * Check to make sure each items keys are in the correct order and their
497 * offsets make sense. We only have to loop through nritems-1 because
498 * we check the current slot against the next slot, which verifies the
499 * next slot's offset+size makes sense and that the current's slot
502 for (slot = 0; slot < nritems - 1; slot++) {
503 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
504 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
506 /* Make sure the keys are in the right order */
507 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
508 CORRUPT("bad key order", leaf, root, slot);
513 * Make sure the offset and ends are right, remember that the
514 * item data starts at the end of the leaf and grows towards the
517 if (btrfs_item_offset_nr(leaf, slot) !=
518 btrfs_item_end_nr(leaf, slot + 1)) {
519 CORRUPT("slot offset bad", leaf, root, slot);
524 * Check to make sure that we don't point outside of the leaf,
525 * just incase all the items are consistent to eachother, but
526 * all point outside of the leaf.
528 if (btrfs_item_end_nr(leaf, slot) >
529 BTRFS_LEAF_DATA_SIZE(root)) {
530 CORRUPT("slot end outside of leaf", leaf, root, slot);
538 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
539 struct extent_state *state)
541 struct extent_io_tree *tree;
545 struct extent_buffer *eb;
546 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
549 tree = &BTRFS_I(page->mapping->host)->io_tree;
550 if (page->private == EXTENT_PAGE_PRIVATE)
555 len = page->private >> 2;
558 eb = alloc_extent_buffer(tree, start, len, page);
564 found_start = btrfs_header_bytenr(eb);
565 if (found_start != start) {
566 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
568 (unsigned long long)found_start,
569 (unsigned long long)eb->start);
573 if (eb->first_page != page) {
574 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
575 eb->first_page->index, page->index);
580 if (check_tree_block_fsid(root, eb)) {
581 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
582 (unsigned long long)eb->start);
586 found_level = btrfs_header_level(eb);
588 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
591 ret = csum_tree_block(root, eb, 1);
598 * If this is a leaf block and it is corrupt, set the corrupt bit so
599 * that we don't try and read the other copies of this block, just
602 if (found_level == 0 && check_leaf(root, eb)) {
603 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
607 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
608 end = eb->start + end - 1;
610 free_extent_buffer(eb);
615 static void end_workqueue_bio(struct bio *bio, int err)
617 struct end_io_wq *end_io_wq = bio->bi_private;
618 struct btrfs_fs_info *fs_info;
620 fs_info = end_io_wq->info;
621 end_io_wq->error = err;
622 end_io_wq->work.func = end_workqueue_fn;
623 end_io_wq->work.flags = 0;
625 if (bio->bi_rw & REQ_WRITE) {
626 if (end_io_wq->metadata == 1)
627 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
629 else if (end_io_wq->metadata == 2)
630 btrfs_queue_worker(&fs_info->endio_freespace_worker,
633 btrfs_queue_worker(&fs_info->endio_write_workers,
636 if (end_io_wq->metadata)
637 btrfs_queue_worker(&fs_info->endio_meta_workers,
640 btrfs_queue_worker(&fs_info->endio_workers,
646 * For the metadata arg you want
649 * 1 - if normal metadta
650 * 2 - if writing to the free space cache area
652 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
655 struct end_io_wq *end_io_wq;
656 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
660 end_io_wq->private = bio->bi_private;
661 end_io_wq->end_io = bio->bi_end_io;
662 end_io_wq->info = info;
663 end_io_wq->error = 0;
664 end_io_wq->bio = bio;
665 end_io_wq->metadata = metadata;
667 bio->bi_private = end_io_wq;
668 bio->bi_end_io = end_workqueue_bio;
672 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
674 unsigned long limit = min_t(unsigned long,
675 info->workers.max_workers,
676 info->fs_devices->open_devices);
680 static void run_one_async_start(struct btrfs_work *work)
682 struct async_submit_bio *async;
684 async = container_of(work, struct async_submit_bio, work);
685 async->submit_bio_start(async->inode, async->rw, async->bio,
686 async->mirror_num, async->bio_flags,
690 static void run_one_async_done(struct btrfs_work *work)
692 struct btrfs_fs_info *fs_info;
693 struct async_submit_bio *async;
696 async = container_of(work, struct async_submit_bio, work);
697 fs_info = BTRFS_I(async->inode)->root->fs_info;
699 limit = btrfs_async_submit_limit(fs_info);
700 limit = limit * 2 / 3;
702 atomic_dec(&fs_info->nr_async_submits);
704 if (atomic_read(&fs_info->nr_async_submits) < limit &&
705 waitqueue_active(&fs_info->async_submit_wait))
706 wake_up(&fs_info->async_submit_wait);
708 async->submit_bio_done(async->inode, async->rw, async->bio,
709 async->mirror_num, async->bio_flags,
713 static void run_one_async_free(struct btrfs_work *work)
715 struct async_submit_bio *async;
717 async = container_of(work, struct async_submit_bio, work);
721 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
722 int rw, struct bio *bio, int mirror_num,
723 unsigned long bio_flags,
725 extent_submit_bio_hook_t *submit_bio_start,
726 extent_submit_bio_hook_t *submit_bio_done)
728 struct async_submit_bio *async;
730 async = kmalloc(sizeof(*async), GFP_NOFS);
734 async->inode = inode;
737 async->mirror_num = mirror_num;
738 async->submit_bio_start = submit_bio_start;
739 async->submit_bio_done = submit_bio_done;
741 async->work.func = run_one_async_start;
742 async->work.ordered_func = run_one_async_done;
743 async->work.ordered_free = run_one_async_free;
745 async->work.flags = 0;
746 async->bio_flags = bio_flags;
747 async->bio_offset = bio_offset;
749 atomic_inc(&fs_info->nr_async_submits);
752 btrfs_set_work_high_prio(&async->work);
754 btrfs_queue_worker(&fs_info->workers, &async->work);
756 while (atomic_read(&fs_info->async_submit_draining) &&
757 atomic_read(&fs_info->nr_async_submits)) {
758 wait_event(fs_info->async_submit_wait,
759 (atomic_read(&fs_info->nr_async_submits) == 0));
765 static int btree_csum_one_bio(struct bio *bio)
767 struct bio_vec *bvec = bio->bi_io_vec;
769 struct btrfs_root *root;
771 WARN_ON(bio->bi_vcnt <= 0);
772 while (bio_index < bio->bi_vcnt) {
773 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
774 csum_dirty_buffer(root, bvec->bv_page);
781 static int __btree_submit_bio_start(struct inode *inode, int rw,
782 struct bio *bio, int mirror_num,
783 unsigned long bio_flags,
787 * when we're called for a write, we're already in the async
788 * submission context. Just jump into btrfs_map_bio
790 btree_csum_one_bio(bio);
794 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
795 int mirror_num, unsigned long bio_flags,
799 * when we're called for a write, we're already in the async
800 * submission context. Just jump into btrfs_map_bio
802 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
805 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
806 int mirror_num, unsigned long bio_flags,
811 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
815 if (!(rw & REQ_WRITE)) {
817 * called for a read, do the setup so that checksum validation
818 * can happen in the async kernel threads
820 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
825 * kthread helpers are used to submit writes so that checksumming
826 * can happen in parallel across all CPUs
828 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
829 inode, rw, bio, mirror_num, 0,
831 __btree_submit_bio_start,
832 __btree_submit_bio_done);
835 #ifdef CONFIG_MIGRATION
836 static int btree_migratepage(struct address_space *mapping,
837 struct page *newpage, struct page *page)
840 * we can't safely write a btree page from here,
841 * we haven't done the locking hook
846 * Buffers may be managed in a filesystem specific way.
847 * We must have no buffers or drop them.
849 if (page_has_private(page) &&
850 !try_to_release_page(page, GFP_KERNEL))
852 return migrate_page(mapping, newpage, page);
856 static int btree_writepage(struct page *page, struct writeback_control *wbc)
858 struct extent_io_tree *tree;
859 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
860 struct extent_buffer *eb;
863 tree = &BTRFS_I(page->mapping->host)->io_tree;
864 if (!(current->flags & PF_MEMALLOC)) {
865 return extent_write_full_page(tree, page,
866 btree_get_extent, wbc);
869 redirty_page_for_writepage(wbc, page);
870 eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
873 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
875 spin_lock(&root->fs_info->delalloc_lock);
876 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
877 spin_unlock(&root->fs_info->delalloc_lock);
879 free_extent_buffer(eb);
885 static int btree_writepages(struct address_space *mapping,
886 struct writeback_control *wbc)
888 struct extent_io_tree *tree;
889 tree = &BTRFS_I(mapping->host)->io_tree;
890 if (wbc->sync_mode == WB_SYNC_NONE) {
891 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
893 unsigned long thresh = 32 * 1024 * 1024;
895 if (wbc->for_kupdate)
898 /* this is a bit racy, but that's ok */
899 num_dirty = root->fs_info->dirty_metadata_bytes;
900 if (num_dirty < thresh)
903 return extent_writepages(tree, mapping, btree_get_extent, wbc);
906 static int btree_readpage(struct file *file, struct page *page)
908 struct extent_io_tree *tree;
909 tree = &BTRFS_I(page->mapping->host)->io_tree;
910 return extent_read_full_page(tree, page, btree_get_extent);
913 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
915 struct extent_io_tree *tree;
916 struct extent_map_tree *map;
919 if (PageWriteback(page) || PageDirty(page))
922 tree = &BTRFS_I(page->mapping->host)->io_tree;
923 map = &BTRFS_I(page->mapping->host)->extent_tree;
925 ret = try_release_extent_state(map, tree, page, gfp_flags);
929 ret = try_release_extent_buffer(tree, page);
931 ClearPagePrivate(page);
932 set_page_private(page, 0);
933 page_cache_release(page);
939 static void btree_invalidatepage(struct page *page, unsigned long offset)
941 struct extent_io_tree *tree;
942 tree = &BTRFS_I(page->mapping->host)->io_tree;
943 extent_invalidatepage(tree, page, offset);
944 btree_releasepage(page, GFP_NOFS);
945 if (PagePrivate(page)) {
946 printk(KERN_WARNING "btrfs warning page private not zero "
947 "on page %llu\n", (unsigned long long)page_offset(page));
948 ClearPagePrivate(page);
949 set_page_private(page, 0);
950 page_cache_release(page);
954 static const struct address_space_operations btree_aops = {
955 .readpage = btree_readpage,
956 .writepage = btree_writepage,
957 .writepages = btree_writepages,
958 .releasepage = btree_releasepage,
959 .invalidatepage = btree_invalidatepage,
960 #ifdef CONFIG_MIGRATION
961 .migratepage = btree_migratepage,
965 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
968 struct extent_buffer *buf = NULL;
969 struct inode *btree_inode = root->fs_info->btree_inode;
972 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
975 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
976 buf, 0, 0, btree_get_extent, 0);
977 free_extent_buffer(buf);
981 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
982 u64 bytenr, u32 blocksize)
984 struct inode *btree_inode = root->fs_info->btree_inode;
985 struct extent_buffer *eb;
986 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
991 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
992 u64 bytenr, u32 blocksize)
994 struct inode *btree_inode = root->fs_info->btree_inode;
995 struct extent_buffer *eb;
997 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
998 bytenr, blocksize, NULL);
1003 int btrfs_write_tree_block(struct extent_buffer *buf)
1005 return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1006 buf->start + buf->len - 1);
1009 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1011 return filemap_fdatawait_range(buf->first_page->mapping,
1012 buf->start, buf->start + buf->len - 1);
1015 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1016 u32 blocksize, u64 parent_transid)
1018 struct extent_buffer *buf = NULL;
1021 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1025 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1028 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1033 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1034 struct extent_buffer *buf)
1036 struct inode *btree_inode = root->fs_info->btree_inode;
1037 if (btrfs_header_generation(buf) ==
1038 root->fs_info->running_transaction->transid) {
1039 btrfs_assert_tree_locked(buf);
1041 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1042 spin_lock(&root->fs_info->delalloc_lock);
1043 if (root->fs_info->dirty_metadata_bytes >= buf->len)
1044 root->fs_info->dirty_metadata_bytes -= buf->len;
1047 spin_unlock(&root->fs_info->delalloc_lock);
1050 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1051 btrfs_set_lock_blocking(buf);
1052 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1058 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1059 u32 stripesize, struct btrfs_root *root,
1060 struct btrfs_fs_info *fs_info,
1064 root->commit_root = NULL;
1065 root->sectorsize = sectorsize;
1066 root->nodesize = nodesize;
1067 root->leafsize = leafsize;
1068 root->stripesize = stripesize;
1070 root->track_dirty = 0;
1072 root->orphan_item_inserted = 0;
1073 root->orphan_cleanup_state = 0;
1075 root->fs_info = fs_info;
1076 root->objectid = objectid;
1077 root->last_trans = 0;
1078 root->highest_objectid = 0;
1080 root->inode_tree = RB_ROOT;
1081 INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1082 root->block_rsv = NULL;
1083 root->orphan_block_rsv = NULL;
1085 INIT_LIST_HEAD(&root->dirty_list);
1086 INIT_LIST_HEAD(&root->orphan_list);
1087 INIT_LIST_HEAD(&root->root_list);
1088 spin_lock_init(&root->orphan_lock);
1089 spin_lock_init(&root->inode_lock);
1090 spin_lock_init(&root->accounting_lock);
1091 mutex_init(&root->objectid_mutex);
1092 mutex_init(&root->log_mutex);
1093 init_waitqueue_head(&root->log_writer_wait);
1094 init_waitqueue_head(&root->log_commit_wait[0]);
1095 init_waitqueue_head(&root->log_commit_wait[1]);
1096 atomic_set(&root->log_commit[0], 0);
1097 atomic_set(&root->log_commit[1], 0);
1098 atomic_set(&root->log_writers, 0);
1099 root->log_batch = 0;
1100 root->log_transid = 0;
1101 root->last_log_commit = 0;
1102 extent_io_tree_init(&root->dirty_log_pages,
1103 fs_info->btree_inode->i_mapping);
1105 memset(&root->root_key, 0, sizeof(root->root_key));
1106 memset(&root->root_item, 0, sizeof(root->root_item));
1107 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1108 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1109 root->defrag_trans_start = fs_info->generation;
1110 init_completion(&root->kobj_unregister);
1111 root->defrag_running = 0;
1112 root->root_key.objectid = objectid;
1117 static int find_and_setup_root(struct btrfs_root *tree_root,
1118 struct btrfs_fs_info *fs_info,
1120 struct btrfs_root *root)
1126 __setup_root(tree_root->nodesize, tree_root->leafsize,
1127 tree_root->sectorsize, tree_root->stripesize,
1128 root, fs_info, objectid);
1129 ret = btrfs_find_last_root(tree_root, objectid,
1130 &root->root_item, &root->root_key);
1135 generation = btrfs_root_generation(&root->root_item);
1136 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1137 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1138 blocksize, generation);
1139 if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1140 free_extent_buffer(root->node);
1143 root->commit_root = btrfs_root_node(root);
1147 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1148 struct btrfs_fs_info *fs_info)
1150 struct btrfs_root *root;
1151 struct btrfs_root *tree_root = fs_info->tree_root;
1152 struct extent_buffer *leaf;
1154 root = kzalloc(sizeof(*root), GFP_NOFS);
1156 return ERR_PTR(-ENOMEM);
1158 __setup_root(tree_root->nodesize, tree_root->leafsize,
1159 tree_root->sectorsize, tree_root->stripesize,
1160 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1162 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1163 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1164 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1166 * log trees do not get reference counted because they go away
1167 * before a real commit is actually done. They do store pointers
1168 * to file data extents, and those reference counts still get
1169 * updated (along with back refs to the log tree).
1173 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1174 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1177 return ERR_CAST(leaf);
1180 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1181 btrfs_set_header_bytenr(leaf, leaf->start);
1182 btrfs_set_header_generation(leaf, trans->transid);
1183 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1184 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1187 write_extent_buffer(root->node, root->fs_info->fsid,
1188 (unsigned long)btrfs_header_fsid(root->node),
1190 btrfs_mark_buffer_dirty(root->node);
1191 btrfs_tree_unlock(root->node);
1195 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1196 struct btrfs_fs_info *fs_info)
1198 struct btrfs_root *log_root;
1200 log_root = alloc_log_tree(trans, fs_info);
1201 if (IS_ERR(log_root))
1202 return PTR_ERR(log_root);
1203 WARN_ON(fs_info->log_root_tree);
1204 fs_info->log_root_tree = log_root;
1208 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1209 struct btrfs_root *root)
1211 struct btrfs_root *log_root;
1212 struct btrfs_inode_item *inode_item;
1214 log_root = alloc_log_tree(trans, root->fs_info);
1215 if (IS_ERR(log_root))
1216 return PTR_ERR(log_root);
1218 log_root->last_trans = trans->transid;
1219 log_root->root_key.offset = root->root_key.objectid;
1221 inode_item = &log_root->root_item.inode;
1222 inode_item->generation = cpu_to_le64(1);
1223 inode_item->size = cpu_to_le64(3);
1224 inode_item->nlink = cpu_to_le32(1);
1225 inode_item->nbytes = cpu_to_le64(root->leafsize);
1226 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1228 btrfs_set_root_node(&log_root->root_item, log_root->node);
1230 WARN_ON(root->log_root);
1231 root->log_root = log_root;
1232 root->log_transid = 0;
1233 root->last_log_commit = 0;
1237 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1238 struct btrfs_key *location)
1240 struct btrfs_root *root;
1241 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1242 struct btrfs_path *path;
1243 struct extent_buffer *l;
1248 root = kzalloc(sizeof(*root), GFP_NOFS);
1250 return ERR_PTR(-ENOMEM);
1251 if (location->offset == (u64)-1) {
1252 ret = find_and_setup_root(tree_root, fs_info,
1253 location->objectid, root);
1256 return ERR_PTR(ret);
1261 __setup_root(tree_root->nodesize, tree_root->leafsize,
1262 tree_root->sectorsize, tree_root->stripesize,
1263 root, fs_info, location->objectid);
1265 path = btrfs_alloc_path();
1268 return ERR_PTR(-ENOMEM);
1270 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1273 read_extent_buffer(l, &root->root_item,
1274 btrfs_item_ptr_offset(l, path->slots[0]),
1275 sizeof(root->root_item));
1276 memcpy(&root->root_key, location, sizeof(*location));
1278 btrfs_free_path(path);
1283 return ERR_PTR(ret);
1286 generation = btrfs_root_generation(&root->root_item);
1287 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1288 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1289 blocksize, generation);
1290 root->commit_root = btrfs_root_node(root);
1291 BUG_ON(!root->node);
1293 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1295 btrfs_check_and_init_root_item(&root->root_item);
1301 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1302 struct btrfs_key *location)
1304 struct btrfs_root *root;
1307 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1308 return fs_info->tree_root;
1309 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1310 return fs_info->extent_root;
1311 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1312 return fs_info->chunk_root;
1313 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1314 return fs_info->dev_root;
1315 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1316 return fs_info->csum_root;
1318 spin_lock(&fs_info->fs_roots_radix_lock);
1319 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1320 (unsigned long)location->objectid);
1321 spin_unlock(&fs_info->fs_roots_radix_lock);
1325 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1329 root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1330 root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1332 if (!root->free_ino_pinned || !root->free_ino_ctl) {
1337 btrfs_init_free_ino_ctl(root);
1338 mutex_init(&root->fs_commit_mutex);
1339 spin_lock_init(&root->cache_lock);
1340 init_waitqueue_head(&root->cache_wait);
1342 ret = get_anon_bdev(&root->anon_dev);
1346 if (btrfs_root_refs(&root->root_item) == 0) {
1351 ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1355 root->orphan_item_inserted = 1;
1357 ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1361 spin_lock(&fs_info->fs_roots_radix_lock);
1362 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1363 (unsigned long)root->root_key.objectid,
1368 spin_unlock(&fs_info->fs_roots_radix_lock);
1369 radix_tree_preload_end();
1371 if (ret == -EEXIST) {
1378 ret = btrfs_find_dead_roots(fs_info->tree_root,
1379 root->root_key.objectid);
1384 return ERR_PTR(ret);
1387 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1389 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1391 struct btrfs_device *device;
1392 struct backing_dev_info *bdi;
1395 list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1398 bdi = blk_get_backing_dev_info(device->bdev);
1399 if (bdi && bdi_congested(bdi, bdi_bits)) {
1409 * If this fails, caller must call bdi_destroy() to get rid of the
1412 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1416 bdi->capabilities = BDI_CAP_MAP_COPY;
1417 err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1421 bdi->ra_pages = default_backing_dev_info.ra_pages;
1422 bdi->congested_fn = btrfs_congested_fn;
1423 bdi->congested_data = info;
1427 static int bio_ready_for_csum(struct bio *bio)
1433 struct extent_io_tree *io_tree = NULL;
1434 struct bio_vec *bvec;
1438 bio_for_each_segment(bvec, bio, i) {
1439 page = bvec->bv_page;
1440 if (page->private == EXTENT_PAGE_PRIVATE) {
1441 length += bvec->bv_len;
1444 if (!page->private) {
1445 length += bvec->bv_len;
1448 length = bvec->bv_len;
1449 buf_len = page->private >> 2;
1450 start = page_offset(page) + bvec->bv_offset;
1451 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1453 /* are we fully contained in this bio? */
1454 if (buf_len <= length)
1457 ret = extent_range_uptodate(io_tree, start + length,
1458 start + buf_len - 1);
1463 * called by the kthread helper functions to finally call the bio end_io
1464 * functions. This is where read checksum verification actually happens
1466 static void end_workqueue_fn(struct btrfs_work *work)
1469 struct end_io_wq *end_io_wq;
1470 struct btrfs_fs_info *fs_info;
1473 end_io_wq = container_of(work, struct end_io_wq, work);
1474 bio = end_io_wq->bio;
1475 fs_info = end_io_wq->info;
1477 /* metadata bio reads are special because the whole tree block must
1478 * be checksummed at once. This makes sure the entire block is in
1479 * ram and up to date before trying to verify things. For
1480 * blocksize <= pagesize, it is basically a noop
1482 if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1483 !bio_ready_for_csum(bio)) {
1484 btrfs_queue_worker(&fs_info->endio_meta_workers,
1488 error = end_io_wq->error;
1489 bio->bi_private = end_io_wq->private;
1490 bio->bi_end_io = end_io_wq->end_io;
1492 bio_endio(bio, error);
1495 static int cleaner_kthread(void *arg)
1497 struct btrfs_root *root = arg;
1500 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1502 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1503 mutex_trylock(&root->fs_info->cleaner_mutex)) {
1504 btrfs_run_delayed_iputs(root);
1505 btrfs_clean_old_snapshots(root);
1506 mutex_unlock(&root->fs_info->cleaner_mutex);
1507 btrfs_run_defrag_inodes(root->fs_info);
1510 if (freezing(current)) {
1513 set_current_state(TASK_INTERRUPTIBLE);
1514 if (!kthread_should_stop())
1516 __set_current_state(TASK_RUNNING);
1518 } while (!kthread_should_stop());
1522 static int transaction_kthread(void *arg)
1524 struct btrfs_root *root = arg;
1525 struct btrfs_trans_handle *trans;
1526 struct btrfs_transaction *cur;
1529 unsigned long delay;
1534 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1535 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1537 spin_lock(&root->fs_info->trans_lock);
1538 cur = root->fs_info->running_transaction;
1540 spin_unlock(&root->fs_info->trans_lock);
1544 now = get_seconds();
1545 if (!cur->blocked &&
1546 (now < cur->start_time || now - cur->start_time < 30)) {
1547 spin_unlock(&root->fs_info->trans_lock);
1551 transid = cur->transid;
1552 spin_unlock(&root->fs_info->trans_lock);
1554 trans = btrfs_join_transaction(root);
1555 BUG_ON(IS_ERR(trans));
1556 if (transid == trans->transid) {
1557 ret = btrfs_commit_transaction(trans, root);
1560 btrfs_end_transaction(trans, root);
1563 wake_up_process(root->fs_info->cleaner_kthread);
1564 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1566 if (freezing(current)) {
1569 set_current_state(TASK_INTERRUPTIBLE);
1570 if (!kthread_should_stop() &&
1571 !btrfs_transaction_blocked(root->fs_info))
1572 schedule_timeout(delay);
1573 __set_current_state(TASK_RUNNING);
1575 } while (!kthread_should_stop());
1579 struct btrfs_root *open_ctree(struct super_block *sb,
1580 struct btrfs_fs_devices *fs_devices,
1590 struct btrfs_key location;
1591 struct buffer_head *bh;
1592 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1594 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1596 struct btrfs_root *tree_root = btrfs_sb(sb);
1597 struct btrfs_fs_info *fs_info = NULL;
1598 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1600 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1602 struct btrfs_root *log_tree_root;
1607 struct btrfs_super_block *disk_super;
1609 if (!extent_root || !tree_root || !tree_root->fs_info ||
1610 !chunk_root || !dev_root || !csum_root) {
1614 fs_info = tree_root->fs_info;
1616 ret = init_srcu_struct(&fs_info->subvol_srcu);
1622 ret = setup_bdi(fs_info, &fs_info->bdi);
1628 fs_info->btree_inode = new_inode(sb);
1629 if (!fs_info->btree_inode) {
1634 mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1636 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1637 INIT_LIST_HEAD(&fs_info->trans_list);
1638 INIT_LIST_HEAD(&fs_info->dead_roots);
1639 INIT_LIST_HEAD(&fs_info->delayed_iputs);
1640 INIT_LIST_HEAD(&fs_info->hashers);
1641 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1642 INIT_LIST_HEAD(&fs_info->ordered_operations);
1643 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1644 spin_lock_init(&fs_info->delalloc_lock);
1645 spin_lock_init(&fs_info->trans_lock);
1646 spin_lock_init(&fs_info->ref_cache_lock);
1647 spin_lock_init(&fs_info->fs_roots_radix_lock);
1648 spin_lock_init(&fs_info->delayed_iput_lock);
1649 spin_lock_init(&fs_info->defrag_inodes_lock);
1650 spin_lock_init(&fs_info->free_chunk_lock);
1651 mutex_init(&fs_info->reloc_mutex);
1653 init_completion(&fs_info->kobj_unregister);
1654 fs_info->tree_root = tree_root;
1655 fs_info->extent_root = extent_root;
1656 fs_info->csum_root = csum_root;
1657 fs_info->chunk_root = chunk_root;
1658 fs_info->dev_root = dev_root;
1659 fs_info->fs_devices = fs_devices;
1660 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1661 INIT_LIST_HEAD(&fs_info->space_info);
1662 btrfs_mapping_init(&fs_info->mapping_tree);
1663 btrfs_init_block_rsv(&fs_info->global_block_rsv);
1664 btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1665 btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1666 btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1667 btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1668 atomic_set(&fs_info->nr_async_submits, 0);
1669 atomic_set(&fs_info->async_delalloc_pages, 0);
1670 atomic_set(&fs_info->async_submit_draining, 0);
1671 atomic_set(&fs_info->nr_async_bios, 0);
1672 atomic_set(&fs_info->defrag_running, 0);
1674 fs_info->max_inline = 8192 * 1024;
1675 fs_info->metadata_ratio = 0;
1676 fs_info->defrag_inodes = RB_ROOT;
1677 fs_info->trans_no_join = 0;
1678 fs_info->free_chunk_space = 0;
1680 fs_info->thread_pool_size = min_t(unsigned long,
1681 num_online_cpus() + 2, 8);
1683 INIT_LIST_HEAD(&fs_info->ordered_extents);
1684 spin_lock_init(&fs_info->ordered_extent_lock);
1685 fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1687 if (!fs_info->delayed_root) {
1691 btrfs_init_delayed_root(fs_info->delayed_root);
1693 mutex_init(&fs_info->scrub_lock);
1694 atomic_set(&fs_info->scrubs_running, 0);
1695 atomic_set(&fs_info->scrub_pause_req, 0);
1696 atomic_set(&fs_info->scrubs_paused, 0);
1697 atomic_set(&fs_info->scrub_cancel_req, 0);
1698 init_waitqueue_head(&fs_info->scrub_pause_wait);
1699 init_rwsem(&fs_info->scrub_super_lock);
1700 fs_info->scrub_workers_refcnt = 0;
1702 sb->s_blocksize = 4096;
1703 sb->s_blocksize_bits = blksize_bits(4096);
1704 sb->s_bdi = &fs_info->bdi;
1706 fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1707 fs_info->btree_inode->i_nlink = 1;
1709 * we set the i_size on the btree inode to the max possible int.
1710 * the real end of the address space is determined by all of
1711 * the devices in the system
1713 fs_info->btree_inode->i_size = OFFSET_MAX;
1714 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1715 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1717 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1718 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1719 fs_info->btree_inode->i_mapping);
1720 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1722 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1724 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1725 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1726 sizeof(struct btrfs_key));
1727 BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1728 insert_inode_hash(fs_info->btree_inode);
1730 spin_lock_init(&fs_info->block_group_cache_lock);
1731 fs_info->block_group_cache_tree = RB_ROOT;
1733 extent_io_tree_init(&fs_info->freed_extents[0],
1734 fs_info->btree_inode->i_mapping);
1735 extent_io_tree_init(&fs_info->freed_extents[1],
1736 fs_info->btree_inode->i_mapping);
1737 fs_info->pinned_extents = &fs_info->freed_extents[0];
1738 fs_info->do_barriers = 1;
1741 mutex_init(&fs_info->ordered_operations_mutex);
1742 mutex_init(&fs_info->tree_log_mutex);
1743 mutex_init(&fs_info->chunk_mutex);
1744 mutex_init(&fs_info->transaction_kthread_mutex);
1745 mutex_init(&fs_info->cleaner_mutex);
1746 mutex_init(&fs_info->volume_mutex);
1747 init_rwsem(&fs_info->extent_commit_sem);
1748 init_rwsem(&fs_info->cleanup_work_sem);
1749 init_rwsem(&fs_info->subvol_sem);
1751 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1752 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1754 init_waitqueue_head(&fs_info->transaction_throttle);
1755 init_waitqueue_head(&fs_info->transaction_wait);
1756 init_waitqueue_head(&fs_info->transaction_blocked_wait);
1757 init_waitqueue_head(&fs_info->async_submit_wait);
1759 __setup_root(4096, 4096, 4096, 4096, tree_root,
1760 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1762 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1768 memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
1769 memcpy(fs_info->super_for_commit, fs_info->super_copy,
1770 sizeof(*fs_info->super_for_commit));
1773 memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
1775 disk_super = fs_info->super_copy;
1776 if (!btrfs_super_root(disk_super))
1779 /* check FS state, whether FS is broken. */
1780 fs_info->fs_state |= btrfs_super_flags(disk_super);
1782 btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
1785 * In the long term, we'll store the compression type in the super
1786 * block, and it'll be used for per file compression control.
1788 fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
1790 ret = btrfs_parse_options(tree_root, options);
1796 features = btrfs_super_incompat_flags(disk_super) &
1797 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1799 printk(KERN_ERR "BTRFS: couldn't mount because of "
1800 "unsupported optional features (%Lx).\n",
1801 (unsigned long long)features);
1806 features = btrfs_super_incompat_flags(disk_super);
1807 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1808 if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
1809 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
1810 btrfs_set_super_incompat_flags(disk_super, features);
1812 features = btrfs_super_compat_ro_flags(disk_super) &
1813 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1814 if (!(sb->s_flags & MS_RDONLY) && features) {
1815 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1816 "unsupported option features (%Lx).\n",
1817 (unsigned long long)features);
1822 btrfs_init_workers(&fs_info->generic_worker,
1823 "genwork", 1, NULL);
1825 btrfs_init_workers(&fs_info->workers, "worker",
1826 fs_info->thread_pool_size,
1827 &fs_info->generic_worker);
1829 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1830 fs_info->thread_pool_size,
1831 &fs_info->generic_worker);
1833 btrfs_init_workers(&fs_info->submit_workers, "submit",
1834 min_t(u64, fs_devices->num_devices,
1835 fs_info->thread_pool_size),
1836 &fs_info->generic_worker);
1838 btrfs_init_workers(&fs_info->caching_workers, "cache",
1839 2, &fs_info->generic_worker);
1841 /* a higher idle thresh on the submit workers makes it much more
1842 * likely that bios will be send down in a sane order to the
1845 fs_info->submit_workers.idle_thresh = 64;
1847 fs_info->workers.idle_thresh = 16;
1848 fs_info->workers.ordered = 1;
1850 fs_info->delalloc_workers.idle_thresh = 2;
1851 fs_info->delalloc_workers.ordered = 1;
1853 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
1854 &fs_info->generic_worker);
1855 btrfs_init_workers(&fs_info->endio_workers, "endio",
1856 fs_info->thread_pool_size,
1857 &fs_info->generic_worker);
1858 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1859 fs_info->thread_pool_size,
1860 &fs_info->generic_worker);
1861 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1862 "endio-meta-write", fs_info->thread_pool_size,
1863 &fs_info->generic_worker);
1864 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1865 fs_info->thread_pool_size,
1866 &fs_info->generic_worker);
1867 btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
1868 1, &fs_info->generic_worker);
1869 btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
1870 fs_info->thread_pool_size,
1871 &fs_info->generic_worker);
1874 * endios are largely parallel and should have a very
1877 fs_info->endio_workers.idle_thresh = 4;
1878 fs_info->endio_meta_workers.idle_thresh = 4;
1880 fs_info->endio_write_workers.idle_thresh = 2;
1881 fs_info->endio_meta_write_workers.idle_thresh = 2;
1883 btrfs_start_workers(&fs_info->workers, 1);
1884 btrfs_start_workers(&fs_info->generic_worker, 1);
1885 btrfs_start_workers(&fs_info->submit_workers, 1);
1886 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1887 btrfs_start_workers(&fs_info->fixup_workers, 1);
1888 btrfs_start_workers(&fs_info->endio_workers, 1);
1889 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1890 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1891 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1892 btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
1893 btrfs_start_workers(&fs_info->delayed_workers, 1);
1894 btrfs_start_workers(&fs_info->caching_workers, 1);
1896 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1897 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1898 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1900 nodesize = btrfs_super_nodesize(disk_super);
1901 leafsize = btrfs_super_leafsize(disk_super);
1902 sectorsize = btrfs_super_sectorsize(disk_super);
1903 stripesize = btrfs_super_stripesize(disk_super);
1904 tree_root->nodesize = nodesize;
1905 tree_root->leafsize = leafsize;
1906 tree_root->sectorsize = sectorsize;
1907 tree_root->stripesize = stripesize;
1909 sb->s_blocksize = sectorsize;
1910 sb->s_blocksize_bits = blksize_bits(sectorsize);
1912 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1913 sizeof(disk_super->magic))) {
1914 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1915 goto fail_sb_buffer;
1918 mutex_lock(&fs_info->chunk_mutex);
1919 ret = btrfs_read_sys_array(tree_root);
1920 mutex_unlock(&fs_info->chunk_mutex);
1922 printk(KERN_WARNING "btrfs: failed to read the system "
1923 "array on %s\n", sb->s_id);
1924 goto fail_sb_buffer;
1927 blocksize = btrfs_level_size(tree_root,
1928 btrfs_super_chunk_root_level(disk_super));
1929 generation = btrfs_super_chunk_root_generation(disk_super);
1931 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1932 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1934 chunk_root->node = read_tree_block(chunk_root,
1935 btrfs_super_chunk_root(disk_super),
1936 blocksize, generation);
1937 BUG_ON(!chunk_root->node);
1938 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1939 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1941 goto fail_chunk_root;
1943 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1944 chunk_root->commit_root = btrfs_root_node(chunk_root);
1946 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1947 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1950 mutex_lock(&fs_info->chunk_mutex);
1951 ret = btrfs_read_chunk_tree(chunk_root);
1952 mutex_unlock(&fs_info->chunk_mutex);
1954 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1956 goto fail_chunk_root;
1959 btrfs_close_extra_devices(fs_devices);
1961 blocksize = btrfs_level_size(tree_root,
1962 btrfs_super_root_level(disk_super));
1963 generation = btrfs_super_generation(disk_super);
1965 tree_root->node = read_tree_block(tree_root,
1966 btrfs_super_root(disk_super),
1967 blocksize, generation);
1968 if (!tree_root->node)
1969 goto fail_chunk_root;
1970 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1971 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1973 goto fail_tree_root;
1975 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1976 tree_root->commit_root = btrfs_root_node(tree_root);
1978 ret = find_and_setup_root(tree_root, fs_info,
1979 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1981 goto fail_tree_root;
1982 extent_root->track_dirty = 1;
1984 ret = find_and_setup_root(tree_root, fs_info,
1985 BTRFS_DEV_TREE_OBJECTID, dev_root);
1987 goto fail_extent_root;
1988 dev_root->track_dirty = 1;
1990 ret = find_and_setup_root(tree_root, fs_info,
1991 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1995 csum_root->track_dirty = 1;
1997 fs_info->generation = generation;
1998 fs_info->last_trans_committed = generation;
1999 fs_info->data_alloc_profile = (u64)-1;
2000 fs_info->metadata_alloc_profile = (u64)-1;
2001 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2003 ret = btrfs_init_space_info(fs_info);
2005 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2006 goto fail_block_groups;
2009 ret = btrfs_read_block_groups(extent_root);
2011 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2012 goto fail_block_groups;
2015 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2017 if (IS_ERR(fs_info->cleaner_kthread))
2018 goto fail_block_groups;
2020 fs_info->transaction_kthread = kthread_run(transaction_kthread,
2022 "btrfs-transaction");
2023 if (IS_ERR(fs_info->transaction_kthread))
2026 if (!btrfs_test_opt(tree_root, SSD) &&
2027 !btrfs_test_opt(tree_root, NOSSD) &&
2028 !fs_info->fs_devices->rotating) {
2029 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2031 btrfs_set_opt(fs_info->mount_opt, SSD);
2034 /* do not make disk changes in broken FS */
2035 if (btrfs_super_log_root(disk_super) != 0 &&
2036 !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2037 u64 bytenr = btrfs_super_log_root(disk_super);
2039 if (fs_devices->rw_devices == 0) {
2040 printk(KERN_WARNING "Btrfs log replay required "
2043 goto fail_trans_kthread;
2046 btrfs_level_size(tree_root,
2047 btrfs_super_log_root_level(disk_super));
2049 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2050 if (!log_tree_root) {
2052 goto fail_trans_kthread;
2055 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2056 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2058 log_tree_root->node = read_tree_block(tree_root, bytenr,
2061 ret = btrfs_recover_log_trees(log_tree_root);
2064 if (sb->s_flags & MS_RDONLY) {
2065 ret = btrfs_commit_super(tree_root);
2070 ret = btrfs_find_orphan_roots(tree_root);
2073 if (!(sb->s_flags & MS_RDONLY)) {
2074 ret = btrfs_cleanup_fs_roots(fs_info);
2077 ret = btrfs_recover_relocation(tree_root);
2080 "btrfs: failed to recover relocation\n");
2082 goto fail_trans_kthread;
2086 location.objectid = BTRFS_FS_TREE_OBJECTID;
2087 location.type = BTRFS_ROOT_ITEM_KEY;
2088 location.offset = (u64)-1;
2090 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2091 if (!fs_info->fs_root)
2092 goto fail_trans_kthread;
2093 if (IS_ERR(fs_info->fs_root)) {
2094 err = PTR_ERR(fs_info->fs_root);
2095 goto fail_trans_kthread;
2098 if (!(sb->s_flags & MS_RDONLY)) {
2099 down_read(&fs_info->cleanup_work_sem);
2100 err = btrfs_orphan_cleanup(fs_info->fs_root);
2102 err = btrfs_orphan_cleanup(fs_info->tree_root);
2103 up_read(&fs_info->cleanup_work_sem);
2105 close_ctree(tree_root);
2106 return ERR_PTR(err);
2113 kthread_stop(fs_info->transaction_kthread);
2115 kthread_stop(fs_info->cleaner_kthread);
2118 * make sure we're done with the btree inode before we stop our
2121 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2122 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2125 btrfs_free_block_groups(fs_info);
2126 free_extent_buffer(csum_root->node);
2127 free_extent_buffer(csum_root->commit_root);
2129 free_extent_buffer(dev_root->node);
2130 free_extent_buffer(dev_root->commit_root);
2132 free_extent_buffer(extent_root->node);
2133 free_extent_buffer(extent_root->commit_root);
2135 free_extent_buffer(tree_root->node);
2136 free_extent_buffer(tree_root->commit_root);
2138 free_extent_buffer(chunk_root->node);
2139 free_extent_buffer(chunk_root->commit_root);
2141 btrfs_stop_workers(&fs_info->generic_worker);
2142 btrfs_stop_workers(&fs_info->fixup_workers);
2143 btrfs_stop_workers(&fs_info->delalloc_workers);
2144 btrfs_stop_workers(&fs_info->workers);
2145 btrfs_stop_workers(&fs_info->endio_workers);
2146 btrfs_stop_workers(&fs_info->endio_meta_workers);
2147 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2148 btrfs_stop_workers(&fs_info->endio_write_workers);
2149 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2150 btrfs_stop_workers(&fs_info->submit_workers);
2151 btrfs_stop_workers(&fs_info->delayed_workers);
2152 btrfs_stop_workers(&fs_info->caching_workers);
2155 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2156 iput(fs_info->btree_inode);
2158 btrfs_close_devices(fs_info->fs_devices);
2159 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2161 bdi_destroy(&fs_info->bdi);
2163 cleanup_srcu_struct(&fs_info->subvol_srcu);
2165 free_fs_info(fs_info);
2166 return ERR_PTR(err);
2169 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2171 char b[BDEVNAME_SIZE];
2174 set_buffer_uptodate(bh);
2176 printk_ratelimited(KERN_WARNING "lost page write due to "
2177 "I/O error on %s\n",
2178 bdevname(bh->b_bdev, b));
2179 /* note, we dont' set_buffer_write_io_error because we have
2180 * our own ways of dealing with the IO errors
2182 clear_buffer_uptodate(bh);
2188 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2190 struct buffer_head *bh;
2191 struct buffer_head *latest = NULL;
2192 struct btrfs_super_block *super;
2197 /* we would like to check all the supers, but that would make
2198 * a btrfs mount succeed after a mkfs from a different FS.
2199 * So, we need to add a special mount option to scan for
2200 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2202 for (i = 0; i < 1; i++) {
2203 bytenr = btrfs_sb_offset(i);
2204 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2206 bh = __bread(bdev, bytenr / 4096, 4096);
2210 super = (struct btrfs_super_block *)bh->b_data;
2211 if (btrfs_super_bytenr(super) != bytenr ||
2212 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2213 sizeof(super->magic))) {
2218 if (!latest || btrfs_super_generation(super) > transid) {
2221 transid = btrfs_super_generation(super);
2230 * this should be called twice, once with wait == 0 and
2231 * once with wait == 1. When wait == 0 is done, all the buffer heads
2232 * we write are pinned.
2234 * They are released when wait == 1 is done.
2235 * max_mirrors must be the same for both runs, and it indicates how
2236 * many supers on this one device should be written.
2238 * max_mirrors == 0 means to write them all.
2240 static int write_dev_supers(struct btrfs_device *device,
2241 struct btrfs_super_block *sb,
2242 int do_barriers, int wait, int max_mirrors)
2244 struct buffer_head *bh;
2250 int last_barrier = 0;
2252 if (max_mirrors == 0)
2253 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2255 /* make sure only the last submit_bh does a barrier */
2257 for (i = 0; i < max_mirrors; i++) {
2258 bytenr = btrfs_sb_offset(i);
2259 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2260 device->total_bytes)
2266 for (i = 0; i < max_mirrors; i++) {
2267 bytenr = btrfs_sb_offset(i);
2268 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2272 bh = __find_get_block(device->bdev, bytenr / 4096,
2273 BTRFS_SUPER_INFO_SIZE);
2276 if (!buffer_uptodate(bh))
2279 /* drop our reference */
2282 /* drop the reference from the wait == 0 run */
2286 btrfs_set_super_bytenr(sb, bytenr);
2289 crc = btrfs_csum_data(NULL, (char *)sb +
2290 BTRFS_CSUM_SIZE, crc,
2291 BTRFS_SUPER_INFO_SIZE -
2293 btrfs_csum_final(crc, sb->csum);
2296 * one reference for us, and we leave it for the
2299 bh = __getblk(device->bdev, bytenr / 4096,
2300 BTRFS_SUPER_INFO_SIZE);
2301 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2303 /* one reference for submit_bh */
2306 set_buffer_uptodate(bh);
2308 bh->b_end_io = btrfs_end_buffer_write_sync;
2311 if (i == last_barrier && do_barriers)
2312 ret = submit_bh(WRITE_FLUSH_FUA, bh);
2314 ret = submit_bh(WRITE_SYNC, bh);
2319 return errors < i ? 0 : -1;
2322 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2324 struct list_head *head;
2325 struct btrfs_device *dev;
2326 struct btrfs_super_block *sb;
2327 struct btrfs_dev_item *dev_item;
2331 int total_errors = 0;
2334 max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2335 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2337 sb = root->fs_info->super_for_commit;
2338 dev_item = &sb->dev_item;
2340 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2341 head = &root->fs_info->fs_devices->devices;
2342 list_for_each_entry_rcu(dev, head, dev_list) {
2347 if (!dev->in_fs_metadata || !dev->writeable)
2350 btrfs_set_stack_device_generation(dev_item, 0);
2351 btrfs_set_stack_device_type(dev_item, dev->type);
2352 btrfs_set_stack_device_id(dev_item, dev->devid);
2353 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2354 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2355 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2356 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2357 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2358 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2359 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2361 flags = btrfs_super_flags(sb);
2362 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2364 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2368 if (total_errors > max_errors) {
2369 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2375 list_for_each_entry_rcu(dev, head, dev_list) {
2378 if (!dev->in_fs_metadata || !dev->writeable)
2381 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2385 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2386 if (total_errors > max_errors) {
2387 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2394 int write_ctree_super(struct btrfs_trans_handle *trans,
2395 struct btrfs_root *root, int max_mirrors)
2399 ret = write_all_supers(root, max_mirrors);
2403 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2405 spin_lock(&fs_info->fs_roots_radix_lock);
2406 radix_tree_delete(&fs_info->fs_roots_radix,
2407 (unsigned long)root->root_key.objectid);
2408 spin_unlock(&fs_info->fs_roots_radix_lock);
2410 if (btrfs_root_refs(&root->root_item) == 0)
2411 synchronize_srcu(&fs_info->subvol_srcu);
2413 __btrfs_remove_free_space_cache(root->free_ino_pinned);
2414 __btrfs_remove_free_space_cache(root->free_ino_ctl);
2419 static void free_fs_root(struct btrfs_root *root)
2421 iput(root->cache_inode);
2422 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2424 free_anon_bdev(root->anon_dev);
2425 free_extent_buffer(root->node);
2426 free_extent_buffer(root->commit_root);
2427 kfree(root->free_ino_ctl);
2428 kfree(root->free_ino_pinned);
2433 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2436 struct btrfs_root *gang[8];
2439 while (!list_empty(&fs_info->dead_roots)) {
2440 gang[0] = list_entry(fs_info->dead_roots.next,
2441 struct btrfs_root, root_list);
2442 list_del(&gang[0]->root_list);
2444 if (gang[0]->in_radix) {
2445 btrfs_free_fs_root(fs_info, gang[0]);
2447 free_extent_buffer(gang[0]->node);
2448 free_extent_buffer(gang[0]->commit_root);
2454 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2459 for (i = 0; i < ret; i++)
2460 btrfs_free_fs_root(fs_info, gang[i]);
2465 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2467 u64 root_objectid = 0;
2468 struct btrfs_root *gang[8];
2473 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2474 (void **)gang, root_objectid,
2479 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2480 for (i = 0; i < ret; i++) {
2483 root_objectid = gang[i]->root_key.objectid;
2484 err = btrfs_orphan_cleanup(gang[i]);
2493 int btrfs_commit_super(struct btrfs_root *root)
2495 struct btrfs_trans_handle *trans;
2498 mutex_lock(&root->fs_info->cleaner_mutex);
2499 btrfs_run_delayed_iputs(root);
2500 btrfs_clean_old_snapshots(root);
2501 mutex_unlock(&root->fs_info->cleaner_mutex);
2503 /* wait until ongoing cleanup work done */
2504 down_write(&root->fs_info->cleanup_work_sem);
2505 up_write(&root->fs_info->cleanup_work_sem);
2507 trans = btrfs_join_transaction(root);
2509 return PTR_ERR(trans);
2510 ret = btrfs_commit_transaction(trans, root);
2512 /* run commit again to drop the original snapshot */
2513 trans = btrfs_join_transaction(root);
2515 return PTR_ERR(trans);
2516 btrfs_commit_transaction(trans, root);
2517 ret = btrfs_write_and_wait_transaction(NULL, root);
2520 ret = write_ctree_super(NULL, root, 0);
2524 int close_ctree(struct btrfs_root *root)
2526 struct btrfs_fs_info *fs_info = root->fs_info;
2529 fs_info->closing = 1;
2532 btrfs_scrub_cancel(root);
2534 /* wait for any defraggers to finish */
2535 wait_event(fs_info->transaction_wait,
2536 (atomic_read(&fs_info->defrag_running) == 0));
2538 /* clear out the rbtree of defraggable inodes */
2539 btrfs_run_defrag_inodes(root->fs_info);
2542 * Here come 2 situations when btrfs is broken to flip readonly:
2544 * 1. when btrfs flips readonly somewhere else before
2545 * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2546 * and btrfs will skip to write sb directly to keep
2547 * ERROR state on disk.
2549 * 2. when btrfs flips readonly just in btrfs_commit_super,
2550 * and in such case, btrfs cannot write sb via btrfs_commit_super,
2551 * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2552 * btrfs will cleanup all FS resources first and write sb then.
2554 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2555 ret = btrfs_commit_super(root);
2557 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2560 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2561 ret = btrfs_error_commit_super(root);
2563 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2566 btrfs_put_block_group_cache(fs_info);
2568 kthread_stop(root->fs_info->transaction_kthread);
2569 kthread_stop(root->fs_info->cleaner_kthread);
2571 fs_info->closing = 2;
2574 if (fs_info->delalloc_bytes) {
2575 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2576 (unsigned long long)fs_info->delalloc_bytes);
2578 if (fs_info->total_ref_cache_size) {
2579 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2580 (unsigned long long)fs_info->total_ref_cache_size);
2583 free_extent_buffer(fs_info->extent_root->node);
2584 free_extent_buffer(fs_info->extent_root->commit_root);
2585 free_extent_buffer(fs_info->tree_root->node);
2586 free_extent_buffer(fs_info->tree_root->commit_root);
2587 free_extent_buffer(root->fs_info->chunk_root->node);
2588 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2589 free_extent_buffer(root->fs_info->dev_root->node);
2590 free_extent_buffer(root->fs_info->dev_root->commit_root);
2591 free_extent_buffer(root->fs_info->csum_root->node);
2592 free_extent_buffer(root->fs_info->csum_root->commit_root);
2594 btrfs_free_block_groups(root->fs_info);
2596 del_fs_roots(fs_info);
2598 iput(fs_info->btree_inode);
2600 btrfs_stop_workers(&fs_info->generic_worker);
2601 btrfs_stop_workers(&fs_info->fixup_workers);
2602 btrfs_stop_workers(&fs_info->delalloc_workers);
2603 btrfs_stop_workers(&fs_info->workers);
2604 btrfs_stop_workers(&fs_info->endio_workers);
2605 btrfs_stop_workers(&fs_info->endio_meta_workers);
2606 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2607 btrfs_stop_workers(&fs_info->endio_write_workers);
2608 btrfs_stop_workers(&fs_info->endio_freespace_worker);
2609 btrfs_stop_workers(&fs_info->submit_workers);
2610 btrfs_stop_workers(&fs_info->delayed_workers);
2611 btrfs_stop_workers(&fs_info->caching_workers);
2613 btrfs_close_devices(fs_info->fs_devices);
2614 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2616 bdi_destroy(&fs_info->bdi);
2617 cleanup_srcu_struct(&fs_info->subvol_srcu);
2619 free_fs_info(fs_info);
2624 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2627 struct inode *btree_inode = buf->first_page->mapping->host;
2629 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2634 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2639 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2641 struct inode *btree_inode = buf->first_page->mapping->host;
2642 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2646 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2648 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2649 u64 transid = btrfs_header_generation(buf);
2650 struct inode *btree_inode = root->fs_info->btree_inode;
2653 btrfs_assert_tree_locked(buf);
2654 if (transid != root->fs_info->generation) {
2655 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2656 "found %llu running %llu\n",
2657 (unsigned long long)buf->start,
2658 (unsigned long long)transid,
2659 (unsigned long long)root->fs_info->generation);
2662 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2665 spin_lock(&root->fs_info->delalloc_lock);
2666 root->fs_info->dirty_metadata_bytes += buf->len;
2667 spin_unlock(&root->fs_info->delalloc_lock);
2671 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2674 * looks as though older kernels can get into trouble with
2675 * this code, they end up stuck in balance_dirty_pages forever
2678 unsigned long thresh = 32 * 1024 * 1024;
2680 if (current->flags & PF_MEMALLOC)
2683 btrfs_balance_delayed_items(root);
2685 num_dirty = root->fs_info->dirty_metadata_bytes;
2687 if (num_dirty > thresh) {
2688 balance_dirty_pages_ratelimited_nr(
2689 root->fs_info->btree_inode->i_mapping, 1);
2694 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2697 * looks as though older kernels can get into trouble with
2698 * this code, they end up stuck in balance_dirty_pages forever
2701 unsigned long thresh = 32 * 1024 * 1024;
2703 if (current->flags & PF_MEMALLOC)
2706 num_dirty = root->fs_info->dirty_metadata_bytes;
2708 if (num_dirty > thresh) {
2709 balance_dirty_pages_ratelimited_nr(
2710 root->fs_info->btree_inode->i_mapping, 1);
2715 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2717 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2719 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2721 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2725 static int btree_lock_page_hook(struct page *page, void *data,
2726 void (*flush_fn)(void *))
2728 struct inode *inode = page->mapping->host;
2729 struct btrfs_root *root = BTRFS_I(inode)->root;
2730 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2731 struct extent_buffer *eb;
2733 u64 bytenr = page_offset(page);
2735 if (page->private == EXTENT_PAGE_PRIVATE)
2738 len = page->private >> 2;
2739 eb = find_extent_buffer(io_tree, bytenr, len);
2743 if (!btrfs_try_tree_write_lock(eb)) {
2745 btrfs_tree_lock(eb);
2747 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2749 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2750 spin_lock(&root->fs_info->delalloc_lock);
2751 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2752 root->fs_info->dirty_metadata_bytes -= eb->len;
2755 spin_unlock(&root->fs_info->delalloc_lock);
2758 btrfs_tree_unlock(eb);
2759 free_extent_buffer(eb);
2761 if (!trylock_page(page)) {
2768 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
2774 if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
2775 printk(KERN_WARNING "warning: mount fs with errors, "
2776 "running btrfsck is recommended\n");
2779 int btrfs_error_commit_super(struct btrfs_root *root)
2783 mutex_lock(&root->fs_info->cleaner_mutex);
2784 btrfs_run_delayed_iputs(root);
2785 mutex_unlock(&root->fs_info->cleaner_mutex);
2787 down_write(&root->fs_info->cleanup_work_sem);
2788 up_write(&root->fs_info->cleanup_work_sem);
2790 /* cleanup FS via transaction */
2791 btrfs_cleanup_transaction(root);
2793 ret = write_ctree_super(NULL, root, 0);
2798 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
2800 struct btrfs_inode *btrfs_inode;
2801 struct list_head splice;
2803 INIT_LIST_HEAD(&splice);
2805 mutex_lock(&root->fs_info->ordered_operations_mutex);
2806 spin_lock(&root->fs_info->ordered_extent_lock);
2808 list_splice_init(&root->fs_info->ordered_operations, &splice);
2809 while (!list_empty(&splice)) {
2810 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2811 ordered_operations);
2813 list_del_init(&btrfs_inode->ordered_operations);
2815 btrfs_invalidate_inodes(btrfs_inode->root);
2818 spin_unlock(&root->fs_info->ordered_extent_lock);
2819 mutex_unlock(&root->fs_info->ordered_operations_mutex);
2824 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
2826 struct list_head splice;
2827 struct btrfs_ordered_extent *ordered;
2828 struct inode *inode;
2830 INIT_LIST_HEAD(&splice);
2832 spin_lock(&root->fs_info->ordered_extent_lock);
2834 list_splice_init(&root->fs_info->ordered_extents, &splice);
2835 while (!list_empty(&splice)) {
2836 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
2839 list_del_init(&ordered->root_extent_list);
2840 atomic_inc(&ordered->refs);
2842 /* the inode may be getting freed (in sys_unlink path). */
2843 inode = igrab(ordered->inode);
2845 spin_unlock(&root->fs_info->ordered_extent_lock);
2849 atomic_set(&ordered->refs, 1);
2850 btrfs_put_ordered_extent(ordered);
2852 spin_lock(&root->fs_info->ordered_extent_lock);
2855 spin_unlock(&root->fs_info->ordered_extent_lock);
2860 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
2861 struct btrfs_root *root)
2863 struct rb_node *node;
2864 struct btrfs_delayed_ref_root *delayed_refs;
2865 struct btrfs_delayed_ref_node *ref;
2868 delayed_refs = &trans->delayed_refs;
2870 spin_lock(&delayed_refs->lock);
2871 if (delayed_refs->num_entries == 0) {
2872 spin_unlock(&delayed_refs->lock);
2873 printk(KERN_INFO "delayed_refs has NO entry\n");
2877 node = rb_first(&delayed_refs->root);
2879 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
2880 node = rb_next(node);
2883 rb_erase(&ref->rb_node, &delayed_refs->root);
2884 delayed_refs->num_entries--;
2886 atomic_set(&ref->refs, 1);
2887 if (btrfs_delayed_ref_is_head(ref)) {
2888 struct btrfs_delayed_ref_head *head;
2890 head = btrfs_delayed_node_to_head(ref);
2891 mutex_lock(&head->mutex);
2892 kfree(head->extent_op);
2893 delayed_refs->num_heads--;
2894 if (list_empty(&head->cluster))
2895 delayed_refs->num_heads_ready--;
2896 list_del_init(&head->cluster);
2897 mutex_unlock(&head->mutex);
2900 spin_unlock(&delayed_refs->lock);
2901 btrfs_put_delayed_ref(ref);
2904 spin_lock(&delayed_refs->lock);
2907 spin_unlock(&delayed_refs->lock);
2912 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
2914 struct btrfs_pending_snapshot *snapshot;
2915 struct list_head splice;
2917 INIT_LIST_HEAD(&splice);
2919 list_splice_init(&t->pending_snapshots, &splice);
2921 while (!list_empty(&splice)) {
2922 snapshot = list_entry(splice.next,
2923 struct btrfs_pending_snapshot,
2926 list_del_init(&snapshot->list);
2934 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
2936 struct btrfs_inode *btrfs_inode;
2937 struct list_head splice;
2939 INIT_LIST_HEAD(&splice);
2941 spin_lock(&root->fs_info->delalloc_lock);
2942 list_splice_init(&root->fs_info->delalloc_inodes, &splice);
2944 while (!list_empty(&splice)) {
2945 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
2948 list_del_init(&btrfs_inode->delalloc_inodes);
2950 btrfs_invalidate_inodes(btrfs_inode->root);
2953 spin_unlock(&root->fs_info->delalloc_lock);
2958 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
2959 struct extent_io_tree *dirty_pages,
2964 struct inode *btree_inode = root->fs_info->btree_inode;
2965 struct extent_buffer *eb;
2969 unsigned long index;
2972 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
2977 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
2978 while (start <= end) {
2979 index = start >> PAGE_CACHE_SHIFT;
2980 start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
2981 page = find_get_page(btree_inode->i_mapping, index);
2984 offset = page_offset(page);
2986 spin_lock(&dirty_pages->buffer_lock);
2987 eb = radix_tree_lookup(
2988 &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
2989 offset >> PAGE_CACHE_SHIFT);
2990 spin_unlock(&dirty_pages->buffer_lock);
2992 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
2994 atomic_set(&eb->refs, 1);
2996 if (PageWriteback(page))
2997 end_page_writeback(page);
3000 if (PageDirty(page)) {
3001 clear_page_dirty_for_io(page);
3002 spin_lock_irq(&page->mapping->tree_lock);
3003 radix_tree_tag_clear(&page->mapping->page_tree,
3005 PAGECACHE_TAG_DIRTY);
3006 spin_unlock_irq(&page->mapping->tree_lock);
3009 page->mapping->a_ops->invalidatepage(page, 0);
3017 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3018 struct extent_io_tree *pinned_extents)
3020 struct extent_io_tree *unpin;
3025 unpin = pinned_extents;
3027 ret = find_first_extent_bit(unpin, 0, &start, &end,
3033 if (btrfs_test_opt(root, DISCARD))
3034 ret = btrfs_error_discard_extent(root, start,
3038 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3039 btrfs_error_unpin_extent_range(root, start, end);
3046 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3048 struct btrfs_transaction *t;
3053 mutex_lock(&root->fs_info->transaction_kthread_mutex);
3055 spin_lock(&root->fs_info->trans_lock);
3056 list_splice_init(&root->fs_info->trans_list, &list);
3057 root->fs_info->trans_no_join = 1;
3058 spin_unlock(&root->fs_info->trans_lock);
3060 while (!list_empty(&list)) {
3061 t = list_entry(list.next, struct btrfs_transaction, list);
3065 btrfs_destroy_ordered_operations(root);
3067 btrfs_destroy_ordered_extents(root);
3069 btrfs_destroy_delayed_refs(t, root);
3071 btrfs_block_rsv_release(root,
3072 &root->fs_info->trans_block_rsv,
3073 t->dirty_pages.dirty_bytes);
3075 /* FIXME: cleanup wait for commit */
3078 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3079 wake_up(&root->fs_info->transaction_blocked_wait);
3082 if (waitqueue_active(&root->fs_info->transaction_wait))
3083 wake_up(&root->fs_info->transaction_wait);
3086 if (waitqueue_active(&t->commit_wait))
3087 wake_up(&t->commit_wait);
3089 btrfs_destroy_pending_snapshots(t);
3091 btrfs_destroy_delalloc_inodes(root);
3093 spin_lock(&root->fs_info->trans_lock);
3094 root->fs_info->running_transaction = NULL;
3095 spin_unlock(&root->fs_info->trans_lock);
3097 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3100 btrfs_destroy_pinned_extent(root,
3101 root->fs_info->pinned_extents);
3103 atomic_set(&t->use_count, 0);
3104 list_del_init(&t->list);
3105 memset(t, 0, sizeof(*t));
3106 kmem_cache_free(btrfs_transaction_cachep, t);
3109 spin_lock(&root->fs_info->trans_lock);
3110 root->fs_info->trans_no_join = 0;
3111 spin_unlock(&root->fs_info->trans_lock);
3112 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3117 static struct extent_io_ops btree_extent_io_ops = {
3118 .write_cache_pages_lock_hook = btree_lock_page_hook,
3119 .readpage_end_io_hook = btree_readpage_end_io_hook,
3120 .submit_bio_hook = btree_submit_bio_hook,
3121 /* note we're sharing with inode.c for the merge bio hook */
3122 .merge_bio_hook = btrfs_merge_bio_hook,