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>
33 #include "transaction.h"
34 #include "btrfs_inode.h"
36 #include "print-tree.h"
37 #include "async-thread.h"
39 #include "ref-cache.h"
41 #include "free-space-cache.h"
43 static struct extent_io_ops btree_extent_io_ops;
44 static void end_workqueue_fn(struct btrfs_work *work);
47 * end_io_wq structs are used to do processing in task context when an IO is
48 * complete. This is used during reads to verify checksums, and it is used
49 * by writes to insert metadata for new file extents after IO is complete.
55 struct btrfs_fs_info *info;
58 struct list_head list;
59 struct btrfs_work work;
63 * async submit bios are used to offload expensive checksumming
64 * onto the worker threads. They checksum file and metadata bios
65 * just before they are sent down the IO stack.
67 struct async_submit_bio {
70 struct list_head list;
71 extent_submit_bio_hook_t *submit_bio_start;
72 extent_submit_bio_hook_t *submit_bio_done;
75 unsigned long bio_flags;
76 struct btrfs_work work;
79 /* These are used to set the lockdep class on the extent buffer locks.
80 * The class is set by the readpage_end_io_hook after the buffer has
81 * passed csum validation but before the pages are unlocked.
83 * The lockdep class is also set by btrfs_init_new_buffer on freshly
86 * The class is based on the level in the tree block, which allows lockdep
87 * to know that lower nodes nest inside the locks of higher nodes.
89 * We also add a check to make sure the highest level of the tree is
90 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
91 * code needs update as well.
93 #ifdef CONFIG_DEBUG_LOCK_ALLOC
94 # if BTRFS_MAX_LEVEL != 8
97 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
98 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
108 /* highest possible level */
114 * extents on the btree inode are pretty simple, there's one extent
115 * that covers the entire device
117 static struct extent_map *btree_get_extent(struct inode *inode,
118 struct page *page, size_t page_offset, u64 start, u64 len,
121 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
122 struct extent_map *em;
125 spin_lock(&em_tree->lock);
126 em = lookup_extent_mapping(em_tree, start, len);
129 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
130 spin_unlock(&em_tree->lock);
133 spin_unlock(&em_tree->lock);
135 em = alloc_extent_map(GFP_NOFS);
137 em = ERR_PTR(-ENOMEM);
142 em->block_len = (u64)-1;
144 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
146 spin_lock(&em_tree->lock);
147 ret = add_extent_mapping(em_tree, em);
148 if (ret == -EEXIST) {
149 u64 failed_start = em->start;
150 u64 failed_len = em->len;
153 em = lookup_extent_mapping(em_tree, start, len);
157 em = lookup_extent_mapping(em_tree, failed_start,
165 spin_unlock(&em_tree->lock);
173 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
175 return btrfs_crc32c(seed, data, len);
178 void btrfs_csum_final(u32 crc, char *result)
180 *(__le32 *)result = ~cpu_to_le32(crc);
184 * compute the csum for a btree block, and either verify it or write it
185 * into the csum field of the block.
187 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
191 btrfs_super_csum_size(&root->fs_info->super_copy);
194 unsigned long cur_len;
195 unsigned long offset = BTRFS_CSUM_SIZE;
196 char *map_token = NULL;
198 unsigned long map_start;
199 unsigned long map_len;
202 unsigned long inline_result;
204 len = buf->len - offset;
206 err = map_private_extent_buffer(buf, offset, 32,
208 &map_start, &map_len, KM_USER0);
211 cur_len = min(len, map_len - (offset - map_start));
212 crc = btrfs_csum_data(root, kaddr + offset - map_start,
216 unmap_extent_buffer(buf, map_token, KM_USER0);
218 if (csum_size > sizeof(inline_result)) {
219 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
223 result = (char *)&inline_result;
226 btrfs_csum_final(crc, result);
229 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
232 memcpy(&found, result, csum_size);
234 read_extent_buffer(buf, &val, 0, csum_size);
235 printk(KERN_INFO "btrfs: %s checksum verify failed "
236 "on %llu wanted %X found %X level %d\n",
237 root->fs_info->sb->s_id,
238 buf->start, val, found, btrfs_header_level(buf));
239 if (result != (char *)&inline_result)
244 write_extent_buffer(buf, result, 0, csum_size);
246 if (result != (char *)&inline_result)
252 * we can't consider a given block up to date unless the transid of the
253 * block matches the transid in the parent node's pointer. This is how we
254 * detect blocks that either didn't get written at all or got written
255 * in the wrong place.
257 static int verify_parent_transid(struct extent_io_tree *io_tree,
258 struct extent_buffer *eb, u64 parent_transid)
262 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
265 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
266 if (extent_buffer_uptodate(io_tree, eb) &&
267 btrfs_header_generation(eb) == parent_transid) {
271 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
272 (unsigned long long)eb->start,
273 (unsigned long long)parent_transid,
274 (unsigned long long)btrfs_header_generation(eb));
276 clear_extent_buffer_uptodate(io_tree, eb);
278 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
284 * helper to read a given tree block, doing retries as required when
285 * the checksums don't match and we have alternate mirrors to try.
287 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
288 struct extent_buffer *eb,
289 u64 start, u64 parent_transid)
291 struct extent_io_tree *io_tree;
296 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
298 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
299 btree_get_extent, mirror_num);
301 !verify_parent_transid(io_tree, eb, parent_transid))
304 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
310 if (mirror_num > num_copies)
317 * checksum a dirty tree block before IO. This has extra checks to make sure
318 * we only fill in the checksum field in the first page of a multi-page block
321 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
323 struct extent_io_tree *tree;
324 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
328 struct extent_buffer *eb;
331 tree = &BTRFS_I(page->mapping->host)->io_tree;
333 if (page->private == EXTENT_PAGE_PRIVATE)
337 len = page->private >> 2;
340 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
341 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
342 btrfs_header_generation(eb));
344 found_start = btrfs_header_bytenr(eb);
345 if (found_start != start) {
349 if (eb->first_page != page) {
353 if (!PageUptodate(page)) {
357 found_level = btrfs_header_level(eb);
359 csum_tree_block(root, eb, 0);
361 free_extent_buffer(eb);
366 static int check_tree_block_fsid(struct btrfs_root *root,
367 struct extent_buffer *eb)
369 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
370 u8 fsid[BTRFS_UUID_SIZE];
373 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
376 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
380 fs_devices = fs_devices->seed;
385 #ifdef CONFIG_DEBUG_LOCK_ALLOC
386 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
388 lockdep_set_class_and_name(&eb->lock,
389 &btrfs_eb_class[level],
390 btrfs_eb_name[level]);
394 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
395 struct extent_state *state)
397 struct extent_io_tree *tree;
401 struct extent_buffer *eb;
402 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
405 tree = &BTRFS_I(page->mapping->host)->io_tree;
406 if (page->private == EXTENT_PAGE_PRIVATE)
411 len = page->private >> 2;
414 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
416 found_start = btrfs_header_bytenr(eb);
417 if (found_start != start) {
418 printk(KERN_INFO "btrfs bad tree block start %llu %llu\n",
419 (unsigned long long)found_start,
420 (unsigned long long)eb->start);
424 if (eb->first_page != page) {
425 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
426 eb->first_page->index, page->index);
431 if (check_tree_block_fsid(root, eb)) {
432 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
433 (unsigned long long)eb->start);
437 found_level = btrfs_header_level(eb);
439 btrfs_set_buffer_lockdep_class(eb, found_level);
441 ret = csum_tree_block(root, eb, 1);
445 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
446 end = eb->start + end - 1;
448 free_extent_buffer(eb);
453 static void end_workqueue_bio(struct bio *bio, int err)
455 struct end_io_wq *end_io_wq = bio->bi_private;
456 struct btrfs_fs_info *fs_info;
458 fs_info = end_io_wq->info;
459 end_io_wq->error = err;
460 end_io_wq->work.func = end_workqueue_fn;
461 end_io_wq->work.flags = 0;
463 if (bio->bi_rw & (1 << BIO_RW)) {
464 if (end_io_wq->metadata)
465 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
468 btrfs_queue_worker(&fs_info->endio_write_workers,
471 if (end_io_wq->metadata)
472 btrfs_queue_worker(&fs_info->endio_meta_workers,
475 btrfs_queue_worker(&fs_info->endio_workers,
480 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
483 struct end_io_wq *end_io_wq;
484 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
488 end_io_wq->private = bio->bi_private;
489 end_io_wq->end_io = bio->bi_end_io;
490 end_io_wq->info = info;
491 end_io_wq->error = 0;
492 end_io_wq->bio = bio;
493 end_io_wq->metadata = metadata;
495 bio->bi_private = end_io_wq;
496 bio->bi_end_io = end_workqueue_bio;
500 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
502 unsigned long limit = min_t(unsigned long,
503 info->workers.max_workers,
504 info->fs_devices->open_devices);
508 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
510 return atomic_read(&info->nr_async_bios) >
511 btrfs_async_submit_limit(info);
514 static void run_one_async_start(struct btrfs_work *work)
516 struct btrfs_fs_info *fs_info;
517 struct async_submit_bio *async;
519 async = container_of(work, struct async_submit_bio, work);
520 fs_info = BTRFS_I(async->inode)->root->fs_info;
521 async->submit_bio_start(async->inode, async->rw, async->bio,
522 async->mirror_num, async->bio_flags);
525 static void run_one_async_done(struct btrfs_work *work)
527 struct btrfs_fs_info *fs_info;
528 struct async_submit_bio *async;
531 async = container_of(work, struct async_submit_bio, work);
532 fs_info = BTRFS_I(async->inode)->root->fs_info;
534 limit = btrfs_async_submit_limit(fs_info);
535 limit = limit * 2 / 3;
537 atomic_dec(&fs_info->nr_async_submits);
539 if (atomic_read(&fs_info->nr_async_submits) < limit &&
540 waitqueue_active(&fs_info->async_submit_wait))
541 wake_up(&fs_info->async_submit_wait);
543 async->submit_bio_done(async->inode, async->rw, async->bio,
544 async->mirror_num, async->bio_flags);
547 static void run_one_async_free(struct btrfs_work *work)
549 struct async_submit_bio *async;
551 async = container_of(work, struct async_submit_bio, work);
555 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
556 int rw, struct bio *bio, int mirror_num,
557 unsigned long bio_flags,
558 extent_submit_bio_hook_t *submit_bio_start,
559 extent_submit_bio_hook_t *submit_bio_done)
561 struct async_submit_bio *async;
563 async = kmalloc(sizeof(*async), GFP_NOFS);
567 async->inode = inode;
570 async->mirror_num = mirror_num;
571 async->submit_bio_start = submit_bio_start;
572 async->submit_bio_done = submit_bio_done;
574 async->work.func = run_one_async_start;
575 async->work.ordered_func = run_one_async_done;
576 async->work.ordered_free = run_one_async_free;
578 async->work.flags = 0;
579 async->bio_flags = bio_flags;
581 atomic_inc(&fs_info->nr_async_submits);
583 if (rw & (1 << BIO_RW_SYNCIO))
584 btrfs_set_work_high_prio(&async->work);
586 btrfs_queue_worker(&fs_info->workers, &async->work);
588 while (atomic_read(&fs_info->async_submit_draining) &&
589 atomic_read(&fs_info->nr_async_submits)) {
590 wait_event(fs_info->async_submit_wait,
591 (atomic_read(&fs_info->nr_async_submits) == 0));
597 static int btree_csum_one_bio(struct bio *bio)
599 struct bio_vec *bvec = bio->bi_io_vec;
601 struct btrfs_root *root;
603 WARN_ON(bio->bi_vcnt <= 0);
604 while (bio_index < bio->bi_vcnt) {
605 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
606 csum_dirty_buffer(root, bvec->bv_page);
613 static int __btree_submit_bio_start(struct inode *inode, int rw,
614 struct bio *bio, int mirror_num,
615 unsigned long bio_flags)
618 * when we're called for a write, we're already in the async
619 * submission context. Just jump into btrfs_map_bio
621 btree_csum_one_bio(bio);
625 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
626 int mirror_num, unsigned long bio_flags)
629 * when we're called for a write, we're already in the async
630 * submission context. Just jump into btrfs_map_bio
632 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
635 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
636 int mirror_num, unsigned long bio_flags)
640 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
644 if (!(rw & (1 << BIO_RW))) {
646 * called for a read, do the setup so that checksum validation
647 * can happen in the async kernel threads
649 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
654 * kthread helpers are used to submit writes so that checksumming
655 * can happen in parallel across all CPUs
657 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
658 inode, rw, bio, mirror_num, 0,
659 __btree_submit_bio_start,
660 __btree_submit_bio_done);
663 static int btree_writepage(struct page *page, struct writeback_control *wbc)
665 struct extent_io_tree *tree;
666 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
667 struct extent_buffer *eb;
670 tree = &BTRFS_I(page->mapping->host)->io_tree;
671 if (!(current->flags & PF_MEMALLOC)) {
672 return extent_write_full_page(tree, page,
673 btree_get_extent, wbc);
676 redirty_page_for_writepage(wbc, page);
677 eb = btrfs_find_tree_block(root, page_offset(page),
681 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
683 spin_lock(&root->fs_info->delalloc_lock);
684 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
685 spin_unlock(&root->fs_info->delalloc_lock);
687 free_extent_buffer(eb);
693 static int btree_writepages(struct address_space *mapping,
694 struct writeback_control *wbc)
696 struct extent_io_tree *tree;
697 tree = &BTRFS_I(mapping->host)->io_tree;
698 if (wbc->sync_mode == WB_SYNC_NONE) {
699 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
701 unsigned long thresh = 32 * 1024 * 1024;
703 if (wbc->for_kupdate)
706 /* this is a bit racy, but that's ok */
707 num_dirty = root->fs_info->dirty_metadata_bytes;
708 if (num_dirty < thresh)
711 return extent_writepages(tree, mapping, btree_get_extent, wbc);
714 static int btree_readpage(struct file *file, struct page *page)
716 struct extent_io_tree *tree;
717 tree = &BTRFS_I(page->mapping->host)->io_tree;
718 return extent_read_full_page(tree, page, btree_get_extent);
721 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
723 struct extent_io_tree *tree;
724 struct extent_map_tree *map;
727 if (PageWriteback(page) || PageDirty(page))
730 tree = &BTRFS_I(page->mapping->host)->io_tree;
731 map = &BTRFS_I(page->mapping->host)->extent_tree;
733 ret = try_release_extent_state(map, tree, page, gfp_flags);
737 ret = try_release_extent_buffer(tree, page);
739 ClearPagePrivate(page);
740 set_page_private(page, 0);
741 page_cache_release(page);
747 static void btree_invalidatepage(struct page *page, unsigned long offset)
749 struct extent_io_tree *tree;
750 tree = &BTRFS_I(page->mapping->host)->io_tree;
751 extent_invalidatepage(tree, page, offset);
752 btree_releasepage(page, GFP_NOFS);
753 if (PagePrivate(page)) {
754 printk(KERN_WARNING "btrfs warning page private not zero "
755 "on page %llu\n", (unsigned long long)page_offset(page));
756 ClearPagePrivate(page);
757 set_page_private(page, 0);
758 page_cache_release(page);
762 static struct address_space_operations btree_aops = {
763 .readpage = btree_readpage,
764 .writepage = btree_writepage,
765 .writepages = btree_writepages,
766 .releasepage = btree_releasepage,
767 .invalidatepage = btree_invalidatepage,
768 .sync_page = block_sync_page,
771 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
774 struct extent_buffer *buf = NULL;
775 struct inode *btree_inode = root->fs_info->btree_inode;
778 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
781 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
782 buf, 0, 0, btree_get_extent, 0);
783 free_extent_buffer(buf);
787 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
788 u64 bytenr, u32 blocksize)
790 struct inode *btree_inode = root->fs_info->btree_inode;
791 struct extent_buffer *eb;
792 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
793 bytenr, blocksize, GFP_NOFS);
797 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
798 u64 bytenr, u32 blocksize)
800 struct inode *btree_inode = root->fs_info->btree_inode;
801 struct extent_buffer *eb;
803 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
804 bytenr, blocksize, NULL, GFP_NOFS);
809 int btrfs_write_tree_block(struct extent_buffer *buf)
811 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
812 buf->start + buf->len - 1, WB_SYNC_ALL);
815 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
817 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
818 buf->start, buf->start + buf->len - 1);
821 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
822 u32 blocksize, u64 parent_transid)
824 struct extent_buffer *buf = NULL;
825 struct inode *btree_inode = root->fs_info->btree_inode;
826 struct extent_io_tree *io_tree;
829 io_tree = &BTRFS_I(btree_inode)->io_tree;
831 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
835 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
838 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
845 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
846 struct extent_buffer *buf)
848 struct inode *btree_inode = root->fs_info->btree_inode;
849 if (btrfs_header_generation(buf) ==
850 root->fs_info->running_transaction->transid) {
851 btrfs_assert_tree_locked(buf);
853 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
854 spin_lock(&root->fs_info->delalloc_lock);
855 if (root->fs_info->dirty_metadata_bytes >= buf->len)
856 root->fs_info->dirty_metadata_bytes -= buf->len;
859 spin_unlock(&root->fs_info->delalloc_lock);
862 /* ugh, clear_extent_buffer_dirty needs to lock the page */
863 btrfs_set_lock_blocking(buf);
864 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
870 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
871 u32 stripesize, struct btrfs_root *root,
872 struct btrfs_fs_info *fs_info,
876 root->commit_root = NULL;
877 root->ref_tree = NULL;
878 root->sectorsize = sectorsize;
879 root->nodesize = nodesize;
880 root->leafsize = leafsize;
881 root->stripesize = stripesize;
883 root->track_dirty = 0;
885 root->fs_info = fs_info;
886 root->objectid = objectid;
887 root->last_trans = 0;
888 root->highest_inode = 0;
889 root->last_inode_alloc = 0;
893 INIT_LIST_HEAD(&root->dirty_list);
894 INIT_LIST_HEAD(&root->orphan_list);
895 INIT_LIST_HEAD(&root->dead_list);
896 spin_lock_init(&root->node_lock);
897 spin_lock_init(&root->list_lock);
898 mutex_init(&root->objectid_mutex);
899 mutex_init(&root->log_mutex);
900 init_waitqueue_head(&root->log_writer_wait);
901 init_waitqueue_head(&root->log_commit_wait[0]);
902 init_waitqueue_head(&root->log_commit_wait[1]);
903 atomic_set(&root->log_commit[0], 0);
904 atomic_set(&root->log_commit[1], 0);
905 atomic_set(&root->log_writers, 0);
907 root->log_transid = 0;
908 extent_io_tree_init(&root->dirty_log_pages,
909 fs_info->btree_inode->i_mapping, GFP_NOFS);
911 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
912 root->ref_tree = &root->ref_tree_struct;
914 memset(&root->root_key, 0, sizeof(root->root_key));
915 memset(&root->root_item, 0, sizeof(root->root_item));
916 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
917 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
918 root->defrag_trans_start = fs_info->generation;
919 init_completion(&root->kobj_unregister);
920 root->defrag_running = 0;
921 root->defrag_level = 0;
922 root->root_key.objectid = objectid;
923 root->anon_super.s_root = NULL;
924 root->anon_super.s_dev = 0;
925 INIT_LIST_HEAD(&root->anon_super.s_list);
926 INIT_LIST_HEAD(&root->anon_super.s_instances);
927 init_rwsem(&root->anon_super.s_umount);
932 static int find_and_setup_root(struct btrfs_root *tree_root,
933 struct btrfs_fs_info *fs_info,
935 struct btrfs_root *root)
941 __setup_root(tree_root->nodesize, tree_root->leafsize,
942 tree_root->sectorsize, tree_root->stripesize,
943 root, fs_info, objectid);
944 ret = btrfs_find_last_root(tree_root, objectid,
945 &root->root_item, &root->root_key);
948 generation = btrfs_root_generation(&root->root_item);
949 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
950 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
951 blocksize, generation);
956 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
957 struct btrfs_fs_info *fs_info)
959 struct extent_buffer *eb;
960 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
969 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
970 0, &start, &end, EXTENT_DIRTY);
974 clear_extent_dirty(&log_root_tree->dirty_log_pages,
975 start, end, GFP_NOFS);
977 eb = fs_info->log_root_tree->node;
979 WARN_ON(btrfs_header_level(eb) != 0);
980 WARN_ON(btrfs_header_nritems(eb) != 0);
982 ret = btrfs_free_reserved_extent(fs_info->tree_root,
986 free_extent_buffer(eb);
987 kfree(fs_info->log_root_tree);
988 fs_info->log_root_tree = NULL;
992 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
993 struct btrfs_fs_info *fs_info)
995 struct btrfs_root *root;
996 struct btrfs_root *tree_root = fs_info->tree_root;
997 struct extent_buffer *leaf;
999 root = kzalloc(sizeof(*root), GFP_NOFS);
1001 return ERR_PTR(-ENOMEM);
1003 __setup_root(tree_root->nodesize, tree_root->leafsize,
1004 tree_root->sectorsize, tree_root->stripesize,
1005 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1007 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1008 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1009 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1011 * log trees do not get reference counted because they go away
1012 * before a real commit is actually done. They do store pointers
1013 * to file data extents, and those reference counts still get
1014 * updated (along with back refs to the log tree).
1018 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1019 0, BTRFS_TREE_LOG_OBJECTID,
1020 trans->transid, 0, 0, 0);
1023 return ERR_CAST(leaf);
1027 btrfs_set_header_nritems(root->node, 0);
1028 btrfs_set_header_level(root->node, 0);
1029 btrfs_set_header_bytenr(root->node, root->node->start);
1030 btrfs_set_header_generation(root->node, trans->transid);
1031 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
1033 write_extent_buffer(root->node, root->fs_info->fsid,
1034 (unsigned long)btrfs_header_fsid(root->node),
1036 btrfs_mark_buffer_dirty(root->node);
1037 btrfs_tree_unlock(root->node);
1041 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1042 struct btrfs_fs_info *fs_info)
1044 struct btrfs_root *log_root;
1046 log_root = alloc_log_tree(trans, fs_info);
1047 if (IS_ERR(log_root))
1048 return PTR_ERR(log_root);
1049 WARN_ON(fs_info->log_root_tree);
1050 fs_info->log_root_tree = log_root;
1054 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1055 struct btrfs_root *root)
1057 struct btrfs_root *log_root;
1058 struct btrfs_inode_item *inode_item;
1060 log_root = alloc_log_tree(trans, root->fs_info);
1061 if (IS_ERR(log_root))
1062 return PTR_ERR(log_root);
1064 log_root->last_trans = trans->transid;
1065 log_root->root_key.offset = root->root_key.objectid;
1067 inode_item = &log_root->root_item.inode;
1068 inode_item->generation = cpu_to_le64(1);
1069 inode_item->size = cpu_to_le64(3);
1070 inode_item->nlink = cpu_to_le32(1);
1071 inode_item->nbytes = cpu_to_le64(root->leafsize);
1072 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1074 btrfs_set_root_bytenr(&log_root->root_item, log_root->node->start);
1075 btrfs_set_root_generation(&log_root->root_item, trans->transid);
1077 WARN_ON(root->log_root);
1078 root->log_root = log_root;
1079 root->log_transid = 0;
1083 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1084 struct btrfs_key *location)
1086 struct btrfs_root *root;
1087 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1088 struct btrfs_path *path;
1089 struct extent_buffer *l;
1095 root = kzalloc(sizeof(*root), GFP_NOFS);
1097 return ERR_PTR(-ENOMEM);
1098 if (location->offset == (u64)-1) {
1099 ret = find_and_setup_root(tree_root, fs_info,
1100 location->objectid, root);
1103 return ERR_PTR(ret);
1108 __setup_root(tree_root->nodesize, tree_root->leafsize,
1109 tree_root->sectorsize, tree_root->stripesize,
1110 root, fs_info, location->objectid);
1112 path = btrfs_alloc_path();
1114 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1121 read_extent_buffer(l, &root->root_item,
1122 btrfs_item_ptr_offset(l, path->slots[0]),
1123 sizeof(root->root_item));
1124 memcpy(&root->root_key, location, sizeof(*location));
1127 btrfs_release_path(root, path);
1128 btrfs_free_path(path);
1131 return ERR_PTR(ret);
1133 generation = btrfs_root_generation(&root->root_item);
1134 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1135 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1136 blocksize, generation);
1137 BUG_ON(!root->node);
1139 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1141 ret = btrfs_find_highest_inode(root, &highest_inode);
1143 root->highest_inode = highest_inode;
1144 root->last_inode_alloc = highest_inode;
1150 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1153 struct btrfs_root *root;
1155 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1156 return fs_info->tree_root;
1157 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1158 return fs_info->extent_root;
1160 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1161 (unsigned long)root_objectid);
1165 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1166 struct btrfs_key *location)
1168 struct btrfs_root *root;
1171 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1172 return fs_info->tree_root;
1173 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1174 return fs_info->extent_root;
1175 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1176 return fs_info->chunk_root;
1177 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1178 return fs_info->dev_root;
1179 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1180 return fs_info->csum_root;
1182 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1183 (unsigned long)location->objectid);
1187 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1191 set_anon_super(&root->anon_super, NULL);
1193 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1194 (unsigned long)root->root_key.objectid,
1197 free_extent_buffer(root->node);
1199 return ERR_PTR(ret);
1201 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1202 ret = btrfs_find_dead_roots(fs_info->tree_root,
1203 root->root_key.objectid, root);
1205 btrfs_orphan_cleanup(root);
1210 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1211 struct btrfs_key *location,
1212 const char *name, int namelen)
1214 struct btrfs_root *root;
1217 root = btrfs_read_fs_root_no_name(fs_info, location);
1224 ret = btrfs_set_root_name(root, name, namelen);
1226 free_extent_buffer(root->node);
1228 return ERR_PTR(ret);
1231 ret = btrfs_sysfs_add_root(root);
1233 free_extent_buffer(root->node);
1236 return ERR_PTR(ret);
1243 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1245 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1247 struct btrfs_device *device;
1248 struct backing_dev_info *bdi;
1250 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1253 bdi = blk_get_backing_dev_info(device->bdev);
1254 if (bdi && bdi_congested(bdi, bdi_bits)) {
1263 * this unplugs every device on the box, and it is only used when page
1266 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1268 struct btrfs_device *device;
1269 struct btrfs_fs_info *info;
1271 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1272 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1276 bdi = blk_get_backing_dev_info(device->bdev);
1277 if (bdi->unplug_io_fn)
1278 bdi->unplug_io_fn(bdi, page);
1282 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1284 struct inode *inode;
1285 struct extent_map_tree *em_tree;
1286 struct extent_map *em;
1287 struct address_space *mapping;
1290 /* the generic O_DIRECT read code does this */
1292 __unplug_io_fn(bdi, page);
1297 * page->mapping may change at any time. Get a consistent copy
1298 * and use that for everything below
1301 mapping = page->mapping;
1305 inode = mapping->host;
1308 * don't do the expensive searching for a small number of
1311 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1312 __unplug_io_fn(bdi, page);
1316 offset = page_offset(page);
1318 em_tree = &BTRFS_I(inode)->extent_tree;
1319 spin_lock(&em_tree->lock);
1320 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1321 spin_unlock(&em_tree->lock);
1323 __unplug_io_fn(bdi, page);
1327 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1328 free_extent_map(em);
1329 __unplug_io_fn(bdi, page);
1332 offset = offset - em->start;
1333 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1334 em->block_start + offset, page);
1335 free_extent_map(em);
1338 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1341 bdi->ra_pages = default_backing_dev_info.ra_pages;
1343 bdi->capabilities = default_backing_dev_info.capabilities;
1344 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1345 bdi->unplug_io_data = info;
1346 bdi->congested_fn = btrfs_congested_fn;
1347 bdi->congested_data = info;
1351 static int bio_ready_for_csum(struct bio *bio)
1357 struct extent_io_tree *io_tree = NULL;
1358 struct btrfs_fs_info *info = NULL;
1359 struct bio_vec *bvec;
1363 bio_for_each_segment(bvec, bio, i) {
1364 page = bvec->bv_page;
1365 if (page->private == EXTENT_PAGE_PRIVATE) {
1366 length += bvec->bv_len;
1369 if (!page->private) {
1370 length += bvec->bv_len;
1373 length = bvec->bv_len;
1374 buf_len = page->private >> 2;
1375 start = page_offset(page) + bvec->bv_offset;
1376 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1377 info = BTRFS_I(page->mapping->host)->root->fs_info;
1379 /* are we fully contained in this bio? */
1380 if (buf_len <= length)
1383 ret = extent_range_uptodate(io_tree, start + length,
1384 start + buf_len - 1);
1389 * called by the kthread helper functions to finally call the bio end_io
1390 * functions. This is where read checksum verification actually happens
1392 static void end_workqueue_fn(struct btrfs_work *work)
1395 struct end_io_wq *end_io_wq;
1396 struct btrfs_fs_info *fs_info;
1399 end_io_wq = container_of(work, struct end_io_wq, work);
1400 bio = end_io_wq->bio;
1401 fs_info = end_io_wq->info;
1403 /* metadata bio reads are special because the whole tree block must
1404 * be checksummed at once. This makes sure the entire block is in
1405 * ram and up to date before trying to verify things. For
1406 * blocksize <= pagesize, it is basically a noop
1408 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1409 !bio_ready_for_csum(bio)) {
1410 btrfs_queue_worker(&fs_info->endio_meta_workers,
1414 error = end_io_wq->error;
1415 bio->bi_private = end_io_wq->private;
1416 bio->bi_end_io = end_io_wq->end_io;
1418 bio_endio(bio, error);
1421 static int cleaner_kthread(void *arg)
1423 struct btrfs_root *root = arg;
1427 if (root->fs_info->closing)
1430 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1431 mutex_lock(&root->fs_info->cleaner_mutex);
1432 btrfs_clean_old_snapshots(root);
1433 mutex_unlock(&root->fs_info->cleaner_mutex);
1435 if (freezing(current)) {
1439 if (root->fs_info->closing)
1441 set_current_state(TASK_INTERRUPTIBLE);
1443 __set_current_state(TASK_RUNNING);
1445 } while (!kthread_should_stop());
1449 static int transaction_kthread(void *arg)
1451 struct btrfs_root *root = arg;
1452 struct btrfs_trans_handle *trans;
1453 struct btrfs_transaction *cur;
1455 unsigned long delay;
1460 if (root->fs_info->closing)
1464 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1465 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1467 mutex_lock(&root->fs_info->trans_mutex);
1468 cur = root->fs_info->running_transaction;
1470 mutex_unlock(&root->fs_info->trans_mutex);
1474 now = get_seconds();
1475 if (now < cur->start_time || now - cur->start_time < 30) {
1476 mutex_unlock(&root->fs_info->trans_mutex);
1480 mutex_unlock(&root->fs_info->trans_mutex);
1481 trans = btrfs_start_transaction(root, 1);
1482 ret = btrfs_commit_transaction(trans, root);
1485 wake_up_process(root->fs_info->cleaner_kthread);
1486 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1488 if (freezing(current)) {
1491 if (root->fs_info->closing)
1493 set_current_state(TASK_INTERRUPTIBLE);
1494 schedule_timeout(delay);
1495 __set_current_state(TASK_RUNNING);
1497 } while (!kthread_should_stop());
1501 struct btrfs_root *open_ctree(struct super_block *sb,
1502 struct btrfs_fs_devices *fs_devices,
1512 struct btrfs_key location;
1513 struct buffer_head *bh;
1514 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1516 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1518 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1520 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1522 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1524 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1526 struct btrfs_root *log_tree_root;
1531 struct btrfs_super_block *disk_super;
1533 if (!extent_root || !tree_root || !fs_info ||
1534 !chunk_root || !dev_root || !csum_root) {
1538 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1539 INIT_LIST_HEAD(&fs_info->trans_list);
1540 INIT_LIST_HEAD(&fs_info->dead_roots);
1541 INIT_LIST_HEAD(&fs_info->hashers);
1542 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1543 INIT_LIST_HEAD(&fs_info->ordered_operations);
1544 spin_lock_init(&fs_info->delalloc_lock);
1545 spin_lock_init(&fs_info->new_trans_lock);
1546 spin_lock_init(&fs_info->ref_cache_lock);
1548 init_completion(&fs_info->kobj_unregister);
1549 fs_info->tree_root = tree_root;
1550 fs_info->extent_root = extent_root;
1551 fs_info->csum_root = csum_root;
1552 fs_info->chunk_root = chunk_root;
1553 fs_info->dev_root = dev_root;
1554 fs_info->fs_devices = fs_devices;
1555 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1556 INIT_LIST_HEAD(&fs_info->space_info);
1557 btrfs_mapping_init(&fs_info->mapping_tree);
1558 atomic_set(&fs_info->nr_async_submits, 0);
1559 atomic_set(&fs_info->async_delalloc_pages, 0);
1560 atomic_set(&fs_info->async_submit_draining, 0);
1561 atomic_set(&fs_info->nr_async_bios, 0);
1562 atomic_set(&fs_info->throttles, 0);
1563 atomic_set(&fs_info->throttle_gen, 0);
1565 fs_info->max_extent = (u64)-1;
1566 fs_info->max_inline = 8192 * 1024;
1567 setup_bdi(fs_info, &fs_info->bdi);
1568 fs_info->btree_inode = new_inode(sb);
1569 fs_info->btree_inode->i_ino = 1;
1570 fs_info->btree_inode->i_nlink = 1;
1571 fs_info->metadata_ratio = 8;
1573 fs_info->thread_pool_size = min_t(unsigned long,
1574 num_online_cpus() + 2, 8);
1576 INIT_LIST_HEAD(&fs_info->ordered_extents);
1577 spin_lock_init(&fs_info->ordered_extent_lock);
1579 sb->s_blocksize = 4096;
1580 sb->s_blocksize_bits = blksize_bits(4096);
1583 * we set the i_size on the btree inode to the max possible int.
1584 * the real end of the address space is determined by all of
1585 * the devices in the system
1587 fs_info->btree_inode->i_size = OFFSET_MAX;
1588 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1589 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1591 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1592 fs_info->btree_inode->i_mapping,
1594 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1597 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1599 spin_lock_init(&fs_info->block_group_cache_lock);
1600 fs_info->block_group_cache_tree.rb_node = NULL;
1602 extent_io_tree_init(&fs_info->pinned_extents,
1603 fs_info->btree_inode->i_mapping, GFP_NOFS);
1604 fs_info->do_barriers = 1;
1606 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1607 btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1608 btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1610 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1611 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1612 sizeof(struct btrfs_key));
1613 insert_inode_hash(fs_info->btree_inode);
1615 mutex_init(&fs_info->trans_mutex);
1616 mutex_init(&fs_info->ordered_operations_mutex);
1617 mutex_init(&fs_info->tree_log_mutex);
1618 mutex_init(&fs_info->drop_mutex);
1619 mutex_init(&fs_info->chunk_mutex);
1620 mutex_init(&fs_info->transaction_kthread_mutex);
1621 mutex_init(&fs_info->cleaner_mutex);
1622 mutex_init(&fs_info->volume_mutex);
1623 mutex_init(&fs_info->tree_reloc_mutex);
1625 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1626 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1628 init_waitqueue_head(&fs_info->transaction_throttle);
1629 init_waitqueue_head(&fs_info->transaction_wait);
1630 init_waitqueue_head(&fs_info->async_submit_wait);
1632 __setup_root(4096, 4096, 4096, 4096, tree_root,
1633 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1636 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1640 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1641 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1642 sizeof(fs_info->super_for_commit));
1645 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1647 disk_super = &fs_info->super_copy;
1648 if (!btrfs_super_root(disk_super))
1651 ret = btrfs_parse_options(tree_root, options);
1657 features = btrfs_super_incompat_flags(disk_super) &
1658 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1660 printk(KERN_ERR "BTRFS: couldn't mount because of "
1661 "unsupported optional features (%Lx).\n",
1667 features = btrfs_super_compat_ro_flags(disk_super) &
1668 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1669 if (!(sb->s_flags & MS_RDONLY) && features) {
1670 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1671 "unsupported option features (%Lx).\n",
1678 * we need to start all the end_io workers up front because the
1679 * queue work function gets called at interrupt time, and so it
1680 * cannot dynamically grow.
1682 btrfs_init_workers(&fs_info->workers, "worker",
1683 fs_info->thread_pool_size);
1685 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1686 fs_info->thread_pool_size);
1688 btrfs_init_workers(&fs_info->submit_workers, "submit",
1689 min_t(u64, fs_devices->num_devices,
1690 fs_info->thread_pool_size));
1692 /* a higher idle thresh on the submit workers makes it much more
1693 * likely that bios will be send down in a sane order to the
1696 fs_info->submit_workers.idle_thresh = 64;
1698 fs_info->workers.idle_thresh = 16;
1699 fs_info->workers.ordered = 1;
1701 fs_info->delalloc_workers.idle_thresh = 2;
1702 fs_info->delalloc_workers.ordered = 1;
1704 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1705 btrfs_init_workers(&fs_info->endio_workers, "endio",
1706 fs_info->thread_pool_size);
1707 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1708 fs_info->thread_pool_size);
1709 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1710 "endio-meta-write", fs_info->thread_pool_size);
1711 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1712 fs_info->thread_pool_size);
1715 * endios are largely parallel and should have a very
1718 fs_info->endio_workers.idle_thresh = 4;
1719 fs_info->endio_meta_workers.idle_thresh = 4;
1721 fs_info->endio_write_workers.idle_thresh = 64;
1722 fs_info->endio_meta_write_workers.idle_thresh = 64;
1724 btrfs_start_workers(&fs_info->workers, 1);
1725 btrfs_start_workers(&fs_info->submit_workers, 1);
1726 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1727 btrfs_start_workers(&fs_info->fixup_workers, 1);
1728 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1729 btrfs_start_workers(&fs_info->endio_meta_workers,
1730 fs_info->thread_pool_size);
1731 btrfs_start_workers(&fs_info->endio_meta_write_workers,
1732 fs_info->thread_pool_size);
1733 btrfs_start_workers(&fs_info->endio_write_workers,
1734 fs_info->thread_pool_size);
1736 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1737 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1738 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1740 nodesize = btrfs_super_nodesize(disk_super);
1741 leafsize = btrfs_super_leafsize(disk_super);
1742 sectorsize = btrfs_super_sectorsize(disk_super);
1743 stripesize = btrfs_super_stripesize(disk_super);
1744 tree_root->nodesize = nodesize;
1745 tree_root->leafsize = leafsize;
1746 tree_root->sectorsize = sectorsize;
1747 tree_root->stripesize = stripesize;
1749 sb->s_blocksize = sectorsize;
1750 sb->s_blocksize_bits = blksize_bits(sectorsize);
1752 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1753 sizeof(disk_super->magic))) {
1754 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1755 goto fail_sb_buffer;
1758 mutex_lock(&fs_info->chunk_mutex);
1759 ret = btrfs_read_sys_array(tree_root);
1760 mutex_unlock(&fs_info->chunk_mutex);
1762 printk(KERN_WARNING "btrfs: failed to read the system "
1763 "array on %s\n", sb->s_id);
1764 goto fail_sys_array;
1767 blocksize = btrfs_level_size(tree_root,
1768 btrfs_super_chunk_root_level(disk_super));
1769 generation = btrfs_super_chunk_root_generation(disk_super);
1771 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1772 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1774 chunk_root->node = read_tree_block(chunk_root,
1775 btrfs_super_chunk_root(disk_super),
1776 blocksize, generation);
1777 BUG_ON(!chunk_root->node);
1779 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1780 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1783 mutex_lock(&fs_info->chunk_mutex);
1784 ret = btrfs_read_chunk_tree(chunk_root);
1785 mutex_unlock(&fs_info->chunk_mutex);
1787 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1789 goto fail_chunk_root;
1792 btrfs_close_extra_devices(fs_devices);
1794 blocksize = btrfs_level_size(tree_root,
1795 btrfs_super_root_level(disk_super));
1796 generation = btrfs_super_generation(disk_super);
1798 tree_root->node = read_tree_block(tree_root,
1799 btrfs_super_root(disk_super),
1800 blocksize, generation);
1801 if (!tree_root->node)
1802 goto fail_chunk_root;
1805 ret = find_and_setup_root(tree_root, fs_info,
1806 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1808 goto fail_tree_root;
1809 extent_root->track_dirty = 1;
1811 ret = find_and_setup_root(tree_root, fs_info,
1812 BTRFS_DEV_TREE_OBJECTID, dev_root);
1813 dev_root->track_dirty = 1;
1815 goto fail_extent_root;
1817 ret = find_and_setup_root(tree_root, fs_info,
1818 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1820 goto fail_extent_root;
1822 csum_root->track_dirty = 1;
1824 btrfs_read_block_groups(extent_root);
1826 fs_info->generation = generation;
1827 fs_info->last_trans_committed = generation;
1828 fs_info->data_alloc_profile = (u64)-1;
1829 fs_info->metadata_alloc_profile = (u64)-1;
1830 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1831 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1833 if (IS_ERR(fs_info->cleaner_kthread))
1834 goto fail_csum_root;
1836 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1838 "btrfs-transaction");
1839 if (IS_ERR(fs_info->transaction_kthread))
1842 if (btrfs_super_log_root(disk_super) != 0) {
1843 u64 bytenr = btrfs_super_log_root(disk_super);
1845 if (fs_devices->rw_devices == 0) {
1846 printk(KERN_WARNING "Btrfs log replay required "
1849 goto fail_trans_kthread;
1852 btrfs_level_size(tree_root,
1853 btrfs_super_log_root_level(disk_super));
1855 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1858 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1859 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1861 log_tree_root->node = read_tree_block(tree_root, bytenr,
1864 ret = btrfs_recover_log_trees(log_tree_root);
1867 if (sb->s_flags & MS_RDONLY) {
1868 ret = btrfs_commit_super(tree_root);
1873 if (!(sb->s_flags & MS_RDONLY)) {
1874 ret = btrfs_cleanup_reloc_trees(tree_root);
1878 location.objectid = BTRFS_FS_TREE_OBJECTID;
1879 location.type = BTRFS_ROOT_ITEM_KEY;
1880 location.offset = (u64)-1;
1882 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1883 if (!fs_info->fs_root)
1884 goto fail_trans_kthread;
1888 kthread_stop(fs_info->transaction_kthread);
1890 kthread_stop(fs_info->cleaner_kthread);
1893 * make sure we're done with the btree inode before we stop our
1896 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1897 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1900 free_extent_buffer(csum_root->node);
1902 free_extent_buffer(extent_root->node);
1904 free_extent_buffer(tree_root->node);
1906 free_extent_buffer(chunk_root->node);
1908 free_extent_buffer(dev_root->node);
1910 btrfs_stop_workers(&fs_info->fixup_workers);
1911 btrfs_stop_workers(&fs_info->delalloc_workers);
1912 btrfs_stop_workers(&fs_info->workers);
1913 btrfs_stop_workers(&fs_info->endio_workers);
1914 btrfs_stop_workers(&fs_info->endio_meta_workers);
1915 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1916 btrfs_stop_workers(&fs_info->endio_write_workers);
1917 btrfs_stop_workers(&fs_info->submit_workers);
1919 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1920 iput(fs_info->btree_inode);
1922 btrfs_close_devices(fs_info->fs_devices);
1923 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1924 bdi_destroy(&fs_info->bdi);
1933 return ERR_PTR(err);
1936 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1938 char b[BDEVNAME_SIZE];
1941 set_buffer_uptodate(bh);
1943 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1944 printk(KERN_WARNING "lost page write due to "
1945 "I/O error on %s\n",
1946 bdevname(bh->b_bdev, b));
1948 /* note, we dont' set_buffer_write_io_error because we have
1949 * our own ways of dealing with the IO errors
1951 clear_buffer_uptodate(bh);
1957 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1959 struct buffer_head *bh;
1960 struct buffer_head *latest = NULL;
1961 struct btrfs_super_block *super;
1966 /* we would like to check all the supers, but that would make
1967 * a btrfs mount succeed after a mkfs from a different FS.
1968 * So, we need to add a special mount option to scan for
1969 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
1971 for (i = 0; i < 1; i++) {
1972 bytenr = btrfs_sb_offset(i);
1973 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
1975 bh = __bread(bdev, bytenr / 4096, 4096);
1979 super = (struct btrfs_super_block *)bh->b_data;
1980 if (btrfs_super_bytenr(super) != bytenr ||
1981 strncmp((char *)(&super->magic), BTRFS_MAGIC,
1982 sizeof(super->magic))) {
1987 if (!latest || btrfs_super_generation(super) > transid) {
1990 transid = btrfs_super_generation(super);
1998 static int write_dev_supers(struct btrfs_device *device,
1999 struct btrfs_super_block *sb,
2000 int do_barriers, int wait, int max_mirrors)
2002 struct buffer_head *bh;
2008 int last_barrier = 0;
2010 if (max_mirrors == 0)
2011 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2013 /* make sure only the last submit_bh does a barrier */
2015 for (i = 0; i < max_mirrors; i++) {
2016 bytenr = btrfs_sb_offset(i);
2017 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2018 device->total_bytes)
2024 for (i = 0; i < max_mirrors; i++) {
2025 bytenr = btrfs_sb_offset(i);
2026 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2030 bh = __find_get_block(device->bdev, bytenr / 4096,
2031 BTRFS_SUPER_INFO_SIZE);
2035 if (buffer_uptodate(bh)) {
2040 btrfs_set_super_bytenr(sb, bytenr);
2043 crc = btrfs_csum_data(NULL, (char *)sb +
2044 BTRFS_CSUM_SIZE, crc,
2045 BTRFS_SUPER_INFO_SIZE -
2047 btrfs_csum_final(crc, sb->csum);
2049 bh = __getblk(device->bdev, bytenr / 4096,
2050 BTRFS_SUPER_INFO_SIZE);
2051 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2053 set_buffer_uptodate(bh);
2056 bh->b_end_io = btrfs_end_buffer_write_sync;
2059 if (i == last_barrier && do_barriers && device->barriers) {
2060 ret = submit_bh(WRITE_BARRIER, bh);
2061 if (ret == -EOPNOTSUPP) {
2062 printk("btrfs: disabling barriers on dev %s\n",
2064 set_buffer_uptodate(bh);
2065 device->barriers = 0;
2068 ret = submit_bh(WRITE_SYNC, bh);
2071 ret = submit_bh(WRITE_SYNC, bh);
2076 if (!buffer_uptodate(bh))
2084 return errors < i ? 0 : -1;
2087 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2089 struct list_head *head = &root->fs_info->fs_devices->devices;
2090 struct btrfs_device *dev;
2091 struct btrfs_super_block *sb;
2092 struct btrfs_dev_item *dev_item;
2096 int total_errors = 0;
2099 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2100 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2102 sb = &root->fs_info->super_for_commit;
2103 dev_item = &sb->dev_item;
2104 list_for_each_entry(dev, head, dev_list) {
2109 if (!dev->in_fs_metadata || !dev->writeable)
2112 btrfs_set_stack_device_generation(dev_item, 0);
2113 btrfs_set_stack_device_type(dev_item, dev->type);
2114 btrfs_set_stack_device_id(dev_item, dev->devid);
2115 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2116 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2117 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2118 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2119 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2120 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2121 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2123 flags = btrfs_super_flags(sb);
2124 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2126 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2130 if (total_errors > max_errors) {
2131 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2137 list_for_each_entry(dev, head, dev_list) {
2140 if (!dev->in_fs_metadata || !dev->writeable)
2143 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2147 if (total_errors > max_errors) {
2148 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2155 int write_ctree_super(struct btrfs_trans_handle *trans,
2156 struct btrfs_root *root, int max_mirrors)
2160 ret = write_all_supers(root, max_mirrors);
2164 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2166 radix_tree_delete(&fs_info->fs_roots_radix,
2167 (unsigned long)root->root_key.objectid);
2168 if (root->anon_super.s_dev) {
2169 down_write(&root->anon_super.s_umount);
2170 kill_anon_super(&root->anon_super);
2173 free_extent_buffer(root->node);
2174 if (root->commit_root)
2175 free_extent_buffer(root->commit_root);
2181 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2184 struct btrfs_root *gang[8];
2188 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2193 for (i = 0; i < ret; i++)
2194 btrfs_free_fs_root(fs_info, gang[i]);
2199 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2201 u64 root_objectid = 0;
2202 struct btrfs_root *gang[8];
2207 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2208 (void **)gang, root_objectid,
2212 for (i = 0; i < ret; i++) {
2213 root_objectid = gang[i]->root_key.objectid;
2214 ret = btrfs_find_dead_roots(fs_info->tree_root,
2215 root_objectid, gang[i]);
2217 btrfs_orphan_cleanup(gang[i]);
2224 int btrfs_commit_super(struct btrfs_root *root)
2226 struct btrfs_trans_handle *trans;
2229 mutex_lock(&root->fs_info->cleaner_mutex);
2230 btrfs_clean_old_snapshots(root);
2231 mutex_unlock(&root->fs_info->cleaner_mutex);
2232 trans = btrfs_start_transaction(root, 1);
2233 ret = btrfs_commit_transaction(trans, root);
2235 /* run commit again to drop the original snapshot */
2236 trans = btrfs_start_transaction(root, 1);
2237 btrfs_commit_transaction(trans, root);
2238 ret = btrfs_write_and_wait_transaction(NULL, root);
2241 ret = write_ctree_super(NULL, root, 0);
2245 int close_ctree(struct btrfs_root *root)
2247 struct btrfs_fs_info *fs_info = root->fs_info;
2250 fs_info->closing = 1;
2253 kthread_stop(root->fs_info->transaction_kthread);
2254 kthread_stop(root->fs_info->cleaner_kthread);
2256 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2257 ret = btrfs_commit_super(root);
2259 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2262 if (fs_info->delalloc_bytes) {
2263 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2264 fs_info->delalloc_bytes);
2266 if (fs_info->total_ref_cache_size) {
2267 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2268 (unsigned long long)fs_info->total_ref_cache_size);
2271 if (fs_info->extent_root->node)
2272 free_extent_buffer(fs_info->extent_root->node);
2274 if (fs_info->tree_root->node)
2275 free_extent_buffer(fs_info->tree_root->node);
2277 if (root->fs_info->chunk_root->node)
2278 free_extent_buffer(root->fs_info->chunk_root->node);
2280 if (root->fs_info->dev_root->node)
2281 free_extent_buffer(root->fs_info->dev_root->node);
2283 if (root->fs_info->csum_root->node)
2284 free_extent_buffer(root->fs_info->csum_root->node);
2286 btrfs_free_block_groups(root->fs_info);
2288 del_fs_roots(fs_info);
2290 iput(fs_info->btree_inode);
2292 btrfs_stop_workers(&fs_info->fixup_workers);
2293 btrfs_stop_workers(&fs_info->delalloc_workers);
2294 btrfs_stop_workers(&fs_info->workers);
2295 btrfs_stop_workers(&fs_info->endio_workers);
2296 btrfs_stop_workers(&fs_info->endio_meta_workers);
2297 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2298 btrfs_stop_workers(&fs_info->endio_write_workers);
2299 btrfs_stop_workers(&fs_info->submit_workers);
2301 btrfs_close_devices(fs_info->fs_devices);
2302 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2304 bdi_destroy(&fs_info->bdi);
2306 kfree(fs_info->extent_root);
2307 kfree(fs_info->tree_root);
2308 kfree(fs_info->chunk_root);
2309 kfree(fs_info->dev_root);
2310 kfree(fs_info->csum_root);
2314 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2317 struct inode *btree_inode = buf->first_page->mapping->host;
2319 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2323 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2328 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2330 struct inode *btree_inode = buf->first_page->mapping->host;
2331 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2335 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2337 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2338 u64 transid = btrfs_header_generation(buf);
2339 struct inode *btree_inode = root->fs_info->btree_inode;
2342 btrfs_assert_tree_locked(buf);
2343 if (transid != root->fs_info->generation) {
2344 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2345 "found %llu running %llu\n",
2346 (unsigned long long)buf->start,
2347 (unsigned long long)transid,
2348 (unsigned long long)root->fs_info->generation);
2351 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2354 spin_lock(&root->fs_info->delalloc_lock);
2355 root->fs_info->dirty_metadata_bytes += buf->len;
2356 spin_unlock(&root->fs_info->delalloc_lock);
2360 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2363 * looks as though older kernels can get into trouble with
2364 * this code, they end up stuck in balance_dirty_pages forever
2366 struct extent_io_tree *tree;
2369 unsigned long thresh = 32 * 1024 * 1024;
2370 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2372 if (current->flags & PF_MEMALLOC)
2375 num_dirty = count_range_bits(tree, &start, (u64)-1,
2376 thresh, EXTENT_DIRTY);
2377 if (num_dirty > thresh) {
2378 balance_dirty_pages_ratelimited_nr(
2379 root->fs_info->btree_inode->i_mapping, 1);
2384 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2386 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2388 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2390 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2394 int btree_lock_page_hook(struct page *page)
2396 struct inode *inode = page->mapping->host;
2397 struct btrfs_root *root = BTRFS_I(inode)->root;
2398 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2399 struct extent_buffer *eb;
2401 u64 bytenr = page_offset(page);
2403 if (page->private == EXTENT_PAGE_PRIVATE)
2406 len = page->private >> 2;
2407 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2411 btrfs_tree_lock(eb);
2412 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2414 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2415 spin_lock(&root->fs_info->delalloc_lock);
2416 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2417 root->fs_info->dirty_metadata_bytes -= eb->len;
2420 spin_unlock(&root->fs_info->delalloc_lock);
2423 btrfs_tree_unlock(eb);
2424 free_extent_buffer(eb);
2430 static struct extent_io_ops btree_extent_io_ops = {
2431 .write_cache_pages_lock_hook = btree_lock_page_hook,
2432 .readpage_end_io_hook = btree_readpage_end_io_hook,
2433 .submit_bio_hook = btree_submit_bio_hook,
2434 /* note we're sharing with inode.c for the merge bio hook */
2435 .merge_bio_hook = btrfs_merge_bio_hook,