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>
33 #include "transaction.h"
34 #include "btrfs_inode.h"
36 #include "print-tree.h"
37 #include "async-thread.h"
40 #include "free-space-cache.h"
42 static struct extent_io_ops btree_extent_io_ops;
43 static void end_workqueue_fn(struct btrfs_work *work);
45 static atomic_t btrfs_bdi_num = ATOMIC_INIT(0);
48 * end_io_wq structs are used to do processing in task context when an IO is
49 * complete. This is used during reads to verify checksums, and it is used
50 * by writes to insert metadata for new file extents after IO is complete.
56 struct btrfs_fs_info *info;
59 struct list_head list;
60 struct btrfs_work work;
64 * async submit bios are used to offload expensive checksumming
65 * onto the worker threads. They checksum file and metadata bios
66 * just before they are sent down the IO stack.
68 struct async_submit_bio {
71 struct list_head list;
72 extent_submit_bio_hook_t *submit_bio_start;
73 extent_submit_bio_hook_t *submit_bio_done;
76 unsigned long bio_flags;
77 struct btrfs_work work;
80 /* These are used to set the lockdep class on the extent buffer locks.
81 * The class is set by the readpage_end_io_hook after the buffer has
82 * passed csum validation but before the pages are unlocked.
84 * The lockdep class is also set by btrfs_init_new_buffer on freshly
87 * The class is based on the level in the tree block, which allows lockdep
88 * to know that lower nodes nest inside the locks of higher nodes.
90 * We also add a check to make sure the highest level of the tree is
91 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
92 * code needs update as well.
94 #ifdef CONFIG_DEBUG_LOCK_ALLOC
95 # if BTRFS_MAX_LEVEL != 8
98 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
99 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
109 /* highest possible level */
115 * extents on the btree inode are pretty simple, there's one extent
116 * that covers the entire device
118 static struct extent_map *btree_get_extent(struct inode *inode,
119 struct page *page, size_t page_offset, u64 start, u64 len,
122 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
123 struct extent_map *em;
126 read_lock(&em_tree->lock);
127 em = lookup_extent_mapping(em_tree, start, len);
130 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
131 read_unlock(&em_tree->lock);
134 read_unlock(&em_tree->lock);
136 em = alloc_extent_map(GFP_NOFS);
138 em = ERR_PTR(-ENOMEM);
143 em->block_len = (u64)-1;
145 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
147 write_lock(&em_tree->lock);
148 ret = add_extent_mapping(em_tree, em);
149 if (ret == -EEXIST) {
150 u64 failed_start = em->start;
151 u64 failed_len = em->len;
154 em = lookup_extent_mapping(em_tree, start, len);
158 em = lookup_extent_mapping(em_tree, failed_start,
166 write_unlock(&em_tree->lock);
174 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
176 return crc32c(seed, data, len);
179 void btrfs_csum_final(u32 crc, char *result)
181 *(__le32 *)result = ~cpu_to_le32(crc);
185 * compute the csum for a btree block, and either verify it or write it
186 * into the csum field of the block.
188 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
192 btrfs_super_csum_size(&root->fs_info->super_copy);
195 unsigned long cur_len;
196 unsigned long offset = BTRFS_CSUM_SIZE;
197 char *map_token = NULL;
199 unsigned long map_start;
200 unsigned long map_len;
203 unsigned long inline_result;
205 len = buf->len - offset;
207 err = map_private_extent_buffer(buf, offset, 32,
209 &map_start, &map_len, KM_USER0);
212 cur_len = min(len, map_len - (offset - map_start));
213 crc = btrfs_csum_data(root, kaddr + offset - map_start,
217 unmap_extent_buffer(buf, map_token, KM_USER0);
219 if (csum_size > sizeof(inline_result)) {
220 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
224 result = (char *)&inline_result;
227 btrfs_csum_final(crc, result);
230 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
233 memcpy(&found, result, csum_size);
235 read_extent_buffer(buf, &val, 0, csum_size);
236 if (printk_ratelimit()) {
237 printk(KERN_INFO "btrfs: %s checksum verify "
238 "failed on %llu wanted %X found %X "
240 root->fs_info->sb->s_id,
241 (unsigned long long)buf->start, val, found,
242 btrfs_header_level(buf));
244 if (result != (char *)&inline_result)
249 write_extent_buffer(buf, result, 0, csum_size);
251 if (result != (char *)&inline_result)
257 * we can't consider a given block up to date unless the transid of the
258 * block matches the transid in the parent node's pointer. This is how we
259 * detect blocks that either didn't get written at all or got written
260 * in the wrong place.
262 static int verify_parent_transid(struct extent_io_tree *io_tree,
263 struct extent_buffer *eb, u64 parent_transid)
267 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
270 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
271 if (extent_buffer_uptodate(io_tree, eb) &&
272 btrfs_header_generation(eb) == parent_transid) {
276 if (printk_ratelimit()) {
277 printk("parent transid verify failed on %llu wanted %llu "
279 (unsigned long long)eb->start,
280 (unsigned long long)parent_transid,
281 (unsigned long long)btrfs_header_generation(eb));
284 clear_extent_buffer_uptodate(io_tree, eb);
286 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
292 * helper to read a given tree block, doing retries as required when
293 * the checksums don't match and we have alternate mirrors to try.
295 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
296 struct extent_buffer *eb,
297 u64 start, u64 parent_transid)
299 struct extent_io_tree *io_tree;
304 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
306 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
307 btree_get_extent, mirror_num);
309 !verify_parent_transid(io_tree, eb, parent_transid))
312 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
318 if (mirror_num > num_copies)
325 * checksum a dirty tree block before IO. This has extra checks to make sure
326 * we only fill in the checksum field in the first page of a multi-page block
329 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
331 struct extent_io_tree *tree;
332 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
336 struct extent_buffer *eb;
339 tree = &BTRFS_I(page->mapping->host)->io_tree;
341 if (page->private == EXTENT_PAGE_PRIVATE)
345 len = page->private >> 2;
348 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
349 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
350 btrfs_header_generation(eb));
352 found_start = btrfs_header_bytenr(eb);
353 if (found_start != start) {
357 if (eb->first_page != page) {
361 if (!PageUptodate(page)) {
365 found_level = btrfs_header_level(eb);
367 csum_tree_block(root, eb, 0);
369 free_extent_buffer(eb);
374 static int check_tree_block_fsid(struct btrfs_root *root,
375 struct extent_buffer *eb)
377 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
378 u8 fsid[BTRFS_UUID_SIZE];
381 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
384 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
388 fs_devices = fs_devices->seed;
393 #ifdef CONFIG_DEBUG_LOCK_ALLOC
394 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
396 lockdep_set_class_and_name(&eb->lock,
397 &btrfs_eb_class[level],
398 btrfs_eb_name[level]);
402 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
403 struct extent_state *state)
405 struct extent_io_tree *tree;
409 struct extent_buffer *eb;
410 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
413 tree = &BTRFS_I(page->mapping->host)->io_tree;
414 if (page->private == EXTENT_PAGE_PRIVATE)
419 len = page->private >> 2;
422 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
424 found_start = btrfs_header_bytenr(eb);
425 if (found_start != start) {
426 if (printk_ratelimit()) {
427 printk(KERN_INFO "btrfs bad tree block start "
429 (unsigned long long)found_start,
430 (unsigned long long)eb->start);
435 if (eb->first_page != page) {
436 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
437 eb->first_page->index, page->index);
442 if (check_tree_block_fsid(root, eb)) {
443 if (printk_ratelimit()) {
444 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
445 (unsigned long long)eb->start);
450 found_level = btrfs_header_level(eb);
452 btrfs_set_buffer_lockdep_class(eb, found_level);
454 ret = csum_tree_block(root, eb, 1);
458 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
459 end = eb->start + end - 1;
461 free_extent_buffer(eb);
466 static void end_workqueue_bio(struct bio *bio, int err)
468 struct end_io_wq *end_io_wq = bio->bi_private;
469 struct btrfs_fs_info *fs_info;
471 fs_info = end_io_wq->info;
472 end_io_wq->error = err;
473 end_io_wq->work.func = end_workqueue_fn;
474 end_io_wq->work.flags = 0;
476 if (bio->bi_rw & (1 << BIO_RW)) {
477 if (end_io_wq->metadata)
478 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
481 btrfs_queue_worker(&fs_info->endio_write_workers,
484 if (end_io_wq->metadata)
485 btrfs_queue_worker(&fs_info->endio_meta_workers,
488 btrfs_queue_worker(&fs_info->endio_workers,
493 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
496 struct end_io_wq *end_io_wq;
497 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
501 end_io_wq->private = bio->bi_private;
502 end_io_wq->end_io = bio->bi_end_io;
503 end_io_wq->info = info;
504 end_io_wq->error = 0;
505 end_io_wq->bio = bio;
506 end_io_wq->metadata = metadata;
508 bio->bi_private = end_io_wq;
509 bio->bi_end_io = end_workqueue_bio;
513 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
515 unsigned long limit = min_t(unsigned long,
516 info->workers.max_workers,
517 info->fs_devices->open_devices);
521 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
523 return atomic_read(&info->nr_async_bios) >
524 btrfs_async_submit_limit(info);
527 static void run_one_async_start(struct btrfs_work *work)
529 struct btrfs_fs_info *fs_info;
530 struct async_submit_bio *async;
532 async = container_of(work, struct async_submit_bio, work);
533 fs_info = BTRFS_I(async->inode)->root->fs_info;
534 async->submit_bio_start(async->inode, async->rw, async->bio,
535 async->mirror_num, async->bio_flags);
538 static void run_one_async_done(struct btrfs_work *work)
540 struct btrfs_fs_info *fs_info;
541 struct async_submit_bio *async;
544 async = container_of(work, struct async_submit_bio, work);
545 fs_info = BTRFS_I(async->inode)->root->fs_info;
547 limit = btrfs_async_submit_limit(fs_info);
548 limit = limit * 2 / 3;
550 atomic_dec(&fs_info->nr_async_submits);
552 if (atomic_read(&fs_info->nr_async_submits) < limit &&
553 waitqueue_active(&fs_info->async_submit_wait))
554 wake_up(&fs_info->async_submit_wait);
556 async->submit_bio_done(async->inode, async->rw, async->bio,
557 async->mirror_num, async->bio_flags);
560 static void run_one_async_free(struct btrfs_work *work)
562 struct async_submit_bio *async;
564 async = container_of(work, struct async_submit_bio, work);
568 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
569 int rw, struct bio *bio, int mirror_num,
570 unsigned long bio_flags,
571 extent_submit_bio_hook_t *submit_bio_start,
572 extent_submit_bio_hook_t *submit_bio_done)
574 struct async_submit_bio *async;
576 async = kmalloc(sizeof(*async), GFP_NOFS);
580 async->inode = inode;
583 async->mirror_num = mirror_num;
584 async->submit_bio_start = submit_bio_start;
585 async->submit_bio_done = submit_bio_done;
587 async->work.func = run_one_async_start;
588 async->work.ordered_func = run_one_async_done;
589 async->work.ordered_free = run_one_async_free;
591 async->work.flags = 0;
592 async->bio_flags = bio_flags;
594 atomic_inc(&fs_info->nr_async_submits);
596 if (rw & (1 << BIO_RW_SYNCIO))
597 btrfs_set_work_high_prio(&async->work);
599 btrfs_queue_worker(&fs_info->workers, &async->work);
601 while (atomic_read(&fs_info->async_submit_draining) &&
602 atomic_read(&fs_info->nr_async_submits)) {
603 wait_event(fs_info->async_submit_wait,
604 (atomic_read(&fs_info->nr_async_submits) == 0));
610 static int btree_csum_one_bio(struct bio *bio)
612 struct bio_vec *bvec = bio->bi_io_vec;
614 struct btrfs_root *root;
616 WARN_ON(bio->bi_vcnt <= 0);
617 while (bio_index < bio->bi_vcnt) {
618 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
619 csum_dirty_buffer(root, bvec->bv_page);
626 static int __btree_submit_bio_start(struct inode *inode, int rw,
627 struct bio *bio, int mirror_num,
628 unsigned long bio_flags)
631 * when we're called for a write, we're already in the async
632 * submission context. Just jump into btrfs_map_bio
634 btree_csum_one_bio(bio);
638 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
639 int mirror_num, unsigned long bio_flags)
642 * when we're called for a write, we're already in the async
643 * submission context. Just jump into btrfs_map_bio
645 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
648 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
649 int mirror_num, unsigned long bio_flags)
653 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
657 if (!(rw & (1 << BIO_RW))) {
659 * called for a read, do the setup so that checksum validation
660 * can happen in the async kernel threads
662 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
667 * kthread helpers are used to submit writes so that checksumming
668 * can happen in parallel across all CPUs
670 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
671 inode, rw, bio, mirror_num, 0,
672 __btree_submit_bio_start,
673 __btree_submit_bio_done);
676 static int btree_writepage(struct page *page, struct writeback_control *wbc)
678 struct extent_io_tree *tree;
679 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
680 struct extent_buffer *eb;
683 tree = &BTRFS_I(page->mapping->host)->io_tree;
684 if (!(current->flags & PF_MEMALLOC)) {
685 return extent_write_full_page(tree, page,
686 btree_get_extent, wbc);
689 redirty_page_for_writepage(wbc, page);
690 eb = btrfs_find_tree_block(root, page_offset(page),
694 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
696 spin_lock(&root->fs_info->delalloc_lock);
697 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
698 spin_unlock(&root->fs_info->delalloc_lock);
700 free_extent_buffer(eb);
706 static int btree_writepages(struct address_space *mapping,
707 struct writeback_control *wbc)
709 struct extent_io_tree *tree;
710 tree = &BTRFS_I(mapping->host)->io_tree;
711 if (wbc->sync_mode == WB_SYNC_NONE) {
712 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
714 unsigned long thresh = 32 * 1024 * 1024;
716 if (wbc->for_kupdate)
719 /* this is a bit racy, but that's ok */
720 num_dirty = root->fs_info->dirty_metadata_bytes;
721 if (num_dirty < thresh)
724 return extent_writepages(tree, mapping, btree_get_extent, wbc);
727 static int btree_readpage(struct file *file, struct page *page)
729 struct extent_io_tree *tree;
730 tree = &BTRFS_I(page->mapping->host)->io_tree;
731 return extent_read_full_page(tree, page, btree_get_extent);
734 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
736 struct extent_io_tree *tree;
737 struct extent_map_tree *map;
740 if (PageWriteback(page) || PageDirty(page))
743 tree = &BTRFS_I(page->mapping->host)->io_tree;
744 map = &BTRFS_I(page->mapping->host)->extent_tree;
746 ret = try_release_extent_state(map, tree, page, gfp_flags);
750 ret = try_release_extent_buffer(tree, page);
752 ClearPagePrivate(page);
753 set_page_private(page, 0);
754 page_cache_release(page);
760 static void btree_invalidatepage(struct page *page, unsigned long offset)
762 struct extent_io_tree *tree;
763 tree = &BTRFS_I(page->mapping->host)->io_tree;
764 extent_invalidatepage(tree, page, offset);
765 btree_releasepage(page, GFP_NOFS);
766 if (PagePrivate(page)) {
767 printk(KERN_WARNING "btrfs warning page private not zero "
768 "on page %llu\n", (unsigned long long)page_offset(page));
769 ClearPagePrivate(page);
770 set_page_private(page, 0);
771 page_cache_release(page);
775 static struct address_space_operations btree_aops = {
776 .readpage = btree_readpage,
777 .writepage = btree_writepage,
778 .writepages = btree_writepages,
779 .releasepage = btree_releasepage,
780 .invalidatepage = btree_invalidatepage,
781 .sync_page = block_sync_page,
784 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
787 struct extent_buffer *buf = NULL;
788 struct inode *btree_inode = root->fs_info->btree_inode;
791 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
794 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
795 buf, 0, 0, btree_get_extent, 0);
796 free_extent_buffer(buf);
800 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
801 u64 bytenr, u32 blocksize)
803 struct inode *btree_inode = root->fs_info->btree_inode;
804 struct extent_buffer *eb;
805 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
806 bytenr, blocksize, GFP_NOFS);
810 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
811 u64 bytenr, u32 blocksize)
813 struct inode *btree_inode = root->fs_info->btree_inode;
814 struct extent_buffer *eb;
816 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
817 bytenr, blocksize, NULL, GFP_NOFS);
822 int btrfs_write_tree_block(struct extent_buffer *buf)
824 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
825 buf->start + buf->len - 1, WB_SYNC_ALL);
828 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
830 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
831 buf->start, buf->start + buf->len - 1);
834 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
835 u32 blocksize, u64 parent_transid)
837 struct extent_buffer *buf = NULL;
838 struct inode *btree_inode = root->fs_info->btree_inode;
839 struct extent_io_tree *io_tree;
842 io_tree = &BTRFS_I(btree_inode)->io_tree;
844 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
848 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
851 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
856 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
857 struct extent_buffer *buf)
859 struct inode *btree_inode = root->fs_info->btree_inode;
860 if (btrfs_header_generation(buf) ==
861 root->fs_info->running_transaction->transid) {
862 btrfs_assert_tree_locked(buf);
864 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
865 spin_lock(&root->fs_info->delalloc_lock);
866 if (root->fs_info->dirty_metadata_bytes >= buf->len)
867 root->fs_info->dirty_metadata_bytes -= buf->len;
870 spin_unlock(&root->fs_info->delalloc_lock);
873 /* ugh, clear_extent_buffer_dirty needs to lock the page */
874 btrfs_set_lock_blocking(buf);
875 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
881 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
882 u32 stripesize, struct btrfs_root *root,
883 struct btrfs_fs_info *fs_info,
887 root->commit_root = NULL;
888 root->sectorsize = sectorsize;
889 root->nodesize = nodesize;
890 root->leafsize = leafsize;
891 root->stripesize = stripesize;
893 root->track_dirty = 0;
895 root->fs_info = fs_info;
896 root->objectid = objectid;
897 root->last_trans = 0;
898 root->highest_inode = 0;
899 root->last_inode_alloc = 0;
902 root->inode_tree.rb_node = NULL;
904 INIT_LIST_HEAD(&root->dirty_list);
905 INIT_LIST_HEAD(&root->orphan_list);
906 INIT_LIST_HEAD(&root->root_list);
907 spin_lock_init(&root->node_lock);
908 spin_lock_init(&root->list_lock);
909 spin_lock_init(&root->inode_lock);
910 mutex_init(&root->objectid_mutex);
911 mutex_init(&root->log_mutex);
912 init_waitqueue_head(&root->log_writer_wait);
913 init_waitqueue_head(&root->log_commit_wait[0]);
914 init_waitqueue_head(&root->log_commit_wait[1]);
915 atomic_set(&root->log_commit[0], 0);
916 atomic_set(&root->log_commit[1], 0);
917 atomic_set(&root->log_writers, 0);
919 root->log_transid = 0;
920 extent_io_tree_init(&root->dirty_log_pages,
921 fs_info->btree_inode->i_mapping, GFP_NOFS);
923 memset(&root->root_key, 0, sizeof(root->root_key));
924 memset(&root->root_item, 0, sizeof(root->root_item));
925 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
926 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
927 root->defrag_trans_start = fs_info->generation;
928 init_completion(&root->kobj_unregister);
929 root->defrag_running = 0;
930 root->defrag_level = 0;
931 root->root_key.objectid = objectid;
932 root->anon_super.s_root = NULL;
933 root->anon_super.s_dev = 0;
934 INIT_LIST_HEAD(&root->anon_super.s_list);
935 INIT_LIST_HEAD(&root->anon_super.s_instances);
936 init_rwsem(&root->anon_super.s_umount);
941 static int find_and_setup_root(struct btrfs_root *tree_root,
942 struct btrfs_fs_info *fs_info,
944 struct btrfs_root *root)
950 __setup_root(tree_root->nodesize, tree_root->leafsize,
951 tree_root->sectorsize, tree_root->stripesize,
952 root, fs_info, objectid);
953 ret = btrfs_find_last_root(tree_root, objectid,
954 &root->root_item, &root->root_key);
957 generation = btrfs_root_generation(&root->root_item);
958 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
959 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
960 blocksize, generation);
961 root->commit_root = btrfs_root_node(root);
966 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
967 struct btrfs_fs_info *fs_info)
969 struct extent_buffer *eb;
970 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
979 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
980 0, &start, &end, EXTENT_DIRTY);
984 clear_extent_dirty(&log_root_tree->dirty_log_pages,
985 start, end, GFP_NOFS);
987 eb = fs_info->log_root_tree->node;
989 WARN_ON(btrfs_header_level(eb) != 0);
990 WARN_ON(btrfs_header_nritems(eb) != 0);
992 ret = btrfs_free_reserved_extent(fs_info->tree_root,
996 free_extent_buffer(eb);
997 kfree(fs_info->log_root_tree);
998 fs_info->log_root_tree = NULL;
1002 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1003 struct btrfs_fs_info *fs_info)
1005 struct btrfs_root *root;
1006 struct btrfs_root *tree_root = fs_info->tree_root;
1007 struct extent_buffer *leaf;
1009 root = kzalloc(sizeof(*root), GFP_NOFS);
1011 return ERR_PTR(-ENOMEM);
1013 __setup_root(tree_root->nodesize, tree_root->leafsize,
1014 tree_root->sectorsize, tree_root->stripesize,
1015 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1017 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1018 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1019 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1021 * log trees do not get reference counted because they go away
1022 * before a real commit is actually done. They do store pointers
1023 * to file data extents, and those reference counts still get
1024 * updated (along with back refs to the log tree).
1028 leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1029 BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1032 return ERR_CAST(leaf);
1035 memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1036 btrfs_set_header_bytenr(leaf, leaf->start);
1037 btrfs_set_header_generation(leaf, trans->transid);
1038 btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1039 btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1042 write_extent_buffer(root->node, root->fs_info->fsid,
1043 (unsigned long)btrfs_header_fsid(root->node),
1045 btrfs_mark_buffer_dirty(root->node);
1046 btrfs_tree_unlock(root->node);
1050 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1051 struct btrfs_fs_info *fs_info)
1053 struct btrfs_root *log_root;
1055 log_root = alloc_log_tree(trans, fs_info);
1056 if (IS_ERR(log_root))
1057 return PTR_ERR(log_root);
1058 WARN_ON(fs_info->log_root_tree);
1059 fs_info->log_root_tree = log_root;
1063 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1064 struct btrfs_root *root)
1066 struct btrfs_root *log_root;
1067 struct btrfs_inode_item *inode_item;
1069 log_root = alloc_log_tree(trans, root->fs_info);
1070 if (IS_ERR(log_root))
1071 return PTR_ERR(log_root);
1073 log_root->last_trans = trans->transid;
1074 log_root->root_key.offset = root->root_key.objectid;
1076 inode_item = &log_root->root_item.inode;
1077 inode_item->generation = cpu_to_le64(1);
1078 inode_item->size = cpu_to_le64(3);
1079 inode_item->nlink = cpu_to_le32(1);
1080 inode_item->nbytes = cpu_to_le64(root->leafsize);
1081 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1083 btrfs_set_root_node(&log_root->root_item, log_root->node);
1085 WARN_ON(root->log_root);
1086 root->log_root = log_root;
1087 root->log_transid = 0;
1091 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1092 struct btrfs_key *location)
1094 struct btrfs_root *root;
1095 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1096 struct btrfs_path *path;
1097 struct extent_buffer *l;
1103 root = kzalloc(sizeof(*root), GFP_NOFS);
1105 return ERR_PTR(-ENOMEM);
1106 if (location->offset == (u64)-1) {
1107 ret = find_and_setup_root(tree_root, fs_info,
1108 location->objectid, root);
1111 return ERR_PTR(ret);
1116 __setup_root(tree_root->nodesize, tree_root->leafsize,
1117 tree_root->sectorsize, tree_root->stripesize,
1118 root, fs_info, location->objectid);
1120 path = btrfs_alloc_path();
1122 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1129 read_extent_buffer(l, &root->root_item,
1130 btrfs_item_ptr_offset(l, path->slots[0]),
1131 sizeof(root->root_item));
1132 memcpy(&root->root_key, location, sizeof(*location));
1135 btrfs_release_path(root, path);
1136 btrfs_free_path(path);
1139 return ERR_PTR(ret);
1141 generation = btrfs_root_generation(&root->root_item);
1142 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1143 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1144 blocksize, generation);
1145 root->commit_root = btrfs_root_node(root);
1146 BUG_ON(!root->node);
1148 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1150 ret = btrfs_find_highest_inode(root, &highest_inode);
1152 root->highest_inode = highest_inode;
1153 root->last_inode_alloc = highest_inode;
1159 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1162 struct btrfs_root *root;
1164 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1165 return fs_info->tree_root;
1166 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1167 return fs_info->extent_root;
1169 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1170 (unsigned long)root_objectid);
1174 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1175 struct btrfs_key *location)
1177 struct btrfs_root *root;
1180 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1181 return fs_info->tree_root;
1182 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1183 return fs_info->extent_root;
1184 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1185 return fs_info->chunk_root;
1186 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1187 return fs_info->dev_root;
1188 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1189 return fs_info->csum_root;
1191 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1192 (unsigned long)location->objectid);
1196 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1200 set_anon_super(&root->anon_super, NULL);
1202 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1203 (unsigned long)root->root_key.objectid,
1206 free_extent_buffer(root->node);
1208 return ERR_PTR(ret);
1210 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1211 ret = btrfs_find_dead_roots(fs_info->tree_root,
1212 root->root_key.objectid);
1214 btrfs_orphan_cleanup(root);
1219 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1220 struct btrfs_key *location,
1221 const char *name, int namelen)
1223 struct btrfs_root *root;
1226 root = btrfs_read_fs_root_no_name(fs_info, location);
1233 ret = btrfs_set_root_name(root, name, namelen);
1235 free_extent_buffer(root->node);
1237 return ERR_PTR(ret);
1240 ret = btrfs_sysfs_add_root(root);
1242 free_extent_buffer(root->node);
1245 return ERR_PTR(ret);
1252 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1254 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1256 struct btrfs_device *device;
1257 struct backing_dev_info *bdi;
1259 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1262 bdi = blk_get_backing_dev_info(device->bdev);
1263 if (bdi && bdi_congested(bdi, bdi_bits)) {
1272 * this unplugs every device on the box, and it is only used when page
1275 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1277 struct btrfs_device *device;
1278 struct btrfs_fs_info *info;
1280 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1281 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1285 bdi = blk_get_backing_dev_info(device->bdev);
1286 if (bdi->unplug_io_fn)
1287 bdi->unplug_io_fn(bdi, page);
1291 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1293 struct inode *inode;
1294 struct extent_map_tree *em_tree;
1295 struct extent_map *em;
1296 struct address_space *mapping;
1299 /* the generic O_DIRECT read code does this */
1301 __unplug_io_fn(bdi, page);
1306 * page->mapping may change at any time. Get a consistent copy
1307 * and use that for everything below
1310 mapping = page->mapping;
1314 inode = mapping->host;
1317 * don't do the expensive searching for a small number of
1320 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1321 __unplug_io_fn(bdi, page);
1325 offset = page_offset(page);
1327 em_tree = &BTRFS_I(inode)->extent_tree;
1328 read_lock(&em_tree->lock);
1329 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1330 read_unlock(&em_tree->lock);
1332 __unplug_io_fn(bdi, page);
1336 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1337 free_extent_map(em);
1338 __unplug_io_fn(bdi, page);
1341 offset = offset - em->start;
1342 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1343 em->block_start + offset, page);
1344 free_extent_map(em);
1348 * If this fails, caller must call bdi_destroy() to get rid of the
1351 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1355 bdi->capabilities = BDI_CAP_MAP_COPY;
1356 err = bdi_init(bdi);
1360 err = bdi_register(bdi, NULL, "btrfs-%d",
1361 atomic_inc_return(&btrfs_bdi_num));
1365 bdi->ra_pages = default_backing_dev_info.ra_pages;
1366 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1367 bdi->unplug_io_data = info;
1368 bdi->congested_fn = btrfs_congested_fn;
1369 bdi->congested_data = info;
1373 static int bio_ready_for_csum(struct bio *bio)
1379 struct extent_io_tree *io_tree = NULL;
1380 struct btrfs_fs_info *info = NULL;
1381 struct bio_vec *bvec;
1385 bio_for_each_segment(bvec, bio, i) {
1386 page = bvec->bv_page;
1387 if (page->private == EXTENT_PAGE_PRIVATE) {
1388 length += bvec->bv_len;
1391 if (!page->private) {
1392 length += bvec->bv_len;
1395 length = bvec->bv_len;
1396 buf_len = page->private >> 2;
1397 start = page_offset(page) + bvec->bv_offset;
1398 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1399 info = BTRFS_I(page->mapping->host)->root->fs_info;
1401 /* are we fully contained in this bio? */
1402 if (buf_len <= length)
1405 ret = extent_range_uptodate(io_tree, start + length,
1406 start + buf_len - 1);
1411 * called by the kthread helper functions to finally call the bio end_io
1412 * functions. This is where read checksum verification actually happens
1414 static void end_workqueue_fn(struct btrfs_work *work)
1417 struct end_io_wq *end_io_wq;
1418 struct btrfs_fs_info *fs_info;
1421 end_io_wq = container_of(work, struct end_io_wq, work);
1422 bio = end_io_wq->bio;
1423 fs_info = end_io_wq->info;
1425 /* metadata bio reads are special because the whole tree block must
1426 * be checksummed at once. This makes sure the entire block is in
1427 * ram and up to date before trying to verify things. For
1428 * blocksize <= pagesize, it is basically a noop
1430 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1431 !bio_ready_for_csum(bio)) {
1432 btrfs_queue_worker(&fs_info->endio_meta_workers,
1436 error = end_io_wq->error;
1437 bio->bi_private = end_io_wq->private;
1438 bio->bi_end_io = end_io_wq->end_io;
1440 bio_endio(bio, error);
1443 static int cleaner_kthread(void *arg)
1445 struct btrfs_root *root = arg;
1449 if (root->fs_info->closing)
1452 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1453 mutex_lock(&root->fs_info->cleaner_mutex);
1454 btrfs_clean_old_snapshots(root);
1455 mutex_unlock(&root->fs_info->cleaner_mutex);
1457 if (freezing(current)) {
1461 if (root->fs_info->closing)
1463 set_current_state(TASK_INTERRUPTIBLE);
1465 __set_current_state(TASK_RUNNING);
1467 } while (!kthread_should_stop());
1471 static int transaction_kthread(void *arg)
1473 struct btrfs_root *root = arg;
1474 struct btrfs_trans_handle *trans;
1475 struct btrfs_transaction *cur;
1477 unsigned long delay;
1482 if (root->fs_info->closing)
1486 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1487 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1489 mutex_lock(&root->fs_info->trans_mutex);
1490 cur = root->fs_info->running_transaction;
1492 mutex_unlock(&root->fs_info->trans_mutex);
1496 now = get_seconds();
1497 if (now < cur->start_time || now - cur->start_time < 30) {
1498 mutex_unlock(&root->fs_info->trans_mutex);
1502 mutex_unlock(&root->fs_info->trans_mutex);
1503 trans = btrfs_start_transaction(root, 1);
1504 ret = btrfs_commit_transaction(trans, root);
1507 wake_up_process(root->fs_info->cleaner_kthread);
1508 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1510 if (freezing(current)) {
1513 if (root->fs_info->closing)
1515 set_current_state(TASK_INTERRUPTIBLE);
1516 schedule_timeout(delay);
1517 __set_current_state(TASK_RUNNING);
1519 } while (!kthread_should_stop());
1523 struct btrfs_root *open_ctree(struct super_block *sb,
1524 struct btrfs_fs_devices *fs_devices,
1534 struct btrfs_key location;
1535 struct buffer_head *bh;
1536 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1538 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1540 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1542 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1544 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1546 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1548 struct btrfs_root *log_tree_root;
1553 struct btrfs_super_block *disk_super;
1555 if (!extent_root || !tree_root || !fs_info ||
1556 !chunk_root || !dev_root || !csum_root) {
1560 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1561 INIT_LIST_HEAD(&fs_info->trans_list);
1562 INIT_LIST_HEAD(&fs_info->dead_roots);
1563 INIT_LIST_HEAD(&fs_info->hashers);
1564 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1565 INIT_LIST_HEAD(&fs_info->ordered_operations);
1566 INIT_LIST_HEAD(&fs_info->caching_block_groups);
1567 spin_lock_init(&fs_info->delalloc_lock);
1568 spin_lock_init(&fs_info->new_trans_lock);
1569 spin_lock_init(&fs_info->ref_cache_lock);
1571 init_completion(&fs_info->kobj_unregister);
1572 fs_info->tree_root = tree_root;
1573 fs_info->extent_root = extent_root;
1574 fs_info->csum_root = csum_root;
1575 fs_info->chunk_root = chunk_root;
1576 fs_info->dev_root = dev_root;
1577 fs_info->fs_devices = fs_devices;
1578 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1579 INIT_LIST_HEAD(&fs_info->space_info);
1580 btrfs_mapping_init(&fs_info->mapping_tree);
1581 atomic_set(&fs_info->nr_async_submits, 0);
1582 atomic_set(&fs_info->async_delalloc_pages, 0);
1583 atomic_set(&fs_info->async_submit_draining, 0);
1584 atomic_set(&fs_info->nr_async_bios, 0);
1586 fs_info->max_extent = (u64)-1;
1587 fs_info->max_inline = 8192 * 1024;
1588 if (setup_bdi(fs_info, &fs_info->bdi))
1590 fs_info->btree_inode = new_inode(sb);
1591 fs_info->btree_inode->i_ino = 1;
1592 fs_info->btree_inode->i_nlink = 1;
1593 fs_info->metadata_ratio = 8;
1595 fs_info->thread_pool_size = min_t(unsigned long,
1596 num_online_cpus() + 2, 8);
1598 INIT_LIST_HEAD(&fs_info->ordered_extents);
1599 spin_lock_init(&fs_info->ordered_extent_lock);
1601 sb->s_blocksize = 4096;
1602 sb->s_blocksize_bits = blksize_bits(4096);
1605 * we set the i_size on the btree inode to the max possible int.
1606 * the real end of the address space is determined by all of
1607 * the devices in the system
1609 fs_info->btree_inode->i_size = OFFSET_MAX;
1610 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1611 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1613 RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1614 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1615 fs_info->btree_inode->i_mapping,
1617 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1620 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1622 spin_lock_init(&fs_info->block_group_cache_lock);
1623 fs_info->block_group_cache_tree.rb_node = NULL;
1625 extent_io_tree_init(&fs_info->freed_extents[0],
1626 fs_info->btree_inode->i_mapping, GFP_NOFS);
1627 extent_io_tree_init(&fs_info->freed_extents[1],
1628 fs_info->btree_inode->i_mapping, GFP_NOFS);
1629 fs_info->pinned_extents = &fs_info->freed_extents[0];
1630 fs_info->do_barriers = 1;
1632 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1633 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1634 sizeof(struct btrfs_key));
1635 insert_inode_hash(fs_info->btree_inode);
1637 mutex_init(&fs_info->trans_mutex);
1638 mutex_init(&fs_info->ordered_operations_mutex);
1639 mutex_init(&fs_info->tree_log_mutex);
1640 mutex_init(&fs_info->drop_mutex);
1641 mutex_init(&fs_info->chunk_mutex);
1642 mutex_init(&fs_info->transaction_kthread_mutex);
1643 mutex_init(&fs_info->cleaner_mutex);
1644 mutex_init(&fs_info->volume_mutex);
1645 mutex_init(&fs_info->tree_reloc_mutex);
1646 init_rwsem(&fs_info->extent_commit_sem);
1648 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1649 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1651 init_waitqueue_head(&fs_info->transaction_throttle);
1652 init_waitqueue_head(&fs_info->transaction_wait);
1653 init_waitqueue_head(&fs_info->async_submit_wait);
1655 __setup_root(4096, 4096, 4096, 4096, tree_root,
1656 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1659 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1663 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1664 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1665 sizeof(fs_info->super_for_commit));
1668 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1670 disk_super = &fs_info->super_copy;
1671 if (!btrfs_super_root(disk_super))
1674 ret = btrfs_parse_options(tree_root, options);
1680 features = btrfs_super_incompat_flags(disk_super) &
1681 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1683 printk(KERN_ERR "BTRFS: couldn't mount because of "
1684 "unsupported optional features (%Lx).\n",
1685 (unsigned long long)features);
1690 features = btrfs_super_incompat_flags(disk_super);
1691 if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1692 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1693 btrfs_set_super_incompat_flags(disk_super, features);
1696 features = btrfs_super_compat_ro_flags(disk_super) &
1697 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1698 if (!(sb->s_flags & MS_RDONLY) && features) {
1699 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1700 "unsupported option features (%Lx).\n",
1701 (unsigned long long)features);
1705 printk("thread pool is %d\n", fs_info->thread_pool_size);
1707 * we need to start all the end_io workers up front because the
1708 * queue work function gets called at interrupt time, and so it
1709 * cannot dynamically grow.
1711 btrfs_init_workers(&fs_info->workers, "worker",
1712 fs_info->thread_pool_size);
1714 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1715 fs_info->thread_pool_size);
1717 btrfs_init_workers(&fs_info->submit_workers, "submit",
1718 min_t(u64, fs_devices->num_devices,
1719 fs_info->thread_pool_size));
1721 /* a higher idle thresh on the submit workers makes it much more
1722 * likely that bios will be send down in a sane order to the
1725 fs_info->submit_workers.idle_thresh = 64;
1727 fs_info->workers.idle_thresh = 16;
1728 fs_info->workers.ordered = 1;
1730 fs_info->delalloc_workers.idle_thresh = 2;
1731 fs_info->delalloc_workers.ordered = 1;
1733 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1734 btrfs_init_workers(&fs_info->endio_workers, "endio",
1735 fs_info->thread_pool_size);
1736 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1737 fs_info->thread_pool_size);
1738 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1739 "endio-meta-write", fs_info->thread_pool_size);
1740 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1741 fs_info->thread_pool_size);
1744 * endios are largely parallel and should have a very
1747 fs_info->endio_workers.idle_thresh = 4;
1748 fs_info->endio_meta_workers.idle_thresh = 4;
1750 fs_info->endio_write_workers.idle_thresh = 2;
1751 fs_info->endio_meta_write_workers.idle_thresh = 2;
1753 fs_info->endio_workers.atomic_worker_start = 1;
1754 fs_info->endio_meta_workers.atomic_worker_start = 1;
1755 fs_info->endio_write_workers.atomic_worker_start = 1;
1756 fs_info->endio_meta_write_workers.atomic_worker_start = 1;
1758 btrfs_start_workers(&fs_info->workers, 1);
1759 btrfs_start_workers(&fs_info->submit_workers, 1);
1760 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1761 btrfs_start_workers(&fs_info->fixup_workers, 1);
1762 btrfs_start_workers(&fs_info->endio_workers, 1);
1763 btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1764 btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1765 btrfs_start_workers(&fs_info->endio_write_workers, 1);
1767 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1768 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1769 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1771 nodesize = btrfs_super_nodesize(disk_super);
1772 leafsize = btrfs_super_leafsize(disk_super);
1773 sectorsize = btrfs_super_sectorsize(disk_super);
1774 stripesize = btrfs_super_stripesize(disk_super);
1775 tree_root->nodesize = nodesize;
1776 tree_root->leafsize = leafsize;
1777 tree_root->sectorsize = sectorsize;
1778 tree_root->stripesize = stripesize;
1780 sb->s_blocksize = sectorsize;
1781 sb->s_blocksize_bits = blksize_bits(sectorsize);
1783 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1784 sizeof(disk_super->magic))) {
1785 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1786 goto fail_sb_buffer;
1789 mutex_lock(&fs_info->chunk_mutex);
1790 ret = btrfs_read_sys_array(tree_root);
1791 mutex_unlock(&fs_info->chunk_mutex);
1793 printk(KERN_WARNING "btrfs: failed to read the system "
1794 "array on %s\n", sb->s_id);
1795 goto fail_sb_buffer;
1798 blocksize = btrfs_level_size(tree_root,
1799 btrfs_super_chunk_root_level(disk_super));
1800 generation = btrfs_super_chunk_root_generation(disk_super);
1802 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1803 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1805 chunk_root->node = read_tree_block(chunk_root,
1806 btrfs_super_chunk_root(disk_super),
1807 blocksize, generation);
1808 BUG_ON(!chunk_root->node);
1809 if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1810 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1812 goto fail_chunk_root;
1814 btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1815 chunk_root->commit_root = btrfs_root_node(chunk_root);
1817 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1818 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1821 mutex_lock(&fs_info->chunk_mutex);
1822 ret = btrfs_read_chunk_tree(chunk_root);
1823 mutex_unlock(&fs_info->chunk_mutex);
1825 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1827 goto fail_chunk_root;
1830 btrfs_close_extra_devices(fs_devices);
1832 blocksize = btrfs_level_size(tree_root,
1833 btrfs_super_root_level(disk_super));
1834 generation = btrfs_super_generation(disk_super);
1836 tree_root->node = read_tree_block(tree_root,
1837 btrfs_super_root(disk_super),
1838 blocksize, generation);
1839 if (!tree_root->node)
1840 goto fail_chunk_root;
1841 if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1842 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1844 goto fail_tree_root;
1846 btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1847 tree_root->commit_root = btrfs_root_node(tree_root);
1849 ret = find_and_setup_root(tree_root, fs_info,
1850 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1852 goto fail_tree_root;
1853 extent_root->track_dirty = 1;
1855 ret = find_and_setup_root(tree_root, fs_info,
1856 BTRFS_DEV_TREE_OBJECTID, dev_root);
1858 goto fail_extent_root;
1859 dev_root->track_dirty = 1;
1861 ret = find_and_setup_root(tree_root, fs_info,
1862 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1866 csum_root->track_dirty = 1;
1868 btrfs_read_block_groups(extent_root);
1870 fs_info->generation = generation;
1871 fs_info->last_trans_committed = generation;
1872 fs_info->data_alloc_profile = (u64)-1;
1873 fs_info->metadata_alloc_profile = (u64)-1;
1874 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1875 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1877 if (IS_ERR(fs_info->cleaner_kthread))
1878 goto fail_csum_root;
1880 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1882 "btrfs-transaction");
1883 if (IS_ERR(fs_info->transaction_kthread))
1886 if (!btrfs_test_opt(tree_root, SSD) &&
1887 !btrfs_test_opt(tree_root, NOSSD) &&
1888 !fs_info->fs_devices->rotating) {
1889 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1891 btrfs_set_opt(fs_info->mount_opt, SSD);
1894 if (btrfs_super_log_root(disk_super) != 0) {
1895 u64 bytenr = btrfs_super_log_root(disk_super);
1897 if (fs_devices->rw_devices == 0) {
1898 printk(KERN_WARNING "Btrfs log replay required "
1901 goto fail_trans_kthread;
1904 btrfs_level_size(tree_root,
1905 btrfs_super_log_root_level(disk_super));
1907 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1910 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1911 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1913 log_tree_root->node = read_tree_block(tree_root, bytenr,
1916 ret = btrfs_recover_log_trees(log_tree_root);
1919 if (sb->s_flags & MS_RDONLY) {
1920 ret = btrfs_commit_super(tree_root);
1925 if (!(sb->s_flags & MS_RDONLY)) {
1926 ret = btrfs_recover_relocation(tree_root);
1930 location.objectid = BTRFS_FS_TREE_OBJECTID;
1931 location.type = BTRFS_ROOT_ITEM_KEY;
1932 location.offset = (u64)-1;
1934 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1935 if (!fs_info->fs_root)
1936 goto fail_trans_kthread;
1941 kthread_stop(fs_info->transaction_kthread);
1943 kthread_stop(fs_info->cleaner_kthread);
1946 * make sure we're done with the btree inode before we stop our
1949 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1950 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1953 free_extent_buffer(csum_root->node);
1954 free_extent_buffer(csum_root->commit_root);
1956 free_extent_buffer(dev_root->node);
1957 free_extent_buffer(dev_root->commit_root);
1959 free_extent_buffer(extent_root->node);
1960 free_extent_buffer(extent_root->commit_root);
1962 free_extent_buffer(tree_root->node);
1963 free_extent_buffer(tree_root->commit_root);
1965 free_extent_buffer(chunk_root->node);
1966 free_extent_buffer(chunk_root->commit_root);
1968 btrfs_stop_workers(&fs_info->fixup_workers);
1969 btrfs_stop_workers(&fs_info->delalloc_workers);
1970 btrfs_stop_workers(&fs_info->workers);
1971 btrfs_stop_workers(&fs_info->endio_workers);
1972 btrfs_stop_workers(&fs_info->endio_meta_workers);
1973 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1974 btrfs_stop_workers(&fs_info->endio_write_workers);
1975 btrfs_stop_workers(&fs_info->submit_workers);
1977 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1978 iput(fs_info->btree_inode);
1980 btrfs_close_devices(fs_info->fs_devices);
1981 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1983 bdi_destroy(&fs_info->bdi);
1991 return ERR_PTR(err);
1994 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1996 char b[BDEVNAME_SIZE];
1999 set_buffer_uptodate(bh);
2001 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
2002 printk(KERN_WARNING "lost page write due to "
2003 "I/O error on %s\n",
2004 bdevname(bh->b_bdev, b));
2006 /* note, we dont' set_buffer_write_io_error because we have
2007 * our own ways of dealing with the IO errors
2009 clear_buffer_uptodate(bh);
2015 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2017 struct buffer_head *bh;
2018 struct buffer_head *latest = NULL;
2019 struct btrfs_super_block *super;
2024 /* we would like to check all the supers, but that would make
2025 * a btrfs mount succeed after a mkfs from a different FS.
2026 * So, we need to add a special mount option to scan for
2027 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2029 for (i = 0; i < 1; i++) {
2030 bytenr = btrfs_sb_offset(i);
2031 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2033 bh = __bread(bdev, bytenr / 4096, 4096);
2037 super = (struct btrfs_super_block *)bh->b_data;
2038 if (btrfs_super_bytenr(super) != bytenr ||
2039 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2040 sizeof(super->magic))) {
2045 if (!latest || btrfs_super_generation(super) > transid) {
2048 transid = btrfs_super_generation(super);
2057 * this should be called twice, once with wait == 0 and
2058 * once with wait == 1. When wait == 0 is done, all the buffer heads
2059 * we write are pinned.
2061 * They are released when wait == 1 is done.
2062 * max_mirrors must be the same for both runs, and it indicates how
2063 * many supers on this one device should be written.
2065 * max_mirrors == 0 means to write them all.
2067 static int write_dev_supers(struct btrfs_device *device,
2068 struct btrfs_super_block *sb,
2069 int do_barriers, int wait, int max_mirrors)
2071 struct buffer_head *bh;
2077 int last_barrier = 0;
2079 if (max_mirrors == 0)
2080 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2082 /* make sure only the last submit_bh does a barrier */
2084 for (i = 0; i < max_mirrors; i++) {
2085 bytenr = btrfs_sb_offset(i);
2086 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2087 device->total_bytes)
2093 for (i = 0; i < max_mirrors; i++) {
2094 bytenr = btrfs_sb_offset(i);
2095 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2099 bh = __find_get_block(device->bdev, bytenr / 4096,
2100 BTRFS_SUPER_INFO_SIZE);
2103 if (!buffer_uptodate(bh))
2106 /* drop our reference */
2109 /* drop the reference from the wait == 0 run */
2113 btrfs_set_super_bytenr(sb, bytenr);
2116 crc = btrfs_csum_data(NULL, (char *)sb +
2117 BTRFS_CSUM_SIZE, crc,
2118 BTRFS_SUPER_INFO_SIZE -
2120 btrfs_csum_final(crc, sb->csum);
2123 * one reference for us, and we leave it for the
2126 bh = __getblk(device->bdev, bytenr / 4096,
2127 BTRFS_SUPER_INFO_SIZE);
2128 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2130 /* one reference for submit_bh */
2133 set_buffer_uptodate(bh);
2135 bh->b_end_io = btrfs_end_buffer_write_sync;
2138 if (i == last_barrier && do_barriers && device->barriers) {
2139 ret = submit_bh(WRITE_BARRIER, bh);
2140 if (ret == -EOPNOTSUPP) {
2141 printk("btrfs: disabling barriers on dev %s\n",
2143 set_buffer_uptodate(bh);
2144 device->barriers = 0;
2145 /* one reference for submit_bh */
2148 ret = submit_bh(WRITE_SYNC, bh);
2151 ret = submit_bh(WRITE_SYNC, bh);
2157 return errors < i ? 0 : -1;
2160 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2162 struct list_head *head;
2163 struct btrfs_device *dev;
2164 struct btrfs_super_block *sb;
2165 struct btrfs_dev_item *dev_item;
2169 int total_errors = 0;
2172 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2173 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2175 sb = &root->fs_info->super_for_commit;
2176 dev_item = &sb->dev_item;
2178 mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2179 head = &root->fs_info->fs_devices->devices;
2180 list_for_each_entry(dev, head, dev_list) {
2185 if (!dev->in_fs_metadata || !dev->writeable)
2188 btrfs_set_stack_device_generation(dev_item, 0);
2189 btrfs_set_stack_device_type(dev_item, dev->type);
2190 btrfs_set_stack_device_id(dev_item, dev->devid);
2191 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2192 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2193 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2194 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2195 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2196 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2197 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2199 flags = btrfs_super_flags(sb);
2200 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2202 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2206 if (total_errors > max_errors) {
2207 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2213 list_for_each_entry(dev, head, dev_list) {
2216 if (!dev->in_fs_metadata || !dev->writeable)
2219 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2223 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2224 if (total_errors > max_errors) {
2225 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2232 int write_ctree_super(struct btrfs_trans_handle *trans,
2233 struct btrfs_root *root, int max_mirrors)
2237 ret = write_all_supers(root, max_mirrors);
2241 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2243 WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2244 radix_tree_delete(&fs_info->fs_roots_radix,
2245 (unsigned long)root->root_key.objectid);
2246 if (root->anon_super.s_dev) {
2247 down_write(&root->anon_super.s_umount);
2248 kill_anon_super(&root->anon_super);
2251 free_extent_buffer(root->node);
2252 if (root->commit_root)
2253 free_extent_buffer(root->commit_root);
2259 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2262 struct btrfs_root *gang[8];
2266 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2271 for (i = 0; i < ret; i++)
2272 btrfs_free_fs_root(fs_info, gang[i]);
2277 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2279 u64 root_objectid = 0;
2280 struct btrfs_root *gang[8];
2285 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2286 (void **)gang, root_objectid,
2291 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2292 for (i = 0; i < ret; i++) {
2293 root_objectid = gang[i]->root_key.objectid;
2294 ret = btrfs_find_dead_roots(fs_info->tree_root,
2297 btrfs_orphan_cleanup(gang[i]);
2304 int btrfs_commit_super(struct btrfs_root *root)
2306 struct btrfs_trans_handle *trans;
2309 mutex_lock(&root->fs_info->cleaner_mutex);
2310 btrfs_clean_old_snapshots(root);
2311 mutex_unlock(&root->fs_info->cleaner_mutex);
2312 trans = btrfs_start_transaction(root, 1);
2313 ret = btrfs_commit_transaction(trans, root);
2315 /* run commit again to drop the original snapshot */
2316 trans = btrfs_start_transaction(root, 1);
2317 btrfs_commit_transaction(trans, root);
2318 ret = btrfs_write_and_wait_transaction(NULL, root);
2321 ret = write_ctree_super(NULL, root, 0);
2325 int close_ctree(struct btrfs_root *root)
2327 struct btrfs_fs_info *fs_info = root->fs_info;
2330 fs_info->closing = 1;
2333 kthread_stop(root->fs_info->transaction_kthread);
2334 kthread_stop(root->fs_info->cleaner_kthread);
2336 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2337 ret = btrfs_commit_super(root);
2339 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2342 fs_info->closing = 2;
2345 if (fs_info->delalloc_bytes) {
2346 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2347 (unsigned long long)fs_info->delalloc_bytes);
2349 if (fs_info->total_ref_cache_size) {
2350 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2351 (unsigned long long)fs_info->total_ref_cache_size);
2354 free_extent_buffer(fs_info->extent_root->node);
2355 free_extent_buffer(fs_info->extent_root->commit_root);
2356 free_extent_buffer(fs_info->tree_root->node);
2357 free_extent_buffer(fs_info->tree_root->commit_root);
2358 free_extent_buffer(root->fs_info->chunk_root->node);
2359 free_extent_buffer(root->fs_info->chunk_root->commit_root);
2360 free_extent_buffer(root->fs_info->dev_root->node);
2361 free_extent_buffer(root->fs_info->dev_root->commit_root);
2362 free_extent_buffer(root->fs_info->csum_root->node);
2363 free_extent_buffer(root->fs_info->csum_root->commit_root);
2365 btrfs_free_block_groups(root->fs_info);
2367 del_fs_roots(fs_info);
2369 iput(fs_info->btree_inode);
2371 btrfs_stop_workers(&fs_info->fixup_workers);
2372 btrfs_stop_workers(&fs_info->delalloc_workers);
2373 btrfs_stop_workers(&fs_info->workers);
2374 btrfs_stop_workers(&fs_info->endio_workers);
2375 btrfs_stop_workers(&fs_info->endio_meta_workers);
2376 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2377 btrfs_stop_workers(&fs_info->endio_write_workers);
2378 btrfs_stop_workers(&fs_info->submit_workers);
2380 btrfs_close_devices(fs_info->fs_devices);
2381 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2383 bdi_destroy(&fs_info->bdi);
2385 kfree(fs_info->extent_root);
2386 kfree(fs_info->tree_root);
2387 kfree(fs_info->chunk_root);
2388 kfree(fs_info->dev_root);
2389 kfree(fs_info->csum_root);
2393 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2396 struct inode *btree_inode = buf->first_page->mapping->host;
2398 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2402 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2407 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2409 struct inode *btree_inode = buf->first_page->mapping->host;
2410 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2414 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2416 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2417 u64 transid = btrfs_header_generation(buf);
2418 struct inode *btree_inode = root->fs_info->btree_inode;
2421 btrfs_assert_tree_locked(buf);
2422 if (transid != root->fs_info->generation) {
2423 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2424 "found %llu running %llu\n",
2425 (unsigned long long)buf->start,
2426 (unsigned long long)transid,
2427 (unsigned long long)root->fs_info->generation);
2430 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2433 spin_lock(&root->fs_info->delalloc_lock);
2434 root->fs_info->dirty_metadata_bytes += buf->len;
2435 spin_unlock(&root->fs_info->delalloc_lock);
2439 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2442 * looks as though older kernels can get into trouble with
2443 * this code, they end up stuck in balance_dirty_pages forever
2446 unsigned long thresh = 32 * 1024 * 1024;
2448 if (current->flags & PF_MEMALLOC)
2451 num_dirty = root->fs_info->dirty_metadata_bytes;
2453 if (num_dirty > thresh) {
2454 balance_dirty_pages_ratelimited_nr(
2455 root->fs_info->btree_inode->i_mapping, 1);
2460 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2462 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2464 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2466 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2470 int btree_lock_page_hook(struct page *page)
2472 struct inode *inode = page->mapping->host;
2473 struct btrfs_root *root = BTRFS_I(inode)->root;
2474 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2475 struct extent_buffer *eb;
2477 u64 bytenr = page_offset(page);
2479 if (page->private == EXTENT_PAGE_PRIVATE)
2482 len = page->private >> 2;
2483 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2487 btrfs_tree_lock(eb);
2488 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2490 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2491 spin_lock(&root->fs_info->delalloc_lock);
2492 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2493 root->fs_info->dirty_metadata_bytes -= eb->len;
2496 spin_unlock(&root->fs_info->delalloc_lock);
2499 btrfs_tree_unlock(eb);
2500 free_extent_buffer(eb);
2506 static struct extent_io_ops btree_extent_io_ops = {
2507 .write_cache_pages_lock_hook = btree_lock_page_hook,
2508 .readpage_end_io_hook = btree_readpage_end_io_hook,
2509 .submit_bio_hook = btree_submit_bio_hook,
2510 /* note we're sharing with inode.c for the merge bio hook */
2511 .merge_bio_hook = btrfs_merge_bio_hook,