Btrfs: use percpu counter for dirty metadata count
->dirty_metadata_bytes is accessed very frequently, so use percpu
counter instead of the u64 variant to reduce the contention of
the lock.
This patch also fixed the problem that we access it without
lock protection in __btrfs_btree_balance_dirty(), which may
cause we skip the dirty pages flush.
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Josef Bacik <jbacik@fusionio.com>
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 04f98e3..34ace16 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -946,18 +946,20 @@
struct writeback_control *wbc)
{
struct extent_io_tree *tree;
+ struct btrfs_fs_info *fs_info;
+ int ret;
+
tree = &BTRFS_I(mapping->host)->io_tree;
if (wbc->sync_mode == WB_SYNC_NONE) {
- struct btrfs_root *root = BTRFS_I(mapping->host)->root;
- u64 num_dirty;
- unsigned long thresh = 32 * 1024 * 1024;
if (wbc->for_kupdate)
return 0;
+ fs_info = BTRFS_I(mapping->host)->root->fs_info;
/* this is a bit racy, but that's ok */
- num_dirty = root->fs_info->dirty_metadata_bytes;
- if (num_dirty < thresh)
+ ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
+ BTRFS_DIRTY_METADATA_THRESH);
+ if (ret < 0)
return 0;
}
return btree_write_cache_pages(mapping, wbc);
@@ -1125,24 +1127,16 @@
void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
struct extent_buffer *buf)
{
+ struct btrfs_fs_info *fs_info = root->fs_info;
+
if (btrfs_header_generation(buf) ==
- root->fs_info->running_transaction->transid) {
+ fs_info->running_transaction->transid) {
btrfs_assert_tree_locked(buf);
if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
- spin_lock(&root->fs_info->delalloc_lock);
- if (root->fs_info->dirty_metadata_bytes >= buf->len)
- root->fs_info->dirty_metadata_bytes -= buf->len;
- else {
- spin_unlock(&root->fs_info->delalloc_lock);
- btrfs_panic(root->fs_info, -EOVERFLOW,
- "Can't clear %lu bytes from "
- " dirty_mdatadata_bytes (%llu)",
- buf->len,
- root->fs_info->dirty_metadata_bytes);
- }
- spin_unlock(&root->fs_info->delalloc_lock);
-
+ __percpu_counter_add(&fs_info->dirty_metadata_bytes,
+ -buf->len,
+ fs_info->dirty_metadata_batch);
/* ugh, clear_extent_buffer_dirty needs to lock the page */
btrfs_set_lock_blocking(buf);
clear_extent_buffer_dirty(buf);
@@ -2008,10 +2002,18 @@
goto fail_srcu;
}
+ ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
+ if (ret) {
+ err = ret;
+ goto fail_bdi;
+ }
+ fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
+ (1 + ilog2(nr_cpu_ids));
+
fs_info->btree_inode = new_inode(sb);
if (!fs_info->btree_inode) {
err = -ENOMEM;
- goto fail_bdi;
+ goto fail_dirty_metadata_bytes;
}
mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
@@ -2266,6 +2268,7 @@
leafsize = btrfs_super_leafsize(disk_super);
sectorsize = btrfs_super_sectorsize(disk_super);
stripesize = btrfs_super_stripesize(disk_super);
+ fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
/*
* mixed block groups end up with duplicate but slightly offset
@@ -2728,6 +2731,8 @@
invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
iput(fs_info->btree_inode);
+fail_dirty_metadata_bytes:
+ percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
fail_bdi:
bdi_destroy(&fs_info->bdi);
fail_srcu:
@@ -3406,6 +3411,7 @@
btrfs_close_devices(fs_info->fs_devices);
btrfs_mapping_tree_free(&fs_info->mapping_tree);
+ percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
bdi_destroy(&fs_info->bdi);
cleanup_srcu_struct(&fs_info->subvol_srcu);
@@ -3448,11 +3454,10 @@
(unsigned long long)transid,
(unsigned long long)root->fs_info->generation);
was_dirty = set_extent_buffer_dirty(buf);
- if (!was_dirty) {
- spin_lock(&root->fs_info->delalloc_lock);
- root->fs_info->dirty_metadata_bytes += buf->len;
- spin_unlock(&root->fs_info->delalloc_lock);
- }
+ if (!was_dirty)
+ __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
+ buf->len,
+ root->fs_info->dirty_metadata_batch);
}
static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
@@ -3462,8 +3467,7 @@
* looks as though older kernels can get into trouble with
* this code, they end up stuck in balance_dirty_pages forever
*/
- u64 num_dirty;
- unsigned long thresh = 32 * 1024 * 1024;
+ int ret;
if (current->flags & PF_MEMALLOC)
return;
@@ -3471,9 +3475,9 @@
if (flush_delayed)
btrfs_balance_delayed_items(root);
- num_dirty = root->fs_info->dirty_metadata_bytes;
-
- if (num_dirty > thresh) {
+ ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
+ BTRFS_DIRTY_METADATA_THRESH);
+ if (ret > 0) {
balance_dirty_pages_ratelimited_nr(
root->fs_info->btree_inode->i_mapping, 1);
}
