Btrfs: inline csums if we're fsyncing

The tree logging stuff needs the csums to be on the ordered extents in order to log them properly, so mark that we're sync and inline the csum creation so we don't have to wait on the csumming to be done when logging extents that are still in flight. Thanks, Signed-off-by: Josef Bacik <jbacik@fusionio.com> Signed-off-by: Chris Mason <chris.mason@fusionio.com>

Btrfs: inline csums if we're fsyncing
The tree logging stuff needs the csums to be on the ordered extents in order to log them properly, so mark that we're sync and inline the csum creation so we don't have to wait on the csumming to be done when logging extents that are still in flight. Thanks, Signed-off-by: Josef Bacik <jbacik@fusionio.com> Signed-off-by: Chris Mason <chris.mason@fusionio.com>
Josef Bacik · Chris Mason
1 parent a95249b392
Showing 3 changed files with 21 additions and 1 deletions Side-by-side Diff
fs/btrfs/btrfs_inode.h
fs/btrfs/file.c
fs/btrfs/inode.c
@@ -91,6 +91,9 @@
  
 	unsigned long runtime_flags;
  
+	/* Keep track of who's O_SYNC/fsycing currently */
+	atomic_t sync_writers;
+
 	/* full 64 bit generation number, struct vfs_inode doesn't have a big
 	 * enough field for this.
 	 */
@@ -1472,6 +1472,7 @@
 	ssize_t num_written = 0;
 	ssize_t err = 0;
 	size_t count, ocount;
+	bool sync = (file->f_flags & O_DSYNC) || IS_SYNC(file->f_mapping->host);
  
 	sb_start_write(inode->i_sb);
  
@@ -1529,6 +1530,9 @@
 		}
 	}
  
+	if (sync)
+		atomic_inc(&BTRFS_I(inode)->sync_writers);
+
 	if (unlikely(file->f_flags & O_DIRECT)) {
 		num_written = __btrfs_direct_write(iocb, iov, nr_segs,
 						   pos, ppos, count, ocount);
@@ -1563,6 +1567,8 @@
 			num_written = err;
 	}
 out:
+	if (sync)
+		atomic_dec(&BTRFS_I(inode)->sync_writers);
 	sb_end_write(inode->i_sb);
 	current->backing_dev_info = NULL;
 	return num_written ? num_written : err;
  
@@ -1613,7 +1619,9 @@
 	 * out of the ->i_mutex. If so, we can flush the dirty pages by
 	 * multi-task, and make the performance up.
 	 */
+	atomic_inc(&BTRFS_I(inode)->sync_writers);
 	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
+	atomic_dec(&BTRFS_I(inode)->sync_writers);
 	if (ret)
 		return ret;
  
@@ -1622,6 +1622,7 @@
 	int ret = 0;
 	int skip_sum;
 	int metadata = 0;
+	int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
  
 	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
  
@@ -1644,7 +1645,7 @@
 				goto out;
 		}
 		goto mapit;
-	} else if (!skip_sum) {
+	} else if (async && !skip_sum) {
 		/* csum items have already been cloned */
 		if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
 			goto mapit;
@@ -1655,6 +1656,10 @@
 				   __btrfs_submit_bio_start,
 				   __btrfs_submit_bio_done);
 		goto out;
+	} else if (!skip_sum) {
+		ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
+		if (ret)
+			goto out;
 	}
  
 mapit:
@@ -6333,6 +6338,9 @@
 	struct btrfs_root *root = BTRFS_I(inode)->root;
 	int ret;
  
+	if (async_submit)
+		async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
+
 	bio_get(bio);
  
 	if (!write) {
@@ -7113,6 +7121,7 @@
 	extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
 	ei->io_tree.track_uptodate = 1;
 	ei->io_failure_tree.track_uptodate = 1;
+	atomic_set(&ei->sync_writers, 0);
 	mutex_init(&ei->log_mutex);
 	mutex_init(&ei->delalloc_mutex);
 	btrfs_ordered_inode_tree_init(&ei->ordered_tree);
...	...	@@ -91,6 +91,9 @@
91	91
92	92	unsigned long runtime_flags;
93	93
	94	+ /* Keep track of who's O_SYNC/fsycing currently */
	95	+ atomic_t sync_writers;
	96	+
94	97	/* full 64 bit generation number, struct vfs_inode doesn't have a big
95	98	* enough field for this.
96	99	*/
...	...	@@ -1472,6 +1472,7 @@
1472	1472	ssize_t num_written = 0;
1473	1473	ssize_t err = 0;
1474	1474	size_t count, ocount;
	1475	+ bool sync = (file->f_flags & O_DSYNC) \|\| IS_SYNC(file->f_mapping->host);
1475	1476
1476	1477	sb_start_write(inode->i_sb);
1477	1478
...	...	@@ -1529,6 +1530,9 @@
1529	1530	}
1530	1531	}
1531	1532
	1533	+ if (sync)
	1534	+ atomic_inc(&BTRFS_I(inode)->sync_writers);
	1535	+
1532	1536	if (unlikely(file->f_flags & O_DIRECT)) {
1533	1537	num_written = __btrfs_direct_write(iocb, iov, nr_segs,
1534	1538	pos, ppos, count, ocount);
...	...	@@ -1563,6 +1567,8 @@
1563	1567	num_written = err;
1564	1568	}
1565	1569	out:
	1570	+ if (sync)
	1571	+ atomic_dec(&BTRFS_I(inode)->sync_writers);
1566	1572	sb_end_write(inode->i_sb);
1567	1573	current->backing_dev_info = NULL;
1568	1574	return num_written ? num_written : err;
1569	1575
...	...	@@ -1613,7 +1619,9 @@
1613	1619	* out of the ->i_mutex. If so, we can flush the dirty pages by
1614	1620	* multi-task, and make the performance up.
1615	1621	*/
	1622	+ atomic_inc(&BTRFS_I(inode)->sync_writers);
1616	1623	ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
	1624	+ atomic_dec(&BTRFS_I(inode)->sync_writers);
1617	1625	if (ret)
1618	1626	return ret;
1619	1627
...	...	@@ -1622,6 +1622,7 @@
1622	1622	int ret = 0;
1623	1623	int skip_sum;
1624	1624	int metadata = 0;
	1625	+ int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1625	1626
1626	1627	skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1627	1628
...	...	@@ -1644,7 +1645,7 @@
1644	1645	goto out;
1645	1646	}
1646	1647	goto mapit;
1647		- } else if (!skip_sum) {
	1648	+ } else if (async && !skip_sum) {
1648	1649	/* csum items have already been cloned */
1649	1650	if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1650	1651	goto mapit;
...	...	@@ -1655,6 +1656,10 @@
1655	1656	__btrfs_submit_bio_start,
1656	1657	__btrfs_submit_bio_done);
1657	1658	goto out;
	1659	+ } else if (!skip_sum) {
	1660	+ ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
	1661	+ if (ret)
	1662	+ goto out;
1658	1663	}
1659	1664
1660	1665	mapit:
...	...	@@ -6333,6 +6338,9 @@
6333	6338	struct btrfs_root *root = BTRFS_I(inode)->root;
6334	6339	int ret;
6335	6340
	6341	+ if (async_submit)
	6342	+ async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
	6343	+
6336	6344	bio_get(bio);
6337	6345
6338	6346	if (!write) {
...	...	@@ -7113,6 +7121,7 @@
7113	7121	extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7114	7122	ei->io_tree.track_uptodate = 1;
7115	7123	ei->io_failure_tree.track_uptodate = 1;
	7124	+ atomic_set(&ei->sync_writers, 0);
7116	7125	mutex_init(&ei->log_mutex);
7117	7126	mutex_init(&ei->delalloc_mutex);
7118	7127	btrfs_ordered_inode_tree_init(&ei->ordered_tree);