nfs4: serialize layoutcommit

Current pnfs_layoutcommit_inode can not handle parallel layoutcommit. And as Trond suggested , there is no need for client to optimize for parallel layoutcommit. So add NFS_INO_LAYOUTCOMMITTING flag to mark inflight layoutcommit and serialize lalyoutcommit with it. Also mark_inode_dirty_sync if pnfs_layoutcommit_inode fails to issue layoutcommit. Reported-by: Vitaliy Gusev <gusev.vitaliy@nexenta.com> Signed-off-by: Peng Tao <peng_tao@emc.com> Signed-off-by: Jim Rees <rees@umich.edu> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

nfs4: serialize layoutcommit
Current pnfs_layoutcommit_inode can not handle parallel layoutcommit. And as Trond suggested , there is no need for client to optimize for parallel layoutcommit. So add NFS_INO_LAYOUTCOMMITTING flag to mark inflight layoutcommit and serialize lalyoutcommit with it. Also mark_inode_dirty_sync if pnfs_layoutcommit_inode fails to issue layoutcommit. Reported-by: Vitaliy Gusev <gusev.vitaliy@nexenta.com> Signed-off-by: Peng Tao <peng_tao@emc.com> Signed-off-by: Jim Rees <rees@umich.edu> Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Peng Tao · Trond Myklebust
1 parent e20de37757
Showing 3 changed files with 29 additions and 3 deletions Side-by-side Diff
fs/nfs/nfs4proc.c
fs/nfs/pnfs.c
include/linux/nfs_fs.h
@@ -5950,6 +5950,7 @@
 {
 	struct nfs4_layoutcommit_data *data = calldata;
 	struct pnfs_layout_segment *lseg, *tmp;
+	unsigned long *bitlock = &NFS_I(data->args.inode)->flags;
  
 	pnfs_cleanup_layoutcommit(data);
 	/* Matched by references in pnfs_set_layoutcommit */
@@ -5959,6 +5960,11 @@
 				       &lseg->pls_flags))
 			put_lseg(lseg);
 	}
+
+	clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
+	smp_mb__after_clear_bit();
+	wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
+
 	put_rpccred(data->cred);
 	kfree(data);
 }
@@ -1443,17 +1443,31 @@
 	/* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
 	data = kzalloc(sizeof(*data), GFP_NOFS);
 	if (!data) {
-		mark_inode_dirty_sync(inode);
 		status = -ENOMEM;
 		goto out;
 	}
  
+	if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
+		goto out_free;
+
+	if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
+		if (!sync) {
+			status = -EAGAIN;
+			goto out_free;
+		}
+		status = wait_on_bit_lock(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING,
+					nfs_wait_bit_killable, TASK_KILLABLE);
+		if (status)
+			goto out_free;
+	}
+
 	INIT_LIST_HEAD(&data->lseg_list);
 	spin_lock(&inode->i_lock);
 	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
+		clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags);
 		spin_unlock(&inode->i_lock);
-		kfree(data);
-		goto out;
+		wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING);
+		goto out_free;
 	}
  
 	pnfs_list_write_lseg(inode, &data->lseg_list);
  
@@ -1475,7 +1489,12 @@
  
 	status = nfs4_proc_layoutcommit(data, sync);
 out:
+	if (status)
+		mark_inode_dirty_sync(inode);
 	dprintk("<-- %s status %d\n", __func__, status);
 	return status;
+out_free:
+	kfree(data);
+	goto out;
 }
@@ -229,6 +229,7 @@
 #define NFS_INO_COMMIT		(7)		/* inode is committing unstable writes */
 #define NFS_INO_PNFS_COMMIT	(8)		/* use pnfs code for commit */
 #define NFS_INO_LAYOUTCOMMIT	(9)		/* layoutcommit required */
+#define NFS_INO_LAYOUTCOMMITTING (10)		/* layoutcommit inflight */
  
 static inline struct nfs_inode *NFS_I(const struct inode *inode)
 {
...	...	@@ -5950,6 +5950,7 @@
5950	5950	{
5951	5951	struct nfs4_layoutcommit_data *data = calldata;
5952	5952	struct pnfs_layout_segment lseg, tmp;
	5953	+ unsigned long *bitlock = &NFS_I(data->args.inode)->flags;
5953	5954
5954	5955	pnfs_cleanup_layoutcommit(data);
5955	5956	/* Matched by references in pnfs_set_layoutcommit */
...	...	@@ -5959,6 +5960,11 @@
5959	5960	&lseg->pls_flags))
5960	5961	put_lseg(lseg);
5961	5962	}
	5963	+
	5964	+ clear_bit_unlock(NFS_INO_LAYOUTCOMMITTING, bitlock);
	5965	+ smp_mb__after_clear_bit();
	5966	+ wake_up_bit(bitlock, NFS_INO_LAYOUTCOMMITTING);
	5967	+
5962	5968	put_rpccred(data->cred);
5963	5969	kfree(data);
5964	5970	}
...	...	@@ -1443,17 +1443,31 @@
1443	1443	/* Note kzalloc ensures data->res.seq_res.sr_slot == NULL */
1444	1444	data = kzalloc(sizeof(*data), GFP_NOFS);
1445	1445	if (!data) {
1446		- mark_inode_dirty_sync(inode);
1447	1446	status = -ENOMEM;
1448	1447	goto out;
1449	1448	}
1450	1449
	1450	+ if (!test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags))
	1451	+ goto out_free;
	1452	+
	1453	+ if (test_and_set_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags)) {
	1454	+ if (!sync) {
	1455	+ status = -EAGAIN;
	1456	+ goto out_free;
	1457	+ }
	1458	+ status = wait_on_bit_lock(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING,
	1459	+ nfs_wait_bit_killable, TASK_KILLABLE);
	1460	+ if (status)
	1461	+ goto out_free;
	1462	+ }
	1463	+
1451	1464	INIT_LIST_HEAD(&data->lseg_list);
1452	1465	spin_lock(&inode->i_lock);
1453	1466	if (!test_and_clear_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) {
	1467	+ clear_bit(NFS_INO_LAYOUTCOMMITTING, &nfsi->flags);
1454	1468	spin_unlock(&inode->i_lock);
1455		- kfree(data);
1456		- goto out;
	1469	+ wake_up_bit(&nfsi->flags, NFS_INO_LAYOUTCOMMITTING);
	1470	+ goto out_free;
1457	1471	}
1458	1472
1459	1473	pnfs_list_write_lseg(inode, &data->lseg_list);
1460	1474
...	...	@@ -1475,7 +1489,12 @@
1475	1489
1476	1490	status = nfs4_proc_layoutcommit(data, sync);
1477	1491	out:
	1492	+ if (status)
	1493	+ mark_inode_dirty_sync(inode);
1478	1494	dprintk("<-- %s status %d\n", __func__, status);
1479	1495	return status;
	1496	+out_free:
	1497	+ kfree(data);
	1498	+ goto out;
1480	1499	}
...	...	@@ -229,6 +229,7 @@
229	229	#define NFS_INO_COMMIT (7) /* inode is committing unstable writes */
230	230	#define NFS_INO_PNFS_COMMIT (8) /* use pnfs code for commit */
231	231	#define NFS_INO_LAYOUTCOMMIT (9) /* layoutcommit required */
	232	+#define NFS_INO_LAYOUTCOMMITTING (10) /* layoutcommit inflight */
232	233
233	234	static inline struct nfs_inode NFS_I(const struct inode inode)
234	235	{