NFSv4: Add a parameter to limit the number of retries after NFS4ERR_DELAY

When using a 'softerr' mount, the NFSv4 client can get stuck waiting
forever while the server just returns NFS4ERR_DELAY. Among other things,
this causes the knfsd server threads to busy wait.
Add a parameter that tells the NFSv4 client how many times to retry
before giving up.

Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>

3 files changed, 34 insertions, 1 deletions

diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h
index 47c5c1f86d66..97e91c33ca91 100644
--- a/fs/nfs/nfs4_fs.h
+++ b/fs/nfs/nfs4_fs.h
@@ -209,6 +209,7 @@ struct nfs4_exception {
 	struct inode *inode;
 	nfs4_stateid *stateid;
 	long timeout;
+	unsigned short retrans;
 	unsigned char task_is_privileged : 1;
 	unsigned char delay : 1,
 		      recovering : 1,
@@ -546,6 +547,7 @@ extern unsigned short max_session_slots;
 extern unsigned short max_session_cb_slots;
 extern unsigned short send_implementation_id;
 extern bool recover_lost_locks;
+extern short nfs_delay_retrans;
 
 #define NFS4_CLIENT_ID_UNIQ_LEN		(64)
 extern char nfs4_client_id_uniquifier[NFS4_CLIENT_ID_UNIQ_LEN];
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 5ee283eb9660..6f64fe97e72f 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -585,6 +585,21 @@ wait_on_recovery:
 	return 0;
 }
 
+/*
+ * Track the number of NFS4ERR_DELAY related retransmissions and return
+ * EAGAIN if the 'softerr' mount option is set, and we've exceeded the limit
+ * set by 'nfs_delay_retrans'.
+ */
+static int nfs4_exception_should_retrans(const struct nfs_server *server,
+					 struct nfs4_exception *exception)
+{
+	if (server->flags & NFS_MOUNT_SOFTERR && nfs_delay_retrans >= 0) {
+		if (exception->retrans++ >= (unsigned short)nfs_delay_retrans)
+			return -EAGAIN;
+	}
+	return 0;
+}
+
 /* This is the error handling routine for processes that are allowed
  * to sleep.
  */
@@ -595,6 +610,11 @@ int nfs4_handle_exception(struct nfs_server *server, int errorcode, struct nfs4_
 
 	ret = nfs4_do_handle_exception(server, errorcode, exception);
 	if (exception->delay) {
+		int ret2 = nfs4_exception_should_retrans(server, exception);
+		if (ret2 < 0) {
+			exception->retry = 0;
+			return ret2;
+		}
 		ret = nfs4_delay(&exception->timeout,
 				exception->interruptible);
 		goto out_retry;
@@ -623,6 +643,11 @@ nfs4_async_handle_exception(struct rpc_task *task, struct nfs_server *server,
 
 	ret = nfs4_do_handle_exception(server, errorcode, exception);
 	if (exception->delay) {
+		int ret2 = nfs4_exception_should_retrans(server, exception);
+		if (ret2 < 0) {
+			exception->retry = 0;
+			return ret2;
+		}
 		rpc_delay(task, nfs4_update_delay(&exception->timeout));
 		goto out_retry;
 	}
diff --git a/fs/nfs/super.c b/fs/nfs/super.c
index 0d6473cb00cb..bd00cfab73d1 100644
--- a/fs/nfs/super.c
+++ b/fs/nfs/super.c
@@ -1366,6 +1366,7 @@ unsigned short max_session_cb_slots = NFS4_DEF_CB_SLOT_TABLE_SIZE;
 unsigned short send_implementation_id = 1;
 char nfs4_client_id_uniquifier[NFS4_CLIENT_ID_UNIQ_LEN] = "";
 bool recover_lost_locks = false;
+short nfs_delay_retrans = -1;
 
 EXPORT_SYMBOL_GPL(nfs_callback_nr_threads);
 EXPORT_SYMBOL_GPL(nfs_callback_set_tcpport);
@@ -1376,6 +1377,7 @@ EXPORT_SYMBOL_GPL(max_session_cb_slots);
 EXPORT_SYMBOL_GPL(send_implementation_id);
 EXPORT_SYMBOL_GPL(nfs4_client_id_uniquifier);
 EXPORT_SYMBOL_GPL(recover_lost_locks);
+EXPORT_SYMBOL_GPL(nfs_delay_retrans);
 
 #define NFS_CALLBACK_MAXPORTNR (65535U)
 
@@ -1424,5 +1426,9 @@ MODULE_PARM_DESC(recover_lost_locks,
 		 "If the server reports that a lock might be lost, "
 		 "try to recover it risking data corruption.");
 
-
+module_param_named(delay_retrans, nfs_delay_retrans, short, 0644);
+MODULE_PARM_DESC(delay_retrans,
+		 "Unless negative, specifies the number of times the NFSv4 "
+		 "client retries a request before returning an EAGAIN error, "
+		 "after a reply of NFS4ERR_DELAY from the server.");
 #endif /* CONFIG_NFS_V4 */