From: mchristi@redhat.com <mchristi@redhat.com>
Date: Mon, 25 Aug 2008 19:15:54 -0500
Subject: [scsi] iscsi: fix nop timeout detection
Message-id: 1219709754-10592-1-git-send-email-mchristi@redhat.com
O-Subject: [RHEL 5.3 PATCH] fix iscsi nop timeout detection
Bugzilla: 453969

From: Mike Christie <mchristi@redhat.com>

This is for BZ: 453969

The following patch fixes two bugs in the iscsi nop processing.
The target sends iscsi nops to ping the initiator and the
initiator has to send nops to reply and can send nops to
ping the target.

The first bug was that whenever the timer woke up we checked
if a ping timedout when most of the time we want to check
if we wanted to send a ping.

The second bug was that when we got a ping response we did not
reset the timer so if the ping timeout was long and the
recv timeout was short we might have missed a chance to check
the transport. This delays error detection which delays
dm-multipath failovers.

This patch checks if a iscsi ping is outstanding and if the ping has
timed out, to determine if we need to signal a connection problem.
And it resets the timer when getting a response to one of our nops.

The upstream commit is here
http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=c8611f975403dd20e6503aff8aded5dcb718f75b
http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=4cf1043593db6a337f10e006c23c69e5fc93e722

diff --git a/drivers/scsi/libiscsi.c b/drivers/scsi/libiscsi.c
index 6af7606..603fc50 100644
--- a/drivers/scsi/libiscsi.c
+++ b/drivers/scsi/libiscsi.c
@@ -584,7 +584,9 @@ int __iscsi_complete_pdu(struct iscsi_conn *conn, struct iscsi_hdr *hdr,
 				if (iscsi_recv_pdu(conn->cls_conn, hdr, data,
 						   datalen))
 					rc = ISCSI_ERR_CONN_FAILED;
-			}
+			} else
+				mod_timer(&conn->transport_timer,
+					  jiffies + conn->recv_timeout);
 			iscsi_free_mgmt_task(conn, mtask);
 			break;
 		default:
@@ -1306,19 +1308,20 @@ static void iscsi_check_transport_timeouts(unsigned long data)
 {
 	struct iscsi_conn *conn = (struct iscsi_conn *)data;	
 	struct iscsi_session *session = conn->session;
-	unsigned long timeout, next_timeout = 0, last_recv;
+	unsigned long recv_timeout, next_timeout = 0, last_recv;
 
 	spin_lock(&session->lock);
 	if (session->state != ISCSI_STATE_LOGGED_IN)
 		goto done;
 
-	timeout = conn->recv_timeout;
-	if (!timeout)
+	recv_timeout = conn->recv_timeout;
+	if (!recv_timeout)
 		goto done;
 
-	timeout *= HZ;
+	recv_timeout *= HZ;
 	last_recv = conn->last_recv;
-	if (time_before_eq(last_recv + timeout + (conn->ping_timeout * HZ),
+	if (conn->ping_mtask &&
+	    time_before_eq(conn->last_ping + (conn->ping_timeout * HZ),
 			   jiffies)) {
 		printk(KERN_ERR "ping timeout of %d secs expired, "
 		       "last rx %lu, last ping %lu, now %lu\n",
@@ -1329,15 +1332,13 @@ static void iscsi_check_transport_timeouts(unsigned long data)
 		return;
 	}
 
-	if (time_before_eq(last_recv + timeout, jiffies)) {
-		if (time_before_eq(conn->last_ping, last_recv)) {
-			/* send a ping to try to provoke some traffic */
-			debug_scsi("Sending nopout as ping on conn %p\n", conn);
-			iscsi_send_nopout(conn, NULL);
-		}
-		next_timeout = last_recv + timeout + (conn->ping_timeout * HZ);
+	if (time_before_eq(last_recv + recv_timeout, jiffies)) {
+		/* send a ping to try to provoke some traffic */
+		debug_scsi("Sending nopout as ping on conn %p\n", conn);
+		iscsi_send_nopout(conn, NULL);
+		next_timeout = conn->last_ping + (conn->ping_timeout * HZ);
 	} else
-		next_timeout = last_recv + timeout;
+		next_timeout = last_recv + recv_timeout;
 
 	debug_scsi("Setting next tmo %lu\n", next_timeout);
 	mod_timer(&conn->transport_timer, next_timeout);