[SCSI] ibmvscsi: Prevent IO during partner login

By setting the request_limit in send_srp_login to 1 we allowed login
requests to be sent to the server adapter.  If this was not an initial
login, but was a login after a disconnect with the server, other I/O
requests could attempt to be processed before the login occured.  These
I/O requests would fail, sometimes resulting in filesystems getting
marked read-only.

To address this we can set the request_limit to 0 while doing the login
and add an exception where login requests, along with task management
events, are always passed to the server.

There is a case where the request_limit had already reached 0 would result
in all events being sent rather than returning SCSI_MLQUEUE_HOST_BUSY; this
has also been fixed by this patch.
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>

drivers/scsi/ibmvscsi/ibmvscsi.c

@@ -556,7 +556,7 @@ static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
 				   unsigned long timeout)
 {
 	u64 *crq_as_u64 = (u64 *) &evt_struct->crq;
-	int request_status;
+	int request_status = 0;
 	int rc;
 
 	/* If we have exhausted our request limit, just fail this request,
@@ -574,6 +574,13 @@ static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
 		if (request_status < -1)
 			goto send_error;
 		/* Otherwise, we may have run out of requests. */
+		/* If request limit was 0 when we started the adapter is in the
+		 * process of performing a login with the server adapter, or
+		 * we may have run out of requests.
+		 */
+		else if (request_status == -1 &&
+		         evt_struct->iu.srp.login_req.opcode != SRP_LOGIN_REQ)
+			goto send_busy;
 		/* Abort and reset calls should make it through.
 		 * Nothing except abort and reset should use the last two
 		 * slots unless we had two or less to begin with.
@@ -633,6 +640,7 @@ static int ibmvscsi_send_srp_event(struct srp_event_struct *evt_struct,
 	unmap_cmd_data(&evt_struct->iu.srp.cmd, evt_struct, hostdata->dev);
 
 	free_event_struct(&hostdata->pool, evt_struct);
+	if (request_status != -1)
 		atomic_inc(&hostdata->request_limit);
 	return SCSI_MLQUEUE_HOST_BUSY;
 
@@ -927,10 +935,11 @@ static int send_srp_login(struct ibmvscsi_host_data *hostdata)
 	login->req_buf_fmt = SRP_BUF_FORMAT_DIRECT | SRP_BUF_FORMAT_INDIRECT;
 	
 	spin_lock_irqsave(hostdata->host->host_lock, flags);
-	/* Start out with a request limit of 1, since this is negotiated in
-	 * the login request we are just sending
+	/* Start out with a request limit of 0, since this is negotiated in
+	 * the login request we are just sending and login requests always
+	 * get sent by the driver regardless of request_limit.
 	 */
-	atomic_set(&hostdata->request_limit, 1);
+	atomic_set(&hostdata->request_limit, 0);
 
 	rc = ibmvscsi_send_srp_event(evt_struct, hostdata, init_timeout * 2);
 	spin_unlock_irqrestore(hostdata->host->host_lock, flags);