From: Derrick Brashear <shadow@dementia.org>
Date: Thu, 5 Jun 2003 22:42:41 +0000 (+0000)
Subject: STABLE12-rx-avoid-leaking-nwaiting-20030605
X-Git-Tag: openafs-stable-1_2_10~46
X-Git-Url: https://git.michaelhowe.org/gitweb/?a=commitdiff_plain;h=2165b62571318ac03f1a0e9e1d00273bb642821f;p=packages%2Fo%2Fopenafs.git

STABLE12-rx-avoid-leaking-nwaiting-20030605


thanks to kolya@mit.edu for suggesting moving rather than adding the
decrement and confirming sanity.

basically if a call ended up call->error it was possible to queue_Remove()
it, drop the call, and not rx_nWaiting--; result should be observed as
rxdebug reporting waitprocs when threads were idle.


(cherry picked from commit 3fd6efb40fb4b81874cd1d48a1b529a175383e55)
---

diff --git a/src/rx/rx.c b/src/rx/rx.c
index f59d1a677..93cdb1a56 100644
--- a/src/rx/rx.c
+++ b/src/rx/rx.c
@@ -1492,6 +1492,13 @@ osi_socket *socketp;
 	    MUTEX_EXIT(&rx_serverPool_lock);
 	    MUTEX_ENTER(&call->lock);
 
+	    if (call->flags & RX_CALL_WAIT_PROC) {
+		call->flags &= ~RX_CALL_WAIT_PROC;
+		MUTEX_ENTER(&rx_stats_mutex);
+		rx_nWaiting--;
+		MUTEX_EXIT(&rx_stats_mutex);
+	    }
+
 	    if (call->state != RX_STATE_PRECALL || call->error) {
 		MUTEX_EXIT(&call->lock);
 		MUTEX_ENTER(&rx_serverPool_lock);
@@ -1505,10 +1512,6 @@ osi_socket *socketp;
 		rxi_SendAck(call, 0, 0, 0, 0, RX_ACK_DELAY, 0);
 
 	    CLEAR_CALL_QUEUE_LOCK(call);
-	    call->flags &= ~RX_CALL_WAIT_PROC;
-	    MUTEX_ENTER(&rx_stats_mutex);
-	    rx_nWaiting--;
-	    MUTEX_EXIT(&rx_stats_mutex);
 	    break;
 	}
 	else {