Keep track of how responsive clients have been and don't time out good ones.

6 files changed, 28 insertions, 19 deletions

diff --git a/server/client.c b/server/client.c
index 93638e8..2dca538 100644
--- a/server/client.c
+++ b/server/client.c
@@ -89,6 +89,7 @@ client_register(notice, host, client_p, wantdefaults)
 	memset(&client->session_key, 0, sizeof(client->session_key));
 #endif
 	client->last_send = 0;
+	client->last_ack = NOW;
 	client->addr.sin_family = AF_INET;
 	client->addr.sin_addr.s_addr = host->s_addr;
 	client->addr.sin_port = notice->z_port;
diff --git a/server/dispatch.c b/server/dispatch.c
index 4519812..5659516 100644
--- a/server/dispatch.c
+++ b/server/dispatch.c
@@ -499,9 +499,7 @@ nack_release(client)
     for (i = 0; i < NACKTAB_HASHSIZE; i++) {
 	for (nacked = nacktab[i]; nacked; nacked = next) {
 	    next = nacked->next;
-	    if (nacked->dest.addr.sin_addr.s_addr ==
-		client->addr.sin_addr.s_addr &&
-		nacked->dest.addr.sin_port == client->addr.sin_port) {
+	    if (nacked->client == client) {
 		timer_reset(nacked->timer);
 		LIST_DELETE(nacked);
 		free(nacked->packet);
@@ -558,6 +556,7 @@ xmit_frag(notice, buf, len, waitforack)
     memcpy(savebuf, buf, len);
 
     sin = ZGetDestAddr();
+    nacked->client = NULL;
     nacked->rexmits = (sendfail) ? -1 : 0;
     nacked->packet = savebuf;
     nacked->dest.addr = sin;
@@ -660,6 +659,7 @@ xmit(notice, dest, auth, client)
 	return;
     }
 
+    nacked->client = client;
     nacked->rexmits = (sendfail) ? -1 : 0;
     nacked->packet = noticepack;
     nacked->dest.addr = *dest;
@@ -681,7 +681,6 @@ rexmit(arg)
 {
     Unacked *nacked = (Unacked *) arg;
     int retval;
-    Client *client;
 
 #if 1
     syslog(LOG_DEBUG, "rexmit %s/%d #%d time %d",
@@ -691,22 +690,23 @@ rexmit(arg)
 
     nacked->rexmits++;
     if (rexmit_times[nacked->rexmits] == -1) {
-	/* Unresponsive client, find it in our database. */
-	client = client_find(&nacked->dest.addr.sin_addr,
-			     nacked->dest.addr.sin_port);
-
-	/* unlink & free nacked */
-	LIST_DELETE(nacked);
-	free(nacked->packet);
-	free(nacked);
-
-	/* Kill the client. */
-	if (client) {
-	    server_kill_clt(client);
-	    client_deregister(client, 1);
+	if (!nacked->client
+	    || NOW - nacked->client->last_ack >= CLIENT_GIVEUP_MIN) {
+	    /* The client (if there was one) has been unresponsive.  Give up
+	     * sending this packet, and kill the client if there was one. */
+	    LIST_DELETE(nacked);
+	    free(nacked->packet);
+	    free(nacked);
+	    if (nacked->client) {
+		server_kill_clt(nacked->client);
+		client_deregister(nacked->client, 1);
+	    }
+	    return;
+	} else {
+	    /* The client has sent us an ack recently.  Retry with the maximum
+	     * retransmit time. */
+	    nacked->rexmits--;
 	}
-
-	return;
     }
 
     /* retransmit the packet */
@@ -841,6 +841,8 @@ nack_cancel(notice, who)
 	if (nacked->dest.addr.sin_addr.s_addr == who->sin_addr.s_addr
 	    && nacked->dest.addr.sin_port == who->sin_port
 	    && ZCompareUID(&nacked->uid, &notice->z_uid)) {
+	    if (nacked->client)
+		nacked->client->last_ack = NOW;
 	    timer_reset(nacked->timer);
 	    free(nacked->packet);
 	    LIST_DELETE(nacked);
diff --git a/server/realm.c b/server/realm.c
index 425cc36..9cf8c92 100644
--- a/server/realm.c
+++ b/server/realm.c
@@ -451,6 +451,7 @@ realm_init()
     sprintf(rlmprinc, "%s.%s@%s", SERVER_SERVICE, SERVER_INSTANCE, rlm->name);
     client->principal = make_string(rlmprinc, 0);
     client->last_send = 0;
+    client->last_ack = NOW;
     client->subs = NULL;
     client->realm = rlm;
     client->addr.sin_family = 0;
@@ -627,6 +628,7 @@ realm_sendit(notice, who, auth, realm, ack_to_sender)
     return;
   }
 
+  nacked->client = NULL;
   nacked->rexmits = 0;
   nacked->packet = pack;
   nacked->dest.rlm.rlm_idx = realm - otherrealms;
diff --git a/server/server.c b/server/server.c
index b034dc2..f5b9678 100644
--- a/server/server.c
+++ b/server/server.c
@@ -1419,6 +1419,7 @@ server_forw_reliable(server, pack, packlen, notice)
 	return;
     }
 
+    nacked->client = NULL;
     nacked->rexmits = 0;
     nacked->packet = pack;
     nacked->dest.srv_idx = server - otherservers;
diff --git a/server/zserver.h b/server/zserver.h
index 219e82b..2cad00e 100644
--- a/server/zserver.h
+++ b/server/zserver.h
@@ -114,6 +114,7 @@ struct _Client {
 #endif /* HAVE_KRB4 */
     String		*principal;	/* krb principal of user */
     int			last_send;	/* Counter for last sent packet. */
+    time_t		last_ack;	/* Time of last received ack */
     Realm		*realm;
     struct _Client	*next, **prev_p;
 };
@@ -135,6 +136,7 @@ enum _Server_state {
 
 struct _Unacked {
     Timer		*timer;		/* timer for retransmit */
+    Client		*client;	/* responsible client, or NULL */
     short		rexmits;	/* number of retransmits */
     short		packsz;		/* size of packet */
     char		*packet;	/* ptr to packet */
diff --git a/server/zsrv_conf.h b/server/zsrv_conf.h
index 9c2cbce..947b4ab 100644
--- a/server/zsrv_conf.h
+++ b/server/zsrv_conf.h
@@ -32,6 +32,7 @@
 
 #define REXMIT_TIMES { 2, 2, 4, 4, 8, 8, 16, 32, 64, 128, 256, 512, -1 }
 #define NUM_REXMIT_TIMES 12
+#define CLIENT_GIVEUP_MIN 512
 
 /* hostmanager defines */
 #define	LOSE_TIMO	(60)		/* time during which a losing host