make dispatcher hash sizes be primes, and provide an increment to help

resolve hash collisions

bin/named/interfacemgr.c

@@ -146,11 +146,11 @@ ns_interface_create(ns_interfacemgr_t *mgr, isc_sockaddr_t *addr,
 	/* 
 	 * XXXRTH hardwired constants.  We're going to need to determine if
 	 * this UDP socket will be shared with the resolver, and if so, we
-	 * need to set the hashsize to be be something bigger than 4.
+	 * need to set the hashsize to be be something bigger than 17.
 	 */
 	ifp->udpdispatch = NULL;
 	result = dns_dispatch_create(mgr->mctx, ifp->udpsocket, ifp->task,
-				     4096, 50, 50, 4, &ifp->udpdispatch);
+				     4096, 50, 50, 17, 19, &ifp->udpdispatch);
 	if (result != ISC_R_SUCCESS) {
 		UNEXPECTED_ERROR(__FILE__, __LINE__,
 				 "UDP dns_dispatch_create(): %s",

bin/tests/dispatch_tcp_test.c

@@ -107,7 +107,7 @@ my_accept(isc_task_t *task, isc_event_t *ev_in)
 	 */
 	disp = NULL;
 	RUNTIME_CHECK(dns_dispatch_create(mctx, ev->newsocket, task,
-					  512, 6, 1024, 4, &disp)
+					  512, 6, 1024, 17, 19, &disp)
 		      == ISC_R_SUCCESS);
 
 	resp = NULL;

bin/tests/dispatch_test.c

@@ -405,7 +405,7 @@ main(int argc, char *argv[])
 	 */
 	disp = NULL;
 	RUNTIME_CHECK(dns_dispatch_create(mctx, s0, t0, 512, 6, 1024,
-					 4, &disp) == ISC_R_SUCCESS);
+					 17, 19, &disp) == ISC_R_SUCCESS);
 
 	RUNTIME_CHECK(isc_mutex_init(&client_lock) == ISC_R_SUCCESS);
 	RUNTIME_CHECK(isc_mutex_lock(&client_lock) == ISC_R_SUCCESS);

bin/tests/res_test.c

@@ -205,8 +205,8 @@ main(int argc, char *argv[]) {
 				      &view) == ISC_R_SUCCESS);
 
 	dispatch = NULL;
-	RUNTIME_CHECK(dns_dispatch_create(mctx, s, task1, 4096, 1000, 1000, 4,
+	RUNTIME_CHECK(dns_dispatch_create(mctx, s, task1, 4096, 1000, 1000,
-					  &dispatch) == DNS_R_SUCCESS);
+					  17, 19, &dispatch) == DNS_R_SUCCESS);
 
 #ifdef notyet
 	res = NULL;

lib/dns/dispatch.c

@@ -57,7 +57,7 @@ struct dns_dispentry {
 	ISC_LINK(dns_dispentry_t)	link;
 };
 
-#define INVALID_BUCKET (0xffffdead)
+#define INVALID_BUCKET		(0xffffdead)
 
 typedef ISC_LIST(dns_dispentry_t)	dns_displist_t;
 
@@ -89,8 +89,8 @@ struct dns_dispatch {
 	ISC_LIST(dns_dispatchevent_t) rq_events; /* holder for rq events */
 	dns_tcpmsg_t		tcpmsg;		/* for tcp streams */
 	isc_lfsr_t		qid_lfsr;	/* state generator info */
-	unsigned int		qid_hashsize;	/* hash table size */
+	unsigned int		qid_nbuckets;	/* hash table size */
-	unsigned int		qid_mask;	/* mask for hash table */
+	unsigned int		qid_increment;	/* id increment on collision */
 	dns_displist_t	        *qid_table;	/* the table itself */
 };
 
@@ -146,7 +146,7 @@ linear_first(dns_dispatch_t *disp)
 
 	bucket = 0;
 
-	while (bucket < disp->qid_hashsize) {
+	while (bucket < disp->qid_nbuckets) {
 		ret = ISC_LIST_HEAD(disp->qid_table[bucket]);
 		if (ret != NULL)
 			return (ret);
@@ -167,7 +167,7 @@ linear_next(dns_dispatch_t *disp, dns_dispentry_t *resp)
 		return (ret);
 
 	bucket = resp->bucket;
-	while (bucket < disp->qid_hashsize) {
+	while (bucket < disp->qid_nbuckets) {
 		ret = ISC_LIST_HEAD(disp->qid_table[bucket]);
 		if (ret != NULL)
 			return (ret);
@@ -178,20 +178,18 @@ linear_next(dns_dispatch_t *disp, dns_dispentry_t *resp)
 }
 
 /*
- * Return a hash of the destination and message id.  For now, just return
+ * Return a hash of the destination and message id.
- * the message id bits, and mask off the low order bits of that.
  */
 static unsigned int
 hash(dns_dispatch_t *disp, isc_sockaddr_t *dest, dns_messageid_t id)
 {
 	unsigned int ret;
 
-	(void)dest;  /* shut up compiler warning. */
+	ret = isc_sockaddr_hash(dest, ISC_TRUE);
+	ret ^= (id & 0x0000ffff); /* important to mask off garbage bits */
+	ret %= disp->qid_nbuckets;
 
-	ret = id;
+	INSIST(ret < disp->qid_nbuckets);
-	ret &= disp->qid_mask;
-
-	INSIST(ret < disp->qid_hashsize);
 
 	return (ret);
 }
@@ -235,7 +233,7 @@ destroy(dns_dispatch_t *disp)
 	isc_mempool_destroy(&disp->bpool);
 	isc_mempool_destroy(&disp->epool);
 	isc_mem_put(disp->mctx, disp->qid_table,
-		    disp->qid_hashsize * sizeof(dns_displist_t));
+		    disp->qid_nbuckets * sizeof(dns_displist_t));
 
 	isc_mem_put(disp->mctx, disp, sizeof(dns_dispatch_t));
 }
@@ -247,7 +245,7 @@ bucket_search(dns_dispatch_t *disp, isc_sockaddr_t *dest, dns_messageid_t id,
 {
 	dns_dispentry_t *res;
 
-	REQUIRE(bucket < disp->qid_hashsize);
+	REQUIRE(bucket < disp->qid_nbuckets);
 
 	res = ISC_LIST_HEAD(disp->qid_table[bucket]);
 
@@ -775,11 +773,10 @@ isc_result_t
 dns_dispatch_create(isc_mem_t *mctx, isc_socket_t *sock, isc_task_t *task,
 		    unsigned int maxbuffersize,
 		    unsigned int maxbuffers, unsigned int maxrequests,
-		    unsigned int hashsize,
+		    unsigned int buckets, unsigned int increment,
 		    dns_dispatch_t **dispp)
 {
 	dns_dispatch_t *disp;
-	unsigned int tablesize;
 	isc_result_t res;
 	isc_sockettype_t socktype;
 	unsigned int i;
@@ -787,7 +784,8 @@ dns_dispatch_create(isc_mem_t *mctx, isc_socket_t *sock, isc_task_t *task,
 	REQUIRE(mctx != NULL);
 	REQUIRE(sock != NULL);
 	REQUIRE(task != NULL);
-	REQUIRE(hashsize <= 24);
+	REQUIRE(buckets < 2097169);  /* next prime > 65536 * 32 */
+	REQUIRE(increment > buckets);
 	REQUIRE(maxbuffersize >= 512 && maxbuffersize < (64 * 1024));
 	REQUIRE(maxbuffers > 0);
 	REQUIRE(dispp != NULL && *dispp == NULL);
@@ -822,20 +820,18 @@ dns_dispatch_create(isc_mem_t *mctx, isc_socket_t *sock, isc_task_t *task,
 	ISC_LIST_INIT(disp->rq_handlers);
 	ISC_LIST_INIT(disp->rq_events);
 
-	tablesize = (1 << hashsize);
-
 	disp->qid_table = isc_mem_get(disp->mctx,
-				      tablesize * sizeof(dns_displist_t));
+				      buckets * sizeof(dns_displist_t));
 	if (disp->qid_table == NULL) {
 		res = DNS_R_NOMEMORY;
 		goto out1;
 	}
 
-	for (i = 0 ; i < tablesize ; i++)
+	for (i = 0 ; i < buckets ; i++)
 		ISC_LIST_INIT(disp->qid_table[i]);
 
-	disp->qid_mask = tablesize - 1;
+	disp->qid_nbuckets = buckets;
-	disp->qid_hashsize = tablesize;
+	disp->qid_increment = increment;
 
 	if (isc_mutex_init(&disp->lock) != ISC_R_SUCCESS) {
 		res = DNS_R_UNEXPECTED;
@@ -924,7 +920,7 @@ dns_dispatch_create(isc_mem_t *mctx, isc_socket_t *sock, isc_task_t *task,
  out3:
 	isc_mutex_destroy(&disp->lock);
  out2:
-	isc_mem_put(mctx, disp->mctx, disp->qid_hashsize * sizeof(void *));
+	isc_mem_put(mctx, disp->mctx, disp->qid_nbuckets * sizeof(void *));
  out1:
 	isc_mem_put(mctx, disp, sizeof(dns_dispatch_t));
 
@@ -1015,7 +1011,8 @@ dns_dispatch_addresponse(dns_dispatch_t *disp, isc_sockaddr_t *dest,
 			ok = ISC_TRUE;
 			break;
 		}
-		id = randomid(disp);
+		id += disp->qid_increment;
+		id &= 0x0000ffff;
 		bucket = hash(disp, dest, id);
 	}
 

lib/dns/include/dns/dispatch.h

@@ -91,7 +91,8 @@ isc_result_t
 dns_dispatch_create(isc_mem_t *mctx, isc_socket_t *sock, isc_task_t *task,
 		    unsigned int maxbuffersize,
 		    unsigned int maxbuffers, unsigned int maxrequests,
-		    unsigned int hashsize, dns_dispatch_t **dispp);
+		    unsigned int buckets, unsigned int increment,
+		    dns_dispatch_t **dispp);
 /*
  * Create a new dns_dispatch and attach it to the provided isc_socket_t.
  *
@@ -102,8 +103,10 @@ dns_dispatch_create(isc_mem_t *mctx, isc_socket_t *sock, isc_task_t *task,
  * overall system and the number of buffers which can be allocated to
  * requests.
  *
- * "hashsize" is the size of the hash table used to handle responses.  The
+ * "buckets" is the number of buckets to use, and should be prime.
- * size is 2^hashsize, maximum of 2^24.
+ *
+ * "increment" is used in a collision avoidance function, and needs to be
+ * a prime > buckets, and not 2.
  *
  * Requires:
  *
@@ -120,7 +123,9 @@ dns_dispatch_create(isc_mem_t *mctx, isc_socket_t *sock, isc_task_t *task,
  *
  *	maxrequests <= maxbuffers.
  *
- *	hashsize <= 24.
+ *	buckets < 2097169 (the next prime after 65536 * 32)
+ *
+ *	increment > buckets (and prime)
  */
 
 void

lib/dns/resolver.c

@@ -594,7 +594,8 @@ fctx_query(fetchctx_t *fctx, dns_adbaddrinfo_t *addrinfo,
 		if (result != ISC_R_SUCCESS)
 			goto cleanup_query;
 		result = dns_dispatch_create(res->mctx, socket, task,
-					     4096, 2, 1, 1, &query->dispatch);
+					     4096, 2, 1, 1, 3,
+					     &query->dispatch);
 		/*
 		 * Regardless of whether dns_dispatch_create() succeeded or
 		 * not, we don't need our reference to the socket anymore.
@@ -3226,7 +3227,8 @@ dns_resolver_create(dns_view_t *view,
 		}
 		result = dns_dispatch_create(res->mctx, res->udpsocket4,
 					     res->buckets[0].task, 4096,
-					     1000, 32768, 14, &res->dispatch4);
+					     1000, 32768, 16411, 16433,
+					     &res->dispatch4);
 		if (result != ISC_R_SUCCESS)
 			goto cleanup_udpsocket4;
 	}
@@ -3247,7 +3249,8 @@ dns_resolver_create(dns_view_t *view,
 			goto cleanup_dispatch4;
 		result = dns_dispatch_create(res->mctx, res->udpsocket6,
 					     res->buckets[0].task, 4096, 
-					     1000, 32768, 14, &res->dispatch6);
+					     1000, 32768, 16411, 16433,
+					     &res->dispatch6);
 		if (result != ISC_R_SUCCESS)
 			goto cleanup_udpsocket6;
 	}

lib/dns/zone.c

@@ -15,7 +15,7 @@
  * SOFTWARE.
  */
 
- /* $Id: zone.c,v 1.44 1999/12/13 07:56:26 marka Exp $ */
+ /* $Id: zone.c,v 1.45 1999/12/15 17:14:52 explorer Exp $ */
 
 #include <config.h>
 
@@ -1815,7 +1815,8 @@ dns_zone_manage(dns_zone_t *zone, isc_taskmgr_t *tmgr) {
 			      isc_sockettype_udp, &s) == ISC_R_SUCCESS);
 		dispatch = NULL;
 		RUNTIME_CHECK(dns_dispatch_create(zone->mctx, s, zone->task,
-			      4096, 1000, 1000, 4, &dispatch) == DNS_R_SUCCESS);
+						  4096, 1000, 1000, 17, 19,
+						  &dispatch) == DNS_R_SUCCESS);
 		result = dns_resolver_create(zone->mctx, tmgr, 10, zone->timgr,
 				             zone->rdclass, dispatch,
 					     &zone->res);