2398. [bug] Improve file descriptor management. New,

temporary, named.conf option reserved-sockets,
                        default 512. [RT #18344]

CHANGES

@@ -106,14 +106,12 @@
 
 2399.	[placeholder]
 
-2398.	[placeholder]
-
-2397.	[bug]		gssapi_functions had too many elements. [RT #18355]
-
-	[bug]           Improve file descriptor management.  New,
+2398.	[bug]           Improve file descriptor management.  New,
 			temporary, named.conf option reserved-sockets,
 			default 512. [RT #18344]
 
+2397.	[bug]		gssapi_functions had too many elements. [RT #18355]
+
 2396.	[bug]		Don't set SO_REUSEADDR for randomized ports.
 			[RT #18336]
 

bin/named/config.c

@@ -15,7 +15,7 @@
  * PERFORMANCE OF THIS SOFTWARE.
  */
 
-/* $Id: config.c,v 1.89 2008/07/20 09:29:52 fdupont Exp $ */
+/* $Id: config.c,v 1.90 2008/09/04 05:56:42 marka Exp $ */
 
 /*! \file */
 
@@ -100,6 +100,7 @@ options {\n\
 	edns-udp-size 4096;\n\
 	max-udp-size 4096;\n\
 	request-nsid false;\n\
+	reserved-sockets 512;\n\
 \n\
 	/* view */\n\
 	allow-notify {none;};\n\

bin/named/named.conf.docbook

@@ -17,7 +17,7 @@
  - PERFORMANCE OF THIS SOFTWARE.
 -->
 
-<!-- $Id: named.conf.docbook,v 1.37 2008/04/02 02:37:41 marka Exp $ -->
+<!-- $Id: named.conf.docbook,v 1.38 2008/09/04 05:56:42 marka Exp $ -->
 <refentry>
   <refentryinfo>
     <date>Aug 13, 2004</date>
@@ -202,6 +202,7 @@ options {
 	port <replaceable>integer</replaceable>;
 	querylog <replaceable>boolean</replaceable>;
 	recursing-file <replaceable>quoted_string</replaceable>;
+	reserved-sockets <replaceable>integer</replaceable>;
 	random-device <replaceable>quoted_string</replaceable>;
 	recursive-clients <replaceable>integer</replaceable>;
 	serial-query-rate <replaceable>integer</replaceable>;

bin/named/server.c

@@ -15,7 +15,7 @@
  * PERFORMANCE OF THIS SOFTWARE.
  */
 
-/* $Id: server.c,v 1.513 2008/07/18 01:26:20 marka Exp $ */
+/* $Id: server.c,v 1.514 2008/09/04 05:56:42 marka Exp $ */
 
 /*! \file */
 
@@ -31,6 +31,7 @@
 #include <isc/file.h>
 #include <isc/hash.h>
 #include <isc/httpd.h>
+#include <isc/socket.h>
 #include <isc/lex.h>
 #include <isc/parseint.h>
 #include <isc/portset.h>
@@ -2883,30 +2884,32 @@ static isc_result_t
 load_configuration(const char *filename, ns_server_t *server,
 		   isc_boolean_t first_time)
 {
-	isc_result_t result;
-	isc_interval_t interval;
-	cfg_parser_t *parser = NULL;
+	cfg_aclconfctx_t aclconfctx;
 	cfg_obj_t *config;
-	const cfg_obj_t *options;
-	const cfg_obj_t *views;
-	const cfg_obj_t *obj;
-	const cfg_obj_t *usev4ports, *avoidv4ports, *usev6ports, *avoidv6ports;
-	const cfg_obj_t *maps[3];
-	const cfg_obj_t *builtin_views;
+	cfg_parser_t *parser = NULL;
 	const cfg_listelt_t *element;
+	const cfg_obj_t *builtin_views;
+	const cfg_obj_t *maps[3];
+	const cfg_obj_t *obj;
+	const cfg_obj_t *options;
+	const cfg_obj_t *usev4ports, *avoidv4ports, *usev6ports, *avoidv6ports;
+	const cfg_obj_t *views;
 	dns_view_t *view = NULL;
 	dns_view_t *view_next;
-	dns_viewlist_t viewlist;
 	dns_viewlist_t tmpviewlist;
-	cfg_aclconfctx_t aclconfctx;
-	isc_uint32_t interface_interval;
-	isc_uint32_t heartbeat_interval;
-	isc_uint32_t udpsize;
+	dns_viewlist_t viewlist;
 	in_port_t listen_port, udpport_low, udpport_high;
+	int i;
+	isc_interval_t interval;
 	isc_portset_t *v4portset = NULL;
 	isc_portset_t *v6portset = NULL;
 	isc_resourcevalue_t nfiles;
-	int i;
+	isc_result_t result;
+	isc_uint32_t heartbeat_interval;
+	isc_uint32_t interface_interval;
+	isc_uint32_t reserved;
+	isc_uint32_t udpsize;
+	unsigned int maxsocks;
 
 	cfg_aclconfctx_init(&aclconfctx);
 	ISC_LIST_INIT(viewlist);
@@ -2992,21 +2995,41 @@ load_configuration(const char *filename, ns_server_t *server,
 	 * but may cause subsequent runtime failures for a busy recursive
 	 * server.
 	 */
-	result = isc_resource_getcurlimit(isc_resource_openfiles, &nfiles);
-	if (result == ISC_R_SUCCESS) {
-		unsigned int maxsocks;
-
 	result = isc_socketmgr_getmaxsockets(ns_g_socketmgr, &maxsocks);
-		if (result == ISC_R_SUCCESS &&
-		    (isc_resourcevalue_t)maxsocks > nfiles) {
+	if (result != ISC_R_SUCCESS)
+		maxsocks = 0;
+	result = isc_resource_getcurlimit(isc_resource_openfiles, &nfiles);
+	if (result == ISC_R_SUCCESS && (isc_resourcevalue_t)maxsocks > nfiles) {
 		isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL,
 			      NS_LOGMODULE_SERVER, ISC_LOG_WARNING,
-				      "max open files "
-				      "(%" ISC_PRINT_QUADFORMAT "u)"
+			      "max open files (%" ISC_PRINT_QUADFORMAT "u)"
 			      " is smaller than max sockets (%u)",
 			      nfiles, maxsocks);
 	}
+
+	/*
+	 * Set the number of socket reserved for TCP, stdio etc.
+	 */
+	obj = NULL;
+	result = ns_config_get(maps, "reserved-sockets", &obj);
+	INSIST(result == ISC_R_SUCCESS);
+	reserved = cfg_obj_asuint32(obj);
+	if (maxsocks != 0) {
+		if (maxsocks < 128U)			/* Prevent underflow. */
+			reserved = 0;
+		else if (reserved > maxsocks - 128U)	/* Minimum UDP space. */
+			reserved = maxsocks - 128;
 	}
+	/* Minimum TCP/stdio space. */
+	if (reserved < 128U)
+		reserved = 128;
+	if (reserved + 128U > maxsocks && maxsocks != 0) {
+		isc_log_write(ns_g_lctx, NS_LOGCATEGORY_GENERAL,
+			      NS_LOGMODULE_SERVER, ISC_LOG_WARNING,
+			      "less than 128 UDP sockets available after "
+			      "applying 'reserved-sockets' and 'maxsockets'");
+	}
+	isc__socketmgr_setreserved(ns_g_socketmgr, reserved);
 	
 	/*
 	 * Configure various server options.

doc/arm/Bv9ARM-book.xml

@@ -18,7 +18,7 @@
  - PERFORMANCE OF THIS SOFTWARE.
 -->
 
-<!-- File: $Id: Bv9ARM-book.xml,v 1.364 2008/08/12 14:40:26 jreed Exp $ -->
+<!-- File: $Id: Bv9ARM-book.xml,v 1.365 2008/09/04 05:56:42 marka Exp $ -->
 <book xmlns:xi="http://www.w3.org/2001/XInclude">
   <title>BIND 9 Administrator Reference Manual</title>
 
@@ -4562,6 +4562,7 @@ category notify { null; };
     <optional> max-transfer-idle-in <replaceable>number</replaceable>; </optional>
     <optional> max-transfer-idle-out <replaceable>number</replaceable>; </optional>
     <optional> tcp-clients <replaceable>number</replaceable>; </optional>
+    <optional> reserved-sockets <replaceable>number</replaceable>; </optional>
     <optional> recursive-clients <replaceable>number</replaceable>; </optional>
     <optional> serial-query-rate <replaceable>number</replaceable>; </optional>
     <optional> serial-queries <replaceable>number</replaceable>; </optional>
@@ -6969,6 +6970,25 @@ avoid-v6-udp-ports { 40000; range 50000 60000; };
               </listitem>
             </varlistentry>
 
+            <varlistentry>
+              <term><command>reserved-sockets</command></term>
+              <listitem>
+                <para>
+		  The number of file descriptors reserved for TCP, stdio,
+		  etc.  This needs to be big enough to cover the number of
+		  interfaces named listens on, tcp-clients as well as
+		  to provide room for outgoing TCP queries and incoming zone
+		  transfers.  The default is <literal>512</literal>.
+		  The minimum value is <literal>128</literal> and the
+		  maximum value is <literal>128</literal> less than
+		  maxsockets (-S).  This option may be removed in the future.
+                </para>
+                <para>
+		  This option has little effect on Windows.
+                </para>
+              </listitem>
+            </varlistentry>
+
             <varlistentry>
               <term><command>max-cache-size</command></term>
               <listitem>

lib/isc/include/isc/socket.h

@@ -15,7 +15,7 @@
  * PERFORMANCE OF THIS SOFTWARE.
  */
 
-/* $Id: socket.h,v 1.83 2008/08/20 23:57:59 jinmei Exp $ */
+/* $Id: socket.h,v 1.84 2008/09/04 05:56:43 marka Exp $ */
 
 #ifndef ISC_SOCKET_H
 #define ISC_SOCKET_H 1
@@ -912,6 +912,12 @@ isc_socketmgr_renderxml(isc_socketmgr_t *mgr, xmlTextWriterPtr writer);
  * Render internal statistics and other state into the XML document.
  */
 
+void
+isc__socketmgr_setreserved(isc_socketmgr_t *mgr, isc_uint32_t);
+/*%<
+ * Temporary.  For use by named only.
+ */
+
 #endif /* HAVE_LIBXML2 */
 
 ISC_LANG_ENDDECLS

lib/isc/unix/socket.c

@@ -15,7 +15,7 @@
  * PERFORMANCE OF THIS SOFTWARE.
  */
 
-/* $Id: socket.c,v 1.301 2008/08/20 23:57:59 jinmei Exp $ */
+/* $Id: socket.c,v 1.302 2008/09/04 05:56:43 marka Exp $ */
 
 /*! \file */
 
@@ -349,6 +349,7 @@ struct isc_socketmgr {
 	fd_set			*write_fds_copy;
 	int			maxfd;
 #endif	/* USE_SELECT */
+	int			reserved;	/* unlocked */
 #ifdef ISC_PLATFORM_USETHREADS
 	isc_thread_t		watcher;
 	isc_condition_t		shutdown_ok;
@@ -1856,9 +1857,18 @@ opensocket(isc_socketmgr_t *manager, isc_socket_t *sock) {
 
 #ifdef F_DUPFD
 	/*
-	 * Leave a space for stdio to work in.
+	 * Leave a space for stdio and TCP to work in.
 	 */
-	if (sock->fd >= 0 && sock->fd < 20) {
+	if (manager->reserved != 0 && sock->type == isc_sockettype_udp &&
+	    sock->fd >= 0 && sock->fd < manager->reserved) {
+		int new, tmp;
+		new = fcntl(sock->fd, F_DUPFD, manager->reserved);
+		tmp = errno;
+		(void)close(sock->fd);
+		errno = tmp;
+		sock->fd = new;
+		err = "isc_socket_create: fcntl/reserved";
+	} else if (sock->fd >= 0 && sock->fd < 20) {
 		int new, tmp;
 		new = fcntl(sock->fd, F_DUPFD, 20);
 		tmp = errno;
@@ -3257,6 +3267,14 @@ watcher(void *uap) {
 }
 #endif /* ISC_PLATFORM_USETHREADS */
 
+void
+isc__socketmgr_setreserved(isc_socketmgr_t *manager, isc_uint32_t reserved) {
+
+	REQUIRE(VALID_MANAGER(manager));
+
+	manager->reserved = reserved;
+}
+
 /*
  * Create a new socket manager.
  */
@@ -3511,6 +3529,7 @@ isc_socketmgr_create2(isc_mem_t *mctx, isc_socketmgr_t **managerp,
 	/* zero-clear so that necessary cleanup on failure will be easy */
 	memset(manager, 0, sizeof(*manager));
 	manager->maxsocks = maxsocks;
+	manager->reserved = 0;
 	manager->fds = isc_mem_get(mctx,
 				   manager->maxsocks * sizeof(isc_socket_t *));
 	if (manager->fds == NULL) {
@@ -3587,7 +3606,6 @@ isc_socketmgr_create2(isc_mem_t *mctx, isc_socketmgr_t **managerp,
 	if (result != ISC_R_SUCCESS)
 		goto cleanup;
 	memset(manager->fdstate, 0, manager->maxsocks * sizeof(int));
-
 #ifdef ISC_PLATFORM_USETHREADS
 	/*
 	 * Start up the select/poll thread.

lib/isc/win32/libisc.def

@@ -34,6 +34,7 @@ isc__mem_putanddetach
 isc__mem_strdup
 isc__mempool_get
 isc__mempool_put
+isc__socketmgr_setreserved
 isc__strerror
 isc_app_block
 isc_app_finish
@@ -336,8 +337,8 @@ isc_ratelimiter_setpertic
 isc_ratelimiter_shutdown
 isc_refcount_init
 isc_region_compare
-isc_resource_getlimit
 isc_resource_getcurlimit
+isc_resource_getlimit
 isc_resource_setlimit
 isc_result_register
 isc_result_totext

lib/isc/win32/socket.c

@@ -15,7 +15,7 @@
  * PERFORMANCE OF THIS SOFTWARE.
  */
 
-/* $Id: socket.c,v 1.66 2008/08/20 23:57:59 jinmei Exp $ */
+/* $Id: socket.c,v 1.67 2008/09/04 05:56:43 marka Exp $ */
 
 /* This code has been rewritten to take advantage of Windows Sockets
  * I/O Completion Ports and Events. I/O Completion Ports is ONLY
@@ -4006,3 +4006,9 @@ void *
 isc_socket_gettag(isc_socket_t *socket) {
 	return (socket->tag);
 }
+
+void
+isc__socketmgr_setreserved(isc_socketmgr_t *manager, isc_uint32_t reserved) {
+	UNUSED(manager);
+	UNUSED(reserved);
+}

lib/isccfg/namedconf.c

@@ -15,7 +15,7 @@
  * PERFORMANCE OF THIS SOFTWARE.
  */
 
-/* $Id: namedconf.c,v 1.89 2008/06/23 23:47:11 tbox Exp $ */
+/* $Id: namedconf.c,v 1.90 2008/09/04 05:56:43 marka Exp $ */
 
 /*! \file */
 
@@ -692,6 +692,7 @@ options_clauses[] = {
 	{ "recursing-file", &cfg_type_qstring, 0 },
 	{ "random-device", &cfg_type_qstring, 0 },
 	{ "recursive-clients", &cfg_type_uint32, 0 },
+	{ "reserved-sockets", &cfg_type_uint32, 0 },
 	{ "serial-queries", &cfg_type_uint32, CFG_CLAUSEFLAG_OBSOLETE },
 	{ "serial-query-rate", &cfg_type_uint32, 0 },
 	{ "server-id", &cfg_type_serverid, 0 },