Merge pull request #5108 from donaldsharp/sendbuffer_size_bgp

Sendbuffer size bgp

bgpd/bgp_main.c

@@ -75,6 +75,7 @@ static const struct option longopts[] = {
 	{"ecmp", required_argument, NULL, 'e'},
 	{"int_num", required_argument, NULL, 'I'},
 	{"no_zebra", no_argument, NULL, 'Z'},
+	{"socket_size", required_argument, NULL, 's'},
 	{0}};
 
 /* signal definitions */
@@ -386,17 +387,19 @@ int main(int argc, char **argv)
 	int no_zebra_flag = 0;
 	int skip_runas = 0;
 	int instance = 0;
+	int buffer_size = BGP_SOCKET_SNDBUF_SIZE;
 
 	frr_preinit(&bgpd_di, argc, argv);
 	frr_opt_add(
-		"p:l:SnZe:I:" DEPRECATED_OPTIONS, longopts,
+		"p:l:SnZe:I:s:" DEPRECATED_OPTIONS, longopts,
 		"  -p, --bgp_port     Set BGP listen port number (0 means do not listen).\n"
 		"  -l, --listenon     Listen on specified address (implies -n)\n"
 		"  -n, --no_kernel    Do not install route to kernel.\n"
 		"  -Z, --no_zebra     Do not communicate with Zebra.\n"
 		"  -S, --skip_runas   Skip capabilities checks, and changing user and group IDs.\n"
 		"  -e, --ecmp         Specify ECMP to use.\n"
-		"  -I, --int_num      Set instance number (label-manager)\n");
+		"  -I, --int_num      Set instance number (label-manager)\n"
+		"  -s, --socket_size  Set BGP peer socket send buffer size\n");
 
 	/* Command line argument treatment. */
 	while (1) {
@@ -452,6 +455,9 @@ int main(int argc, char **argv)
 				zlog_err("Instance %i out of range (0..%u)",
 					 instance, (unsigned short)-1);
 			break;
+		case 's':
+			buffer_size = atoi(optarg);
+			break;
 		default:
 			frr_help_exit(1);
 			break;
@@ -461,7 +467,7 @@ int main(int argc, char **argv)
 		memset(&bgpd_privs, 0, sizeof(bgpd_privs));
 
 	/* BGP master init. */
-	bgp_master_init(frr_init());
+	bgp_master_init(frr_init(), buffer_size);
 	bm->port = bgp_port;
 	if (bgp_port == 0)
 		bgp_option_set(BGP_OPT_NO_LISTEN);

bgpd/bgp_network.c

@@ -320,6 +320,14 @@ static int bgp_get_instance_for_inc_conn(int sock, struct bgp **bgp_inst)
 #endif
 }
 
+static void bgp_socket_set_buffer_size(const int fd)
+{
+	if (getsockopt_so_sendbuf(fd) < (int)bm->socket_buffer)
+		setsockopt_so_sendbuf(fd, bm->socket_buffer);
+	if (getsockopt_so_recvbuf(fd) < (int)bm->socket_buffer)
+		setsockopt_so_recvbuf(fd, bm->socket_buffer);
+}
+
 /* Accept bgp connection. */
 static int bgp_accept(struct thread *thread)
 {
@@ -371,8 +379,7 @@ static int bgp_accept(struct thread *thread)
 		return -1;
 	}
 
-	/* Set socket send buffer size */
-	setsockopt_so_sendbuf(bgp_sock, BGP_SOCKET_SNDBUF_SIZE);
+	bgp_socket_set_buffer_size(bgp_sock);
 
 	/* Check remote IP address */
 	peer1 = peer_lookup(bgp, &su);
@@ -621,8 +628,7 @@ int bgp_connect(struct peer *peer)
 
 	set_nonblocking(peer->fd);
 
-	/* Set socket send buffer size */
-	setsockopt_so_sendbuf(peer->fd, BGP_SOCKET_SNDBUF_SIZE);
+	bgp_socket_set_buffer_size(peer->fd);
 
 	if (bgp_set_socket_ttl(peer, peer->fd) < 0)
 		return -1;

bgpd/bgpd.c

@@ -7823,7 +7823,7 @@ int bgp_config_write(struct vty *vty)
 	return 0;
 }
 
-void bgp_master_init(struct thread_master *master)
+void bgp_master_init(struct thread_master *master, const int buffer_size)
 {
 	qobj_init();
 
@@ -7838,6 +7838,7 @@ void bgp_master_init(struct thread_master *master)
 	bm->t_rmap_update = NULL;
 	bm->rmap_update_timer = RMAP_DEFAULT_UPDATE_TIMER;
 	bm->terminating = false;
+	bm->socket_buffer = buffer_size;
 
 	bgp_process_queue_init();
 

bgpd/bgpd.h

@@ -156,6 +156,9 @@ struct bgp_master {
 	/* BGP-EVPN VRF ID. Defaults to default VRF (if any) */
 	struct bgp* bgp_evpn;
 
+	/* How big should we set the socket buffer size */
+	uint32_t socket_buffer;
+
 	bool terminating;	/* global flag that sigint terminate seen */
 	QOBJ_FIELDS
 };
@@ -1575,7 +1578,8 @@ extern char *peer_uptime(time_t uptime2, char *buf, size_t len, bool use_json,
 
 extern int bgp_config_write(struct vty *);
 
-extern void bgp_master_init(struct thread_master *master);
+extern void bgp_master_init(struct thread_master *master,
+			    const int buffer_size);
 
 extern void bgp_init(unsigned short instance);
 extern void bgp_pthreads_run(void);

doc/manpages/bgpd.rst

@@ -21,6 +21,48 @@ OPTIONS available for the |DAEMON| command:
 
 .. include:: common-options.rst
 
+.. option:: -p, --bgp_port <port>
+
+   Set the bgp protocol's port number. When port number is 0, that means do not
+   listen bgp port.
+
+.. option:: -l, --listenon
+
+   Specify a specific IP address for bgpd to listen on, rather than its default
+   of ``0.0.0.0`` / ``::``. This can be useful to constrain bgpd to an internal
+   address, or to run multiple bgpd processes on one host.
+
+.. option:: -n, --no_kernel
+
+   Do not install learned routes into the linux kernel.  This option is useful
+   for a route-reflector environment or if you are running multiple bgp
+   processes in the same namespace.  This option is different than the --no_zebra
+   option in that a ZAPI connection is made.
+
+.. option:: -S, --skip_runas
+
+   Skip the normal process of checking capabilities and changing user and group
+   information.
+
+.. option:: -e, --ecmp
+
+   Run BGP with a limited ecmp capability, that is different than what BGP
+   was compiled with.  The value specified must be greater than 0 and less
+   than or equal to the MULTIPATH_NUM specified on compilation.
+
+.. option:: -Z, --no_zebra
+
+   Do not communicate with zebra at all.  This is different than the --no_kernel
+   option in that we do not even open a ZAPI connection to the zebra process.
+
+.. option:: -s, --socket_size
+
+   When opening tcp connections to our peers, set the socket send buffer
+   size that the kernel will use for the peers socket.  This option
+   is only really useful at a very large scale.  Experimentation should
+   be done to see if this is helping or not at the scale you are running
+   at.
+
 LABEL MANAGER
 -------------
 

doc/user/bgp.rst

@@ -35,6 +35,44 @@ be specified (:ref:`common-invocation-options`).
    of ``0.0.0.0`` / ``::``. This can be useful to constrain bgpd to an internal
    address, or to run multiple bgpd processes on one host.
 
+.. option:: -n, --no_kernel
+
+   Do not install learned routes into the linux kernel.  This option is useful
+   for a route-reflector environment or if you are running multiple bgp
+   processes in the same namespace.  This option is different than the --no_zebra
+   option in that a ZAPI connection is made.
+
+.. option:: -S, --skip_runas
+
+   Skip the normal process of checking capabilities and changing user and group
+   information.
+
+.. option:: -e, --ecmp
+
+   Run BGP with a limited ecmp capability, that is different than what BGP
+   was compiled with.  The value specified must be greater than 0 and less
+   than or equal to the MULTIPATH_NUM specified on compilation.
+
+.. option:: -Z, --no_zebra
+
+   Do not communicate with zebra at all.  This is different than the --no_kernel
+   option in that we do not even open a ZAPI connection to the zebra process.
+
+.. option:: -s, --socket_size
+
+   When opening tcp connections to our peers, set the socket send buffer
+   size that the kernel will use for the peers socket.  This option
+   is only really useful at a very large scale.  Experimentation should
+   be done to see if this is helping or not at the scale you are running
+   at.
+
+LABEL MANAGER
+-------------
+
+.. option:: -I, --int_num
+
+   Set zclient id. This is required when using Zebra label manager in proxy mode.
+
 .. _bgp-basic-concepts:
 
 Basic Concepts

lib/sockopt.c

@@ -72,6 +72,21 @@ int getsockopt_so_sendbuf(const int sock)
 	return optval;
 }
 
+int getsockopt_so_recvbuf(const int sock)
+{
+	uint32_t optval;
+	socklen_t optlen = sizeof(optval);
+	int ret = getsockopt(sock, SOL_SOCKET, SO_RCVBUF, (char *)&optval,
+			     &optlen);
+	if (ret < 0) {
+		flog_err_sys(EC_LIB_SYSTEM_CALL,
+			     "fd %d: can't getsockopt SO_RCVBUF: %d (%s)", sock,
+			     errno, safe_strerror(errno));
+		return ret;
+	}
+	return optval;
+}
+
 static void *getsockopt_cmsg_data(struct msghdr *msgh, int level, int type)
 {
 	struct cmsghdr *cmsg;

lib/sockopt.h

@@ -30,6 +30,7 @@ extern "C" {
 extern void setsockopt_so_recvbuf(int sock, int size);
 extern void setsockopt_so_sendbuf(const int sock, int size);
 extern int getsockopt_so_sendbuf(const int sock);
+extern int getsockopt_so_recvbuf(const int sock);
 
 extern int setsockopt_ipv6_pktinfo(int, int);
 extern int setsockopt_ipv6_checksum(int, int);

tests/bgpd/test_aspath.c

@@ -1339,7 +1339,7 @@ int main(void)
 {
 	int i = 0;
 	qobj_init();
-	bgp_master_init(thread_master_create(NULL));
+	bgp_master_init(thread_master_create(NULL), BGP_SOCKET_SNDBUF_SIZE);
 	master = bm->master;
 	bgp_option_set(BGP_OPT_NO_LISTEN);
 	bgp_attr_init();

tests/bgpd/test_capability.c

@@ -912,7 +912,7 @@ int main(void)
 
 	qobj_init();
 	master = thread_master_create(NULL);
-	bgp_master_init(master);
+	bgp_master_init(master, BGP_SOCKET_SNDBUF_SIZE);
 	vrf_init(NULL, NULL, NULL, NULL, NULL);
 	bgp_option_set(BGP_OPT_NO_LISTEN);
 

tests/bgpd/test_mp_attr.c

@@ -37,6 +37,7 @@
 #include "bgpd/bgp_mplsvpn.h"
 #include "bgpd/bgp_nexthop.h"
 #include "bgpd/bgp_vty.h"
+#include "bgpd/bgp_network.h"
 
 #define VT100_RESET "\x1b[0m"
 #define VT100_RED "\x1b[31m"
@@ -1078,7 +1079,7 @@ int main(void)
 	cmd_init(0);
 	bgp_vty_init();
 	master = thread_master_create("test mp attr");
-	bgp_master_init(master);
+	bgp_master_init(master, BGP_SOCKET_SNDBUF_SIZE);
 	vrf_init(NULL, NULL, NULL, NULL, NULL);
 	bgp_option_set(BGP_OPT_NO_LISTEN);
 	bgp_attr_init();

tests/bgpd/test_mpath.c

@@ -38,6 +38,7 @@
 #include "bgpd/bgp_nexthop.h"
 #include "bgpd/bgp_mpath.h"
 #include "bgpd/bgp_evpn.h"
+#include "bgpd/bgp_network.h"
 
 #define VT100_RESET "\x1b[0m"
 #define VT100_RED "\x1b[31m"
@@ -379,7 +380,7 @@ static int global_test_init(void)
 	qobj_init();
 	master = thread_master_create(NULL);
 	zclient = zclient_new(master, &zclient_options_default);
-	bgp_master_init(master);
+	bgp_master_init(master, BGP_SOCKET_SNDBUF_SIZE);
 	vrf_init(NULL, NULL, NULL, NULL, NULL);
 	bgp_option_set(BGP_OPT_NO_LISTEN);
 

tests/bgpd/test_packet.c

@@ -34,6 +34,7 @@
 #include "bgpd/bgp_debug.h"
 #include "bgpd/bgp_packet.h"
 #include "bgpd/bgp_aspath.h"
+#include "bgpd/bgp_network.h"
 
 /* need these to link in libbgp */
 struct zebra_privs_t *bgpd_privs = NULL;
@@ -58,7 +59,7 @@ int main(int argc, char *argv[])
 	qobj_init();
 	bgp_attr_init();
 	master = thread_master_create(NULL);
-	bgp_master_init(master);
+	bgp_master_init(master, BGP_SOCKET_SNDBUF_SIZE);
 	vrf_init(NULL, NULL, NULL, NULL, NULL);
 	bgp_option_set(BGP_OPT_NO_LISTEN);
 

tests/bgpd/test_peer_attr.c

@@ -29,6 +29,7 @@
 #include "bgpd/bgp_route.h"
 #include "bgpd/bgp_vty.h"
 #include "bgpd/bgp_zebra.h"
+#include "bgpd/bgp_network.h"
 
 #ifdef ENABLE_BGP_VNC
 #include "bgpd/rfapi/rfapi_backend.h"
@@ -1388,7 +1389,7 @@ static void bgp_startup(void)
 	master = thread_master_create(NULL);
 	yang_init();
 	nb_init(master, NULL, 0);
-	bgp_master_init(master);
+	bgp_master_init(master, BGP_SOCKET_SNDBUF_SIZE);
 	bgp_option_set(BGP_OPT_NO_LISTEN);
 	vrf_init(NULL, NULL, NULL, NULL, NULL);
 	frr_pthread_init();