zebra: Allow *BSD to specify a receive buffer size

End operator is reporting that they are receiving buffer overruns
when attempting to read from the kernel receive socket.  It is
possible to adjust this size to more modern levels especially
for when the system is under load.  Modify the code base
so that *BSD operators can use the zebra `-s XXX` option
to specify a read buffer.

Additionally setup the default receive buffer size on *BSD
to be 128k instead of the 8k so that FRR does not run into
this issue again.

Fixes: #10666
Signed-off-by: Donald Sharp <sharpd@nvidia.com>
(cherry picked from commit 9fb83b55066e5cafa344d6a049e363fb631683f6)

diff --git a/zebra/kernel_netlink.c b/zebra/kernel_netlink.c
index 08f9782cf7c132c5e13323317bc634e8e4d6c081..ccafc8f0c02bccc72485b18ffccf43bcd070e59b 100644 (file)
--- a/zebra/kernel_netlink.c
+++ b/zebra/kernel_netlink.c
@@ -154,7 +154,6 @@ static const struct message rttype_str[] = {{RTN_UNSPEC, "none"},
                                            {0}};
 
 extern struct thread_master *master;
-extern uint32_t nl_rcvbufsize;
 
 extern struct zebra_privs_t zserv_privs;
 
@@ -251,12 +250,11 @@ static int netlink_recvbuf(struct nlsock *nl, uint32_t newsize)
        /* Try force option (linux >= 2.6.14) and fall back to normal set */
        frr_with_privs(&zserv_privs) {
                ret = setsockopt(nl->sock, SOL_SOCKET, SO_RCVBUFFORCE,
-                                &nl_rcvbufsize,
-                                sizeof(nl_rcvbufsize));
+                                &rcvbufsize, sizeof(rcvbufsize));
        }
        if (ret < 0)
-               ret = setsockopt(nl->sock, SOL_SOCKET, SO_RCVBUF,
-                                &nl_rcvbufsize, sizeof(nl_rcvbufsize));
+               ret = setsockopt(nl->sock, SOL_SOCKET, SO_RCVBUF, &rcvbufsize,
+                                sizeof(rcvbufsize));
        if (ret < 0) {
                flog_err_sys(EC_LIB_SOCKET,
                             "Can't set %s receive buffer size: %s", nl->name,
@@ -1624,11 +1622,11 @@ void kernel_init(struct zebra_ns *zns)
                         errno);
 
        /* Set receive buffer size if it's set from command line */
-       if (nl_rcvbufsize) {
-               netlink_recvbuf(&zns->netlink, nl_rcvbufsize);
-               netlink_recvbuf(&zns->netlink_cmd, nl_rcvbufsize);
-               netlink_recvbuf(&zns->netlink_dplane_out, nl_rcvbufsize);
-               netlink_recvbuf(&zns->netlink_dplane_in, nl_rcvbufsize);
+       if (rcvbufsize) {
+               netlink_recvbuf(&zns->netlink, rcvbufsize);
+               netlink_recvbuf(&zns->netlink_cmd, rcvbufsize);
+               netlink_recvbuf(&zns->netlink_dplane_out, rcvbufsize);
+               netlink_recvbuf(&zns->netlink_dplane_in, rcvbufsize);
        }
 
        /* Set filter for inbound sockets, to exclude events we've generated

diff --git a/zebra/kernel_socket.c b/zebra/kernel_socket.c
index 5ff66f7c63371a8aa9c73d0999e633cef44e610c..499b9cccf6582fc8ea9cf4849943ed2bbd796aca 100644 (file)
--- a/zebra/kernel_socket.c
+++ b/zebra/kernel_socket.c
@@ -1409,6 +1409,9 @@ static int kernel_read(struct thread *thread)
 /* Make routing socket. */
 static void routing_socket(struct zebra_ns *zns)
 {
+       uint32_t default_rcvbuf;
+       socklen_t optlen;
+
        frr_with_privs(&zserv_privs) {
                routing_sock = ns_socket(AF_ROUTE, SOCK_RAW, 0, zns->ns_id);
 
@@ -1443,6 +1446,23 @@ static void routing_socket(struct zebra_ns *zns)
        /*if (fcntl (routing_sock, F_SETFL, O_NONBLOCK) < 0)
          zlog_warn ("Can't set O_NONBLOCK to routing socket");*/
 
+       /*
+        * Attempt to set a more useful receive buffer size
+        */
+       optlen = sizeof(default_rcvbuf);
+       if (getsockopt(routing_sock, SOL_SOCKET, SO_RCVBUF, &default_rcvbuf,
+                      &optlen) == -1)
+               flog_err_sys(EC_LIB_SOCKET,
+                            "routing_sock sockopt SOL_SOCKET SO_RCVBUF");
+       else {
+               for (; rcvbufsize > default_rcvbuf &&
+                      setsockopt(routing_sock, SOL_SOCKET, SO_RCVBUF,
+                                 &rcvbufsize, sizeof(rcvbufsize)) == -1 &&
+                      errno == ENOBUFS;
+                    rcvbufsize /= 2)
+                       ;
+       }
+
        /* kernel_read needs rewrite. */
        thread_add_read(zrouter.master, kernel_read, NULL, routing_sock, NULL);
 }

diff --git a/zebra/main.c b/zebra/main.c
index 079751af0ad3c1a3ef0b55c6875a18abcd909314..6a5bdfed7811d59488d64102aefa8718ccfff50c 100644 (file)
--- a/zebra/main.c
+++ b/zebra/main.c
@@ -78,10 +78,12 @@ int graceful_restart;
 
 bool v6_rr_semantics = false;
 
+/* Receive buffer size for kernel control sockets */
 #ifdef HAVE_NETLINK
-/* Receive buffer size for netlink socket */
-uint32_t nl_rcvbufsize = 4194304;
-#endif /* HAVE_NETLINK */
+uint32_t rcvbufsize = 4194304;
+#else
+uint32_t rcvbufsize = 128 * 1024;
+#endif
 
 #define OPTION_V6_RR_SEMANTICS 2000
 #define OPTION_ASIC_OFFLOAD    2001
@@ -294,9 +296,9 @@ int main(int argc, char **argv)
        frr_preinit(&zebra_di, argc, argv);
 
        frr_opt_add(
-               "baz:e:rK:"
+               "baz:e:rK:s:"
 #ifdef HAVE_NETLINK
-               "s:n"
+               "n"
 #endif
                ,
                longopts,
@@ -308,9 +310,11 @@ int main(int argc, char **argv)
                "  -K, --graceful_restart   Graceful restart at the kernel level, timer in seconds for expiration\n"
                "  -A, --asic-offload       FRR is interacting with an asic underneath the linux kernel\n"
 #ifdef HAVE_NETLINK
-               "  -n, --vrfwnetns          Use NetNS as VRF backend\n"
                "  -s, --nl-bufsize         Set netlink receive buffer size\n"
+               "  -n, --vrfwnetns          Use NetNS as VRF backend\n"
                "      --v6-rr-semantics    Use v6 RR semantics\n"
+#else
+               "  -s,                      Set kernel socket receive buffer size\n"
 #endif /* HAVE_NETLINK */
        );
 
@@ -359,10 +363,10 @@ int main(int argc, char **argv)
                case 'K':
                        graceful_restart = atoi(optarg);
                        break;
-#ifdef HAVE_NETLINK
                case 's':
-                       nl_rcvbufsize = atoi(optarg);
+                       rcvbufsize = atoi(optarg);
                        break;
+#ifdef HAVE_NETLINK
                case 'n':
                        vrf_configure_backend(VRF_BACKEND_NETNS);
                        break;

diff --git a/zebra/zebra_router.h b/zebra/zebra_router.h
index dd788216c77b75589500bc51c759caf8941ff53e..c0eab9cd6eb610e700b03497952c7231a24007c6 100644 (file)
--- a/zebra/zebra_router.h
+++ b/zebra/zebra_router.h
@@ -214,6 +214,7 @@ struct zebra_router {
 #define GRACEFUL_RESTART_TIME 60
 
 extern struct zebra_router zrouter;
+extern uint32_t rcvbufsize;
 
 extern void zebra_router_init(bool asic_offload, bool notify_on_ack);
 extern void zebra_router_cleanup(void);