Netlink: Add option to specify netlink socket receive buffer size

Add option 'netlink rx buffer' to specify netlink socket receive buffer
size. Uses SO_RCVBUFFORCE, so it can override rmem_max limit.

Thanks to Trisha Biswas and Michal for the original patches.

1 files changed, 54 insertions, 0 deletions

diff --git a/sysdep/linux/netlink.c b/sysdep/linux/netlink.c
index ccd62f26..f662301d 100644
--- a/sysdep/linux/netlink.c
+++ b/sysdep/linux/netlink.c
@@ -175,6 +175,27 @@ nl_set_strict_dump(struct nl_sock *nl, int strict)
 }
 
 static void
+nl_set_rcvbuf(int fd, uint val)
+{
+  if (setsockopt(fd, SOL_SOCKET, SO_RCVBUFFORCE, &val, sizeof(val)) < 0)
+    log(L_WARN "KRT: Cannot set netlink rx buffer size to %u: %m", val);
+}
+
+static uint
+nl_cfg_rx_buffer_size(struct config *cfg)
+{
+  uint bufsize = 0;
+
+  struct proto_config *pc;
+  WALK_LIST(pc, cfg->protos)
+    if ((pc->protocol == &proto_unix_kernel) && !pc->disabled)
+      bufsize = MAX(bufsize, ((struct krt_config *) pc)->sys.netlink_rx_buffer);
+
+  return bufsize;
+}
+
+
+static void
 nl_open(void)
 {
   nl_open_sock(&nl_scan);
@@ -1976,6 +1997,8 @@ krt_do_scan(struct krt_proto *p UNUSED)	/* CONFIG_ALL_TABLES_AT_ONCE => p is NUL
 
 static sock *nl_async_sk;		/* BIRD socket for asynchronous notifications */
 static byte *nl_async_rx_buffer;	/* Receive buffer */
+static uint nl_async_bufsize;		/* Kernel rx buffer size for the netlink socket */
+static struct config *nl_last_config;	/* For tracking changes to nl_async_bufsize */
 
 static void
 nl_async_msg(struct nlmsghdr *h)
@@ -2111,6 +2134,32 @@ nl_open_async(void)
     bug("Netlink: sk_open failed");
 }
 
+static void
+nl_update_async_bufsize(void)
+{
+  /* No async socket */
+  if (!nl_async_sk)
+    return;
+
+  /* Already reconfigured */
+  if (nl_last_config == config)
+    return;
+
+  /* Update netlink buffer size */
+  uint bufsize = nl_cfg_rx_buffer_size(config);
+  if (bufsize && (bufsize != nl_async_bufsize))
+  {
+    /* Log message for reconfigurations only */
+    if (nl_last_config)
+      log(L_INFO "KRT: Changing netlink rx buffer size to %u", bufsize);
+
+    nl_set_rcvbuf(nl_async_sk->fd, bufsize);
+    nl_async_bufsize = bufsize;
+  }
+
+  nl_last_config = config;
+}
+
 
 /*
  *	Interface to the UNIX krt module
@@ -2139,6 +2188,7 @@ krt_sys_start(struct krt_proto *p)
 
   nl_open();
   nl_open_async();
+  nl_update_async_bufsize();
 
   return 1;
 }
@@ -2146,12 +2196,16 @@ krt_sys_start(struct krt_proto *p)
 void
 krt_sys_shutdown(struct krt_proto *p)
 {
+  nl_update_async_bufsize();
+
   HASH_REMOVE2(nl_table_map, RTH, krt_pool, p);
 }
 
 int
 krt_sys_reconfigure(struct krt_proto *p UNUSED, struct krt_config *n, struct krt_config *o)
 {
+  nl_update_async_bufsize();
+
   return (n->sys.table_id == o->sys.table_id) && (n->sys.metric == o->sys.metric);
 }