[zebra/linux] Use BPF to filter out responses, to try avoid netlink overruns

2008-05-29 Stephen Hemminger <stephen.hemminger@vyatta.com>

	* rt_netlink.c: (netlink_install_filter) BPF filter to catch and
	  drop responses to zebra's own route messages.
	  (kernel_init) add BPF filter on the netlink socket.

3 files changed, 62 insertions, 1 deletions

diff --git a/lib/zebra.h b/lib/zebra.h
index 150aa2c5..2716460f 100644
--- a/lib/zebra.h
+++ b/lib/zebra.h
@@ -162,6 +162,8 @@ typedef int socklen_t;
 #ifdef HAVE_NETLINK
 #include <linux/netlink.h>
 #include <linux/rtnetlink.h>
+#include <linux/filter.h>
+#include <stddef.h>
 #else
 #define RT_TABLE_MAIN		0
 #endif /* HAVE_NETLINK */
diff --git a/zebra/ChangeLog b/zebra/ChangeLog
index d9cae283..6f6dfa2c 100644
--- a/zebra/ChangeLog
+++ b/zebra/ChangeLog
@@ -1,3 +1,9 @@
+2008-05-29 Stephen Hemminger <stephen.hemminger@vyatta.com>
+
+	* rt_netlink.c: (netlink_install_filter) BPF filter to catch and
+	  drop responses to zebra's own route messages.
+	  (kernel_init) add BPF filter on the netlink socket.
+
 2008-02-26 Denis Ovsienko
 	* zebra_rib.[ch]: (rib_lookup_and_pushup) New function, which makes sure,
 	  that if_set_prefix() has nothing in its way of assigning an address.
diff --git a/zebra/rt_netlink.c b/zebra/rt_netlink.c
index 5b592f94..0bf2d9eb 100644
--- a/zebra/rt_netlink.c
+++ b/zebra/rt_netlink.c
@@ -1938,6 +1938,56 @@ kernel_read (struct thread *thread)
   return 0;
 }
 
+/* Filter out messages from self that occur on listener socket */
+static void netlink_install_filter (int sock)
+{
+  /*
+   * Filter is equivalent to netlink_route_change
+   *
+   * if (h->nlmsg_type == RTM_DELROUTE || h->nlmsg_type == RTM_NEWROUTE) {
+   *    if (rtm->rtm_type != RTM_UNICAST)
+   *    	return 0;
+   *    if (rtm->rtm_flags & RTM_F_CLONED)
+   *    	return 0;
+   *    if (rtm->rtm_protocol == RTPROT_REDIRECT)
+   *    	return 0;
+   *    if (rtm->rtm_protocol == RTPROT_KERNEL)
+   *        return 0;
+   *    if (rtm->rtm_protocol == RTPROT_ZEBRA && h->nlmsg_type == RTM_NEWROUTE)
+   * 	return 0;
+   * }
+   * return 0xffff;
+   */
+  struct sock_filter filter[] = {
+    /* 0*/ BPF_STMT(BPF_LD|BPF_ABS|BPF_H, offsetof(struct nlmsghdr, nlmsg_type)),
+    /* 1*/ BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, htons(RTM_DELROUTE), 1, 0),
+    /* 2*/ BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, htons(RTM_NEWROUTE), 0, 11),
+    /* 3*/ BPF_STMT(BPF_LD|BPF_ABS|BPF_B,
+		    sizeof(struct nlmsghdr) + offsetof(struct rtmsg, rtm_type)),
+    /* 4*/ BPF_JUMP(BPF_JMP|BPF_B, RTN_UNICAST, 0, 8),
+    /* 5*/ BPF_STMT(BPF_LD|BPF_ABS|BPF_B,
+		    sizeof(struct nlmsghdr) + offsetof(struct rtmsg, rtm_flags)),
+    /* 6*/ BPF_JUMP(BPF_JMP|BPF_JSET|BPF_K, RTM_F_CLONED, 6, 0),
+    /* 7*/ BPF_STMT(BPF_LD|BPF_ABS|BPF_B,
+		    sizeof(struct nlmsghdr) + offsetof(struct rtmsg, rtm_protocol)),
+    /* 8*/ BPF_JUMP(BPF_JMP+ BPF_B, RTPROT_REDIRECT, 4, 0),
+    /* 9*/ BPF_JUMP(BPF_JMP+ BPF_B, RTPROT_KERNEL, 0, 1),
+    /*10*/ BPF_JUMP(BPF_JMP+ BPF_B, RTPROT_ZEBRA, 0, 3),
+    /*11*/ BPF_STMT(BPF_LD|BPF_ABS|BPF_H, offsetof(struct nlmsghdr, nlmsg_type)),
+    /*12*/ BPF_JUMP(BPF_JMP|BPF_JEQ|BPF_K, htons(RTM_NEWROUTE), 0, 1),
+    /*13*/ BPF_STMT(BPF_RET|BPF_K, 0),		/* drop */
+    /*14*/ BPF_STMT(BPF_RET|BPF_K, 0xffff),	/* keep */
+  };
+
+  struct sock_fprog prog = {
+    .len = sizeof(filter) / sizeof(filter[0]),
+    .filter = filter,
+  };
+
+  if (setsockopt(sock, SOL_SOCKET, SO_ATTACH_FILTER, &prog, sizeof(prog)) < 0)
+    zlog_warn ("Can't install socket filter: %s\n", safe_strerror(errno));
+}
+
 /* Exported interface function.  This function simply calls
    netlink_socket (). */
 void
@@ -1954,5 +2004,8 @@ kernel_init (void)
 
   /* Register kernel socket. */
   if (netlink.sock > 0)
-    thread_add_read (zebrad.master, kernel_read, NULL, netlink.sock);
+    {
+      netlink_install_filter (netlink.sock);
+      thread_add_read (zebrad.master, kernel_read, NULL, netlink.sock);
+    }
 }