OSDN Git Service

ipv4: guard IP_MINTTL with a static key
authorEric Dumazet <edumazet@google.com>
Mon, 25 Oct 2021 16:48:24 +0000 (09:48 -0700)
committerJakub Kicinski <kuba@kernel.org>
Tue, 26 Oct 2021 01:02:14 +0000 (18:02 -0700)
RFC 5082 IP_MINTTL option is rarely used on hosts.

Add a static key to remove from TCP fast path useless code,
and potential cache line miss to fetch inet_sk(sk)->min_ttl

Note that once ip4_min_ttl static key has been enabled,
it stays enabled until next boot.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Soheil Hassas Yeganeh <soheil@google.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
include/net/ip.h
net/ipv4/ip_sockglue.c
net/ipv4/tcp_ipv4.c

index cf229a5..b71e885 100644 (file)
@@ -24,6 +24,7 @@
 #include <linux/skbuff.h>
 #include <linux/jhash.h>
 #include <linux/sockptr.h>
+#include <linux/static_key.h>
 
 #include <net/inet_sock.h>
 #include <net/route.h>
@@ -750,6 +751,7 @@ void ip_cmsg_recv_offset(struct msghdr *msg, struct sock *sk,
                         struct sk_buff *skb, int tlen, int offset);
 int ip_cmsg_send(struct sock *sk, struct msghdr *msg,
                 struct ipcm_cookie *ipc, bool allow_ipv6);
+DECLARE_STATIC_KEY_FALSE(ip4_min_ttl);
 int ip_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
                  unsigned int optlen);
 int ip_getsockopt(struct sock *sk, int level, int optname, char __user *optval,
index d5487c8..38d29b1 100644 (file)
@@ -886,6 +886,8 @@ static int compat_ip_mcast_join_leave(struct sock *sk, int optname,
        return ip_mc_leave_group(sk, &mreq);
 }
 
+DEFINE_STATIC_KEY_FALSE(ip4_min_ttl);
+
 static int do_ip_setsockopt(struct sock *sk, int level, int optname,
                sockptr_t optval, unsigned int optlen)
 {
@@ -1352,6 +1354,10 @@ static int do_ip_setsockopt(struct sock *sk, int level, int optname,
                        goto e_inval;
                if (val < 0 || val > 255)
                        goto e_inval;
+
+               if (val)
+                       static_branch_enable(&ip4_min_ttl);
+
                /* tcp_v4_err() and tcp_v4_rcv() might read min_ttl
                 * while we are changint it.
                 */
index a9cbc8e..13d868c 100644 (file)
@@ -508,10 +508,12 @@ int tcp_v4_err(struct sk_buff *skb, u32 info)
        if (sk->sk_state == TCP_CLOSE)
                goto out;
 
-       /* min_ttl can be changed concurrently from do_ip_setsockopt() */
-       if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
-               __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
-               goto out;
+       if (static_branch_unlikely(&ip4_min_ttl)) {
+               /* min_ttl can be changed concurrently from do_ip_setsockopt() */
+               if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
+                       __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
+                       goto out;
+               }
        }
 
        tp = tcp_sk(sk);
@@ -2070,10 +2072,12 @@ process:
                }
        }
 
-       /* min_ttl can be changed concurrently from do_ip_setsockopt() */
-       if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
-               __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
-               goto discard_and_relse;
+       if (static_branch_unlikely(&ip4_min_ttl)) {
+               /* min_ttl can be changed concurrently from do_ip_setsockopt() */
+               if (unlikely(iph->ttl < READ_ONCE(inet_sk(sk)->min_ttl))) {
+                       __NET_INC_STATS(net, LINUX_MIB_TCPMINTTLDROP);
+                       goto discard_and_relse;
+               }
        }
 
        if (!xfrm4_policy_check(sk, XFRM_POLICY_IN, skb))