Instead of relying on rps_needed, it is safer to use a separate
static key, since we do not want to enable TCP rx_skb_cache
by default. This feature can cause huge increase of memory
usage on hosts with millions of sockets.

Signed-off-by: Eric Dumazet <eduma...@google.com>
---
 Documentation/networking/ip-sysctl.txt | 8 ++++++++
 include/net/sock.h                     | 6 ++----
 net/ipv4/sysctl_net_ipv4.c             | 9 +++++++++
 3 files changed, 19 insertions(+), 4 deletions(-)

diff --git a/Documentation/networking/ip-sysctl.txt 
b/Documentation/networking/ip-sysctl.txt
index 
14fe93049d28e965d7349b03c5c8782c3d386e7d..288aa264ac26d98637a5bb1babc334bfc699bef1
 100644
--- a/Documentation/networking/ip-sysctl.txt
+++ b/Documentation/networking/ip-sysctl.txt
@@ -772,6 +772,14 @@ tcp_challenge_ack_limit - INTEGER
        in RFC 5961 (Improving TCP's Robustness to Blind In-Window Attacks)
        Default: 100
 
+tcp_rx_skb_cache - BOOLEAN
+       Controls a per TCP socket cache of one skb, that might help
+       performance of some workloads. This might be dangerous
+       on systems with a lot of TCP sockets, since it increases
+       memory usage.
+
+       Default: 0 (disabled)
+
 UDP variables:
 
 udp_l3mdev_accept - BOOLEAN
diff --git a/include/net/sock.h b/include/net/sock.h
index 
e9d769c04637a3c0b967c9bfa6def724834796b9..b02645e2dfad722769c1455bcde76e46da9fc5ac
 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -2433,13 +2433,11 @@ static inline void skb_setup_tx_timestamp(struct 
sk_buff *skb, __u16 tsflags)
  * This routine must be called with interrupts disabled or with the socket
  * locked so that the sk_buff queue operation is ok.
 */
+DECLARE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
 static inline void sk_eat_skb(struct sock *sk, struct sk_buff *skb)
 {
        __skb_unlink(skb, &sk->sk_receive_queue);
-       if (
-#ifdef CONFIG_RPS
-           !static_branch_unlikely(&rps_needed) &&
-#endif
+       if (static_branch_unlikely(&tcp_rx_skb_cache_key) &&
            !sk->sk_rx_skb_cache) {
                sk->sk_rx_skb_cache = skb;
                skb_orphan(skb);
diff --git a/net/ipv4/sysctl_net_ipv4.c b/net/ipv4/sysctl_net_ipv4.c
index 
875867b64d6a6597bf4fcd3498ed55741cbe33f7..886b58d31351df44725bdc34081e798bcb89ecf0
 100644
--- a/net/ipv4/sysctl_net_ipv4.c
+++ b/net/ipv4/sysctl_net_ipv4.c
@@ -51,6 +51,9 @@ static int comp_sack_nr_max = 255;
 static u32 u32_max_div_HZ = UINT_MAX / HZ;
 static int one_day_secs = 24 * 3600;
 
+DEFINE_STATIC_KEY_FALSE(tcp_rx_skb_cache_key);
+EXPORT_SYMBOL(tcp_rx_skb_cache_key);
+
 /* obsolete */
 static int sysctl_tcp_low_latency __read_mostly;
 
@@ -559,6 +562,12 @@ static struct ctl_table ipv4_table[] = {
                .extra1         = &sysctl_fib_sync_mem_min,
                .extra2         = &sysctl_fib_sync_mem_max,
        },
+       {
+               .procname       = "tcp_rx_skb_cache",
+               .data           = &tcp_rx_skb_cache_key.key,
+               .mode           = 0644,
+               .proc_handler   = proc_do_static_key,
+       },
        { }
 };
 
-- 
2.22.0.410.gd8fdbe21b5-goog

Reply via email to