[PATCH net-next] tcp: syncookies: extend validity range

[PATCH net-next] tcp: syncookies: extend validity range

[Eric Dumazet]

From: Eric Dumazet <edumazet@google.com>

Now we allow storing more request socks per listener, we might
hit syncookie mode less often and hit following bug in our stack :

When we send a burst of syncookies, then exit this mode,
tcp_synq_no_recent_overflow() can return false if the ACK packets coming
from clients are coming three seconds after the end of syncookie
episode.

This is a way too strong requirement and conflicts with rest of
syncookie code which allows ACK to be aged up to 2 minutes.

Perfectly valid ACK packets are dropped just because clients might be
in a crowded wifi environment or on another planet.

So let's fix this, and also change tcp_synq_overflow() to not
dirty a cache line for every syncookie we send, as we are under attack.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
As this is an old bug, I chose net-next tree.

 include/net/tcp.h |   38 ++++++++++++++++++++++++--------------
 1 file changed, 24 insertions(+), 14 deletions(-)

diff --git a/include/net/tcp.h b/include/net/tcp.h
index b8ea12880fd9..7ace6acbf5fd 100644
--- a/include/net/tcp.h
+++ b/include/net/tcp.h
@@ -326,18 +326,6 @@ static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
 
 bool tcp_check_oom(struct sock *sk, int shift);
 
-/* syncookies: remember time of last synqueue overflow */
-static inline void tcp_synq_overflow(struct sock *sk)
-{
-	tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
-}
-
-/* syncookies: no recent synqueue overflow on this listening socket? */
-static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
-{
-	unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
-	return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
-}
 
 extern struct proto tcp_prot;
 
@@ -483,13 +471,35 @@ struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
  * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
  * the counter advances immediately after a cookie is generated).
  */
-#define MAX_SYNCOOKIE_AGE 2
+#define MAX_SYNCOOKIE_AGE	2
+#define TCP_SYNCOOKIE_PERIOD	(60 * HZ)
+#define TCP_SYNCOOKIE_VALID	(MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
+
+/* syncookies: remember time of last synqueue overflow
+ * But do not dirty this field too often (once per second is enough)
+ */
+static inline void tcp_synq_overflow(struct sock *sk)
+{
+	unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
+	unsigned long now = jiffies;
+
+	if (time_after(now, last_overflow + HZ))
+		tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
+}
+
+/* syncookies: no recent synqueue overflow on this listening socket? */
+static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
+{
+	unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
+
+	return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
+}
 
 static inline u32 tcp_cookie_time(void)
 {
 	u64 val = get_jiffies_64();
 
-	do_div(val, 60 * HZ);
+	do_div(val, TCP_SYNCOOKIE_PERIOD);
 	return val;
 }
 

Re: [PATCH net-next] tcp: syncookies: extend validity range

[Florian Westphal]

Eric Dumazet <eric.dumazet@gmail.com> wrote:
> From: Eric Dumazet <edumazet@google.com>
> 
> Now we allow storing more request socks per listener, we might
> hit syncookie mode less often and hit following bug in our stack :
> 
> When we send a burst of syncookies, then exit this mode,
> tcp_synq_no_recent_overflow() can return false if the ACK packets coming
> from clients are coming three seconds after the end of syncookie
> episode.

Heh.  Indeed.  This dates back to original from 1997...

> This is a way too strong requirement and conflicts with rest of
> syncookie code which allows ACK to be aged up to 2 minutes.
> 
> Perfectly valid ACK packets are dropped just because clients might be
> in a crowded wifi environment or on another planet.
> 
> So let's fix this, and also change tcp_synq_overflow() to not
> dirty a cache line for every syncookie we send, as we are under attack.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>
> ---
> As this is an old bug, I chose net-next tree.

Looks great, thanks Eric!

Acked-by: Florian Westphal <fw@strlen.de>

Re: [PATCH net-next] tcp: syncookies: extend validity range

[Yuchung Cheng]

On Thu, May 14, 2015 at 2:26 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> From: Eric Dumazet <edumazet@google.com>
>
> Now we allow storing more request socks per listener, we might
> hit syncookie mode less often and hit following bug in our stack :
>
> When we send a burst of syncookies, then exit this mode,
> tcp_synq_no_recent_overflow() can return false if the ACK packets coming
> from clients are coming three seconds after the end of syncookie
> episode.
>
> This is a way too strong requirement and conflicts with rest of
> syncookie code which allows ACK to be aged up to 2 minutes.
>
> Perfectly valid ACK packets are dropped just because clients might be
> in a crowded wifi environment or on another planet.
>
> So let's fix this, and also change tcp_synq_overflow() to not
> dirty a cache line for every syncookie we send, as we are under attack.
>
> Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>

Hopefully this reduces phony LINUX_MIB_SYNCOOKIESFAILED counts!

> ---
> As this is an old bug, I chose net-next tree.
>
>  include/net/tcp.h |   38 ++++++++++++++++++++++++--------------
>  1 file changed, 24 insertions(+), 14 deletions(-)
>
> diff --git a/include/net/tcp.h b/include/net/tcp.h
> index b8ea12880fd9..7ace6acbf5fd 100644
> --- a/include/net/tcp.h
> +++ b/include/net/tcp.h
> @@ -326,18 +326,6 @@ static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
>
>  bool tcp_check_oom(struct sock *sk, int shift);
>
> -/* syncookies: remember time of last synqueue overflow */
> -static inline void tcp_synq_overflow(struct sock *sk)
> -{
> -       tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
> -}
> -
> -/* syncookies: no recent synqueue overflow on this listening socket? */
> -static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
> -{
> -       unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
> -       return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
> -}
>
>  extern struct proto tcp_prot;
>
> @@ -483,13 +471,35 @@ struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
>   * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
>   * the counter advances immediately after a cookie is generated).
>   */
> -#define MAX_SYNCOOKIE_AGE 2
> +#define MAX_SYNCOOKIE_AGE      2
> +#define TCP_SYNCOOKIE_PERIOD   (60 * HZ)
> +#define TCP_SYNCOOKIE_VALID    (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
> +
> +/* syncookies: remember time of last synqueue overflow
> + * But do not dirty this field too often (once per second is enough)
> + */
> +static inline void tcp_synq_overflow(struct sock *sk)
> +{
> +       unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
> +       unsigned long now = jiffies;
> +
> +       if (time_after(now, last_overflow + HZ))
> +               tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
> +}
> +
> +/* syncookies: no recent synqueue overflow on this listening socket? */
> +static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
> +{
> +       unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
> +
> +       return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
> +}
>
>  static inline u32 tcp_cookie_time(void)
>  {
>         u64 val = get_jiffies_64();
>
> -       do_div(val, 60 * HZ);
> +       do_div(val, TCP_SYNCOOKIE_PERIOD);
>         return val;
>  }
>
>
>

Re: [PATCH net-next] tcp: syncookies: extend validity range

[David Miller]

From: Eric Dumazet <eric.dumazet@gmail.com>
Date: Thu, 14 May 2015 14:26:56 -0700

> From: Eric Dumazet <edumazet@google.com>
> 
> Now we allow storing more request socks per listener, we might
> hit syncookie mode less often and hit following bug in our stack :
> 
> When we send a burst of syncookies, then exit this mode,
> tcp_synq_no_recent_overflow() can return false if the ACK packets coming
> from clients are coming three seconds after the end of syncookie
> episode.
> 
> This is a way too strong requirement and conflicts with rest of
> syncookie code which allows ACK to be aged up to 2 minutes.
> 
> Perfectly valid ACK packets are dropped just because clients might be
> in a crowded wifi environment or on another planet.
> 
> So let's fix this, and also change tcp_synq_overflow() to not
> dirty a cache line for every syncookie we send, as we are under attack.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Applied, thanks.