[PATCH net-next] inetpeer: avoid false sharing in inet_peer_xrlim_allow()

[PATCH net-next] inetpeer: avoid false sharing in inet_peer_xrlim_allow()

[Eric Dumazet]

Under DOS, inet_peer_xrlim_allow() might be called millions
of times per second from different cpus.

Make sure to write over peer->rate_tokens and peer->rate_last
only when really needed.

Note the inherent races of this function are still there,
we do not care of precise ICMP rate limiting.

Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv4/inetpeer.c | 18 +++++++++++-------
 1 file changed, 11 insertions(+), 7 deletions(-)

diff --git a/net/ipv4/inetpeer.c b/net/ipv4/inetpeer.c
index e02484f4d22b8ea47cbaeed46c5fb0a7411462a1..b8b23a77ceb4f0f1a3d3adaacea2a7c59a7da3c9 100644
--- a/net/ipv4/inetpeer.c
+++ b/net/ipv4/inetpeer.c
@@ -246,23 +246,27 @@ void inet_putpeer(struct inet_peer *p)
 #define XRLIM_BURST_FACTOR 6
 bool inet_peer_xrlim_allow(struct inet_peer *peer, int timeout)
 {
-	unsigned long now, token;
+	unsigned long now, token, otoken, delta;
 	bool rc = false;
 
 	if (!peer)
 		return true;
 
-	token = peer->rate_tokens;
+	token = otoken = READ_ONCE(peer->rate_tokens);
 	now = jiffies;
-	token += now - peer->rate_last;
-	peer->rate_last = now;
-	if (token > XRLIM_BURST_FACTOR * timeout)
-		token = XRLIM_BURST_FACTOR * timeout;
+	delta = now - READ_ONCE(peer->rate_last);
+	if (delta) {
+		WRITE_ONCE(peer->rate_last, now);
+		token += delta;
+		if (token > XRLIM_BURST_FACTOR * timeout)
+			token = XRLIM_BURST_FACTOR * timeout;
+	}
 	if (token >= timeout) {
 		token -= timeout;
 		rc = true;
 	}
-	peer->rate_tokens = token;
+	if (token != otoken)
+		WRITE_ONCE(peer->rate_tokens, token);
 	return rc;
 }
 EXPORT_SYMBOL(inet_peer_xrlim_allow);
-- 
2.47.1.613.gc27f4b7a9f-goog

Re: [PATCH net-next] inetpeer: avoid false sharing in inet_peer_xrlim_allow()

[Willem de Bruijn]

Eric Dumazet wrote:
> Under DOS, inet_peer_xrlim_allow() might be called millions
> of times per second from different cpus.
> 
> Make sure to write over peer->rate_tokens and peer->rate_last
> only when really needed.
> 
> Note the inherent races of this function are still there,
> we do not care of precise ICMP rate limiting.
> 
> Signed-off-by: Eric Dumazet <edumazet@google.com>

Reviewed-by: Willem de Bruijn <willemb@google.com>

Re: [PATCH net-next] inetpeer: avoid false sharing in inet_peer_xrlim_allow()

[patchwork-bot+netdevbpf]

Hello:

This patch was applied to netdev/net-next.git (main)
by Jakub Kicinski <kuba@kernel.org>:

On Thu, 19 Dec 2024 15:03:30 +0000 you wrote:
> Under DOS, inet_peer_xrlim_allow() might be called millions
> of times per second from different cpus.
> 
> Make sure to write over peer->rate_tokens and peer->rate_last
> only when really needed.
> 
> Note the inherent races of this function are still there,
> we do not care of precise ICMP rate limiting.
> 
> [...]

Here is the summary with links:
  - [net-next] inetpeer: avoid false sharing in inet_peer_xrlim_allow()
    https://git.kernel.org/netdev/net-next/c/05dd04b218f4

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html