Thread (4 messages) 4 messages, 2 authors, 13d ago

Re: [PATCH net] tcp: ignore locked zero CWND route metric

flat view

From: Weiming Shi <hidden>
Date: 2026-10-04 16:51:40

Eric Dumazet [off-list ref] 于2026年10月4日周日 01:09写道:
quoted hunk ↗ jump to hunk


On 10/3/26 18:10, Eric Dumazet wrote:
quoted

On 10/3/26 17:32, bestswngs@gmail.com wrote:
quoted
From: Weiming Shi <redacted>

A locked RTAX_CWND route metric of zero is copied into the TCP metrics
cache and then into snd_cwnd_clamp. This sets the initial cwnd to zero.
With a cached SSTHRESH metric, a later Reno ACK can reach a divide by
zero in tcp_cong_avoid_ai(). A user with CAP_NET_ADMIN in a user-created
network namespace can trigger the oops.

Ignore a zero cached CWND metric when applying the locked clamp, leaving
the socket's initialized nonzero clamp in place.

   Oops: divide error: 0000 [#1] SMP KASAN NOPTI
   RIP: 0010:tcp_cong_avoid_ai (net/ipv4/tcp_cong.c:482)
   Call Trace:
   tcp_ack (net/ipv4/tcp_input.c:4440)
   tcp_rcv_established (net/ipv4/tcp_input.c:6678)
   tcp_v4_do_rcv (net/ipv4/tcp_ipv4.c:1854)
   tcp_v4_rcv (net/ipv4/tcp_ipv4.c:2248)
   Kernel panic - not syncing: Fatal exception in interrupt
 > Fixes: 51c5d0c4b169 ("tcp: Maintain dynamic metrics in local
cache.")> Reported-by: [off-list ref]
quoted
Assisted-by: LLM
Signed-off-by: Weiming Shi <redacted>
---
  net/ipv4/tcp_metrics.c | 7 +++++--
  1 file changed, 5 insertions(+), 2 deletions(-)
diff --git a/net/ipv4/tcp_metrics.c b/net/ipv4/tcp_metrics.c
index dc0c081fc1f3..ff41804ad086 100644
--- a/net/ipv4/tcp_metrics.c
+++ b/net/ipv4/tcp_metrics.c
@@ -484,8 +484,11 @@ void tcp_init_metrics(struct sock *sk)
          goto reset;
      }
-    if (tcp_metric_locked(tm, TCP_METRIC_CWND))
-        tp->snd_cwnd_clamp = tcp_metric_get(tm, TCP_METRIC_CWND);
+    if (tcp_metric_locked(tm, TCP_METRIC_CWND)) {
+        val = tcp_metric_get(tm, TCP_METRIC_CWND);
Please explain how we can come to this situation?

cwnd can never be zero, so how a save metric could contain zero?
That is to say :

The right fix is in ip_metrics_convert(), one time instead of at each
TCP session init.
diff --git a/net/ipv4/metrics.c b/net/ipv4/metrics.c
index
ad40762a8b38379e4ccdc505e3885adb8dfff763..50a08e592b207e85ccc2bb6f917379de61af4579
100644
--- a/net/ipv4/metrics.c
+++ b/net/ipv4/metrics.c
@@ -54,6 +54,10 @@ static int ip_metrics_convert(struct nlattr *fc_mx,
                         NL_SET_ERR_MSG(extack, "Unknown flag set in
feature mask in metrics attribute");
                         return -EINVAL;
                 }
+               if ((type == RTAX_CWND || type == RTAX_INITCWND) && !val) {
+                       NL_SET_ERR_MSG(extack, "CWND metric must be
greater than zero");
+                       return -EINVAL;
+               }
                 metrics[type - 1] = val;
         }

pw-bot: cr
Thanks for the review. v2 sent.

Best regards,
Weiming Shi
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help