Skip to contentSkip to conversation

netfilter: ctnetlink: always include remaining timeout

PATCH In review Florian Westphal 2 participants 3 messages Dec 10, 2020

Subsystems (2)
lore
Share & tools
lore

Florian Westphal

author maintainer
#0 lore raw

PATCH 1/1 netfilter: ctnetlink: always include remaining timeout

DESTROY events do not include the remaining timeout.

Unconditionally including the timeout allows to see if the entry timed timed out or was removed explicitly.

The latter case can happen when a conntrack gets deleted prematurely, e.g. due to a tcp reset, module removal, netdev notifier (nat/masquerade device went down), ctnetlink and so on.

---
Might make sense to further extend nf_ct_delete and also pass a reason code in the future.

 net/netfilter/nf_conntrack_netlink.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)
net/netfilter/nf_conntrack_netlink.c +2 -3
index 3d0fd33be018..3f957769cd72 100644
@@ -778,15 +778,14 @@ ctnetlink_conntrack_event(unsigned int events, struct nf_ct_event *item)
 
 	if (ctnetlink_dump_status(skb, ct) < 0)
 		goto nla_put_failure;
+	if (ctnetlink_dump_timeout(skb, ct) < 0)
+		goto nla_put_failure;
 
 	if (events & (1 << IPCT_DESTROY)) {
 		if (ctnetlink_dump_acct(skb, ct, type) < 0 ||
 		    ctnetlink_dump_timestamp(skb, ct) < 0)
 			goto nla_put_failure;
 	} else {
-		if (ctnetlink_dump_timeout(skb, ct) < 0)
-			goto nla_put_failure;
-
 		if (events & (1 << IPCT_PROTOINFO)
 		    && ctnetlink_dump_protoinfo(skb, ct) < 0)
 			goto nla_put_failure;
Signature · 1 line
2.26.2
#1 lore raw
Quoted 5 lines · Florian Westphal

DESTROY events do not include the remaining timeout.

Unconditionally including the timeout allows to see if the entry timed timed out or was removed explicitly.

Quoted 8 lines

The latter case can happen when a conntrack gets deleted prematurely, e.g. due to a tcp reset, module removal, netdev notifier (nat/masquerade device went down), ctnetlink and so on.

---
Might make sense to further extend nf_ct_delete and also pass a reason code in the future.

IIRC, TCP state is not included in the event, right?

This has been requested many times in the past, to debug connectivity issues too.

Probably extending .to_nlattr to take a bool parameter to specify if this is the destroy event path, then _only_ include the TCP state information there (other TCP information is not relevant and netlink bandwidth is limited from the event path).

Thanks.

author maintainer
#2 lore raw
Quoted 10 lines · Pablo Neira Ayuso
Quoted 9 lines · Florian Westphal

DESTROY events do not include the remaining timeout.

Unconditionally including the timeout allows to see if the entry timed timed out or was removed explicitly.

The latter case can happen when a conntrack gets deleted prematurely, e.g. due to a tcp reset, module removal, netdev notifier (nat/masquerade device went down), ctnetlink and so on.

Quoted 6 lines
Quoted 4 lines

---
Might make sense to further extend nf_ct_delete and also pass a reason code in the future.

IIRC, TCP state is not included in the event, right?

No, protoinfo is only dumped for non-destroy case.

Quoted 7 lines

This has been requested many times in the past, to debug connectivity issues too.

Probably extending .to_nlattr to take a bool parameter to specify if this is the destroy event path, then _only_ include the TCP state information there (other TCP information is not relevant and netlink bandwidth is limited from the event path).

Sounds reasonable, will send a v2.