[PATCH net] act_ct: prepare for stolen verdict coming from conntrack and nat engine

[PATCH net] act_ct: prepare for stolen verdict coming from conntrack and nat engine

[Florian Westphal]

At this time, conntrack either returns NF_ACCEPT or NF_DROP.
To improve debuging it would be nice to be able to replace NF_DROP verdict
with NF_DROP_REASON() helper,

This helper releases the skb instantly (so drop_monitor can pinpoint
exact location) and returns NF_STOLEN.

Prepare call sites to deal with this before introducing such changes
in conntrack and nat core.

Signed-off-by: Florian Westphal <fw@strlen.de>
---
 './tdc.py -c ct' passes for me.

 net/sched/act_ct.c | 31 +++++++++++++++++++++++++------
 1 file changed, 25 insertions(+), 6 deletions(-)

diff --git a/net/sched/act_ct.c b/net/sched/act_ct.c
index 2a96d9c1db65..a6b7c514a181 100644
--- a/net/sched/act_ct.c
+++ b/net/sched/act_ct.c
@@ -944,6 +944,8 @@ static int tcf_ct_act_nat(struct sk_buff *skb,
 		action |= BIT(NF_NAT_MANIP_DST);
 
 	err = nf_ct_nat(skb, ct, ctinfo, &action, range, commit);
+	if (err != NF_ACCEPT)
+		return err & NF_VERDICT_MASK;
 
 	if (action & BIT(NF_NAT_MANIP_SRC))
 		tc_skb_cb(skb)->post_ct_snat = 1;
@@ -1035,7 +1037,7 @@ TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
 		state.pf = family;
 		err = nf_conntrack_in(skb, &state);
 		if (err != NF_ACCEPT)
-			goto out_push;
+			goto nf_error;
 	}
 
 do_nat:
@@ -1047,7 +1049,7 @@ TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
 
 	err = tcf_ct_act_nat(skb, ct, ctinfo, p->ct_action, &p->range, commit);
 	if (err != NF_ACCEPT)
-		goto drop;
+		goto nf_error;
 
 	if (!nf_ct_is_confirmed(ct) && commit && p->helper && !nfct_help(ct)) {
 		err = __nf_ct_try_assign_helper(ct, p->tmpl, GFP_ATOMIC);
@@ -1061,8 +1063,9 @@ TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
 	}
 
 	if (nf_ct_is_confirmed(ct) ? ((!cached && !skip_add) || add_helper) : commit) {
-		if (nf_ct_helper(skb, ct, ctinfo, family) != NF_ACCEPT)
-			goto drop;
+		err = nf_ct_helper(skb, ct, ctinfo, family);
+		if (err != NF_ACCEPT)
+			goto nf_error;
 	}
 
 	if (commit) {
@@ -1075,8 +1078,9 @@ TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
 		/* This will take care of sending queued events
 		 * even if the connection is already confirmed.
 		 */
-		if (nf_conntrack_confirm(skb) != NF_ACCEPT)
-			goto drop;
+		err = nf_conntrack_confirm(skb);
+		if (err != NF_ACCEPT)
+			goto nf_error;
 	}
 
 	if (!skip_add)
@@ -1100,6 +1104,21 @@ TC_INDIRECT_SCOPE int tcf_ct_act(struct sk_buff *skb, const struct tc_action *a,
 drop:
 	tcf_action_inc_drop_qstats(&c->common);
 	return TC_ACT_SHOT;
+
+nf_error:
+	/* some verdicts store extra data in upper bits, such
+	 * as errno or queue number.
+	 */
+	switch (err & NF_VERDICT_MASK) {
+	case NF_DROP:
+		goto drop;
+	case NF_STOLEN:
+		tcf_action_inc_drop_qstats(&c->common);
+		return TC_ACT_CONSUMED;
+	default:
+		DEBUG_NET_WARN_ON_ONCE(1);
+		goto drop;
+	}
 }
 
 static const struct nla_policy ct_policy[TCA_CT_MAX + 1] = {
-- 
2.44.2

Re: [PATCH net] act_ct: prepare for stolen verdict coming from conntrack and nat engine

[Florian Westphal]

Florian Westphal <fw@strlen.de> wrote:
> At this time, conntrack either returns NF_ACCEPT or NF_DROP.
> To improve debuging it would be nice to be able to replace NF_DROP verdict
> with NF_DROP_REASON() helper,

Sigh, this patch targets *net-next*, not net.

Re: [PATCH net] act_ct: prepare for stolen verdict coming from conntrack and nat engine

[Paolo Abeni]

On Wed, 2024-07-03 at 13:13 +0200, Florian Westphal wrote:
> Florian Westphal <fw@strlen.de> wrote:
> > At this time, conntrack either returns NF_ACCEPT or NF_DROP.
> > To improve debuging it would be nice to be able to replace NF_DROP verdict
> > with NF_DROP_REASON() helper,
> 
> Sigh, this patch targets *net-next*, not net.

Given that the current ML traffic is manageable, would you mind re-
submitting for net-next, so it goes through the correct CI?

Thanks!

Paolo