Skip to contentSkip to conversation

net/sched: act_api: fix TOCTOU NULL deref on a->goto_chain

PATCH Applied Jamal Hadi Salim 4 participants 6 messages Aug 9 – Aug 12

Subsystems (4)
lore
Share & tools
lore

Jamal Hadi Salim

author maintainer
#0 lore raw

PATCH 1/1 net/sched: act_api: fix TOCTOU NULL deref on a->goto_chain

tcf_action_exec() handles TC_ACT_GOTO_CHAIN by first checking rcu_access_pointer(a->goto_chain) and then calling tcf_action_goto_chain_exec(), which does a second, independent rcu_dereference_bh(a->goto_chain) read and immediately dereferences chain->filter_chain. A concurrent tcf_action_set_ctrlact() (e.g. the gact replace path) can clear a->goto_chain between the two reads, so the second read returns NULL and tcf_action_goto_chain_exec() dereferences NULL.

Fix the race by doing a single rcu_dereference_bh() read of a->goto_chain in tcf_action_exec(), checking it once for NULL, and passing the resulting chain pointer into tcf_action_goto_chain_exec(). This turns the split check/use into a single check/use on one value.

Fixes ee3bbfe806cd ("net/sched: let actions use RCU to access 'goto_chain'")
Reported-by [email protected]
---
 net/sched/act_api.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
net/sched/act_api.c +5 -5
index f141634df214..600b7804befd 100644
@@ -41,11 +41,9 @@ int tcf_dev_queue_xmit(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb))
 }
 EXPORT_SYMBOL_GPL(tcf_dev_queue_xmit);
 
-static void tcf_action_goto_chain_exec(const struct tc_action *a,
+static void tcf_action_goto_chain_exec(const struct tcf_chain *chain,
 				       struct tcf_result *res)
 {
-	const struct tcf_chain *chain = rcu_dereference_bh(a->goto_chain);
-
 	res->goto_tp = rcu_dereference_bh(chain->filter_chain);
 }
 
@@ -1170,12 +1168,14 @@ int tcf_action_exec(struct sk_buff *skb, struct tc_action **actions,
 					return TC_ACT_OK;
 			}
 		} else if (TC_ACT_EXT_CMP(ret, TC_ACT_GOTO_CHAIN)) {
-			if (unlikely(!rcu_access_pointer(a->goto_chain))) {
+			struct tcf_chain *chain = rcu_dereference_bh(a->goto_chain);
+
+			if (unlikely(!chain)) {
 				tcf_set_drop_reason(skb,
 						    SKB_DROP_REASON_TC_CHAIN_NOTFOUND);
 				return TC_ACT_SHOT;
 			}
-			tcf_action_goto_chain_exec(a, res);
+			tcf_action_goto_chain_exec(chain, res);
 		}
 
 		if (ret != TC_ACT_PIPE)
Signature · 1 line
2.34.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 17 lines · Jamal Hadi Salim

tcf_action_exec() handles TC_ACT_GOTO_CHAIN by first checking rcu_access_pointer(a->goto_chain) and then calling tcf_action_goto_chain_exec(), which does a second, independent rcu_dereference_bh(a->goto_chain) read and immediately dereferences chain->filter_chain. A concurrent tcf_action_set_ctrlact() (e.g. the gact replace path) can clear a->goto_chain between the two reads, so the second read returns NULL and tcf_action_goto_chain_exec() dereferences NULL.

Fix the race by doing a single rcu_dereference_bh() read of a->goto_chain in tcf_action_exec(), checking it once for NULL, and passing the resulting chain pointer into tcf_action_goto_chain_exec(). This turns the split check/use into a single check/use on one value.

Fixes ee3bbfe806cd ("net/sched: let actions use RCU to access 'goto_chain'")
Reported-by [email protected]

[...]

Thanks!

Signature · 1 line
davide
maintainer
#2 lore raw
In reply to Jamal Hadi Salim
Quoted 7 lines · On Sun, 9 Aug 2026 05:09:28 -0400 Jamal Hadi Salim wrote:

tcf_action_exec() handles TC_ACT_GOTO_CHAIN by first checking rcu_access_pointer(a->goto_chain) and then calling tcf_action_goto_chain_exec(), which does a second, independent rcu_dereference_bh(a->goto_chain) read and immediately dereferences chain->filter_chain. A concurrent tcf_action_set_ctrlact() (e.g. the gact replace path) can clear a->goto_chain between the two reads, so the second read returns NULL and tcf_action_goto_chain_exec() dereferences NULL.

FWIW *shiko suggests another tweak but looks orthogonal, LMK if you disagree:

https://netdev-ai.bots.linux.dev/sashiko/#/patchset/[email protected]

(the patch is "too fresh" for me to apply right now anyway)

author maintainer
#3 lore raw
Quoted 10 lines · Jakub Kicinski
Quoted 7 lines · On Sun, 9 Aug 2026 05:09:28 -0400 Jamal Hadi Salim wrote:

tcf_action_exec() handles TC_ACT_GOTO_CHAIN by first checking rcu_access_pointer(a->goto_chain) and then calling tcf_action_goto_chain_exec(), which does a second, independent rcu_dereference_bh(a->goto_chain) read and immediately dereferences chain->filter_chain. A concurrent tcf_action_set_ctrlact() (e.g. the gact replace path) can clear a->goto_chain between the two reads, so the second read returns NULL and tcf_action_goto_chain_exec() dereferences NULL.

Quoted 6 lines

FWIW *shiko suggests another tweak but looks orthogonal, LMK if you disagree:

https://netdev-ai.bots.linux.dev/sashiko/#/patchset/[email protected]

(the patch is "too fresh" for me to apply right now anyway)

The concern is valid but pre-existing and orthogonal to this patch (and a lot less severe than the posted fix)
I had this discussion with Paolo: When the sashikos raise a concern on "pre-existing" issues, what should be the reaction?
In general the conclusion was to follow up later if worth it; however, sometimes we need to make a judgement call - if the pointed to issue is serious (and yes, the AI bots are now reading what Sashikos are saying and constructing bug reports) then a v2 is needed. In this case, I was planning to follow up. I will start more actively looking at sashiko reports and analyzing if worth a followup or a v2. I dont know how to do these pw signals, but in case i see it as "needs v2" it won't be worth waiting for one of you guys to comment.

Do we need a written policy somewhere?

cheers,
jamal

maintainer
#4 lore raw
Quoted 18 lines · On Tue, 11 Aug 2026 06:43:32 -0400 Jamal Hadi Salim wrote:
Quoted 10 lines · Jakub Kicinski
Quoted 7 lines · On Sun, 9 Aug 2026 05:09:28 -0400 Jamal Hadi Salim wrote:

tcf_action_exec() handles TC_ACT_GOTO_CHAIN by first checking rcu_access_pointer(a->goto_chain) and then calling tcf_action_goto_chain_exec(), which does a second, independent rcu_dereference_bh(a->goto_chain) read and immediately dereferences chain->filter_chain. A concurrent tcf_action_set_ctrlact() (e.g. the gact replace path) can clear a->goto_chain between the two reads, so the second read returns NULL and tcf_action_goto_chain_exec() dereferences NULL.

Quoted 6 lines

FWIW *shiko suggests another tweak but looks orthogonal, LMK if you disagree:

https://netdev-ai.bots.linux.dev/sashiko/#/patchset/[email protected]

(the patch is "too fresh" for me to apply right now anyway)

Quoted 12 lines

The concern is valid but pre-existing and orthogonal to this patch (and a lot less severe than the posted fix)
I had this discussion with Paolo: When the sashikos raise a concern on "pre-existing" issues, what should be the reaction?
In general the conclusion was to follow up later if worth it; however, sometimes we need to make a judgement call - if the pointed to issue is serious (and yes, the AI bots are now reading what Sashikos are saying and constructing bug reports) then a v2 is needed. In this case, I was planning to follow up. I will start more actively looking at sashiko reports and analyzing if worth a followup or a v2. I dont know how to do these pw signals, but in case i see it as "needs v2" it won't be worth waiting for one of you guys to comment.

Right, I was hoping my question was clear enough. I don't think v2 was needed here either. But we had time to confirm...

Quoted 1 line

Do we need a written policy somewhere?

The only written policy should be that everyone who asks for a written policy in this rapidly changing environment owes maintainers a beer :)

More seriously I tried to float two written policies recently - for net vs net-next and requirements for information in fixes. And each time there was a long discussion and questions. So y'all need to either stop making written policies so painful, or stop asking for the policies. *%$#.

#5 lore raw
In reply to Jamal Hadi Salim
This patch was applied to netdev/net.git (main) by Jakub Kicinski <[email protected]>:

Hello:

This patch was applied to netdev/net.git (main)
by Jakub Kicinski <[email protected]>:

Quoted 9 lines · On Sun, 9 Aug 2026 05:09:28 -0400 you wrote:

tcf_action_exec() handles TC_ACT_GOTO_CHAIN by first checking rcu_access_pointer(a->goto_chain) and then calling tcf_action_goto_chain_exec(), which does a second, independent rcu_dereference_bh(a->goto_chain) read and immediately dereferences chain->filter_chain. A concurrent tcf_action_set_ctrlact() (e.g. the gact replace path) can clear a->goto_chain between the two reads, so the second read returns NULL and tcf_action_goto_chain_exec() dereferences NULL.

[...]

Here is the summary with links:
- [net] net/sched: act_api: fix TOCTOU NULL deref on a->goto_chain

    https://git.kernel.org/netdev/net/c/f60b396ee174

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html