Re: Possible netfilter-related memory corruption in 2.6.37

[Patrick McHardy <kaber@trash.net>]

[-- Attachment #1: Type: text/plain, Size: 1130 bytes --]

Am 14.02.2011 16:50, schrieb Eric Dumazet:
> Le lundi 14 février 2011 à 16:18 +0100, Jan Engelhardt a écrit :
>> On Monday 2011-02-14 16:11, Eric Dumazet wrote:
>>
>>> Le lundi 14 février 2011 à 16:58 +0200, Avi Kivity a écrit :
>>>> We see severe memory corruption in kvm while used in conjunction with 
>>>> bridge/netfilter.  Enabling slab debugging points the finger at a 
>>>> netfilter chain invoked from the bridge code.
>>>>
>>>> Can someone take a look?
>>>>
>>>> https://bugzilla.kernel.org/show_bug.cgi?id=27052
>>
>> Maybe looks familiar to https://lkml.org/lkml/2011/2/3/147
> 
> Are you sure Jan ?
> 
> IMHO it looks like in your case, a NULL ->hook() is called, from
> nf_iterate()
> 
> BTW, list_for_each_continue_rcu() really should be converted to 
> list_for_each_entry_continue_rcu()
> 
> This is a bit ugly :
> 
> list_for_each_continue_rcu(*i, head) {
> 	struct nf_hook_ops *elem = (struct nf_hook_ops *)*i;
> 
> Also, I wonder if RCU rules are respected in nf_iterate().
> For example this line is really suspicious :
> 
> *i = (*i)->prev;

Yeah, that definitely looks wrong. How about this instead?


[-- Attachment #2: x --]
[-- Type: text/plain, Size: 640 bytes --]

diff --git a/net/netfilter/core.c b/net/netfilter/core.c
index 1e00bf7..899b71c 100644
--- a/net/netfilter/core.c
+++ b/net/netfilter/core.c
@@ -133,6 +133,7 @@ unsigned int nf_iterate(struct list_head *head,
 
 		/* Optimization: we don't need to hold module
 		   reference here, since function can't sleep. --RR */
+repeat:
 		verdict = elem->hook(hook, skb, indev, outdev, okfn);
 		if (verdict != NF_ACCEPT) {
 #ifdef CONFIG_NETFILTER_DEBUG
@@ -145,7 +146,7 @@ unsigned int nf_iterate(struct list_head *head,
 #endif
 			if (verdict != NF_REPEAT)
 				return verdict;
-			*i = (*i)->prev;
+			goto repeat;
 		}
 	}
 	return NF_ACCEPT;