[lvc-project] [PATCH] netfilter: ebtables: remove unnecessary NULL check

[lvc-project] [PATCH] netfilter: ebtables: remove unnecessary NULL check

[Igor Artemiev]

In ebt_do_table() 'private->chainstack' cannot be NULL
and the 'cs' pointer is dereferenced below, so it does not make
sense to compare 'private->chainstack' with NULL. 

Found by Linux Verification Center (linuxtesting.org) with SVACE.

Signed-off-by: Igor Artemiev <Igor.A.Artemiev@mcst.ru>
---
 net/bridge/netfilter/ebtables.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/net/bridge/netfilter/ebtables.c b/net/bridge/netfilter/ebtables.c
index 757ec46fc45a..74daca8a5142 100644
--- a/net/bridge/netfilter/ebtables.c
+++ b/net/bridge/netfilter/ebtables.c
@@ -212,10 +212,7 @@ unsigned int ebt_do_table(void *priv, struct sk_buff *skb,
 	private = table->private;
 	cb_base = COUNTER_BASE(private->counters, private->nentries,
 	   smp_processor_id());
-	if (private->chainstack)
-		cs = private->chainstack[smp_processor_id()];
-	else
-		cs = NULL;
+	cs = private->chainstack[smp_processor_id()];
 	chaininfo = private->hook_entry[hook];
 	nentries = private->hook_entry[hook]->nentries;
 	point = (struct ebt_entry *)(private->hook_entry[hook]->data);
-- 
2.30.2

Re: [lvc-project] [PATCH] netfilter: ebtables: remove unnecessary NULL check

[Florian Westphal]

Igor Artemiev <Igor.A.Artemiev@mcst.ru> wrote:
> In ebt_do_table() 'private->chainstack' cannot be NULL
> and the 'cs' pointer is dereferenced below, so it does not make
> sense to compare 'private->chainstack' with NULL. 

?  Why do you think that?

> +	cs = private->chainstack[smp_processor_id()];

Looks like NULL deref to me.  Did you test this?

Re: [lvc-project] [PATCH] netfilter: ebtables: remove unnecessary NULL check

[Florian Westphal]

Igor A. Artemiev <Igor.A.Artemiev@mcst.ru> wrote:
> On 6/20/23 19:38, Florian Westphal wrote:
> > Igor Artemiev <Igor.A.Artemiev@mcst.ru> wrote:
> > > In ebt_do_table() 'private->chainstack' cannot be NULL
> > > and the 'cs' pointer is dereferenced below, so it does not make
> > > sense to compare 'private->chainstack' with NULL.
> > ?  Why do you think that?
> > 
> The 'cs' pointer is dereferenced below without checking, as it is assumed to
> always be initialized with 'private->chainstack[smp_processor_id()]'.

No, its not.  The dereferencing is conditional, as is the allocation
of the chainstack.

No user defined chains, no chain stack.

With this change, "ebtables-legacy -A INPUT" causes kernel panic.

Re: [lvc-project] [PATCH] netfilter: ebtables: remove unnecessary NULL check

[Igor A. Artemiev]

On 6/20/23 19:38, Florian Westphal wrote:
> Igor Artemiev <Igor.A.Artemiev@mcst.ru> wrote:
>> In ebt_do_table() 'private->chainstack' cannot be NULL
>> and the 'cs' pointer is dereferenced below, so it does not make
>> sense to compare 'private->chainstack' with NULL.
> ?  Why do you think that?
>
The 'cs' pointer is dereferenced below without checking, as it is 
assumed to always be initialized with 
'private->chainstack[smp_processor_id()]'.
>> +	cs = private->chainstack[smp_processor_id()];
> Looks like NULL deref to me.  Did you test this?
>
No, I didn't test this.

Thanks,
Igor