Re: [PATCH] net_sched: check skb_transport_header_was_set() in qdisc_pkt_len_init()

[Matthew Rosato <mjrosato@linux.vnet.ibm.com>]

On 05/20/2013 02:58 PM, Eric Dumazet wrote:
> On Mon, 2013-05-20 at 14:47 -0400, Matthew Rosato wrote:
>> commit 1def9238d4aa2 (net_sched: more precise pkt_len computation) does
>> not check to see if skb_transport_header is valid prior to using it in
>> qdisc_pkt_len_init(), which can lead to a kernel panic if
>> skb_transport_header is not valid but gso_size is nonzero.  This patch
>> adds a check for skb_transport_header_was_set().
>>
>> I managed to hit this scenario by driving a burst of traffic from a
>> qemu guest through a macvtap interface, causing a panic in the host kernel.
>>
>> Signed-off-by: Matthew Rosato <mjrosato@linux.vnet.ibm.com>
>> ---
>>   net/core/dev.c |    2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/net/core/dev.c b/net/core/dev.c
>> index fc1e289..5d0d3af 100644
>> --- a/net/core/dev.c
>> +++ b/net/core/dev.c
>> @@ -2596,7 +2596,7 @@ static void qdisc_pkt_len_init(struct sk_buff *skb)
>>   	/* To get more precise estimation of bytes sent on wire,
>>   	 * we add to pkt_len the headers size of all segments
>>   	 */
>> -	if (shinfo->gso_size)  {
>> +	if (shinfo->gso_size && skb_transport_header_was_set(skb))  {
>>   		unsigned int hdr_len;
>>   		u16 gso_segs = shinfo->gso_segs;
>>
>
> Have you tried a recent kernel ?
>
> I believe this was already discussed.

Mea Culpa.  You are correct, I was missing a macvtap fix that already 
resolves this issue. For the sake of documentation, that fix is 
9b4d669bc06c2 macvtap: set transport header before passing skb to lower 
device.

Applied and verified that problem is resolved.  You can consider this 
patch withdrawn.  Thanks.

>
> Please fix macvtap instead if its still buggy.
>
>