Re: [PATCH net] ipv6/gro: fix an out of bounds memory bug in ipv6_gro_receive()

[Ziyang Xuan (William)]

> On Thu, Oct 27, 2022 at 6:01 AM Ziyang Xuan (William)
> <william.xuanziyang@xxxxxxxxxx> wrote:
>>
>>> On Thu, Oct 27, 2022 at 3:25 AM Ziyang Xuan
>>> <william.xuanziyang@xxxxxxxxxx> wrote:
>>>>
>>>> IPv6 packets without NEXTHDR_NONE extension header can make continuous
>>>> __skb_pull() until pskb_may_pull() failed in ipv6_gso_pull_exthdrs().
>>>> That results in a big value of skb_gro_offset(), and after __skb_push()
>>>> in ipv6_gro_receive(), skb->data will less than skb->head, an out of
>>>> bounds memory bug occurs. That will trigger the problem as following:
>>>>
>>>> ==================================================================
>>>> BUG: KASAN: use-after-free in eth_type_trans+0x100/0x260
>>>> ...
>>>> Call trace:
>>>>  dump_backtrace+0xd8/0x130
>>>>  show_stack+0x1c/0x50
>>>>  dump_stack_lvl+0x64/0x7c
>>>>  print_address_description.constprop.0+0xbc/0x2e8
>>>>  print_report+0x100/0x1e4
>>>>  kasan_report+0x80/0x120
>>>>  __asan_load8+0x78/0xa0
>>>>  eth_type_trans+0x100/0x260
>>>
>>> Crash happens from eth_type_trans() , this should happen before
>>> ipv6_gro_receive() ?
>>>
>>> It seems your patch is unrelated.
>>>
>>> Please provide a repro.
>>
>> C repro put in attachment.
> 
> This seems to be a bug in tun device.
> 
> Please take more time to root cause this issue, instead of adding work
> arounds all over the place.

Hi Eric,

Thank you for your suggestion.

I have analyzed the problem more deeply. The odd IPv6 packet and
big packet length value(IPv6 payload length more than 65535)
together cause the problem.

skb->network_header and skb->transport_header are all u16 type.
They would occuer overflow errors during ipv6_gro_receive() processing.
That cause the value error for __skb_push(skb, value).

So the problem is a bug in tun device.

I will combine my previous problem "net: tun: limit first seg size to avoid oversized linearization"
together to give the fix patch later.

Thanks.

> 
> Thanks.
> 
>>
>>>
>>>
>>>>  napi_gro_frags+0x164/0x550
>>>>  tun_get_user+0xda4/0x1270
>>>>  tun_chr_write_iter+0x74/0x130
>>>>  do_iter_readv_writev+0x130/0x1ec
>>>>  do_iter_write+0xbc/0x1e0
>>>>  vfs_writev+0x13c/0x26c
>>>>
>>>> Add comparison between skb->data - skb_gro_offset() and skb->head
>>>> and exception handler before __skb_push() to fix the bug.
>>>>
>>>> Fixes: 86911732d399 ("gro: Avoid copying headers of unmerged packets")
>>>> Signed-off-by: Ziyang Xuan <william.xuanziyang@xxxxxxxxxx>
>>>> ---
>>>>  net/ipv6/ip6_offload.c | 4 ++++
>>>>  1 file changed, 4 insertions(+)
>>>>
>>>> diff --git a/net/ipv6/ip6_offload.c b/net/ipv6/ip6_offload.c
>>>> index 3ee345672849..6659ccf25387 100644
>>>> --- a/net/ipv6/ip6_offload.c
>>>> +++ b/net/ipv6/ip6_offload.c
>>>> @@ -237,6 +237,10 @@ INDIRECT_CALLABLE_SCOPE struct sk_buff *ipv6_gro_receive(struct list_head *head,
>>>>                 proto = ipv6_gso_pull_exthdrs(skb, proto);
>>>>                 skb_gro_pull(skb, -skb_transport_offset(skb));
>>>>                 skb_reset_transport_header(skb);
>>>> +               if (unlikely(skb_headroom(skb) < skb_gro_offset(skb))) {
>>>
>>> This makes no sense to me.
>>>
>>> If there is a bug, it should be fixed earlier.
>>
>> Maybe it is good to validate IPv6 packet earlier in ipv6_gro_receive() or more earlier?
>>
>>>
>>>> +                       kfree_skb(skb);
>>>> +                       return ERR_PTR(-EINPROGRESS);
>>>> +               }
>>>>                 __skb_push(skb, skb_gro_offset(skb));
>>>>
>>>>                 ops = rcu_dereference(inet6_offloads[proto]);
>>>> --
>>>> 2.25.1
>>>>
>>> .
>>>
> .
>