Re: [Patch net v2] mlx5: fixup checksum for short ethernet frame padding

[Cong Wang <xiyou.wangcong@gmail.com>]

On Wed, Nov 28, 2018 at 7:00 AM Eric Dumazet <edumazet@google.com> wrote:
>
> Nice packet of death alert.
>
> pad_len can be 0xFFFFFF67  here, if frame_len is smaller than pad_offset.

Unless IP header is malformed, how could it be?

Speaking of IP header sanity, I am totally aware of it, I don't check it because
I know get_ip_proto() doesn't check either, it must be hardware which verifies
the sanity.


>
> Really I suggest you set ip_summed to CHECKSUM_NONE, then remove the
> initial test ( if (likely(frame_len > ETH_ZLEN)) ...)
>
> Until the firmware is fixed.

Hmm, why setting to CHECKSUM_NONE could get rid of the minimum ethernet
frame check? I am lost, there is no bug for packet > ETH_ZLEN _for me_, what
needs to fix here?

Overall, you keep pushing me to fix a bug I don't observe. I don't understand
why. If you see it, please come up with your own patch? Why do I have to fix
the problem you see??


>
> Otherwise frames with a wrong checksum and some non zero padding could
> potentially
> be seen as correct frames. (Probability of 1/65536)
>
> Do not focus on your immediate problem (small packets being padded by
> a non malicious entity)

Again, why _I_ should fix a problem I never observe? Why is it not you who
fix the problem you find during code review? No to mention I have no environment
to test it even if I really want to fix. I can' take such a risk.

Thanks.