Re: [PATCH net-next v1] rtnetlink: Return error when message too short

[Donald Hunter <donald.hunter@gmail.com>]

Jakub Kicinski <kuba@kernel.org> writes:

> On Mon, 20 Mar 2023 23:18:34 +0000 Donald Hunter wrote:
>> rtnetlink_rcv_msg currently returns 0 when the message length is too
>> short. This leads to either no response at all, or an ack response
>> if NLM_F_ACK was set in the request.
>>
>> Change rtnetlink_rcv_msg to return -EINVAL which tells af_netlink to
>> generate a proper error response.
>
> It's a touch risky to start returning an error now.
> Some application could possibly have been passing an empty netlink
> message just because.

It seemed harmless enough to me, but you make a good point.

> We should give the user a heads up (pr_warn_once() with the name+pid
> of current process). Or continue returning a 0 but add a warning via
> the extack. The latter is cleaner but will not help old / sloppy apps,
> your call.

A pr_warn_once() should be enough to help the next person and also to
give visibility of any sloppy apps.

I was considering pr_warn_once() for the nlmsghdr length check in
netlink_rcv_skb since there's nowhere to reply to when the header is
malformed. 

>> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
>
> We can't put a Fixes tag on it. It could break uAPI, we don't want
> it backported for sure.