On Tue, Oct 02, 2018 at 01:06:14PM +0200, Jiri Benc wrote:

On Mon,  1 Oct 2018 17:28:29 -0700, David Ahern wrote:

quoted

Add a new flag, NLM_F_DUMP_PROPER_HDR, for userspace to indicate to the
kernel that it believes it is sending the right header struct for the
dump message type (ifinfomsg, ifaddrmsg, rtmsg, fib_rule_hdr, ...).

Why is this limited to dumps? Other kind of netlink messages contain
the common struct, too. When introducing such mechanism, please make it
generic.

Last time when we were discussing strict checking in netlink, it was
suggested to add a socket option instead of adding NLM flags[1].

I didn't find this in the linked thread. What I find interesting and
convincing is one of Dave's points:

"I'm beginning to wonder if we can just change this unilaterally to
not ignore unrecognized attributes.

I am increasingly certain that things that would "break" we wouldn't
want to succeed anyways." [1]

:)

But a socket option or this header flag both sound acceptable to me. Was
there any more detail on how a socket option would look like, i.e. an
api proposal or something?

[1]: https://marc.info/?l=linux-netdev&m=144522081220166&w=2

It makes a lot of sense: the number of flags is very limited and we'd
run out of them pretty fast. It's not just the header structure that
is currently checked sloppily. It's also attributes, flags in
attributes, etc. We can't assign a flag to all of them.

You should also consider a different name for the flag: it should
reflect what the effect of the flag is. "Proper header" is not an
effect, it's a requirement for the message to pass. The effect is
enforced strict checking of the header.

 Jiri

[1] https://marc.info/?l=linux-netdev&m=144492718118955