tech-net archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: route_protosw.pr_input

Le 07/09/2018 à 11:02, Ryota Ozaki a écrit :
On Fri, Sep 7, 2018 at 4:01 PM Maxime Villard <> wrote:
Le 06/09/2018 à 10:45, Ryota Ozaki a écrit :
On Wed, Sep 5, 2018 at 4:37 PM Maxime Villard <> wrote:
In net/rtsock.c, I don't understand where we use the 'pr_input' field of
COMPATNAME(route_protosw)[]. To me it can't be used: .pr_input takes a
variadic function, but if you grep through the tree, you can see that
the variadic .pr_input functions we call always take three arguments, as
opposed to raw_input() which takes five.

Therefore raw_input() is never called via .pr_input, and we can set the
field to NULL. Then we can switch all the .pr_input functions to be static
and not variadic.

This will clear a lot of confusion, and will allow us to find problems
related to types -- like this one.


No objection.

And a next step would be to change the signature of protosw#pr_input
to that of ip6protosw#pr_input and then we can merge protosw and
ip6protosw into one.

Is it a really good idea? To me it would be better if the IPv4 pr_input
stayed as

         void x_input(struct mbuf *m, int off, int proto)

I don't see a lot of point in wanting to mimic IPv6; IPv6 has constraints
that don't apply to IPv4. Eg the second arg is a pointer, because of the
format of the IPv6 options -- but that doesn't apply to IPv4.

I just prefer having just one structure to having two similar structures,
which reduces some codes.

(Well, and FreeBSD / OpenBSD / Dragonfly BSD did the change.)

In this particular case I don't think it's a really good move. The IPv6
code is already a bit difficult to track because of *offp, so if we start
doing the same for IPv4 there will be blood.

We can still make the change later if it turns out to be really better.

Home | Main Index | Thread Index | Old Index