Re: patch: sockaddr instead of mbuf to carry addresses

[Tyler Retzlaff <rtr%netbsd.org@localhost>]

On 3/25/2015 7:27 AM, Alan Barrett wrote:
> On Tue, 24 Mar 2015, Christos Zoulas wrote:
>
> I like the idea of making struct sockaddr_storage the largest of
> all the sockaddr_foo types.

As it stands today it already is.

> sockaddr_storage is used in struct ifreq, which is used in a bunch of
> ioctls.  It is also mentioned in a few other places in /usr/include/**.
> If we increased the size of struct sockaddr_storage, then we'd need to
> provide binary compatibility, which is tedious but not difficult in most
> cases.
>
> The difficult cases are that struct sockaddr_storage is embedded in
> struct utmpx and struct lastlogx, which are stored on disk in
> /var/run/utmpx, /var/log/wtmpx, and /var/log/lastlogx.  I don't see an
> easy way for a process reading the file to be able to tell whether it
> uses the old or new struct sockaddr_storage.

After thinking about it for a couple of days I don't see how changing 
the size of sockaddr_storage has any benefit.  As you point out doing so 
would require even more change for ABI compatibility.

The only gain appears to be avoiding definition of a new structure that 
would be used only in the kernel.  Maybe there's more?

In the absence of any outright objection to a new structure I think I'd 
prefer to stay with what I've done in the patch.

* It allows us to use a single structure consistently within the kernel.
* It has no user-space visible change.
* It decouples the kernel from the user-space structure definitions.

regards,

rtr