possible rose recvmsg behavior change after msg_namelen fixes

possible rose recvmsg behavior change after msg_namelen fixes

[Florian Westphal]

Hi.

recvmsg handler in net/rose/af_rose.c performs size-check
on  ->msg_namelen:

if (msg->msg_namelen >= sizeof(struct full_sockaddr_rose)) {
 /* something */
} else {
  /* else */
}

Now that msg_namelen is zeroed before the handler calls we always
hit the 'else' branch.

TBH i don't comprehend how this was/is supposed to work.

Could someone who is familiar with AF_ROSE look at this, before
I send untested patch that blindly chops off code?

Thanks!

Re: possible rose recvmsg behavior change after msg_namelen fixes

[Hannes Frederic Sowa]

On Fri, Dec 20, 2013 at 05:02:07PM +0100, Florian Westphal wrote:
> Hi.
> 
> recvmsg handler in net/rose/af_rose.c performs size-check
> on  ->msg_namelen:
> 
> if (msg->msg_namelen >= sizeof(struct full_sockaddr_rose)) {
>  /* something */
> } else {
>   /* else */
> }
> 
> Now that msg_namelen is zeroed before the handler calls we always
> hit the 'else' branch.
> 
> TBH i don't comprehend how this was/is supposed to work.
> 
> Could someone who is familiar with AF_ROSE look at this, before
> I send untested patch that blindly chops off code?

Good catch, thanks!

If in doubt I would just replace msg->msg_namelen with sizeof(struct
sockaddr_storage) and do manual constant folding. ;)

I still have some patches in this area (somehow validating msg_namelen before
calling down to recvmsg, so that we don't dequeue a skb if the msg_name logic
would error out when writing the result back to user space).

The patch is still a mess but maybe we can pass the actual user-supplied
msg_namelen down then, eventually.