Re: Add a SOCK_DESTROY operation to close sockets from userspace

[Tom Herbert <tom@herbertland.com>]

On Wed, Dec 2, 2015 at 8:30 AM, Lorenzo Colitti <lorenzo@google.com> wrote:
> On Thu, Dec 3, 2015 at 1:12 AM, Tom Herbert <tom@herbertland.com> wrote:
>> The in kernel caller would already have a pointer to the socket so the
>> call would just be sk->sk_prot->destroy(sk). That call should make its
>> way down to same backend function in TCP that the diag path would use.
>> We need this in the kernel for the same reasons you want this in
>> userspace, if a third party hits an unrecoverable error on the socket
>> it needs to signal this condition to the owner of the socket but can't
>> actually close the socket (useful to RDS, KCM, probably
>> TLS-in-kernel).
>
> Oh, I see, yes. You're thinking of a case where multiple protocols
> support this destroy operation. An in-kernel caller has a pointer to a
> sk that it wants to unwedge, but doesn't need to know which destroy
> function to use for which protocol, it just does "if
> (sk->sk_prot->destroy != NULL) sk->sk_prot_destroy(sk)". I think that
> should be a trivial change on top of what I have.

Thanks, and that's a good description for the commit log :-)