Why ECONNREFUSED from unix_dgram_sendmsg?

Why ECONNREFUSED from unix_dgram_sendmsg?

[Andy Lutomirski]

What is unix_dgram_sendmsg trying to tell me here?

        if (sock_flag(other, SOCK_DEAD)) {
                /*
                 *      Check with 1003.1g - what should
                 *      datagram error
                 */
                unix_state_unlock(other);
                sock_put(other);

                err = 0;
                unix_state_lock(sk);
                if (unix_peer(sk) == other) {
                        unix_peer(sk) = NULL;
                        unix_state_unlock(sk);

                        unix_dgram_disconnected(sk, other);
                        sock_put(other);
                        err = -ECONNREFUSED;

I'm getting this intermittently on a seqpacket socket when the other
end dies.  Shouldn't this be -ECONNRESET?

git blame doesn't tell me anything here -- this code predates git.

--Andy

Re: Why ECONNREFUSED from unix_dgram_sendmsg?

[Hannes Frederic Sowa]

On Thu, Sep 11, 2014, at 17:11, Andy Lutomirski wrote:
> What is unix_dgram_sendmsg trying to tell me here?
> 
>         if (sock_flag(other, SOCK_DEAD)) {
>                 /*
>                  *      Check with 1003.1g - what should
>                  *      datagram error
>                  */
>                 unix_state_unlock(other);
>                 sock_put(other);
> 
>                 err = 0;
>                 unix_state_lock(sk);
>                 if (unix_peer(sk) == other) {
>                         unix_peer(sk) = NULL;
>                         unix_state_unlock(sk);
> 
>                         unix_dgram_disconnected(sk, other);
>                         sock_put(other);
>                         err = -ECONNREFUSED;
> 
> I'm getting this intermittently on a seqpacket socket when the other
> end dies.  Shouldn't this be -ECONNRESET?
> 
> git blame doesn't tell me anything here -- this code predates git.

Changelogs were not that descriptive as they were today: :)

commit 283fd6cf0be690a83b5dbb34e79130f8592c0765
Author: davem <davem>
Date:   Fri Aug 20 10:59:51 1999 +0000

    Merge in ANK networking jumbo patch,
    which includes ATM from Werner Almesberger, Netfilter
    from Paul Russell, and massive SMP networking enhancements
    by Alexey.

(in
https://git.kernel.org/cgit/linux/kernel/git/davem/netdev-vger-cvs.git/)

I agree, -ECONNRESET would be a better errno value, but this is hard to
change because of backwards compatibility.

Bye,
Hannes

Re: Why ECONNREFUSED from unix_dgram_sendmsg?

[Andy Lutomirski]

On Sep 11, 2014 8:59 AM, "Hannes Frederic Sowa"
<hannes@stressinduktion.org> wrote:
>
>
>
> On Thu, Sep 11, 2014, at 17:11, Andy Lutomirski wrote:
> > What is unix_dgram_sendmsg trying to tell me here?
> >
> >         if (sock_flag(other, SOCK_DEAD)) {
> >                 /*
> >                  *      Check with 1003.1g - what should
> >                  *      datagram error
> >                  */
> >                 unix_state_unlock(other);
> >                 sock_put(other);
> >
> >                 err = 0;
> >                 unix_state_lock(sk);
> >                 if (unix_peer(sk) == other) {
> >                         unix_peer(sk) = NULL;
> >                         unix_state_unlock(sk);
> >
> >                         unix_dgram_disconnected(sk, other);
> >                         sock_put(other);
> >                         err = -ECONNREFUSED;
> >
> > I'm getting this intermittently on a seqpacket socket when the other
> > end dies.  Shouldn't this be -ECONNRESET?
> >
> > git blame doesn't tell me anything here -- this code predates git.
>
> Changelogs were not that descriptive as they were today: :)
>
> commit 283fd6cf0be690a83b5dbb34e79130f8592c0765
> Author: davem <davem>
> Date:   Fri Aug 20 10:59:51 1999 +0000
>
>     Merge in ANK networking jumbo patch,
>     which includes ATM from Werner Almesberger, Netfilter
>     from Paul Russell, and massive SMP networking enhancements
>     by Alexey.
>
> (in
> https://git.kernel.org/cgit/linux/kernel/git/davem/netdev-vger-cvs.git/)
>
> I agree, -ECONNRESET would be a better errno value, but this is hard to
> change because of backwards compatibility.

Might it be safe to change it for just seqpacket sockets?  -ECONRESET
actually makes some sense for dgram sockets, assuming this code is
even reachable for dgram sockets.

Anyway, I'll have to fix my code regardless.

--Andy