close() behavior when there are multiple fds for a socket

close() behavior when there are multiple fds for a socket

[Ben Pfaff]

I guess this isn't the best place to ask this question, but I think there
are knowledgeable people here. I don't know the right place, so I'd
welcome being redirected to a better one.

The POSIX spec for close() is here:
https://pubs.opengroup.org/onlinepubs/9699919799/

This specification has a lot of cases that explicitly apply only to the
last close of an object, either using that language or by saying that
something happens when "all file descriptors associated with" an
object are closed. But it doesn't say that for sockets. Instead, it
just says, "If fildes refers to a socket, close() shall cause the
socket to be destroyed," which implies that closing any file
descriptor for a socket destroys the socket. I think that must be
wrong, because I don't think that's the behavior on the systems
I've used.

Do I misunderstand the spec? Or the behavior on real systems?
(Should I submit a request for clarification to the spec maintainers?)

Thanks,

Ben.

Re: close() behavior when there are multiple fds for a socket

[Paul Eggert]

On 10/14/21 16:03, Ben Pfaff wrote:

> https://pubs.opengroup.org/onlinepubs/9699919799/
> 
> This specification has a lot of cases that explicitly apply only to the
> last close of an object, either using that language or by saying that
> something happens when "all file descriptors associated with" an
> object are closed. But it doesn't say that for sockets.

Yes, and you're right that this is a typo.


> (Should I submit a request for clarification to the spec maintainers?)

Sure, and you can do so as described here:

https://www.opengroup.org/austin/

(look for "Defect Reports").

Re: close() behavior when there are multiple fds for a socket

[Ben Pfaff]

On Thu, Oct 14, 2021 at 4:09 PM Paul Eggert <eggert@cs.ucla.edu> wrote:
>
> On 10/14/21 16:03, Ben Pfaff wrote:
>
> > https://pubs.opengroup.org/onlinepubs/9699919799/
> >
> > This specification has a lot of cases that explicitly apply only to the
> > last close of an object, either using that language or by saying that
> > something happens when "all file descriptors associated with" an
> > object are closed. But it doesn't say that for sockets.
>
> Yes, and you're right that this is a typo.

Thanks.

> > (Should I submit a request for clarification to the spec maintainers?)
>
> Sure, and you can do so as described here:
>
> https://www.opengroup.org/austin/
>
> (look for "Defect Reports").

That's harder than it should be. To get a login to their bug tracker,
you have to email the chair of the group. The chair of the group
appears to be Andrew Josey, but he hides his email so you can't
find it. Anyway, after some web searches I think I found his real
email address, so I'll report it once he creates my account.

Thanks again!

Re: close() behavior when there are multiple fds for a socket

[Paul Eggert]

On 10/14/21 16:53, Ben Pfaff wrote:
> The chair of the group
> appears to be Andrew Josey, but he hides his email

Sorry, I should have told you it's Andrew Josey <ajosey@opengroup.org>.

Re: close() behavior when there are multiple fds for a socket

[Ben Pfaff]

On Thu, Oct 14, 2021 at 4:55 PM Paul Eggert <eggert@cs.ucla.edu> wrote:
>
> On 10/14/21 16:53, Ben Pfaff wrote:
> > The chair of the group
> > appears to be Andrew Josey, but he hides his email
>
> Sorry, I should have told you it's Andrew Josey <ajosey@opengroup.org>.

Thanks!

What a bureaucratic organization! The bug report form, once I got a login,
requires one to specify a page number and line number, but it seems that
you can only get the PDF version of the spec that has that information  if
you belong to an organization that is a member, so I filled those in with "?"
and am hoping for the best.

Re: close() behavior when there are multiple fds for a socket

[Eric Blake]

On Fri, Oct 15, 2021 at 09:10:04AM -0700, Ben Pfaff wrote:
> What a bureaucratic organization! The bug report form, once I got a login,
> requires one to specify a page number and line number, but it seems that
> you can only get the PDF version of the spec that has that information  if
> you belong to an organization that is a member, so I filled those in with "?"
> and am hoping for the best.

That's fine.  Adding page numbers after the fact is something that
happens rather frequently; and I've done in for your bug in particular
just now.

-- 
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3266
Virtualization:  qemu.org | libvirt.org

Re: close() behavior when there are multiple fds for a socket

[Ben Pfaff]

On Fri, Oct 15, 2021 at 11:40 AM Eric Blake <eblake@redhat.com> wrote:
> On Fri, Oct 15, 2021 at 09:10:04AM -0700, Ben Pfaff wrote:
> > What a bureaucratic organization! The bug report form, once I got a login,
> > requires one to specify a page number and line number, but it seems that
> > you can only get the PDF version of the spec that has that information  if
> > you belong to an organization that is a member, so I filled those in with "?"
> > and am hoping for the best.
>
> That's fine.  Adding page numbers after the fact is something that
> happens rather frequently; and I've done in for your bug in particular
> just now.

Thanks so much!

Re: close() behavior when there are multiple fds for a socket

[Bernhard Voelker]

On 10/15/21 20:41, Ben Pfaff wrote:
> On Fri, Oct 15, 2021 at 11:40 AM Eric Blake <eblake@redhat.com> wrote:
>> On Fri, Oct 15, 2021 at 09:10:04AM -0700, Ben Pfaff wrote:
>>> What a bureaucratic organization! The bug report form, once I got a login,
>>> requires one to specify a page number and line number, but it seems that
>>> you can only get the PDF version of the spec that has that information  if
>>> you belong to an organization that is a member, so I filled those in with "?"
>>> and am hoping for the best.
>>
>> That's fine.  Adding page numbers after the fact is something that
>> happens rather frequently; and I've done in for your bug in particular
>> just now.
> 
> Thanks so much!

Would you mind adding the opengroup bug here for reference?

Thanks & have a nice day,
Berny

Re: close() behavior when there are multiple fds for a socket

[Ben Pfaff]

On Thu, Oct 21, 2021 at 3:02 PM Bernhard Voelker
<mail@bernhard-voelker.de> wrote:
> Would you mind adding the opengroup bug here for reference?

Here you go:
https://www.austingroupbugs.net/view.php?id=1525