Persistence of file implemented objects

Persistence of file implemented objects

[Richard H Lee]

Various objects like fifo pipes, POSIX IPC shared memory and semaphores 
that are implemented through the filesystem in cygwin persist when the 
program abnornally terminates. They also persist through reboots, which 
is different to the behaviour on linux. This is also different with the 
case of sysV IPC + cygserver on cygwin which clears after reboots.

If there anyway to clear out orphaned objects or to make sure they are 
released in the event of an abnormal termination?

Does the POSIX specification even require the release of these objects 
in these situations?

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Persistence of file implemented objects

[Christopher Faylor]

On Sat, Jun 30, 2012 at 11:29:10PM +0100, Richard H Lee wrote:
>Various objects like fifo pipes, POSIX IPC shared memory and semaphores 
>that are implemented through the filesystem in cygwin persist when the 
>program abnornally terminates. They also persist through reboots, which 
>is different to the behaviour on linux. This is also different with the 
>case of sysV IPC + cygserver on cygwin which clears after reboots.

Please be more precise about what you are talking about.

Fifos persist on reboot on Linux or Cygwin.  They live on the
filesystem.  I don't see how POSIX IPC shared memory and semaphores
could persist.

cgf

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Persistence of file implemented objects

[Richard H Lee]

> Fifos persist on reboot on Linux or Cygwin.  They live on the
> filesystem.  I don't see how POSIX IPC shared memory and semaphores
> could persist.

Sorry, I meant unix/bsd sockets.

Regarding the POSIX IPC's, they are stored in /dev . In regular *nix, 
/dev do not represent "physical" files on the filesystem, hence they do 
not persist over boot.

In cygwin, they actually do represent physical files. So if they are not 
freed correctly by the program, the persist over to the next boot.

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Persistence of file implemented objects

[Christopher Faylor]

On Sun, Jul 01, 2012 at 07:17:15PM +0100, Richard H Lee wrote:
>> Fifos persist on reboot on Linux or Cygwin.  They live on the
>> filesystem.  I don't see how POSIX IPC shared memory and semaphores
>> could persist.
>
>Sorry, I meant unix/bsd sockets.

...which also persist on linux.

>Regarding the POSIX IPC's, they are stored in /dev . In regular *nix, 
>/dev do not represent "physical" files on the filesystem, hence they do 
>not persist over boot.
>
>In cygwin, they actually do represent physical files. So if they are not 
>freed correctly by the program, the persist over to the next boot.

what specific interface are you talking about which creates physical
files in /dev?

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

Re: Persistence of file implemented objects

[Corinna Vinschen]

On Jul  1 19:17, Richard H Lee wrote:
> >Fifos persist on reboot on Linux or Cygwin.  They live on the
> >filesystem.  I don't see how POSIX IPC shared memory and semaphores
> >could persist.
> 
> Sorry, I meant unix/bsd sockets.

AF_UNIX/AF_LOCAL sockets are implemented on filesystem level as well,
not only on Cygwin.  If they are not explicitely unlinked, they will be
in the way when trying to create a new socket of the same name.

> Regarding the POSIX IPC's, they are stored in /dev . In regular
> *nix, /dev do not represent "physical" files on the filesystem,
> hence they do not persist over boot.
> 
> In cygwin, they actually do represent physical files. So if they are
> not freed correctly by the program, the persist over to the next
> boot.

Right.  The named POSIX IPC objects are created as normal files under
/dev/shm and /dev/mqueue.  These objects are supposed to follow POSIX
file name naming rules, and they are supposed to be persistent until you
call mq_unlink, sem_unlink, or shm_unlink respectively.  They are also
meant to persist if a process crashes(*).

So in contrast to XSI IPC implemented by running cygserver they also
persist when all Cygwin processes have stopped and even over reboot,
because they don't live on a virtual filesystem like on Linux, but on
the real filesystem.

From the POSIX point of view that's ok.  If you have a problems with
that, you might consider to create some autostart script which deletes
all files below /dev/shm and /dev/mqueue.


Corinna


(*) See http://pubs.opengroup.org/onlinepubs/007908799/xsh/mq_open.html
        http://pubs.opengroup.org/onlinepubs/007908799/xsh/sem_open.html
	http://pubs.opengroup.org/onlinepubs/007908799/xsh/shm_open.html

-- 
Corinna Vinschen                  Please, send mails regarding Cygwin to
Cygwin Project Co-Leader          cygwin AT cygwin DOT com
Red Hat

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple