On Fri, Dec 26, 2003 at 09:13:36PM -0500, Pierre A. Humblet wrote:
>At 05:59 PM 12/26/2003 -0500, Christopher Faylor wrote:
>>I added strace debugging. You can use that as a clue for what I was
>>talking about and make it visible if you can't run exim under strace.
>
>I put a try_to_debug(1). Waiting for something to happen.
Thanks, for debugging this but that wasn't specifically what I was
talking about. If you go back and read the message, you'll see that I
was talking about your 'get_proc_lock' problem.
>>>Also, while running sysinternals I noticed to the exim daemon still had tty
>>>related handles, despite setsid(). Ditto for inetd.
>>
>>Calling setsid does not automatically eliminate tty handles.
>
>This is what inetd does:
> if (setsid() == -1)
> return (-1);
>
> if (!noclose && (fd = open(PATH_DEVNULL, O_RDWR, 0)) != -1) {
> (void)dup2(fd, STDIN_FILENO);
> (void)dup2(fd, STDOUT_FILENO);
> (void)dup2(fd, STDERR_FILENO);
>*******
>Before your last changes, strace was showing
> 160 282749 [main] inetd 34569887 close: close (2)
> 361 283110 [main] inetd 34569887 fhandler_tty_slave::close: decremented open_fhs
> 0, archetype usecount 2
> 198 283308 [main] inetd 34569887 fhandler_tty_slave::close: just exiting because
> archetype usecount is > 0
> 163 283471 [main] inetd 34569887 close: 0 = close (2)
> 160 283631 [main] inetd 34569887 dtable::dup2: 2 = dup2 (3, 2)
>*******
>
>Now usecount is even higher
>
> 159 292878 [main] inetd 1086339 close: close (2)
> 181 293059 [main] inetd 1086339 fhandler_tty_slave::close: decremented open_fhs 0,
> archetype usecount 3
> 313 293372 [main] inetd 1086339 fhandler_tty_slave::close: just returning because
> archetype usecount is > 0
> 175 293547 [main] inetd 1086339 close: 0 = close (2)
> 158 293705 [main] inetd 1086339 dtable::dup2: 2 = dup2 (3, 2)
>
>
>I looked at your latest change in syscalls.cc.
>It seems to me that setsid should simply call close all the time if (cygheap->ctty)
>and let close() take care of usecount.
I had already checked in a change to this effect but I missed...
>Ditto in pinfo::set_ctty where a problem similar to the one you
>addressed also exists.
this. Thanks. I've checked in a new fix as well as a similar change
for close_all_files.
>I don't think that will take care of everything, but I can't follow
>what's happening when tracing sh -c inetd
I tried the current CVS version and I don't see any stray tty garbage
with inetd. I never tried this with an older snapshot, however, so I
don't know if I would have been lucky before. I did try a much simpler
test case which worked incorrectly with CYGWIN=tty and correctly after
today's initial setsid change.
cgf
--
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple
Problem reports: http://cygwin.com/problems.html
Documentation: http://cygwin.com/docs.html
FAQ: http://cygwin.com/faq/
