Subj : Issues with multi-threading and fork
To   : comp.programming.threads
From : jmarant
Date : Thu Mar 24 2005 06:36 am

Hi,

I'm writing a multi-threaded application.

A subset of the threads in the application do fork children.
Here is basically what is being done:

Child:
-----

change process user and group
change directory
redirect standard output and error to a trace file
execv a shell script

Parent:
------

busy-wait for the child process to terminate by the means of:

cnt =3D timeout;
while (waitpid (pid , &status, WNOHANG) =3D=3D 0 && cnt !=3D 0)
{
  sleep (1);
  --cnt;
}

(I didn't find any other way to implement timeouted waitpid in
a mutil-threaded context since one cannot use signals)


>From time to times, the timeout mysteriously expires, so I searched
for the causes.

After some investigation, it seems that the child process hangs right
after the fork and does not go further.
'strace -f' shows that it is stuck in a rt_sigsuspend following
a sigprocmask(SIG_SETMASK, NULL, ..) which could correspond to
__pthread_wait_for_restart_signal, after some diging into the
linuxthreads sources.

Some additional notes:
- The parent process locks mutexes but the child process does not
use them, so I think that using pthread_atfork is unnecessary.
- I do not use async-signal-unsafe functions within signals and
the child is not interrupted by any signal
- I removed calls to syslog because it is said to be
async-signal-unsafe
but it does not change anything.

I spent some time searching the Web, and followed some tracks without
any success (anything about threading, forking and
async-signal-safety).

Any idea about what might be happening?

Many thanks in advance.

--
J=E9r=F4me Marant

.