Same problem with the last CVS Snapshot.

Assigned to the author of this stuff.

Same problem with 5.1.3RC4-dev (latest CVS snapshot at the moment) on Linux/i386.

I independently reproduced the bug with the following piece of code:

error_reporting(E_ALL);
$spec = array(
        0 => array("pipe", "r"),
        1 => array("pipe", "w"),
        2 => array("pipe", "w")
        );
$p1 = proc_open("/bin/cat", $spec, $fd1);
$p2 = proc_open("/bin/cat", $spec, $fd2);

fclose($fd1[0]);
fclose($fd1[1]);
fclose($fd1[2]);
echo "closing p1... "; flush();
proc_close($p1);
echo "success\n"; flush();

This code hangs in proc_close(). This doesn't happen if the second proc_open() is commented.

Although the parent process seems to correctly close all pipes, the child process still remains blocked in read(0,...).

I have a similar problem, it seems at I can't concurrently call the same page with proc_open until the first proc_open returns. It looks like proc_open is running through a critical section even though it is opening separate processes.

From experience it is like php is using the same pipes for each proc_open and can't continue to the next proc_open until the original as ended. I would normally use temporary files instead of pipes ? but this makes life difficult.

Dallas

http://www.ekkySoftware.com/

This is actually not a bug, or just documentation bug.

proc_close() blocks if the child has not terminated yet. 

You must use proc_terminate() instead of proc_close() if you can't guarantee the child has already exited, or use proc_get_status() to check if the child has already exited if you want to avoid blocking.

Actually yes, there is a severe pipe setup problem.

The problem is that the second spawned process inherits the descriptor of parent end of pipe to the first spawned process, created when setting up the process1 pipes. Since PHP doesn't set the close-on-exec flag, the descriptor stays open in process2.

So, when the parent-end of pipes1[0] is closed in master script, the descriptor is still open by process2, thus the pipe write end is not closed yet and thus cat in process1 doesn't end.

Note this is also potential security problem, since the second process is able to send data to the first.

Fix:

--- ext/standard/proc_open.c.orig   2006-05-28 19:10:35.000000000 +0200
+++ ext/standard/proc_open.c
@@ -929,6 +929,16 @@ PHP_FUNCTION(proc_open)
                            descriptors[i].mode_flags), mode_string, NULL);
 #else
                stream = php_stream_fopen_from_fd(descriptors[i].parentend, mode_string, NULL);
+
+#if defined(F_SETFD) && defined(FD_CLOEXEC)
+               /*
+                * Mark the descriptor close-on-exec, so that it
+                * it won't be inherited by potential other children.
+                * This avoids first child deadlock on proc_close().
+                */
+               fcntl(descriptors[i].parentend, F_SETFD, FD_CLOEXEC);
+#endif
+
 #endif
                if (stream) {
                    zval *retfp;

This bug has been fixed in CVS.

Snapshots of the sources are packaged every three hours; this change
will be in the next snapshot. You can grab the snapshot at
http://snaps.php.net/.
 
Thank you for the report, and for helping us make PHP better.