New submission from Ross Lagerwall <rosslagerw...@gmail.com>: There is an issue where if a python program closes all the std. file descriptors (e.g. a daemon) and then uses the subprocess module, the file descriptors may not be set up properly in the subprocess. This may actually be a fairly common use case in a daemon program that needs to run a subprocess and set up pipes to it.
Here is an example: import os, subprocess, sys x=os.open('/dev/null', os.O_RDWR) os.close(0) os.close(1) os.close(2) res = subprocess.Popen([sys.executable, "-c", 'import sys;' 'sys.stdout.write("apple");' 'sys.stdout.flush();' 'sys.stderr.write("orange")'], stdin=x, stdout=subprocess.PIPE, stderr=subprocess.PIPE).communicate() with open('/tmp/out', 'w') as f: f.write(repr(res) + '\n') f.write(repr((b'apple', b'orange')) + '\n') The expected output in /tmp/out is: ('apple', 'orange') ('apple', 'orange') but we get: (b'', b"Fatal Python error: Py_Initialize: can't initialize sys standard streams\nOSError: [Errno 9] Bad file descriptor\n") (b'apple', b'orange') The problem comes about where the calls are made (this applies to the python & c versions): os.dup2(p2cread, 0) os.dup2(c2pwrite, 1) os.dup2(errwrite, 2) if c2pwrite or p2cread or errwrite is the same as what it's being dupped() to (eg if c2pwrite == 1) then the dup2 call does nothing. But, if we're using pipes, the close-on-exec flags are set initially and the dup2() call would normally remove the flag but it doesn't. Attached is a patch which basically uses fcntl if necessary to remove the close-on-exec flag, and tests. ---------- components: Library (Lib) files: subprocess.patch keywords: patch messages: 125067 nosy: georg.brandl, giampaolo.rodola, gregory.p.smith, pitrou, rosslagerwall priority: normal severity: normal status: open title: Subprocess error if fds 0,1,2 are closed type: behavior versions: Python 3.2 Added file: http://bugs.python.org/file20223/subprocess.patch _______________________________________ Python tracker <rep...@bugs.python.org> <http://bugs.python.org/issue10806> _______________________________________ _______________________________________________ Python-bugs-list mailing list Unsubscribe: http://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com