[Linux] Properly raise ZombieProcess in exe() method

[Jongy]

Summary

• OS: Linux
• Bug fix: yes
• Type: core
• Fixes: None

Description

psutil.Process(pid).exe() returns "" for zombie processes, incorrectly. It should raise ZombieProcess, and return "" only for kernel threads.

This PR fixes the logic (I wrote something quick, I suppose that the logic can be reordered more nicely maybe).

Now we get:

• Alive processes - their exe
• Non-existing PIDs - NoSuchProcess
• Kernel threads - ""
• Zombie processes - ZombieProcess.

Tested on kernel 5.4.

Also fixed & improved the comment along the way.

[Jongy]

I will look into the failing test later today

[Jongy]

I looked into the test. The logic here is incorrect:

                # ...but if /proc/pid no longer exist we're supposed to treat
                # it as an alias for zombie process
                with mock.patch('psutil._pslinux.os.path.lexists',
                                return_value=False):
                    self.assertRaises(
                        psutil.ZombieProcess, psutil.Process().exe)

if /proc/pid no longer exists, it's not a zombie. Zombies have their /proc/pid. It's a zombie if /exe link is missing & .status() is "zombie".

How should I fix the test? The lexists patch is relevant for NoSuchProcess, so I can change that, & add another mock for status() which makes it look like a zombie?

[giampaolo]

psutil.Process(pid).exe() returns "" for zombie processes, incorrectly. It should raise ZombieProcess, and return "" only for kernel threads.

Mmm why are you saying so? Can you provide a test case?
We do have a unit test for zombie processes (and it passes on Linux):

psutil/psutil/tests/test_process.py

Line 1297 in 1da9f79

def test_zombie_process(self):

Also, if there really is a problem with zombie detection, then we probably also have a problem with other Process methods, not only exe().

With that said, by looking at the Linux code I also suspect we're not detecting/handling zombie properly, because BSD and OSX implementations do something different, similar to what you're doing in this PR:

• OSX:

  psutil/psutil/_psosx.py

  Lines 331 to 336 in 1da9f79

  	def is_zombie(pid):
  	try:
  	st = cext.proc_kinfo_oneshot(pid)[kinfo_proc_map['status']]
  	return st == cext.SZOMB
  	except Exception:
  	return False

• BSD:

  psutil/psutil/_psbsd.py

  Lines 545 to 550 in 1da9f79

  	def is_zombie(pid):
  	try:
  	st = cext.proc_oneshot_info(pid)[kinfo_proc_map['status']]
  	return st == cext.SZOMB
  	except Exception:
  	return False

[giampaolo]

OK, I see what you mean now. exe() returns an empty string and test_zombie_process is too relaxed and does not test this condition (assumes "" is a legit return value).

[Jongy]

OK, I see what you mean now. exe() returns an empty string and test_zombie_process is too relaxed and does not test this condition (assumes "" is a legit return value).

Precisely :)

I'm available to work on the PR if you want any changes to get this merged.

[Jongy]

We're using a patched version to work around this bug:

def process_exe(process: psutil.Process) -> str:
    """
    psutil.Process(pid).exe() returns "" for zombie processes, incorrectly. It should raise ZombieProcess, and return ""
    only for kernel threads.

    See https://github.com/giampaolo/psutil/pull/2062
    """
    exe = process.exe()
    if exe == "":
        if is_process_zombie(process):
            raise psutil.ZombieProcess(process.pid)
        raise MissingExePath(process)
    return exe

I'd love to remove it and get back to mainline haha.

[giampaolo]

I'm afraid this may affect other methods other than exe() and requires a more generic solution. I still didn't properly understood the whole picture though. Zombie processes are hard to deal with, especially on Linux apparently.

[giampaolo]

As I suspected, it turns out the problem not only affects process exe(). We have it also in cmdline() and memory_maps(). I created #2288 which fixes the problem in a more generic way, and also expands tests for zombie processes.

[Jongy]

As I suspected, it turns out the problem not only affects process exe(). We have it also in cmdline() and memory_maps(). I created #2288 which fixes the problem in a more generic way, and also expands tests for zombie processes.

That's great! Thanks.