Re: [PATCH v2] fs/exec: require argv[0] presence in do_execveat_common()

[Ariadne Conill <ariadne@dereferenced.org>]

Hi,

On Wed, 26 Jan 2022, Kees Cook wrote:

> On Wed, Jan 26, 2022 at 03:30:13PM -0600, Ariadne Conill wrote:
>> Hi,
>>
>> On Wed, 26 Jan 2022, Kees Cook wrote:
>>
>>> On Wed, Jan 26, 2022 at 03:13:10PM -0600, Ariadne Conill wrote:
>>>> Looks good to me, but I wonder if we shouldn't set an argv of
>>>> {bprm->filename, NULL} instead of {"", NULL}.  Discussion in IRC led to the
>>>> realization that multicall programs will try to use argv[0] and might crash
>>>> in this scenario.  If we're going to fake an argv, I guess we should try to
>>>> do it right.
>>>
>>> They're crashing currently, though, yes? I think the goal is to move
>>> toward making execve(..., NULL, NULL) just not work at all. Using the
>>> {"", NULL} injection just gets us closer to protecting a bad userspace
>>> program. I think things _should_ crash if they try to start depending
>>> on this work-around.
>>
>> Is there a reason to spawn a program, just to have it crash, rather than
>> just denying it to begin with, though?
>
> I think the correct behavior here is to unconditionally reject a NULL
> argv -- and I wish this had gotten fixed in 2008. :P Given the code we've
> found that depends on NULL argv, I think we likely can't make the change
> outright, so we're down this weird rabbit hole of trying to reject what we
> can and create work-around behaviors for the cases that currently exist.
> I think new users of the new work-around shouldn't be considered. We'd
> prefer they get a rejection, etc.
>
>> I mean, it all seems fine enough, and perhaps I'm just a bit picky on the
>> colors and flavors of my bikesheds, so if you want to go with this patch,
>> I'll be glad to carry it in the Alpine security update I am doing to make
>> sure the *other* GLib-using SUID programs people find don't get exploited in
>> the same way.
>
> They "don't break userspace" guideline is really "don't break userspace
> if someone notices". :P Since this is a mitigation (not strictly a
> security flaw fix), changes to userspace behavior tend to be very
> conservatively viewed by Linus. ;)
>
> My preference is the earlier very simple version to fix this:
>
> diff --git a/fs/exec.c b/fs/exec.c
> index 79f2c9483302..aabadcf4a525 100644
> --- a/fs/exec.c
> +++ b/fs/exec.c
> @@ -1897,6 +1897,8 @@ static int do_execveat_common(int fd, struct filename *filename,
> 	}
>
> 	retval = count(argv, MAX_ARG_STRINGS);
> +	if (reval == 0)
> +		retval = -EINVAL;
> 	if (retval < 0)
> 		goto out_free;
> 	bprm->argc = retval;
>
> So, I guess we should start there and send a patch to valgrind?

Yes, seems reasonable, though without the typo :)

Since you've already written the patch, do you want to proceed with it?
If so, I can work on the Valgrind tests.

Ariadne