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

[Kees Cook]

On Wed, Jan 26, 2022 at 06:02:25PM +0300, Alexey Dobriyan wrote:
> >	execve("...", NULL, NULL);
> 
> I personally wrote a program which relies on execve(NULL) to succeed.
> It wasn't an exploit, it was test program against IMA-like kernel
> "only whitelisted executables can run" feature.
> 
> Test copies and "corrupts" itself by appending \0 to the end, then tries
> to reexec itself with execve("/proc/self/exe", NULL, NULL);
> main() if run with argc==0 exits with specific error code.
> 
> Appending \0 breaks checksum so working kernel protection scheme must
> not allow it, therefore if execve(NULL) succeeded, than the parent
> process doing test hard fails.
> 
> Also appending \0 doesn't break ELF structure. In other words,
> if executable A is working (and it is working because it is running)
> then A||\0 is valid executable as well and will run too.
> 
> This is independent from filesystem layout, libc, kernel, dynamic
> libraries, compile options and what not.
> 
> Now QNX doesn't allow execve(NULL) and I don't remember if I changed it
> to the next simplest variant and I don't work anymore at that company,
> so I can't check :^)
> 
> 	execve("/proc/self/exe", (char*[]){"Alexey", NULL}, NULL);

One of the various suggestions was to inject { path, NULL } when argc=0.

Given that execve(path, NULL, ...) is being used at least a little,
hopefully there is nothing that depends on argc==0... :P

-- 
Kees Cook