Re: [GIT PULL] execve updates for v6.13-rc1 (take 2)

[Linus Torvalds]

On Wed, 27 Nov 2024 at 18:06, Al Viro <viro@xxxxxxxxxxxxxxxxxx> wrote:
>
> > So if pathname exists and isn't empty, AT_EMPTY_PATH does nothing.
>
> ... so let's tie that to pathname _being_ empty - it's not as if it
> had been hard to check.

This is not some kind of new system call, and AT_EMPTY_PATH isn't some
Linux-only thing.

It has well-defined and documented semantics:

  AT_EMPTY_PATH
     If this flag is specified, oldname can be an empty string.

Note the "can be". Not "will/must be".

> What's more, let's allow userland pointer to be NULL - use getname_maybe_null()
> and treat NULL returned by it as "we have an empty pathname".

Now, that's separate, and I agree with that extension.  That just
suppresses another "empty string" error case.

But no, I do not accept changing well-documented behaviour of
AT_EMPTY_PATH, much less the insanity of making "execveat()" have
completely different semantics for AT_EMPTY_PATH than a plain openat.

             Linus