On Sat, May 30, 2020 at 05:17:24AM +0200, Jann Horn wrote:

On Sat, May 30, 2020 at 4:43 AM Kees Cook [off-list ref] wrote:

quoted

I mean, yes, that's certainly better, but it just seems a shame that
everyone has to do the get_unused/put_unused dance just because of how
SCM_RIGHTS does this weird put_user() in the middle.

Can anyone clarify the expected failure mode from SCM_RIGHTS? Can we
move the put_user() after instead?

Honestly, I think trying to remove file descriptors and such after
-EFAULT is a waste of time. If userspace runs into -EFAULT, userspace
is beyond saving and can't really do much other than exit immediately.
There are a bunch of places that will change state and then throw
-EFAULT at the end if userspace supplied an invalid address, because
trying to hold locks across userspace accesses just in case userspace
supplied a bogus address is kinda silly (and often borderline
impossible).

Logically, I agree. I'm more worried about the behavioral change -- if
we don't remove the fd on failure, the fd is installed with no
indication to the process that it exists (it won't know the close it --
if it keeps running -- and it may survive across exec). Before, it never
entered the file table.

You can actually see that even scm_detach_fds() currently just
silently swallows errors if writing some header fields fails at the
end.

Yeah, and it's a corner case. But it should be possible (trivial, even)
to clean up on failure to retain the original results.

-- 
Kees Cook