Re: uudecode bug (?)

[Paul Eggert]

> From: Santiago Vila <address@hidden>
> Date: Mon, 10 Jun 2002 13:41:17 +0200 (CEST)
> 
> Moreover, uudecode(1) says:
> 
> STANDARDS
>        This implementation is compliant with P1003.2b/D11.
> 
> Does someone knows whether the proposed change is compatible with
> this standard?

The proposed change does not conform to POSIX 1003.1-2001, which
supersedes P1003.2b/D11.

To conform to POSIX, you'd have to enable the proposed behavior only
if a new option were set, or if POSIXLY_CORRECT were not set, or
something like that.

Also, the proposed change in
<ftp://updates.redhat.com/7.0/en/os/SRPMS/sharutils-4.2.1-8.7.x.src.rpm>
does not fix the security problem entirely; a determined attacker can
still get through, even if that patch is installed.


> From: Andrew D Jewell <address@hidden>
> Date: Mon, 10 Jun 2002 10:44:37 -0400
> 
> Perhaps there is a common option that many gnu utils should add, 
> which is a little bit like the '-f' option in cp :
> 
> -P, --paranoid : always open output files with
>                   "unlink(foo); open(foo, O_EXCL | O_CREATE)"

That is not sufficiently paranoid, since it is vulnerable to
denial-of-service attacks.  Also, unlink(foo) can unlink directories
on some hosts.


It wouldn't hurt to add some documentation about the problem, though.