Re: BUGTRAQ ALERT: Solaris 2.x vulnerability

[avalon () coombs anu edu au (Darren Reed)]

In some mail from Nathan Lawson, sie said:
> > On Wed, 16 Aug 1995, Nathan Lawson wrote:
> >
> > > The /tmp/.psXXXXXX is open to a race.  The directory is safe as long as it
> > > isn't world writable.
> >
> > Mind to expand on this? Who would tiss work. The /tmp directory *does*
> > have the sticky bit so we cant rm the file and substitute it. Are you talking
>
> Yes.  Basically, Irix uses an insecure method of making a temp file based on
> pid, just like any other OS that uses blahXXXXXX.  If you are going to use a
> predictable filename, then you'd better open it securely.
>
> Hobbit posted a very good method a while back during some binmail fiasco.
> Basically, with a temp file, you are creating a file.  The way he described was
> to open the file with O_EXCL so it doesn't follow symlinks.  Then, lstat the
> file and make sure it is not a link.  Lastly, stat the file and make sure it
> has the same inode as found in the lstat call.

I think this is what Sun did...and fell over.

Even doing all of those things would not have prevented the Sun bug in ps.

What all those checks do is fail to realise that the file they have open is
not necessarily that is there on disk.  What they needed to do was use
fchown().  Had they used fchown instead of chown, there would be no bug,
regardless of the permissions on /tmp.  Similarly, you should be using
fstat above, not just stat/lstat.  Once you've opened a file, any further
check on it which doesn't use the fd is raceable, if it can be changed, such
as permissions on /tmp usually allow.

darren