On Tue, 2008-04-08 at 17:13 -0400, Paul Moore wrote:
> On Tuesday 08 April 2008 10:43:44 am Stephen Smalley wrote:
> > On Mon, 2008-04-07 at 19:11 -0400, Paul Moore wrote:
> > > @@ -709,73 +697,46 @@ static int security_context_to_sid_core(char
> > > *scontext, u32 scontext_len, u32 *s null suffix to the copy to
> > > avoid problems with the existing attr package, which doesn't view
> > > the null terminator as part of the attribute value. */
> > > - scontext2 = kmalloc(scontext_len+1,GFP_KERNEL);
> > > - if (!scontext2) {
> > > - rc = -ENOMEM;
> > > - goto out;
> > > - }
> > > - memcpy(scontext2, scontext, scontext_len);
> > > - scontext2[scontext_len] = 0;
> > > + scontext_dup = kmemdup(scontext, scontext_len + 1, GFP_KERNEL);
> >
> > Also, in addition to the gfp_flags change, I'm not clear that the
> > above change is correct. We are taking a byte array "scontext" of
> > length "scontext_len" and copying it into a buffer of length
> > "scontext_len+1" so that we can ensure that it is NUL terminated
> > prior to parsing. Won't kmemdup with scontext_len+1 ultimately run
> > off the end of the original string?
>
> Good point, I believe you're right. I'll add this and the gfp stuff to
> the list of needed changes.
>
> I think I may also suggest shelving this patch for 2.6.26 as a little
> birdie mentioned it would be a good idea to give this a through testing
> on non-MLS/MCS systems which I haven't yet done and don't expect to be
> able to do so before the merge window opens.
Just for future reference for others, since I didn't cc the list on that
reply: When making a change that substantively affects the secid/secctx
conversion functions, it would be good idea (albeit painful) to test
each of the following cases:
1) kernel with MLS/MCS policy, filesystem without MLS/MCS labels (to
exercise the compatibility code to map to a default MLS/MCS label).
This is encountered e.g. when mounting a filesystem created on a
non-MCS/MLS distribution on a MCS/MLS distribution, as with an upgrade
from RHEL4 to RHEL5 or Fedora <= 4 to Fedora >= 5.
2) kernel without MLS/MCS policy, filesystem with MLS/MCS label (to
exercise the compatibility code to skip the MCS/MLS label). This is
encountered in the reverse direction, particularly if sharing a
filesystem across multiple distributions on a multi-boot system.
3) kernel with MLS/MCS policy, filesystem with MLS/MCS labels. This is
a clean install of RHEL5 or Fedora >= 5.
4) kernel without MLS/MCS policy, filesystem without MLS/MCS label.
This is a RHEL4 or Fedora <= 4 or Ubuntu Hardy Heron or Hardened Gentoo
install.
--
Stephen Smalley
National Security Agency
--
This message was distributed to subscribers of the selinux mailing list.
If you no longer wish to subscribe, send mail to majordomo@xxxxxxxxxxxxx with
the words "unsubscribe selinux" without quotes as the message.
