Re: Secure Linux From Scratch
Robert Day <[email protected]>
| Newsgroups | gmane.linux.lfs.security |
|---|---|
| Message-ID | <[email protected]> |
On Fri, 2003-12-19 at 07:53, Ken Moffat wrote: > On Thu, 18 Dec 2003, ashes wrote: > > > > > >From what I understand et_dyn is broken in gcc-3.3. So without adding more > > patches, we could downgrade to gcc-3.2, or upgrade to gcc-3.4 with a > > backport/patch.. and that needs glibc-2.3.3. Just something to consider, but > > we'll need to do one of these to take advantage of Pax. > > > > I think that using a different version of gcc, let alone a different > version of glibc, should be a non-starter for an LFS project. I can > cope with patching the version which is in LFS, obviously, but if SLFS > becomes too far removed from the things in LFS then the amount of > testing required will increase exponentially (and I don't mean the > security testing people here are focussed on, I mean the "can it build > everything correctly" testing). Simple answer, if Pax requires that much, then we don;t use it. Not till the required subsystems are available as part of LFS. We're not replacing LFS here with a cutting edge system. One of the key points in security is using tested, stable packages. You can't build a secure box using insecure packages as a foundation. If the gcc 3.4, being so new, has a flaw, that flaw propagates to every package in the system. Imagine the errata we'd have to release if a bug was found in gcc that allowed a buffer overflow to occur in any package compiled with that compiler that used XXXX system call? "A bug has been found in SLFS 5.1 (aka Valentine) released Feb 14, 2004. This bug allows a remote attacker to overflow the stack of any program compiled with gcc 3.4 that uses XXXX function call, and execute arbitrary code, leading to a possible root comprimise. The fix for this bug is a complete rebuild of your SLFS system from the ground up. Please refer to SLFS 5.1.1 (aka Lion) release March 1, 2004." Not such a good idea ;) Now whether or not such a thing could happen, I dunno - and frankly it is a moot point.. the essence of my argument still stands. No application and no Os can be secure, if the underlying technology is insecure. (ie. your box is only as secure as the local users allow. Any local user can reboot the box, clear the CMOS if it is passworded, boot floppy or CD, and comprimise root on the box. Unless the box is in a safe, with alarms around it, in a building with armed guards, and a direct link to the police should anyone enter the building, the system can be comprimised.) My suggestion is only use released packages, that have been tested, and secured. Rob Day (BOFH) -- http://linuxfromscratch.org/mailman/listinfo/lfs-security FAQ: http://www.linuxfromscratch.org/faq/ Unsubscribe: See the above information page