RBAC Integration
Are there any plans to provide integration between iptables and the various rule based access control architectures available on the Linux platform such as RSBAC, grsecurity or selinux?
Thanks,
Joe Saland, CISSP
gideon#gideonolam.com
Quoted 3 lines · gideon olam
Are there any plans to provide integration between iptables and the various rule based access control architectures available on the Linux platform such as RSBAC, grsecurity or selinux?
I cannot remember any discussion on that on this list. Care to elaborate what you imagine?
best regards
Patrick (who knows those projects exist...)
Quoted 11 lines · Patrick Schaaf
Quoted 3 lines · gideon olam
Are there any plans to provide integration between iptables and the various rule based access control architectures available on the Linux platform such as RSBAC, grsecurity or selinux?
I cannot remember any discussion on that on this list. Care to elaborate what you imagine?
best regards
Patrick (who knows those projects exist...)
Well, I was thinking about "role" matches for filters and other policy type matches, be it MAC or whatever. Userspace could use roles to
determine who has what access rights to a table. Matches like owner
might be able to benefit. I would think there would be a large variety of implications for integration.
Joe Saland, CISSP
gideon#gideonolam.com
Quoted 5 lines · gideon olam
On Fri, 2003-06-27 at 11:40, Patrick Schaaf wrote:
Userspace could use roles to determine who has what access rights to a table.
Matches like owner might be able to benefit. I would think there
would be a large variety
of implications for integration.
I think you don't go far enough. A framework doing external autorisation thanks to the QUEUE target could be very useful.
In fact, I'm coding an auth request system (based on UDP) which permits to any protocol compliant auth server to take decision over packet.
BR,
Signature · 1 line
Eric Leblond <[email protected]>
Quoted 7 lines · Eric Leblond
Quoted 5 lines · gideon olam
On Fri, 2003-06-27 at 11:40, Patrick Schaaf wrote:
Userspace could use roles to determine who has what access rights to a table.
Matches like owner might be able to benefit. I would think there
would be a large variety
of implications for integration.
Quoted 6 lines
I think you don't go far enough. A framework doing external autorisation thanks to the QUEUE target could be very useful.
In fact, I'm coding an auth request system (based on UDP) which permits to any protocol compliant auth server to take decision over packet.BR,
I agree, I was just saying that I felt there were many benefits to integrating netfilter with the various security models. The QUEUE is great for some apps, but personally though I think a netfilter module that provided additional matches would provide instant integration. It would become easy to tie specific filters down to authenticated Roles on
the system. Some of this might even require patches for other
netfilter code to integrate role management.
Netfilter integrated into an RBAC would provide a truly least privilege network layer for secure servers. It would also move administration of iptables out of root's domain in a secure manner and into secure roles allowed to modify netfilter objects.
Hi Joe,
Quoted 2 lines
Well, I was thinking about "role" matches for filters and other policy type matches, be it MAC or whatever.
In the existing RBAC systems, is there already a well defined notion of "role" for individual network packets (aka sk_buffs), as they pass through the stack?
Quoted 1 line
Matches like owner might be able to benefit.
That one is obvious. It should be straightforward to do, in that case, right? Processes have role tags, the owner match is the sole point which knows process context, so process roles could be checked by some clone of the owner match.
Quoted 2 lines
Userspace could use roles to determine who has what access rights to a table.
You mean something like this: who is permitted to modify this or that chain, e.g. a userspace role "permitted to set MARKs from 1 to 100 in mangle/PREROUTING"?
That could be very useful. Also very complex to get all interesting criteria (like the MARK range I mention) covered.
A last comment: an owner like role match, and probably other similar things, would better be part of the respective RBAC system. At least until the time when that RBAC system is part of the base kernel. I don't think that the iptables core implementation could / should contain parts useful solely by out-of-kernel patches / modules. Maybe patch-o-matic/extra could cope with this.
best regards
Patrick
Quoted 9 lines · Patrick Schaaf
Hi Joe,
Quoted 2 lines
Well, I was thinking about "role" matches for filters and other policy type matches, be it MAC or whatever.
In the existing RBAC systems, is there already a well defined notion of "role" for individual network packets (aka
sk_buffs), as they pass through the stack?
The state of network control in the different systems are at varying
levels. Grsecurity is adding network controls in 2.0.
Quoted 7 lines
Quoted 1 line
Matches like owner might be able to benefit.
That one is obvious. It should be straightforward to do, in that case, right? Processes have role tags, the owner match is the sole point which knows process context, so process roles could be checked by some clone of the owner match.
This would follow from the above, if there is already a context for the objects in the RBAC, then it follows there would be a role associated with the object.
Quoted 10 lines
Quoted 2 lines
Userspace could use roles to determine who has what access rights to a table.
You mean something like this: who is permitted to modify this or that chain, e.g. a userspace role "permitted to set MARKs from 1 to 100 in mangle/PREROUTING"?
That could be very useful. Also very complex to get all interesting criteria (like the MARK range I mention) covered.
Yes, but even more simply, on a secure system root shouldn't have to manage filters. The RBAC MAC could control access to the iptables objects at a more granular level then just read/write.
Perhaps even new filters could be checked against some policy to
determine whether or not to apply it. This could be very useful in a
dynamic firewall application.
Quoted 7 lines
A last comment: an owner like role match, and probably other similar things, would better be part of the respective RBAC system. At least until the time when that RBAC system is part of the base kernel. I don't think that the iptables core implementation could / should contain parts useful solely by out-of-kernel patches / modules. Maybe patch-o-matic/extra could cope with this.
Well, most of these modules are kernel patches and the like. I suspect any project would become part of the RBAC's patches to the kernel, an add-on to the RBAC to provide iptables integration. That being the case, the RBAC and how it integrates, what capabilities are feasible, really depend upon the specific RBAC chosen. It would be nice if any interface could satisfy them all, but I do not think that's realistic considering the nature of these projects.
http://www.rsbac.org
http://www.grsecurity.org
http://www.nasa.gov/selinux
Selinux has some possible patent issues (search google on selinux patent) It is available as part of the Hardened Gentoo project (see below)
RSBAC is used in Adamantix (previously Trusted Debian)
http://www.adamantix.org
GRSecurity is getting more mature and is available as part of the Hardened Gentoo Project
http://www.gentoo.org/proj/en/hardened/
Most of these can go right on top of any distro, though some of these features can break apps on an OS. For instance, some PaX options in grsecurity can wreak havoc on XFree, but it's other adds are sweet, things like IP ID randomization, TCP Seq randomization, source port randomization, pid randomization, etc..
Quoted 2 lines
best regards
Patrick
--jrs