Preferences

agentultra parent
From the same article:

> Theo de Raadt's problem is that he views security the way cryptography experts view cyphers: as an absolute. But security isn't like math. It's not absolute. There are right and wrong ways of doing security.

Not an absolute but there is a right and wrong way to "do security?"

To be fair Boender is attacking the naive maxim that "virtualization is secure." It's just another layer if you isolate processes inside of virtualized run-time environments. Makes sense... it's not snake oil.

However there is no need for "right" and "wrong" in these discussions. The security of any given system exists in a continuum and it will only be a matter of time before the next vulnerability is discovered. I get the sense that all we can do is limit the damage that can be done by any particular system.

It seems that virtualization is just one path towards providing those limits just as chroot and other attempts have been.

I'm most interested in seeing how jitsu and unikernels can turn the tables... not only can a process be wrapped in a virtualization layer but it is short lived and only runs when it is requested. It puts the onus on us to set up the summoner properly and provide safe-guards... but it's just another layer of complexity for attackers to manage.


> I'm most interested in seeing how jitsu and unikernels can turn the tables

The useful aspect about the MirageOS unikernels is that they use the pure Xen PV interface, which has almost no dependency on qemu. No floppy or block/net emulation, timers through the direct Xen shared_info page, and generally as "native" to x86 as Xen permits.

cthalupa
>timers through the direct Xen shared_info page

HVM has support PV timers (and interrupt controllers, and spinlocks) for quite some time now

>generally as "native" to x86 as Xen permits.

I really don't know if I agree with this. With hardware extensions basic CPU performance is going to be significantly better on HVM instances (No longer having to bounce to the hypervisor every time you make a system call since you once again have three CPU protection rings with ring -1), and that's before getting into SR-IOV, etc.

Unikernels are sweet and all, but without PVH, PV will outperform modern "PVHVM" implementations, and with PVH you're still running in a partial HVM shell.

cthalupa
Can't edit this anymore apparently, but correction: PV will not outperform modern "PVHVM"

This item has no comments currently.