DragonflyBSD going to have it hypervisor

[Deleted member 63539]

They are going to port NetBSD's NVMM to DragonflyBSD: http://lists.dragonflybsd.org/pipermail/kernel/2020-August/287891.html

So another BSD join the hypervisor game. If it's done, we now could use both the four major BSDs to host virtual machine.

Excluding OpenBSD, it vmd currently is just a joke:

https://twitter.com/x/status/1291257985734410244

View: https://twitter.com/m00nbsd/status/1291257985734410244?s=20
Who on earth would use a hypervisor that only supports maximum 1 vcpu for each guest? And with that security issues above, no, no one other than OpenBSD fanboys would use it if they don't fix all of these security issues and support more vcpus per each guest!

This news is indeed a good news for us. The more non-Linux OSes on the game, the better for us!

 

[Deleted member 63539]

Currently on FreeBSD we have Bhyve, VirtualBox, Xen. If we could port NVMM and HAXM we would even have more choices!

 

[kpedersen]

To be fair, I think OpenBSD's leader did have a fair point in saying:

"You are deluded if you think that a worldwide collection of software engineers who can't write operating systems or applications without security holes, can then turn around and suddenly write virtualization layers without security holes"

For security (which OpenBSD has many great innovations in), I can see why vmm development is coming along slowly. However once it is a little more mature, I will trust it considerably more than perhaps some operating systems who have a quick 'n dirty port of an existing hypervisor with minimal code audit.

As it stands, vmm is still useful for a number of solutions where guests should only have one vcpu. This requirement is actually fairly common.

Try to keep an open mind with this project. Having an almost zero compromise approach to quality and security is absolutely unique in this day and age and I think good things will come of it.

 

[kpedersen]

What's the use of an operating system if it's only secure if you install no apps and don't make any changes after installation?

To the contrary, what use is an operating system if it isn't secure, even before you install applications or make any changes?

As for vmm and other type 2 hypervisors, they still have a place, but they are being pushed aside more by containerization and use of namespaces.

I kind of agree with this. OS level virtualisation (such as Jails) is likely going to remain the better option for a number of use-cases. But, it has always been around so it can't really be called "the future". The jail command has been in FreeBSD since version 4.0!

Admittedly, it has taken Linux a *long* time to catch up here and implement a container system in its "own little way". And brand names like Docker will be around until the native Linux tools mature a little more.

Type 2 hypervisors do have their uses but I feel the main reason they are popular at the moment is because people who don't understand UNIX can't quite get their heads round the idea of a (glorified) chroot and can only really think in terms of "Ooh, an emulated computer. Thats like a PC in a PC!".

But both solutions have their share of problems. For example, an interesting read if you have time: http://www.watson.org/~robert/2007woot/2007usenixwoot-exploitingconcurrency.pdf.
Discusses a few flaws with how many container systems are implemented. It is one of the reasons why OpenBSD discontinued their own jail system (sysjail).

 

[Deleted member 63539]

For security (which OpenBSD has many great innovations in), I can see why vmm development is coming along slowly. However once it is a little more mature, I will trust it considerably more than perhaps some operating systems who have a quick 'n dirty port of an existing hypervisor with minimal code audit.

Security? Really? Read this:

https://twitter.com/x/status/1291257985734410244

View: https://twitter.com/m00nbsd/status/1291257985734410244?s=20


No, people here still underestimate other BSDs a lot. In their view BSDs are only about FreeBSD and OpenBSD. It's plain wrong. And how could you know it would be a quick 'n dirty port I wonder? Given how I observed Kamil Rytarowski's past works on NetBSD, I think it will definitely not a quick 'n dirty port! You could check his past and current works on NetBSD on their blog here: https://blog.netbsd.org/

 

[Deleted member 63539]

To the contrary, what use is an operating system if it isn't secure, even before you install applications or make any changes?


I kind of agree with this. OS level virtualisation (such as Jails) is likely going to remain the better option for a number of use-cases. But, it has always been around so it can't really be called "the future". The jail command has been in FreeBSD since version 4.0!

Admittedly, it has taken Linux a *long* time to catch up here and implement a container system in its "own little way". And brand names like Docker will be around until the native Linux tools mature a little more.

Type 2 hypervisors do have their uses but I feel the main reason they are popular at the moment is because people who don't understand UNIX can't quite get their heads round the idea of a (glorified) chroot and can only really think in terms of "Ooh, an emulated computer. Thats like a PC in a PC!".

But both solutions have their share of problems. For example, an interesting read if you have time: http://www.watson.org/~robert/2007woot/2007usenixwoot-exploitingconcurrency.pdf.
Discusses a few flaws with how many container systems are implemented. It is one of the reasons why OpenBSD discontinued their own jail system (sysjail).

Given your comments, I think I should not discuss with you anymore. I don't want to deal with fanboyism. And I also not intended to make this thread a flame war. I created the thread only to inform people don't track the DragonflyBSD mailing list to know that they are going to have a hypervisor. Just that. I quit this discussion.

 

[kpedersen]

I created the thread only to inform people don't track the DragonflyBSD mailing list to know that they are going to have a hypervisor. Just that. I quit this discussion.

That's fine, you brought up OpenBSD in your first post so it was fair game to correct you.

Now that you have quit this discussion. I invite you to spend the time to check out some of OpenBSD's novel ideas such as the pledge system (https://www.openbsd.org/papers/BeckPledgeUnveilBSDCan2018.pdf) and have a read up on why the project is still seen as one of the most secure.

 

[shkhln]

novel ideas such as the pledge system

 

[Deleted member 63539]

Did you mean fanboy? what about girls? what flame war? only three replies. I can't see any flame here!

Being warned by the Mod since the gh_origin account, I have to protect myself first. I took 10 days and 10 posts to be a full member!

 

[kpedersen]

You don't agree?

So far it is the only thing I have seen to be in a good enough state to try to contain the madness of Google's Chromium without resorting to a full on Jail (which can make things like audio / mic / webcam / gpu fairly fiddly).

If as much software as possible had this functionality passively, a whole host of vulnerabilities would be much harder to exploit. And it all stacks. Imagine a pledged program running in a Jail, running on Bhyve.

 

[shkhln]

Pledge/unveil is basically an OpenBSD's answer to Capsicum. We can argue whether it has a better API, but novel it is not.

 

[kpedersen]

Pledge/unveil is basically an OpenBSD's answer to Capsicum. We can argue whether it has a better API, but novel it is not.

Oh right. I see. Well yeah, sandboxing isn't new at all.
The novelty is (effectively the mindset) that they are pushing for it to be included everywhere where other platforms tend not to bother.

We have some attempts: https://wiki.freebsd.org/Chromium/Capsicum

But their security focus makes things like this a priority.

Same with this: https://www.openbsd.org/papers/eurobsdcon2018-rop.pdf
Yes, there isn't anything truely novel about this (the mitigations in clang?)... but the fact that they are actually doing it is novel in itself.