Privilege drop, separation, and restricted-service operating mode in OpenBSD

https://sha256.net/privsep.html

by brynet • 3 years ago

74 44 3 years ago

sha256.net

g0xA52A2A3• 3 years ago

Changed "privilege separation" -> "separation" to fit within character limits.

brynet• 3 years ago

Indeed. :-)

g0xA52A2A3• 3 years ago

LOL I thought it was my submission, totally missed the redirect.

Zurrrrr13• 3 years ago

Security is more than just auditing code to remove bugs - it's also being prepared for worst case scenarios for bugs that inevitably exist but have not yet been found.

For a security focused OS, OpenBSD is terrible in this regard. Pledge and Unveil are simply not sufficient.

Something like SELinux, RSBAC, Even AppArmor is sorely needed.

yazzku• 3 years ago

Aren't pledge/unveil precisely preparing for unknown bugs? Also they seem to be much easier to use than doing AppArmor chops.

_0w8t• 3 years ago

Linux approach delegates security to the system administrator. Hence Linux provides containers, SELinux etc. focusing on securing the existing binaries.

OpenBSD assumes it is the programmer who is responsible for delivering secure code. Hence it provides the tools that require to modify the sources.

One can argue for one or another approach, but both at the end deliver similar reduction of the attack surface.

Zurrrrr13• 3 years ago

I think that is a flawed way of looking at it. Stuff like SELinux assumes the system administrator or software can't be trusted, and provides ways to protect against future attacks while enforcing only the minimal access needed.

It isn't letting programmers off the hook, and isn't an alternate approach.

Unless you can be absolutely certain of the code you are running, which no one really can, then it's the better approach.

traceroute66• 3 years ago

Something like SELinux, RSBAC, Even AppArmor

The problem with AppArmor and friends is that they are hell on earth to configure and manage.

And I'm saying that as a security conscious sysadmin !

I don't mind spending little time to lock-down my systems a bit, but writing new AppArmor rules is about as attractive as bashing my head against the proverbial brick wall.

Of course I leave AppArmor enabled to some degree, and I'm happy if a package comes with AppArmor rules ... but I ain't gonna be sacrificing days of my life writing, tweaking, troubleshooting my own rules.

That's why I feel OpenBSD's answer is likely a good compromise. It adds a demonstrable level of security, and it doesn't stress out the sysadmin.

binkHN• 3 years ago

The problem with AppArmor and friends is that they are hell on earth to configure and manage.
I don't mind spending little time to lock-down my systems a bit, but writing new AppArmor rules is about as attractive as bashing my head against the proverbial brick wall.

Yep. This is one of the prime motivators as to why OpenBSD did what they did. Often times, when AppArmor and similar solutions prevent a program from working properly, the first reaction is to disable it, and then these protections tend to stay disabled.

Zurrrrr13• 3 years ago

Rules might be a pain, but no more so than writing good firewall rules, and once done don't have to be done again.

It's not the issue some make it out to be

daneel_w3• 3 years ago

"For a security focused OS, OpenBSD is terrible in this regard. Pledge and Unveil are simply not sufficient."

Does that mean that OpenBSD's mitigations are trivial to exploit? "Terrible" and "simply not sufficient" are strong words. If so, can you provide a working proof-of-concept? I keep asking this of people who make this statement because I'm genuinely interested in seeing it demonstrated as it has negative implications for me, but no one wants to spend time proving their point.

beachhead7• 3 years ago

It'll never happen. Every single time this comes up it turns into the same thing over and over again. You must remember the person that said they'd "write a blogpost bypassing OpenBSD mitigations next week" and that's been well over a month now and, surprise, there's no blog about this.

Everything OpenBSD does is wrong and trivial to bypass but everyone's too busy to do it. Maybe the dumbest part about this is that nobody on the other side of this is making claims that these mitigations are perfect in any way.

Qualys has bypassed some OpenBSD malloc hardening features recently but then they don't go around making wild or insulting claims about how wrong and trivial they are either. Go figure.

Zurrrrr13• 3 years ago

No.

It means if you look at any number of the security vulnerabilities that affected OpenBSD base system that they disclosed, or any number of bugs in third party software they haven't audieted, out of the vulns that allow root and/or rce, a lot of damage can be done on a 'secure' OpenBSD system, which something like SELinux can prevent.

Lets say Apache or something had a remote root bug (despite not running as root) that granted a shell was was pretty easy to exploit. An SELinux system can protect an unkown but actively exploited bug like that more than OpenBSD can.

The OpenBSD approach is just to keep looking and hope you've found every bug that would possibly allow for anything like that, and hope they've found them all.

mrweasel• 3 years ago

Pledge and Unveil are better in one, maybe two respects.

1) The restrictions are put in place by the developer, who understand the correct and expected behaviour better than a random systems administrator.

2) Neither Pledge nor Unveil can be disabled by the user. Both SELinux and AppArmor (I don't know RSBAC) can be disabled, and is FREQUENTLY disabled because dealing with either systems are a major hassle to configure.

I'm not quite sure how SELinux is better, it's perhaps more granular, but is that better?

sillywalk• 3 years ago

I believe that for the OpenBSD developers, SELinux/apparmor/seccmp are against their philosophy. [0]

While they are very powerful tools, they require complex rules maintained outside the actual code, and their complexity may lead you to disable the Security stuff entirely, rather than try to re-write the rules your distro/package came with.

OpenBSD had systrace back in the day to restrict system calls, but it suffered from the Time of Check/Time of Use/ problem, and was only used on one (sshd) program in the base system.

[0] https://www.openbsd.org/papers/hackfest2015-pledge/

rustifico• 3 years ago

Yet they are still writing everything in C.

A well-polished turd is still a turd.

chungy• 3 years ago

Now name the OS that isn't written in C.

Hint: It's not Windows. It's not Linux. It's not Mac OS. It's not FreeBSD. It's not illumos...

hulitu• 3 years ago

MULTICS ?

Apollo Domain OS ?

josephcsible• 3 years ago

Redox?

unused0• 3 years ago

IIRC,Primos was written in FORTRAN (with language extensions including the ability to pass a statement number as a parameter, allowing longjmp() like behavior).

Multics (as noted above) was written in PL/I.

mvdwoord• 3 years ago

Sour much?

crest• 3 years ago

Where is your shiny rust kernel and userland useable as desktop/laptop or server? Don't pull others down just because you have yet to deliver.

commandersaki2• 3 years ago

We first get a user with getpwnam(3) to drop to. The user has /var/empty configured as its home directory so we can use that in chroot(2). Next we chdir(2) to the new file-system root to have a valid current working directory. This prevents us accidentally marking a file-system as busy depending on from where the daemon was started, preventing unmounting file-systems while the daemon is running.

I think another reason for chdir(2) is to avoid a vulnerability where you can repeatedly chain chdir("..") in the process and then chroot(".") to escape chroot.

Privilege drop, separation, and restricted-service operating mode in OpenBSD

Related Stories