The code is less, but there are fewer developers too. For example, basic support for disk encryption (an important security feature) has been added to the installer in 2023 in OpenBSD.
Perhaps people familiar with the software development process in both could chime in.
While might be more pre-oackaged containers available for linux, our OCI stack has nothing on BSD's Jails. ZFS was ported to linux from Solaris, and still runs better on the BSD's (at least in regards to license controversies, if not feature support or performance, which have mostly converged in OpenZFS). The linux-native option would be BTRFS (love them both :p).
One is the mandatory access control of SELinux, used extensively in Android (probably the most widely-used Linux distribution of them all). This has no OpenBSD equivalent, as far as I know.
Another is that the Linux kernel has some compiled-in exploit mitigations enabled that OpenBSD doesn't, such as automatic bounds checks on fixed-size array access (based on UBSan).
Besides this, it changes the perspective of who is the best person to configure the secure features. OpenBSD makes this a decision for the developer of the application, who should know better about it than a random sysadmin who most probably won't even open the source code to know better (this is the approach SELinux takes).
https://man.openbsd.org/pledge https://man.openbsd.org/unveil
For example FreeBSD has a MAC framework (a massive one btw) and also the "SELinux/SEBSD" framework on top of it (FLASK/TE), but you don't need to use it (not on Linux nor on FBSD).
OpenBSD has no MAC implementation, and with that no framework (SE*) on top of it, but has different/other way's to secure a system.
And TBH i have seen just 3 Customers until now who really develop highly secure/complicated policies (Two use MLS and one uses Brewer-Nash)
I think MAC should be used much more, but it's time intensive and hard to do it right, also to keep the policies clean and understandable need's LOTS of documentation and dedication.
https://www.diva-portal.org/smash/get/diva2:5365/FULLTEXT01.... (2006)
The 'different/other/ ways to secure the system are inferior since they offer no protection if root is compromised.
I don't think MAC is as hard to use as it was, there are so many policies and issues known this much later, but people still just disable it by default because they don't want to put in the time.
Capsicum?
That's NOT what i said, the FreeBSD MAC implementation is big and pretty much feature complete, NOT SEBSD.
>The 'different/other/ ways to secure the system are inferior since they offer no protection if root is compromised.
There is no such thing as "inferior" but different approaches, from completely deleting root as a user to using Container/Jail/Zones, Sandbox's, VM's etc. MAC is one of just many methods and OpenBSD voted against it and went another route (and that is totally fine and understandable).
>I don't think MAC is as hard to use as it was
MAC is still very hard, you are talking about SELinux that is just one implementation called FLASK/TE.
Try to implement Brewer-Nash MAC-policy on a Fileserver and i will see you sweating ;)
But as you can see, there is you and me (in this thread) who understand what a MAC even is, and that on HN....that just tells you how many people really have even a understanding what it even is.
It is what you said. I never said you claimed SEBSD.
You said FreeBSD has a massive MAC framework. I was asking which one, and the only one I know of is SEBSD, which is not at all massive.
You are saying now FreeBSD has its own MAC framework, but I've never heard of it. What is it called?
> There is no such thing as "inferior" but different approaches,
Well that's not true. A screen door vs a heavy deadbolted door is clearly an inferior approach, not just a different approach to security, and that analogy extends to OS security technologies.
MAC is the only system that can 100% protect against an attacker getting remote root.
> There is no such thing as "inferior" but different approaches,
I've been dealing with MAC for 20 years, so I don't find it hard at all, and if people are willing to put in the effort to learn it the reward is worth it. But this is a world where most people want to get home to watch their latest story instead of doing any kind of mental work, and admins are no different.
SEBSE is a Framework, MAC is an implementation, those are two different things on different levels.
>MAC framework, but I've never heard of it. What is it called?
It's called MAC...you still don't see the difference?
https://docs.freebsd.org/en/books/handbook/mac/
Look i stop here you have obviously no knowledge of MAC.
>I've been dealing with MAC for 20 years
Yeah no you don't since you don't even know the difference of SELinux and the/a MAC implementation.
> SEBSE is a Framework, MAC is an implementation, those are two different things on different levels.
This is incredibly wrong unless you are referring to something other than mandatory access controls when you say MAC.
MAC is a concept. SELinux AND SEBSD are implementations. And yes, you can say they are implementations of FLASK, or call them frameworks, but semantics aside none of that changes that SELinux and SEBSD are implementations of a concept.
Saying MAC is an implementation is just flat out wrong.
And for what it's worth, I was correct when I said it was SEBSD, even though it isn't called that anymore. That's what the project started off as before it was merged: http://www.trustedbsd.org/sebsd.html
> Yeah no you don't since you don't even know the difference of SELinux and the/a MAC implementation.
The irony here lol.
Please don't reply to me again.
And this is exactly what I'm talking about. Putting more energy into hoping no one ever gets root rather than providing anything to protect against the scenario where it is obtained.
You can say that, but unveil(2) will hide whole portions of any file system from a application. I do not think selinux does that (correct?). For example, firefox can only access ~/.mozilla/firefox, ~/Downloads and that is it (IIRC). There may be other directories I forgot.
pledge(2) will hide system calls to prevent unintended processing.
But, for the application programmer, it may be a bit harder and you need to know what you are doing. One hopes the programmer knows what he is doing :)
IMHO, pledge(2) and unveil(2) is far easier to maintain than selinux, containers etc. But of course the programmer needs to code it. And not to mention, these calls are far more efficient than the multiple containers out there.
(edit) I also saw somewhere a person is tring to get pledge(2) and unveil(2) into Linux, but I forgot all the details.
Things like SELinux by contrast allow you to remove all access and power from the root account except for what it specifically needs, while allowing you to grant root capabilities to other accounts as needed.
pledge and unveil are toys by comparison.
Can you show me a proof of concept here?
Yes. There is no mechanism at all to unmask the vnode. If you believe otherwise, I'd like to see a proof of concept.
It's cute that the OS might play some tricks to hide them from you, but at the end of the day, that's all it is doing, is hiding. Security by obscurity.
You're just betting you hid them really really well and that no one will be able to find them. With zero preparation in the event someone does.
If you have root, you have access to the files, period. Someone determined enough could search, read and write to the disk directly, bypassing the filesystem.
unveil is defense in depth, but it shouldn't be treated as a final line of defense which too many obsd fans seem to want to do.
Again, proof of concept please.
So you just want to have blind faith in what is basically security via obscurity with extra steps.
That's fine, but it's not an approach to be taken seriously or advocated.
I won't be replying again as I don't see how this is wroth my time when you can't even support anything you've said.
You don’t get any kind of god powers like kernel memory access, if you don’t have access to the system call you can’t debug.
Like I don’t understand how to explain to you that SELinux locks down system calls in an identical way conceptually.
If you can actually exploit a system call, neither a MAC based approach or a pledge will help.
You're making a distinction about 'privileged' system calls, why, exactly? You really think something like Oracle won't require access to a ton of syscalls to work correctly?
> If you can actually exploit a system call, neither a MAC based approach or a pledge will help.
MAC will, pledge won't.
For example with SELinux:https://www.kernel.org/doc/Documentation/prctl/seccomp_filte...
Linux has added a direct pledge+unveil clone to improve the situation: https://raw.githubusercontent.com/torvalds/linux/master/Docu...
That's not a 'weaker version of pledge', it's part of a much larger framework with much greater enforcement capabilities.
seccomp-bpf is a more fragile version of pledge; you need to keep changing your sandbox whenever you upgrade glibc, because there's no mechanism to keep syscall usage in sync between the kernel and userspace.
It was fine for my case, because I was implemeting my own direct system calls, and froze any external dependencies, but it's typically very fragile across system and dependency upgrades.
Also, can you please not perpetuate flamewars on HN generally? It's not what this site is for, and destroys what it is for.
I've banned the other account and I'm not banning yours because as far as I can tell you didn't break the rules as badly and you haven't been making a habit of it—but if you'd please review https://news.ycombinator.com/newsguidelines.html and not do this kind of thing in the future, we'd appreciate it.
That statement indicates you do not fully understand pledge(2) and unveil(2).
For example, if you are running Firefox using the default settings as root on OpenBSD. Firefox will not be able to see anything outside of the directories allowed by unveil(2). So an errant JS will not be able to peruse "veiled" Directories. For instance, Firefox cannot see any data in ~/.ssh no matter what ID you are using.
Not so, not so at all. You have to take my statement in context.
The point was those technologies, while nifty, are entirely irrelevant if an attacker gains remote root.
https://www.openbsd.org/faq/faq14.html#softraidFDE
Above is the section of the FAQ about full disk encryption for OpenBSD 7.2 which is the most recent release.
PS: I agree fully with a parent comment that the OpenBSD installer is superb. It is like working your way through a checklist. Nice and straight forward once you have read the FAQ first.
Semantics doesn't change that.
My point is that OpenBSD wants to focus on trying to remove bugs as much as possible, and securing the system in case a bug is exploited is very much a secondary objective.
Pledge and unveil are nice, but insufficient compared to what SELinux can do, or similar systems. That's simply objectively true, regardless of if your bias towards OpenBSD will let you admit it or not. So no, my point isn't wrong, you're just possibly out of your depth in this convo.
OpenBSD provides NOTHING to protect against an attacker who gets remote root, it just really really really hopes they won't.
Alright, if we're talking about things that are objectively true, this is objectively false. They specifically go to great lengths to make it impossible to get root.
I'm not talking about GETTING root, I'm talking about in the situation where an attacker already HAS root.
Besides, by that logic, your systems are totally insecure because you're just really really really hoping the attacker doesn't get an unconfined process.
Saying OpenBSD 'really really really hopes' an attacker doesn't get root is still 100% correct. They are still prioritizing removing bugs ahead of time instead of providing stronger tools to prepare for a scenario where someone does exploit a bug they didn't get to.
> Besides, by that logic, your systems are totally insecure because you're just really really really hoping the attacker doesn't get an unconfined process.
Which is where things like SELinux come in, which is my whole point.....
They're absolutely dealing with that situation as well. So let's say for instance that the default OpenBSD httpd turns out to have some bug that allows an attacker to execute any arbitrary code of their choosing on the server. That's a problem, but it's hardly the end of the world. First, the server is using pledge+unveil, so it can't actually do all that much other than serving web pages. Second, even if the attacker can work within those restrictions, they're running as a completely unprivileged user, because OpenBSD is actually a great fan of defense in depth.
> Which is where things like SELinux come in, which is my whole point.....
If we're assuming an attacker can run things as root, then obviously we have to assume that they can run things in an unconfined context as well, at which point you've just basically added a second word for root and nothing helps.
So, as I said, good luck trying to call something outside pledge. Being root? You wish if you can even run something alonside the process. Because even the subprocesses from a pledged process are pledged too.
You're just proving what I said above 100% correct, that the OBSD approach is to just really hope that no one ever gets remote root.
Tell you what, getting root on a properly configured MAC system is 100% impossible, so I'll choose that over a system where the developers just really really hope something bad won't happen every day of the week.
That's why I said they were prioritizing. The very use of that word obviously means they are not just working on one thing....
Why do you keep arguing against strawmen?
I won't address your scenario because you chose one that is overly favorable to your point, it doesn't really show anything.
> If we're assuming an attacker can run things as root, then obviously we have to assume that they can run things in an unconfined context as well,
That's a really, really dumb assumption, because that kind of issue is addressed. It's an entirely different user from root that administers selinux policies....no reason at all to assume root can run unconfined when that's counter to the whole point of running SELinux....
These are really, really bad arguments you are making.
SELinux and similar frameworks do prevent further compromise.
That's the point.
The reason it's wrong, the reason it's clear that you have no clue what you are talking about, is it's a basic design decision that it is not the root user that configures or removes policies. Root has no ability to do what you describe, and under SELinux is just another user.
But hey, keep spouting nonsense because your feelings are hurt because you have some weird attachment to an obscure OS.
> If you believe that there's a way to become root, why don't you believe there's a way to become some other non-root user?
SELinux maintains it's own set of users that map to unix user. SELinux users have their own roles and capabilities that are very granular. It's frequently configured to make root 'just another user' as much as possible, and remove any access root doesn't need.
So if you 'get root' on a properly configured SELinux system, you don't actually 'have root', you just have a user with a user id of 0 but no real access.
You can switch to whatever user administers SELinux policies, but you will only be switching in the unix sense, not the SELinux sense, so you won't actually be able to do anything.
To do what you are suggesting would require a kernel exploit, but even if you get root you won't have any access to run it.
What you suggest simply is not possible barring very specific vulnerabilities being found, which afaik, have not been.
Pledge and unveil maintains it's own set of system calls and allowed paths that a process may access. Pledge contexts have their own system call masks and capabilities that are very granular. It's hard coded to make root 'just another system call maker' and remove any access root doesn't need. The permissions are ratcheted down as the program initializes, so a process typically doesn't even have access to its own configuration, let alone anything important. Once httpd is started, it can't even start listening on a port -- no matter how much root it has.
So if you 'get root' on a properly pledged system, you don't actually 'have root', you just have a program with a user id of 0 but no real access to make system calls.
You can switch to whatever user you want, but you will only be switching in the unix sense, and not changing the allowed set of system calls and data accesses, so you won't actually be able to do anything.
To do what you are suggesting would require a kernel exploit, but even if you get root you won't have any access to run it.
What you suggest simply is not possible barring very specific vulnerabilities being found, which afaik, have not been.
What is with you and putting so much effort just to argue out of loyalty and emotion when you don't even know the thing you are arguing against, aside from assumptions?
Pledge is a goddamn toy and it's hilarious you think otherwise. Pledge isn't going to do ANYTHING to protect, say, a complex database server or any number of third party programs that need all kinds of access, and require being able to make all types of syscalls.
You've parroted my own reply back at me, without bothering to comprehend it or realize that it doesn't apply at all in the way you've bastardized it and tried to force it to.
You clearly, CLEARLY have no clue at all what you are talking about, and are basically just the OBSD version of a linux zealot stereotype.
In short, you're a tool. You clearly need to have the last word due to your ego, so have at it, but I won't waste my time by humoring you further. It's genuinely a shame I can't block you and will still be notified of whatever noise you make next.
Here's a program that will run an arbitrary command under the same restrictions as Chrome. Presumably, Chrome counts as a complex third party program.
#include <stdio.h>
#include <unistd.h>
#include <err.h>
#define CHROME_MAIN_CONTENT_PLEDGE \
"rpath wpath cpath stdio tmppath dns inet flock "
"unix proc prot_exec tty sendfd recvfd exec mcast "
"ps drm fattr route getpw vminfo video"
int
main(int argc, char **argv)
{
if(argc < 2)
errx(1, "usage: %s cmd\n", argv[0]);
if(pledge(CHROME_MAIN_CONTENT_PLEDGE, CHROME_MAIN_CONTENT_PLEDGE) == -1)
err(1, "pledge");
if(unveil("/var/empty", "r") == -1)
err(1, "unveil");
if(unveil(NULL, NULL) == -1)
err(1, "unveil");
execv(argv[1], &argv[1]);
err(1, "exec");
}
cc -o asroot asroot.c
doas ./asroot your-exploit-command
How about you show me how you lock down an OBSD system that is running, say Oracle. With pledge.
Though I'm not sure why you think it'd be much harder than pledging Chrome or Firefox.
If you want t discuss this further, illustrate how pledge can protect against a remote root exploit in Oracle as good or better than SELinux can.
What operations would you use to break out of the above sandbox? Once you give me that, I'll happily discuss the details.
I'm not interested in retreading anything. I don't even want to continue discussing anything with you as it is, but I'm willing to see where this goes.
Show how pledge can protect against a remote root compromise in oracle. I doubt it can, you've been insisting pledge is all OpenBSD needs to protect against such attacks.
I'm not answering any more questions or taking any more detours. Stop being intellectually dishonest.
Either admit that you don't know for sure that pledge can protect against a remote root compromise in complex third party software like oracle and simply think it can, or show how it can.
If you can't do that, we're at an impasse and there won't be further discussion. Unless you post some kind of baiting comment I suppose, which wouldn't surprise me at this point.
The program upthread will let you do that.
It won't be as good as Oracle doing its own pledges, of course, because well pledged programs narrow down their pledge set as they run -- but if you have a binary that you can't modify, this is as good as you can do with any security framework. For example, Linux has landed on the same model as pledge when they added landlock, where a program is expected to landlock itself after opening the initial handles to resources.
And if it needs access to an awful lot of data across the file system or raw devices (which is not uncommon for complex proprietary commercial software), what then?
> It won't be as good as Oracle doing its own pledges, of course
And that's part of the issue, because you can't count on or only use software which is making pledges, and no commercial software will at all.
> but if you have a binary that you can't modify, this is as good as you can do with any security framework.
This is where a MAC framework has a strength and advantage over pledge.
With pledge/unveil and oracle, you likely won't be able to sufficiently restrict access to data or devices to prevent lateral movement and prevent further compromise, as complex programs like that require too much access to function correctly.
With a MAC framework you can. Now the binary has no access to certain parts of the filesystem in a way that is much more secure than a chroot, only has access to syscalls it needs (on par with pledge here), can't execute other binaries, can't make library calls, can't access devices (e.g. tty, network), can't switch user context, can't do anything as root even if root were obtained, can't write to ANY files only append, can't delete ANY files, etc etc etc. You have complete granular control over many more aspects of what to permit or deny.
Pledge and unveil simply can't compare. In some scenarios they may be sufficient, but in most, and certainly in most where an attacker has root, they are not.
The same thing as any other framework. You either grant it access, or you don't. The list of files just gets long.
> And that's part of the issue, because you can't count on or only use software which is making pledges, and no commercial software will at all.
Luckily, you can pledge and unveil from outside the process, as the code above does. Once more, that runs an arbitrary command under pledges and unveils.
The thing lost is the ability reduction in access during the program lifetime, which is something that can only be done with the cooperation of the binary running; you don't have enough information to know when (for example) the code setting up the last network listener needed has been run without the help of the binary. You don't get that in most security frameworks anyways, so this is not a weakness in comparison to them.
> With a MAC framework you can. Now the binary has no access to certain parts of the filesystem in a way that is much more secure than a chroot, only has access to syscalls it needs (on par with pledge here), can't execute other binaries, can't make library calls, can't access devices (e.g. tty, network), can't switch user context, can't do anything as root even if root were obtained, can't write to ANY files only append, can't delete ANY files, etc etc etc. You have complete granular control over many more aspects of what
Yes, that's what you get with pledge and unveil.
If you think otherwise -- I gave you a program that runs any arbitrary binary; please show a demo of a jailbreak.
No. Not even close. You're missing the point.
There is a fundamental difference which you are ignoring or blind to.
With something like pledge, you can't limit anywhere near as much as what you can limit with a MAC framework.
This means with pledge, you can only limit to what the program needs, which may well be enough (and with something like Oracle it's almost guaranteed) for an attacker to escape whatever confines you think you have in place. That simply isn't true for a MAC framework where you can limit to a greater extent and with a lot more granularity.
> Luckily, you can pledge and unveil from outside the process, as the code above does.
That doesn't help even slightly and has no bearing on the point I made.
> Yes, that's what you get with pledge and unveil.
Jesus christ no you don't. Not even close. At this point it's abundantly clear you are not familiar with MAC frameworks at all, and you're just convinced that pledge and unveil are the same thing. That's willful ignorance and nothing more.
> If you think otherwise -- I gave you a program that runs any arbitrary binary; please show a demo of a jailbreak.
You haven't shown that pledge and unveil are as good at SELinux at locking down something like Oracle, you're just claiming it is with hand waiving. pledge and unveil simply don't offer what SELinux does, and you're just dismissing that and insisting they do.
Ridiculous.
Go ahead, tell me a specific thing that I can't limit with pledge and unveil. Note that everything you mentioned when discussing Oracle could be done with them.
I've provided code. You haven't even listed a threat model. You just yelled a bit about root escapes, and then when challenged to provide an example, yelled about Oracle. And then yelled a bit about how MAC provides more, without any specifics about which relevant 'more' it provides. (I certainly agree that it's more flexible, more complicated, and more difficult to configure -- but not that it's more secure).
Again, there's example code sitting right there, which can be used for something like Oracle. Prove that there's an escape.
Sigh. I've listed numerous examples, and like a dipshit you just change a few words in my reply and parrot it back at me, without seemingly comprehended a thing.
At this point you are just displaying zealotry. Even Theo disagrees with you.
> I've provided code.
Fucking exactly. That shows that you are not following the discussion at all, just blindly defending your niche OS.
We were talking about complex third party programs that are hard to lock down. A short code example using pledge is, now make sure you read these next few words and try to absorb them, entirely fucking irrelevant.
We're not talking about software developers trying to prevent exploitation of trusted code, we're talking about trying to lock down untrusted code in the event of a compromise.
> You just yelled a bit about root escapes, and then when challenged to provide an example, yelled about Oracle. And then yelled a bit about how MAC provides more, without any specifics about which relevant 'more' it provides.
No, I gave specific examples, and you just hand waived them and asserted pledge and unveil offer the same capabilities and defenses without bothering to show how.
> Again, there's example code sitting right there, which can be used for something like Oracle
No it fucking can't, at all, it's entirely irrelevant and has no relationship to oracle at all. Are you on the spectrum or something? How are you failing to follow this discussion to this extent? It's honestly baffling.
That short code example LOCKS DOWN AN ARBITRARY THIRD PARTY PROGRAM that you pass it as a command line argument. I skipped arg parsing for simplicity, but it does exactly what you're asking for.
> No, I gave specific examples
Can you point to the code you posted that breaks out of the sandbox I gave? I seem to have missed it.
> That short code example LCOKS DOWN A COMPLEX THIRD PARTY PROGRAM. The fact that you haven't understood this shows that you're not understanding the discussion.
It doesn't do a fucking thing to lock down Oracle you absolute twit. Oracle won't even start if you try to limit it as though it were Chrome.
Why is it somehow not in the least bit surprising that you a) think that would work and b) think it's a viable solution.
Sigh. This is honestly so fucking stupid.
Pledge can ONLY restrict a program to the syscalls it needs. If the program needs syscalls which can be used to leverage an escape or escalation, pledge is fucking useless. SELinux doesn't have that weakness. That you don't understand this fundamental difference shows you're out of your depth, and that as suspected you're a waste of time to try and educate.
I'm not going to reply to you again given how technically illiterate you are in this context. Given how insecure and religious towards OpenBSD you seem to be, I'm assuming you're going to need to have the last word, and likely will yet again just parrot back a paraphrased version of one of my replies to you...yet again demonstrating your lack of originality or comprehension.
Have fun with that, and I hope having the last word gives you that little boost of self-confidence you seem to be craving to compensate for your clear lack of knowledge.
Best of luck to you in life. You need it.
I'm sure you know a lot and I normally wouldn't ban an account without warning at least once first, but the way you behaved in this thread was just way too extreme. I know this sort of no-holds-barred flamewar is standard in some systems-oriented communities and I don't have any judgment about that, but HN is much larger and we can't afford to have accounts posting like this—it would lead straight to the self-incineration of this site (by masses of people who likely know a lot less than you do).
If you don't want to be banned, you're welcome to email hn@ycombinator.com and give us reason to believe that you'll follow the rules in the future. They're here: https://news.ycombinator.com/newsguidelines.html.
And you're wrong about OpenBSD. I'm not going to get as angry about this as you appear to be, but I recommend you run "man 7 securelevel" on one of the OpenBSD systems you've been playing with for 20 years.
I'm not remotely wrong about OpenBSD, and that manpage doesn't contain anything that contradicts a point I've made.
You claim that OpenBSD has no conpromised-root mitigation. I have provided evidence to back my counterclaim. I will assume your rejection of actual evidence indicates a disinterest in good-faith argument, so I won't bother pressing the point, except to mention that you may be more successful persuading people to support your case if you adopt a reality-based approach to argument instead of just making wild-ass assumptions about other people.
Major external packages are patched to improve security. For example, a compromised chrome or firefox can't observe any directory other than ~/Downloads; pledge and unveil take care of that.
As a concrete example, Firefox can only load/save or download files to ~/Download by default.