Between this and the pip install break all system packages
This has to be about the dumbest change I could possibly gather in the last 20 years of computing. I can’t even imagine breaking this many things all at once. I’m still dealing with the side effects of people’s installers from docker-compose and the pip problems - ansible will just never be the same again. Now this.
The vast majority of Linux users consider systemd as a good thing because it apparently makes system administration easier. They also don’t agree that systemd is monolithic, because it’s actually designed modular.
But of course there are detractors. The only thing I like about systemd is its declarative service definition and parallel service startup. But if I wanted to run an OS with bloated and inscrutable software (even with the source code), my choice wouldn’t be Linux or Systemd.
I also routinely switch parts of my OS. This is harder with systemd. Although it is modular, the modules are so tightly coupled that it will prevent the replacement of modular components with alternatives. Frankly, I think systemd is killing the innovation in system component development.
I use Gentoo with OpenRC. So my position in this matter should be clear. Anyway, check the last paragraph again to see what I think about systemd’s modularity.
Yes. I agreed with you. But I made it sound like something else. Bad wording on my side.
As I’m too Gentoo openrc user. I also use seatd+greetd instead of (e)logind and replacing sysvinit with openrc-init. The availability of choices made me do it!
I really don’t care about the command I use to get elevated privileges. On my Debian servers I use su and maybe in future, if Fedora decides to make the switch, I will use run0.
Will this be an integral part of systemd, or will they release it as a separate thing? I mean, if I like it, but I’m not using systemd (I do use it, but I’m just thinking about it), could I use this run0 (horrible name) without having to buy into all of systemd?
The article says it works by messaging systemd to run the process as the given user, rather than being a SUID binary. So it wouldn’t work without systemd.
This is fine, but why does everything need to be part of Systemd? Like, seriously, why can’t this just be an independent project? Why must everything be tied into this one knot of interdependent programs, and what’s going to happen to all of them when the people who are passionate about it and actually understand all the stupid ways they interrelate move on with their lives? Are we looking at the formation of the next Xorg? Will everybody being scrambling to undo all of this in another 20 years when we all realize it’s become an unmaintainable mess?
It seems a fairly explicit goal of systemd to redefine Linux as a unified platform rather than as a kernel that can run any one of many implementations of many different services. I assume this is not just the systemd lead but also a goal of Red Hat.
Personally, while I am ok with systemd defining itself as a single source for all this functionality, I hate that they are taking away ( or making it hard at least ) to have independent implementations of these services.
What Chinera is doing with dinit and turnstile is really interesting. It would be nice to have feature comparable approaches to the systemd monolith that distributions could choose from.
What Chinera is doing with dinit and turnstile is really interesting. It would be nice to have feature comparable approaches to the systemd monolith that distributions could choose from.
Systemd does a lot of things that could probably be separate projects, but run0 is an example of something that benefits from being a part of systemd. It ties directly into the existing service manager to spawn new processes.
Yeah, if all those complainers want something more modular, they’re free to push for protocols that allow to leverage existing components while also allowing for them to come from multiple vendors.
Systemd does a lot of things that could probably be separate projects,
I dont get the hate for this - Linux is full of projects that do the same thing: coreutils, busybox, kde, gnome, different office suites, even the kernel itself. It is very common for different related projects to be maintained together under the same project/branding with various different levels of integration between them. But people really seem to only hate on systemd for this…
What standards? The old init systems were a loose collection of shell scripts that were wildly different on every distro. Other tools like sudo also broke the established standards of the time, before it you had to login as root with the root password.
Even gnome and KDE have their own themeing standards as well as other ways of doing things. Even network manager is its own standard not following things that came before it. Then there are flatpack, snaps and app images. Not to mention deb vs rpm vs pacman vs nix package formats. Loads of things in Linux userland have broken or evolved the standards of oldern times.
Systemd breaks its own standards. Oh, were you making a replacement for this component of systemd that does some things the systemd version doesn’t? Well the latest version of systemd just changed the Unix socket protocol that it uses to communicate with the rest of systemd from text based to binary. Sorry for the lack of warning.
@AVincentInSpace@nous I've always disliked the arrogance of the lead Dev & the inexorable incremental usurping of Linux functionality. I'm deeply uncomfortable with so much being absorbed into a big binary black box
I guess for me the difference is that the kernel is just way beyond what I can understand and has never had any viable alternatives, gnome I really don’t like, and everything else you listed is just collections of simple stuff that aren’t actually very interdependent. Systemd is a giant mess of weirdly interdependent things that used to be simple things. Sure, some of them weren’t great, but every major distro abandoning all of the alternatives feels like putting all of our eggs in one basket that’s simultaneously getting more important and more fragile the bigger it gets.
Except desktop environments - they are far from a simple loosely collection of simple stuff. They coordinate your whole desktop experience. Apps need to talk to them a lot and often in ways specific to a single DE. Theming applications is done differently for every toolkit there is, startup applications (before systemd) is configured differently, global shortcuts are configured differently by each one… If anything it is something you interact with far more than systemd and has far more inconsistencies between each one. Yet few people complain about this as much as they complain about systemd.
Systemd is a giant mess of weirdly interdependent things that used to be simple things.
They used to be simple things back when hardware and the way we use computers were much simpler. Nowadays hardware and computers are much more dynamic and hotplugable and handle a lot more state that needs to persist and be kept track of. www.youtube.com/watch?v=o_AIw9bGogo is a great talk on the subject and talks about why systemd does what it does.
Okay, but why go about it that way? That can’t be the only way of making a viable alternative to sudo. Why does everything need to be part of one project? If you want to reuse code why not spin it out into a library so each component can be installed with just the libraries it needs and not the depending on the whole gigantic thing? KDE works that way. It’s obviously possible for some things, at least.
One of my favorite things about Linux is simply fiddling around and finding the things I like and don’t and just using the ones I do. I can’t do that effectively with systemd though. Sure, it’s theoretically modular, and there are even a couple parts left that can work independently, but mostly it’s just one big block of half an operating system that all gets lumped together into one gigantic mess, and I can’t effectively just use the bits I like. It’s kind of all or nothing, and then maybe being allowed to double up on some of the things I’d like to use an alternative to… for now. It just kinda sucks the joy out of using my computer, but trying to avoid it completely is a massive pain in the butt.
There’s no big dramatic thing wrong with systemd. Using systemd and being happy with it is a good thing. I do not object to the existence of systemd. Systemd is fine. It just makes me like Linux less is all. I am enjoying my time with my computer less than I used to, and the universal dominance of systemd is probably the biggest reason for that.
That’s rich, coming from a company that sued a child whose website domain name was mikerowesoft.com. (His name was Mike Rowe, and the site was about the software he made).
I never understood the hate, tbh. A lot of users don’t even care if Sysd is used, as long as it works. So… Since the majority of distros use it… I think it works enough.
I understand the concern about the future and we have seen overbloated projects have issues. In the long run though I will use what works best for me and only get into philosophical comparisons if im making the choice between relatively equal options.
I think some of the hate is from the main systemd dev, Poettering, being so abrasive on social media. He’s got a hateboner for certain distros (which don’t ship with systemd as the default).
It seems to me to be mainly from people who are dedicated to the Unix philosophy that programs should do only one thing, and do it well. Tying everything up into systemd doesn’t follow that. I don’t care either, and I don’t mind systemd, but some people care about it enough to throw paragraphs of hate on it wherever it’s mentioned online. And apparently it’s “bloat”, and to some " bloat" is worse than the devil himself.
If you dig deeper into systemd, it’s not all that far off the Unix philosophy either. Some people seem to think the entirety of systemd runs as PID1, but it really only spawns and tracks processes. Most systemd components are separate processes that focus on their own thing, like journald and log management. It’s kinda nice that they all work very similarly, it makes for a nice clean integrated experience.
Because it all lives in one repo doesn’t mean it makes one big fat binary that runs as PID1 and does everything.
This is what turned me around: investigating and realizing that it is following the unix philosophy, it’s just under the hood (under the other hood inside the bigger under the hood).
My main issues are that it obfuscates things and seems to consume everything it can into itself.
Honestly, if it were more transparent and designed in a way to easily facilitate swapping out components with alternatives, I’d be a lot more okay with it.
This just sounds like a bad idea, a solution in search of a problem. Sure, sudo is a setuid binary, but it’s a fairly simple program, and at some point, you have to trust the code. It’s also a very fundamental piece of the system that you want to always work, even (especially!) when other things get borked. The brief description of run0 already has too many potential points of failure.
sudo is not a fairly simple program. Last I checked, it had ~177k lines of code. It provides functionality far beyond what is needed of an average user. doas is a simpler alternative (also using SUID) at ~3k lines of code. It comes from OpenBSD. There is absolutely a problem when it comes to SUID binaries. If you can find a way to exploit the permissions given at the start of the SUID binary before user authentication occurs (since the UID is set before the binary runs), you have yourself a full privilege escalation vulnerability. systemd is very well integrated with the distros that use it, being the first process to run after the kernel is initialized. There will never be a point at which systemd is not functioning, but the rest of your system is perfectly fine. It is an absolutely necessary part of the system (assuming your distro uses it), and if it goes down, you have to restart your system. As such, I don’t see any validity to the statement “you want to always work, even (especially!) when other things get borked”. What exactly do you see as being an issue with run0? What specific part of its implementation do you seem to have a problem with? It’s just a symlink to systemd-run, which is already very well tested and has been around for a long time. It’s also far simpler than sudo, and removes the attack surface of running an SUID binary of its size. What “points of failure” do you see here, exactly?
I’ve actually ran into some of those problems. If you run sudo su --login someuser, it’s still part of your user’s process group and session. With run0 that would actually give you a shell equivalent to as if you logged in locally, and manage user units, all the PAM modules.
systemd-run can do a lot of stuff, basically anything you can possibly do in a systemd unit, which is basically every property you can set on a process. Processor affinity, memory limits, cgroups, capabilities, NUMA node binding, namespaces, everything.
I’m not sure I would adopt run0 as my goto since if D-Bus is hosed you’re really locked out and stuck. But it’s got its uses, and it’s just a symlink, it’s basically free so its existence is kBs of bloat at most. There’s always good ol su when you’re really stuck.
I have 0 knowledge of these things, but I do know that people always comment that sudo is bloated, that nobody is truly using everything that sudo can do, only one basic command.
Nobody is using all of sudo’s features because those features are for different use cases. Case in point, LDAP support. At home, pretty much nobody uses it. But on the job, where there are tens to hundreds of machines that someone might need, and they’re all hooked into LDAP for centralized authentication management, it makes sense to have that built into sudo. Same with Kerberos support - at home, forget it, but in a campus environment where Kerberos (and possibly AFS) are part of the network, it makes sense.
sudo is a setuid binary, but it’s a fairly simple program
Some people would disagree to this.
The brief description of run0 already has too many potential points of failure.
If the “listener” is PID1, which will run the privileged command, in theory, it would be quite bullet proof (in a working system PID1 is always there). But since this is systemd, PID1 is much more than that and much more complex. On the other hand spawning another daemon from PID1 to be the “listener” makes it, perhaps, even more complicated. You’d have to make sure the listener is always running and have some process supervisor there to watch if it exits… and maybe even a watchdog polling it to make sure it isn’t frozen.
So my conclusion is the same as yours:
a solution in search of a problem
We already have a working solution. Have a well written SUID program. I’ve been using doas for some years now. It’s simple enough that I trust it.
On a server, it allows you to track who initiates which root season session. It also greatly minimizes the attack surface from a security perspective to have admin privileged accounts unable to be remotely connected to.
Separate ssh keys for the user and the admin? Yeah, see point 2, admins should not be remotely accessible.
Really? How, exactly? Break the ssh key authentication? And wouldn’t that apply to all accounts equally?
Keys aren't perfect security. They can easily be mishandled, sometimes getting published to GitHub, copied to USB drives which can easily be lost, etc.
Further, there have been attacks against SSH that let malicious actors connect remotely to any session, or take over existing sessions. By not allowing remote access on privileged accounts, you minimize risk.
Forcing a non privileged remote session to authenticate with a password establishes a second factor of security that is different from the first. This means a cracked password or a lost key is still not enough for a malicious actor to accomplish administrative privileges.
A key is something you have
A password is something you know
So, by not allowing remote privileged sessions, we're forcing a malicious actor to take one more non-trivial step before arriving at their goals. A step that will likely be fairly obvious in logs on a monitored machine.
If I get into your non-privileged account I can set up a program that acts like sudo and I bet 99% of people will never notice they just gave their password away. And even if they do it’s too late anyway because I’ve just compromised root and locked everybody out and I’m in there shitting on the filesystems or whatever. Because root can do anything.
And if I can’t break into your non-privileged account then I can’t break into a privileged account either.
These artificial distinctions between “non-privileged” and “superuser” accounts need to stop. This is not good security, this is not zero trust. Either you don’t trust anybody and enforce explicit privilege escalation for specific things, or just accept that you’re using a “super” paradigm and once you’ve got access to that user all bets are off.
I strongly disagree with your premise. Separating authentication and privilege escalation adds layers of security that are non-trivial and greatly enhance resilience. Many attacks are detected and stopped at privilege escalation, because it happens locally before a user can stop or delete the flow of logs.
If I get into your non-privileged account I can set up a program that acts like sudo
No you cannot. A non privileged user doesn't have the access necessary to run a program that can accomplish this.
And even if they do it’s too late anyway because I’ve just compromised root and locked everybody out and I’m in there shitting on the filesystems or whatever. Because root can do anything.
Once again, you didn't privilege escalate, because once you have a foothold (authentication) you don't have the necessary privileges, so you must perform reconnaissance to identify an exploitable vector to privilage escalate with. This can be any number of things, but it's always noisy and slow, usually easy to detect in logs. There is a reason the most sophisticated attacks against well protected targets are "low and slow".
And if I can’t break into your non-privileged account then I can’t break into a privileged account either.
You're ignoring my points given regarding the risks of compromised keys. If there are no admin keys, there are no remote admin sessions.
These artificial distinctions between “non-privileged” and “superuser” accounts need to stop. This is not good security, this is not zero trust. Either you don’t trust anybody and enforce explicit privilege escalation for specific things, or just accept that you’re using a “super” paradigm and once you’ve got access to that user all bets are off.
Spoken like someone who has never red teamed or purple teamed. Even admin accounts are untrusted, given only privileges specific to their role, and closely monitored. That doesn't mean they should have valid security measures thrown away.
That's called 'privilege escalation', and replacing system level calls with user level calls is closely watched for and guarded against with many different security measures including SELinux.
You've already outed yourself multiple times in this thread as someone who doesn't understand how security in the real world works. Take the L and try to learn from this. It's okay not to understand something. But it's very important to recognize when that happens and not claim to understand better than someone else.
It’s not privilege escalation because it doesn’t subvert the correct authentication mechanism, it leverages it. This particular technique is called UI input capture. It’s a script that shows a password prompt then uses your password to do things as root. Nothing untowards would be detected. The main defense is to always run commands with full path – which nobody does.
Separating the process of reaching root in two steps does nothing to improve security, it actually increases complexity and subverts security. If the system is set up to SSH as root you either have a good key or you don’t. If you force people to SSH as individual users and then use a complex mechanism to reach root you create opportunity for a hundred more attack methods, and add a false sense of security.
Input capture btw is not the only method. Sudo has a lot of them. Another very common one is leveraging the password cache timeout.
You’ve already outed yourself
How about we skip the dick measuring contest and we stick to the discussion at hand.
sudo and friends allow you to gain root access while not enabling the root account. If the root account has no credentials then nobody is guessing your password and logging in as an admin.
On a multi-user system it allows for multiple admins without sharing a password. It also allows providing admin access for “some” things but not others.
If the root account has no credentials then nobody is guessing your password and logging in as an admin.
They just need to log in as you and trick you into entering your password in a seemingly legit prompt.
On a multi-user system it allows for multiple admins without sharing a password.
Multiple distinct ssh keys do the same. As long as everybody ends up doing things as the same user it’s all moot anyway.
It also allows providing admin access for “some” things but not others.
Can I provide selective access to just some files? Just some network interfaces? Just some ports? Just some parts of RAM or CPU? Without being able to change those limits?
The point I’m trying to make is that having just one “super” account for everything is a very poor idea. A lot of work has gone into filtering access to the root account and very little into getting rid of the root account. Ideally nothing should run as root, it should run as individual accounts with varying levels of access on a need-to-have basis.
Sure, sudo is a setuid binary, but it’s a fairly simple program, and at some point, you have to trust the code.
Have to trust the code ? doas for OpenBSD was created because of issues with sudo.
Talking with deraadt and millert, however, I wasn’t quite alone. There were some concerns that sudo was too big, running too much code in a privileged process. And there was also pressure to enable even more options, because the feature set shipped in base wasn’t big enough.
I mostly agree with your write-up here. That said, I do think that systemd does want to eliminate SUID. I also think they want to absorb most of the low level system plumbing.
There are other applications that use suid (like newuidmap). And there are programs that use capabilities (like ping). I’m pretty sure that this logic will be used to justify assimilating those applications too. But I’m sure that the crowd will cheer them on as if they did something revolutionary.
it took less than a day for someone to break run0 totally open, so basically, you have a choice between a well tested/debugged sudo and this new thing which may eventually mature
As far as I know, the exploit you are referring to, wasn’t actually a vulnerability. youtu.be/awkoa_WxFIg?feature=shared&t=659 Although feel free to correct me on that one
SUID stands for Set User ID. An SUID binary is a file that is always run with the UID of the owner user (almost always root). Note that this does not require that the user running them has root permissions, the UID is always changed. For instance, the ping command needs to set up network sockets, which requires root permissions, but is also often used by non-root users to check their network connections. Instead of having to sudo ping, any normal user is able to just run ping, as it uses SUID to run as the root user. sudo and doas also require functions that necessitate them running as root, and so if you can find out how to exploit these commands to run some arbitrary code without having to authenticate (since authentication happens after the binary has started running), there is a potential for vulnerabilities. Specifically, there is the privilege escalation, which is one of the most severe types of vulnerabilities.
run0 starts using systemd-run, which does not use SUID. Instead, it runs with the permissions of the current user, and then authenticates to the root user after the binary has already started to run. systemd-run contacts polkit for authentication, and if it succeeds, it creates a root PTY (pseudo-terminal/virtual terminal), and sends information between your session and the root PTY. So this means that in order to achieve privilege escalation with run0 as root, you have to actually authenticate first, removing the “before authentication” attack surface of sudo and doas.
TL;DR SUID binaries will always run as the owner (usually root), even before any form of authentication. run0 will start with the permissions of the current user, and then authenticate before running anything with root permissions.
Suid is a bit set on executables that results in them being run as the user that owns the file without needing a password, for example, passwd as root.
But also completely useless. Run0 ignores the suid bit for the same reason as 99% of command line apps do: it ignores because it isn’t relevant to its functionality.
Basically, the SUID bit makes a program get the permissions of the owner when executed. If you set /bin/bash as SUID, suddenly every bash shell would be a root shell, kind of. Processes on Linux have a real user ID, an effective user ID, and also a saved user ID that can be used to temporarily drop privileges and gain them back again later.
So tools like sudo and doas use this mechanism to temporarily become root, then run checks to make sure you’re allowed to use sudo, then run your command. But that process is still in your user’s session and process group, and you’re still its real user ID. If anything goes wrong between sudo being root and checking permissions, that can lead to a root shell when you weren’t supposed to, and you have a root exploit. Sudo is entirely responsible for cleaning the environment before launching the child process so that it’s safe.
Run0/systemd-run acts more like an API client. The client, running as your user, asks systemd to create a process and give you its inputs and outputs, which then creates it on your behalf on a clean process tree completely separate from your user session’s process tree and group. The client never ever gets permissions, never has to check for the permissions, it’s systemd that does over D-Bus through PolKit which are both isolated and unprivileged services. So there’s no dangerous code running anywhere to exploit to gain privileges. And it makes run0 very non-special and boring in the process, it really does practically nothing. Want to make your own in Python? You can, safely and quite easily. Any app can easily integrate sudo functionnality fairly safely, and it’ll even trigger the DE’s elevated permission prompt, which is a separate process so you can grant sudo access to an app without it being able to know about your password.
Run0 takes care of interpreting what you want to do, D-Bus passes the message around, PolKit adds its stamp of approval to it, systemd takes care of spawning of the process and only the spawning of the process. Every bit does its job in isolation from the others so it’s hard to exploit.
But I’ve had so many issues with D-Bus fucking shit up on my systems that I’d be very reluctant to hinge my only way of recovering from failures upon something so brittle.
Granted, D-Bus hasn’t given me any trouble since moving to NixOS. The hell of trying to recover my arch systems from a perpetually failing D-Bus would make me very apprehensive to adopt this. I could see myself using run0 by default, but keeping sudo-rs or doas around with a much stricter configuration as a failsafe until the run0 + D-Bus + PolKit is absolutely stable and bulletproof.
I haven’t had D-Bus problems in quite a while but actually run0 should help with some of those issues. Like, systemctl --user will actually work when used with run0, or at least systemd-run can.
Haven’t used it yet so it’s all theoretical, but it makes sense to me especially at work. I’ve used systemd-run to run processes in very precise contexts, it’s worth using even if just to smush together schedtool, numactl, nice, taskset and sudo in one command and one syntax. Anything a systemd unit can do, systemd-run and run0 can do as well.
I’m definitely going to keep su around just in case because I will break it the same I’ve broken sudo a few times, but I might give it a shot and see if it’s any good just for funsies.
Just trying to explain what it does and what it can do as accurately as possible, because out of context “systemd adds sudo clone” people immediately jump to conclusions. It might not be the best idea in the end but it’s also worth exploring.
Some executables are special. When you run them, they automagically run as root instead! But if sudo isn’t very, very careful, you can trick it into letting you run things as root that you shouldn’t be able to.
Run0 DM’s systemd asking it to go fork a process as root for you, and serves as the middleman between you and the other process.
(I’ll attempt this based on my understanding of both)
Pouring a cup of juice is something an adult needs to be involved with.
sudo is when you ask for permission to pour your own cup of juice. You ask an adult, they give you the cup and the juice, and then you’re responsible for pouring it. If the adult isn’t paying attention they may leave the fridge open for you to go back for more juice or another beverage, but otherwise you’re limited to the amount of juice the adult has given you.
run0 is when the adult just gets you a cup of juice. You tell them what you want, they go and pour the juice, and just give you the cup with the juice in it. You never enter the kitchen, so you don’t have access to the fridge, just your cup of juice.
Some people are opposed to sudo being a fairly complex program with an awkward to understand configuration language and a couple of methods that can fetch config from elsewhere. Fixing upstream sudo can't happen because those features exist and are presumably used by some subset of people, so straight up removing them is not good, but luckily doas and sudo-rs exist as alternatives with a somewhat stripped featureset and less footguns.
Others are opposed to the concept of SUID. Underneath all the SUID stuff lies far more complexity than is obvious at first sight. There's a pretty decent chunk of code in glibc's libdl that will treat all kinds of environment variables differently based on whether an executable is SUID, and when that goes wrong, it's reported as a glibc bug (last year's glibc CVE-2023-4911 was this). And that gets all the more weird when fancy Linux features like namespaces get involved.
Removing SUID requires an entirely different implementation and the service manager is the logical place for that. That's not just Lennart's idea; s6, as minimal and straight to the point as it tends to be, also implements s6-sudo{,d,c}. It's a bit more awkward to use but is a perfectly "Unix philosophy" style implementation of this very same idea.
news.itsfoss.com
Newest