Security is not a binary variable, but managed in terms of risk. Update your stuff, don't expose it to the open Internet if it doesn't need it, and so on. If it's a server, it should probably have unattended upgrades.
If it's a server, it should probably have unattended upgrades.
Interesting opinion, I've always heard that unattended upgrades were a terrible option for servers because it might randomly break your system or reboot when an important service is running.
There are two schools of thought here. The "never risk anything that could potentially break something" school and the "make stuff robust enough that it will deal with broken states". Usually the former doesn't work so well once something actually breaks.
Both my Debian 12 servers run with unattended upgrades. I've never had anything break from the changes in packages, I think. I tend to use docker and on one even lxc containers (proxmox), but the lxc containers also have unattended upgrades running.
Do you just update your stuff manually or do you not update at all? I'm subscribed to the Debian security mailing list, and they frequently find something that means people should upgrade, recently something with the glibc.
Debian especially is focused on being very stable, so updating should never break anything that wasn't broken before. Sometimes docker containers don't like to restart so they refuse, but then I did something stupid.
I used to check the cockpit web interface every once in a while, but I've tried to enable unattended updates today. It doesn't actually seem to work, but I planned on switching to Nix anyway.
I don't use Cockpit, I just followed the Debian wiki guide to enabling unattended upgrades. As fast as I remember you have to apt install something and change a few lines in the config file.
It's also good to have SMTP set up, so your server will notify you when something happens, you can configure what exactly.
That only applies to unstable distros. Stable distros, like debian, maintain their own versions of packages.
Debian in particular, only includes security patches and changes in their packages - no new features at all.* This means risk of breakage and incompatibilitu is very low, basically nil.
*exceot for certain packages which aren't viable to maintain, like Firefox or other browsers.
I mean, this isn't any different for Windows or macos. The difference is the culture around the kernel.
With Linux there are easily orders of magnitude more eyeballs on it than the others combined. And fixes are something anyone with a desire to do so can apply. You don't have to wait for a fix to be packaged and delivered.
Just because it has a CVE number doesn't mean it's exploitable. Of the 800 CVEs, which ones are in the KEV catalogue? What are the attack vectors? What mitigations are available?
The idea that it is somehow possible to determine that for each and every bug is a crazy fantasy by the people who don't like to update to the latest version.
The fact that you think it's not possible means that you're not familiar with CVSS scores, which every CVE includes and which are widely used in regulated fields.
And if you think that always updating to the latest version keeps you safe then you've forgotten about the recent xz backdoor.
I am familiar with CVSS and its upsides and downsides. I am talking about the amount of resources required to determine that kind of information for every single bug, resources that far exceed the resources required to fix the bug.
New bugs are introduced in backports as well, think of that Debian issue where generated keys had flaws for years because of some backport. The idea that any version, whether the same you have been using, the latest one or a backported one, will not gain new exploits or new known bugs is not something that holds up in practice.
I don't know where you got the idea that I'm arguing that old versions don't get new vulnerabilities. I'm saying that just because a CVE exists it does not necessarily make a system immediately vulnerable, because many CVEs rely on theoretical scenarios or specific attack vectors that are not exploitable in a hardened system or that have limited impact.
And I am saying that that information you are referring to is unknown for any given CVE unless it is unlocked by some investment of effort that usually far exceeds the effort to actually fix it and we already don't have enough resources to fix all the bugs, much less assess the impact of every bug.
Assessing the impact on the other hand is an activity that is only really useful for two things
a risk / impact assessment of an update to decide if you want to update or not
determining if you were theoretically vulnerable in the past
You could add prioritizing fixes to that list but then, as mentioned, impact assessments are usually more work than actual fixes and spending more effort prioritizing than actually fixing makes no sense.
That is actually perfectly reasonable assumption to make in the absence of resources to determine the opposite, which would probably be many times the resources needed to actually fix the bug.
It's a wild assumption to claim "All bugs in the Linux kernel are security issues", without any backing, whoever is making that claim needs to provide evidence since the default position for any program is that there are bugs that are not security issues.
defend one out there assumption with another, i guess.
who can tell if sidewinder force feedback (11684) is a security bug or just one that affects people using old joysticks. better treat it with all the seriousness of xv just to be sure!
I'm genuinely worried sometimes that a Ken hack has been introduced. I don't know by who, but possibly some government agency. Then again, we also have a Minix system built into the CPU doing god knows what and we just accept that.
I'd just like to interject for a moment. What you’re refering to as Linux, is in fact, GNU/LInux, or as I’ve recently taken to calling it, GNU plus Linux. Linux is not an operating system unto itself, but rather another free component of a fully functioning GNU system made useful by the GNU corelibs, shell utilities and vital system components comprising a full OS as defined by POSIX.
It means anyone including microsoft or apple can use the code contribution or take the entire softwarw and make some modifications and sell it proprietary. Any optimisations or features made by community can be proprietarised
L4. HURD never panned out, and L4 is where the microkernel research settled: Memory protection, scheduling, IPC in the kernel the rest outside and there's also important insights as to the APIs to do that with. In particular the IPC mechanism is opaque, the kernel doesn't actually read the messages which was the main innovation over Mach.
Literally billions of devices run OKL4, seL4 systems are also in mass production. Think broadband processors, automotive, that kind of stuff.
The kernel being watertight doesn't mean that your system is, though, you generally don't need kernel privileges to exfiltrate any data or generally mess around, root suffices.
If you want to see this happening -- I guess port AMDGPU to an L4?
seL4 is the world’s only hypervisor with a sound worst-case execution-time (WCET) analysis, and as such the only one that can give you actual real-time guarantees, no matter what others may be claiming. (If someone else tells you they can make such guarantees, ask them to make them in public so Gernot can call out their bullshit.)
Everybody knows it was. Even Linus said a microkernel architecture was better. He just wanted something working “now” for his hobby project, and microkernel research was still ongoing then.
