For context for other readers: this is referring to NAT64. NAT64 maps the entire IPv4 address space to an IPv6 subnet (typically 64:ff9b). The router (which has an IPv4 address) drops the IPv6 prefix and does a normal IPv4 NAT from there. After that, you forward back the response over v6.

This lets IPv6 hosts reach the IPv4 internet, and let you run v6 only internally (unlike dual stack which requires all hosts having v6 and v4).

You can do rollbacks if you’re using something like home-manager on a foreign distribution. It’s just a bit more janky admittedly.

There’s a transaction fee, the higher you pay the more priority you have (since miners get a cut).

The vulnerability has nothing to do with accidentally logging sensitive information, but crafting a special payload to be logged which gets glibc to write memory it isn’t supposed to write into because it didn’t allocate memory properly. glibc goes too far outside of the scope of its allocation and writes into other memory regions, which an attacked could carefully hand craft to look how they want.

Other languages wouldn’t have this issue because

1. they wouldn’t willy nilly allocate a pointer directly like this, but rather make a safer abstraction type on top (like a C++ vector), and

2. they’d have bounds checking when the compiler can’t prove you can go outside of valid memory regions. (Manually calling .at() in C++, or even better - using a language like rust which makes bounds checks default and unchecked access be opt in with a special method).

Edit: C’s bad security is well known - it’s the primary motivator for introducing rust into the kernel. Google / Microsoft both report 70% of their security vulnerabilities come from C specific issues, curl maintainer talks about how they use different sanitizers and best practices and still run into the same issues, and even ubiquitous and security critical libraries and tools like sudo + polkit suffer from them regularly.

The solution here generally afaik is to give a specific deadline before you go public. It forces the other party to either patch it, or see the problem happen when they go live. 90 days is the standard timeframe for that since it’s enough time to patch and rollout, but still puts pressure on making it happen.

It’s not complicated until your reputation drops for a multitude of reasons, many not even directly your fault.

Neighboring bad acting IPs, too many automated emails sent out while you were testing, compromised account, or pretty much any number of things means everyone on your domain is hosed. And email is critical.

Not in this one, iirc they actually reverse engineered and were working off of apple libraries, rather than proxies.

True, but that doesn’t necessarily matter if I can compromise the privileged app instead. I could replace it, modify it on disk, or really any number of things in order to get myself a hook into a privileged position.

Just injecting code in some function call which launches malware.exe would do the trick. Ofc signature checks and the like can help here - but those aren’t a given. There’s any number of ways you can elevate yourself on a system based off of user security if your threat model is malicious processes. Linux (and windows) will stop users from accessing each other’s crap by default, but not processes.

Or: supply chain attacks. Now your official app without any modifications is malicious.

Yep! You can also get pretty far even without containers. At the end of the day containers are just sandboxing using namespaces, and systemd can expose that pretty trivially for services, and tools like bubble wrap / flatpak let you do it for desktop apps. In an ideal world every package would only use the namespaces it needs, and stuff like this would largely not be a concern.

The idea is malware you installed would presumably run under your user account and have access. You could explicitly give it different UIDs or even containerize it to counteract that, but by default a process can access everything it’s UID can, which isn’t great. And even still to this day that’s how users execute a lot of processes.

Windows isn’t much better here, though.

Containers don’t typically have inits, your process is the init - so no extra processes are started for things other than what you care about.

The proper way to handle issues like these is process level permissions (i.e. capability systems), instead of user level. Linux CGroups, namespaces, etc. are already moving that way, and in effect that’s the way windows is trying to head too. (Windows has its own form of containerization called AppContainers, which UWP apps use. Windows also has its own capability system).

As a third party, my understanding is that both the implementation and the protocol are really hard, if not next to impossible to iterate on. Modern hardware doesn’t work like how it did when X did, and X assumes a lot of things that made sense in the 90s that don’t now. Despite that, we cram a square peg into the round hole and it mostly works - and as the peg becomes a worse shape we just cram it harder. At this point no one wants to keep working on X.

And I know your point is that it works and we don’t need too, but we do need too. New hardware needs to support X - at least the asahi guys found bugs in the X implementation that only exists on their hardware and no one who wants to fix them. Wayland and X are vastly different, because X doesn’t make sense in the modern day. It breaks things, and a lot of old assumptions aren’t true. That sucks, especially for app devs that rely on those assumptions. But keeping around X isn’t the solution - iterating on Wayland is. Adding protocols to different parts of the stack with proper permission models, moving different pieces of X to different parts of the stack, etc. are a long term viable strategy. Even if it is painful.

manpages aren’t guides though - they don’t help much in learning new tools, especially complicated ones. They’re comprehensive references, some can literally span hundreds of pages. Useful when you know what you’re doing and what you’re looking for, not great for learning new tools.

This is good for precisely the single user case - potentially malicious services on your system can’t view things they otherwise would be able to, or access resources they don’t need. Even if it’s under the same user.

As a Linux user (and ex arch user btw), I’m deeply offended.

Linus has stepped away from kernel development before, and probably will again. Life continues on.

Second person excited for bcachefs, I’m planning on swapping over as soon as it supports scrubbing.

Right, but squashed commits don’t scale for large PRs. You could argue that large PRs should be avoided, but sometimes they make sense. And in the case where you do have a large PR, a commit by commit review makes a lot of sense to keep your history clean.

Large features that are relatively isolated from the rest of the codebase make perfect sense to do in a different branch before merging it in - you don’t merge in half broken code. Squashing a large feature into one commit gets rid of any useful history that branch may have had.

Yeah, but phabricator and Gerrit are entirely separate workflows from GitHub, and a lot of people prefer that workflow because it leads to encouraging better histories and reviews. It helps you in getting rid of the “fixed typos” type of commits, while still letting you make larger PRs.

GitHub obviously does let you keep a clean git history, but the code review workflow in GH just doesn’t encourage reviewing commits.