• 1 Post
  • 92 Comments
Joined 1 year ago
cake
Cake day: July 1st, 2025

help-circle

  • I didn’t test it (of course), but I think they do for this case.

    So we have a man in the middle attack, that proxies traffic to a website through it and just decrypts it without changing it during the login. This is similar to what cloudflare does. The browser has no way to know somebody is spying: HTTPS cerificate is ok(stolen cert), the domain matches the expected(dns poisoning), even the page is exactly the same, so it performs the passkey challenge/response. The only thing that changed is the certificate, which is not suspicious (certificates can change normally!).

    The attacker is not able to clone the passkey, but they are able though to steal your session cookies after the login. Instead of giving you the “login success page” they keep you waiting on a loading page while they navigate (with a script) with your stolen session cookies to the “add passkey” page of the site to add theirs.

    If the website asks them again for the passkey as confirmation, instead of serving the “login successful page”, they serve the new 2FA prompt to you. You are left wondering “Didn’t I just do the 2FA? What a buggy site!”. And unknowingly approve adding the attacker’s passkey.

    Game over, account fully compromised.

    We are lucky that HTTPs is very good for security, they revoke certificates very quickly if this happens, and browsers check the certificate transparency logs making it impossible to forge a new certificate, even if you got hold of a root certificate with private key “4”. This attack also needs unrealistic level of compromise: stolen and unrevoked certificate + a poisoned DNS. Completely unrealistic scenario, but it’s not passkeys that saved you, it was HTTPS.

    When they say passkeys are unphisheable they mean you can’t usually copy them and give them to an attacker, but an attacker can absolutely still add theirs to your account.




  • Try losing your phone and see how easy is to recover from that, or having to explain to your grandma how they lost their accounts everywhere because they didn’t have a backup strategy for passkeys, even though they remember the password for the account (!).

    There’s also the very big issue of attestation, where websites can force users to use a specific password manager, which doesn’t allow you to move the key to a fully free solution. This was made for companies to force the users to use real hardware tokens or only approved authenticator apps, but nothing prevents google to start forcing you to use only google authenticator to login to google, or for apple to own an iphone to login to a site. Usually all big-company passwors managers don’t have any way to export the key, and you have to wonder why.

    This attestation threat is very real: here is somebody that AFAIK was part of the w3c committee that made passkeys threatening to use attestation against keepassxc for allowing the user to export keys: https://github.com/keepassxreboot/keepassxc/issues/10407

    Overall, it feels like yet another complicated technology, that also allows big companies to consolidate power.



  • Ah sorry to hear. A year ago I had to disable kernel modesetting to make the nvidia drivers work for me, but I did not lose hardware acceleration, so I thought modesetting was optional for the drivers.

    Your experience though sounds like the driver fails to load entirely, and the kernel switches to the standard VESA drivers, which are not really usable. Maybe it’s the VESA drivers not being compatible with modesetting, or they set a weird resolution/refresh rate not supported by the actual panel?

    Best of luck!


  • Depends on your risk model. To simplify my point of view: risk = severity * likelyhood. If it’s unlikely and relatively nonconsequential to overcome, it’s still low risk.

    There is a world where running an old kernel and latest firefox gets you hacked, say maybe buggy GPU drivers with a really really bad opengl vulnerability?. I think it’s quite unlikely. Like what hacker targets your weird GPU driver several years out of date?

    I think it’s much more likely your few-days-outdated LTS version of firefox gets hacked instead, because even if you have an old and fully supported distro, updates are often a bit more delayed (backporting patches takes still some time from maintainers). LTS distro maintainers also do choose what bugs and vulnerability get backported or not, and not everything crosses the line.

    If you do things like accessing your bank with your life savings or your crypto wallet then please do get a more modern distro and live with the hardware incompatibilities. If you use it to browse facebook and the news, read emails and write some random documents, there’s not much there to steal for a hacker anyway.

    IMHO just the fact you are not running windows and are on the latest version of the software means you’ll be all fine, for many, many years, even with an old, unsupported kernel. Your old pc will probably die before a hacker gets in.

    If you have compliance requirements from your company… get your company to buy you a better, more modern pc please. Like, wtf, that thing has probably hardware and firmware vulnerabilities too!



  • Often you just need to install it. If it’s a really really old system then pick something that is minimal or designed to be lightweight. In the old times something like xubuntu would do the trick, but nowadays I don’t know what’s out there.

    You can also try some more “heavy” distros. I have a 2014 macbook air (4GB of RAM), with full gnome 3 and it works really really smoothly. If you have some money to spend, adding a small ssd, even with adapters, will make a whole lot of difference.

    If the drivers have been removed from the kernel, you’ll need to find a more “antique” distro too, or go deep into customization (I can tell you more if you want, do ask!)



  • So, HTTPS prevents people on the network from hijacking the connection or spying on what you do with the website or your passwords. If you have a VPN and no evil devices inside your network you are likely fine, and many things don’t need HTTPS-level of safety.

    That said, browsers make using HTTP almost unusable nowadays, because over the internet you really want your connection to not be hyjacked, and almost nobody self-hosts (sadly).

    I would reccommend to set up a reverse proxy to do https for all services you host, even if you only use it via lan/vpn. Adding a new service then becomes just hosting it over HTTP, and have your reverse proxy converting the connection to HTTPS.

    If you have a domain, you can request a wildcard cert from let’s encrypt and use that also for internal domains (they don’t need to be reachable from the public internet, but you might need to expose the ACME challenge file).

    If you want to learn a lot about certificates you can go the self-signed route, but you’ll have to install the certificate on every device you use to access your services (guests will get huge warnings when trying to access your pages, if you give them access to them). Once set up it’s more flexible though, and if you configure your dns server you can have whatever domain you want, even nonexistent or unavailable ones.



  • Omg it’s brutal and a very entertaining read.

    It starts with plain false advertisement:

    ParticleDB’s home page advertised that ParticleDB was “formally proven” with a TLA+ proof. However, during our collaboration no TLA+ specification was available

    It then goes into a detailed list of issues, such as:

    In ParticleDB, 0.0 was not equal to 0 (#63), but 0.90... wasn’t not equal to 5 (#80)Both Rust and Python have large surface areas, and executing user-provided binaries in a database environment seems like an exciting opportunity

    l…what? How? How can you even “accidentally” program something like this?

    Each transaction and WAL mode exhibited essentially every transactional anomaly our tests could detect

    Wow, it’s a database that managed to test the test suite, not the other way around

    Orkes has since published [a release history] which describes versions 1.1.4 through 1.1.18.11.rc-22 as “unsupported pre-release builds"

    Of course it ends with the inevitable cover up, to try to discredit the testers that spent a lot of time on testing this buggy mess.


  • Same for me, but that’s because I have a rule saying to the claude to not interact with git unless instructed, and I commit manually after reviewing every line of code changed. The bot runs in its own container, isolated from all credentials, so it can’t push commits and can’t access production. I also hate that the super long commit messages, so I just want to write it myself.

    Basically if you are surrenderinf your entire PC to AI, then you’ll not get those commit messages authored by claude.

    All my colleagues instead gave it access to the whole work pc including ssh keys, and the bot can do whatever the fuck it wants with the whole repo on github. It co-authors every single commit. Our master branch pushed straight to prod, and is not protected, and all our team is repo admin anyway. It’s complete insanity, bordering criminal negligence.






  • It’s actually possible to prevent impersonation:

    In theory all the generated keys are added to a public certificate transparancy log. Browsers should check that the key is there (but your http client library probably doesn’t - and this is why these compromised keys are a problem).

    Website owners should also check that there are no other certificates in there, and if they see an extra misissued certificate something went really wrong at a CA and that CA would probably lose all the credibility and trust.

    Also when you request a certificate the CA doesn’t get your private keys, so they can’t reuse your cert.

    Example: https://ctlogs.dev/search?q=Programming.dev