• aard@kyu.de
      link
      fedilink
      arrow-up
      19
      ·
      1 year ago

      A surprising amount of services (including Azure last I tried) can only handle RSA keys, so after trying ecdsa only for a while I ended up adding a RSA key again.

      With that said - it’s 2023, in almost all cases you should have your keys in a hardware module nowadays, in which case you’d use a different command for keygeneration.

      • FOSS Is Fun@lemmy.ml
        link
        fedilink
        English
        arrow-up
        10
        ·
        edit-2
        1 year ago

        Actually it is the same story with TLS 1.3 and TLS 1.2. A bunch of sites still doesn’t support TLS 1.3 (e. g. arstechnica.com, startpage.com) and some of them only support TLS 1.2 with RSA (e. g. startpage.com).

        You can try this yourself in Firefox by disabling ciphers (search for security.ssl3 in about:config) or by setting the minimum TLS version to 1.3 (security.tls.version.min = 4 in about:config).

        • deepdive@lemmy.world
          link
          fedilink
          English
          arrow-up
          3
          ·
          1 year ago

          Strange enough TLS 1.3 still doesn’t support signed ed25519 certificates :| P‐256, NIST P‐384 or NIST P‐521 curves are known to be “backdoored” or having deliberately chosen mathematical weakness. I’m not an expert and just a noob security/selfhoster enthusiast but I don’t want to depend on curves made by NSA or other spy agencies !

          I also wondering if the EU isn’t going to implement something similar with all their new spying laws currently discussed…

        • aard@kyu.de
          link
          fedilink
          arrow-up
          5
          ·
          1 year ago

          Easiest and most affordable is probably a security key like the Nitrokey or the https://www.yubico.com/. I personally don’t like the company behind yubikey much, but if you want something small you can always leave in the device that’s pretty much your only option.

          For “cheaper, but a bit more effort” would be just getting a smartcard blank, a card reader (if you’re not lucky enough to have a notebook or computer with one built in), and then either write your own applet, or use one of the available opensource ones, and upload it to the card. A variant of that would be the Fidesmo card, where you get a card and their applet.

          Or you just use the TPM you may have in your system - though you’ll need to be careful with that: Typically one reason for using a hardware token is to make sure keys can’t get extracted, while TPMs often do allow key extraction. Software to make that work would be opencryptoki.

          Generally you’d use PKCS#11 to have the various components talk to each other. On your average Linux pretty much everything but GnuPG place nice. with PKCS#11. Typically you end up with pcscd to interface with the smartcard (the above USB tokens are technically also just USB smartcards), OpenSC as layer to provide PKCS#11 on top, and software (like OpenSSH) then talks to that.

          All of that should be available as packages in any Linux distribution nowadays - and typically will also provide p11-kit configured to use a proxy library to make multiple token sources easily available, and avoid blocking on concurrent access.

          ssh-add supports adding keys from pkcs#11 providers to the SSH agent (search pkcs11 in ssh-add manpage), with some distribution (like RedHat) also carrying patches allowing you to only select individual tokens for adding.

          If you’re also using GnuPG it gets more complicated - you pretty much have two options: Stick with PKCS#11, in which case you’d replace GPGs own smartcard agent with gnupg-pkcs11-scd, or you use GPGs own card implementation, in which case you can forget pretty much everything I wrote above, and just follow the security key manual for setting up a GPG card, enable SSH agent support in the GPG agent, and just use that for SSH authentication.

    • 𝒍𝒆𝒎𝒂𝒏𝒏@lemmy.one
      link
      fedilink
      arrow-up
      11
      ·
      1 year ago

      I delete them from the ssh config folder after installation, along with the DSA and ECDSA keys. No ed25519? No auth.

      Also prevents a handful of bots from attempting SSH login into your cloud infra, a lot of them don’t support ed25519 kex

    • 018118055@sopuli.xyz
      link
      fedilink
      arrow-up
      5
      ·
      1 year ago

      I had to create one this year after discovering that connectbot (ssh client on Android) didn’t support agent forwarding otherwise.

      • 𝒍𝒆𝒎𝒂𝒏𝒏@lemmy.one
        link
        fedilink
        arrow-up
        6
        ·
        1 year ago

        Probably a good idea to look for a different client, call me tinfoil but I wouldn’t want to touch a very old mechanism that is supported/pushed by a very recognisable 3 letter agency

        • LiveLM@lemmy.zip
          link
          fedilink
          English
          arrow-up
          6
          ·
          1 year ago

          I’ve just started using SSH inside of Termux, got tired of all the weird pitfalls SSH Clients for Android usually have

        • 018118055@sopuli.xyz
          link
          fedilink
          arrow-up
          2
          ·
          1 year ago

          Probably. It’s in f-droid but increasingly looking not quite unmaintained, but not developed actively enough.

  • Pantherina@feddit.de
    link
    fedilink
    arrow-up
    9
    arrow-down
    1
    ·
    1 year ago

    Nice no ChatGPT anymore to remember how that damn Algorithm is spelled.

    Why not just call it RSB ? People, really!

  • maniel@beehaw.org
    link
    fedilink
    arrow-up
    7
    arrow-down
    1
    ·
    1 year ago

    Isn’t elliptic curves cryptography sensitive to quantum computers attack? Shor’s algorithm etc

    • duncesplayed@lemmy.one
      link
      fedilink
      English
      arrow-up
      3
      ·
      1 year ago

      Yes, it is. ed25519 depends upon discrete log for its security, which Shor’s algorithm can (theoretically, of course, not like it’s ever been done) efficiently solve.

      The post-quantum algorithms are in active research right now. I don’t blame anyone for avoiding those at least until we’ve quantum computers big enough to solve baby toy elliptic curves.

    • LaggyKar@programming.dev
      link
      fedilink
      arrow-up
      3
      ·
      edit-2
      1 year ago

      Yes, though OpenSSH has already switched to a quantum resistant algorithm for key exchange (Streamlined NTRU Prime, combined with x25519 in case SNTRUPrime turns out to be weak), and that’s the stuff that needs to be switched as soon as possible to preserve forward secrecy. Authentication keys are less urgent.