cross-posted from: https://lemmy.ml/post/1840134

This is a changeset adding encryption to btrfs. It is not complete; it
does not support inline data or verity or authenticated encryption. It
is primarily intended as a proof that the fscrypt extent encryption
changeset it builds on work. 

As per the design doc refined in the fall of last year [1], btrfs
encryption has several steps: first, adding extent encryption to fscrypt
and then btrfs; second, adding authenticated encryption support to the
block layer, fscrypt, and then btrfs; and later adding potentially the
ability to change the key used by a directory (either for all data or
just newly written data) and/or allowing use of inline extents and
verity items in combination with encryption and/or enabling send/receive
of encrypted volumes. As such, this change is only the first step and is
unsafe.

This change does not pass a couple of encryption xfstests, because of
different properties of extent encryption. It hasn't been tested with
direct IO or RAID. Because currently extent encryption always uses inline
encryption (i.e. IO-block-only) for data encryption, it does not support
encryption of inline extents; similarly, since btrfs stores verity items
in the tree instead of in inline encryptable blocks on disk as other
filesystems do, btrfs cannot currently encrypt verity items. Finally,
this is insecure; the checksums are calculated on the unencrypted data
and stored unencrypted, which is a potential information leak. (This
will be addressed by authenticated encryption).

This changeset is built on two prior changesets to fscrypt: [2] and [3]
and should have no effect on unencrypted usage.

[1] https://docs.google.com/document/d/1janjxewlewtVPqctkWOjSa7OhCgB8Gdx7iDaCDQQNZA/edit?usp=sharing
[2]
https://lore.kernel.org/linux-fscrypt/cover.1687988119.git.sweettea-kernel@dorminy.me/
[3]
https://lore.kernel.org/linux-fscrypt/cover.1687988246.git.sweettea-kernel@dorminy.me
  • ozymandias117@lemmy.world
    link
    fedilink
    arrow-up
    8
    ·
    1 year ago

    Is there a major performance benefit to encryption using LUKS over this?

    Or is this more useful for like, encrypting a single directory of a filesystem?

    I’ve never quite understood the impetus for each filesystem handling encryption rather than doing it at the block layer

    • Atemu@lemmy.mlOP
      link
      fedilink
      arrow-up
      4
      ·
      1 year ago

      Possibilities at the block layer are generally quite limited since it only has limited means to work with. It’s very low-level. For example, it is not possible to do authentication in LUKS. An attacker can’t read the data but they can modify it; undetected.
      You need to stack another layer on top and I’m not sure that’s even a thing.

      The patch mentions that authenticated hashes aren’t supported yet either but with effectively limitless metadata to work with, it’s at least possible to do.

      Per-directory/subvolume encryption is also a useful feature. You could encrypt the root fs which generally does not contain sensitive information using a key in TPM but then require a password to unlock the user’s home. That’s basically how it works in Android and it builds on top of fscrypt.

      • ozymandias117@lemmy.world
        link
        fedilink
        arrow-up
        1
        ·
        edit-2
        1 year ago

        An attacker can’t read the data, but they can modify it; undetected

        Wouldn’t it then decrypt to gibberish data unless they already had the encryption keys?

        If it decrypts incorrectly, shouldn’t BTRFS checksumming then return an I/O error to user space as well?

        • Atemu@lemmy.mlOP
          link
          fedilink
          English
          arrow-up
          1
          ·
          edit-2
          1 year ago

          Note that this is of course a very theoretical attack vector.

          Wouldn’t it then decrypt to gibberish data unless they already had the encryption keys?

          Depends. I don’t know the situation of LUKS and its commonly used ciphers in particulare but even some commonly used ciphers are vulnerable to things like bitflip attacks.

          This is usually “fixed” by authenticating them but that’s not easily possible at the block layer.

          If it decrypts incorrectly, shouldn’t BTRFS checksumming then return an I/O error to user space as well?

          Note that btrfs usually uses CRCs, not cryptographic checksums. They’re designed to catch “naturally” occuring corruption, not crafted corruption. Naturally, it’d still be extremely hard to break them when working with encrypted data but it’s a “uh, sounds pretty hard” situtation, not a “we can prove you’d need billions of years to do it” one.

          You can use cryptographic checksums but note again here that the attacker could be able to modify the checksum aswell.

          I don’t know how feasible this really is a but a possible attack could be to tell btrfs that the extent you modified is a nochecksum extent (you can turn off checksums in btrfs) which would make btrfs simply not check the checksum.

          Actual authentication fixes all of that.