Expose Go library functions (please review the API!) about age

[schollz]

I made this for now: https://github.com/schollz/age/ but I'd rather not have to merge upstream to keep it up to date, so it would be nice if this repo is the library.

@FiloSottile, do you plan on having this repo be used as a library?

[FiloSottile]

It's out! There is now a Go API for age!

https://pkg.go.dev/filippo.io/[email protected]
https://pkg.go.dev/filippo.io/[email protected]/agessh
https://pkg.go.dev/filippo.io/[email protected]/armor

Still in beta so we can fix it if we realize I made some horrible mistake (let me know!) but otherwise I expect it to be pretty stable (and probably to grow an EncryptWithOptions later).

[FiloSottile]

Hey folks, sorry for taking so long, I am indeed preparing to expose an API and would love some feedback on the current plan which is to expose the age, agessh, and armor packages!

(Sorry for not just taking #120, but you won't be surprised to learn I have strong opinions on APIs :))

[schollz]

@ovalseven8 That's great! I should have been more specific - I'm actually wondering if this implementation (filippo.io/age) is meant to serve as the Go library.

[sashabaranov]

Exposing age as a library would very much benefit building mobile apps with good encryption through gomobile!

[stchris]

FYI https://twitter.com/FiloSottile/status/1262279790377291776?s=20

[dominikschulz]

Thanks for the initial API draft. It's a nice and clean API that seems to provide anything that gopass needs.

Some questions:

• What about EncryptedSSHIdentity for encrypted SSH private keys? This would need to be handled by the caller?
• Any thoughts about Agent support, e.g. yubikey-agent?
• What about the hard coded recipient limit (20)? Any chance of raising / exposing this?

[ovalseven8]

According to the age documentation it's intended to be a library:

This is a design for a simple file encryption CLI tool, Go library, and format.

[xt0fer]

perhaps we're just getting caught by the 1.4/1.5 design stricture? If internal was renamed to library or impl would that change the intention of the project?

[schollz]

@xt0fer exactly, any name change would allow it to be used as a library. However it was probably purposely done to be internal, just wanted to double check

[abarisani]

I'd welcome this for attempting to compile this under TamaGo (https://github.com/inversepath/tamago).

[richo]

Is there any word on this? I'd love to avoid shelling out to age if I can avoid it.

[cyb3rz3us]

Apologies in advance for not fully understanding the ask here but since 'age' is using the Go crypto library, why would someone then want to have 'age' be a crypto library for use in a Go program?

I guess I would think one would just use the same libs that 'age' uses for their Go program...please help me see what I am missing...

[hairyhenderson]

@cyb3rz3us it's not "just" using the Go crypto package... it's also implementing a specific format.

The simplest (and probably most naïve) solution to this issue would be to expose all the packages that are currently in https://github.com/FiloSottile/age/tree/master/internal; but I don't think @FiloSottile really wants to do that 😉...

[dncohen]

Count me in as wanting this all this age goodness as an importable package.

I think internal/age is the only one that needs moving out of the internal/ directory. That's what PR #119 does.

[FiloSottile]

Thank you for the feedback, all very useful!

• What about EncryptedSSHIdentity for encrypted SSH private keys? This would need to be handled by the caller?

It's specifically for decryption deferral, if you are sure you need to decrypt it you can just use the x/crypto/ssh functions to do so. But I guess it is general enough to be useful, so I moved it to agessh, thanks for the suggestion.

• Any thoughts about Agent support, e.g. yubikey-agent?

@str4d and I worked on a plugin support which should make it possible to implement an agent as a plugin.

• What about the hard coded recipient limit (20)? Any chance of raising / exposing this?

Hmm, yeah, this is a great example of something that would be nice to configure. The current API allows neither returning information about the file nor passing options.

[lucor]

Thanks for the tool and the API nice and clean!

Just a question, any thoughts about to expose in some way the generation of a X25519Identity using a seed (i.e. exposing the newX25519IdentityFromScalar) ?

[FiloSottile]

Just a question, any thoughts about to expose in some way the generation of a X25519Identity using a seed (i.e. exposing the newX25519IdentityFromScalar) ?

It used to be exposed but I realized it was probably more of a footgun than useful. What's your use case?

[lucor]

The use case is to encrypt a file using an existing private key derived from a user's password with scrypt.

[FiloSottile]

I see. Can't you store a fresh private key encrypted with the user's password? That would also let you rotate the password without changing the public key.

[lucor]

Apologies, I'd have shared more details about the use case.
I'm experimenting with the idea of password manager that allow to generate services' password using a pure function and (optionally) fallback to custom ones where this is not applicable (i.e. imported passwords). Neither the user's password nor the private key are stored, but the last one is used to derive passwords and store some settings along with the custom passwords into an encrypted file.
I know it is an edge case and probably itself could not justify the addition of the new public method :-)

[FiloSottile]

Neither the user's password nor the private key are stored, but the last one is used to derive passwords and store some settings along with the custom passwords into an encrypted file.

That sounds like it doesn't need a public key, so maybe you can use the ScryptIdentity / ScryptRecipient?

[lucor]

Indeed, this was my first try too. In this case I'm able to encrypt but still need to generate another private key with scrypt to generate the services' password :(. But this is something related to my project's UX not to age.

Right now I'm using with success agessh as below:

// error handling omitted for brevity
key, _ := scrypt.Key([]byte(password), salt, 1<<16, 8, 1, 32)
ed25519Key := ed25519.NewKeyFromSeed(key)
ageIdentity, _ := agessh.NewEd25519Identity(ed25519Key)
servicePassword, _ := fn(key, serviceInfo)

[FiloSottile]

still need to generate another private key with scrypt to generate the services' password

I don't really understand how this is not compatible with ScryptIdentity / ScryptRecipient, you can use the same passphrase, as the salts will be different.

(This is probably a better fit for the mailing list than the issue tracker.)

[lucor]

Agreed. Moved the discussion to the mailing list. Thanks @FiloSottile

[FiloSottile]

The API has now been around for a bit, closing this issue!

[littlecxm]

age/scrypt.go

Line 38 in 2e09054

func NewScryptRecipient(password string) (*ScryptRecipient, error) {

use []byte to replace exists arg password string should be better?