Re: [PATCH v2 02/11] qcrypto-luks: extend the create options for upcoming encryption key management

[Maxim Levitsky]

On Thu, 2019-10-10 at 15:44 +0200, Kevin Wolf wrote:
> Am 13.09.2019 um 00:30 hat Maxim Levitsky geschrieben:
> > Now you can specify which slot to put the encryption key to
> > Plus add 'active' option which will let  user erase the key secret
> > instead of adding it.
> > Check that active=true it when creating.
> > 
> > Signed-off-by: Maxim Levitsky <address@hidden>
> > diff --git a/qapi/crypto.json b/qapi/crypto.json
> > index b2a4cff683..9b83a70634 100644
> > --- a/qapi/crypto.json
> > +++ b/qapi/crypto.json
> > @@ -190,6 +190,20 @@
> >  #                  Currently defaults to 'sha256'
> >  # @hash-alg: the master key hash algorithm
> >  #            Currently defaults to 'sha256'
> > +#
> > +# @active: Should the new secret be added (true) or erased (false)
> > +#          (amend only, since 4.2)
> > +#
> > +# @slot: The slot in which to put/erase the secret
> > +#        if not given, will select first free slot for secret addtion
> > +#        and erase all matching keyslots for erase. except last one
> > +#        (optional, since 4.2)
> > +#
> > +# @unlock-secret: The secret to use to unlock the image
> > +#        If not given, will use the secret that was used
> > +#        when opening the image.
> > +#        (optional, for amend only, since 4.2)
> > +#
> >  # @iter-time: number of milliseconds to spend in
> >  #             PBKDF passphrase processing. Currently defaults
> >  #             to 2000. (since 2.8)
> 
> This approach doesn't look right to me. BlockdevCreateOptions should
> describe the state of the image after the operation. You're describing
> an update instead (and in a way that doesn't allow you to change
> everything that you may want to change, so that you need to call the
> operation multiple times).
> 
> I imagined the syntax of a blockdev-amend QMP command similar to
> x-blockdev-reopen: Describe the full set of options that you want to
> have in effect after the operation; if you don't want to change some
> option, you just specify it again with its old value.

This approach is a compromise trying to create more or less usable interface.
In particular we (I and Daniel) wanted the following to work:

1. ability to add a new password to an empty keyslot and then remove the
old password. This is probably the most common operation and it won't
require the caller to know anything about the keyslots.

2. Allow the user to not know the passwords of some keyslots.
For example if I want to add a new keyslot, I might  not know
some of the other keyslots. Specifying all the active keyslots,
on each amend would force the user to know all the passwords
(you can't 'extract' a password by reading a keyslot, since only
hash of it is stored there for the security reasons)


Thus the amend interface either allows you to add a keyslot (either a specific 
one,
or first free one), and to remove a keyslot (again, either a specific one, or
one that matches given password).




> 
> Specifically for luks, this probably means that you have a @slots, which
> is a list that contains at least the secret for each slot, or JSON null
> for a slot that should be left empty.
> 
> With the same approach, you don't have to make 'size' optional in later
> patches, you can just require that the current size is re-specified. And
> later, blockdev-amend could actually allow changing the size of images
> if you provide a different value.

This can be done IMHO.

> 
> Kevin


Best regards,   
        Maxim Levitsky