Re: [Qemu-devel] [PATCH 11/34] qmp: query-block: add 'valid_encryption_key' field

[Luiz Capitulino <lcapitulino@redhat.com>]

On Fri, 10 Aug 2012 18:35:26 +0200
Markus Armbruster <armbru@redhat.com> wrote:

> Luiz Capitulino <lcapitulino@redhat.com> writes:
> 
> > On Fri, 10 Aug 2012 09:56:11 +0200
> > Markus Armbruster <armbru@redhat.com> wrote:
> >
> >> Revisited this one on review of v2, replying here for context.
> >> 
> >> Luiz Capitulino <lcapitulino@redhat.com> writes:
> >> 
> >> > On Thu, 02 Aug 2012 13:35:54 +0200
> >> > Markus Armbruster <armbru@redhat.com> wrote:
> >> >
> >> >> Luiz Capitulino <lcapitulino@redhat.com> writes:
> >> >> 
> >> >> > Signed-off-by: Luiz Capitulino <lcapitulino@redhat.com>
> >> >> > ---
> >> >> >  block.c          | 1 +
> >> >> >  qapi-schema.json | 7 +++++--
> >> >> >  2 files changed, 6 insertions(+), 2 deletions(-)
> >> >> >
> >> >> > diff --git a/block.c b/block.c
> >> >> > index b38940b..9c113b8 100644
> >> >> > --- a/block.c
> >> >> > +++ b/block.c
> >> >> > @@ -2445,6 +2445,7 @@ BlockInfoList *qmp_query_block(Error **errp)
> >> >> >              info->value->inserted->ro = bs->read_only;
> >> >> >              info->value->inserted->drv = g_strdup(bs->drv->format_name);
> >> >> >              info->value->inserted->encrypted = bs->encrypted;
> >> >> > +            info->value->inserted->valid_encryption_key = bs->valid_key;
> >> >> >              if (bs->backing_file[0]) {
> >> >> >                  info->value->inserted->has_backing_file = true;
> >> >> >                  info->value->inserted->backing_file = g_strdup(bs->backing_file);
> >> >> > diff --git a/qapi-schema.json b/qapi-schema.json
> >> >> > index bc55ed2..1b2d7f5 100644
> >> >> > --- a/qapi-schema.json
> >> >> > +++ b/qapi-schema.json
> >> >> > @@ -400,6 +400,8 @@
> >> >> >  #
> >> >> >  # @encrypted: true if the backing device is encrypted
> >> >> >  #
> >> >> > +# @valid_encryption_key: true if a valid encryption key has been set
> >> >> > +#
> >> >> >  # @bps: total throughput limit in bytes per second is specified
> >> >> >  #
> >> >> >  # @bps_rd: read throughput limit in bytes per second is specified
> >> >> > @@ -419,8 +421,9 @@
> >> >> >  { 'type': 'BlockDeviceInfo',
> >> >> >    'data': { 'file': 'str', 'ro': 'bool', 'drv': 'str',
> >> >> >              '*backing_file': 'str', 'encrypted': 'bool',
> >> >> > -            'bps': 'int', 'bps_rd': 'int', 'bps_wr': 'int',
> >> >> > -            'iops': 'int', 'iops_rd': 'int', 'iops_wr': 'int'} }
> >> >> > +            'valid_encryption_key': 'bool', 'bps': 'int',
> >> >> > +            'bps_rd': 'int', 'bps_wr': 'int', 'iops': 'int',
> >> >> > +            'iops_rd': 'int', 'iops_wr': 'int'} }
> >> >> >  
> >> >> >  ##
> >> >> >  # @BlockDeviceIoStatus:
> >> >> 
> >> >> BlockDeviceInfo is API, isn't it?
> >> >
> >> > Yes.
> >> >
> >> >> Note that bs->valid_key currently implies bs->encrypted.  bs->valid_key
> >> >> && !bs->encrypted is impossible.  Should we make valid_encryption_key
> >> >> only available when encrypted?
> >> >
> >> > I don't think so. It's a bool, so it's ok for it to be false when
> >> > encrypted is false.
> >> 
> >> What bothers me is encrypted=false, valid_encryption_key=true.
> >
> > Disappearing keys is worse, IMHO (assuming that that situation is impossible
> > in practice, of course).
> 
> It's fundamentally three states: unencrypted, encrypted-no-key,
> encrypted-got-key.  I'm fine with mapping these onto two bools, it's how
> the block layer does it.  You may want to consider a single enumeration
> instead.

That's arguable. But I like the bools slightly better because they allow
clients to do a true/false check vs. having to check against an enum value.

Again, that's arguable.

> >> >> valid_encryption_key is a bit long for my taste.  Yours may be
> >> >> different.
> >> >
> >> > We should choose more descriptive and self-documenting names for the
> >> > protocol. Besides, I can't think of anything shorter that won't get
> >> > cryptic.
> >> >
> >> > Suggestions are always welcome though :)
> >> 
> >> valid_encryption_key sounds like the value is the valid key.
> >
> > That's exactly what it is.
> 
> Err, isn't the value bool?  The key value is a string...

Ah, sorry, I read "sounds like true means the key is valid even for an
invalid key". I've renamed it to encryption_key_missing, should be better
(although I could also do encryption_key_is_missing).

> >> got_crypt_key?  Also avoids "valid".  Good, because current encrypted
> >> formats don't actually validate the key; they happily accept any key.
> >
> > That's a block layer bug, not QMP's.
> >
> > QMP clients are going to be misguided by valid_encryption_key the same way
> > they are with the block_passwd command or how we suffer from it internally
> > when calling bdrv_set_key() (which also manifests itself in HMP).
> >
> > Fixing the bug where it is will automatically fix all its instances.
> 
> It's not fixable for existing image formats, and thus existing images.

Why not? I'd expect that changing AES_set_decrypt_key() to fail for an
invalid key wouldn't affect images, am I wrong?

> You could even call it a feature that makes it (marginally) harder to
> brute-force keys (I don't buy that argument myself).

I don't, either.