On Mon, Aug 08, 2016 at 10:49:27AM +0200, Jan Cholasta wrote:
> On 8.8.2016 09:06, Fraser Tweedale wrote:
> > On Mon, Aug 08, 2016 at 08:54:05AM +0200, Jan Cholasta wrote:
> > > Hi,
> > > 
> > > On 8.8.2016 06:34, Fraser Tweedale wrote:
> > > > Please review the attached patch with adds --certificate-out and
> > > > --certificate-chain-out options to `ca-show' command.
> > > > 
> > > > Note that --certificate-chain-out currently writes a bogus file due
> > > > to a bug in Dogtag that will be fixed in this week's build.
> > > > 
> > > > https://fedorahosted.org/freeipa/ticket/6178
> > > 
> > > 1) The client-side *-out options should be defined on the client side, not
> > > on the server side.
> > > 
> > Will option defined on client side be propagated to, and observable
> > in the ipaserver plugin?  The ipaserver plugin needs to observe that
> > *-out has been requested and executes additional command(s) on that
> > basis.
> 
> Is there a reason not to *always* return the certs?
> 
We hit Dogtag to retrieve them.

> > 
> > > 
> > > 2) I don't think there should be additional information included in 
> > > summary
> > > (and it definitely should not be multi-line). I would rather inform the 
> > > user
> > > via an error message when unable to write the files.
> > > 
> > I was just following the pattern of other commands that write certs,
> > profile config, etc.  Apart from consistency with other commands I
> > agree that there is no need to have it.  So I will remove it.
> > 
> > > If you think there is an actual value in informing the user about
> > > successfully writing the files, please use ipalib.messages for the job.
> > > 
> > > 
> > > 3) IMO a better format for the certificate chain than PKCS#7 would be
> > > concatenated PEM, as that's the most commonly used format in IPA (in
> > > installers, there are no cert chains in API commands ATM).
> > > 
> > Sure, but the main use case isn't IPA.  Other apps require PKCS #7
> > or concatenated PEMs, but sometimes they must be concatenated
> > forward, and othertimes backwards.  There is no one size fits all.
> 
> True, which is exactly why I think we should at least be self-consistent and
> use concatenated PEM (and multi-value DER over the wire).
> 
Dogtag returns a PKCS7 (either DER or PEM, according to HTTP Accept
header).

If we want list-of-PEMs between server and client we have to convert
on the server.  Do we have a good way of doing this without exec'ing
`openssl pkcs7' on the server?  Is it acceptable to exec 'openssl'
to do the conversion on the server?  python-nss does not have PKCS7
functions and I am not keen on adding a pyasn1 PKCS7 parser just for
the sake of pushing bits as list-of-PEMs.

FWIW, man pages and code suggest that PKCS #7 is accepted in
installer, etc.

> > We can add an option to control the format later, but for now,
> > Dogtag returns a PKCS #7 (PEM or DER) so let's go with that.  Worst
> > case is an admin has to invoke `openssl pkcs7' and concat the certs
> > themselves.
> 
> AFAIK none of NSS, OpenSSL or p11-kit can use PKCS#7 cert chains directly,
> so I'm afraid the worst case would happen virtually always.
> 
If you're OK with invoking OpenSSL on the client to convert PKCS #7
to list-of-PEMs (similar to what is done in
ipapython.certdb.NSSDatabase) then we can have the client perform
the conversion.

> > 
> > > 
> > > 4) Over the wire, the certs should be DER-formatted, as that's the most
> > > common wire format in other API commands.
> > > 
> > OK.
> > 
> > > 
> > > 5) What is the benefit in having the CA cert and the rest of the chain
> > > separate? For end-entity certs it makes sense to separate the cert from 
> > > the
> > > CA chain, but for CA certs, you usually want the full chain, no?
> > > 
> > If you want to anchor trust directly at a subca (e.g. restrict VPN
> > login to certs issued by VPN sub-CA) then you often just want the
> > cert.  The chain option does subsume it, at cost of more work for
> > administrators with this use case.  I think it makes sense to keep
> > both options.
> 
> Does it? From what you described above, you either want just the sub-CA
> cert, or the full chain including the sub-CA cert, in which case it might
> make more sense to have a single --out option and a --chain flag.
> 
How about --certificate-out which defaults to single cert, but does
chain (as list-of-PEMs) when --chain flag given.

Per https://fedorahosted.org/freeipa/ticket/5166 let's not add more
`--out' options.

Thanks,
Fraser

-- 
Manage your subscription for the Freeipa-devel mailing list:
https://www.redhat.com/mailman/listinfo/freeipa-devel
Contribute to FreeIPA: http://www.freeipa.org/page/Contribute/Code

Reply via email to