Re: Setuid programs

["Gábor Boskovits" <boskovits@gmail.com>]

Hello,

Christopher Lemmer Webber <cwebber@dustycloud.org> ezt írta (időpont:
2020. szept. 10., Cs, 0:52):
>
> Christopher Lemmer Webber writes:
>
> > Gábor Boskovits writes:
> >
> >> Hello,
> >>
> >> Christopher Lemmer Webber <cwebber@dustycloud.org> ezt írta (időpont:
> >> 2020. szept. 9., Sze, 21:00):
> >>>
> >>> Maxim Cournoyer writes:
> >>>
> >>> > Hello Gabor!
> >>> >
> >>> > Gábor Boskovits <boskovits@gmail.com> writes:
> >>> >
> >>> >> Hello guix,
> >>> >>
> >>> >> I would like to propose an extension to how setuid programs are
> >>> >> currently handled. The last time I checked it could only do setuid and
> >>> >> setgid root. Some services, such as postfix need a more fine grained
> >>> >> setuid setup. I would propose a record type, such as:
> >>> >> (setuid
> >>> >> (program setuid-program)
> >>> >> (setuid setuid-setuid)
> >>> >> (setgid setuid-setgid)
> >>> >> (user setuid-user)
> >>> >> (group setuid-group))
> >>> >>
> >>> >> So that there is more fine grained control.
> >>> >>
> >>> >> I would also propose to move this to the services framework, so that
> >>> >> services could extend this field on demand.
> >>> >>
> >>> >> Wdyt?
> >>> >
> >>> > This sounds great!  I also encountered such limitation and tried to
> >>> > fixing it in https://issues.guix.info/41763, with some success (and an
> >>> > unresolved limitation pointed by Chriistopher) but I agree that using a
> >>> > record makes more sense and is more future proof.
> >>> >
> >>> > Maxim
> >>>
> >>> I'm eager to use Postfix on Guix (maybe it's me, but I just can't make
> >>> sense of the weird DSL that opensmtpd uses) so I guess if that's what's
> >>> necessary it already makes it a good idea.
> >>>
> >>> However I don't fully understand the syntax of what you proposed.  Let's
> >>> see if I can guess with a fake entry
> >>>
> >>> #~(setuid
> >>>    ;; The program to run, from the shady package
> >>>    (program (string-append #$shady "/bin/scaryfoo")
> >>>    ;; Would this be a boolean?  If so should it be `setuid?`
> >> yes, this should be a bool, studi? looks good to me.
> >>>    (setuid setuid-setuid)
> >>>    ;; Likewise?
> >>>    (setgid setuid-setgid)
> >> yes, the same thing applies here.
> >>>    ;; Presumably the use we want to set this to
> >>>    (user setuid-user)
> >> yes, this should just be the uid of the owner
> >>>    ;; Presumably the group we want to se this to
> >> this should be the gid.
> >>>    (group setuid-group))
> >>>
> >>> ... right?
> >>>
> >>> I guess this could be done in a backwards compatible way;
> >>> %setuid-programs could either evaluate to strings or records, so the
> >>> "simpler" version can remain an option?
> >> Yes, it can be done this way. Actually I had a bit more general
> >> solution in mind,
> >> I feel there should be service to install a file from a store to a
> >> given place, and with all the access control available,
> >> like acl-s, if supported. And then provide the whole setuid thing as a
> >> backwards compatibility layer, somehow like you described.
> >> For now I guess creating this record type and implementing the
> >> extended setuid functionality would be a good first step.
> >
> > A service seems like a really good idea to me in that it feels the most
> > composable with how Guix currently approaches things.
>
> I feel like this one needs more "Guix maintainer" overview.
I agree, this would be nice.

  The current
> setuid-programs could be kept as legacy behavior that installs an
> additional service.  Thoughts?

I believe it should be kept and install an additional service.

I have two reasons for that: backwards compatibility is really
important, so we should not break it, and I believe this would not be
hard to do.
On the other hand it would be nice to have a more integrated backend,
and move as many things into the services infrastructure as practical,
and I think this is a good candidate for that. Wdyt?

Best regards,
g_bor
-- 
OpenPGP Key Fingerprint: 7988:3B9F:7D6A:4DBF:3719:0367:2506:A96C:CF63:0B21