[Libguestfs] [PATCH nbdkit v2 1/3] server: Add new APIs for reading the client’s SO_PEERCRED.

[Daniel P. Berrangé]

On Mon, Oct 05, 2020 at 08:21:50AM -0500, Eric Blake wrote:
> On 10/3/20 1:50 PM, Richard W.M. Jones wrote:
> > New nbdkit_peer_pid, nbdkit_peer_uid and nbdkit_peer_gid calls can be
> > used on Linux (only) to read the peer PID, UID and GID from clients
> > connected over a Unix domain socket.  This can be used in the
> > preconnect phase to add additional filtering.
> > 
> > One use for this is to add an extra layer of authentication for local
> > connections.  A subsequent commit will enhance the now misnamed
> > nbdkit-ip-filter to allow filtering on these extra fields.
> > 
> > It appears as if it would be possible to implement this for FreeBSD
> > too (see comment in code).
> > ---
> >  docs/nbdkit-plugin.pod  |  47 +++++++++++++++--
> >  include/nbdkit-common.h |   3 ++
> >  server/nbdkit.syms      |   3 ++
> >  server/public.c         | 108 ++++++++++++++++++++++++++++++++++++++++
> >  4 files changed, 156 insertions(+), 5 deletions(-)
> > 
> 
> > +=head2 C<nbdkit_peer_pid>
> > +
> > +(nbdkit E<ge> 1.24)
> > +
> > + int nbdkit_peer_pid (void);
> > +
> > +Return the peer process ID.  This is only available when the client
> > +connected over a Unix domain socket, and only works for Linux.
> > +
> > +On success this returns the peer process ID.  On error,
> > +C<nbdkit_error> is called and this call returns C<-1>.
> 
> Is int always going to be sufficient?  Or are there platforms with
> 64-bit pid_t?  Mingw is an interesting beast; I've seen conflicting
> stories on whether 64-bit windows has 32- or 64-bit pids (the spawn APIs
> manage 64-bit handles, but other windows APIs return 32-bit int), so
> 64-bit pid_t on mingw does seem to be a real concern.

IIUC,  POSIX says  pid_t is a signed integer, but doesn't specify the
size.  Thus libvirt exposed  pid_t as  "signed long long" in our APIs
to be futureproof.

> > +
> > +=head2 C<nbdkit_peer_uid>
> > +
> > +(nbdkit E<ge> 1.24)
> > +
> > + int nbdkit_peer_uid (void);
> > +
> > +Return the peer user ID.  This is only available when the client
> > +connected over a Unix domain socket, and only works for Linux.
> > +
> > +On success this returns the user ID.  On error, C<nbdkit_error> is
> > +called and this call returns C<-1>.
> > +
> > +=head2 C<nbdkit_peer_gid>
> > +
> > +(nbdkit E<ge> 1.24)
> > +
> > + int nbdkit_peer_gid (void);
> 
> int for these two is probably fine.

IIUC, gid_t/uid_t don't have their signed-ness specified by POSIX,
nor size, but you're required to cast negative values eg

    gid_t foo = (gid_t)-1;

based on this, libvirt chose to expose them as "unsigned long long" to
maximise future proofing.


Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|