On May 11, 2011, at 10:51 AM, Sigbjorn Lie wrote:
On Wed, May 11, 2011 14:42, Stephen Gallagher wrote:
On Tue, 2011-05-10 at 23:42 +0200, Sigbjorn Lie wrote:
I would like to see the ipa client scripts and possibly the admin
in a nice Solaris package. This would make my job a lot easier as
we have a lot of customers
running Solaris. :)
For the server part I agree with you, keep it at RHEL.
SSSD @ Solaris / HP-UX / AIX ... well there isn't much (if any) of the
UNIX vendors selling their iron as client machines anymore. And I
see a considerable benefit in adding SSSD to servers, who will be
well connected to the network
Actually, SSSD is still valuable on server systems (and is used very
often in datacenters). The reason is that it can allow a server to
ride out an outage in the LDAP
and/or Kerberos server and still handle authentication and identity
requests from its cache.
We've expressed interest several times in working WITH other platforms
to help them port the SSSD, but we've received no real commitment to
assisting with it. We have a
lot on our plates already, so it is difficult for us to justify
spending time improving our
competitors' offerings :)
Also, SSSD has additional features with FreeIPA integration that
nss_ldap and pam_krb5 do not. Specifically, it has support for
managing access-control using
FreeIPA's host-based access control model. This is
a very valuable piece of the puzzle and should not be ignored.
I see you're having a valid point about the outage support. This
could be worked around using the
"High Availability Add-on" in RHEL, sharing an IP address between
your IPA servers, which you
would switch to the currently active IPA server.
Not only is there a question of high availability with regard to
lookups into ldap. But there is also a problem of scale and overhead.
nss_ldap and pam_ldap perform a lookup per iteration in many cases.
Consider for example. 4 data centers with 100 servers each, all tied
back to ldap for uid/gid mappings and pam_ldap for authentication and
If you have a task that logs into each of these 400 servers and
performs a 'sudo ls -la /home' for example,
your ldap servers are going to incur the cost of looking up each file
on each server, the cost of each authentication, and the cost of
performing several ldap lookups from the sudo binary.
SSSD is not only beneficial during periods of network inaccessibility,
but also crucial with regard to scale.
With regards to IPA's host-based access control: What about doing
access control through using
netgroups via the tcp wrappers?
You could still be configuring host based access control in IPA as
it's creating transparent
netgroups for the host groups.
Host based access control is currently a mess in the Linux Community.
There are currently a few ways to go about it.
^ This method implies that the changes in your central database must
eventually be pushed to flatfile configs on the end hosts.
While this works pretty well in small environments, it can fall apart
and have serious scale issues when dealing with hundreds or thousands
(Yes, even when using something like Satellite or Puppet)
Consider the case of Active Directory where you scratch your head and
go: "Gee, I'm SURE that i pushed that GPO, but for some reason, this
set of hosts didn't get the memo"
pam_ldap + pam_check_host_attr
^ This issue has a sheer drop off problem with scale. In this
approach, you need to fill the user objects with every host that the
user is permitted to login to.
When the number of users/administrators grow along with the number of
hosts you have, you get: n^users * n^hosts and the administrative
overhead becomes overwhelming.
These are all workarounds, I assume having the functionality
available trough the native sssd
would be of an advantage. But this way you would the mentioned extra
functionality of SSSD without
having to do the work of supporting your competitors operating
There have been _some_ discussions surrounding a pam module that could
be used as a very base level of hbac support since there are a lot of
pre-required dependancies for sssd.
The advantage would be theoretical portability, and the loss would be
I have personally written such a pam plugin prototype in python, and
it functions just fine in linux installations. the c code that calls
the python script is not compatible with open_pam,
so there is still work to be done to support the BSD / MAC solutions,
but I believe its just a matter of some syntax changes...
I hope this information helps clarify these points.