Man I just can't seem to reply to this list correctly.  I hope the other
two didn't actually go anywhere.


I do not mean to store system environment variables but MY environment

dn: uid=alan,dc=example,dc=com
objectclass: top
objectclass: posixAccount
objectclass: inetOrgPerson
objectclass: posixAccountEnv
uid: alan
cn: Alan Evans
posixEnv: EDITOR=vim
# or
posixEnv;EDITOR: vim

In my opinion while these do not necessarily meet the definition you just
described they are distinctly MY preferences and I would argue they belong
with my object in LDAP.  Storing some environment variables isn't that
different from storing my preferred shell, or my preferred contact
information and so on.

As for implementing it could be all done in pam_ldap which would be pretty
easily portable.


On Wed, Jul 10, 2013 at 3:20 PM, Dmitri Pal <> wrote:

>  On 07/10/2013 04:48 PM, Alan Evans wrote:
>            So I have been kicking around an idea for a while now and
> thought I would develop it but its out of my league. The FreeIPA community
> is very very active in pam/sssd/ldap/so on and so on so I thought I would
> float the idea here before I made a Trac [RFE] ticket.
>  Would anyone else find it useful to store environment variables in
> LDAP?  In the environment (no pun intended) I work in we have a few
> thousand servers all authenticating to LDAP and a home grown
> LDAP+sshPublicKey implementation which is great.  But we have a bunch of
> different distros which all have different default editors.  Unless I feel
> like touching a lot of servers or using cfengine3 to distribute my
> preferred environment variables I am at the mercy of the editor on the
> system.
> How wonderful would it be to set EDITOR=vim, LESS='-S',
> TZ='America/Hawaii', LANG='Klingon' in LDAP and have pam/sss pull/store
> that information for me.  Sort of like pam_env but backed by LDAP.
>  So this got me thinking the other thing that would be wonderful to store
> in LDAP would be shell profiles...  Consider having your ~/.profile or
> ~/.bashrc  or ~/.my.cnf or what-have-you in LDAP?
> Maybe this modified pam_ldap could do things like append, remove, replace
> or unset environment variables.  Consider:
>  dn: uid=me,dc=example,dc=com
>  objectClass: posixAccountEnv
>  ...
>  # replace EDITOR
>  posixEnv: EDITOR=vim
>  # unset TZ
>  posixEnv: TZ-=
>  # append PATH
>  posixEnv: PATH=+:~/bin
>  # prepend PATH
>  posixEnv: PATH+=/opt/foo/bin:
>  Further perhaps the PAM module could be configured to only allow certain
> environment variables to be manipulated this way admins can control which
> variables users can set.
>  /etc/ldap/pam_ldap.conf:
> ...
>  # allow
>  pam_allow_env_vars PATH,EDITOR
>  # deny
>  pam_deny_env_vars PATH,TZ
>  # wildcards? regex?
>  pam_allow_env_vars LC_*,PATH,EDITOR
>  pam_deny_env_vars TZ
>  So if we're storing environment variables in LDAP why not profiles or
> small files?  ~/.bashrc, ~/.my.cnf, ~/.ssh/config?  Sure there's some
> overlap with env vars because you could set them in your profile but with
> both options an admin is free to choose.
>  I can think of a couple of ways to implement this.
>  1. create subortinate objects to the user object
>  dn: cn=~/.bashrc,uid=me,dc=example,dc=com
> ...
>  objectClass: posixAccountProfile
>  posixProfile: <octet string>
>  Advantages: The advantage of this setup is that the profileScript class
> could contain any number of attributes, perhaps a modified time so that the
> script isn't rewritten if the subordinate object hasn't been modified since
> the script was last modified.
>  Disadvantages: Kinda kludgy.  Extra objects (more lookups).
>  2. Use LDAP attribute options (See 
> 2251 "4.1.5. Attribute Description" if not familiar)
>  dn: uid=me,dc=example,dc=com
> ...
>  posixProfile;~/.bashrc: <octet string>
>  posixProfile;~/.my.cnf: <octet string>
>  Advantages: No extra objects, makes use of oft unused LDAP attribute
> options :), can have ACI's applied to them.
> Disadvantages: Only modified time to track is modified time of the LDAP
> object not individual profileScript attrs
>  In both cases it might be wise to consider how file names are specified.
> Perhaps leave off the ~/ and make everything relative to ~ no matter what.
> Or make everything relative to ~/ even if it starts w/ a '/'.  Maybe simply
> reject anything that begins with '/'.
> dn: uid=me,dc=example,dc=com
> ...
>  posixProfile;.bashrc: <octetString>
>  posixProfile;.foo/foorc: <octetstring>
> Plus I don't know if / and . are legitimate characters in attribute
> options.
>  So thanks for sticking with me if you got this far.  What do you think?
>  Regards,
>  -Alan
> _______________________________________________
> Freeipa-users mailing 
> listFreeipa-users@redhat.com
> I see couple problems but in a different plane.
> 1) You are talking about the server storing this data, it is not standard
> but extension can be made. The bigger problem is the client. Creating the
> client and porting it to multiple distros is a challenge. SSSD is in Linux
> but not in classical UNIXes. As you start looking at the client side effort
> solutions like Puppet, Chef and friends become much more attractive.
> 2) Which ENV vars need to be served to which groups of hosts. You need the
> infrastructure to define and manage it. Puppet, Chef and others are already
> good at it.
> So I am not really sure that adding the suggested data to LDAP is the
> right place. LDAP is just a storage format and client protocol. This is the
> smallest part of the effort. This would effectively lead to duplicating
> some of existing management systems.
> We discussed things like that several years ago when we were starting IPA
> project. We needed to draw the line about what to store and what not to
> store in LDAP. We came up with a following definition:
> Store things that are either traditionally stored in LDAP and already have
> LDAP schema of some sort, store things that are dynamic and must be looked
> up on the fly because security decision or configuration relies on it.
> That leaves LDAP to be storage of identities (user, groups, hosts, host
> groups) and remappings of those identities to the externally managed
> objects, i.e. user(s) to command(s) = sudo, user(s) to host access = HBAC,
> user(s) to SELinux policies = SELinux user mappings, users and hosts to SSH
> keys = SSH integration.
> So realistically you might want to have something similar to what we have
> for SELinux user mapping that would deliver a "tag" to a host and then pass
> it to some pam module that would configure system using that tag or use
> this tag to fetch something from a central policy server. But that would
> again require a client and server change and if you want it to be available
> on multiple platforms you face the same challenge as in 1).
> --
> Thank you,
> Dmitri Pal
> Sr. Engineering Manager for IdM portfolio
> Red Hat Inc.
> -------------------------------
> Looking to carve out IT costs?
> _______________________________________________
> Freeipa-users mailing list
Freeipa-users mailing list

Reply via email to