[Freeipa-users] documentation or example of using S42U for NFS
Charles Hedrick
hedrick at rutgers.edu
Thu Mar 2 18:15:15 UTC 2017
Thanks. That’s what I was originally looking for. However since asking it I realized that doing it without further limitations defeats the purpose of using Kerberos in the first place, since it means that anyone who becomes the user in Linux can access their files.
I’m trying to make sure that users can decide which machines they’re willing to expose to access without a credential. In our department we have shared systems, which we run, student labs, which we run but that are not physically secure, and user-maintained machines. We’d like to say that normally no one can access your files without having your credentials. So if someone manages to become root on one system they don’t suddenly have access to everyone’s files.
We have to relax that where users need to run cron jobs. But I’d like that relaxation to be in a controlled way. That is, I want the user to have to say that on a specific machine it’s possible to access their files based just on being their uid. But that only applies to the specific user on the specific machine. That at least restricts exposure.
If S42U is turned on for NFS as described below, then anyone who becomes root can become any user, and then access any user’s files.
The design documentation says:
"The ACL system also provides a means of limiting which user's a ticket may be obtained for using the ipaAllowToImpersonate attribute. This is not implemented."
There’s no sign of this feature in the command line tools, so I assume it’s still not implemented. Until it is I don’t think that will do what I need.
Kgetcred, however, will do the job.
On Mar 1, 2017, at 9:29 PM, Greg <gkubok at gmail.com<mailto:gkubok at gmail.com>> wrote:
I've been at this as well for a while now, and managed to make it work for my NFS needs (automounting user homes with password-less logons).
Whether and how this would apply to cron or other services, I don't know yet, but presumably would/should work in a similar manner.
My env:
$ lsb_release -d
Description: Red Hat Enterprise Linux Server release 7.3 (Maipo)
$ rpm -q ipa-server
ipa-server-4.4.0-14.el7_3.4.x86_64
$ ipa --version
VERSION: 4.4.0, API_VERSION: 2.213
This is what worked for me in the end:
1. Create "nfs/nfsserver.dom.com at DOM.COM<mailto:nfsserver.dom.com at DOM.COM>" service, and add to NFS server's /etc/krb5.keytab.
2. Create IPA "servicedelegation" rule/targets, so that they look like so:
$ ipa servicedelegationrule-show ipa-nfs-delegation
Delegation name: ipa-nfs-delegation
Allowed Target: ipa-nfs-delegation-targets
Member principals: host/nfsclient.dom.com at DOM.COM<mailto:dom.com at DOM.COM>
$ ipa servicedelegationtarget-show ipa-nfs-delegation-targets
Delegation name: ipa-nfs-delegation-targets
Member principals: nfs/nfsserver.dom.com at DOM.COM<mailto:dom.com at DOM.COM>
Only niggle here is IPA CLI didn't let me add "host/..." principal to the rule, or perhaps there's a default LDAP ACI of some sort and it requires a privilege/permission to be granted. The "ipa servicedelegationrule-add-member ..." command simply says "no such entry" for "host/..." type principals. Maybe IPA folks can comment.
I force added it to the delegation rule via LDAP instead using this ldif:
dn: cn=ipa-nfs-delegation,cn=s4u2proxy,cn=etc,dc=dom,dc=com
changetype: modify
add: memberPrincipal
memberPrincipal: host/nfsclient.dom.com at DOM.COM<mailto:nfsclient.dom.com at DOM.COM>
The "nfs/..." principal can be added using CLI "ipa servicedelegationtarget-add-member ..." just fine.
3. Allow the "nfsclient" host to impersonate users:
$ ipa host-mod nfsclient.dom.com<http://nfsclient.dom.com/> --ok-to-auth-as-delegate=true
4. On the "nfsclient" machine, add "impersonate = true" line in the "[service/nfs-client]" section of /etc/gssproxy/gssproxy.conf.
5. Restart nfs/gssproxy/rpc services on client and server (it's probably just gssproxy on the client that needs a kick, but just to be sure). I was also religiously doing "sss_cache -E" for good measure, unmounting anything that got mounted, and clearing /var/lib/gssproxy/clients of all caches, to start as cleanly before each attempt at user logon. Obv make sure the user does not have an existing/valid ticket in their own cache ("kdestroy -A" as the user), otherwise it'll just mount successfully without delegation.
I think that's it, I've messed about with the config for a long time and in many places, so there's a small chance there's something else that I did and don't remember. Gssproxy config on "nfsserver" is vanilla, as are my sssd.confs and krb5.confs on both machines, can't think of much else that I might've changed for now.
So my IPA automount config now mounts users' home dirs on the "nfsclient", without any tickets or keytabs from users.
There's also a "krb5_principal" option available in gssproxy.conf - I tried to set that to "nfs/nfsclient at DOM.COM<mailto:nfsclient at DOM.COM>" in "[service/nfs-client]" section on the "nfsclient" machine, to try and force gssproxy to use that principal instead of "host/...", but it didn't seem to work, gssproxy defaults to "host/...". Possibly mis-understanding what this option is for, and possibly "host/..." is the safer/standard option? I'm assuming it's default for a reason, or maybe just operational convenience (not having to pollute /etc/krb5.keytabs with more principals than necessary).
Hope this helps.
--
Thanks,
Greg Kubok.
On 1 March 2017 at 22:47, Orion Poplawski <orion at cora.nwra.com<mailto:orion at cora.nwra.com>> wrote:
If you ever get this into a repository, I'd love to see it. Thanks.
On 01/17/2017 07:44 PM, Charles Hedrick wrote:
> Instructions like that are several places. But NFS is different, and I believe the configuration would be different from other services.
>
> I’ve given up on this approach, and have written my own utilities. I’ve actually got three. The first two assume that users who want to do cron jobs on a machine register a keytab for that machine on a secure system. I don’t want to put the actual keytab on the machine where the user is running, because if someone can become root, they can take the keytab and use it anywhere. (I’m in a computer science dept with systems run by faculty and grad students; also systems in public labs.)
>
> So instead, I allow users to register that they want to do cron on a specific machine by putting a keytab on a secure server, associated with their username and the hostname. I expect to write a web app to set that up.
>
> Credserv runs on the machine with the keytabs. It accepts a request from a server, authenticated using the host’s host key (i.e. /etc/krb5.keytab), with a username in the request. If the user has registered a keytab for that host, credserv returns a credential, locked to that IP address, with forwarding off.
>
> Kgetcred is the client side. It runs setuid root (so it can read /etc/krb5.keytab), though it drops privs as quickly as possible. It creates a credentials cache for the user from the credentials returned from credserv.
>
> This gives the best granularity of control I can come up with.
>
> There’s a third utility in the family: renewd. It gets the uids of all current processes, and renews credentials for all of the users. It’s more complex than you’d expect because a normal renewal (as in kinit -R or kstart) leaves a brief period of time when the credentials cache is unusable. This can result in NFS failing. I use a special approach that depends upon the details of the KEYRING implementation. It should be free of race conditions for NFS. If desired, a different approach to avoid race conditions could be used for caches located in /tmp. I haven’t written that code but there’s a comment in the source outlining it.
>
> I’ll be creating a git repository for the code.
>
>
>> On Jan 17, 2017, at 6:41:13 PM, Orion Poplawski <orion at cora.nwra.com<mailto:orion at cora.nwra.com>> wrote:
>>
>> On 01/09/2017 09:52 AM, Charles Hedrick wrote:
>>> Various documentation suggests that it is possible for Gssproxy to get tickets for users who need to use NFS. This is a possible way to handle things like cron jobs.
>>>
>>> However while a gssproxy.conf example is given, there’s no sign of what needs to be done in freeipa to authorize it. I tried following instructions for LDAP access, but it doesn’t work. NFS seems to use a different, two-stage method for getting credentials, so that’s not a surprise. There are, not surprisingly, no useful error messages even with logging turned all the way up.
>>>
>>>
>>
>> I'm interested in this as well. All I've been able to find so far is:
>>
>> https://vda.li/en/posts/2013/07/29/Setting-up-S4U2Proxy-with-FreeIPA/
>>
>> haven't tried anything.
>>
>> --
>> Orion Poplawski
>> Technical Manager 720-772-5637
>> NWRA, Boulder/CoRA Office FAX: 303-415-9702
>> 3380 Mitchell Lane orion at nwra.com<mailto:orion at nwra.com>
>> Boulder, CO 80301 http://www.nwra.com<http://www.nwra.com/>
>
--
Orion Poplawski
Technical Manager 720-772-5637
NWRA, Boulder/CoRA Office FAX: 303-415-9702
3380 Mitchell Lane orion at nwra.com<mailto:orion at nwra.com>
Boulder, CO 80301 http://www.nwra.com<http://www.nwra.com/>
--
Manage your subscription for the Freeipa-users mailing list:
https://www.redhat.com/mailman/listinfo/freeipa-users
Go to http://freeipa.org<http://freeipa.org/> for more info on the project
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/freeipa-users/attachments/20170302/8b3092d9/attachment.htm>
More information about the Freeipa-users
mailing list