After getting sssd logins working yesterday (thanks again, Sumit), I was pleasantly surprised to find I was able to login this morning with my domain credentials from home /before/ I had established my VPN connection to the office. (I know I shouldn't have necessarily been surprised, that's the expected behavior, but I've been fiddling with this for weeks and only yesterday finally got things working as 'expected'.)
Before I made my VPN connection, I did a klist to see the cached credentials, and did a double-take when I saw the TGT:
At first I thought I was back in the U.S. Navy boot camp (which is where I was on December 31, 1969) but then I decided this timestamp might have been chosen intentionally to pre-date UNIX epoch time. But why go to all that trouble rather than just use the valid TGT I had received yesterday when I made a live, valid connection? Wasn't that cached, along with my authentication credentials?
Once I established my tunnel connection, I checked again, saw the same (old) TGT, so I logged out of the session (without dropping the tunnel connection) and when I logged back in I had a TGT dated today. I'm guessing (something I can test easily enough) that if I had waiting long enough before logging out and back in again, the TGT would have been re-issued correctly.
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 04/11/2013 08:15 AM, Sutton, Harry (GSSE) wrote:
After getting sssd logins working yesterday (thanks again, Sumit), I was pleasantly surprised to find I was able to login this morning with my domain credentials from home /before/ I had established my VPN connection to the office. (I know I shouldn't have necessarily been surprised, that's the expected behavior, but I've been fiddling with this for weeks and only yesterday finally got things working as 'expected'.)
Before I made my VPN connection, I did a klist to see the cached credentials, and did a double-take when I saw the TGT:
At first I thought I was back in the U.S. Navy boot camp (which is where I was on December 31, 1969) but then I decided this timestamp might have been chosen intentionally to pre-date UNIX epoch time. But why go to all that trouble rather than just use the valid TGT I had received yesterday when I made a live, valid connection? Wasn't that cached, along with my authentication credentials?
Our default behavior on modern systems is actually to store the kerberos credential cache in volatile storage (a tmpfs on Fedora). This is intentional as a security precaution, as it means that on reboot you need to have human intervention in order to gain network access again. This can be changed in sssd.conf by setting the krb5_ccachedir option to point to somewhere persistent.
We create that special cache file in order to play better with other kerberos-enabled applications (such as krb5-auth-dialog) which would otherwise try to create their own cache in a location that SSSD couldn't be aware of. So we generate a cache file, populate it with intentionally expired data and then establish the user's session with the KRB5CCNAME environment variable set to point to it for the benefit of those other kerberized apps.
Once I established my tunnel connection, I checked again, saw the same (old) TGT, so I logged out of the session (without dropping the tunnel connection) and when I logged back in I had a TGT dated today. I'm guessing (something I can test easily enough) that if I had waiting long enough before logging out and back in again, the TGT would have been re-issued correctly.
If this is a single-user machine, we recommend enabling the (disabled-by-default) option krb5_store_password_if_offline (set it to True). What this does is temporarily store your plaintext password in the kernel keyring until SSSD detects that it has gone "online". This may take as much as two minutes on some systems (though it may be close to instantaneous when using VPN, as we monitor for changes in the network and DNS resolver state as a cue to try reconnecting to LDAP). Then the SSSD will kinit on your behalf and update that stale cache.
Also, locking your screen and unlocking it with your password is a less-invasive way to force a kinit as well, rather than logging out and back in.
On 04/11/2013 08:40 AM, Stephen Gallagher wrote:
Our default behavior on modern systems is actually to store the kerberos credential cache in volatile storage (a tmpfs on Fedora). This is intentional as a security precaution, as it means that on reboot you need to have human intervention in order to gain network access again. This can be changed in sssd.conf by setting the krb5_ccachedir option to point to somewhere persistent.
We create that special cache file in order to play better with other kerberos-enabled applications (such as krb5-auth-dialog) which would otherwise try to create their own cache in a location that SSSD couldn't be aware of. So we generate a cache file, populate it with intentionally expired data and then establish the user's session with the KRB5CCNAME environment variable set to point to it for the benefit of those other kerberized apps.
That makes sense, I was overlooking the security vulnerability that caching a (potentially) still valid credential would present.
If this is a single-user machine, we recommend enabling the (disabled-by-default) option krb5_store_password_if_offline (set it to True). What this does is temporarily store your plaintext password in the kernel keyring until SSSD detects that it has gone "online". This may take as much as two minutes on some systems (though it may be close to instantaneous when using VPN, as we monitor for changes in the network and DNS resolver state as a cue to try reconnecting to LDAP). Then the SSSD will kinit on your behalf and update that stale cache.
Also, locking your screen and unlocking it with your password is a less-invasive way to force a kinit as well, rather than logging out and back in.
Both good suggestions, thanks.
/Harry
On Thu, Apr 11, 2013 at 08:15:41AM -0400, Sutton, Harry (GSSE) wrote:
After getting sssd logins working yesterday (thanks again, Sumit), I was pleasantly surprised to find I was able to login this morning with my domain credentials from home /before/ I had established my VPN connection to the office. (I know I shouldn't have necessarily been surprised, that's the expected behavior, but I've been fiddling with this for weeks and only yesterday finally got things working as 'expected'.)
Before I made my VPN connection, I did a klist to see the cached credentials, and did a double-take when I saw the TGT:
At first I thought I was back in the U.S. Navy boot camp (which is where I was on December 31, 1969) but then I decided this timestamp might have been chosen intentionally to pre-date UNIX epoch time. But why go to all that trouble rather than just use the valid TGT I had received yesterday when I made a live, valid connection? Wasn't that cached, along with my authentication credentials?
Once I established my tunnel connection, I checked again, saw the same (old) TGT, so I logged out of the session (without dropping the tunnel connection) and when I logged back in I had a TGT dated today. I'm guessing (something I can test easily enough) that if I had waiting long enough before logging out and back in again, the TGT would have been re-issued correctly.
I think krb5_store_password_if_offline (see man sssd-krb5) is the option you are looking for. About the strange date, sssd creates and empty credential cache with UNIX epoch time to allow other desktop application which tries to re-new the Kerberos ticket start working. I think that you see December 31 is due to your timezone.
HTH
bye, Sumit
-- *Harry Sutton* Global Solutions Support Engineering (GSSE) GSD Customer Solution Center Technology Services, Enterprise Group
sssd-users mailing list sssd-users@lists.fedorahosted.org https://lists.fedorahosted.org/mailman/listinfo/sssd-users
On 04/11/2013 08:42 AM, Sumit Bose wrote:
I think krb5_store_password_if_offline (see man sssd-krb5) is the option you are looking for. About the strange date, sssd creates and empty credential cache with UNIX epoch time to allow other desktop application which tries to re-new the Kerberos ticket start working. I think that you see December 31 is due to your timezone.
HTH
bye, Sumit
That fits - I'm at UTC -5 (US East Coast on Daylight Saving Time,) and the timestamp was 19:00.
/Harry
sssd-users@lists.fedorahosted.org
-
Stephen Gallagher -
Sumit Bose -
Sutton, Harry (GSSE)