Description of problem:

When an NFS server is rebooted while an NFS client has an NFSv4/krb5 export mounted, the NFS client can not access the content of the mountpoint even a few minutes after the NFS server has completed the reboot.

Following the reboot of the NFS server, it receives RPC_GSS_PROC_DATA packets from the NFS client. In svcauth_gss_accept():
------------------------------------------------------------
 830         case RPC_GSS_PROC_DATA:
 831         case RPC_GSS_PROC_DESTROY:
 832                 *authp = rpcsec_gsserr_credproblem;
 833                 rsci = gss_svc_searchbyctx(&gc->gc_ctx);
 834                 if (!rsci)
 835                         goto auth_err;

 937 auth_err:
 938         /* Restore write pointer to original value: */
 939         xdr_ressize_check(rqstp, reject_stat);
 940         ret = SVC_DENIED;
 941         goto out;
------------------------------------------------------------
gss_svc_searchbyctx() returns NULL which results in authp being set to rpcsec_gsserr_credproblem and svcauth_gss_accept() returning SVC_DENIED.

Eventually svc_authenticate() returns SVC_DENIED and auth_stat is set to rpcsec_gsserr_credproblem:
------------------------------------------------------------
255 int
256 svc_process(struct svc_serv *serv, struct svc_rqst *rqstp)
257 {

320         auth_res = svc_authenticate(rqstp, &auth_stat);

328         switch (auth_res) {
329         case SVC_OK:
330                 break;
331         case SVC_GARBAGE:
332                 rpc_stat = rpc_garbage_args;
333                 goto err_bad;
334         case SVC_SYSERR:
335                 rpc_stat = rpc_system_err;
336                 goto err_bad;
337         case SVC_DENIED:
=> 338                 auth_stat = rpc_autherr_badcred;
339                 goto err_bad_auth;
340         case SVC_DROP:
341                 goto dropit;
342         case SVC_COMPLETE:
343                 goto sendit;
344         }
------------------------------------------------------------
In RHEL4 the auth_stat value of rpcsec_gsserr_credproblem is overwritten and set to rpc_autherr_badcred.

If the NFS server returns RPCSEC_GSS_CREDPROBLEM, then the NFS client calls call_refresh() and refreshes the credentials, which is the RHEL5 behaviour. An RPC_AUTH_BADCRED results in the NFS client doing a retry instead.


Version-Release number of selected component (if applicable):
* Red Hat Enterprise Linux AS release 4 (Nahant Update 7)
* Kernel 2.6.9-78

How reproducible:
* Always

Steps to Reproduce:
0. Setup a kerberized NFS environment.
1. On the NFS client, mount the NFSv4/krb5 export at /mnt/test
2. Do an ls /mnt/test
3. Reboot the NFS server. Wait till the NFS server restarts.
4. Do an ls /mnt/test
  
Actual results:
ls: /mnt/test: Permission denied

Expected results:
* The content of the directory.

Additional info:
* If you wait ~ 45 minutes, the ls /mnt/test will succeed. I suspect there is a timeout on the NFS client that results in the NFS client sending an RPC_GSS_PROC_INIT packet.

Regards,
Harshula