NFS locking bug

NFS locking bug

[Ragnar Kjørstad]

Let me first ask if I've understood this correctly:
Is statd on the NFS-server supposed to keep track of all NFS-clients
that have locks on the server?
And if the NFS-server gets restarted, statd look at the list of clients
that held locks, and notify them about the restart?
And then statd/lockd on the clientside will relock the files, to make
sure no other clients can lock the same file?


This is what I do:
* mount NFS filesystem on client1
* mount NFS filesystem on client2
* lock file on client1
* verify that I can not lock the same file on client2
* reboot the NFS-server
* When I start statd on the nfs-server, I see the following
  in the logfiles for the clients:

  Jul 31 18:33:55 client rpc.statd[455]: recv_rply: [127.0.0.1] RPC status 3 
  Jul 31 18:33:55 client kernel: svc: unknown procedure (24)
  Jul 31 18:34:01 client kernel: svc: unknown procedure (24)
  Jul 31 18:34:01 client rpc.statd[455]: recv_rply: [127.0.0.1] RPC status 3 
  Jul 31 18:34:07 client kernel: svc: unknown procedure (24)
  Jul 31 18:34:07 client rpc.statd[455]: recv_rply: [127.0.0.1] RPC status 3 
  Jul 31 18:34:13 client kernel: svc: unknown procedure (24)
  Jul 31 18:34:13 client rpc.statd[455]: recv_rply: [127.0.0.1] RPC status 3 
  Jul 31 18:34:19 client rpc.statd[455]: Can't callback client (100021,1), giving up.

* I can no lock the file on client2, even if the client1 still thinks
  the file is locked.

Is this a bug?


-- 
Ragnar Kjorstad
Big Storage

Re: NFS locking bug

[Trond Myklebust]

>>>>> " " == Ragnar Kjørstad <kernel@ragnark.vestdata.no> writes:

     > This is what I do:
     > * mount NFS filesystem on client1
     > * mount NFS filesystem on client2
     > * lock file on client1
     > * verify that I can not lock the same file on client2
     > * reboot the NFS-server
     > * When I start statd on the nfs-server, I see the following
     >   in the logfiles for the clients:

     >   Jul 31 18:33:55 client rpc.statd[455]: recv_rply: [127.0.0.1]
     >   RPC status 3 Jul 31 18:33:55 client kernel: svc: unknown
     >   procedure (24) Jul 31 18:34:01 client kernel: svc: unknown
     >   procedure (24) Jul 31 18:34:01 client rpc.statd[455]:
     >   recv_rply: [127.0.0.1] RPC status 3 Jul 31 18:34:07 client
     >   kernel: svc: unknown procedure (24) Jul 31 18:34:07 client
     >   rpc.statd[455]: recv_rply: [127.0.0.1] RPC status 3 Jul 31
     >   18:34:13 client kernel: svc: unknown procedure (24) Jul 31
     >   18:34:13 client rpc.statd[455]: recv_rply: [127.0.0.1] RPC
     >   status 3 Jul 31 18:34:19 client rpc.statd[455]: Can't
     >   callback client (100021,1), giving up.

     > * I can no lock the file on client2, even if the client1 still thinks
     >   the file is locked.

     > Is this a bug?

The NLM lock reclaiming code in the stock 2.4.x kernel is
incomplete. Please apply the patch on

   http://www.fys.uio.no/~trondmy/src/2.4.7/linux-2.4.7-reclaim.dif

Cheers,
   Trond