core: Return BAD_STATEID for NFSv4.0 special "stateless" stateids

Under some rare circumstances, especially when under I/O pressure, NFS
clients such as those on macOS may try to send NFS requests with an
invalid state id (seq=-1, other:ffffffffffffffffffffffff; or seq:0,
other:0).

This usually happens after the NFS server has been restarted and lost
information about the previous client state. Unfortunately, the macOS
client does not recover from this situation and keeps making requests to
the server. This slows down both server and client to a halt.

Send a "bad stateid" error upon occurring such a sequence. This lets the
NFS client reconnect and re-establish the correct state with the server.

Fixes: #160
Related: https://datatracker.ietf.org/doc/html/rfc7530#section-9.1.4.3
Signed-off-by: Christian Kohlschütter <christian@kohlschutter.com>

Author: kohlschuetter

core/src/main/java/org/dcache/nfs/v4/NFSv4StateHandler.java

@@ -238,6 +238,17 @@ public NFS4Client getClient(clientid4 clientid) throws StaleClientidException {
     }
 
     public NFS4Client getClientIdByStateId(stateid4 stateId) throws ChimeraNFSException {
+        if (Stateids.isStateLess(stateId)) {
+            // We do not support NFSv4.0 Anonymous/Bypass Stateids:
+            // https://datatracker.ietf.org/doc/html/rfc7530#section-9.1.4.3
+            //
+            // This is occasionally used by macOS NFS when it loses track of its state.
+            // Returning BAD_STATEID forces a connection-restart.
+            //
+            // https://github.com/apple-oss-distributions/NFS/blob/NFS-327.120.3/kext/nfs4_vnops.c#L2813
+            // https://github.com/apple-oss-distributions/NFS/blob/NFS-327.120.3/kext/nfs4_vnops.c#L1863L1880
+            throw new BadStateidException();
+        }
 
         _readLock.lock();
         try {