NFSv4: Fix a potential state manager deadlock when returning delegations
The nfsv4 state manager could potentially deadlock inside
__nfs_inode_return_delegation() if the server reboots, so that the calls to
nfs_msync_inode() end up waiting on state recovery to complete.
Also ensure that if a server reboot or network partition causes us to have
to stop returning delegations, that NFS4CLNT_DELEGRETURN is set so that
the state manager can resume any outstanding delegation returns after it
has dealt with the state recovery situation.
Finally, ensure that the state manager doesn't wait for the DELEGRETURN
call to complete. It doesn't need to, and that too can cause a deadlock.
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
diff --git a/fs/nfs/delegation.h b/fs/nfs/delegation.h
index 09f3837..e225a12 100644
--- a/fs/nfs/delegation.h
+++ b/fs/nfs/delegation.h
@@ -42,7 +42,7 @@
void nfs_expire_all_delegations(struct nfs_client *clp);
void nfs_expire_unreferenced_delegations(struct nfs_client *clp);
void nfs_handle_cb_pathdown(struct nfs_client *clp);
-void nfs_client_return_marked_delegations(struct nfs_client *clp);
+int nfs_client_return_marked_delegations(struct nfs_client *clp);
void nfs_delegation_mark_reclaim(struct nfs_client *clp);
void nfs_delegation_reap_unclaimed(struct nfs_client *clp);