[nfsd PATCH 0/3] address issues with shutdown while portmap is dead

[NeilBrown <neilb@suse.de>]

Hi all.
Currently if you shutdown nfsd while portmap is dead, it takes about
3.5 minutes.  This isn't ideal.

The following patch series address some issues surrounding that.
The first two a minor fixes to nfsd which get it behave a little bit
more consistently.  They don't directly affect the issue.

The third changes the timeouts used when rpcb_client talks to
portmap or rpcbind to unregister a service.  It simply reduces the
timeout to 0.5 seconds with no retries.
As there should be no packet loss, this should only fail if
portmap/rpcbind isn't working.  And in that case unregistering
wont serve any purpose anyway.

And as unregistering is merely for cleanliness, not for correctness,
there wouldn't be a problem even if something did go wrong.

With this patch, the 3.5 minutes drops to 3 seconds.

Comments welcome.

Thanks,
NeilBrown


---

NeilBrown (3):
      sunrpc: reduce timeout when unregistering rpcbind registrations.
      nfsd: optimise the starting of zero threads when none are running.
      nfsd: don't take nfsd_mutex twice when setting number of threads.


 fs/nfsd/nfsctl.c       |   13 +++++++++----
 fs/nfsd/nfssvc.c       |    8 ++++++--
 net/sunrpc/rpcb_clnt.c |   17 +++++++++++++++--
 3 files changed, 30 insertions(+), 8 deletions(-)

-- 
Signature