[Devel] [PATCH v2 0/3] lockd: use per-net refrence-counted NSM clients

Stanislav Kinsbursky skinsbursky at parallels.com
Tue Sep 18 02:37:06 PDT 2012


v2:
1) NSM transport is TCP based now. And all RPC tasks now called with
RPC_TASK_SOFTCONN. The advantage of this is that the kernel could discover
when statd is not running and fail the upcall immediately, rather than waiting
possibly many seconds for each upcall RPC to time out.
2) XDR layer violation (reference to upper RPC client cl_hostname) was
replaced by passing the string as a part of nlm_args structure.

This is a bug fix for https://bugzilla.redhat.com/show_bug.cgi?id=830862.

The problem is that with NFSv4 mount in container (with separated mount
namesapce) and active lock on it, dying child reaped of this container will
try to umount NFS and doing this will try to create RPC client to send
unmonitor request to statd.
But creation of RCP client requires valid current->nsproxy (for operation with
utsname()) and during umount on child reaper exit it's equal to zero.

Proposed solution is to introduce refrence-counter per-net NSM client, which
is created on fist monitor call and destroyed after the lst monitor call.

The following series implements...

---

Stanislav Kinsbursky (3):
      lockd: per-net NSM client creation and destruction helpers introduced
      lockd: use rpc client's cl_nodename for id encoding
      lockd: create and use per-net NSM RPC clients on MON/UNMON requests


 fs/lockd/mon.c   |   86 +++++++++++++++++++++++++++++++++++++++++++-----------
 fs/lockd/netns.h |    4 +++
 fs/lockd/svc.c   |    1 +
 3 files changed, 74 insertions(+), 17 deletions(-)




More information about the Devel mailing list