On Wed, May 02, 2012 at 05:58:37PM -0400, J. Bruce Fields wrote:Thanks, applying.
Whoops, no--I'm hitting the BUG at net/sunrpc/svc.c:558 (the first of
the two BUG_ON()s in svc_destroy()) when restarting nfsd. Could you
look into this?
--b.
--b.
On Wed, May 02, 2012 at 04:08:29PM +0400, Stanislav Kinsbursky wrote:creation
v3: "SUNRPC: new svc_bind() routine introduced" patch was squashed with the
"SUNRPC: check rpcbind clients usage counter before decrement" patch.
v2: Increase per-net usage counted in lockd_up_net().
This is a cleanup patch set.
It will be followed my LockD start/stop cleanup patch set and NFS callback
service containerization patch set (yes, I forgot to implement it).
Today per-net data is created with service, and then is service is starting in
other network namespace. And thus it's destroyed with service too. Moreover,
network context for destroying of per-net data is taken from current process.
This is correct, but code looks ugly.
This patch set separates per-net data allocation from service allocation and
destruction.
IOW, per-net data have to be destroyed by service users - not service itself.
BTW, NFSd code become uglier with this patch set. Sorry.
But I assume, that these new ugly parts will be replaced later by NFSd service
containerization code.
The following series implements...
---
Stanislav Kinsbursky (2):
SUNRPC: new svc_bind() routine introduced
SUNRPC: move per-net operations from svc_destroy()
fs/lockd/svc.c | 33 +++++++++++++++++++++------------
fs/nfs/callback.c | 11 +++++++++++
fs/nfsd/nfsctl.c | 4 ++++
fs/nfsd/nfssvc.c | 16 ++++++++++++++++
include/linux/sunrpc/svc.h | 1 +
net/sunrpc/rpcb_clnt.c | 12 +++++++-----
net/sunrpc/svc.c | 23 ++++++++++-------------
7 files changed, 70 insertions(+), 30 deletions(-)