netfs: don't race NFS umount to network shutdown

RHEL6 customers have been reporting hangs when restaring due to IO
for NFS filesystems being unable to flush after network shutdown.
The current __umount_loop allows newly created processes to continue
to open files to NFS filesystems, which can create thisi problem.

Change the umount logic to perform a MNT_DETACH, then search for
processes that have open file descriptors on the detached filesystems.
The detach prevents newly created processes from opening new files
during our search.  Finally, after making every attempt to clean up
processes with open files, perform a sync to flush NFS filesystems
before continuing onto network shutdown.

0 files changed, 0 insertions, 0 deletions