The following patches address a couple of deadlocks that we've been
seeing during reboot recovery and/or delegation recovery.
For NFSv4, it was safe for ordinary processes to grab the nfs_inode->rwsem
across an NFSv4 RPC call, since if the stateid needed recovery, the server
would return an appropriate error, the rpc call would return, and the
process would release the nfs_inode->rwsem so that the state recovery
thread could proceed.
In NFSv4.1, things get a little more interesting: the state recovery
thread will block all outstanding RPC calls using the NFS4_SESSION_DRAINING
flag. It needs to be able to do this so that we can safely recover
session state, stateids, etc.
The problem is that if a process is holding a lock, then that lock will
remain held until we call nfs4_end_drain_session(). So if the recovery
process needs to take that lock, we end up deadlocking...
The following patchset attempts to address this issue by adjusting the
locking in the state recovery and delegation recovery.
Trond Myklebust (6):
NFSv4: Allow the state manager to mark an open_owner as being
recovered
NFSv4.1: Prevent deadlocks between state recovery and file locking
NFSv4.1: Don't lose locks when a server reboots during delegation
return
NFSv4: Fix up the return values of nfs4_open_delegation_recall
NFSv4: Ensure delegation recall and byte range lock removal don't
conflict
NFSv4: Fix a reboot recovery race when opening a file
fs/nfs/delegation.c | 154 +++++++++++++++++++++++++++++++++++++---------------
fs/nfs/delegation.h | 1 +
fs/nfs/nfs4_fs.h | 4 ++
fs/nfs/nfs4proc.c | 96 +++++++++++++++++++++++---------
fs/nfs/nfs4state.c | 10 +++-
5 files changed, 193 insertions(+), 72 deletions(-)
--
1.8.1.2
--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
