Date: Sat, 22 Mar 2008 11:51:39 +0000 From: Doug Rabson <dfr@rabson.org> To: Doug Rabson <dfr@rabson.org> Cc: current@freebsd.org Subject: Re: CFR: New NFS Lock Manager Message-ID: <E392F264-BCBC-4752-8E99-91ADA2228EBF@rabson.org> In-Reply-To: <E7E41AFB-FB94-47C1-9169-71992D70C320@rabson.org> References: <E7E41AFB-FB94-47C1-9169-71992D70C320@rabson.org>
next in thread | previous in thread | raw e-mail | index | archive | help
--Apple-Mail-106-246233165 Content-Type: text/plain; charset=US-ASCII; format=flowed; delsp=yes Content-Transfer-Encoding: 7bit I've just uploaded a new patch at http://people.freebsd.org/~dfr/lockd-RC2-22032008.diff . This fixes a serious problem on kernels not compiled with the LOCKF_DEBUG option (I misplaced a #endif). It also includes minor fixes to support 64bit architectures and RELENG_7 (the patch does not apply cleanly to RELENG_7 but does work when you work around the patch rejects manually). On 21 Mar 2008, at 10:27, Doug Rabson wrote: > As I mentioned previously, I have been working on a brand new NFS > Lock Manager which runs in kernel mode and uses the normal local > locking infrastructure for its state. I'm currently trying to tie up > the last few loose ends before committing this work to current. You > can find a snapshot of this code at http://people.freebsd.org/~dfr/lockd-RC1-20032008.diff > . > > To try it out, take a recent current (I last merged with current on > 20th March) and apply the patch. Build a kernel with the NFSLOCKD > option and add '-k' to 'rpc_lockd_flags' in rc.conf. You will need > to build and install at least a new libc and rpc.lockd. > > At this point, it would be useful to get some extra eyes to look > over my changes. In particular the following: > > 1. Choice of syscall number - I found one spare next to the NFS > syscall and took that. The new syscall is listed in the FBSD_1.1 > namespace, possibly it should be somewhere else. > > 2. ABI compatibility - I extended the flock structure by one member > (adding l_sysid). I have added new operations to fcntl to support > the new extended structure, leaving the old operations in place to > work on the old structure. The kernel translates old to new and vice > versa. No attempt is made to allow a new userland to work with an > old kernel. > > 3. The local lock manager has had a complete rewrite to support > required features. The new local lock manager supports a more > flexible model of lock ownership (which can support remote lock > owners). I have replaced the inadequate deadlock detection code with > a new (and fast) graph based system. Using the deadlock graph, I was > able to avoid the 'thundering herd' issues the old lock code had > when many processes were contending for the same locked region. > Given the extent of the changes, wider testing and review would be > extremely welcome. > > 4. The NFS lock manager itself is brand new code and as such ought > to be reviewed. I have also ported the userland sunrpc code to run > in the kernel environment which may prove useful in future. > > Highlights include: > > * Thread-safe kernel RPC client - many threads can use the same RPC > client handle safely with replies being de-multiplexed at the socket > upcall (typically driven directly by the NIC interrupt) and handed > off to whichever thread matches the reply. For UDP sockets, many RPC > clients can share the same socket. This allows the use of a single > privileged UDP port number to talk to an arbitrary number of remote > hosts. > > * Single-threaded kernel RPC server. Adding support for multi- > threaded server would be relatively straightforward and would follow > approximately the Solaris KPI. A single thread should be sufficient > for the NLM since it should rarely block in normal operation. > > * Kernel mode NLM server supporting cancel requests and granted > callbacks. I've tested the NLM server reasonably extensively - it > passes both my own tests and the NFS Connectathon locking tests > running on Solaris, Mac OS X and Ubuntu Linux. > > * Userland NLM client supported. While the NLM server doesn't have > support for the local NFS client's locking needs, it does have to > field async replies and granted callbacks from remote NLMs that the > local client has contacted. We relay these replies to the userland > rpc.lockd over a local domain RPC socket. > > * IPv6 should be supported but has not been tested since I've been > unable to get IPv6 to work properly with the Parallels virtual > machines that I've been using for development. > > * Robust deadlock detection for the local lock manager. In > particular it will detect deadlocks caused by a lock request that > covers more than one blocking request. As required by the NLM > protocol, all deadlock detection happens synchronously - a user is > guaranteed that if a lock request isn't rejected immediately, the > lock will eventually be granted. The old system allowed for a > 'deferred deadlock' condition where a blocked lock request could > wake up and find that some other deadlock-causing lock owner had > beaten them to the lock. > > * Since both local and remote locks are managed by the same kernel > locking code, local and remote processes can safely use file locks > for mutual exclusion. Local processes have no fairness advantage > compared to remote processes when contending to lock a region that > has just been unlocked - the local lock manager enforces a strict > first-come first-served model for both local and remote lockers. --Apple-Mail-106-246233165--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?E392F264-BCBC-4752-8E99-91ADA2228EBF>