NFS locking routines (nlmsvc_lock and others in svclock.c) go directly to the default kernel operations (eg., posix_lock_file) and do not invoke any filesystem lock operation that may have been provided. They SHOULD invoke the lock file_operation (if one exists for the filesystem) similarly to the non-NFS fcntl_setlk which does the following: if (filp->f_op->lock != NULL) { error = filp->f_op->lock(filp, cmd, &file_lock); Without this, a filesystem that needs to provide the lock operation has no way to achieve proper locking via nfs clients.
assigned to johnsonm
Alan, you wanted lots of testing on this one; how can we get this?
Has there been any update on this bug?
I am happy that this is a real bug. No existing code Red Hat ship is effected by it but a hypothetical third party file system (I suspect a not so hypothetical AFS too) would be affected by this and it is a reasonable change. I thought IBM provided 2.4.x patches for this. If they have a 2.4 set for that bug can they send me a copy and I'll add it to the 2.4ac base
Can I please be updated on the latest for this bug? I was informed that IBM sent you this patch on 2/23/01. Will this be resolved in our next release? Thanks, Kyu
Its in the current -ac tree for testing. I had to fix some formatting mess and bits they didnt do but it seems ok. For Linus tree IBM should discuss the -ac changes with Matthew Wilcox <matthew> who is the locking maintainer. He is aware of the changes Alan
Thank you for the update Alan. Will patch be a part of our next release?
Kyu, please send linux-2.4.2-lockd.patch to IBM so they can look at it.
I have sent and IBM reviewed the lockd.patch. They informed me that everything looks great.
Great! I will close this bug now as fixed, if something comes up, feel free to reopen it again.
This patch (svclock.c) apparently did not make it into a Linus kernel.org release although it was accepted into 2.2.4-ac23 - why did this happen?
I assume because Linus didn't decide to adopt it. You want to discuss that with Linus.
Before we discuss this with Linus, is there any indication as to why it was not accepted.
I dont have any, its been a long time
Is this the hooks for IBM's binary only filesystem ?
Well it matters for AFS (which is now not binary only anyway) but its also a genuine 100% kosher bug.
What would be the most expeditious approach get this patch accepted into the Linus kernel?
How about sending it to Linus and Marcelo with explenation ?
Thanks for the bug report. However, Red Hat no longer maintains this version of the product. Please upgrade to the latest version and open a new bug if the problem persists. The Fedora Legacy project (http://fedoralegacy.org/) maintains some older releases, and if you believe this bug is interesting to them, please report the problem in the bug tracker at: http://bugzilla.fedora.us/