Re: 2.2.10-ac<x> / 2.2.11-pre<x> NFS client problem & bugfix

Frank van Maarseveen (fvm@tasking.nl)
Sat, 7 Aug 1999 21:26:40 +0200

On Sat, Aug 07, 1999 at 09:30:14AM -0400, Steven N. Hirsch wrote:
>
> Just to avoid confusing the masses (among which I count myself), can you
> clarify whether this problem exists in stock 2.2.10 or Alan's
> proposed-2.2.11-* patches? From a quick read of nfs/dir.c, it does not
> appear that nfs_renew_times() is being called on "goto out_valid".
>
> May I assume that this discussion is particular to -acXX and/or the NFS3
> patches?
>
NFS v3 for Linux hasn't been used in any of the tests.
There are two problems, not just one:
1) negative dentry caching or not
2) live time of negative dentries and under what conditions.

Regarding (1): whether or not this is a problem depends on the
situation. IMNSHO: should be possible to switch this feature off!

Regarding (2) I have tested only a very few combinations of
parameters which affect the behavior:
- doing a readdir or not.
- parent dir is >15 minutes old or not
- stat() calls are repeated within 1 second or not
- mount options (acregmin, acdirmin, etc)
- kernel version
- NFS version

Regarding the mount options I did not use any ac* option. NFS version
is always 2. Readdirs are uncertain (I'm not sure what bash does in
some of the test scripts). Here's the result:

ac11: infinite negative dentries provided the stat() is
repeated within every second. This is the case even
when the parent dir mtime is < 15 minutes old. Once
the stat() calls stop, there remains a >1 second
live time for negative dentries. pre4 seems ok.
ac11/pre4 negative dentries for about 30 seconds provided the
mtime of the parent dir is >15minutes old. No repeated
stat() in this case.
ac11/pre4 negative dentries for about 1 second otherwise.

-- 
Frank


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/