svn commit: r212293 - in head/sys: fs/nfsclient nfsclient

[John Baldwin]

On Tuesday, September 07, 2010 10:29:45 am John Baldwin wrote:
> Author: jhb
> Date: Tue Sep  7 14:29:45 2010
> New Revision: 212293
> URL: http://svn.freebsd.org/changeset/base/212293
> 
> Log:
>   Store the full timestamp when caching timestamps of files and
>   directories for purposes of validating name cache entries.  This
>   closes races where two updates to a file or directory within the same
>   second could result in stale entries in the name cache.  While here,
>   remove the 'n_expiry' field as it is no longer used.
>   
>   Reviewed by:	rmacklem
>   MFC after:	1 week

Note that this only reduces races if the NFS server supports sub-second 
timestamps.  FreeBSD NFS servers do not by default since the global vfs 
timestamp is only updated once a second by default.  In particular, the name 
cache entries in the NFS client can become stale any time that multiple 
updates can happen to a directory within the same timestamp.  Put another way, 
if a directory has multiple "states" that share the same timestamp, the NFS 
client may cache the earlier state until the next directory update.  It may be 
worth adding an optional timeout on positive cache entries in the NFS client 
just as we support having a timeout on negative cache entries to limit the 
extent of such races at the expense of increased LOOKUP RPCs.

-- 
John Baldwin