From: Brent Callaghan (brent@caribe-86)
Date: 03/11/97-04:17:27 PM Z
Date: Tue, 11 Mar 1997 14:17:27 -0800 From: brent@caribe-86 (Brent Callaghan) Message-Id: <199703112217.OAA01354@terra.eng.sun.com> Subject: Re: Attributes > From lance@eco.twg.com Tue Mar 11 12:49:56 1997 > Would it not be possible for a client to derive a file id from some form of > mapping from the file handle? I know that the opaqueness of the file handle > can make hashing difficult, but there are other techniques which would allow > an efficient mapping to be used. > > For our implementation on NTFS, the file handles and file-ids are closely > related anyway, and I'm can't off-hand think of any failure mode that would > be introduced by the client doing the mapping rather than the server. Yep, I think a fileid derived from mapping (or hash) of the filehandle would be quite suitable for a fileid. If the mapping is good then the chance of a fileid collision is probably small enough to ignore. A Unix client would need to associate a fileid with every directory entry to satisfy the needs of readdir(), so for a server that doesn't supply fileid's the client would have to fetch filehandles for all directory entries to derive the fileid equivalents. It could be argued that it's kinder to the client to have the server do the fudgery and return derived fileids, but then the client has no idea of the quality of the derived fileids. I think you're right in suggesting that the client do it. Brent
This archive was generated by hypermail 2.1.2 : 03/04/05-01:45:31 AM Z CST