Re: [PATCH 1/2] handling 64bit values for st_ino]

[Peter Staubach <staubach@redhat.com>]

Al Viro wrote:

>[My apologies, forgot to Cc the first half...]
>
>Date: Thu, 10 Nov 2005 00:27:29 +0000
>From: Al Viro <viro@ftp.linux.org.uk>
>To: Linus Torvalds <torvalds@osdl.org>
>Subject: [PATCH 1/2] handling 64bit values for st_ino
>User-Agent: Mutt/1.4.1i
>
>	We certainly do not want 64bit kernel ino_t, since that would
>screw icache lookups for no good reason; fs with 64bit keys used to
>identify inodes can just use iget5().
>  
>

Has this potential degradation been measured?  This is a lot of extra
complexity which needs to justified by the resulting performance.

>	Fix is pretty cheap and consists of two parts:
>1) widen struct kstat ->ino to u64, add a macro (check_inumber()) to
>be used in callers of ->getattr() that want to store ->ino in possibly
>narrower fields and care about overflows (stuff like sys_old_stat() with
>its 16bit st_ino clearly doesn't ;-)
>

It seems to me that a type with a name which better matches the intended
semantics would be a better choice than u64.  Even something like ino64_t
would help file systems maintainers to correctly implement the appropriate
support.

    Thanx...

       ps