svn commit: r225154 - in projects/ino64/sys/ufs: ffs ufs

John Baldwin jhb at freebsd.org
Thu Aug 25 11:49:34 UTC 2011


On Wednesday, August 24, 2011 7:09:24 pm mdf at freebsd.org wrote:
> On Wed, Aug 24, 2011 at 3:14 PM, Matthew D Fleming <mdf at freebsd.org> wrote:
> > Author: mdf
> > Date: Wed Aug 24 22:14:55 2011
> > New Revision: 225154
> > URL: http://svn.freebsd.org/changeset/base/225154
> >
> > Log:
> >  Use fixed-width types in on-disk structures.  This takes care of all the
> >  ino_t references I could find that were on-disk.
> >
> >  GSoC r223157.
> >  Code by Gleb Kurtsou.
> 
> Note that I've explicitly left all the XXXfid structs alone; when
> ino_t is changed to 64-bit they will use 64-bit members for inode
> number.  The only in-tree one that may be a problem is ReiserFS which
> will have a 32 byte FID after compiler padding is applied.  This can
> be reduced to 24 bytes by reordering the members.  In theory, IIRC,
> even 32 bits is an okay FID, however I have seen a note at $WORK that
> NFSv2 has 32 byte file handles and BSD is using 8 bytes for the fsid.
> 
> If anyone knows anything more, please let me know.  For the moment I
> think my plan is to re-order the members of various implementations to
> keep them smaller, and leave it as an ino_t in the struct definition.

One suggestion I do have in regards to this branch.  Earlier there was a
request to increase the userland dev_t to 32-bits.  The primary user of
this would seem to be OpenAFS.  Gleb did not agree to do this, but I
actually think we should go ahead and bump dev_t.  Changing for the
layout of stat is a lot of work and I'd rather err on the side of being
too future proof than deciding 2 years from now "gee, we really could
have used a bigger dev_t".

-- 
John Baldwin


More information about the svn-src-projects mailing list