large ufs2 partitions and 'df'

[Kirk McKusick]

> Date: Sun, 11 May 2003 15:11:59 -0700 (PDT)
> From: Julian Elischer <julian at elischer.org>
> To: Kirk McKusick <mckusick at mckusick.com>
> cc: freebsd-current at freebsd.org
> Subject: Re: large ufs2 partitions and 'df' 
> In-Reply-To: <200305112138.h4BLcVTh036013 at beastie.mckusick.com>
> X-ASK-Info: Whitelist match
> 
> On Sun, 11 May 2003, Kirk McKusick wrote:
> 
> > I think that we should use merge the two spares at the start of the
> > structure into a 64-bit f_bfree field. Rename the existing f_bfree
> > to f_old_bfree and set it to the same value as f_bfree if it will fit,
> > otherwise the largest number that will fit (ox7fffffff). Similarly,
> > take the two spare longs at the end and merge them into a 64-bit
> > b_avail field. Rename f_bavail to f_old_bavail and set as with
> > f_old_bfree. Those are the only two fields that need to grow, and
> > this will provide binary compatibility for existing applications.
> > Seem reasonable?
> 
> It would be a good idea except that there is only one spare long
> at the start of the structure.. its name is f_spare2 but it is only
> one long in length.
> 
> I think that swithing to a new syscall with a fixed structure
> and using the rules you mention above to populate the structure in an 
> ostatfs call might be the best answer.
> Old binaries probably only need to know that there is > X blocks free
> and not necessarily the correct number.
> New binaries can use the new syscall.

So right you are. It would be possible to get the space by nibbling
a bit more space from MNAMELEN, but at some point we need to just bite
the bullet and define a new structure. I am leaning towards believing
that time is now. If we do define a new structure, I would like to
clean up the existing one a bit. I would propose this:

#define	MFSNAMELEN	16	/* length of fs type name, including null */
#define	MNAMELEN	80	/* size of on/from name bufs */
struct statfs {
	u_int_32 f_bsize;		/* fundamental filesystem block size */
	u_int_32 f_iosize;		/* optimal transfer block size */
	int_64	 f_blocks;		/* total data blocks in filesystem */
	int_64	 f_bfree;		/* free blocks in fs */
	int_64	 f_bavail;		/* free blocks avail to non-superuser */
	int_64	 f_files;		/* total file nodes in filesystem */
	int_64	 f_ffree;		/* free file nodes in fs */
	u_int_64 f_syncwrites;		/* count of sync writes since mount */
	u_int_64 f_asyncwrites;		/* count of async writes since mount */
	u_int_64 f_syncreads;		/* count of sync reads since mount */
	u_int_64 f_asyncreads;		/* count of async reads since mount */
	u_int_64 f_spare[10];		/* unused spare */
	fsid_t	 f_fsid;		/* filesystem id */
	uid_t	 f_owner;		/* user that mounted the filesystem */
	u_int_32 f_type;		/* type of filesystem */
	u_int_32 f_flags;		/* copy of mount exported flags */
	char	 f_fstypename[MFSNAMELEN]; /* fs type name */
	char	 f_mntfromname[MNAMELEN];  /* mounted filesystem */
	char	 f_mntonname[MNAMELEN];	   /* directory on which mounted */
};

It reorganizes things back into a rational order. It increases the
sizes of everything that might ever want to be 64-bits to that size,
and leaves plenty (10) of spares for future growth. Comments?

	Kirk McKusick