Re: [RFC] Slimming down struct inode

["Brian F. G. Bidulock" <bidulock@openss7.org>]

Nathan,

On Thu, 15 Jun 2006, Nathan Scott wrote:
> 
> There is no IRIX/Linux compatibility layer, you're misunderstanding
> the code (which is understandable, its erm a bit crufty in places).

You gotta be kidding.  It does everything in terms of an SVR 4 VFS
vnode and then converts that to Linux VFS on dentries/inodes.  It was
obviously built by stuffing the Linux VFS under SVR 4 VFS code and
even documents the code as such.  Things like:

| /*
|  * XFS arguments structure, constructed from the arguments we
|  * are passed via the mount system call.
|  *
|  * NOTE: The mount system call is handled differently between
|  * Linux and IRIX.  In IRIX we worked work with a binary data
|  * structure coming in across the syscall interface from user
|  * space (the mount userspace knows about each filesystem type
|  * and the set of valid options for it, and converts the users
|  * argument string into a binary structure _before_ making the
|  * system call), and the ABI issues that this implies.
|  *
|  * In Linux, we are passed a comma separated set of options;
|  * ie. a NULL terminated string of characters.  Userspace mount
|  * code does not have any knowledge of mount options expected by
|  * each filesystem type and so each filesystem parses its mount
|  * options in kernel space.
|  *
|  * For the Linux port, we kept this structure pretty much intact
|  * and use it internally (because the existing code groks it).
|  */
| struct xfs_mount_args {
| 	int	flags;		/* flags -> see XFSMNT_... macros below */
| 	int	flags2;		/* flags -> see XFSMNT2_... macros below */
| 	int	logbufs;	/* Number of log buffers, -1 to default */
| 	int	logbufsize;	/* Size of log buffers, -1 to default */
| 	char	fsname[MAXNAMELEN+1];	/* data device name */
| 	char	rtname[MAXNAMELEN+1];	/* realtime device filename */
| 	char	logname[MAXNAMELEN+1];	/* journal device filename */
| 	char	mtpt[MAXNAMELEN+1];	/* filesystem mount point */
| 	int	sunit;		/* stripe unit (BBs) */
| 	int	swidth;		/* stripe width (BBs), multiple of sunit */
| 	uchar_t iosizelog;	/* log2 of the preferred I/O size */
| 	int	ihashsize;	/* inode hash table size (buckets) */
| };

No... No compatibility layer there.