Re: fdclone KPI
Content-Type: text/plain; charset=us-ascii
On Wed, Oct 31, 2007 at 05:06:20PM +0200, Kostik Belousov wrote:[color=blue]
> Dear arch@ readers,
> with the important help from Peter Holm I have implemented the KPI that
> provides the ability for the driver to implement cloning on the open(2).
> This is another (IMHO, more UNIXy) way to provide per-fd private data
> for the driver. It seems that at least /dev/apm, /dev/drm and /dev/sg
> could immediately benefit from the fdclone() KPI.
> The patch is at
As Oleksandr Tymoshenko pointed out, the patch was not really available
at this address. Fixed, sorry.
> Sample dumb driver that uses the KPI is at
> The driver that uses the fdclone() shall provide cdevsw for master device,
> and cdevsw for clones. Master shall have d_fdopen() method that could call
> int fdclone(struct cdevsw *_csw, struct file *_fp, int _fmode,
> struct cdev **_clone, void *si_drv1, struct thread *td);
> to replace the reference in the _fp with newly created cdev.
> After successfull fdclone() call, all further calls on the _fp
> are dispatched to the clone _csw instead of master one. si_drv1 of the
> new cdev is set to respective argument, allowing the clone to find the
> The cloned cdev is not accessible for lookup through the devfs, and is
> destroyed automatically on the final close of the last filedescriptor that
> references the vnode.
> Please, review. Your feedback is welcome ![/color]
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.7 (FreeBSD)
-----END PGP SIGNATURE-----