On Sunday 09 March 2003 08:14, Christoph Badura wrote:
> >I'm either being unclear or dense.  What I was getting at was the idea of
> >having mount simply look for the mount process for any particular mount
> > and send that process the signal.  This may mean storing the PID o fthe
> > mount process in the kernel structure but that's all.
> >
> >Now, this is based on the (perhaps erroneous) assumption that each mount
> >command causes a unique process to run and manage the mount point.  Is
> > this an incorrect assumption?
>
> Yes, this is an incorrect assumption.  From the file systems I have used
> only MFS uses a process to "manage" the mount point.  And the mount_mfs
> process doesn't really manage anything, because it is just a hack to get
> some swap- backed VM space.

OK then I guess we are back to storing the FS type in the mount structure, 
right?  So does someone want to tackle src/sys/kern/vfs_syscalls.c.  Can we 
agree on what generally needs to be done?

1.  Change src/sys/mount.h to have the mount struct hold a (16 byte?) copy of 
the FS type.
2.  Change sys_mount() to store the FS type in that area.
3.  Change umount to get the FS (sys_fstatfs?) and call umount_${fs} if it 
exists.

-- 
D'Arcy J.M. Cain <darcy@netbsd.org>
http://www.NetBSD.org/