I keep thinking, btw, that we should construct an "mfs v2" that
dynamically sizes itself and does not attempt to pretend to be a disk, 
but instead just allocates chains of memory in the obvious manner...

Perry

Ross Harvey writes:
> Anyone who wants to know just how the MFS works need only read mfs_doio() in:
> 
> 	src/sys/ufs/mfs/mfs_vnops.c.
> 
> It's a grand total of only _8_ lines, and it does all the MFS I/O by just
> calling copyin() or copyout() to the mount_mfs process.
> 
> And that process is just the newfs process: see line 172 in
> 
> 	src/sbin/newfs/mkfs.c
> 
> Which contains:
> 
> 	membase = malloc(fssize * sectorsize)
> 
> Yes, the entire MFS is just a thing returned by a user program calling
> malloc(3). The blocks will be created as they are touched.  They are never
> released, but for the most part don't exist until used.
> 
> 
>   -- Ross Harvey