From: Jeff Dike <jdike@ka...> - 2001-08-08 00:49:26
> I think something akin to BSD's unionfs would solve a lot of these
> problems. For those not familiar, unionfs let's you mount two
> filesystems at the same mount point. All changes go to one specific
> filesystem that overloads the other.
Yup, it would. But Linux doesn't have a unionfs.
OTOH, if UML were to be ported to *BSD, then this would work fine there.
<hint>And the port might not even be that hard</hint>
From: Jeff Dike <jdike@ka...> - 2001-08-09 04:08:51
> I 'intended' to start porting the code to use the native /proc
> filesystem, as an alternative to learning FreeBSD kernel hacking, but
> that inertia I mentioned kicked in about then.
> A 'native' port should also be possible, I would think...
Yeah, FreeBSD looks like a good port to try. I was talking about it with
someone on #kernelnewbies a while back who had a FreeBSD box, and the odd
Linux system calls had pretty close equivalents on FreeBSD. I was concerned
about clone + CLONE_FS, which is extremely handy and whose lack would be a
pain. FreeBSD has something equivalent (rproc?).
It doesn't have the equivalent of ptrace(PTRACE_SYSCALL, ...), but that's OK.
FreeBSD will segfault Linux 0x80 system calls, and the segfault handler can
do the system call virtualization.
Amusingly, this should run a lot faster than UML running on Linux.
So, if anyone else is interested and has a FreeBSD box, then this would be
a good project.