cowloop technology

[Ivan Voras]

Thomas E. Zander wrote:

Well, it blocks raw device write if the reference count of this device
is not zero, i.e. you can't change the partition table of a disk in
multi-user mode if you have mounted filesystems on that disk.
It's a bit uncomfortable sometimes but I wouldn't go as far as to speak
of a particular problem. It reminds me that it's better to avoid some
things in multi-user mode :-)
As you said, it's better not to do it in multi-user mode, but it's 
actually a non-problem. AFAIK, a partition slicer could be implemented 
in a way that allows manipulating the partition table(s), but it would 
have to be done from inside the kernel (using the builtin userland <-> 
kernel communication facility for GEOM classes).

As for "cowloop", this is very similar what I did with "gjournal" 
(http://wikitest.freebsd.org/moin.cgi/gjournal). (note: this version 
gjournal is discintinued; future version(s) are in development).

GEOM is a *very* nice framework for this kind of thing.