2 Sep
2014
2 Sep
'14
9:57 p.m.
Mumble -- I would say that BSD portals was more in response to an MIT idea
that the Apollo guys (Jim Rees I believe) put in domain as the Apollo Typed
File System. At the time of the BSD, there was a lots of discussion going
on about how to move things out the kernel and a number of techniques were
considered -- Portals was one of them.
The idea of /dev/mumble/some_other_mumble was kicking around early UNIX for
a long time. I think I first encountered it in the MIT Chaos code, but
the Arpanet NCP code may have redated its use. The idea was simple, the
kernel's nami() function that was called by kernel open code (or any other
call the had to walk a pathname), left the pointer to the next undecoded
part of the path in the user's input buffer alone. So a hack was put in
the kernel that caused open to vector to needed helper code for the network
to do the rest of the pathname to get the parameters. The advantage of
this of course is open returns a real file descriptor.
BTW: One idea I had many years ago, and I remember talking to Dennis about
it at a couple of USENIX was similar to Apollo's typed scheme. As dmr put
it to me at the time, UNIX has a typed file system with very few types.
The question was if we opened up the typing scheme to allow for arbitrary
types, would you get more of mess or would the new feature be of an
advantage. I was not doing research at the time - I was employed at
Masscomp and just getting a distributed FS was enough. Since I did not
have a good use for the hack for EFS, I never tried to implement it. It
was always on my "TODO" list. Maybe when I retire and I can hack just for
the fun of it, I can try it ;-)
Clem
On Tue, Sep 2, 2014 at 3:15 PM, Aaron J. Grier <agrier(a)poofygoof.com> wrote:
On Fri, Aug 22, 2014 at 11:57:01AM -0400, John Cowan
wrote:
# filedes = open( "/dev/net/harv",2
);
# if( filedes < 0 )
# printf(" harvard is dead");
# else
# while( (nbytes=read(filedes,buf,80)) > 0 )
# write( 0,buf,nbytes );
If only this code still worked on modern Unixes! The socket API is
fine, but there really was no need to break good old open, at least
for client-side operations. Plan 9 got it right here, as usual.
was this the impetus behind mount_portal in 4.4BSD? It's still
available in NetBSD, although I've never played with it myself.
--
Aaron J. Grier | "Not your ordinary poofy goof." | agrier(a)poofygoof.com
_______________________________________________
TUHS mailing list
TUHS(a)minnie.tuhs.org
https://minnie.tuhs.org/mailman/listinfo/tuhs