> > > The following code gets moved into the kernel proper... > > > - mbuf's > > > - pools (although he has soemthing similar already so I'll have a > > > look at > > > thats > > > - sockets (including the main syscalls, i.e. accept, bind, > > > select, > > > listen > > > etc etc) > > > - sockbufs > > I remain in the position that there is absolutely no reason > > whatever > > that sockets need have their own syscalls: the current system works > > quite well and is vastly more extensible. > > Not that I would really object against you, but it has an advantage: > you can easily use networking functions from kernel drivers or file > systems; you can use that API from everywhere. A number of other things would also allow this: kernel modules or modifying the kernel so that drivers so actually use libraries (special kernel libraries, which would be kept in a seperate directory), which seems a much more worthwhile change than mucking about with kernel internals just for networking. -Nathan -- Fortune Cookie Says: Never settle with words what you can accomplish with a flame thrower.