Possible solution to max FD probs...

Gregory Maxwell (nullc@nightshade.ml.org)
Mon, 6 Jan 1997 18:33:53 -0500 (EST)

<warning> I may not know what the heck I'm talking about.</warning>

Hmm.. Would this work:
Add new kernel funct(s) that would be able to look at a processes
fd table as a list of references to lists... (which could be allocated
elseware..). Make the old interface a wrapper on the new one.... Programs
can only get FD>256 if they either A) directly use the new function or B)
make a syscall that lets the old function use fd>256... The magic syscall
could be called as the program is forked so that compatible programs could
be run with fd>256 without mod...

Using that method would make FD virtualy limitless (only limited by
performance, memory, and a sane system config) on a process by process
basis... Just cause StupidServer(tm) wants 2048 fds that doesn't mean all
my proggies need that..