From: hargrove@theory.TC.Cornell.EDU (Paul H. Hargrove) Subject: Need testers for FIFO patches Date: Mon, 1 Jun 1992 23:16:10 GMT
I have a set of patches to the 0.96 and/or 0.96a kernels which add
FIFO special files to the kernel, and 'fix' the existing pipe code to
perform the atomic writes required by POSIX. Additional changes have
been made to allow non-root users to create FIFO special files with the
mknod() call, or the program by the same name.
The modified system seems pretty stable on my machine, and I am now asking
for people willing to try my FIFOs out on their own systems. I would, of
course, prefer people who have some application for FIFOs so I can get some
feedback on the standards adhearance and general usability of my FIFOs.
Once I can find three people who have applications requiring FIFOs who can
get the applications to run w/o significant modifications, I will post
the patches and send Linus a copy for possible inclusion in the next
release.
NOTE: The FIFOs implement, to the best of my knowledge, the POSIX-required
behavior. The brain-dead O_NDELAYisms of SysV are not implemented, so
applications written for SysV which open the FIFOs with O_NDELAY will
probably not work.
The ability to open a FIFO with O_RDWR (to force read() to block when there
is no process writing the FIFO) is supported, so servers/daemons which
use a FIFO as a request queue will work.
My thanks go to Brandon Allbery and Leslie Mikesell for bringing the
atomic write requirement to my attention, and to Bruce Evans for making
the need for O_RDWR support clear to me.
Thanks also go to Jim Burns who has agreed to be my first guinea pig.
-- Paul H. Hargrove hargrove@theory.tc.cornell.edu "A witty saying proves nothing." --Voltaire