back to the main page
Last updated 22 May 1995
If you would like to work on one of these, please contact
It's important that you let me know what's being
worked on, because someone else might also be interested, and you should
Items that we want done with more priority are marked with .
If you plan to work on the microkernel, you should be on the mailing list
for the University of Utah's Mach4 distribution; that is the microkernel
distribution that GNU will use. To get on the mailing list, send mail to
You should discuss proposed microkernel work there as well as
Mach 3.0 needs many new device drivers for popular PC devices.
A replacement for MiG that understood C .h files.
Bootstrap tools and documentation to help people set up Mach 3.0
machines if they already have Linux; if they already have Net BSD;
if they don't have anything.
Mach 3.0 needs to provide support for task virtual timers similar
in functionality to the Unix ITIMER_PROF and ITIMER_VIRTUAL timers.
Mach 3.0 needs to provide a way for users to do statistical PC
profiling similar to the Unix profil system call.
Mach 3.0 needs a facility to automatically send task and thread
status on task/thread exit to a port that can only be changed by
a privileged user; this would be used to implement process
Mach 3.0 needs a facility to find out what task is the parent of
a given task.
Mach 3.0 needs a facility to find out which pages of a task's
address space are in core to implement Unix's mincore call.
Mach 3.0 needs a facility to do msync.
Mach 3.0 needs a replacement for MEMORY_OBJECT_COPY_CALL that
works at least for the cases needed in ordinary files. (Write firstname.lastname@example.org if
you want to know what the problem is and some ideas about how to
Mach 3.0 needs proxy memory objects. (email@example.com can tell you what these
are and why they are important.)
Mach 3.0 needs a way to do per-task resource counters that are
accessible to servers called by the task.
Mach 3.0 should keep a timestamp of the creation of each task.
Mach 3.0 needs facilities to implement resource limits of various sorts.
Mach 3.0 needs a way to have a thread's CPU time statistics
include time spent by servers on its behalf. [This has been done
for the migrating-threads version of Mach; talk to firstname.lastname@example.org before starting
work on it.]
Of course, free ports are always necessary to machines that don't
already have free ports.
Much work can be done doing research in how to improve Mach VM
performance and timesharing scheduling policy.
We need a good malloc for multi-threaded programs that uses
vm_allocate. It should probably be based on the current GNU malloc,
as well as merge the mmalloc interfaces used by some existing GNU
Mach 3.0 needs facilities to get a list of all the devices which
can be device_open'd, as well as to get the type of a device.
A way to have the kernel send a message on some designated port
everytime a new task is started.
OSF has enhanced the exception_raise protocol to include thread_state
information. This code should be merged into the kernel; OSF people
have said their could could be released to the public (but it has not
These are brief descriptions; email@example.com
can give more information
A port of the Linux filesystem code to the Hurd. This will be
useful once the Hurd is in widespread use, to help people switch
from earlier Linux-based GNU systems to Hurd-based systems. It
might be useful even for Hurd development, since it might permit use
of early Hurd versions that lack network software--you would shut
down the Hurd and start up Linux in order to transfer files on the
Programs that use utmp need to be changed to use /utmp and utmp.defs.
We need some standard translators for /utmp nodes; most importantly
one for ordinary terminals (set up by login) and one for X displays
(set up by xdm).
We need some existing shell programs changed to do Hurd things:
like ls, su (see roland), tar, cpio, etc.
Some new programs need to be written: login (see roland), getty, ps, tools
for new filesystem features.
Shadow directory translators. (Roland has the beginnings of this.)
A system for write, send, talkd and so forth to bleep users;
this should be integrated with the utmp replacement above.
X. (Porting XFree86 should not be hard.)
A filesystem for /tmp that uses virtual memory instead of disk.
(Roland has some ideas about this.)
Filesystem implementations (using libdiskfs) for other popular
Transparent FTP translator.
NFS client implementation. You should start with BSD's 4.4 code
and support the extensions they support; don't worry about Hurd
extensions right now. (The server we want to write ourselves
because it will probably involve changing the Hurd interfaces.)
A fancy terminal driver that uses readline and supports detach/attach.
A notepad program to hold and keep track of ports for shell scripts.
C library work. See firstname.lastname@example.org if you are interested in working
on anything in the C library.
Useful response to SIGINFO.
Translations of socket SIOC* ioctls with hairy data formats into some
RPC protocol with socket servers.
Better integration with cthreads.
Shared libraries. Most of the work for ELF shared libraries has
already been done by Linux volunteers. What is needed is a clean
adaptation of the Linux code to load shared libraries at run time,
which could be used on several operating systems with different minimal
OS-specific filesystem-access modules. Before this can happen, libc
needs to be converted to use ELF, which is planned to happen soon but
is not happening yet.