[GSoC 2017] Original proposal: Port kernel Lua to FreeBSD

[Warner Losh]

On Mon, Feb 27, 2017 at 8:26 AM, Allan Jude <allanjude at freebsd.org> wrote:
> On February 27, 2017 5:28:41 AM PST, Saurav Sachidanand <sauravsachidanand at gmail.com> wrote:
>>Hello FreeBSD community,
>>
>>I'm
>> Saurav Sachidanand, and I'm
>> a CS sophomore studying in India
>>.
>>I have an interest in operating systems development and wish to
>>contribute
>>to the FreeBSD community. I'm proficient with C and have some
>>experience in
>>kernel programming. Hence, I'd like to propose an original project for
>>GSoC
>>2017 that I feel would benefit this community.
>>
>>In past years, the Lua interpreter was ported to run inside the Linux
>>and
>>NetBSD kernel [1]. Lua was chosen because it's interpreter is very
>>small (~240
>>KB) compared to that of Python or Ruby, it's MIT licensed, and is
>>almost
>>freestanding. A working demonstration of it is a packet filtering
>>algorithm
>>written entirely in kernel Lua [2].
>>
>>Specifically, my proposal would be to port the following that are
>>currently
>>written for NetBSD:
>>- the modified Lua VM source code with _KERNEL preprocessor directives
>>to
>>exclude user-space functionality like floating point, the io and os
>>module
>>in the standard library, etc. [3]
>>- the kernel module device driver for /dev/lua, to which Lua scripts
>>are
>>fed to be executed [4], [5]
>>- the luactl user-space program to control the Lua device and a couple
>>of
>>sysctl variables which serve similar purpose [6], [7]
>>
>>And then:
>>- run the Lua test suite targeting whatever we support in the kernel to
>>make sure it works [8]
>>- and write Lua bindings to the kernel interfaces that would interest
>>the
>>FreeBSD community
>>
>>Since NetBSD and FreeBSD have similar kernel interfaces (mutexes,
>>linked
>>lists, device switch interface), the porting shouldn't involve too much
>>code refactoring. Also, this would all be an experiment in that we
>>don't
>>fully know what the real world use cases might be, but it would attract
>>more people to writing kernel code who otherwise wouldn't because of
>>having
>>to do everything in C. And it would be interesting to carry out it out
>>in
>>FreeBSD as well since it has a larger community than NetBSD.
>>
>>I humbly request anyone who is interested in this project to be my
>>potential mentor(s) for GSoC.
>>
>>More slides on kernel Lua in NetBSD - [9], [10].
>>
>>Thanks,
>>Saurav
>>
>>[1] - http://www.netbsd.org/~lneto/dls14.pdf
>>[2] - https://www.netbsd.org/~lneto/eurobsdcon14.pdf
>>[3] - https://github.com/jsonn/src/tree/trunk/external/mit/lua/dist/src
>>[4] -
>>https://github.com/IIJ-NetBSD/netbsd-src/tree/master/sys/modules/lua
>>[5] -
>>https://github.com/IIJ-NetBSD/netbsd-src/tree/master/sys/modules/luasystm
>>[6] - https://github.com/IIJ-NetBSD/netbsd-src/tree/master/sbin/luactl
>>[7] - http://netbsd.gw.com/cgi-bin/man-cgi?lua+4+NetBSD-current
>>[8] - http://www.lua.org/tests/
>>[9] -
>>https://www.netbsd.org/gallery/presentations/mbalmer/fosdem2012/kernel_mode_lua.pdf
>>[10] - https://www.lua.org/wshop13/Cormack.pdf
>>_______________________________________________
>>freebsd-hackers at freebsd.org mailing list
>>https://lists.freebsd.org/mailman/listinfo/freebsd-hackers
>>To unsubscribe, send any mail to
>>"freebsd-hackers-unsubscribe at freebsd.org"
>
> This may be quite a nice thing to have. Another upcoming use for LUA in the kernel is ZFS Channel Programs. These allow a number of ZFS operations to be completed as a single atomic transaction.
>
> I would hope we could structure this in such a way as to not end up with two copies of Lua in the kernel.

There's also a 3/4 finished lua in the boot loader that you might be
able to leverage as well....

Warner