The arrow of time

Ivan Voras' blog

Home · My wish list for FreeBSD

My wish list for FreeBSD

In a perfect world, where I somehow manage to have millions of $currency, I'd sponsor the development of:

  • Host-side virtualization for FreeBSD: either qemu / kvm, virtualbox, vmware or xen-dom0. This is the first and "hot" item on this list - there are many opportunities where I could use FreeBSD as a host for other operating systems instead of reverse (I don't care for "guest" support like xen dom-U since full virtualization solution exist and are used almost everywhere. Well, except Amazon EC2 and such.) One of the best features (for me at least) of virtualization like this is the possibility to move virtual machines between (different) host operating systems. Jails are of very limited usefulness in this scenario.
  • Finalization and inclusion of the FUSE kernel module or port of NetBSD's puffs in the main kernel, where it would be better supported and updated with the rest of the system. FUSE allows for very easy experimentation with file systems, which are my pet interest, but more importantly it enables the usage of many already made FUSE file systems like ntfs3g and various network file systems and emulations of thereof (sshfs, webdav).
  • Porting of either ext3, ext4, JFS or XFS to FreeBSD. Though ext3/4 can be mounted as ext2, journalling is lost in this case. XFS kind of exists but was never finished (write support is missing). For me, the ability to share file systems between FreeBSD and Linux (usually with VMs) is almost as important as having a well known low-maintainance (and low-resource using, as opposed to ZFS) journalling file system.

There will probably be more items on the list, stay tuned.

I'm surprised that to this day there isn't a central sponsorship project or site for features such as these. I know there are some low-profile sponsorship sites:

but they seem either stalled or have a too small user- / sponsor- base to make a difference. To be fair,  I was also thinking about creating or participating in such a site (SponsorBSD in fact) but nothing came out of this - I couldn't do much for SponsorBSD for some reason - I think they decided on a web framework I was completely unfamiliar with, and since the country I'm in is backwards enough, I can't actually handle any money in a way that's suitable for sponsorships without major problems.

Of course, there's FreeBSD Foundation, which actually has nontrivial funds, and which does sponsor relatively big projects, but the projects sponsored always seem to be for very low-level kernel features for which I'm not sure there is large enough user base.

On the other hand, is the project really big enough to sustain serious user-sponsored development? Is any project big enough for that? A very interesting open-source project is Mozilla - which has a whopping $75 million backing, most of which ($66 milllion) comes from Google. Compared to this, the FreeBSD Foundation's goal of $300,000 for 2008 looks practically insignificant (and it's a completely different order of complexity to develop an operating system than a browser, no matter what they say).

Posted on 2008-12-04T14:28 by ivoras

#1 Cant agree with you more.

Added on 2008-12-04T17:13 by Prudhvi Krishna Surapaneni

I can't agree more.

#2 ZFS

Added on 2008-12-04T17:19 by Tomasz

For sharing files you can use ZFS. ZFS-fuse on linux works fairly well.

#3 Re: ZFS

Added on 2008-12-04T21:18 by Ivan Voras

True, ZFS sort of works on Linux via ZFS-fuse so it's a possibility. It still leaves the issue of its high resource consumption - I can't really assign 4 GB of memory per VM just to make its file system happy :)

(Note: the "4 GB" in the last sentence is an exaggeration but not a big one).

#4 Flash player / Ati & Nvidia drivers

Added on 2008-12-04T23:06 by Haris

... there is the deskop side, ofcourse..

- Native adobe flash player for FreeBSD

- Ati X11 drivers for x86/amd64

- Nvidia X11 drivers for amd64

#5 onefs

Added on 2008-12-05T01:04 by mike

anyone else read something about a patchset from:

http://www.isilon.com/products/index.php?page=software_OneFS

being given back to freebsd?

#6 fuse

Added on 2008-12-05T03:19 by ooga

fuse works failry well on my system ... shouldn't be in the kernel tree should it?  puffs maybe could be there though ...

I'd like a easy rock solid secure SSO network login alternative to ActiveDirectory :-\ OpenLDAP is too much (idle it uses 400mb on our system) . Heimdal rox for authentication (using 1.2.2 here) but what about directory info ... NIS can work with kerberos for auth but could be made more secure and easier to admin (RPC firewall headaches)

MS added authorization packets to the kerberos protocol maybe BSD's need to work with Apple and Linux toward an easier to run directory info storage system designed from the gett go to work with kerberos.

#7 Mozilla Foundation dough

Added on 2008-12-05T03:42 by webguy

Yahoo uses a lot of freebsd - they sort of even have their own version (yBSD) .. why not enter into a relations ship with them? Even a tiny bit of ad revenue from Yahoo would be nice - along with say some engineering hours and a few bits of hardware ;-)

Sure there's more ad dollars to be had from refering the searches from "desktop installs" of browsers onto Windows to Google than there is from refering searches from a niche operating system desktop (case in point: Fedora does almost the same thing but gets way less money) ... but how about a few ads on search results pages or money for any search using Yahoo for FreeBSD info or something ...

OK the above is all rot ... Yahoo should just cough up I million dollars and *give* it to the foundation as an endowment.

#8 Yahoo

Added on 2008-12-05T12:28 by Ivan Voras

Yahoo actually does sponsor FreeBSD quite a lot - not in the "millions" range, but it hosts a large part of the FreeBSD development server infrastructure and employs some developers.

#9 Network login

Added on 2008-12-05T16:45 by Sysadhominem

I like the "improve NIS" comment :) 

It's true that Kerberos is a great, robust and actually fairly simple system to setup for network authentication. I think Microsoft's addition (shoehorning)  of "group information" into the protocol  is not a good idea and I wouldn't want Heimdal or MIT to follow this. KRB is for *authentication* and the addition of really rudimentary info like "group" membership"  munges authentication and authorization for no big gain. What can you do that is all that sophisticated with this group info anyway? You still need AD to designate what a particular group can do and where etc etc.

A better simpler approach would be to build on the *excellent* foundation of FreeBSD's login/login.conf facility and somehow make that "network aware". On a network you could have one or more hosts be the network_login "servers" refering authentication to kerberos or SASL or whatever.Not sure if kerberos tickets can already be issued in the form:

user/[loginclass]@REALMNET.NET

which would be neat for restricting/enforcing what people do on different machines but even without that the "network aware" login could read it's machine specific login.conf and do the work like BSD does now. Caching a login readable file for use via cgetent(3) and all the normal BSD login facilities would be easy and the config files could be distributed from a cvs via dhcp or accessed o nthe network during the login phase (sort of like way you can have a static resolv.conf that refers to other DNS servers and/or is built by dhcp). Essentially this could be like NIS but with transport over encrypted tcp/udp and no need for a separate set of tools for administration.

Things could be made simple enough for other BSD's to adopt and even for interoperation with other unices ... maybe the network aware login_conf is just a login conf aware NIS server that communicates via a dedicated port and the normal ypbind clients can connect to it without noticing. Or maybe new "source" entry for nsswitch.conf
comes into existence: login_conf.  The first entry in /etc/login.conf class could be:

default:\
       :network_login_host=: 192.168.2.1,192168,2,4: \   
       :passwd_format=md5:\
       :copyright=/etc/COPYRIGHT:\

or something.

There've been notable attempts to improve NIS (e.g. MIDAS from the university of Manitoba) but few recent attempts to NIS-ify or make network aware basic Unix system level authorization and user management since Hesiod (unless you count libnss-sqlite libnss-mysql etc.)

http://libnss-sqlite.tuxfamily.org/doku.php
http://libnss-mysql.sourceforge.net/purpose.shtml

#10 Suspend&resume

Added on 2008-12-05T17:28 by Sec

More than all of  this, I would like to see working suspend&resume (both to ram and to disk). Thats really making FreeBSD on laptops a non-option :-(

#11 @Network login comment

Added on 2008-12-05T17:44 by webguy

shouln't you do everything you just said with PAM? i.e. you make /etc/pam.d/login use kerboros for authentication and point all the user/host/account/network drectory look up to either be local to the machine (i.e. controlled login.conf and files) and/or pulled from NIS/LDAP/MySQL by changing nsswitch.conf. Am I missing something? sqlite is cool but it's not a "network aware" database AFAIK.

I guess this stuff could be better documented or made easier to switch on or off but I'm not sure login.conf is the place to do that. We do adjust the login class of all "not in front of  console" users to be a bit more strict and use netgroups(5) to limit which users get to login where and control access to each system's root/admin accounts with ksu and per machine /root/.k5login.d/* and /root/.k5users files. I think ksudo would be nicer (for logging etc) :)

#12 sudo

Added on 2008-12-06T02:38 by sun junkie

ksudo ?? :-)  does such a beast exist?

One thing about sudo is that is sort of substitutes for a "real" approach RBAC (like on solaris).  With all the mac_ kernel modules FreeBSD seems very close to actually having RBAC though.

That'd be my wish although sudo is pretty decent and simple substitute.

 

#13 FreeBSD: desktop / server / GSoC

Added on 2008-12-19T02:23 by mato

On desktop side -- Mr.Scott said long time ago that 6-current will be targeted at mobile users.  Unfortunately we are still without working suspend/resume (to disk) and easier support for multihoming and waking up in different networks.

On server side I agree with you that not having a host side virtualisation solution is harming and excluding FreeBSD in too many cases/scenarios.
I don't think ext3/4 is worth implementing, we can use FUSE, however JFS or XFS would be really welcome as they are modern journalling yet not resource intensive like ZFS.

Another thing is that while FreeBSD has rather many GSoC entries each year I do not see many of them to get incorporated into the official system.

#14 KVM port

Added on 2008-12-22T19:56 by Cane

KVM was ported during as a"Google Summer of Code 2007" project.

Here's the site:
http://feanor.sssup.it/~fabio/freebsd/lkvm/

#15 Re: KVM port

Added on 2008-12-22T20:18 by Ivan Voras

The KVM port would have to be integrated into the FreeBSD tree in order for it to be useful to any significant number of users.

Post your comment here!

Your name:
Comment title:
Text:
Type "xxx" here:

Comments are subject to moderation and will be deleted if deemed inappropriate. All content is © Ivan Voras. Comments are owned by their authors (who agree to basically surrender all rights by publishing them :) )..