FreeBSD Status Report January-March, 2010

FreeBSD Quarterly Status Report


This report covers FreeBSD related projects between January and March
2010. Being the first of the four reports planned for 2010 with 46
entries, it shows a good progress of the FreeBSD Project and proves
that our committers are keeping up with the latest trends in the OS
development. During this period, a new minor version of FreeBSD,
7.3-RELEASE, has been released, while the release process for
8.1-RELEASE is soon to begin and is planned to be released later this

Thanks to all the reporters for their excellent work! We hope you enjoy
the reading.

Please note that the deadline for submissions covering the period
between April and June 2010 is July 15th, 2010.

Google Summer of Code

* Google Summer of Code 2010


* Chromium web browser
* Clang replacing GCC in the base system
* EFI support for FreeBSD/i386
* mfsBSD
* Modular Congestion Control
* NAND Flash framework for embedded FreeBSD
* Out of Tree Toolchain
* PC-BSD PC-SysInstall Backend
* The tbemd branch
* webcamd

FreeBSD Team Reports

* FreeBSD Bugbusting Team
* Release Engineering Team
* The FreeBSD Foundation

Network Infrastructure

* (Virtual) Network Stack resource cleanup
* 802.11n support
* Atheros AR9285 support
* Enhancing the FreeBSD TCP Implementation
* Experimental NFS subsystem (NFSv4)
* ipfw and dummynet enhancements
* net80211 rate control framework
* TCP/UDP connection groups


* CAM-based ATA implementation
* Dynamic Ticks in FreeBSD
* geom_sched
* IPv6 without legacy IP kernel
* Multichannel playback in HDA sound driver (snd_hda)
* Rewrite of FreeBSD read/write path using vnode page
* SUJ: Journaled Softupdates


* The FreeBSD German Documentation Project
* The FreeBSD Hungarian Documentation Project

Userland Programs

* FreeBSD port for libunwind
* LDAP support in base system


* FreeBSD/arm port for TI DaVinci
* FreeBSD/ia64
* FreeBSD/mips on D-Link DIR-320
* FreeBSD/powerpc
* FreeBSD/powerpc64 port
* FreeBSD/sparc64


* Portmaster
* Ports Collection


* BSDCan 2010 -- The BSD Conference
* meetBSD 2010 -- The BSD Conference

(Virtual) Network Stack resource cleanup

Contact: Bjoern A. Zeeb <bz@xxxxxxxxxxx>

In February work was done to address resource leaks in the (virtual)
network stack, especially on teardown.

During that time also multiple general run-time problems and leaks were
identified and fixed including leaked ipfw tables on module unload,
routing entries leaked, in case of interfaces going away, as well as
leaked link-layer entries in interaction with flowtable and timers.

For virtual network stacks resources are are no longer allocated
multiple times or freed upon teardown for eventhandlers, IP and upper
level layers, like TCP syncache and host cache, flowtable, and
especially radix/routing table memory.
In addition epair(4) was enhanced and debugging was improved.

This work was sponsored by ISPsystem.

Open tasks:

1. Merge the remaining patches.
2. Work on a better teardown model and get to the point where we can
free UMA zones without keeping pages for type stability and timers

802.11n support

Contact: Rui Paulo <rpaulo@xxxxxxxxxxx>

802.11n support in the Atheros driver is being worked on. Right now it
can do AMPDU RX in software and we are working on TX AMPDU. The code
lives in a private Perforce branch, but some bits of it are already
committed to HEAD.

This work is being sponsored by iXsystems, inc.

Atheros AR9285 support

Contact: Rui Paulo <rpaulo@xxxxxxxxxxx>

Atheros AR9285 support was added to FreeBSD HEAD and 8-STABLE. There
are still some issues but in general it works fine.

BSDCan 2010 -- The BSD Conference


Contact: BSDCan Information <info@xxxxxxxxxx>

BSDCan, a BSD conference held in Ottawa, Canada, has quickly
established itself as the technical conference for people working on
and with 4.4BSD based operating systems and related projects. The
organizers have found a fantastic formula that appeals to a wide range
of people from extreme novices to advanced developers.

BSDCan 2010 will be held on 13-14 May 2010 at the University of Ottawa,
and will be preceded by two days of Tutorials on 11-12 May 2010.

There will be related events (of a social nature, for the most part) on
the day before and after the conference.

Please check the conference web site for more information.

CAM-based ATA implementation

Contact: Alexander Motin <mav@xxxxxxxxxxx>

Work on CAM-based ATA implementation continues. Since last report
handling of heavy errors and timeouts was improved, Hot-plug now works
for both Host and Port Multiplier ports. Series of changes were made to
CAM to fix some old issues and honor some new ATA demands.

New drivers ahci(4) and siis(4) got some fixes and are quite stable
now. "options ATA_CAM" kernel option shows good results in supporting
other controllers using existing ata(4) drivers, so it is possible to
start deprecating old ata(4) APIs now.

Started work on new Marvell SATA driver for both PCI-X/PCIe cards and
ARM System-on-Chip SATA controllers. It is expected to support NCQ,
Port Multipliers with FIS-based switching and other new features.

Most of the code is present in 8-STABLE.

Open tasks:

1. Port ataraid(4) functionality to GEOM module.
2. Write SAS-specific transport and drivers for SAS HBAs (specs
wanted). SAS controllers can support SATA devices and multipliers,
so it should fit nicely into new infrastructure.

Chromium web browser


Contact: sprewell <chromium@xxxxxxxxxxx>

Chromium is a Webkit-based web browser that is mostly BSD licensed. It
works very well on FreeBSD and even supports new features like HTML 5
video. I have started offering subscriptions to fund the porting effort
to FreeBSD, funding which has already paid to fix Chromium on BSD-i386.
I am using a new funding model where subscriptions pay for development
that is kept closed for at most 1 year, after which all patches used in
a build are released to subscribers under the same BSD license as
Chromium. Also, parts of the closed patches are continually pushed
upstream, the BSD i386 fix has already been committed upstream. The
goal is to fund Chromium development on BSD while continually pushing
patches back to the BSD-licensed Chromium project. I will spin off a
Chromium port for ports soon, for those who do not mind using an older,
stable build that does not have all the paid features in the subscriber
builds. You can read about the issues that a subscription would pay
for, such as replacing the ALSA audio backend with OSS, and find out
more about subscribing.

Clang replacing GCC in the base system


Contact: Ed Schouten <ed@xxxxxxxxxxx>
Contact: Roman Divacky <rdivacky@xxxxxxxxxxx>
Contact: Brooks Davis <brooks@xxxxxxxxxxx>
Contact: Pawel Worach <pawel.worach@xxxxxxxxx>

Since the last status report we got to the state where we are able to
build all of FreeBSD (the C and C++ bits) on i386/amd64 with clang. The
only exception is the bootloader which does not fit within the given
size constraint. This is where the current efforts are going on. The
C++ part got a big boost now being able to compile all C++ code in
FreeBSD and itself.

We saw some movment on Mips and PowerPC. Mips got its driver
definitions from Oleksander Tymoshenko and Nathan Whitehorn did the
same for PowerPC and tested the kernel. Currently, the PPC kernel seems
to boot but due to lack of va_arg implementation for PowerPC nothing is
printed out. Nathan is working on that.

Overall ClangBSD is selfhosting on i386/amd64 and some progress has
been made on PowerPC/PPC. We also saw some contribution to the Sparc64
but this seems to have stalled.

We need people to try out ClangBSD (see the wiki) and runtime test it.
We also would appreciate help with other archs - namely ARM.

Open tasks:

1. Runtime test ClangBSD on amd64/i386.
2. Help with ARM/Mips/Sparc64.
3. More testing of clang on 3rd party apps (ports).
4. Discussion on integrating LLVM/clang into FreeBSD.

Dynamic Ticks in FreeBSD


Contact: Tsuyoshi Ozawa <ozawa@xxxxxxxxx>

I wrote experimental code (please see my project page) and threw patch
( ) to freebsd-hackers. A lot of FreeBSD
hackers gave me precious advice, so I am going to reflect it as a next

Open tasks:

1. Run hard/stat/prof-clocks irregularly (in progress).
2. Some timers which are added after the kernel's scheduling next
timer interrupt may be ignored (BUG).
3. Make callout queue have the tick when the next timer event rise up.

EFI support for FreeBSD/i386

Contact: Rui Paulo <rpaulo@xxxxxxxxxxx>

Work on supporting EFI booting on FreeBSD/i386 resumed. The boot loader
can now read an ELF file from the EFI FAT partition. We are now working
on trying to boot a kernel.

Enhancing the FreeBSD TCP Implementation


Contact: Lawrence Stewart <lstewart@xxxxxxxxxxx>

The ALQ(9) implementation and KPI has been rototilled and modified (one
more patch needs to be committed) to support variable length messages.
In addition, it can now be compiled and loaded as a kernel module.

With the ALQ changes in head, SIFTR can finally be imported.

Reassembly queue autotuning is in the project branch and needs to be
extracted as a patch people can easily test.

Open tasks:

1. Solicit external testing for and commit SIFTR.
2. Solicit external testing for and commit reassembly queue autotuning

Experimental NFS subsystem (NFSv4)

Contact: Rick Macklem <rmacklem@xxxxxxxxxxx>

Although the bare bones of the NFS Version 4 support was released in
FreeBSD 8.0, the integration has been progressing slowly and support
should be functional for FreeBSD 8.1 for RFC3530 (NFS Version 4.0).

Post FreeBSD 8.1, I believe the focus will be on code cleanup and,
under a projects area of svn, some experimental work on aggressive
whole file caching to client disk.

Open tasks:

1. Handling of delegations on the server w.r.t. local processes
running on the server.
2. Integration of recent changes to the regular NFS client, such as
Dtrace support.

FreeBSD Bugbusting Team


Contact: Gavin Atkinson <gavin@xxxxxxxxxxx>
Contact: Mark Linimon <linimon@xxxxxxxxxxx>
Contact: Remko Lodder <remko@xxxxxxxxxxx>
Contact: Volker Werth <vwe@xxxxxxxxxxx>

Bruce Cran (brucec) has graduated from GNATS-only access to having a
src commit bit. He has been making commits to help us catch up with the
PR backlog. Thanks!

We continue to classify PRs as they arrive, adding 'tags' to the
subject lines corresponding to the kernel subsystem involved, or man
page references for userland PRs. These tags, in turn, produce lists of
PRs sorted both by tag and by manpage. The most recent use of these
tags is the creating of a new report, Summary Chart of PRs With Tags,
which sorts tagged PRs into logical groups such as filesystem, network
drivers, libraries, and so forth. The slice labels are clickable. The
chart is updated once a day. You can consider it as a prototype for
browsing "sub-categories" of kernel PRs.

The "recommended list" has been split up into "non-trivial PRs which
need committer evaluation" and the "easy list" of trivial PRs, to try
to focus some attention on the latter.

New reports were added for "PRs which are from FreeBSD vendors or
OEMs", "PRs containing code for new device drivers", and "PRs
referencing other BSDs". These will primarily be of interest to

Some other bitrot on the "experimental PR reports" pages has been

It is now possible for interested parties to be emailed a weekly,
customized, report along the lines of the above. If you are interested
in setting one up, contact linimon@xxxxxxxxxxxx

The overall PR count has recently jumped to around 6400. This may be
due to increasing uptake of FreeBSD 8.

Our clearance rate of PRs, especially in kern and bin, seems to be

Mark Linimon polled various committers about their interest in specific
PRs. As a result, the AssigningPRs page on the wiki and the
src/MAINTAINERS file were updated based on feedback.

As always, anybody interested in helping out with the PR queue is
welcome to join us in #freebsd-bugbusters on EFnet. We are always
looking for additional help, whether your interests lie in triaging
incoming PRs, generating patches to resolve existing problems, or
simply helping with the database housekeeping (identifying duplicate
PRs, ones that have already been resolved, etc). This is a great way of
getting more involved with FreeBSD!

Open tasks:

1. We will be having a bugbusting session at BSDCan. If you are
developer who will be attending the conference, please stop by.
2. try to find ways to get more committers helping us with closing PRs
that the team has already analyzed.

FreeBSD port for libunwind


Contact: Konstantin Belousov <kib@xxxxxxxxxxx>

The alpha version of libunwind library port for FreeBSD x86 and x86_64
is completed and imported into the official libunwind git repository.
Libunwind is the library to perform dynamic unwinding of stacks, using
dwarf call frame information. The library features remote unwinding
using ptrace(2), very fast setjmp(3) implementation and more
interesting features.

FreeBSD/arm port for TI DaVinci


Contact: Jakub Klama <jceel@xxxxxxxxxxxx>

DaVinci (TMS320DM644x) is an ARM9-based system-on-chip family from
Texas Instruments with built-in DSP core and powerful multimedia/video
features. This work is bringing support for FreeBSD on these systems -
it works in multiuser mode, using root filesystem mounted either via
NFS or from SD/MMC card. The code is available in P4 at

Current DaVinci support includes:
* Booting from U-Boot bootloader
* Serial console
* Interrupt controller
* Integrated timers
* Power and sleep controller
* 10/100 Ethernet controller
* SD/MMC controller

Open tasks:

1. Remaining built-in peripherals drivers (USB, ATA, NAND flash, I2C,
DMA engine, sound, video input/output).
2. Framework for communicating with DSP core.


Contact: Mark Linimon <linimon@xxxxxxxxxxx>

The stability of the machines under package build has been improved by
a number of recent commits. Some rework is underway to run with
WITNESS. However, we are still limited in the number of simultaneous
packages that can be built.

Based on this, we have completed the first full ia64-8 package build.
17187 were built (as compared to 19885 on a recent i386-8.) Mark
Linimon has gone through the results to denote which packages do not
build. A few fixes have already been committed based on this.

We currently have 3 available machines that are stable enough for
package builds.

Support for the SGI Altix 350 has made its start. Porting is done on 2
SGI Altix 350 machines connected with NUMAFlex, giving a total of 4
CPUs and 24GB of DDR. The kernel boots with code on the projects/altix
branch but since ACPI does not enumerate PCI busses, no hardware
devices are found. SMP has been disabled because waking up the APs
result in a machine check.

Open tasks:

1. Continue to try to understand why multiple simultaneous package
builds bring the machines down.
2. Upgrade the firmware on the two machines at Yahoo! to see if that
helps the problem.
3. Figure out why the fourth machine is not stable.
4. Configure a fifth machine that has been made available to us.
5. Figure out the problems with the latest gcc port.
6. We need documentation about the SGI SAL implementation to speed up
porting to the SGI Altix 350.
7. The loader and kernel need to change to allow the kernel to be
loaded at a runtime-determined physical address as well as add
support for NUMA.

FreeBSD/mips on D-Link DIR-320


Contact: Alexandr Rybalko <ray@xxxxxxxx>

FreeBSD/mips has been ported to D-Link DIR-320, wireless router based
on BCM5354 SoC. Project aims to providing several working images
tailored for different purposes (profiles). So far racoon based
router-ipsec image is available.

Open tasks:

1. bfeswitch configuration utility.
2. Add router profile.
3. Add wifi-router profile.
4. Add openvpn-router profile.

FreeBSD/mips on Octeon


Contact: Juli Mallett <jmallett@xxxxxxxxxxx>

Significant progress has been made in terms of stabilizing the
uniprocessor Octeon port and adding support for MIPS ABIs other than
o32 in the toolchain, rtld, libc and the kernel. Kernels built to the
n32 ABI are currently supported with changes that will not be merged
because they make invasive changes throughout the system with regard to
db_expr_t and register_t, which are larger than a pointer in the n32
ABI. Once support for n64 kernels is completed (including the ability
to run n32 worlds) and the n32 hacks are removed, the branch will be
suitable for merging. Many nearby cleanups have occurred, particularly
in the area of TLB and pmap code.

Open tasks:

1. An import of select pieces of the Cavium simple executive as vendor
code is planned to make it possible to remove locally-maintained
copies of Cavium headers and shim functions, many of which are
vastly outdated.
2. The Linux opencrypto port contains an opencrypto driver for the
cryptographic coprocessor which look relatively easy to port.
3. Support for SMP is a high-priority item that will be addressed
after the 64-bit changes are stabilized.
4. PCI and USB bus and device support is planned to follow the import
of the simple executive functions and headers.
5. The rgmx ethernet driver currently copies packets in and out of
mbufs rather than putting pointers to mbuf storage into hardware,
which results in bad network performance.


Contact: Nathan Whitehorn <nwhitehorn@xxxxxxxxxxx>

An Apple XServe G5 has been donated by Peter Grehan for package
building. Based on the last two months' worth of testing, a large
number of commits have been made to increase stability.

We have completed the first full powerpc-8 package build. Only 10918
were built (as compared to 19885 on a recent i386-8), primarily due to
a few high-impact packages failing (such as lang/python25). Mark
Linimon has gone through the results to denote which packages do not
build. A few fixes have already been committed based on this; we have
patches that are being tested in the next run.

Mark Linimon is working on getting us more XServes.

Open tasks:

1. Start the hard work of fixing individual packages.

FreeBSD/powerpc64 port

Contact: Nathan Whitehorn <nwhitehorn@xxxxxxxxxxx>

A full 64-bit PowerPC port of FreeBSD is now complete, and should
shortly be merged to HEAD, likely first appearing in FreeBSD 9.0. This
port supports SLB-based 64-bit server CPUs, such as the IBM POWER4-7,
PowerPC 970 (G5), and Cell Broadband Engine. Current machine support is
limited to Apple single and dual processor G5 systems, with future
support planned for IBM Power Systems servers and the Sony PlayStation


Contact: Marius Strobl <marius@xxxxxxxxxxx>
* Yet another bug causing unaligned accesses in NFS server operation
has been found and fixed in FreeBSD 7 and 8. Unlike as announced in
the last Status Report, no Erratum Notices regarding these problems
have been issued as it quickly became obvious that dealing with so
many of them is impractical, especially since the fixes unveiled
secondary bugs.
* Alexander Motin has fixed several bugs in netgraph(4) nodes in
9.0-CURRENT which also caused unaligned accesses, so these should
work now on sparc64.
* Peter Jeremy has contributed several fixes for the sparc64 FPU
emulation code, which now passes a test suite built around
TestFloat. These fixes were incorporated into FreeBSD 6, 7 and 8
but unfortunately did not quite make it into 7.3-RELEASE but will
be present in 8.1-RELEASE and 7.4-RELEASE.
* Support for UltraSPARC-IV and -IV+ CPUs has been added and will be
present in 8.1-RELEASE and 7.4-RELEASE. Thus Sun Fire V890 is now
supported and stable, though due to the lack of properly working
test hardware, not with configurations consisting of a mix of US-IV
and -IV+ CPUs. However, performance is not yet where it should be,
i.e. a buildworld on a 4x1.5GHz US-IV+ Sun Fire V890 takes nearly 3
hours while on a Sun Fire V440 with (theoretically) less powerful
4x1.5GHz US-IIIi CPUs it takes just over 1 hour. So far it is
unclear what is causing this, it might have to with what appears to
be a silicon bug of US-IV+ CPUs encountered and worked around while
adding support for these.
* Work on getting Sun Fire V1280 supported has been continued. A
third firmware bug has been worked around and a driver for the
BootBus controller, which provides console and time-of-day services
in these machines, has been written. It is now possible to netboot
Sun Fire V1280 into multi-user mode. Unfortunately, they do not run
stable as processes may hang when transitioning to another CPU,
likely due to what the OpenSolaris code refers to as Cheetah+
erratum 25, but which unfortunately is not part of the publicly
available US-III+/++ errata document. Efforts on understanding this
problem are still ongoing.
* Mark Linimon is trying to find volunteers interested in helping to
fix packages on sparc64.



Contact: Luigi Rizzo <luigi@xxxxxxxxxxx>
Contact: Fabio Checconi <fabio@xxxxxxxxxxx>

geom_sched is a GEOM module that supports pluggable schedulers for disk
I/O requests. The main algorithm supported at the moment is an
anticipatory Round Robin scheduler, which is especially effective in
presence of workloads with highly random disk accesses. Other
schedulers are available on the geom_sched page.

Developed in early 2009 and refined as a GSOC2009 project, geom_sched
has been recently introduced in HEAD and is going to be soon merged to
stable/8. A version for stable/7 also exists, with some restrictions.

To use the module, say on disk ad4, all you need to do is:

kldload geom_sched
geom sched insert ad4

A number of sysctl variables under kern.geom.sched allow you to tune
the parameters of the algorithm, or bypass the scheduler entirely so
you can tell the difference of behaviour with and without the

Google Summer of Code 2010


Contact: Brooks Davis <brooks@xxxxxxxxxxx>
Contact: Robert Watson <rwatson@xxxxxxxxxxx>

We are once again participating in the Google Summer of Code. This is
our 6th year of participation and we hope to once again see great
results from our students. Currently applications have all been
submitted and we are in the process of reviewing them. Accepted
students will be announced April 26th and coding officially begins May

ipfw and dummynet enhancements


Contact: Luigi Rizzo <luigi@xxxxxxxxxxx>

We have recently completed a massive revision of ipfw and dummynet, and
the result has been committed to HEAD and stable/8. The main features
introduced with this work are:
* ipfw now has much faster skipto instructions, including table-based
ones. The complexity for rule lookups is now O(1) or O(log N) as
opposed to the O(N) that we had before. People using "skipto
tablearg" or "pipe tablearg" with large numbers of rules or pipes
should see a significant performance improvement;
* Expensive operations in response to userland reconfigurations now
do not interfere with kernel filtering for more than the time
required to swap a pointer;
* You can now use ports and the "tos" field as lookup argument for
tables. This might allow some simplifications in rulesets which in
turn result in faster execution time;
* ipfw can now send packets matching rules with a 'log' attribute to
the "ipfw0" pseudo interface, where you can run tcpdump to
implement additional filtering, logging etc.;
* dummynet now supports many different scheduler types, to adapt to
different needs people may have in terms of performance and service
guarantees. Existing schedulers now include FIFO, WF2Q+, Deficit
Round Robin, Priority, and QFQ. More schedulers can be implemented
as loadable kernel modules.;
* The kernel side has a backward-compatible interface so you can use
a RELENG_7 or RELENG_8 version of /sbin/ipfw to configure the
firewall and dummynet.

Open tasks:

1. There is ongoing work on optimizing the deletion of idle entries in
dummynet. This should be completed shortly.
2. A longer term goal is to parallelize operation in presence of ipfw
dynamic rules, which currently require exclusive lock on a hash
table containing dynamic rules.

IPv6 without legacy IP kernel


Contact: Bjoern A. Zeeb <bz@xxxxxxxxxxx>

During 2009 work was done that allowed us to build a FreeBSD kernel
without INET and without INET6 (again). This work was the foundation
for a prototype to get a kernel to compile and boot with only INET6 but
no INET compiled in earlier this year.

The current focus is to identify general architectural problems and
dependencies we do have between these two address families as well as
with the upper layer protocols. This will at some point allow us to
discuss the issues and seek solutions, preparing for a future where we
can remove either INET or INET6 from the system.

Once we will have a stable, in-tree way to compile out either address
family, optimizations wrt. size, as well as user space will need to be
worked on. In addition to this, the work is believed to help should we
further head in the direction of network stack modularization.

LDAP support in base system

Contact: Xin ZHAO <quakelee@xxxxxxxxxx>
Contact: Xin LI <delphij@xxxxxxxxxxx>

FreeBSD is currently lacking support of LDAP based authentication and
user identity.

We have integrated a stripped down OpenLDAP library (renamed to avoid
conflict with ports OpenLDAP libraries), as well as some changes to
OpenSSH as well as plugins for PAM, NSS and can support.

We have used several existing works and updated them to use new
OpenLDAP API, fixed several bugs and integrated them together. All
these works are under BSD or similar license and our new work would be
under 2-clause BSD license. Currently, we support storing user
identity, password and SSH public keys in LDAP tree.

Open tasks:

1. Further code review.
2. Make the changes less intrusive.
3. Fix issues found in production deployment.

meetBSD 2010 -- The BSD Conference


Contact: meetBSD Information <info@xxxxxxxxxxx>

meetBSD is an annual event gathering users and developers of the BSD
operating systems family, mostly FreeBSD, NetBSD and OpenBSD. Afer the
special California edition, meetBSD Wintercamp in Livigno, this year we
are back to Krakow, Poland.

meetBSD 2010 will be held on 2-3 July at Jagiellonian University.

See the conference main web site for more details.



Contact: Martin Matuska <mm@xxxxxxxxxxx>

mfsBSD is a set of scripts that generate a bootable image (e.g. an ISO
file) that creates a working minimal installation of FreeBSD that is
completely loaded into memory (mfs).

The project has now reached a stable and well tested state. Images can
be created from 8.0-RELEASE or 7.3-RELEASE ISO image files or from a
custom makeworld.

A new feature is a script called "zfsinstall" that automates a ZFS-only
install of FreeBSD from a mfsbsd ISO (script works with 8-STABLE and
9-CURRENT, sample ISO images can be downloaded from the project web

Open tasks:

1. Bundle distribution installation files (target: 8.1-RELEASE).
2. Make zfsinstall 7.3 compatible (mostly gpart syntax).
3. Enable zfsinstall combination with sysinstall (zfsinstall prepares
drives, sysinstall installs distribution).
4. Integrate toolset into FreeBSD source (tools?).

Modular Congestion Control


Contact: Lawrence Stewart <lstewart@xxxxxxxxxxx>

I have just completed the last disruptive change to the KPI, which laid
the groundwork to allow different congestion aware transports to share
congestion control algorithms. The import into the head branch is a big
job and my time is limited, so progress will be slow and I will not
have it done and ready to MFC by 8.1 as I had hoped. I will aim to have
it in 8.2 though.

Open tasks:

1. Solicit external testing.
2. Commit to head.

Multichannel playback in HDA sound driver (snd_hda)

Contact: Alexander Motin <mav@xxxxxxxxxxx>

snd_hda(4) audio driver got real multichannel playback support. It now
supports 4.0 (quadro), 5.1 and 7.1 analog speaker setups. Digital
multichannel AC3/DTS passthrough was already implemented earlier.
Digital multichannel LPCM output via HDMI could also be possible now,
but is not tested.

To use multichannel playback you should have fresh 8-STABLE kernel,
instruct sound(4) vchans subsystem (if you are using it) about your
speaker setup using sysctls and use your
audio/video player application to play multichannel audio content
without down-mixing it to stereo.

Open tasks:

1. HDMI/DisplayPort often require some audio support from X11 video
drivers. This area still should be investigated and tested,
especially relayed to multichannel LPCM playback.

NAND Flash framework for embedded FreeBSD


Contact: Grzegorz Bernacki <gjb@xxxxxxxxxxxx>
Contact: Rafal Jaworowski <raj@xxxxxxxxxxxx>

The purpose of this project is to provide embedded FreeBSD with a
generic and flexible scheme to support NAND Flash devices. The
framework provides a set of KOBJ interfaces inside the kernel, which
allow for uniform and flexible management of the NAND devices:
* NAND Flash Controller (NFC) layer, into which back-end drivers for
individual controllers plug in (implementing low-level routines
specific to a given NAND controller)
* Generic (common) NAND layer which provides means to perform
operations on the flash devices in an abstract way (read, program,
erase, get status etc.)
* NAND character device, which exports chip device as a standard
character device and allows to read/write directly to a device, as
well as perform other specific operations by using ioctl.
* GEOM NAND class for basic access through GEOM.

Part of the infrastructure is a full system simulator of ONFI-compliant
devices (NANDsim), with a userland control application. This allows for
exercising of the framework on platforms without real NAND chips.

Current state highlights:
* The framework is considered functionally complete (including
* Framework compliant back-end drivers are available for the
following NAND Flash controller (NFC) chips:

* Freescale MPC8572 (PowerPC)
* Marvell MV-78100 (ARM)
* Samsung S3C24X0 (ARM)

Open tasks:

1. Extend interface with features / options suggested by early
adopters of the code.
2. Complete, clean up, merge with HEAD.

net80211 rate control framework


Contact: Rui Paulo <rpaulo@xxxxxxxxxxx>

The net80211 (wireless) stack will support a modular rate control
framework soon. The idea is to reduce some code in the drivers and add
more rate control algorithms in the tree. All drivers that do rate
control in software will automatically benefit from this project. On
this stage, we are working on changing all the necessary drivers to
cope with the new framework and making sure it all works as expected.
Later this year we will bring the necessary changes to change the rate
control algorithm with ifconfig(1).

If you are doing rate control algorithm or research on rate control
algorithms for wireless networks, FreeBSD is now an ideal candidate for
testing your project!

Out of Tree Toolchain

Contact: Warner Losh <imp@xxxxxxxxxx>

Work is underway to allow the FreeBSD build system to use out of tree
compilers and binary utililies (loaders, linkers, etc), especially in a
cross compilation environment. While it is possible to swap out the
compiler with a compatible compiler relatively easily, swapping out the
toolchain is more involved. In addition, when using an external
compiler to build the system, certain parts of buildworld can be

Open tasks:

1. Create ports for latest binutils. This work is nearly complete, and
is waiting for integration of two branches that are collapsing soon
(the 'tbemd' branch from Warner and the mips collapse from Juli
2. Create ports for gcc. This work has been started. Native builds are
straight forward, but cross builds have a buildworld dependency at
the moment. These dependencies are being worked out, as well as
some gcc library dependencies.
3. Documentation needs to be written for how to use all of this.

PC-BSD PC-SysInstall Backend


Contact: Kris Moore <kmoore@xxxxxxxxxxx>

We are currently doing a lot of code cleanup in the new System
Installer backend for PC-BSD, pc-sysinstall, which can be used to
install regular FreeBSD as well. Some new features have already been
implemented, such as:
* Improved ZFS support, raidz, mirroring, multiple mount-points
per-pool, etc.
* Support for GPT/EFI on "Full" installations, allowing us to go
beyond the 2TB barrier.
* MBR Slice/Partition manager.
* geli passphrase support.

Open tasks:

1. We are mostly finished migrating to only using gpart instead of
fdisk, which gives us some new functionality for dealing with
GPT/EFI partitioning schemes.



Contact: Doug Barton <dougb@xxxxxxxxxxx>

Portmaster version 2.22 is now in the ports tree and has full support
for the following new features:
* Using the INDEX file to show that an installed port needs updating.
* Support for installation of packages in 'try packages first,'
--packages-only, --packages-if-newer, and --packages-build modes.
* A new --delete-build-only option to delete ports/packages that are
not needed at run time.
* Updating of the terminal title bar to show what is being worked on,
and how much more is left to do.
* Support for custom definitions of the packages repository and INDEX
* The ability to operate without any local ports tree at all with the
--index-only and --packages-only options.
* A new dialog to confirm the list of ports to be installed.

I am very excited about these new features, and owe a debt of gratitude
to the companies and especially the individuals who stepped forward to
support this work. I literally could not have done it without them.

Open tasks:

1. There are still some interesting and oft-requested features listed
on the proposal web site that I would really like to implement,
including (but not limited to) downloading of all packages before
beginning the installation, and writing out a script that can be
re-run either on that machine, or on a set of identical machines.

Ports Collection


Contact: Thomas Abthorpe <portmgr-secretary@xxxxxxxxxxx>
Contact: Port Management Team <portmgr@xxxxxxxxxxx>

Most of quarter one was spent dealing with the 7.3-RELEASE process.
With apparent success enforcing Feature Safe ports commits during the
8.0-RELEASE, it was continued for the recent src/ freeze.

The ports count now exceeds 21,500 ports, and counting. The open PR
count currently is over 1000. With the release of FreeBSD 7.3, it is
hoped this count will drop drastically.

Since the last report, we added four new committers, and had an old
committer rejoin us.

With the donation of an Apple Xserve, powerpc builds have resumed.
Renewed interest in ia64 has brought about new ports builds. A new
sparc64 machine hosted by skreuser will help us with this build.

The Ports Management team have been running -exp runs on an ongoing
basis, verifying how src code updates may affect the ports tree, as
well as providing QA runs for major ports updates. Of note -exp runs
were done for; gabor's BSD licensed bc/dc in src/, mva's OpenAL and SDL
upgrades; brooks' removal of NGROUPS; ed's removal of libcompat and
regexp.h; dinoex's jpeg update; a test run for m4 update; jilles'
update for sh(1); johans' update for bison; and roam's curl update.

Open tasks:

1. Looking for help fixing ports broken on CURRENT.
2. Looking for help with Tier-2 architectures.
3. Most ports PRs are assigned, we now need to focus on testing,
committing and closing.
4. Major commits expected soon include the latest Xorg, KDE4, and
Gnome updates.


Contact: Ion-Mihai Tetcu <itetcu@xxxxxxxxxxx>
Contact: Josh Paetzel <jpaetzel@xxxxxxxxxxx>

QAT has been running on a single server for about two years now and has
proven very effective at catching problems with ports commits. Many of
the problems it cannot catch are architecture or branch related. By
moving QAT to a VMware box capable of running arbitrary versions of
FreeBSD on both amd64 and i386 this limitation will be removed.

Open tasks:

1. Bring VMware server online and provision VMs.
2. Refactor QAT code to handle concurrent builds.
3. Migrate the existing QAT to the new setup.

Release Engineering Team


Contact: Release Engineering Team <re@xxxxxxxxxxx>

The Release Engineering Team announced FreeBSD-7.3 on March 23rd, 2010.
The schedule has been set for FreeBSD-8.1 with the release date planned
for mid July 2010.

Rewrite of FreeBSD read/write path using vnode page


Contact: Konstantin Belousov <kib@xxxxxxxxxxx>
Contact: Peter Holm <pho@xxxxxxxxxxx>

Based on the idea of Jeff Roberton, we reimplemented the path for
read(2)/write(2) syscalls using page cache (in wide sense) to eliminate
the issues with recursive vnode and buffer lock acquisitions. The usual
reads and writes are no longer calls into VOP_READ/VOP_WRITE; the
operation is done by copying user buffers to or from the pages of the
vnode. This fixes known deadlocks when reads or writes are done over
file-mmaped buffers.

The patch changes the performance characteristics of I/O, and we
observed both better and worse behaviour. If filesystem implements
VOP_GETPAGES and VOP_PUTPAGES without referencing buffer cache, buffers
are completely eliminated from the i/o path (not true for UFS or NFS).

Open tasks:

1. We need wider testing and reviews.

SUJ: Journaled Softupdates


Contact: Jeff Roberson <jeff@xxxxxxxxxxx>

The soft-updates journaling project is nearing completion and will be
available in head by the time this status report is released. Backports
to other releases are maintained in SVN. SUJ is fully backwards
compatible with non-journaled softupdates. Existing systems will not be
affected. Journaling may be enabled and disabled by tunefs on unmounted
filesystems. Journaling provides near-instant filesystem recovery after
crash at the expense of some runtime performance and extra disk I/O.

TCP/UDP connection groups

Contact: Robert Watson <rwatson@xxxxxxxxxxx>
Contact: FreeBSD network mailing list <freebsd-net@xxxxxxxxxxx>

This on-going project is to reduce tcbinfo/udbinfo lock and cache line
contention; this global lock protects access to connection lists, and
while it is a read-write lock, it is acquired for every in-bound packet
(briefly) to look up the connection. This project adds a new connection
group table, which assigns connections to groups, each of which has CPU
affinity and aligns with RSS-selected queues in high-end 1gbps and most
10gbps implementations. The following tasks have been completed:
* Teach libkvm to handle dynamic per-cpu storage (DPCPU) to improve
crashdump analysis of per-CPU data.
* Teach netstat to monitor netisr DPCPU queues for live kernels and
* Create a new inpcbgroup abstraction, used for UDP and TCP.
* Distribute UDP and TCP connections (inpcbs) over groups based on
4-tuple bindings.
* Replicate membership across all groups for wildcard socket
* Write new TCP/UDP connection and binding regression tests.

The following tasks remain:
* Migrate from naive work assignment algorithm to RSS assignment.
* Modify device driver KPI to allow consistent initialization and
configuration between stack and hardware.
* Complete migration to dynamic, per-CPU network statistics in TCP,
UDP, and IP.
* Add socket options to query effective CPU affinity of connections
from userspace.
* On supporting hardware, allow affinity for a specific connection to
be explicitly migrated using a socket option.
* Detailed performance evaluation and optimization.

This work is being performed in the FreeBSD Perforce repository, and is
sponsored by Juniper Networks. Connection groups and related features
are slated for inclusion in FreeBSD 9.0 (with possible backports to
8-STABLE of some features).

The FreeBSD Foundation


Contact: Deb Goodkin <deb@xxxxxxxxxxxxxxxxxxxxx>

We were proud to be a sponsor for AsiaBSDCon in March. We also
committed to sponsoring BSDCan 2010 and NYCBSDCon 2010. We provided
travel grants for AsiaBSDCon.

We funded a project by Murray Stokely to provide Closed Captioning of
FreeBSD Technical Videos in the BSD Conferences YouTube Channel. We
were very pleased that the foundation funded HAST project completed.

We solicited project proposals and were very pleased with the number of
proposals we received. With our project spending budget increase, we
will be able to fund more projects this year.

We grew our board of directors by adding Erwin Lansing. This will
expand our representation in Europe. Erwin brings ports knowledge and
expertise to the board.

We continued our work on infrastructure projects to beef up hardware
for package-building, network-testing, etc.

Follow us on Twitter now!

We are fund-raising for 2010 now! Find out more at

The FreeBSD German Documentation Project


Contact: Johann Kois <jkois@xxxxxxxxxxx>
Contact: Benedict Reuschling <bcr@xxxxxxxxxxx>

Our last status report listed a number of documents that needed help.
Thanks to the external contributions of Frank Boerner we were able to
update a substantial amount of documents. This has resulted in a great
reduction of our backlog. Subsequently, Benedict has agreed to take
Frank under mentorship for the German doc project. We are looking
forward to his future contributions and thank him for his past efforts.

Johann was busy keeping the German website in sync with updates to However, there are still parts of the website that remain
untranslated. We are looking for more support in maintaining the German

FreeBSD users with German language skills are always welcome to join
our efforts in translating the documentation and/or fixing bugs.

Open tasks:

1. Translate more parts of the documentation and the German website.
2. Keep the current documentation up to date.
3. Report bugs to de-bsd-translators@xxxxxxxxxxxxxxx

The FreeBSD Hungarian Documentation Project


Contact: Gábor Kövesdán <gabor@xxxxxxxxxxx>
Contact: Gábor Páli <pgj@xxxxxxxxxxx>

We restlessly keep the existing documentation and web page translations
up to date. However, this will not last forever, and help is always
welcome, so if you feel yourself Hungarian with some interests in
translation, please contact our Documentation Project via the email
addresses noted above.

Open tasks:

1. Translate release notes.
2. Translate articles.
3. Translate web pages.
4. Read translations, send feedback.

The tbemd branch

Contact: Warner Losh <imp@xxxxxxxxxx>

'tbemd' stands for Target Big Endian Must Die. The current build
systems requires that one define TARGET_BIG_ENDIAN for either big
endian MIPS or big endian ARM processors. There are many problems with
this approach. The resulting system will not create the proper binaries
without TARGET_BIG_ENDIAN defined. There is no easy way to know what
the endian is of the system you are running. There are many issues with
ports, since they do not use bsd make, so do not pick up the extra
flags that are added if TARGET_BIG_ENDIAN is defined.

The tbemd branch seeks to fix this. We will move from MACHINE_ARCH=mips
for all mips platforms to MACHINE_ARCH=mipsel, mipseb, mips64eb and
mips64el to match NetBSD's conventions. These represent 32-bit mips
little endian, 32-bit mips big endian, 64-bit mips big endian and
64-bit mips little endian respectively. ARM will move to arm (little
endian) and armeb (big endian), again following the standards set
elsewhere. To facilitate a number of different MACHINE_ARCHs all built
from the same source, a new MACHINE_CPUARCH is introduced and
represents the sources needed to build CPU support for a given

In addition, MACHINE_ARCH is overused in the build system today. Many
of its uses are gratuitous and can be simplified. Many of its uses do
not scale well and need to be refactored into a system that will scale
well. A per MACHINE/MACHINE_ARCH/MACHINE_CPUARCH selection mechanism
for makefile snippets will be introduced to move much of the current if
spaghetti into more controlled lists.
The branch can build everything we currently support with the new

Open tasks:

1. Finish migrating to
2. Reduce diffs between the branch and the mainline before the
3. Documentation needs to be written for how to use all of this.



Contact: Hans Petter Selasky <hselasky@xxxxxxxxxxx>

Webcamd is a userland daemon that enables use of hundreds of different
USB based Linux device drivers under the FreeBSD-8/9 operating system.
Current focus has been on USB webcam and USB DVB-T/S/C devices. It is
also possible to use the webcamd framework to make other Linux kernel
USB devices work under the FreeBSD-8/9 operating system, without
violating the GPL license. The daemon currently depends on libc,
pthreads, libusb and libcuse4bsd. Cuse4BSD is a new character device
from userland implementation that fully supports open, read, write,
ioctl, mmap and close file operations.

If you like this project or want me to spend more time on it, you can
support it by transferring money to hselasky@xxxxxxx via paypal.

Open tasks:

1. Testing and bugfixes.
2. Add support for more device drivers.



Contact: Pawel Jakub Dawidek <pjd@xxxxxxxxxxx>
Contact: Martin Matuska <mm@xxxxxxxxxxx>
Contact: Xin LI <delphij@xxxxxxxxxxx>

The ZFS file system has been updated to version 14 on both -HEAD and
8-STABLE. Ongoing work is undergoing to bring bug fixes and performance
improvements from upstream svn -HEAD to approximately ZFS v15 in the
near future, and a full upgrade of ZFS to version 24 including the
de-duplication functionality, etc. The de-duplication functionality is
currently partly supported, which is demonstrated below:

# uname -sr
# zpool create tank ad{4,6,8,10}
# zpool get version tank
tank version 24 default
# zfs set dedup=on tank
# dd if=/dev/random of=/tank/rand0 bs=1m count=1024
# zpool get allocated,dedupratio tank
tank allocated 1.00G -
tank dedupratio 1.00x -
# dd if=/tank/rand0 of=/tank/rand1 bs=1m
# dd if=/tank/rand0 of=/tank/rand2 bs=1m
# dd if=/tank/rand0 of=/tank/rand3 bs=1m
# zpool get allocated,dedupratio tank
tank allocated 1.01G -
tank dedupratio 4.00x -

Open tasks:

1. Bring ZFS v15 changes to svn -HEAD and MFC.
2. Further polish the code in perforce and test for functionality,

(c) 2010 The FreeBSD Project. All rights reserved.
freebsd-current@xxxxxxxxxxx mailing list
To unsubscribe, send any mail to "freebsd-current-unsubscribe@xxxxxxxxxxx"