Jonathan Perkin

New blog design

jonathan@perkin.org.uk (Jonathan Perkin) — Mon, 20 Feb 2012 00:00:00 +0000

I am a minimalist at heart, and I love command line interfaces and text files.

However, I’m also pragmatic, and so for a long time my blog has been powered by Wordpress, which made it relatively easy to manage content and design. I just didn’t want the hassle of maintaining something myself.

Somewhere along the way recently I stumbled across Jekyll, and the idea of generating a static site from templates.

I’d seen a few sites I liked the look of, and I loved the idea of simplifying my site (and editing posts in vim!), and so the end result is the site you see now. The theme is based on skim.la’s site.

I pondered implementing something like disqus for comments, but honestly I get so few proper responses, compared to the hundreds and hundreds of spam attempts, that it’s not worth it - if you want to comment on a post, just use twitter (I’ll probably implement a little button to do that at some point).

Also, I got rid of categories, simpler to just have tags.

It’s likely I’ve messed up conversion of my old blog posts, as I did them all manually. If you spot anything obviously broken, please let me know. One of the great things of the Jekyll approach is that you can simply manage your site via github, and so if you prefer you can comment on breakages here!

Set up anonymous FTP upload on Oracle Linux

jonathan@perkin.org.uk (Jonathan Perkin) — Wed, 11 Jan 2012 00:00:00 +0000

Just because this took a little longer than I expected, here’s a quick howto for setting up an anonymous FTP drop-off on Oracle Linux, which I use as a simple way to transfer files out of my Virtual Machines.

Install vsftpd

$ sudo yum -y install vsftpd

Configure iptables

As FTP is a more complicated protocol than most, there is a special netfilter module required in order to correctly keep track of connections.

# You will perhaps want to change the insert number here.
$ sudo iptables -I INPUT 4 -m state --state NEW -p tcp --dport 21 -j ACCEPT
$ sudo /etc/init.d/iptables save

Add nf_conntrack_ftp to IPTABLES_MODULES

$ sudo vi /etc/sysconfig/iptables-config

Then load the module rather than reboot

$ sudo modprobe nf_conntrack_ftp

Create /incoming

Create /incoming area and ensure it has the correct file permissions and SELinux context. This is the bit which had me stumped for a little while, as I didn’t know about allow_ftpd_anon_write, and while I normally just disable SELinux, I do also like to know how things should work (and be able to write about them!):

$ sudo mkdir /var/ftp/incoming
$ sudo chown ftp:ftp /var/ftp/incoming

# This allows anonymous users to upload, but not see what''s in the directory
$ sudo chmod 750 /var/ftp/incoming

$ sudo chcon -u system_u -t public_content_rw_t /var/ftp/incoming
$ sudo setsebool allow_ftpd_anon_write=1

Configure vsftpd

$ sudo vi /etc/vsftpd/vsftpd.conf

anon_upload_enable=YES

Startup

Finally, enable and start vsftpd:

$ sudo chkconfig vsftpd on
$ sudo /etc/init.d/vsftpd start

And that’s it, you should now be able to FTP as anonymous and upload files into /incoming.

Kickstart Oracle Linux in VirtualBox

jonathan@perkin.org.uk (Jonathan Perkin) — Mon, 09 Jan 2012 00:00:00 +0000

In my previous post I configured an Ubuntu laptop as a Kickstart install server for a physical machine I wanted to build.

Now that everything is configured for automated installs, it makes sense to use the same infrastructure to install virtual machines too. Everything can be done from the command line, and new virtual machines can be provisioned very quickly.

Again, I will use Oracle Linux 6.2 as an example.

Create Virtual Machine

These commands create a new virtual machine, including disk, and configures it for network booting. If you don’t already have VirtualBox installed:

$ sudo apt-get install virtualbox

Set the name of the virtual machine as it appears in VirtualBox, this variable is then used throughout:

$ VM="Oracle Linux 6.2"

Create the VM:

$ VBoxManage createvm --name "${VM}" --ostype "Oracle_64" --register

Create the hard disk (32GB expanding) and attach it via SATA. Note that I store my VMs in ${HOME}/VirtualBox rather than the default ${HOME}/VirtualBox VMs.

$ VBoxManage createhd --filename "VirtualBox/${VM}/${VM}.vdi" --size 32768

$ VBoxManage storagectl "${VM}" --name "SATA Controller" --add sata \
    --controller IntelAHCI

$ VBoxManage storageattach "${VM}" --storagectl "SATA Controller" --port 0 \
    --device 0 --type hdd --medium "VirtualBox/${VM}/${VM}.vdi"

Default RAM is 128MB, Oracle Linux installer requires at least 512MB however. Once installed we can drop back down to 256MB or so:

$ VBoxManage modifyvm "${VM}" --memory 512

Configure boot order. Put disk first, as on the first boot there is nothing on it so it falls through to PXE for install, then after the install the disk is bootable.

$ VBoxManage modifyvm "${VM}" --boot1 disk --boot2 net --boot3 none --boot4 none

Change NIC type from the default e1000, as a vanilla VirtualBox install does not include the firmware necessary to network boot from that device – it is available in the “VirtualBox Extension Pack” add-on. Switch to a plain PCNet Fast III which does include PXE firmware.

$ VBoxManage modifyvm "${VM}" --nictype1 Am79C973

VirtualBox does have the ability to serve TFTP directly from the file system by placing files inside ${HOME}/.VirtualBox/TFTP/ but I prefer to just use the network as it’s already configured.

$ VBoxManage modifyvm "${VM}" --nattftpserver1 10.0.2.2
$ VBoxManage modifyvm "${VM}" --nattftpfile1 pxelinux.0

Configure pxelinux/kickstart

We just need a couple of tweaks to the configs from the last blog entry, as the network addresses are different inside VirtualBox, and we also may want a different kickstart configuration.

$ sudo vi /var/lib/tftpboot/pxelinux.cfg/default

I just amended the existing entry to point to 10.0.2.2 which is the address of the machine running VirtualBox, and a different ks-vm.cfg kickstart configuration file, but you could also create a new label if you wanted to regularly switch between different configurations:

LABEL ol6.2
    KERNEL /ol6.2/vmlinuz
    APPEND initrd=/ol6.2/initrd.img ks=http://10.0.2.2/ks-vm.cfg

For virtual machines I use a slightly different configuration compared to previously. I’ve only shown the changes below, not the full file:

$ sudo cp /usr/share/nginx/www/ks.cfg /usr/share/nginx/www/ks-vm.cfg

# Update network configuration for DHCP instead of static
network --bootproto=dhcp
url --url=http://10.0.2.2/ol6.2

# Don't specify disks, just use default layout
bootloader --location=mbr --driveorder=sda
clearpart --all --initlabel

# Just add 'screen' to the default set of @base and @core packages.
%packages
screen
%end

Start Virtual Machine

All that’s left to do is to boot up the VM, and everything else should run automatically.

$ VirtualBox --startvm "${VM}"

All done!

Kickstart Oracle Linux from Ubuntu

jonathan@perkin.org.uk (Jonathan Perkin) — Mon, 09 Jan 2012 00:00:00 +0000

As my new job involves working on Oracle Linux, I figured I should migrate my home server to it, which would also mean I could move it to a proper RAID10 configuration rather than relying on multiple RAID1′s.

My laptop runs Ubuntu, and I wanted to install the server from it using PXE and Kickstart, so here’s how I did it.

Configure dhcpd

DHCP is required for two things, to give the server its network configuration, and to point it at the pxe boot loader we want to use.

$ sudo apt-get install isc-dhcp-server
$ sudo vi /etc/dhcp/dhcpd.conf

Here’s the configuration I used, which says to configure 192.168.2.0/24 with a dynamic DHCP range between 192.168.2.100 – 192.168.2.200, and to boot machines using pxelinux.0 which is relative to the TFTP root directory (configured in the next section):

subnet 192.168.2.0 netmask 255.255.255.0 {
    range 192.168.2.100 192.168.2.200;
    filename "pxelinux.0";
}

The isc-dhcp-server install automatically tries to start the server, but will fail as it isn’t configured, so we restart it now that there is a working configuration installed

$ sudo /etc/init.d/isc-dhcp-server restart

Configure tftpd

TFTP is a simple protocol used to transfer files over the network, and due to its simplicity it is the primary way to network boot, as it can be easily embedded into firmware.

All we need to do is install the TFTP daemon and syslinux which includes the pxe boot loader, then put the pxelinux file into the tftproot area:

$ sudo apt-get install syslinux tftpd-hpa
$ sudo cp /usr/lib/syslinux/pxelinux.0 /var/lib/tftpboot

Configure Oracle Linux DVD

To save space on the laptop we can just mount the DVD read-only and install from that:

$ sudo mkdir /media/ol6.2
$ sudo mount -o loop,ro /path/to/OracleLinux-R6-U2-Server-x86_64-dvd.iso /media/ol6.2

However, we do need to copy the kernel and initrd image from the DVD into the tftproot as they are required for booting:

$ sudo mkdir /var/lib/tftpboot/ol6.2 /var/lib/tftpboot/pxelinux.cfg
$ sudo cp -a /mnt/images/pxeboot/{initrd.img,vmlinuz} /var/lib/tftpboot/ol6.2/

Configure pxelinux

All that’s left for the PXE stage is to configure the boot loader, and tell it what kernel and initrd we want to use:

$ sudo vi /var/lib/tftpboot/pxelinux.cfg/default

This configuration has just one entry which is booted after a short wait, but pxelinux has many more options, including the ability to boot from local disk.

DEFAULT	ol6.2
PROMPT 1
TIMEOUT 5

LABEL ol6.2
    KERNEL /ol6.2/vmlinuz
    APPEND initrd=/ol6.2/initrd.img ks=http://192.168.2.1/ks.cfg

Note the ks argument which specifies the kickstart file we will use, that and the web server required to serve it will be set up next.

Configure nginx

I chose nginx as it is small and simple to configure, but any web server will do.

$ sudo apt-get install nginx
$ sudo vi /etc/nginx/sites-available/default

The install will be performed over HTTP, so we need to make the DVD we mounted earlier available. This entry in the server { } section makes the DVD available via http://192.168.2.1/ol6.2/:

server {
[...]
    location /ol6.2 {
        root /media;
        autoindex on;
        allow all;
    }

Then start nginx (unlike isc-dhcp-server this isn’t done automatically):

$ sudo /etc/init.d/nginx start

Configure kickstart

Finally, we create a kickstart configuration which specifies exactly how our target machine is to be installed, and this allows a completely unattended installation.

Ideally I should create a specific area for holding files like this, but as a quick hack I simply put it into the default nginx web root (and thus available as http://192.168.2.1/ks.cfg as configured earlier in the pxelinux.cfg/default file:

sudo vi /usr/share/nginx/www/ks.cfg

Here is my ks.cfg file in full. The only thing missing is a rootpw line to automatically set a root password, however for maximum security I am happy to forego a completely unattended installation and instead have the installer prompt me to type it in during the install.

Some notes:

UK keyboard language and timezone selected.
Automatically reboot when the installer is finished.
Point to the Oracle Linux DVD using the url directive.
I have 4 disks configured with RAID1 for /boot, and RAID10 for swap and /.
Disks are referred to by path, to ensure correct ordering.
A small set of packages are installed, containing just the functionality I require.
A small %post section is used to perform any fixes I want for the first boot.

#
# Miscellaneous options.
#
install
keyboard uk
lang en_GB.UTF-8
reboot
selinux --enforcing
timezone Europe/London

#
# User setup.  I'd create my local user here and configure sudoers, but
# kickstart doesn't yet support creating a user with uid/gid of 1000 (the
# gid is always 500, even if you add the named group first).
#
authconfig --enableshadow --passalgo=sha512

#
# Networking.  The 'network' line needs to be on a single line
# for kickstart to work - it is only split here for the blog.
#
firewall --service=ssh
network --bootproto=static \
        --hostname=gromit.adsl.perkin.org.uk \
        --ip=192.168.2.10 \
        --netmask=255.255.255.0 \
        --gateway=192.168.2.1 \
        --nameserver=193.178.223.141,208.72.84.24 \
        --ipv6=auto
url --url=http://192.168.2.1/ol6.2

#
# Disk configuration.
#
bootloader --location=mbr --driveorder=sda,sdb,sdc,sdd
clearpart --all --initlabel
#
# /boot (RAID1 necessary as booting from RAID10 isn't supported)
#
part raid.00 --asprimary --size=1024 --ondisk=/dev/disk/by-path/pci-*-0*0
part raid.01 --asprimary --size=1024 --ondisk=/dev/disk/by-path/pci-*-1*0
part raid.02 --asprimary --size=1024 --ondisk=/dev/disk/by-path/pci-*-2*0
part raid.03 --asprimary --size=1024 --ondisk=/dev/disk/by-path/pci-*-3*0
raid /boot --level=1 --device=md0 --fstype=ext4 raid.00 raid.01 raid.02 raid.03
#
# swap, RAID10 of size RAM+2GB, give or take..
#
part raid.10 --asprimary --size=6144 --ondisk=/dev/disk/by-path/pci-*-0*0
part raid.11 --asprimary --size=6144 --ondisk=/dev/disk/by-path/pci-*-1*0
part raid.12 --asprimary --size=6144 --ondisk=/dev/disk/by-path/pci-*-2*0
part raid.13 --asprimary --size=6144 --ondisk=/dev/disk/by-path/pci-*-3*0
raid swap --level=10 --device=md1 --fstype=swap raid.10 raid.11 raid.12 raid.13
#
# /, RAID10 of remainder (have to specify an arbitrary --size even with --grow)
#
part raid.20 --asprimary --size=1024 --grow --ondisk=/dev/disk/by-path/pci-*-0*0
part raid.21 --asprimary --size=1024 --grow --ondisk=/dev/disk/by-path/pci-*-1*0
part raid.22 --asprimary --size=1024 --grow --ondisk=/dev/disk/by-path/pci-*-2*0
part raid.23 --asprimary --size=1024 --grow --ondisk=/dev/disk/by-path/pci-*-3*0
raid / --level=10 --device=md2 --fstype=ext4 raid.20 raid.21 raid.22 raid.23

#
# Packages.  @base and @core are pre-selected.
#
%packages
@cifs-file-server
@console-internet --optional
@development
@legacy-unix --optional
@mail-server
@network-server --optional
@network-tools
@nfs-file-server
@web-server
screen
%end

#
# Post-install fix-ups.
#
%post
#
# The 'network' directive doesn't support DNS search paths, so set those
# manually, and disable Network Manager.
#
printf "/^NM_CONTROLLED/s/yes/no/\nw\nq\n" \
  | ed /etc/sysconfig/network-scripts/ifcfg-eth0
printf "/^#/s/.*/search adsl.perkin.org.uk perkin.org.uk/\nw\nq\n" \
  | ed /etc/resolv.conf
#
# Disable unwanted services
#
chkconfig --del cups
%end

All done.

Last day at MySQL

jonathan@perkin.org.uk (Jonathan Perkin) — Thu, 22 Dec 2011 00:00:00 +0000

Today is my final day working on MySQL. It has been an amazing 4.5 years, and I’ve loved working on the technical challenges involved in producing a piece of software which runs on so many different platforms, as well as working with so many talented individuals. I have learned a lot, which is just how I like it.

Times have changed since I first joined MySQL AB, of course. But despite going through two acquisitions, the day to day work hasn’t changed much at all. I still get to work from home, which is something I am passionate about and feel many companies are missing out on massively by still being stuck in an outdated 20th century mindset. Oracle have continued to invest in MySQL, providing additional headcount and extra hardware. And the vast majority of people I started working with back in 2007 are still with the company (don’t believe the FUD, folks) and working harder than ever.

I’m most proud of the work we have done internally to ensure our MySQL releases are in excellent shape. When I first joined, our internal ‘PushBuild’ system was very limited, only able to handle a small number of pushes with reduced testing across a few platforms, and the builds produced were completely different to those that were released.

Nowadays, thanks to the investment made by Sun and Oracle, we have a large server farm producing thousands of builds and hundreds of gigabytes of data every day across all our supported platforms. Those which are based on MySQL 5.5 or newer have additional package verification tests, using chroots and virtual machines, to ensure that the RPM/MSI/etc packages can be installed, run, and uninstalled, all automatically and on every push by developers. We run additional nightly and weekly tests which extend the default set of test suites. Our 5.5+ releases are produced directly in PushBuild. And we are looking to extend this to all MySQL products, not just the Server.

If the above sounds fun, keep an eye out for job posts, as the Release Engineering team is looking to expand, and they are an amazing group of people to work with.

I certainly found it fun, which is why I’m going to continue in that line of work, and stay within Oracle, but move to the Linux group. To be honest, databases have never been a passion of mine, however I am rather fond of Operating Systems, and so I am really looking forward to continuing to work on continuous integration and testing, but with Oracle Linux and Oracle VM instead of MySQL. Plus, it will be nice to go back to a purely technical role – I have learned over the last year or so that management isn’t really my thing :)

To all my colleagues, past and present, thank you for the wonderful ride. I have many, many good memories, and hopefully we will keep in touch.

Installing OpenBSD with softraid

jonathan@perkin.org.uk (Jonathan Perkin) — Thu, 15 Dec 2011 00:00:00 +0000

This is more of a log for me than anything else, but perhaps someone will find this useful.

OpenBSD includes a software RAID implementation which supports booting in newer snapshots, and I was itching to install the latest version and use it as my file server, which has 4 750GB disks. There is a small bit of preparation work to do prior to installing, which is the bulk of this entry, most of which is based on this undeadly.org article.

Grab latest amd64 snapshot, boot it, drop into (S)hell mode and set up the disks:

#!/bin/sh

cd /dev
sh MAKEDEV sd1 sd2 sd3 sd4 sd5
for disk in 0 1 2 3
do
  # Clear beginning of disks..
  dd if=/dev/zero of=/dev/rsd${disk}c bs=1m count=10

  # ..and initialise new partition table
  fdisk -iy sd${disk}

  #
  # Create BSD disklabel:
  #
  # - 128m partitions at start to hold kernels for booting
  # - 4g spare raid on each disk for testing
  # - rest raid for main OS and data
  #   - OS and /home on first two mirrored disks
  #   - /store on second two mirrored disks
  #
  print "a a\n\n128m\n\na d\n\n4g\nraid\na e\n\n\nraid\nw\nq\n" \
    | disklabel -E sd${disk}

  # Clear beginning of raid partitions
  dd if=/dev/zero of=/dev/rsd${disk}d bs=1m count=10
  dd if=/dev/zero of=/dev/rsd${disk}e bs=1m count=10
done

# Create RAID1 mirrors
bioctl -c 1 -l sd0e,sd1e softraid0
bioctl -c 1 -l sd2e,sd3e softraid0

# Exit shell and start the (I)nstall
exit

As for the install, go with the sensible defaults, other than:

change keyboard layout to ‘uk’
manually configure network, enable rtsol
start ntpd
do not expect to run X

When it comes to disk selection, choose sd4 as the root disk, and use the following layout:

# partition  size  mount
  sd4a       1G    /
  sd4b       8G    swap
  sd4d       1G    /tmp
  sd4e       8G    /var
  sd4f       16G   /usr
  sd4h       rest  /home

then initialise sd5 with:

# partition  size  mount
  sd5a       2G    /altroot
  sd5d       8G    /scratch
  sd5e       rest  /store

Install the full OS, set the correct timezone, then before rebooting initialize the boot partitions and copy the kernels to them. Doing this on all of them means we can boot from any disk.

#!/bin/sh

for disk in 0 1 2 3
do
  newfs sd${disk}a
  mount /dev/sd${disk}a /mnt2
  cp /mnt/bsd* /mnt2
  umount /mnt2
done
eject cd0a
reboot

Job done.

Create VirtualBox VM from the command line

jonathan@perkin.org.uk (Jonathan Perkin) — Wed, 21 Sep 2011 00:00:00 +0100

As something of a follow-up post to the previous entry, here’s a quick recipe for creating a Virtual Machine using the VirtualBox command line tools:

We’re using Windows Server 2008 64bit as an example, modify to taste.

$ VM='Windows-2008-64bit'

Create a 32GB ‘dynamic’ disk.

$ VBoxManage createhd --filename $VM.vdi --size 32768

You can get a list of the OS types VirtualBox recognises using:

$ VBoxManage list ostypes

Then copy the most appropriate one into here.

$ VBoxManage createvm --name $VM --ostype "Windows2008_64" --register

Add a SATA controller with the dynamic disk attached.

$ VBoxManage storagectl $VM --name "SATA Controller" --add sata \
>  --controller IntelAHCI
$ VBoxManage storageattach $VM --storagectl "SATA Controller" --port 0 \
>  --device 0 --type hdd --medium $VM.vdi

Add an IDE controller with a DVD drive attached, and the install ISO inserted into the drive:

$ VBoxManage storagectl $VM --name "IDE Controller" --add ide
$ VBoxManage storageattach $VM --storagectl "IDE Controller" --port 0 \
>  --device 0 --type dvddrive --medium /path/to/windows_server_2008.iso

Misc system settings.

$ VBoxManage modifyvm $VM --ioapic on
$ VBoxManage modifyvm $VM --boot1 dvd --boot2 disk --boot3 none --boot4 none
$ VBoxManage modifyvm $VM --memory 1024 --vram 128
$ VBoxManage modifyvm $VM --nic1 bridged --bridgeadapter1 e1000g0

Configuration is all done, boot it up! If you’ve done this one a remote machine, you can RDP to the console via vboxhost:3389.

$ VBoxHeadless -s $VM

Once you have configured the operating system, you can shutdown and eject the DVD.

$ VBoxManage storageattach $VM --storagectl "IDE Controller" --port 0 \
>  --device 0 --type dvddrive --medium none

Finally, it’s a good idea to take regular snapshots so that you can always revert back to a known-good state rather than having to completely re-install.

$ VBoxManage snapshot $VM take <name of snapshot>

And, if you need to revert back to a particular snapshot:

$ VBoxManage snapshot $VM restore <name of snapshot>

Enjoy!

Creating chroots for fun and MySQL testing

jonathan@perkin.org.uk (Jonathan Perkin) — Wed, 14 Sep 2011 00:00:00 +0100

Virtualisation is all the rage today, however there are still a few cases where good old-fashioned Unix chroot is still applicable, and testing MySQL across multiple platforms and architectures is one of those cases.

At Oracle we do full automated package verification testing of our MySQL server binaries prior to release, which attempts to install the package, start the server, run some functionality testing, then uninstall. It is of course highly desirable that the testing environment this is performed in is as close to a clean install of the target operating system as possible, to avoid problems such as our packages depending upon some local changes or packages we may have installed which won’t be available on a customer system.

Given the large number of platforms and architectures that MySQL supports, going the virtualisation route would mean having to use many different products: VirtualBox for x86, zones for SPARC, qemu for ia64/PA-RISC/others (if it even supports them), and this gets complicated quickly and is not very maintainable. Thus I chose to use chroot as much as possible. In addition, it’s much faster and less intensive on resources to use a chroot than boot up an entire OS image each time.

I built all images directly from the original installation images (DVD, ISO, RPM, etc), to ensure that there was no contamination from our build environment or local install scripts in the image – they should be as close to what a normal user or customer will be running in their setup. From the install image, the packages are installed to a temporary directory, some final modifications are made, then the directory is tarred up ready to be extracted by the test framework and used.

Here are some operating system specific examples, which set up an extracted chroot image into ${CHROOTDIR}. There may be additional steps required to get a fully functioning chroot, such as copying device files (/dev/zero and /dev/null are usually the minimum requirements) and adding users.

FreeBSD

FreeBSD 7 and 8 come as a number of sets in tar format, and for our purposes we only need to extract the base set. You may wish to add more sets if you want to use your chroot for building packages:

#!/bin/sh
mdunit=$(mdconfig -a -t vnode -o readonly -f /path/to/dvd1/of/freebsd.iso)
mount_cd9660 /dev/${mdunit} /mnt
cat /mnt/*/base/base.?? | tar -xpzf - -C ${CHROOTDIR}
umount /mnt
mdconfig -d -u ${mdunit}

HP-UX

HP-UX has since been EOL’d for MySQL, however this information might still prove useful. The HP-UX installation media contains per-directory packages, with the contents representing how they are laid out on the destination file system with each file gzip compressed.

#!/bin/sh
cd /path/to/extracted/hpux-dvd1
for pkg in $LIST_OF_PKGS
do
  for subpkg in ${pkg}/*
  do
    if [ ! -d "${subpkg}" ]; then
      continue
    fi
    for d in $(find ${subpkg} -type d)
    do
      mkdir -p ${CHROOTDIR}/$(echo ${d} \
        | sed -e "s#${subpkg}/##g" \
              -e "s#usr/newconfig/##g")
    done
    for f in $(find ${subpkg} -type f)
    do
      gzip -dc ${f} >${CHROOTDIR}/$(echo ${f} \
        | sed -e "s#${subpkg}/##g" \
              -e "s#usr/newconfig/##g")
    done
  done
done

Once this is done you’ll need to fix up permissions in bin and lib directories (make files executable), as well as create a bunch of symlinks for e.g. /bin and /lib.

OSX

For OSX we don’t actually use a chroot tarball but instead create a sparse disk image. Currently the size of the “chroot” is very large as there’s no easy way to strip down an OSX install, so mounting a disk image is faster than unpacking a chroot, plus it preserves various HFS-specific attributes.

You will likely need at least the BSD, BaseSystem, and Essentials packages.

# Create a sparse image to hold the chroot (which isn't really a directory)
hdiutil create -fs HFS+ -size 8g -type SPARSE -volname osx-chroot ${CHROOTDIR}
hdiutil attach -mountpoint ${CHROOTDIR} ${CHROOTDIR}.sparseimage
# Either attach a DVD image or the real thing
hdiutil attach -mountpoint /Volumes/osx-install /path/to/dvd
# Install packages
for pkg in BSD BaseSystem Essentials
do
  installer -verbose \
   -pkg /Volumes/osx-install/System/Installation/Packages/${pkg}.pkg \
   -target ${CHROOTDIR}
done
# Unmount
hdiutil detach ${CHROOTDIR}
hdiutil detach /Volumes/osx-install

Red Hat / Oracle Linux / SuSE

For RPM-based distributions we use rpm to directly install packages into the chroot. The list of RPMs we install varies quite a lot from release to release, usually by having to increase the number:

RH3: 81
RH4: 85
RH5: 114
RH6: 203

for the same functionality.

# Kludge for 'setup' RPM to install
mkdir -p ${CHROOTDIR}/var/lock/rpm
# If installing from an ISO:
mount -o loop /path/to/iso /mnt
# Path varies from release to release
cd /mnt/path/to/RPMs
rpm --root=${CHROOTDIR} -Uvh ${LIST_OF_RPMS}

Solaris

Similar to RPM, we use the native package manager to install packages directly into the chroot directory:

# Avoid prompts
sed -e "s/ask$/nocheck/" /var/sadm/install/admin/default > /tmp/admin-$$
pkgadd -a /tmp/admin-$$ -R ${CHROOTDIR} -d . ${LIST_OF_PKGS}
rm /tmp/admin-$$

Windows

Ok, so of course we can’t use chroot images for Windows, as it doesn’t have chroot(2). So here we use VirtualBox and its snapshot ability to load a clean snapshot of a basic Windows install, do the tests, then shut down the virtual machine, restore the snapshot, and boot up again.

Graphing memory usage during an MTR run

jonathan@perkin.org.uk (Jonathan Perkin) — Thu, 30 Jun 2011 00:00:00 +0100

In order to optimally size the amount of RAM to allocate to a set of new machines for running MTR, I ran a few tests to check the memory usage of an MTR run for mysql-trunk and cluster-7.1. As using a RAM disk considerably speeds things up, I set the vardir to be on /ramdisk and logged the usage of that too.

The tests were performed on an 8-core E5450 @ 3.00GHz with 24GB RAM, with 8GB allocated to /ramdisk. Each branch ran the default.daily collection, which generally contains the most testing we do per-run. Between each run I rebooted the machine to clear the buffer cache and /ramdisk.

I used something like the script below, which saved the per-second usage of /ramdisk, the total RAM used, and the RAM used minus buffers.

#!/bin/bash

BRANCH="mysql-trunk"
BUILDDIR="mysql-5.6.3-m5-linux2.6-x86_64"
TESTDIR="${HOME}/mtr-test/${BRANCH}"

stats()
{
  i=1
  rm -f ${TESTDIR}/stats-${BRANCH}
  while [ -f ${TESTDIR}/running ]; do
    rd=$(df -k /ramdisk | awk '/^\// {print $3}')
    mem=$(free | awk '/^Mem/ {print $3}')
    mem1=$(free | awk '/cache:/ {print $3}')
    echo "${i} ${rd} ${mem} ${mem1}" >>${TESTDIR}/stats-${BRANCH}
    i=$((i+1))
    sleep 1
  done
}

export TMPDIR="${TESTDIR}/tmp"
rm -rf ${TMPDIR}
mkdir -p ${TMPDIR}

>${TESTDIR}/running
stats &

(
  cd ${TESTDIR}/${BUILDDIR}/mysql-test

  perl mysql-test-run.pl ... --parallel=8 --vardir=/ramdisk/mtr-${BRANCH}/...
  mv /ramdisk/mtr-${BRANCH}/* ${TMPDIR}/
  ...
)

sync
rm -f ${TESTDIR}/running
wait

First I graphed a straight run of the two branches, using the following gnuplot script:

set terminal png enhanced font "Times,11" size 640,768
set output "mtr-ram.png"
set title "MTR memory usage (8-core Xeon, 24GB, 8GB RAM disk)"
set xlabel "Time (minutes)"
set ylabel "Memory usage (GB)"
set yrange [0:16]
set xtics 10
set key top box
set grid
plot "stats-mysql-trunk" every 60 using (($1)/60):(($2)/1024/1024) \
        title 'mysql-trunk /ramdisk usage' with lines, \
     "stats-mysql-trunk" every 60 using (($1)/60):(($3)/1024/1024) \
        title 'mysql-trunk RAM (inc buf)' with lines, \
     "stats-mysql-trunk" every 60 using (($1)/60):(($4)/1024/1024) \
        title 'mysql-trunk RAM (exc buf)' with lines, \
     "stats-mysql-cluster-7.1" every 60 using (($1)/60):(($2)/1024/1024) \
        title 'cluster-7.1 /ramdisk usage' with lines, \
     "stats-mysql-cluster-7.1" every 60 using (($1)/60):(($3)/1024/1024) \
        title 'cluster-7.1 RAM (inc buf)' with lines, \
     "stats-mysql-cluster-7.1" every 60 using (($1)/60):(($4)/1024/1024) \
        title 'cluster-7.1 RAM (exc buf)' with lines

I then performed a valgrind run on mysql-trunk using similar scripts. As valgrind takes considerably longer (and uses more RAM) I kept it separate as the combined graph isn’t very clear:

So, based on these results, the host machine (16GB RAM + 8GB RAM disk) is probably a sensible guide for now, and allows for some future growth.

Fix input box keybindings in Firefox

jonathan@perkin.org.uk (Jonathan Perkin) — Wed, 29 Jun 2011 00:00:00 +0100

Those of us used to command line editing will no doubt have been frustrated many times in Firefox when editing text in an input box and subconciously hitting ctrl-w to delete-word, only to have the tab close and your work deleted.

Thankfully there is a workaround to this. It used to be a case of adding the following to .gtkrc:

gtk-key-theme-name = "Emacs"

However these days it’s a gconf setting:

$ gconftool-2 --set /desktop/gnome/interface/gtk_key_theme Emacs --type string

This will bind ctrl-w to delete-word when in an input box, but retain the close tab binding elsewhere, a nice implementation of DWIM. See this page for more information.

How to lose weight

jonathan@perkin.org.uk (Jonathan Perkin) — Fri, 24 Jun 2011 00:00:00 +0100

This year I set myself a target to get back to my pre-marital weight. Back in 2001 I was 90kg, but over the past decade I’d steadily gone up and for quite a few years had hovered around 100kg.

I’m a bit of a perfectionist, and like shiny graphs, so from Jan 1st this year I weighed myself every day on the Wii Fit, and kept a record of my progress. This morning I was finally under 90kg. Here’s how things have gone thus far:

And for the geeks, my gnuplot script, assuming an input file named weight.txt containing lines of the format "%Y-%m-%d <weight>":

set terminal png size 640, 400
set output "weight-jun11.png"
set xdata time
set timefmt "%Y-%m-%d"
set format x "%b"
set xlabel 'Date'
set ytics 2
set ylabel 'Weight (kg)'
set multiplot
plot 'weight.txt' using 1:2 title 'Weight' with lines linecolor 2

My three step plan for losing weight

The interwebs are awash with a million different ways to lose weight, and there are thousands of companies who will gladly accept large chunks of your cash in order to provide you with their expert opinion on how to do it. This is my three-step plan, and you can have it for free:

Eat less.
Eat less!
Profit!!!

Really, that’s it. It’s not rocket science, after all. If you eat less stuff (especially saturated fat and sugar), there is less excess for your body to store as fat, and thus over time you weigh less as your body starts to use up the excess you have stored.

Here’s some things I did to achieve this.

Eat less

The basic aim. For me this was mainly about reducing my portion size.

Previously, I’d eat a massive bowl of cereal in the morning. which I thought was a good thing as everyone always says to ensure you have a good breakfast. However I was eating more than my body needed and as a consequence it was likely storing up the excess.

Also at meal times I’d usually serve myself a full plate, and always feel that I needed to finish it. I liked feeling full. However, again, this likely just meant I was eating more than I needed, and I found that just by serving less food I was still satisfying my hunger but with less excess.

After a relatively short period of time I found that by avoiding these large meals, I needed less to feel full – as if my stomach had shrunk and gotten used to the reduced size. This was a great positive feedback loop, as it massively helped avoid the temptation to snack between meals.

Cut down on sugar/fat

I realise this can be hard for some people, I found it relatively easy but as you can see on the graph above there are some upward trends, which were mainly when I went to visit my parents, who have a cupboard full of chocolates, crisps, cakes, sweets, etc, and I find it really hard not to have at least one or two fake Lidl snickers bars per day!

However, also note that after an upward trend, I lost the weight again pretty quickly, so don’t worry too much about a few days of eating junk, I actually noticed how bad I felt afterwards after getting used to a reduced sugar/fat intake, and that provided good incentive to cut down again.

Some practical things I did:

Stopped putting sugar in tea/coffee. I’ve since regressed, but it was helpful to do this for a while, and definitely helped wean my body off desiring sugar.
Avoided snacking on crisps, chocolate, biscuits, etc. When you work from home this can be difficult, but I found it helped to ensure they weren’t in the house to begin with, and that we were stocked up on bananas, apples, and other less sugary/fatty snacks.
Used a fine cheese grater when making sandwiches, beans on toast, etc. Previously I’d put a good few slabs of cheese in, when I didn’t really need that quantity. The fine grater ensured I still got the taste, but with less quantity.
Bought reduced fat mayonnaise, margarine, etc. These actually taste pretty good these days, and in a sandwich with lots of fresh cucumber, tomato, salad etc you don’t notice the difference

All these things helped to lose weight and, in a similar manner to the ‘eating less’ part, trained my body to not require them as much as it used to – and actually, to noticeably feel worse if ever I regressed, which made getting back on track very easy.

Weigh myself daily

It seems this is generally not recommended, and advice is that you should weigh yourself weekly (at the same time each week). Logically this doesn’t make sense to me, as a geek it’s obvious that the more data points you have, the better. Your body weight can fluctuate quite a lot from day to day especially with regards to how much liquid you have drunk, and if you are only weighing yourself once per week you could get caught out by a daily spike.

Aside from the geek factor of a better graph resolution, I also prefer daily weighing as it provides me, as a perfectionist, with incentives both ways. If I weigh less compared to the previous day, I feel good that I am achieving my aim, and am encouraged to continue. If I weigh more, it’s a warning that I may have eaten too much, and I am then motivated to be more careful that day.

Weighing daily also helps to just keep your mind reminded of the task, plus if you’re on the Wii Fit already you might be tempted to do some exercise :)

Next steps

I’m not content to stay here, the next plan is to get closer to my ideal BMI weight which is around 80kg. This is likely to be much harder, as there is less excess fat for me to get rid of now. However, one thing I haven’t done so far is increase the amount of exercise I do, and there is definitely room for improvement there!

Hopefully I can provide another update later this year and perhaps be around 85kg.

How to fix stdio buffering

jonathan@perkin.org.uk (Jonathan Perkin) — Thu, 23 Jun 2011 00:00:00 +0100

It’s a common problem. You write some shell command like:

$ tail -f /var/log/foo | egrep -v 'some|stuff' | sed | awk

and wonder why nothing is printed, even though you know some text has matched. The problem is that stdio is being buffered, and there’s a very good write-up of the problem here so I won’t repeat the technical background.

What I will provide though is how to fix it for common cases.

stdbuf

stdbuf is part of GNU coreutils, and is essentially an LD_PRELOAD hack which calls setvbuf() for an application. Thus it is a generic solution to the problem and can be used to fix most applications. Usage looks like this:

$ tail -f /var/log/foo | stdbuf -o0 app ...

which will disable output buffering for app, assuming it does not do something itself to reverse the setvbuf() call. An example of a misbehaving application is mawk, below.

awk

GNU awk needs no modifications, that is it does not buffer when there is no controlling tty.

mawk however (the default awk in Debian/Ubuntu and possibly others) buffers output, and also does not seem to work with stdbuf. It does however provide a -Winteractive option which will turn off buffering.

$ tail -f /var/log/foo | gawk

$ tail -f /var/log/foo | mawk -Winteractive

sed

GNU sed provides the -u option which calls fflush(), thereby providing unbuffered output. You can also use stdbuf as above.

$ tail -f /var/log/foo | sed -u

$ tail -f /var/log/foo | stdbuf -o0 sed

grep

Similar to sed, GNU grep provides a specific option, --line-buffered, to disable buffering, or again you can use stdbuf.

$ tail -f /var/log/foo | grep --line-buffered

$ tail -f /var/log/foo | stdbuf -o0 grep

Serving multiple DNS search domains in IOS DHCP

jonathan@perkin.org.uk (Jonathan Perkin) — Mon, 13 Jun 2011 00:00:00 +0100

I have a Cisco router at home which I also use as a DHCP server, and it works pretty well. Today I wanted to fix a long-standing issue on my network, in that I want multiple DNS search domains.

First off, the domain-name DHCP option doesn’t support multiple entries so we can’t use that.

So, off to try raw DHCP option codes. You can find the list of options here:

http://www.iana.org/assignments/bootp-dhcp-parameters/bootp-dhcp-parameters.xml

thus 119 is the one I want.

Trying a simple:

ip dhcp pool host.net.example.com
   option 119 ascii net.example.com,example.com

didn’t work at all. A quick prod of lazyweb (in this case Simon on IRC) suggested using hex input instead. In order to do that we need to convert the ASCII string into Cisco’s hex sequence, which is as follows:

Split domain name by dot
Prepend each string by its length (in hex)
NUL terminate each domain
Dot-seperate the final string in 16bit chunks

To do this I wrote a quick Python script:

#!/usr/bin/python

import sys

hexlist = []
for domain in sys.argv[1:]:
    for part in domain.split("."):
        hexlist.append("%02x" % len(part))
        for c in part:
            hexlist.append(c.encode("hex"))
    hexlist.append("00")

print "".join([(".%s" % (x) if i and not i % 2 else x) \
               for i, x in enumerate(hexlist)])

which can be used like this:

$ ./cisco.py net.example.com example.com
036e.6574.0765.7861.6d70.6c65.0363.6f6d.0007.6578.616d.706c.6503.636f.6d00

Then back to IOS and:

ip dhcp pool host.net.example.com
   option 119 hex 036e.6574.0765.7861.6d70.6c65.0363.6f6d.0007.6578.616d.706c.6503.636f.6d00

seems to do what we want, though IOS appears to append a dot to each domain when serving via DHCP.

One last note, if you use this in addition to domain-name then the option 119 list will be appended to the domain-name name in the search list, so you’d actually want something like this:

ip dhcp pool host.net.example.com
   domain-name net.example.com
   option 119 hex 0765.7861.6d70.6c65.0363.6f6d.00

to generate a resolv.conf containing:

domain net.example.com
search net.example.com example.com.

Fix Firefox URL double click behaviour

jonathan@perkin.org.uk (Jonathan Perkin) — Mon, 13 Jun 2011 00:00:00 +0100

One of the best things I find with OS X is the consistent and sane handling of keyboard and mouse bindings. Since having to move back to Linux for work I am constantly frustrated that different applications all have their own idea of what shortcuts to use.

One of the biggest annoyances was the mouse click behaviour in the browser URL bar. OS X behaviour is one click to position cursor, two clicks to select a word, three clicks to select all. I rely on this behaviour a lot as I frequently copy/paste parts of URLs.

Thankfully that’s one that can be fixed easily. Simply navigate to about:config and double click on the browser.urlbar.doubleClickSelectsAll key to set it to false.

One annoyance down, lots more to go..

SSH via HTTP proxy in OSX

jonathan@perkin.org.uk (Jonathan Perkin) — Wed, 20 Apr 2011 00:00:00 +0100

If you happen to be stuck behind a corporate firewall with only HTTP proxies for external access, you might still be able to SSH out through them using the built-in nc on OSX.

First, hope that the proxies haven’t disabled the CONNECT method, then simply add a section to your .ssh/config like this:

Host foobar.example.com
    ProxyCommand          nc -X connect -x proxyhost:proxyport %h %p
    ServerAliveInterval   10

This will tunnel the connection through the HTTP proxy to the remote server. The ServerAliveInterval setting is required as most proxies will drop the connection after a period of inactivity.

To avoid issues with trying to connect to the host when not behind the corporate firewall, replace the above with a fake entry for the proxy method like this:

Host foobar-proxy.example.com
    HostName              foobar.example.com
    ProxyCommand          nc -X connect -x proxyhost:proxyport %h %p
    ServerAliveInterval   10

Then use

$ ssh foobar-proxy.example.com

when inside the firewall, and

$ ssh foobar.example.com

when outside.

Simples, no external software required.

How to build MySQL releases

jonathan@perkin.org.uk (Jonathan Perkin) — Tue, 09 Nov 2010 00:00:00 +0000

One of the major benefits from the CMake work available in MySQL 5.5 is that in the MySQL Release Engineering team we have been able to make it easy for users and developers to build MySQL exactly as we do for the official releases. For too long there has been a disconnect between the binaries produced as part of a regular ’./configure; make’ build and what we ship to users and customers.

We’re still not exactly where we want to be, there are still some parts which need to be integrated into the server tree, but for now it’s relatively straightforward to build exactly as we do.

Here are the instructions, using the mysql-5.5.6-rc.tar.gz source tarball as an example.

tar.gz

These are the generic instructions to build a tarball release. Note that we make use of CMake’s out-of-srcdir support to

ensure the source directory is kept pristine
allow us to easily build both debug and release binaries, and package them together

#!/bin/sh

tar -zxf mysql-5.5.6-rc.tar.gz

# Build debug binaries first, they are picked up in the final 'make package'
mkdir debug
(
  cd debug
  cmake ../mysql-5.5.6-rc -DBUILD_CONFIG=mysql_release -DCMAKE_BUILD_TYPE=Debug
  make VERBOSE=1
)

# Build release binaries and create final package
mkdir release
(
  cd release
  cmake ../mysql-5.5.6-rc -DBUILD_CONFIG=mysql_release
  make VERBOSE=1 package
)

Assuming everything goes ok, you should end up with a tarball in the release/ directory.

Some platforms require extra flags to be specified on the CMake command line to ensure the correct compiler etc is used. Here are some that we use.

Avoid libstdc++ dependancy

In order to create ‘generic’ binaries, on GCC platforms we compile using CXX=gcc. This avoids libstdc++ being pulled in, and means that the server will run across a larger range of releases as you do not rely on having the exact version of libstdc++ installed by your system package manager.

Paste this before running cmake.

# CXXFLAGS is required to trick CMake into believing it is a C++ compiler
$ CXX=gcc; CXXFLAGS="-fno-exceptions"; export CXX CXXFLAGS

Specify target architecture

On systems which support multiple targets, you can specify exactly which one you want, rather than relying on the OS default. These strings are meant to be added to the cmake command lines above.

OSX

$ -DCMAKE_OSX_ARCHITECTURES="i386"    # 32bit
$ -DCMAKE_OSX_ARCHITECTURES="x86_64"  # 64bit

GCC/Sun Studio

$ -DCMAKE_C_FLAGS="-m32" -DCMAKE_CXX_FLAGS="-m32"  # 32bit
$ -DCMAKE_C_FLAGS="-m64" -DCMAKE_CXX_FLAGS="-m64"  # 64bit

HP/UX

$ -DCMAKE_C_FLAGS="+DD64" -DCMAKE_CXX_FLAGS="+DD64"

RPM

We’ve spent a lot of time improving the RPM builds too. Previously some of the configuration was only available in the commercial RPM builds, but now we have merged them into the community version.

#!/bin/sh

mkdir -p rpm/{BUILD,RPMS,SOURCES,SPECS,SRPMS} tmp

# Create spec file.
# XXX: We should probably just include this in the source tarball.
tar -zxf mysql-5.5.6-rc.tar.gz
(
  mkdir bld; cd bld
  cmake ../mysql-5.5.6-rc
)

cp bld/support-files/*.spec rpm/SPECS
cp mysql-5.5.6-rc.tar.gz rpm/SOURCES

rpmbuild -v --define="_topdir $PWD/rpm" --define="_tmppath $PWD/tmp" \
 -ba rpm/SPECS/mysql.5.5.6-rc.spec

You should end up with nice shiny RPMs in rpm/RPMS. One thing to note is that the RPM spec no longer runs the test suite as part of the build, so you will need to run that separately. On the plus side, you get your RPMs much quicker.

Another nice thing about the improved RPM spec is that you can now build targetted distribution RPMs, as we do. These have extra dependancy information in them tailored to the target distribution. Currently the spec file supports the distributions we build on, but we will accept patches for others.

To enable this, use:

$ rpmbuild -v --define="distro_specific 1" ...

Windows

Our Windows builds have relied on CMake since MySQL 5.0, but the procedure has still changed to ensure that you can build as we do. These instructions use cmd.exe but you can use the Visual Studio front end if you prefer.

unzip mysql-5.5.6-rc.zip

rem There is no separate 'debug' directory on Windows, as the CMake
rem infrastructure doesn't yet know to pull files in from there on Windows.
md release
cd release

rem Choose your target architecture, 32bit
set VSTARGET=Visual Studio 9 2008
rem or 64bit
set VSTARGET=Visual Studio 9 2008 Win64

cmake ..\mysql-5.5.6-rc -DBUILD_CONFIG=mysql_release \
 -DCMAKE_BUILD_TYPE=Debug -G "%VSTARGET%"

devenv MySql.sln /build Debug

cmake ..\mysql-5.5.6-rc -DBUILD_CONFIG=mysql_release \
 -DCMAKE_BUILD_TYPE=RelWithDebInfo -G "%VSTARGET%"

You can now choose which type of package to create. 5.5 includes new code to create minimal MSI packages, these should work ok, and only differ from the official MySQL MSI packages in that they do not include the instance config wizard.

To create the MSI packages you will need to install WiX.

rem Standard zip package
devenv MySql.sln /build RelWithDebInfo /project package

rem Full MSI package
devenv MySql.sln /build RelWithDebInfo /project msi

rem 'Essentials' MSI package
devenv MySql.sln /build RelWithDebInfo /project msi_essentials

Assuming everything goes ok, you should have some packages in the release\ directory.

Work still to be done

We still have a number of scripts only available internally, for example those we use to create SVR4, DMG and DEPOT packages. However, we are looking to integrate these into the MySQL Server source tree so that all users can benefit from them.

Enjoy!

'apt-get' and 5,000 packages for Solaris10/x86

jonathan@perkin.org.uk (Jonathan Perkin) — Thu, 29 Apr 2010 00:00:00 +0100

Here’s how:

# Install pkg_* tools and the 'pkgin' package manager
$ pkgadd -d http://www.netbsd.org/~sketch/TNFpkgsrc-x86.pkg all

# Add tools to PATH
$ PATH=/opt/pkg/sbin:/opt/pkg/bin:$PATH

# Update package repository (akin to 'apt-get update')
$ pkgin up

# Search for a particular package (you can use regexp)
$ pkgin search ^ap.*python 

# Install it
$ pkgin install ap22-py25-python

# Update all packages (akin to 'apt-get dist-upgrade')
$ pkgin full-upgrade

# How many packages are available?
$ pkgin avail | wc -l
   4970

Ok, so the headline might be slightly mis-leading, this isn’t really apt-get but a tool which is very similar. This is work which builds upon my previous post using pkgsrc to build binary packages on Solaris.

See http://imil.net/pkgin/ for more information on pkgin.

Hopefully this will prove really useful to people still using Solaris 10 and unable to use the new pkg(5) stuff in OpenSolaris.

Please try it out and provide any feedback to pkgsrc-users@netbsd.org. I’m hoping to keep the packages updated for the 2009Q3 branch.

ZFS+NFS vs OSX

jonathan@perkin.org.uk (Jonathan Perkin) — Wed, 16 Sep 2009 00:00:00 +0100

gromit is my Solaris 10 server, chorlton is my OSX desktop. Explain this:

root@gromit# zfs set sharenfs='anon=shared,sec=none' gromit/store

root@gromit# zfs get sharenfs gromit/store
NAME          PROPERTY  VALUE                 SOURCE
gromit/store  sharenfs  anon=shared,sec=none  local

user@chorlton$ ls /net/gromit/store
ls: cannot open directory /net/gromit/store/: Operation not permitted

root@gromit# zfs set sharenfs='anon=shared' gromit/store

user@chorlton$ ls /net/gromit/store
file1  file2

user@chorlton$ touch /net/gromit/store/file3
touch: /net/gromit/store/file3: Permission denied

root@gromit# zfs set sharenfs='anon=shared,sec=none' gromit/store

user@chorlton$ touch /net/gromit/store/file3
user@chorlton$ echo $?
0

pkgsrc on Solaris

jonathan@perkin.org.uk (Jonathan Perkin) — Sat, 12 Sep 2009 00:00:00 +0100

For many years, building and installing third-party software on Solaris has been a huge pain. For people who do not use pkgsrc, that is.

Originating from the NetBSD project, pkgsrc is a source-based cross-platform package manager. If you’ve used FreeBSD ports, then it is very similar as it derives from the same codebase, so the basic premise is that you:

$ cd /usr/pkgsrc/www/apache22
$ make package

and pkgsrc will download the source for Apache 2.2, compile it and all dependancies, install it on the local system and create a package which can be installed on other similar systems.

However, we’ve taken ports further and applied the NetBSD philosophy of portability, meaning that it not only works on NetBSD, but across all *BSD as well as Linux, OSX, HP/UX, AIX, IRIX, QNX, Windows (via Interix), and of course Solaris.

So while apt-get might be awesome, it only really works on Linux. FreeBSD might have way more ports than us, but only runs on FreeBSD and OSX. pkgsrc provides a consistent interface across all the platforms listed above, and in some cases provides a superior package manager than the system provides.

Here’s how I use pkgsrc on Solaris, in this specific case Solaris 10/x86. Paths are specific to my setup, you can of course change them.

Create a chroot/zone environment

I use the zones feature of Solaris 10 to ensure that all packages are built in a sandbox. This has a number of benefits:

The running system is unaffected by the builds, in that they are not writing to the same file system. This is good when you have misbehaving packages.
You can separate the build and install phases, so that you can verify all the packages have been built and are correct before starting any install/upgrade procedure
It’s easier to catch package mistakes, e.g. unpackaged files.
It avoids pollution from the host environment which may produce bad packages

For creating a zone, I wrote the following create-zone script:

#!/bin/sh

DOMAIN="adsl.perkin.org.uk"
MASQDOMAIN="perkin.org.uk"
PHYSIF="e1000g0"
RPOOL="gromit"

if [ $# -eq 1 ]; then
    name=$1; shift
    ipaddr=`getent hosts ${name} | awk '{print $1}'`
    if [ -z "${ipaddr}" ]; then
        echo "ERROR: Could not determine IP address of $name"
        echo "Either add to hosts database or provide on command line"
        echo
        echo "usage: $0 <name> [ <ipaddr> ]"
        exit 2
    fi
elif [ $# -eq 2 ]; then
    name=$1; shift
    ipaddr=$1; shift
else
    echo "usage: $0 <name> [ <ipaddr> ]"
    exit 2
fi

zfs list ${RPOOL}/zones/$name >/dev/null 2>&1
if [ $? -eq 0 ]; then
    echo "${RPOOL}/zones/$name already exists, not continuing"
    exit 1
fi

# Inherited directories are read-only
cat >/tmp/zonecfg.$$ << EOF
    create
    set zonepath=/zones/$name
    set autoboot=true
    add net
        set address=$ipaddr/24
        set physical=${PHYSIF}
    end
    verify
    commit
    exit
EOF

zonecfg -z $name -f /tmp/zonecfg.$$
rm /tmp/zonecfg.$$

zoneadm -z $name install

#
# Automatically configure new zone with most of the settings from the current
# global, grabbing the encrypted root password directly.  Note that the
# service_profile keyword doesn't seem to be used here, so we configure the
# limited_net profile manually and optionally install a custom site.xml which
# is parsed and activated during first boot.
#
cat << EOF >/zones/$name/root/etc/sysidcfg
network_interface=PRIMARY
{
    hostname=$name.${DOMAIN}
}
name_service=DNS
{
    domain_name=${DOMAIN}
    name_server=`awk '/^nameserver/ { print $NF; exit }' /etc/resolv.conf`
}
nfs4_domain=dynamic
root_password=`awk -F: '/^root/ {print $2}' /etc/shadow`
security_policy=NONE
service_profile=limited_net
system_locale=C
terminal=xterm
timezone=Europe/London
EOF

#
# Use the limited_net profile and install custom site.xml if provided.
#
rm /zones/${name}/root/var/svc/profile/generic.xml
ln -s generic_limited_net.xml /zones/${name}/root/var/svc/profile/generic.xml
if [ -f /install/zones/${name}.xml ]; then
    cp /install/zones/${name}.xml /zones/${name}/root/var/svc/profile/site.xml
fi

#
# Disable nscd host cache
#
ex /zones/$name/root/etc/nscd.conf >/dev/null 2>&1 <<EOF
/enable-cache/s/^#//
wq
EOF

#
# Configure sendmail to masquerade and route via smarthost.
#
ex /zones/$name/root/etc/mail/cf/cf/submit.mc >/dev/null 2>&1 <<EOF
/^dnl/
a
FEATURE(\`masquerade_envelope')dnl
MASQUERADE_AS(\`${MASQDOMAIN}')dnl
dnl
.
/^FEATURE.*msp/s/.127.0.0.1./mail.perkin.org.uk/
wq!
EOF
(cd /zones/$name/root/etc/mail/cf/cf; /usr/ccs/bin/make submit.cf >/dev/null 2>&1)
cp /zones/$name/root/etc/mail/cf/cf/submit.cf /zones/$name/root/etc/mail/submit.cf

#
# Use SHA512 crypted passwords, and change root's home directory.
#
ex /zones/$name/root/etc/security/policy.conf >/dev/null 2>&1 <<EOF
/^CRYPT_DEFAULT/s/__unix__/6/
wq
EOF
mkdir -m 0700 /zones/$name/root/root
ex /zones/$name/root/etc/passwd >/dev/null 2>&1 <<EOF
/^root/s,:/:,:/root:,
/^root/s,:/sbin/sh,:/bin/bash,
wq
EOF

cp /root/.bash_profile /zones/$name/root/root/.bash_profile

#
# Allow root login over SSH (rest of network takes care of external access
# so this isn't a problem) and enforce SSH key authentication.
#
ex /zones/$name/root/etc/ssh/sshd_config >/dev/null 2>&1 <<EOF
/^PermitRootLogin/s/no$/yes/
/^PasswordAuthentication/s/yes$/no/
wq
EOF
mkdir -m 0700 /zones/$name/root/root/.ssh
cp /home/jperkin/.ssh/local_rsa.pub /zones/$name/root/root/.ssh/authorized_keys

#
# Automount shared directories from global zone using direct mounts
#
ex /zones/$name/root/etc/auto_master >/dev/null 2>&1 <<EOF
a
/-        auto_direct
.
wq
EOF
ex /zones/$name/root/etc/auto_direct >/dev/null 2>&1 <<EOF
a
/content    gromit-lan:/content
/install    gromit-lan:/install
.
wq
EOF

#
# All done!  Fire it up...
#
zoneadm -z $name boot

And the corresponding delete-zone script is:

#!/bin/sh

RPOOL="gromit"

if [ $# -ne 1 ]; then
    echo "usage: $0 <zone>"
    exit 2
fi

zoneadm -z $1 halt
zonecfg -z $1 delete -F
zfs destroy ${RPOOL}/zones/$1

If you want to use them then there are some variables to set at the top, and you may want to scan through them for additional bits to change, for example create-zone copies my ssh public key which will most likely be wrong for your setup :-)

One additional piece of configuration for create-zone is an optional SMF xml file. I use this file to disable inetd inside the zone for additional security, like so:

<?xml version='1.0'?>
<!DOCTYPE service_bundle SYSTEM '/usr/share/lib/xml/dtd/service_bundle.dtd.1'>
<service_bundle type='profile' name='extract'>
  <service name='network/inetd' type='service' version='0'>
    <instance name='default' enabled='false'/>
  </service>
</service_bundle>

The file should be named <yourzonename>.xml. Mine is named vm-generic.xml and I then create symlinks to it for each VM I want with that default configuration.

Fetch pkgsrc

pkgsrc is developed very rapidly. Tracking nearly 9,000 pieces of third-party software means there are always many updates. Thankfully, we provide quarterly branches for people who want more stability, and I recommend using the latest quarterly release. At time of writing, this is known as pkgsrc-2009Q2. Within the next month or so we will release pkgsrc-2009Q3, and you can figure out the names of future releases yourself.

The easiest way to get pkgsrc is using cvs. I keep stuff like this under /content as opposed to the default of /usr, you can use whatever you wish but will need to change all my example scripts to match where you put it.

$ cd /content
$ BRANCH="pkgsrc-2009Q2"
$ cvs -d anoncvs@anoncvs.netbsd.org:/cvsroot co -r${BRANCH} -d${BRANCH} -P pkgsrc

Alternatively, you can fetch either bzip2 or gzip archives of the current branch. I recommend the cvs method as, with the branch being updated for security fixes and other important changes, you can easily track it using

$ cd /content/pkgsrc-2009Q2
$ cvs update

pkgsrc configuration

pkgsrc is configured using a mk.conf file, this is mine:

#
# Local settings
#
OS_HOST!= uname -n

#
# Per-host settings
#
#    vm0 builds local packages
#
.if !empty(OS_HOST:Mvm0.*) || !empty(OS_HOST:Mgromit.*)
PKGDIRSUFFIX=
PKG_DEFAULT_OPTIONS=    inet6 perl sasl ssl ncursesw
PKG_OPTIONS.mutt+=      mutt-hcache mutt-smtp
PKG_UID.cyrus=          2000
#PKG_GID.cyrus=         mail
PKG_SHELL.cyrus=        /usr/bin/false
PKG_UID.www=            2001
PKG_GID.www=            2001
PYTHON_VERSION_DEFAULT= 24
X11_TYPE=               modular
#
#    vm1 modular X11 bulk builds
#
.elif !empty(OS_HOST:Mvm1.*)
X11_TYPE=       modular
PKGDIRSUFFIX=   -${X11_TYPE}
#
#    vm2 native X11 bulk builds
#
.elif !empty(OS_HOST:Mvm2.*)
X11_TYPE=       native
PKGDIRSUFFIX=   -${X11_TYPE}
#
#    vm3 pkgsrc dev builds
#
.elif !empty(OS_HOST:Mvm3.*)
X11_TYPE=       modular
PKGDIRSUFFIX=   -${X11_TYPE}
.endif

.if exists(../../CVS/Tag)
PKG_BRANCH!=    ${TOOLS_PLATFORM.awk} -F- '{print $$2}' < ../../CVS/Tag
.else
PKG_BRANCH=     HEAD
.endif

ALLOW_VULNERABLE_PACKAGES=  YES
BULKFILESDIR=               ${WRKOBJDIR}
DISTDIR=                    /install/pkgsrc/distfiles/
FAILOVER_FETCH=             YES
INSTALL_UNSTRIPPED=         YES
LINTPKGSRC_DB=              ${_PKGSRCDIR}/.lintpkgsrc.db
MASTER_SITE_OVERRIDE=       ftp://ftp.netbsd.org/pub/NetBSD/packages/distfiles/
PACKAGES=                   /install/pkgsrc/packages/${PKG_BRANCH}${PKGDIRSUFFIX}
PKGCHK_CONF=                /install/pkgsrc/misc/pkgchk.conf
SKIP_LICENSE_CHECK=         YES
WRKOBJDIR=                  /tmp/pkgsrc

#
# Parse pkgchk.conf and supply list of packages for the bulk build framework.
#
.if defined(SPECIFIC_PKGS)
PKGLIST!= ${TOOLS_PLATFORM.awk} '$$1 !~ /^(\#|$$)/ {print $$1}' ${PKGCHK_CONF}
.  for _pkg_ in ${PKGLIST}
HOST_SPECIFIC_PKGS+= ${_pkg_}
.  endfor
.endif

This is the primary configuration file for pkgsrc. Again, you may need to tailor this to your environment, and may find it useful to read the pkgsrc guide to understand what it all means.

As I do a lot of pkgsrc development I have a number of virtual machines up and running doing various bits and pieces. Obviously I don’t want to copy that mk.conf around, so I also have a small fragment file which is appended to each virtual machine’s mk.conf (using the --mk-fragment argument to bootstrap) and includes the global copy:

.include "/install/pkgsrc/misc/mk.conf"

The bulk build setup in pkgsrc requires its own configuration, and for this you will need to edit a file inside pkgsrc. There is an example file provided, so what I usually do is symlink this to the real copy then I can easily keep it up-to-date via cvs.

$ cd /content/pkgsrc-2009Q2/mk/bulk
$ ln -s build.conf-example build.conf
$ vi build.conf

Again, you can find my personal build.conf here.

Finally, there is a configuration file for pkg_chk which is a package inside pkgsrc which makes managing upgrades easier (ideally it should be a part of the main pkgsrc tools but that’s for another day). pkgchk.conf is a list of package directories, relative to the pkgsrc top level, which are to be built and installed for this setup. If you have a large installation then pkg_chk has extra features to make it possible to share pkgchk.conf across a number of machines and configure packages on a per-host, per-OS etc basis.

Thus, a sample pkgchk.conf:

chat/irssi
chat/irssi-icb
converters/xlhtml
databases/lbdb
databases/rrdtool
devel/autoconf
devel/automake
devel/bzr
devel/bzrtools
devel/diffutils
devel/libtool-base
devel/scmcvs
editors/vim
mail/mutt-devel
misc/screen
net/p5-Net-SNMP
net/p5-Net-SNMP-Interfaces
net/rsync
net/samba
net/tnftp
pkgtools/digest
pkgtools/pkg_chk
pkgtools/rc.subr
print/mp
security/cy2-login
security/cy2-plain
security/sudo
sysutils/coreutils
textproc/antiword
textproc/grep
textproc/urlview
www/apache22
www/w3m

It is highly likely you will want to change the pkgchk.conf file from what I use :-)

Build scripts

Once everything is set up, I have two scripts to build then update my packages, intuitively called build-packages:

#!/bin/sh

PATH="/opt/pkg/sbin:/opt/pkg/bin:/sbin:/usr/sbin:/usr/bin:/usr/ccs/bin"
export PATH

BRANCH="2009Q2"
export BRANCH

if [ ! -f /install/pkgsrc/bootstrap-${BRANCH}.tar ]; then
    cd /content/pkgsrc-${BRANCH}/bootstrap
    ./bootstrap --workdir=/tmp/pkgsrc --prefix=/opt/pkg --varbase=/var/opt/pkg --sysconfdir=/etc/opt/pkg --mk-fragment=/install/pkgsrc/misc/mk-include.conf --binary-kit=/install/pkgsrc/bootstrap-${BRANCH}.tar
    rm -rf /tmp/pkgsrc
fi

if [ ! -f /opt/pkg/bin/bmake ]; then
    (cd /; tar -xf /install/pkgsrc/bootstrap-${BRANCH}.tar)
fi

cd /content/pkgsrc-${BRANCH}

/bin/ksh mk/bulk/build -s  2>&1 | tee -a /install/pkgsrc/logs/bulk-build-${BRANCH}.log
/bin/ksh mk/bulk/upload -n 2>&1 | tee -a /install/pkgsrc/logs/bulk-upload-${BRANCH}.log

and update-packages:

#!/bin/sh

PATH="/opt/pkg/sbin:/opt/pkg/bin:/sbin:/usr/sbin:/usr/bin:/usr/ccs/bin"
export PATH

BRANCH=2009Q2
export BRANCH

PKGSRCDIR=/content/pkgsrc-${BRANCH}
export PKGSRCDIR

if [ ! -f /opt/pkg/bin/bmake ]; then
    (cd /; tar -xf /install/pkgsrc/bootstrap-${BRANCH}.tar)
fi

# Run this anyway in case there are updates to pkg_chk
env PKG_PATH=/install/pkgsrc/packages/${BRANCH}/All pkg_add -u pkg_chk

# Dry run first
pkg_chk -aurbn

printf "Do you wish to install? [y/N]: "
read ans
case "${ans}" in
[Yy])
    pkg_chk -aurb 2>&1 | tee -a /install/pkgsrc/logs/pkg_chk-${BRANCH}.log
    ;;
esac

These are pretty simple as all the hard work has all been done. build-packages is ran inside the zone, then update-packages on the main host. These scripts hardcode the name of the branch currently used, so you will need to update this when moving to newer releases.

Quick recap

Ok, so here is the stuff I have for my setup and where I keep them:

Path	Description
`/content/pkgsrc-2009Q2`	Checked out pkgsrc tree, "2009Q2" branch
`/content/scripts/create-zone`	Creates Solaris zone
`/content/scripts/delete-zone`	Uninstalls and deletes zone
`/content/scripts/build-packages`	Bulk build packages inside the zone
`/content/scripts/update-packages`	Updates installed packages
`/install/pkgsrc/misc/mk.conf`	Main pkgsrc configuration file
`/install/pkgsrc/misc/mk-include.conf`	Fragment file included in each zone's `mk.conf`, sources the global `mk.conf`
`/install/pkgsrc/misc/pkgchk.conf`	`pkg_chk` configuration file
`/install/zones/vm-generic.xml`	Shared SMF configuration file, symlinked to from e.g. "`vm0.xml`"

And these are the paths where stuff will be created:

Path	Description
`/install/pkgsrc/distfiles`	Source tarballs of packages
`/install/pkgsrc/packages/2009Q2`	Resulting binary packages
`/tmp/pkgsrc`	Temporary build area for packages
`/content/vwww/www.adsl.perkin.org.uk/pkgstat`	Bulk build results directory

It’s definitely harder than it should be to get this all setup, but the good news is that once it’s done there’s very little maintenance.

Kicking it all off

Once everything is setup:

$ /content/scripts/create-zone vm0
$ ssh vm0 /content/scripts/build-packages
$ /content/scripts/update-packages
$ /content/scripts/delete-zone vm0

This should do the lot. Once build-packages has finished you should, if you configured your email address in build.conf, get an email with the bulk build results which looks similar to this:

http://mail-index.netbsd.org/pkgsrc-bulk/2009/08/23/msg006883.html

A fuller report is available if you configure a web server to serve the pkgstat directory created by the bulk build, and this can help debug problems (again see the above URL for an example).

You will need to add /opt/pkg/sbin:/opt/pkg/bin to $PATH. Configuration files are in /etc/opt/pkg, and log files and metadata are kept in /var/opt/pkg.

This stuff should be obsolete

While this all works well for me, it’s pretty lame for users who just want to install packages and have stuff work. I’m working on providing regular updates of binary packages, including a SVR4 package of the bootstrap kit, so that in theory all a user needs to do is

$ pkgadd TNFpkgsrc.pkg
$ pkg_add apache22
# Upgrade all installed packages to latest releases
$ pkg_chk -aurb

I’m almost there, just needs some tidying up and regular builds. Please feel free to help out!

Jumpstart from OSX

jonathan@perkin.org.uk (Jonathan Perkin) — Tue, 09 Dec 2008 00:00:00 +0000

I recently built a new file server on which I planned to install Solaris 10 10/08. I’m not a fan of CD/DVD installs, so wanted to jumpstart via PXE, though I only had OSX handy.

Here’s how I installed my new machine (gromit.adsl.perkin.org.uk/192.168.1.10) from my iMac (192.168.1.30) over the local network.

Note that since writing this piece I’ve updated to Solaris 10 10/09, and have changed the examples to use that instead.

Step 1, Prepare File System

First off, create a dedicated file system which we can export our jumpstart configuration from. You can probably skip this and just use any existing file system but this way everything is self-contained and we avoid NFS exporting more than we need.

We use HFSX to ensure that the file system is case sensitive, HFS+ can cause problems with pkgadd(1M).

$ hdiutil create -size 1g -type SPARSE -fs HFSX -volname "install" install
$ hdiutil attach install.sparseimage -mountpoint /install

Next up, download and mount sol-10-u8-ga-x86-dvd.iso.

$ open sol-10-u8-ga-x86-dvd.iso

Step 2, NFS

Share /install and the DVD via NFS with the correct options. -alldirs allows clients to mount from any point within that file system (which jumpstart requires), and -maproot=root is also required by jumpstart. As this allows root-owned files to be created, make sure you understand the security risks.

$ sudo vi /etc/exports

/install                  -alldirs -maproot=root
/Volumes/SOL_10_1009_X86  -alldirs -maproot=root

$ sudo nfsd checkexports && sudo nfsd enable

Step 3, DHCP

For DHCP I happen to already use my Cisco router as a DHCP server on the local network, so added the following configuration:

ip dhcp pool gromit.adsl.perkin.org.uk
   host 192.168.1.10 255.255.255.0
   hardware-address xxxx.xxxx.xxxx
   bootfile /boot/grub/pxegrub
   next-server 192.168.1.30
   client-name gromit
   domain-name adsl.perkin.org.uk
   dns-server xxx.xxx.xxx.xxx
   default-router 192.168.1.1

however, given this is a guide for setting everything up under OSX I also tried using ISC DHCP on OSX to prove it can be done that way too.

I used pkgsrc to install it (I’ll add another blog some time showing how to set up pkgsrc)

$ cd /usr/pkgsrc/net/isc-dhcpd
$ sudo bmake package

And here is my DHCP configuration file in full:

option domain-name "adsl.perkin.org.uk";
option domain-name-servers xxx.xxx.xxx.xxx;
ddns-update-style none;
authoritative;
log-facility local7;

subnet 192.168.1.0 netmask 255.255.255.0 {
    option routers 192.168.1.1;
}

group {
    filename "/boot/grub/pxegrub";
    next-server 192.168.1.30;

    host gromit {
        hardware ethernet xx:xx:xx:xx:xx:xx;
        fixed-address 192.168.1.10;
        option host-name "gromit.adsl.perkin.org.uk";
    }
}

Finally, start DHCP with:

$ sudo /usr/pkg/sbin/dhcpd

Most parts of these configurations should be self-explanatory. The /boot/grub/pxegrub entry is important for our next step, and I’d recommend using that exact pathname for reasons explained later.

Step 4, TFTP

Now, enable the TFTP server which comes with OSX. I added the -s option so tftpd would chroot to the tftpboot directory, both for security reasons and also to ensure that paths specified as /path/to/file would work correctly (relative to /install/tftpboot).

I also changed the location of the tftpboot directory so that everything was self-contained within the UFS image. In previous attempts I didn’t do this and ran into problems with GRUB which I think are again caused by case-insensitive file systems.

$ sudo vi /System/Library/LaunchDaemons/tftp.plist

[...]
    <key>ProgramArguments</key>
    <array>
        <string>/usr/libexec/tftpd</string>
        <string>-i</string>
        <string>-s</string>
        <string>/install/tftpboot</string>
    </array>
[...]

$ mkdir /install/tftpboot
$ sudo launchctl load -w /System/Library/LaunchDaemons/tftp.plist

You can then create a test file and check that it’s working as you expect, using:

$ echo "testing" >/install/tftpboot/testfile
$ printf "verbose\ntrace\nget testfile\n" | tftp localhost
$ rm /install/tftpboot/testfile

Step 5, GRUB

Next up, configure PXE booting using GRUB. We need to copy the GRUB images and configuration from the Solaris install DVD then modify it for our environment:

$ rsync -av /Volumes/SOL_10_1009_X86/boot/grub /install/tftpboot/boot/
$ rsync -av /Volumes/SOL_10_1009_X86/boot/multiboot /install/tftpboot/sol10u8x/
$ rsync -av /Volumes/SOL_10_1009_X86/boot/x86.miniroot /install/tftpboot/sol10u8x/

As we are copying the boot files from the DVD, they come hardcoded with particular pathnames to e.g. the menu.lst file. While it may be possible to pass extra parameters to pxegrub and load this from a different path, I simply recommend doing as I do and replicating the /boot/grub/ path structure so that everything Just Works.

The menu.lst file includes kernel arguments and allows you to choose which type of install to perform at startup. My file listed below has 3 choices:

Unattended install using a graphical environment (if available). The ‘install’ keyword after the kernel instructs it to perform an unattended install, so long as it can find the necessary settings from sysidcfg etc.
As above, but force the use of the console and do not start a graphical environment (using the ‘nowin’ keyword)
A manual install, so you need to go through the steps of layout out disks, selecting packages, etc.

$ vi /install/tftpboot/boot/grub/menu.lst

default=0
timeout=60

title Solaris PXE Unattended Install
    kernel /sol10u8x/multiboot kernel/unix - install -B \
      install_media=192.168.1.30:/Volumes/SOL_10_1009_X86,\
      sysid_config=192.168.1.30:/install/jumpstart,\
      install_config=192.168.1.30:/install/jumpstart
    module /sol10u8x/x86.miniroot

title Solaris PXE Unattended Install (console)
    kernel /sol10u8x/multiboot kernel/unix - install nowin -B \
      install_media=192.168.1.30:/Volumes/SOL_10_1009_X86,\
      sysid_config=192.168.1.30:/install/jumpstart,\
      install_config=192.168.1.30:/install/jumpstart
    module /sol10u8x/x86.miniroot

title Solaris PXE Manual Install
    kernel /sol10u8x/multiboot kernel/unix -B \
      install_media=192.168.1.30:/Volumes/SOL_10_1009_X86
    module /sol10u8x/x86.miniroot

Anyone used to doing jumpstart but with RARP/bootparams will notice the symmetry between install_config etc in the GRUB configuration and similar options in /etc/bootparams. Make sure that the full kernel arguments are all on one line, and that there are no spaces in between the install_media=..,sysid_config=.. options.

Step 6, Jumpstart

Finally, set up your Jumpstart configuration. Here’s what I personally use, you may want something different:

$ mkdir /install/jumpstart
$ cd /install/jumpstart
$ vi sysidcfg

name_service=DNS
{
    domain_name=adsl.perkin.org.uk
    name_server=xxx.xxx.xxx.xxx
}
network_interface=PRIMARY
{
    default_route=192.168.1.1
    netmask=255.255.255.0
    protocol_ipv6=yes
}
nfs4_domain=dynamic
root_password=xxxxxxxx
terminal=xterm
timeserver=localhost
timezone=Europe/London
security_policy=NONE
service_profile=limited_net
system_locale=C

Ordinarily this file is processed using a check script available in the jumpstart_sample directory on the Solaris DVD, however this only works from a Solaris host. To create the rules.ok file, we need to strip out any comments and put entries on one line, then create the checksum (although this isn’t actually necessary).

$ vi rules # emacs sucks :)

hostname gromit.adsl.perkin.org.uk - profile -

$ cp rules rules.ok
$ echo "# version=2 checksum=$(cksum -o 2 rules | awk '{print $1}')" >> rules.ok

Machine profile. This gives me a full Solaris install (minus OEM stuff) on mirrored ZFS disks with additional dump/swap space (the defaults made dump a bit too small I found).

$ vi profile

install_type    initial_install
pool            store auto 4g 4g mirror c1t0d0s0 c1t1d0s0
bootenv         installbe bename sol10u8x
cluster         SUNWCall

Step 7, Make A Cup Of Tea

With everything set up you should be able to enable PXE booting in your BIOS and watch it automatically install. One small minor problem you may have if you don’t have a BIOS which allows you to hit F12 or similar and choose PXE booting for one boot only is that it will infinitely cycle through installing, rebooting, installing, rebooting.. until you change your boot options.

If this happens, I recommend making more cups of tea until you happen to return in time to change the BIOS settings. If you aren’t able to do this for a while, you may need to add the extra steps 8, 9 and 10 titled “Visit The Bathroom”.