vmiklos.hu
shameless self-promoting website
»Root
»Rejourn root
  • Sunday, 16 September 2012
    Frugalware history (Comments)

    https://lh5.googleusercontent.com/-72B3Mo4hOQs/UeegcVH4YZI/AAAAAAAADNU/IwX2_lLQt5k/s128/logo-new-big.png

    Given that Frugalware was founded by me, and recently James recently took over the project leadership, I think it makes sense to look back and summarize the past 6 years.

    The beginning

    It all started with this announcement about 6 years ago. It was a one-man show before, there was a manually written ChangeLog but even then there were already quite a few packages, so don’t ask me when I started hacking on this. Yes, normally there would be a first commit in git, but remember this was before git existed, and I hated centralized CVS so much that we didn’t use anything.

    Looking back, it was all quite lame. :-) I used a mail address called "mamajom" (English translation could be "momonkey"), tied to an ISP, with a lengthy signature at the end of every mail I sent and was using my IRC nick instead of my real one everywhere… OTOH, I made some decisions I’m happy about even today. The first four developers (Ádám Zlehovszky, Krisztián Vasas, Zsolt Szalai and me) were all Hungarian and despite of this, I forced every code, test and documentation to be in English, to possibly turn the project into an international one in the future. And that proved to very, very useful.

    It was between 0.1 and 0.2 that we (Krisztián, Zsolt and me) showed up in a Hungarian TV (video record) explaining Frugalware. This is something never happened later for some reason.

    The way we found a free hosting at a university I never attended may be also interesting. In the grammar school, our IT teacher found my Linux distribution idea interesting enough to support, and his friend was a teacher at the Eötvös Loránd University, which is at the same Budapest city where I live. So, with my friend Botond Balázs (who later was my witness at our wedding) we bought an ultra-slow old PC (some Pentium 1 with a single HDD) and we were extremely happy, as the uni offered to host it for free. I remember it ran some Slackware version, as Frugalware didn’t support i586… ;-)

    Early days

    Then some more developers joined and we started to package all the usual free software which is available in other distributions but I personally didn’t use. Think of GNOME, Xfce, OpenOffice.org and so on.

    Of course we were still lame, the announcements still were not spellchecked by someone native and knowledgeable enough, ending up in words like "splitted" or "optimalization". However, we started to use an SCM (Darcs, which was horribly slow, but at least not the centralized CVS crap). Another developer I should mention was Bence Nagy, he came up with the idea of how we should avoid duplication in FrugalBuilds, leading to the various package templates under source/include/ in the source tree.

    Then we went multiarch, Krisztián Hamar contributed the x86_64 port to the 0.3 release. At the same time, I invented syncpkg, which avoided having to build the same package on multiple architectures manually.

    We also got Gábor Lőcsei (later we met IRL and completed several bike marathons together, resulting in a great friendship) who helped users in general on IRC, but more importantly, for some time he did quite some bug triaging, leading to a much better bugzilla (actually it was running Flyspray at that time) state.

    The constant questions of László Csécsy (who nowadays mostly hacks Drupal core and its modules) generated more and more improving documentation. We first started with a LaTeX documentation, then later turned it to an asciidoc one, in the hope of attracting more non-math contributors. ;-)

    Then we started to have some beautiful artwork, Viktor Gondor contributed some really cool wallpapers.

    Finally, I think the last two Hungarian guys who spent a hell of a time on the project was András Vöröskői (we still link his getting started HOWTO for new contributors) and János Kovács. I remember for a long time I used the home-hosted server of János when we didn’t have a fast dedicated i686 build server and my desktop was still 32bit.

    Going international

    Then things started to happen fast. Developers outside Hungary showed up: Gabriel Craciunescu from Germany and Michel Hermier from France worked a lot on KDE, Gabriel also hacked a lot of the core OS, like kernel, glibc, etc.

    Priyank Gosalia from India started to work on GTK tools like a package manager frontend. David Kimpe contributed a PPC port years before I got my iBook (which I used as my primary box for a few years).

    And needless to say, we still made mistakes. We thought that the development of the original Pacman package manager slowed down, its rewrite (providing a separate library and a console frontend, instead of one monolithic binary) seemed to never complete, and given that we were not Archlinux developers, we never got access to its CVS. We started to use that library API in the installer, testcases, GUI tools, while changing the API for them wasn’t an issue, since officially it was still unreleased. All this frustration (and underestimating the cost of maintaining a fork) resulted in our Pacman-G2 project. First it appeared to be shiny, but once Krisztián Hamar left Frugalware, we constantly lacked manpower and today the consensus is that with git, maintaining a patchset over pacman.git would be far better, just nobody did the work of merging the two projects. Pacman-G2 is in maintenance mode today, and it still well serves its purpose, but it didn’t see any new major feature in years.

    We also saw a rewrite of the installer. The original one was written in bash, then the second was written in C, using libdialog. And currently James works on a third one, hopefully fixing lots of instability, resulting from the design errors of the current one. ;-)

    More and more developers

    And then somehow more and more developers came. Gourdin Gaetan (bouleetbil) and Sébastien Vincent (Baste) worked a lot on the GNOME support, James Buren took over the base system from me. Paolo Cretaro (Melko) does KDE bumps nowadays and Marius Cirsta (mcirsta) did heroic work of including OpenJDK 7 into Frugalware.

    Abdelmoumene Hamza (Slown) is a guy who volunteered to do a log of boring work. When you can install a random Perl on Python module from a package, chances that it was packaged by Slown are high. ;-)

    Regarding infrastructure, Benjamin Nolmans (Xarkam) stepped up, and finally converted the data from our Flyspray instance to the more mainstream and maintained Trac format. Regarding architectures Boris Albar (Elentir) contributed an ARM port. Our press noise was generated by Russell Dickenson (phayz) for a long time: he wrote newsletters regularly, and these days he works as a documentation writer for Red Hat.

    For a long time, no more Hungarians joined the project. Then Bagdán Róbert (kikadf) came, and even if he’s rarely present on IRC, he still contributes from time to time, and these small, but valuable contributions are what keep Frugalware alive. :-)

    Regarding other big upgrades, Daniel Exner (dex) was the guy who stepped up and finally packaged KDE4 for Frugalware. All KDE4 users meeting him should buy a beer for him. ;-) Anthony Jorion (pingax) joined in 2010, and since then he does XFCE updates.

    Finally, I should not forget about Daniel Eledut (Devil505), who probably helped me the most with my "the rest" task: he really helped out in various areas, and he was a major Frugalware contributor for a long time.

    The end of my leadership

    Contributing to other projects is something I did regularly, since every time I got a bugreport for a package, first I created a patch to fix the problem (in case it could not be worked around in the buildscript), and then of course tried to upstream it, knowing the maintenance cost of non-upstreamed patches. Still, these were minor and my main project was Frugalware for years.

    The bigger contributions came when I started to complete Google Summer of Code projects, first SWIG, then git, finally OpenOffice.org — which turned into LibreOffice. The motivation is easy: these were still Free Software hacking, but given that they are not bash hacking (like most packaging work for Frugalware), more interesting problem had to be solved, GSoC also provided nice payments, and needless to say: was useful to many people not using Frugalware.

    The idea of passing over my leadership to someone else first came when I looked back, and I saw that I spend more time on LibreOffice than on Frugalware, even after the paid GSoC ended (later resulting in joining SUSE to work on LibreOffice fulltime). My focus somehow just shifted from hacking a distribution to hacking something that is directly useful for average users as well. It’s not worse or better, it’s just different, and after lots of years, I wanted to do something different.

    To sum up, I did not reach my expectations as a project leader and I waited for a candidate I could teach and finally who could become as good as I was in earlier years, or hopefully even better. And then James came, who first took over the base system, and we agreed that after 1.7, he’ll do releases. So it was not unexpected, but still I was a bit surprised when I saw his role-changing commit. But needless to say, when thinking a bit more, it was completely logical, all I really do these days is fixing things left and right regarding my arm port needs, and maintaining the LibreOffice package.

    Summary

    The idea of writing this post is not mine. However, when I thought about how many individuals helped this project so far, I realized I really need to create a summary, thanking their work. Reading the above, I hope that in the past years my leadership to Frugalware were useful and I realize that these days I have to step back, given a better leader appeared. I honestly wish the project to see at least as many releases as we already have, now that fresh energy leads it. :-)

    I tried to point out all our heroes during the last years, but it’s possible I forgot someone. If you think you’re missing from this post, please leave a comment. Thanks.


  • Sunday, 29 July 2012
    GRUB2 vs. RAID1, cfdisk and serial console (Comments)

    Since the Frugalware 1.7 release is near, and it uses GRUB2 by default, I created a virtual machine, that is similar to the one we use under genesis.frugalware.org, which hosts this blog as well.

    The relevant details:

    • it has two RAID1 arrays with ext3:

    $ mount|grep /md
    /dev/md126 on / type ext3 (rw,relatime,errors=continue,barrier=1,data=ordered)
    /dev/md127 on /home/ftp/pub type ext3 (rw,relatime,errors=continue,barrier=1,data=ordered)
    • the partition tables are created by cfdisk:

       Device Boot      Start         End      Blocks   Id  System
    /dev/sda1              63  1465144064   732572001   fd  Linux raid autodetect
    • the machine has a serial console configured:

    # cat /proc/cmdline
    root=/dev/disk/by-uuid/7e41c95d-cd73-4043-b0ba-4797af6ddeff ro vga=normal nomodeset console=ttyS1,115200

    Now the question is how does this config deal with the GRUB2 upgrade. First, don’t miss the official upgrade howto, it covers most cases. What I want to detail here is how did I avoid starting from scratch and creating a proper partition table using fdisk.

    Here are the steps I needed:

    • Backup. Yes, I did screw up for the first time, so it’s really needed.

    • Resize parts of the / RAID1 (/dev/sda1 and /dev/sdb1) with gparted livecd (resizing ext3 from cmdline parted didn’t work for me).

    • mdadm re-creation as described in the upgrade howto, from fw install cd, so you’ll have the required 1.0 metadata.

    • Reinstall GRUB1 to sda and sdb, since the physical location of GRUB1’s stage* changed.

    • Boot back to 1.6, run pacman-g2 -Syu, and grub-install — again, see the upgrade howto for details.

    • Now given that serial console needs a custom GRUB config and kernel parameters, you need to modify GRUB2’s /etc/default/grub. Here is my diff:

      /etc/default# diff -u grub.orig grub
      --- grub.orig   2012-07-19 01:57:20.000000000 +0200
      +++ grub        2012-07-29 14:45:50.000000000 +0200
      @@ -1,11 +1,12 @@
       GRUB_DEFAULT=0
       GRUB_TIMEOUT=5
       GRUB_DISTRIBUTOR="Frugalware 1.6"
      -GRUB_CMDLINE_LINUX_DEFAULT="quiet splash"
      +GRUB_CMDLINE_LINUX_DEFAULT="vga=normal nomodeset plymouth.enable=0 console=ttyS0,115200"
       GRUB_CMDLINE_LINUX=""
       GRUB_PRELOAD_MODULES=""
      -GRUB_TERMINAL_INPUT=console
      -GRUB_TERMINAL_OUTPUT=gfxterm
      +GRUB_TERMINAL_INPUT=serial
      +GRUB_SERIAL_COMMAND="serial --speed=115200 --port=0x3f8"
      +GRUB_TERMINAL_OUTPUT=serial
       #GRUB_GFXPAYLOAD_LINUX=keep
       GRUB_GFXMODE=auto
       #GRUB_DISABLE_LINUX_UUID=true

      (Refer to this post if you don’t know the needed port number.)

    • Finally run grub-mkconfig to create the real config from the just modified default settings.

    • Reboot, and GRUB2, the boot process and the login prompt should be accessible over the serial console again.


  • Friday, 06 January 2012
    mtd-utils (Comments)

    Quick node about this useful project I packaged two days ago. It has a long FAQ - I was interested in how can one access the builtin nand storage on an arm board using it.

    First, check your dmesg, you should see something like:

    Creating 3 MTD partitions on "orion_nand":
    0x000000000000-0x000000100000 : "u-boot"
    0x000000100000-0x000000500000 : "uImage"
    0x000000500000-0x000020000000 : "root"

    As the names say, the three items here are the bootloader, the kernel and the root filesystem. To access and mount the last one, you need:

    ubiattach /dev/ubi_ctrl -m 2
    mount /dev/ubi0_0 root
    ... hack hack hack ...
    umount root
    ubidetach /dev/ubi_ctrl -m 2


  • Wednesday, 04 January 2012
    Frugalware arm port install HOWTO (Comments)

    I recently got a GuruPlug. It has Debian by default, and it’s apt config is set to stable, while in fact at the moment what’s the factory default is considered as oldstable by upstream. So if you blindly do a few apt-get install foo, soon you’ll have newer userspace than kernel, and your device will no longer boot (based on true story - and yes, this is not Debian’s fault). Moreover, I was interested in how to install Frugalware on this device, so here is a quick howto.

    Install rootfs

    First you need to bootstrap Frugalware from Debian. It’s a good idea to install Frugalware on a USB stick, so you can switch back to Debian in case you messed up something and start from scratch again.

    Partitioning is up to you, you’re recommended to have a small FAT (type: 0x0b) partition (32MB for example) at the beginning, we’ll use that later. The second can be the rest, ext4 or so.

    Format and mount it (your device name may differ!):

    mkfs.ext4 /dev/sda2
    mkdir -p /mnt/sda2
    mount /dev/sda2 /mnt/sda2

    Then install our pacman-g2 binary to the Debian system, so you can bootstrap:

    wget http://ftp.frugalware.org/pub/frugalware/frugalware-stable/frugalware-arm/pacman-g2-3.8.3-2mores2-arm.fpm
    unxz pacman-g2-3.8.3-2mores2-arm.tar.xz
    cd /
    tar xf /path/to/pacman-g2-3.8.3-2mores2-arm.tar
    rm .CHANGELOG .FILELIST .PKGINFO

    Installing the required packages is a single command, as described here:

    pacman-g2.static --noconfirm -Sy core base -r /mnt/sda2/

    Upgrade the bootloader

    Once the rootfs is ready, you need a new bootloader that will be able to boot our vanilla kernel.

    You need a JTAG Board, so you can access the serial console. If you connect the USB cable to you PC, you can use for example

    screen /dev/ttyUSB0 115200

    to access the device.

    Given that we want to boot a vanilla kernel, we need a vanilla bootloader as well. Before you mess with the bootloader, it’s a good idea to make a backup of its config (there is a 3 second timeout during boot - if you press any key there, you get the Marvell prompt). Here is my config:

    Marvell>> printenv
    bootcmd=${x_bootcmd_ethernet}; ${x_bootcmd_usb}; ${x_bootcmd_kernel}; setenv bootargs ${x_bootargs} ${x_bootargs_root}; bootm 0x6400000;
    bootdelay=3
    baudrate=115200
    x_bootcmd_ethernet=ping 192.168.2.1
    x_bootcmd_usb=usb start
    x_bootcmd_kernel=nand read.e 0x6400000 0x100000 0x400000
    x_bootargs=console=ttyS0,115200
    x_bootargs_root=ubi.mtd=2 root=ubi0:rootfs rootfstype=ubifs
    ethact=egiga0
    bootargs=console=ttyS0,115200 ubi.mtd=2 root=ubi0:rootfs rootfstype=ubifs
    ipaddr=10.10.10.10
    serverip=10.10.10.179
    ethaddr=F0:AD:4E:00:CE:C3
    stdin=serial
    stdout=serial
    stderr=serial

    The only semi-unique part is the MAC address of the network interface(s).

    If you want to update the bootloader, a possible way is to put the new binary to a pendrive. Given that the default bootloader does not support ext*, we need a fat filesystem. So format the first small partition we created already (the device name may be different in your case!):

    mkdosfs /dev/sda1

    Till Frugalware 1.6 is released, support for GuruPlug is available in Frugalware -current only, so download the binary package from there, extract the u-boot.kwb file from the guruplug directory, put it to the new partition. (A few other models are explained here).

    Before you reboot, copy also /boot/uImage to the fat partition, you may have problems problems with reading the kernel from the ext4 partition with u-boot.

    Once copying the kernel is done, reboot and in the u-boot shell do:

    usb start
    fatload usb 0:1 0x0800000 u-boot.kwb
    nand erase 0x0 0x60000
    nand write 0x0800000 0x0 0x60000
    reset

    You can verify the updated bootloader with the version command:

    Marvell>> version
    
    U-Boot 2011.12 (Jan 03 2012 - 16:55:38)
    Marvell-GuruPlug
    gcc (Frugalware Linux) 4.6.2
    GNU ld (GNU Binutils) 2.22

    If Frugalware is mentioned, that’s a good sign. :)

    Boot the new rootfs

    Now you can boot your new rootfs:

    usb start
    fatload usb 0:1 0x00800000 /uImage
    setenv bootargs console=ttyS0,115200 root=/dev/sda2 rootdelay=5
    bootm 0x00800000

    If it booted fine, you may want to make this the default:

    setenv bootargs 'console=ttyS0,115200 root=/dev/sda2 rootdelay=5'
    setenv bootcmd_usb 'usb start; fatload usb 0:1 0x00800000 /uImage'
    setenv bootcmd 'run bootcmd_usb; bootm 0x00800000'
    saveenv

    Finalize

    The rest is up to you:

    • setting up a root password

    • setting up network by default using netconfig

    and so on… you know this already, nothing arm-specific.

    For the reference, here is the tested CPU and Frugalware version:

    $ cat /proc/cpuinfo
    Processor       : Feroceon 88FR131 rev 1 (v5l)
    BogoMIPS        : 1191.11
    Features        : swp half thumb fastmult edsp
    CPU implementer : 0x56
    CPU architecture: 5TE
    CPU variant     : 0x2
    CPU part        : 0x131
    CPU revision    : 1
    
    Hardware        : Marvell GuruPlug Reference Board
    Revision        : 0000
    Serial          : 0000000000000000
    $ cat /etc/frugalware-release
    Frugalware 1.5 (Mores)


  • Saturday, 24 September 2011
    Userspace boot speed: less than a second (Comments)

    I did not care about boot speed for a long time, since servers usually restarted only for kernel upgrades, laptops suspend to ram all the time, and I have no desktop machine at home for years now. Though one use case started to motivate me recently: I use virtual machines a lot, and waiting for them to boot up is boring.

    So I looked at the systemd-analyze blame output, and it turned out that for a base-only install the only service that takes a lot of time is netconfig. It’s because it was a oneshot service, so the whole boot process waited for dhcpcd to get an IP. A much better solution is to just start netconfig in the background and move on — and that’s exactly frugalwareutils in git does now.

    Before:

    $ systemd-analyze
    Startup finished in 2030ms (kernel) + 1166ms (initrd) + 6755ms (userspace) = 9953ms

    After:

    $ systemd-analyze
    Startup finished in 2039ms (kernel) + 1146ms (initrd) + 892ms (userspace) = 4079ms
    Note
    Again, this is for a virtual machine (where the host has some IO cache) and for a base-only install.


  • Tuesday, 18 January 2011
    systemd vs. LVM (Comments)

    Yesterday I enabled udev support in our LVM package, and that means LVM volumes under systemd not work as expected. Thanks to Lennart for the quick answer. :)


  • Friday, 14 January 2011
    Nationality of currently active Frugalware developers (Comments)

    /pic/nationality-of-frugalware-developers.t.png

    Just a FYI:

    $ ./chkacc --verbose
    Accepting active developer 'vmiklos' [hu]
    Accepting active developer 'iron' [hu]
    Accepting active developer 'voroskoi' [hu]
    Accepting active developer 'boobaa' [hu]
    Accepting active developer 'crazy' [ro]
    Accepting active developer 'priyank' [in]
    Accepting active developer 'hermier' [fr]
    Accepting active developer 'devil505' [fr]
    Accepting active developer 'bouleetbil' [fr]
    Accepting active developer 'ryuo' [us]
    Accepting active developer 'gholafox' [fr]
    Accepting active developer 'phayz' [au]
    Accepting active developer 'elentir' [fr]
    Accepting active developer 'cedynamix' [fr]
    Accepting active developer 'jercel' [fr]
    Accepting active developer 'exceed' [fr]
    Accepting active developer 'dex' [de]
    Accepting active developer 'kooda' [fr]
    Accepting active developer 'centuri0' [fr]
    Accepting active developer 'kikadf' [hu]
    Accepting active developer 'pingax' [fr]


  • Monday, 27 December 2010
    Systemd: kernel, bluetooth, bitlbee (Comments)

    • Finally I can run systemd on my desktop, and not in a virtual machine only, since there is a patch for the sched/cgroup issue I had. It was not a systemd bug, but systemd triggered a kernel issue, which was hidden so far.

    • I just updated our bluez package, and bus activation works fine. This means bluetoothd is not started till somebody connects to it. Given that I sometimes do not log in to KDE to use kbluetooth, but just want to do something on the console, where having bluetoothd running is totally unnecessary, I consider this really nice.

    • It came up on #bitlbee that there is a Fedora bug where they ask for systemd unit files for BitlBee. So I created a patch, and the nice thing is that the second iteration now seems to be fine for Frugalware/Fedora/Debian, so we managed to build something cross-distro here. (I think I mentioned that’s the #1 feature I like in systemd.)


  • Monday, 20 December 2010
    Systemd Podcast (Comments)

    I just found this podcast, it’s a bit dated since it happened in August and things have been improved a bit since then, but still I was happy to listen it. (The interesting part starts at 1:10:50; 17 minutes.)

    I already had a short post about systemd, and I mentioned the killer feature for me is the compact distro-independent service file (initscript for sysvinit) format it uses, since even if the initscript format we use in Frugalware right now is mostly bullshit-free, it’s still more bloated than the systemd service files.

    Now back to why I began writing this post. :) So there is the post about systemd, but it’s rather long, and it’s easy to miss the point. Since the previous post, I think there are a few more killer features in systemd:

    • Every service is started in a separate control group (cgroup). Do you remember the situation when you wanted to know who the hell started a given particular process? This is typically a problem if the process is running as root. Now this is no longer a problem, systemd-cgls will show you if it was started by a user or a given service, etc. (Yes, I consider this a security feature.)

    • Upstart already had this "restart if it crashed" feature, but systemd does it better: if the daemon supports socket activation, then messages sent to the socket are buffered by the kernel, and no message will be lost during the restart.

    • systemd provides this on-demand feature, which is pretty much like inetd, so hopefully we can get rid of inetd, which was a constant problem previously. (We inherited a fork of the OpenBSD inetd from Slackware, and being a fork it did not improve since years, causing a constant problem.)

    Why writing this post today? Because this morning systemd appeared in -current, so you can even try it out without installing any unofficial package.


  • Wednesday, 15 December 2010
    Splashy vs. systemd (Comments)

    So we have this systemd wip repo and one of the remaining todo items was to fix up splashy to work with systemd.

    First, why doesn’t it out of the box of systemd is a drop-in replacement? Because splashy is not a simple init script, it hooks itself to rc.sysvinit, so it was specific to sysvinit.

    Now given that plymouth was already ported to systemd, it served as a good example. You may also ask: why don’t we just switch to plymouth? Because:

    • we don’t want to introduce multiple changes at the same time

    • it requires kms, while Splashy did not require it - so as long as it’s not true that most video cards support kms (think of via or anything else which supports vesafb, but not kms), such a switch would be a regression

    Okay, enough anti-Plymouth rant. As you can see, right now the task is to fix Splashy.

    The steps I did were:

    • updated the Splashy theme to remove the progressbar: given that now we start services parallel and on-demand, it does not really make too much sense

    • introduced unit files: to start on boot, to stop before getty/xorg, to start before kexec/reboot/halt/poweroff

    • added the usual compatibility symlinks to avoid starting the old rc.splash (which would happen anyway)

    • added a simple wrapper that fires up splashy and once it’s ready it sends the Booting…/Rebooting…/etc text to it

    The last problem is a bit more complex: now that xorg startup wants to stop splashy, a dependency has to be declared. xorg startup is handled by prefdm.service, but in case it depends on splashy-quit.service, what will happen if you don’t have or don’t want a splash?

    The solution for now is to declare an After= relation: that means xorg startup will happen after the quit of splashy, if it happens at all.

    This way:

    • you can disable the splash, it won’t cause any dependency errors

    • if you don’t have prefdm.service enabled, you can even have splashy uninstalled

    • (this one is a bit ugly) you can disable the splash, but if you uninstall it and have prefdm.service enabled, you’ll get a dependency error, as systemd will have no idea what splashy-quit.service is.

    To test the above:

    • you need to enable the systemd wip repo (see the link above)

    • run pacman-g2 -Sy systemd

    • boot with init=/bin/systemd

    Stay tuned, we still have a few blocker issues before merging the (disabled by default, as you need that kernel parameter) systemd support to current.


more »