[einit] v0.10 is out, any beta testers about?

[der bastler]

As I am constructing a minimalistic Gentoo system I can act as some sort of beta tester.

Hardware: VIA EPIA 5000 (x86 arch: C3), 256 MiB RAM, Compact Flash as boot medium.
Intended purpose: Musicbox

• Plays: CD, DVD, USB storage, NFS and Samba network resources, internet radio stations.
  
• Supports as many formats as possible (perhaps even inside archives?).
  
• Provides a sweet little 8" touchscreen GUI (Without Xorg: Qtopia?).
  

system-1: kernel 2.6.17-gentoo-r8, uclibc, baselayout-lite, busybox (approx. 3 MiB on CF)
TinyGentoo. Worked, but not suitable for further development.

system-2 (current): kernel 2.6.17-gentoo-r8, uclibc, baselayout, shadow, util-linux, grep, sed, gawk, module-init-tools (approx. 7 MiB on CF)
After some trial and error and hacking (disable udevsend, mkdir /var/lib/init.d...) I could boot a stripped down standard Gentoo system, but I had to pull in more and more extra tools. Busybox offers these tools, but its init is not compatible with standard baselayout's inittab. As I need the standard baselayout for some future programs I have to go a different way... keep busybox, but use another init. Sysvinit could do, but why not try something new?

system-3 (intended): kernel 2.6.17-gentoo-r8, uclibc, baselayout, busybox compiled without init, einit, udev, coldplug.
Well, let's see how far I can get...

---

I could also offer support as dev. Started programming with Turbo Pascal 7 and its inline assembler (some $a000 magic) in the early 90s, moved on to C and C++ respectively. Started to read Charles Petzold's brick of book. During my studies I got to know PEARL90, Java and MATLAB and switched to Linux. Picked up basic Bash for everyday scripting and finally tried Python this year. And recently I rewrote our institute's website in XML+XSLT=XHTML.
[/ad]

Only catch: free time is precious.
_________________
Tempus fugit.
@frank@troet.cafe

[mdeininger]

@filesystems: seems good, probably something doesn't get killed... i might want to be a bit more agressive about killing processes

i implemented the wait-retry idea from earlier (just commited it, it's in 0.11-cvs-34), haven't tested it in an actual environment (bleedin UML), but it worked in the sandbox...

kktm: that idea is rather good you know! that could work very well... although i wonder how come the remounting r/o doesn't work properly

bastler: i'd love to see some extensive beta-testing. as you probably picked up, einit's design goals are speed, modularity and small size. one note of warning however: last time i tried, uclibc had broken pthreads support (it didn't allow one thread to wait() on a different thread's children, which is vital), so the regular ebuilds won't work on your uclibc box. however, there's a configure flag that should get it to work -- i tested it once or twice on my ipaq (linux 2.6, uclibc) and it worked fine there. of course we could be lucky and someone fixed that bug with uclibc already .

can't hack much myself right now, only feature enhancements and the occasional hack so i know about the time-issue :/ however, there's some feature requests on the wiki that should get you started if you wanted to contribute and get into the code -- just hack along and submit patches if you like but do tell me when you're about to start something bigger.

einit is plain C, no C++, no C# no nothing, so it shouldn't be that hard, but i was told the code isn't the easiest to read (partially that's because my comments suck, but one definite problem with the code is that it isn't running linearly).

btw, if there's any specific questions about the code, just PM me, or IM me (jabber preferred, but i'm on just about any protocol). + i'm sometimes on DALnet and freenode as jyu.
_________________
"Confident, lazy, cocky, dead." -- Felix Jongleur, Otherland

( Twitter | Blog | GitHub )

[KKtm]

new unmount seems to work.
at least /sys and / could be seen attempting to unmount and I didn't remark any fs corruptions last boot. good job!

My little wish list:
* Delete network from *dm requirements. Or treat network as started, as soon as net.lo (or whatever the first network device is) is up.
Because last update when replacing einit.xml I had to wait for ~3 minutes before gdm was started, just because eth0 was still waiting for a non-existent dhcp-server.

* Add my eth1-module to einit.xml
Or should I rather include a modules.xml in local.xml and add this module only at my place?

* A nice framebuffer-bootscreen
Maybe I'll find the time to get started on this myself one day.

[brot]

I am sorry but for me the unmounting doesnt work

I have XFS, and everytime i reboot i have to check my filesystems again. also, my jfs partition doesnt seem to be checked.
And as far as i understood the needed alsa modules should be loaded by default, but it doesnt seem to work here, i have to create a new module group (i have a soundblaster audigy 2, so nothing too uncommon here).

But after some hours of playing with einit i have to say that einit could really be the init of the future. what works does really work great, and if you keep in mind that its 0.11 then you really created something wonderful

(maybe someone can help me with the unmount/alsa/fsck problems... thanks in advance)
brot

[mdeininger]

kktm: glad it seems to work for you! so we definitely made *some* progress here

i'll do as you suggested with the configuration file with the next commit. sorry about those modules you suggested, i seem to have lost track of them ^^8

i had already started on a framebuffer-bootscreen using evas/edje, if you wanted, you could extend that code. if you wanted to extend it, i'll get it up to date so people can hack on it (working with evas/edje to design things really is great, trust me, which is why i chose those libraries )

brot: first of all, thanks for testing, really! (that's to everyone who ever tried it, honestly!)

not sure about those filesystems, einit doesn't have metadata-code for them. do you mean to say that you have to check them manually because they need to be checked but aren't, or that they're always checked while you expect them not to?

the sound-module loader code uses the same way to find modules that gentoo's /etc/init.d/alsa uses, it's kinda weird that they're not loaded... could you post the output of "modprobe -c" somewhere? ( wiki.jyujin.de maybe )

thanks for your enthusiasm towards einit. i'm trying to get stuff done properly, so i never try to over-exagerate with version numbers and releases
_________________
"Confident, lazy, cocky, dead." -- Felix Jongleur, Otherland

( Twitter | Blog | GitHub )

[brot]

Unmounting works with the new version of einit - i reemerged it and now it works, so there definitely is a progress

To my sound problem: i will look at it tomorrow, maybe i am doing something terribly wrong. Its 2:50 am here, so time to go to bed

[rmh3093]

[Lloeki]

KKtm,

[Lloeki]

well, I don't know what's wrong, but even after several reboots (running einit), even emerging udev locked up my computer w/ 100% reproduceability. in the end I booted sysvinit, and the emerge went fine... then I booted back into einit, and the computer rebooted right before I could login. another boot and I could login again. absolutely weird.

these random lockups and reboots seem to only occur since I use einit, and don't seem to happen when I boot sysvinit. I can't even begin to track down what's the cause of it...
_________________
Moved to using Arch Linux
Life is meant to be lived, not given up...
HOLY COW I'M TOTALLY GOING SO FAST OH F***

[brot]

maybe something like a errorlog should be implemented so that we can see what has gone wrong ? This would be really helpfull, especially when trying to track down small problems

[KKtm]

Ok, there still seems to be that problem with unmounting. It actually tries several times to unmount, but failes sometimes (I saw an error-message on / and /home had to be checked when I was booting this time)

My "network config" looks like this:

[mdeininger]

those lockups ARE very weird, I haven't had that happen to me before when running einit I haven't changed anything that could cause such lockups, either...

that is, if the regular einit thread was to receive a SIGINT for whatever reason, it'd commence with the reboot procedure -- but what would send a SIGINT to einit's secondary process?

as to the suggestions:
rmh: I can't promise much eye-candy. potential eye-candy definitely (e17 is so nicely modular), but i'm not raster when it comes to designing things (which should be apparent if you have a look at einit's homepage )

lloeki:
1./2. this is the weirdest series of errors i've ever seen, specially since i didn't change any code anywhere near the modules used by this!

3: einit= works with --check-configuration if you set it as an environment variable, like this:

[KKtm]

btw: I think the "timeout before turning power off, when there's a problem"-option would be very nice to get closer to the roots of all that problems

[Lloeki]

well, 1 and 2 are probably something with my system. such hard lockups are 99% kernel things. anyway it may have something to do with einit, maybe the way or the timing with which things get loaded.

[smlgbl]

Hi,

besides that I can confirm that there are still unmounting issues, meaning /proc and /sys aso. aren't unmounted, so unmounting / fails, I have to admit, I am kind of shocked. As I hadn't played around with einit while everything is running, but ivman didn't come up again, I guess, because sometimes it has to wait a second or so longer for hald, I wanted to start the service again, without rebooting. So without knowing what I do, I type "erc enable ivman" and I get switched back to tty1 and einit starts all over. Except, that everything is running already, so most things fail. This basically isn't funny, even though I could have checked the manpages and stuff on how to do it properly.

Otherwise. to the mounting I don't see any other service fail that would prevent the unmounting. My half a cent.
_________________
samuel.
'Do not let one girding on boast about himself like one unfastening'

[Lloeki]

what's weird is that even in /proc and /sys failed to unmount, we should be able to remount them ro, and then remount / ro... btw Magnus, could there be a last resort remount ro with -l (lazy, see man page) parameter?
_________________
Moved to using Arch Linux
Life is meant to be lived, not given up...
HOLY COW I'M TOTALLY GOING SO FAST OH F***

[mdeininger]

LLoeki, i sometimes noticed sporadic lockups on my system when trying to rmmod some modules...

the two-config-files thing is to make certain things easier. right now i need that 'till there's conditionals in the config file. it's basically so that the core knows a configuration file, and the default profile to load after that... sort of anyway.

the agettys should actually be killed as soon as nothing depends on them... which is rather early actually? weirdness... but i noticed the same. actually that helped me a lot when having to kill portmap (for whatever reason?)

haader, from the sound of it, you actually started "einit" and not "erc" or "einit-control rc"? einit will usually not switch ttys when merely told to enable something... it's actually:

[der bastler]

Downloaded the code of einit-0.10. Will try to improve project management by adding comments.
_________________
Tempus fugit.
@frank@troet.cafe

[mdeininger]

[Maxwell]

Hi!

I'm searching for an init system for my old pentium 1, but i need two things: ADSL support (pppoe) and the firmware for my usb modem being sent correctly.
Gentoo's baselayout (seome versions) works ok. Can eInit accomplish this also? Anyone used einit with a pentium1?

thanks
_________________
Freedom works. Use it!
Linux, by Gentoo

[Lloeki]

things you ask are not dependend of the init system used, so they will work. not OOTB though, since einit is fairly new, but it just requires the proper config entries.
_________________
Moved to using Arch Linux
Life is meant to be lived, not given up...
HOLY COW I'M TOTALLY GOING SO FAST OH F***

[mdeininger]

the only requirement einit has right now is proper pthreads support (or at least "proper enough"), so should work on a pentium1, provided you use a recent kernel (2.6, best with nptl). i recall some 2.4 kernels used to have bugs that prevented glibc from implementing pthreads in a way that allowed all threads of a process to wait on another thread's children to die (which einit relies on, since it's multithreaded and needs to monitor processes).

if you run 2.6 (best with glibc+nptl), it should work. if you know how to load the firmware into your modem using the command-line, you can simply write a small module to do so (you can write regular shell-commands).

if you use an old 2.4 with an old glibc, or you're using uclibc, then you'll have to manually configure and compile einit (there's a command-line switch that enables a work-around, but it's flakey).
_________________
"Confident, lazy, cocky, dead." -- Felix Jongleur, Otherland

( Twitter | Blog | GitHub )

[mdeininger]

a lil' update: (0.11-cvs-40): I added the "multiline-output" feature (sort of, works okay most of the time), the "broadcast message on shutdown/reboot" (only tested with ptys, but the ttys should get it too), a configurable delay on shutdown if something went wrong (default is 10/15 seconds) and i updated the evas-module (it kinda locks all consoles when used from the console, so... better not use it yet).

i'm also trying to get some design ideas for a bootsplash straight on the wiki ( http://wiki.jyujin.de ). don't look like much, but maybe you'd check 'em out and comment on them?

cheers,
Magnus
_________________
"Confident, lazy, cocky, dead." -- Felix Jongleur, Otherland

( Twitter | Blog | GitHub )

[smlgbl]

The definitions for the modules, daemons aso. that are found in einit.xml are not overwritten by what I do in local.xml, is it? I don't like that. I changed some of the definitions for me, but when I update einit, my einit.xml is overwritten. That's why we had the local.xml in the first place, isn't it? But I have to manually delete the service definitions from einit.xml after every update, otherwise my next boot is screwed.
Besides, where can I specify the time to wait on a shutdown failure? Or is it possible to say, like "There where errors in your shutdown sequence. When you're done fixing it, press Ctrl-Alt-Del" or seomething like that instead of a timeout? Then everyone could still decide, if he doesn't care about the unmounting or whatever causes the problem, or whether he wants to investigate the problem. I would actually prefer that very much, because within those 15 seconds, there is not too much one can do. After all you still have to find out what the problem is.
Also, I am still investigating the umount problems I am still experiencing.
_________________
samuel.
'Do not let one girding on boast about himself like one unfastening'

[brot]

somehow einit stopped working - i know that is has to be a configuration issue, but this shouldnt stop einit from shutting down.

When i select to shut down my pc, einit just hangs on the console 1 and does nothing... Maybe it should try to shut down even if some things fail.