Rationalisation of init

[jlst]

I'm ok with that

[gyro]

The current "init" tries to be too "friendly" in what it does when a specified partition or file is not found.
If anything is specified and not found, "init" should act as though the file is not present.
It should not fallback to the default file, and use that, as the current "init" tries to do.

Why?
So that you get immediate feedback that the spec is incorrect.
If you're specifying an alternative sfs to test something, and it boots fine. The obvious assumption is that the test sfs is fine.
But if there is a typo in your spec, and "init" used the default sfs instead, then your assumption is incorrect.
You're more likely to notice if the sfs is missing from the boot.

(Of course it also makes the code simpler.)

gyro

[Karl Godt]

unfortunately bloat is necessary .
there are too many possibilities of HW configurations available,
so testing needs done.

I have my own init code to speed up the search.
My init code accepts parameters like sfsload=oldsyle|new that is to load the sfs
and lastly the puppy.sfs and zdrv.sfs ;
and pupmode=X to force a pupmode and if X is something unknown falls back to 5 something like that ; instead editing the init to change 12 to 13 and such .

But I rub fukk installs on pupmode 2 mostly.
I wish you good luck

[greengeek]

If anything is specified and not found, "init" should act as though the file is not present. It should not fallback to the default file, and use that, as the current "init" tries to do.
Why?
So that you get immediate feedback that the spec is incorrect.

Is it possible to include user interaction at this point so that the following occurs?:

- file not found
- tell user file not found
- ask user if they want further searches done or not (in case new user unsure how to proceed and wants to see if anything else CAN automatically be found - so they may end up with a clue where the file actually was)
- ie: no automatic searches performed without at least asking the user.

[gyro]

Is it possible to include user interaction at this point so that the following occurs?

No.
1. With "huge-kernels" there are no kernel modules ordinarily available to "init", so any device requiring kernel modules to work, will not work while "init" is running, this includes some keyboards. (I have one.)
Currently only the savefile/savefolder code in "init" requires keyboard input.
One of the aims of the rewrite is to get the sfs files loaded so that the "initmodules" facility can "insmod" any keyboard required modules before the keyboard is needed.
Adding more keyboard interaction during "init" only makes an already complicated situation more complicated.

2. I don't consider folk who are specifying alternate sfs files via boot parameters to be complete noobies.

gyro

[technosaurus]

One of the aims of the rewrite is to get the sfs files loaded so that the "initmodules" facility can "insmod" any keyboard required modules before the keyboard is needed.

If you need to load kernel modules during init, the kernel is configured wrong. Just change tne CONFIG_XXXX=M to =Y for everything referenced in init - there should be no modules in the initramfs/initrd.

[jlst]

That leads us to kernel-kit and the huge kernels already built. When compiling a kernel, one needs to be sure that all (or almost all) modules that have to do with keyboard input are builtin.. if a huge kernel does not comply with this, then it should be deleted immediately...

[jlst]

gyro, watching the logs I see vercmp complains about a wrong syntax... i made sure the busybox vercmp works the same as the standalone app.

This happens when one of the operands is empty, so I wonder if there is something to delete there...

[jlst]

i think we should use the full blkid in the initrd, it's just more robust at everything, but a bit slower too

[Sailor Enceladus]

That leads us to kernel-kit and the huge kernels already built. When compiling a kernel, one needs to be sure that all (or almost all) modules that have to do with keyboard input are builtin.. if a huge kernel does not comply with this, then it should be deleted immediately...

Do the templates in kernel-kit/configs_extra comply with this? I think there's 12 of them in there.

edit: Nevermind, I think gyro answered this in the next post.

[gyro]

One of the aims of the rewrite is to get the sfs files loaded so that the "initmodules" facility can "insmod" any keyboard required modules before the keyboard is needed.

If you need to load kernel modules during init, the kernel is configured wrong. Just change tne CONFIG_XXXX=M to =Y for everything referenced in init - there should be no modules in the initramfs/initrd.

Correct, except for, where is this mythical list of all the modules that are needed?
My keyboard works with 666philb's kernels, but not with 01micko's. (not a big deal for me, since I then use "initmodules", with a 01micko kernel.)
666philb used to have a problem with a particular USB port, that required a module, (fixed in his kernels now.)
The end result is a kernel that conatins everything that looks slightly like it might be needed.
But stiil, at any time someone can start using new hardware that is only supported with a new kernel module, thus making all existing kernels obselete.
And until the new kernel's are available, the user has the "no keyboard during init" problem.

Edit:
Making the "huge-kernel" modules available during "init", we can do.
The remaining problem is; how do we sort out what is the minimal set of modules that need to be loaded to enable the keyboard?
The only current solution is "intmodules".

gyro

[gyro]

gyro, watching the logs I see vercmp complains about a wrong syntax... i made sure the busybox vercmp works the same as the standalone app.

This happens when one of the operands is empty, so I wonder if there is something to delete there...

I haven't got to the upgrading code yet.

I'm currently working with my new minimilist init script to sortout all the new stuff.
Once that is working I will start porting this code into the current init. (First action, delete obsolete stuff.)
And/Or port the current stuff that is missing in my new init, from the current init to my new init.

gyro

[gyro]

That leads us to kernel-kit and the huge kernels already built. When compiling a kernel, one needs to be sure that all (or almost all) modules that have to do with keyboard input are builtin.. if a huge kernel does not comply with this, then it should be deleted immediately...

Sounds nice, but at any time, new hardware could invalidate all previous kernels.
My keyboard worked in "init" with old kernels, because there was no module peculiar to it. But then latter kernels introduced a new module to provide full support for my keyboard. It then stopped working during "init". (Fixed in 666philb's kernels.)

gyro

[gyro]

I have found a difficulty with searching for pup...sfs instead of vmlinuz, PDEV1 won't be set if pup...sfs is included in the initrd.gz.
A "humongous initrd" puppy will probably still boot, but PUPSTATE will have a couple of empty variables, unless the user adds a "pupsfs=xxx" boot parameter.

gyro

[gyro]

i think we should use the full blkid in the initrd, it's just more robust at everything, but a bit slower too

My testing has found no problem with the blkid in the tahrpup 6.0.5 initrd.gz.
My test init script relies on blkid to decode Label's into partition names, and so far this has been reliable in both ata and usb boots.
I have not discovered any need to replace it.

gyro

[technosaurus]

My test init script relies on blkid to decode Label's into partition names, and so far this has been reliable in both ata and usb boots.

Out of curiousity, have you tested it with toybox blkid? I wrote and optimized that one a while back (with help from Rob) to be able to detect all filesystems UUIDs and lables in a single pass, so it should be faster/lighter than the busybox and GNU versions. I'm interested to see some feedback on it... BTW, it also has the guess-fstype functionality builtin (used in many puppy inits) I think Rob omitted my hacky NTFS label code because ... well NTFS is really a PITA that requires a lot of workarounds for "correct" code (UTF-16 labels, large spans between like data, weird "UUID"s)

Now that I think about it, with my microsaurus init.c (absolute minimal combo of init and xinit) and my toybox blkid.ccode, I could probably do the whole thing in C

The microsaurus code was something along the lines of this code:

Code: Select all

/*  Copyright 2011 -2016 Brad Conroy.
	Released to the Public Domain.
	Redistributable under any OSI approved license.
	Redistributable under any license meeting the OSI definition of open source.
*/

/* config.h begin */
#define CFG_SH ENABLED // {EN/DIS}ABLE using shell as init fallback
#define SHELL "/bin/sh"

#define ENV "PATH=/bin", "HOME=/", "PS1=# ", "SHELL=" SHELL

#define CFG_X  ENABLED // {EN/DIS}ABLE X server
#define X_CMD "Xvesa","-screen","1280x800x24","-nolisten","tcp","-tst"
#define X_ENV "TERM=rxvt", "DISPLAY=:0"

#define CFG_WM ENABLED // {EN/DIS}ABLE window manager
#define WM_CMD "jwm" , "-display", ":0"

#define CFG_DE DISABLED // {EN/DIS}ABLE desktop environment (icons)
#define DE_CMD "rox","-p","default"
#define DE_ENV "APP_DIR=/usr/share/ROX-Filer", "CHOICESPATH=/usr/share"
/* config.h end */

#define ENABLED(...) __VA_ARGS__
#define DISABLED(...)

#include <unistd.h> /* for exec */
#include <sys/mount.h> /* for mount */
#include <stdlib.h> /* for putenv */

static void forkit(char * const *argv,char * const * envp){
  long x = fork();
  //if (x<0) die(1); //TODO but this really shouldn't happen
  if (x==0) execve(argv[0], argv, envp);
  //TODO error if exec fails - unlikely unless command is bad
}


CFG_WM(static char * const wm[]={ WM_CMD,NULL}; )
CFG_X( static char * const  X[] ={ X_CMD, NULL}; )
CFG_DE(static char * const de[]={ DE_CMD,NULL}; )
CFG_SH(static char * const sh[]={ SHELL, NULL}; )
static char * const env[] = { ENV, CFG_X(X_ENV,) CFG_DE(DE_ENV,) NULL };

int main(int argc, char** argv, char **envp) {
	CFG_X(forkit(X,env);) //should this optionally be /bin/sh ??
	//Xvesa doesn't need anything mounted, so mount while X starts up
	mount("/proc", "/proc", "proc", MS_MGC_VAL, NULL);
	mount("/sys", "/sys", "sys", MS_MGC_VAL, NULL);
	mount("/dev", "/dev", "devtmpfs", MS_MGC_VAL, NULL);
	mount( "/dev/pts", "/dev/pts", "devpts", MS_MGC_VAL, NULL );
	//TODO wait for /tmp/.X11-unix/X0 socket
	CFG_DE(forkit(de,env);)
	CFG_WM(forkit(wm,env);) //should wm be optional too?
/*  The rest of "init", can be setup using jwm's <StartupCommand> etc...*/
	CFG_SH(execve(sh[0],sh,env);)
	/* TODO reap zombies, handle signals and clean shutdown method */
	while (1) sleep(1);
}

Note: Microsaurus had multicall binary with Xvesa, jwm, rxvt and dash included in the kernel builtin initramfs and the kernel size was under 1MB (2MB with gtk1 apps like rox, dillo, aumix, an editor, gtkdialog1 and a few other tools). The additional required kernel options for general purpos use would add an additional 1-3MB (wireless modules can be rather large). Using shared libraries like X11, c, gtk2 and its dependencies (for current ROX-Filer) would add a couple MB also.

[gyro]

Out of curiousity, have you tested it with toybox blkid?

No.
Since I'm having no problem with the blkid I have, I haven't bothered trying others.

Perhaps this is a "busybox" v "toybox" issue, which is outside the scope of my current project.

gyro

[gyro]

I have uploaded a new minimalist test "init" script to the github.com/puppylinux-woof-CE/initrd_progs repository.
It can be downloaded using this command:

Code: Select all

svn export https://github.com/puppylinux-woof-CE/initrd_progs/trunk/0initrd/init

This is the first version of this "init", it can actually boot to a running system.

But it's a work in progess, there are still many missing facilities:
Support for pupmodes other than 5 and 12.
Support for "initmodules".
Support for multisession cd's.
Support for encrypted savefiles.
Support for savefile resizing.
Support for updrading savefile/savefolder.
Support for keymaps.
Support for non-english languages.
May need more savefile/savefolder searching for cd boots.

It does not support full installs, and since it resides in "initrd.gz" it may never do so.
(Unless someonelse changes this.)

Some issues with what it does do:
In pupmode=12, /tmp is in the savefile/savefolder, not tmpfs.
The pupmode management is not good.
Humongus initrd folk, may want to specify a "pdev1=" boot parameter to get something more meaningful into PUPSTATE->PDEV1.

It does implement the new algorithm for puppy file searching; much simpler that the old one.
It builds the aufs stack as it goes, first a pupmode=5 stack, it changes this to a pupmode=12 stack if a savefile/savefolder is found.
It only waits for usb storage if it is needed.
It fully supports the recent changes to specifying puppy files using Label or UUID in boot parameters.

My immediate plans:
1. fix the /tmp issue in pupmode=12
2. Devise a pupmode management scheme.
3. Implement "initmodules" support.
4. Sortout savefile/savefolder searching for cd boots.
5. Sortout what's really necessary for upgrading, because it's messy.

I will be pushing changes directly to initrd_progs on git-hub as soon as I have any patch that's worth backing-up.
Of course, being on git-hub, others are welcome to make changes also.

gyro

[learnhow2code]

hi gyro, sorry to ask something that you might have answered several times:

whats this init for? i know what an init is, what i mean is which distros do you currently work on/with? tahr was mentioned, did you have a hand in maintaining/editing its init? what about slacko? any others? kind of a trivia question but im actually interested, so its a sincere trivia question. good luck with your init, it sounds like an interesting project.

[gyro]

hi learnhow2code,

This project is about updating the "init" script in woof-ce. So improvements produced by this project will be included in all puppies generated using woof-ce, in the future.

The "init" script is inside the "initrd.gz" and is the code that runs at the beginning of the boot in frugal installs. It's major purpose is to setup the aufs stack that is at the heart of a puppy frugal install.

gyro