Avoid "Searching for Puppy files" during bootup - revisited

[mikeb]

Well i never found the mount unmount routine to be significant in time usage.... the use of find and multilevel searches followed by rationalizing the resultant pile of info on the other hand was a big hit ..way slower than the original use of ls <pattern match> .

After all the init goes through a myriad of hoops just to generate the pupsave variable...so why not supply it and save the trip.

mike

ps...I did wonder how grub searches and find files....but probably not something we could use for this ... find... set kernel parameter idea....

[mavrothal]

Another little issue is that $pupsave works fine when savefile/folder is in another disk. In this case /mnt/home is a broken link. Adding psavemark is useless in this case.
BTW savefile/folder is in another partition of the boot disk works fine for /mnt/home, with of without psavemark.
Some code should be added so either pupsave should not be allowed in another disk or /mnt/home also consider safefile/fodler drive in addition to partition.
I think the latter is better.

[mikeb]

It makes coding way way easier if one works on the basis that an install is located in one place rather than scattered over several devices...that one grew out of the live CD usage and a lack of comfortable booting options and has stuck even though methods have moved on/improved.

I don't have floppies and only have an external DVD writer now for starters...and my gear is old. A superfloppy zip drive... never even seen one.

Mike

[mavrothal]

It makes coding way way easier if one works on the basis that an install is located in one place rather than scattered over several devices...that one grew out of the live CD usage and a lack of comfortable booting options and has stuck even though methods have moved on/improved.

The basic assumption is that the user all (s)he wants is to use their machine with no hassle what so ever and no idea of its inner workings and requirements are.
There are mini-distros that require you to define a number of cryptic, complex and precise options to boot, and this is perfectly fine. Puppy, to keep true to its goals, should rather give the ability of "cryptic, complex and precise boot options" if you want to use them, but still be forgiving and relaxed about them.

BTW the /mnt/home issue is not as bad as originally reported. Is only the first time that has an issue. Then it behaves better. Still needs to be looked at though.

[mikeb]

The basic assumption is that the user all (s)he wants is to use their machine with no hassle what so ever and no idea of its inner workings and requirements are.

hmm i found the puppy 2 approach fulfilled that requirement quite nicely and 4+ meant an increasing number of confusing boot parameters and a complex init script with quirky hbehaviour.. As a noob I found puppy 2 straightforward. Hence the interest in gyros re-evaluation of the boot methods in use....bit of a pet subject ..no pun intended.
Anything to streamline and keep user friendly is welcome.

mike

[mavrothal]

Hence the interest in gyros re-evaluation of the boot methods in use....bit of a pet subject ..no pun intended.
Anything to streamline and keep user friendly is welcome.

Gyro's additions are of course welcome, but they do not change anything in the boot process. They do add another option for a kernel command line argument which is very welcome. We just want to make it as much as possible user friendly and false-tolerant.

[gyro]

Updated patch to "init.nosearch-3.diff" see first post for download.

This is a significant rewrite to avoid extra mount and umount.

The code at the beginning of the "Searching" section does not set the "PUPSAVE" variable, rather it just checks that the partition exists and stores data for later.

The decision is now made in the "search_func" after the first partition is mounted.
The variable "LESSPARTS" is modified so that the specified partition is first in the list.
If the specified savefile/savefolder is not found, then normal searching is performed.

@mavrothal,
"psavemark" is irrelevent since that should have been sorted out before "nosearch" is activated, so it is now recorded in "PUPSTATE" and hence gets respected. The idea behind "nosearch" is that you get it all sorted out running normally, until "PUPSTATE" contains all the appropriate information. Then "nosearch" is activated by specifying a "pupsave" parameter.

I don't quite understand the problem with "/mnt/home".
The "pupsave" parameter could be used to choose between multiple savefiles/savefolders but I don't see any use for specifying the savefile/savefolder of another puppy.

Note: I will probably announce "initrd.gz" files for Tahrpup 6.0.2 in the near future. (My site is up again.)
Edit: now done.

gyro

[gyro]

Just a thought:
One of the difficulties in implementing "nosearch" by reading "PUPSTATE" from the savefile/savefolder is that the issues of it being a savefile or savefolder, and is the savefile encrypted, have to be sorted before "PUPSTATE" can be read.
It would probably be easier to implement "nosearch" if it were triggered by "pdev1" boot parameter being supplied, and the presence of "${PDEV1}${PSUBDIR}/PUPSTATE".

gyro

[gyro]

@mavrothal,
Here is a patch to check the existence of an "adrv" boot parameter file:

Code: Select all

--- init.orig	2015-06-24 09:52:21.966056508 +1000
+++ init	2015-06-28 19:47:14.014732397 +1000
@@ -473,6 +473,13 @@
     done
    fi
    
+   if [ "$PADRVDEV" = "$ONEDEV" ];then
+    if [ -s "/mnt/data${PADRVSPEC}" ];then
+     ADRVSFS="`basename $PADRVSPEC`"
+     ADRV="${ONEDEV},${ONEFS},${PADRVSPEC}"
+     echo -n " adrive" > /dev/console
+    fi
+   fi
    if [ "$ADRV" = "" ];then
     FND_FILES="`find /mnt/data${PSUBDIR} -maxdepth ${SEARCHDEPTH} -xdev -type f -iname ${ADRVSFS} | grep -v ' ' | sed -e 's%^/mnt/data%%' | tr '\n' ' '`"
     for ONEPUPFILE in $FND_FILES
@@ -799,11 +806,13 @@
 fi
 
 if [ "$ADRV" ];then
- DEV="`echo "$ADRV" | cut -f 1 -d ':'`"
- FS="`echo "$PCPARTS0" | grep "${DEV}|" | cut -f 2 -d '|'`"
- SPEC="`echo -n "$ADRV" | cut -f 2 -d ':'`"
- ADRVSFS="`basename $SPEC`"
- ADRV="${DEV},${FS},${SPEC}"
+ PADRVDEV="`echo "$ADRV" | cut -f 1 -d ':'`"
+ if [ "`echo "$PCPARTS0" | grep "${PADRVDEV}|"`" != "" ];then
+  PADRVSPEC="`echo -n "$ADRV" | cut -f 2 -d ':'`"
+ else
+  PADRVDEV=""
+ fi
+ ADRV=""
 fi
 
 if [ "$YDRV" ];then

The idea is to delay checking the file until the "search_func", when a partition is already mounted. A bit like my latest "nosearch" patch.
Obviously the same approach could be applied to the other sfs parameters.

gyro

[mavrothal]

@mavrothal,
Here is a patch to check the existence of an "adrv" boot parameter file:

Code: Select all

-[/quote]
Looks good.
You may want to send a pull request with the changes.
If you get ti it pls try to include verifications for $zdrv and $pupsfs as well as.
Thx

[gyro]

You may want to send a pull request with the changes.
If you get ti it pls try to include verifications for $zdrv and $pupsfs as well as.

My plan is to implement a full patch that includes all the sfs parameters, and then announce it in a separate topic to give others a better chance of commenting. After all this topic is about "nosearch". Pull request later.

I've simplified the code a bit so it is in only 2 places, where the boot parameters are first processed, and in search_func.

In testing, I liked being able to specify an adrv with the wrong name, and in a different directory.

gyro

[mavrothal]

I liked being able to specify an adrv with the wrong name, and in a different directory.

Why adrv only?
Theoretically SFSs can be anywhere as long as they are loaded in RAM. You do not care after that.
You can actually have a small 10MB partition for vmlinuz/initrd/grub(2) and then point to anywhere you want.

One thing that should be checked with the "anywhere" approach is how the system will behave if SFSs are not loaded in RAM either because there is not enough or because they are called with pfix=nocopy and are in different drives.

[gyro]

Why adrv only?

Because that is all I had implemented at the time, as a proof of concept.

Theoretically SFSs can be anywhere as long as they are loaded in RAM.

The first limitation is which partitions get mounted during the "Loading puppy files" section of "init". See http://www.murga-linux.com/puppy/viewtopic.php?t=100011. Copying to RAM overcomes only the second challenge; which partitions remain mounted.

Please continue any discussion of sfs boot parameters in the other topic.
This topic is about "nosearch".

gyro

[gyro]

I have uploaded "initrd.gz" files for Tahrpup 6.0.2 for version 3 of the patch.
See the first post for download details.
gyro

[Marv]

Tested with and without pupsave kernel parameter in X-tahr-1b2 non-PAE on a core 2 duo laptop and on one of my Bay Trail desktops with the very slow USB flag returns. In these two tests the modified init does just what it is supposed to. Speedup on both systems is the same as my ignore=usb modification to init. If this were to be incorporated in the 'stock' inits, it would save me folding in my patch when using a new initrd.

Thanks,

Edit: Init script also used in X-slacko-3.0n. Booting correctly with and without pupsave parameter there also. All of these tests are on Grub4Dos frugal installs with standard unencrypted savefiles.

[technosaurus]

WOW, I can't believe how overly complex this has become ... when it could just be a single parameter SFS= with a ',' separated list of files to mount and use full paths vs. search if an item in the list starts with '/'

Here is some blah template code example ... mount_sfs omitted for brevity ... needs to be updated for overlayfs anyhow

Code: Select all

file_missing(){
	echo "$1 not found, add the appropriate mount point to MNT="
}

autoload_sfs(){
	for MTPT in /mnt/*; do
		[ -f "$MTPT/$1" ] && mount_sfs "$MTPT/$1" && break
	done
	file_missing "$1"
}


load_sfs(){
	[ -f "$1" ] && mount_sfs "$1" || file_missing "$1"
}

mount_sfs(){
	echo "mounting $1" #etc... and so forth
}

handle_sfs_list(){
	while ([ "$1" ]) do
		case "$1" in
		/*)load_sfs "$1";;
		*)[ "$NOSEARCH" ] || autoload_sfs "$1";;
		esac
		shift
	done
}

SFS=a.sfs,/mnt/sda1/b.sfs,...
IFS=","
handle_sfs_list $SFS

[gyro]

WOW, I can't believe how overly complex this has become

Some of the complexity comes from trying to efficiently integrate it into the 'init' script.

when it could just be a single parameter SFS= with a ',' separated list of files to mount and use full paths vs. search if an item in the list starts with '/'

Are you suggesting that this code in the 'init' script

Code: Select all

#100915 requested by technosaurus (formats get changed further down)...
[ $pupsfs ] && PUPSFS=$pupsfs #format partition:<path><filename> ex: sda2:/wary071/wary_071.sfs
[ $zdrv ] && ZDRV=$zdrv #ex: sda2:/wary071/zdrv_071.sfs
[ $adrv ] && ADRV=$adrv
[ $ydrv ] && YDRV=$ydrv

is obsolete?

gyro

[gyro]

This version greatly simplifies the code at the expense of a little processing inefficiency.
Instead of reading the file "PUPSTATE" directly from the savefile/savefolder, it looks for it in the frugal install directory. So it doesn't have to wait until the savefile/savefolder is available before making the final decision.

To use:
1) Put patched "initrd.gz" in the frugal install directory.
2) Reboot to show that everything is still the same.
3) Copy "/etc/rc.d/PUPSTATE" to the frugal install directory.
4) Add "nosearch=sda3" as boot parameter (using appropriate partition instead of "sda3").
5) Reboot.

The "nosearch" facility can be deactivated by either removing the "nosearch" boot parameter or deleting the "PUPSTATE" file in the frugal install directory.

To add an adrv or ydrv to the boot:
1) delete the copy of "PUPSTATE" in the frugal install directory.
2) Reboot.
3) Copy "/etc/rc.d/PUPSTATE" to the frugal install directory.
4) Reboot

See first post for download.

gyro

[boof]

have downloaded version 3 of initxxxxx file and done all preparations. USB boots until

searching for sfs files

kernel panic unable to mount VFS root fs not mounted to (2.0)

Then a "debug" readout that I can't copy & paste and machine hangs with caps lock light flashing.

[mikeb]

Some of the complexity comes from trying to efficiently integrate it into the 'init' script.

well that is the heart of the matter... fiddling with the window boxes is fine but sometimes its worth peeking in the basement at the foundations.

Shinobar had the same problem...large scripts trying to cater for all the hoplessly messy puppyisms and variations.

My init in initrd is around 25k...thats a lot less crap to try and work around.

Its not a diff or 2 needed but a complete rewrite/re-evaluation.

I am sure gyro you could do a much better job yourself.
see whats needed...cut out things that are hopelessly obsolete and make a script to do the job...ps the cpio initrd adds to the complexity by the way. You could add nice features like better sfs handling.

mike