STICKY: Making your own custom burn-n-boot Raspbian image

[pe78420]

Hi HawaiianPi,

Very helpfull

Everything is ok with the raspbian_2018-11-15 release even when I tried to add or modify any file (ssh, config.txt, cmdline.txt, wpa_supplicant.conf, etc.)... To do that, I copied them from the windows explorer or after mounting the image on Debian via kpartx or loseup.

But I got a problem
On the latest raspbian image (2019-04-09), booting stucks when I tried to add of modify the same files!
Without modification (the raw image), booting is always successfull...

What's wrong!
Any ideas?

Many thanks in advance and best regards,

Philippe
A french guy...

[LittleMike]

That sounds very useful to me! Is there a way that I can even use apt to install packages that is not in the original image myself? I assume I need to mount the image to an real Raspberry Pi to do that.... if it is even possible.

Right, so begin by installing systemd-container (this will give you the systemd-nspawn utility) and zerofree, if you don't already have them. This will not take long:

Code: Select all

pi@raspberrypi:~ $ sudo apt-get update
pi@raspberrypi:~ $ sudo apt-get install -y systemd-container zerofree

Then, assuming you have already downloaded the zipped image (from here), unpack, and then loop mount it:

Code: Select all

pi@raspberrypi:~ $ unzip raspbian_latest 
Archive:  raspbian_latest
  inflating: 2018-11-13-raspbian-stretch.img 
pi@raspberrypi:~ $ sudo losetup --show -P -f 2018-11-13-raspbian-stretch.img
/dev/loop0

Your reported loop device may differ (loop1, loop2 etc.) in which case modify the following instructions accordingly.
This losetup command should have created partition mappings automatically:

Code: Select all

pi@raspberrypi:~ $ ls /dev/loop0*
/dev/loop0  /dev/loop0p1  /dev/loop0p2

Yep, two partitions, as expected. Now mount them:

Code: Select all

pi@raspberrypi:~ $ sudo mkdir -p /mnt/raspbian
pi@raspberrypi:~ $ sudo mount -v /dev/loop0p2 /mnt/raspbian
mount: /dev/loop0p2 mounted on /mnt/raspbian.
pi@raspberrypi:~ $ sudo mount -v /dev/loop0p1 /mnt/raspbian/boot
mount: /dev/loop0p1 mounted on /mnt/raspbian/boot.

Now we can chroot in. We'll use systemd-nspawn to do this; it will take care of all the necessary behind-the-scenes bind-mounts etc.:

Code: Select all

pi@raspberrypi:~ $ sudo systemd-nspawn --directory=/mnt/raspbian
Spawning container raspbian on /mnt/raspbian.
Press ^] three times within 1s to kill container.
root@raspbian:~#

Can this be done on the x86 Raspi image? The download is an iso, but I used dd to convert it to an img and tried the above steps. Most of it seemed to work, but whenever I get to the systemd-nspawn portion, it yells at me that there is no OS-tree. So I tried putting a /usr/bin and an /etc folder but then it tells me that it's mounted read-only.

My goal is to create a Live USB stick running Raspberry Pi Desktop that has some base packages installed and some minor configuration completed so that I have a "baseline" because I have a habit of breaking things and have needed to start all over from scratch each time as I learn and that's just gotten really annoying.

*EDIT*

So I tried a different method of just extracting all the contents of the ISO to a folder and added the /etc and /usr/bin folders manually. Now when I try to run systemd-nspawn it tells me execv() failed, no such file or directory.

[sakaki]

LittleMike,

in principle this technique should be usable with an x86 (or x86_64) guest linux image, running under an x86 (or x86_64) host linux system, such as the Debian with Raspberry Pi Desktop OS, for example.

Could you provide some more details about your host system (e.g., is it running the RPi Desktop for PC, or some other Linux variant) and what you see at /mnt/raspbian (if you run e.g. "ls -l /mnt/raspbian"; assuming you are using the same paths as given in the instructions).

For disclosure, I have used the systemd-nspawn approach extensively on PCs running Gentoo and Arch (with Gentoo, Arch and vanilla Debian guests) but haven't actually used the "Debian with Raspberry Pi Desktop" image in either capacity to date.

Best, sakaki

[rin67630]

You know, it is so easy to have a headless reference Raspberry Pi (can be a cheap PiZero) with all your loved software that you keep up-to date and from which you copy the SD card each time you need another one...

[LittleMike]

LittleMike,

in principle this technique should be usable with an x86 (or x86_64) guest linux image, running under an x86 (or x86_64) host linux system, such as the Debian with Raspberry Pi Desktop OS, for example.

Could you provide some more details about your host system (e.g., is it running the RPi Desktop for PC, or some other Linux variant) and what you see at /mnt/raspbian (if you run e.g. "ls -l /mnt/raspbian"; assuming you are using the same paths as given in the instructions).

For disclosure, I have used the systemd-nspawn approach extensively on PCs running Gentoo and Arch (with Gentoo, Arch and vanilla Debian guests) but haven't actually used the "Debian with Raspberry Pi Desktop" image in either capacity to date.

Best, sakaki

Sorry, I didn't get an email notification you responded.

I'm running Raspberry Pi Desktop using the Live USB distro, running it with persistence.

I removed/redid everything a few times, so I'll have to go back to the start and try again (I gave up on it, honestly) but there was a boot directory and a bunch of others. No /usr/bin, but I was able to create one, and no /etc either, but I was able to create one. That's how I got as far as I did.

I'm wondering if it's less the Raspberry Pi Desktop being the issue and more that it's the Live USB that's the issue (along with the gnome bootloader). No syslinux file, I don't think.

[MyDearDeveloper]

sakaki wrote: ↑

Right, so begin by installing systemd-container (this will give you the systemd-nspawn utility) and zerofree, if you don't already have them. This will not take long:

Code: Select all

pi@raspberrypi:~ $ sudo apt-get update
pi@raspberrypi:~ $ sudo apt-get install -y systemd-container zerofree

Then, assuming you have already downloaded the zipped image (from here), unpack, and then loop mount it:

Code: Select all

pi@raspberrypi:~ $ unzip raspbian_latest 
Archive:  raspbian_latest
  inflating: 2018-11-13-raspbian-stretch.img 
pi@raspberrypi:~ $ sudo losetup --show -P -f 2018-11-13-raspbian-stretch.img
/dev/loop0

Thanks for this informative post i was struggling but this helped me

[bmanske]

I'm going to resurrect this topic. I have decided to work on a project requiring the use of several raspberry pis and the thought of going through the setup for each one is hard to swallow. So I started looking for this topic.

First:
SET UP A HEADLESS RASPBERRY PI, ALL FROM ANOTHER COMPUTER’S COMMAND LINE
https://hackaday.com/2018/11/24/set-up- ... mand-line/

November 24 2018 this article proceeded this topic. I have to wonder why it was never brought up in this discussion.

Second: Peter Lorenzen deserves credit for this very efficient shell script
Trying to get to http://peter.lorenzen.us/linux/headless ... figuration it never responded. So I went to the way back machine to get the shell script.
https://web.archive.org/web/20190131013 ... figuration
Using the final snapshot of this page is where I started.

I don't know if I got the latest version of the code. I decided to build a VMware Debian 10 machine with the Cinnamon desktop. I don't know what he used. This script would no longer run. The website obviously changed as there is one more redirect now than was accounted for in his code. I had some problems with curl giving me decent results worked almost always from the command line when requesting the https URL and almost always failed when run from the shell script. The format of the redirect message had also changed to instead of doing an awk on the second http in the file I implemented a quick search for it. I also didn't like his defaults so some of those were changed. You get the picture, I really extended the length of the script by adding more error checking and comments. I've used features of bash prior to v4.0 to try to maintain compatibility.

Three: Going Forward
I want to replace the rsync command with code to look at where the file is being copied to and set the ownership and permissions without the need for getting all of the files set up in the right condition before hand. Adding a ability to specify an include file for configuration as the command line is getting long among other ideas.

HERE's THE QUESTION.
I don't want to just set up SSH I also want RealVNC server enabled. I use TightVNC on all of my local machines and have changed a RealVNC config file to allow the 2 to communicate so I know SED and AWK will get that for me in the script. Can someone point me to the source code for the rasberry pi config or just tell me how to do it? The source code would allow me to learn how to enable / disable other devices also.

I've been thinking about putting the project on GitHub. Any interest in this?

[RonR]

Can someone point me to the source code for the rasberry pi config or just tell me how to do it?

If you're referring to 'raspi-config', it's a script located at: /usr/bin/raspi-config

[abraham5]

Nice Write Up Man, keep up the good work.

[dmikester1]

This sounds like something I was just reading about the other day: pibakery. Does that do a similar thing?
https://www.raspberrypi.org/blog/pibakery/

[bmanske]

RonR - thank you for pointing to the raspi-config shell script. I thought it was a compiled executable so I never opened it. It is probably one of the best written shell scripts I have ever seen. I found the project online and it is under the MIT license so I'm free to steal... I mean copy - paste sections of the script.

dmikester1 - I did try out Pi Bakery. It was just too limited for what I wanted to do. For example, checking for a new version before downloading, exiting on error, editing arbitrary files in the file system. This could be fixed by writing custom cards but I really didn't want to learn.

Sorry for the delay with replies but I was out of town for about a week. I'll make another post when I put it on .Git Hub.

[bmanske]

I've put what I have so far at https://github.com/bwmanske/rpido/tree/master for those who are interested.

[bmanske]

It's been a couple of weeks since I last posted and nobody has come after me for copyrights. So I must be good there.

I have added a HostCount option so you can configure the card and then it will create the number of images indicated host HostCount. Each will have a suffix on the file name and the suffix will be on the HostName as well. I'm hoping this helps code clubs, people experimenting with clusters and classrooms. If you fall into one of these categories let me know if this helps you and let me know your suggestions.

I have set up the potential for config directories 1 thru 9. -c 0 attempting to get to config 0 will loop through the possible config directories and print the messages describing the config setup (of course, you have to add the descriptions yourself). If a config number is asked for and it doesn't exist then a template file and directories will be created. I'm trying to detect if there are valid options and all of the variables are set. If a problem is detected, a rpido-configX.sh file will be created so you can compare your file to this latest template to see what is missing.

There is also a rpido-target.sh file that will read the config and actually set the options. Right now nothing makes actual changes to the system but that 's coming in the next few days. I just needed to get all of this structure right first.

The -k for keep file system mounted and the arbitrary command line command have been removed. If you use the -s option then a shell will open and you will be in a chroot shell at the root of the file system. qemu will be running the bash command shell from the image. This should be a quick way to mount and change the image from your Linux system. If using -s with a config the config executes first and then you are dropped into the shell. Type "exit" to get out of the shell when you have finished and all will be unmounted. I felt this was better than just leaving the file system forever mounted.

My next step is to be able to configure SSH, WiFi, VNC after that is country, timezone, keyboard, etc...

While it may sound like I know what I'm doing... I really don't. I don't know what options are needed. So it you have some ideas, requests or suggestions please let me know. I'm really trying to put together something that is useful for more than just myself.

[jaromrax]

Hi,
I was looking for a SD mod solution and I have found very promising
- PiBakery , last activity Oct 2018 - https://github.com/davidferguson/pibakery . Javascript, so it is more complex than I like.
- this RPiDo - https://github.com/bwmanske/rpido - from bmanske, that is just active. And it is command-line, the style.

I didn't go too deep, but it seems to me, that both projects basically mount the existing image and copy stuff into /boot. It is great that somebody works on RPiDo, however it is very brief on usage and examples and thus it makes some barrier to use it.

1/ I would like to see the examples in README.md to see where it really is now.
2/ What combinations do work ? - jessie, stretch, buster, Pi2, 3B, ZeroW, Pi4 ?

3/ One would also need to add his own blocks - e.g. download and start docker. It seems that there is very nice organization model of blocks https://github.com/davidferguson/pibakery-blocks . Maybe using the efforts out of (orphaned?) PiBakery blocks and documenting Pido would make a new quality. And attracted people that would share their new blocks?

Waiting for a comment of bmanske or bwmanske
cheers, jarom

[bmanske]

Jarom,

Thank you for the feedback. I haven't had very much so I really appreciate it. I hope that some more people that it may be useful for would offer some. I will add some examples to the readme and maybe that will help others come up with questions that I could then write about.


Everyone,

Again, I don't have much experience with shell scripts or github projects so I appreciate all kinds of constructive criticism.

To answer another of Jarom's queries... I have tried this on the 3B and 4B. I tried the latest release of Raspbian and the one previous. I'll add this to the readme also. Hopefully, others can try it and let me know if it works for them.

Recently I have added enough code to set the country / keyboard / wifi. I ran into some issues with enabling RealVNC as the card wipes out the settings as the it boots.

I have been concentrating on trying to get this much working so that I can test without using my TV as a monitor. I'll post another update soon.

[jaromrax]

Dear bmanske,
I tried briefly your code.
1. The documentation is necessary. yes, `rpido -h` shows a help. Many things are not clear.
2. crash of curl in the beginning leaves one in doubts, if the retrieved image is ok
3. sudo dd and sudo rm are very dangerous commands, I added confirmations before I have had run the code
4. user must find the necessity to install qemu-user-static
5. chroot ${PWD}/${RPI_ROOT} - adding $PWD worked to me
6. apt upgrade crashed with openssh-server

On the other hand, your code is very clean, nice and you have dedicated lot of effort to it. Congratulations.

I will try to play with it more. It seems that it is possible to install some packages before writing the image. Modular design would make a big change and opened space for other contributors.

Good luck

[bmanske]

Jaromrax,

You are the first person I have heard from that has tried this code! Thank you for your suggestions and feedback.

I'll try writing some documentation over the next day or two in the wiki on github.

I have had problems with the curl command before. Would you please tell me how it crashed for you?

Can you tell me more about #5 & 6?

Have you looked at the sample config file generated? Use the "-c 0" command line option and rpido will make one for you. I tried to leave a bunch of notes in there on use of config files.

I'll put some thought into making more modules out of this...

As always I'm always open to suggestions or even lessons on writing shell scripts.

Thank you

[Lewis-H]

Thanks! These codes are helpful! It worked on my raspberry pi.
Regards,
Lewis

[blee0518]

As others have said, yes absolutely this can be done. For the sake of concreteness, I'll provide a compete workflow for doing so in this post.

sakaki
[/quote]

Can I get a little help?
Followed your instructions on a fresh RPi4 with Rasbpian Lite.

Code: Select all

root@raspberrypi:/home/admin# losetup --show -P -f ./root/openwrt-RaspberryPi3-GO2020-03-01.img
/dev/loop0
root@raspberrypi:/home/admin# ls /dev//loop0*
/dev//loop0  /dev//loop0p1  /dev//loop0p2
root@raspberrypi:/home/admin# mount -v /dev/loop0p2 /mnt/rooterroot/
mount: /dev/loop0p2 mounted on /mnt/rooterroot.
root@raspberrypi:/home/admin# mount -v /dev/loop0p1 /mnt/rooterroot/boot
mount: /mnt/rooterroot/boot: mount point does not exist.
root@raspberrypi:/home/admin# mkdir /mnt/rooterroot/boot
root@raspberrypi:/home/admin# mount -v /dev/loop0p1 /mnt/rooterroot/boot
mount: /dev/loop0p1 mounted on /mnt/rooterroot/boot.
root@raspberrypi:/home/admin# systemd-nspawn --directory=/mnt/rooterroot/
Spawning container rooterroot on /mnt/rooterroot.
Press ^] three times within 1s to kill container.
execv(/bin/bash, /bin/sh) failed: Exec format error
Container rooterroot failed with error code 1.
root@raspberrypi:/home/admin#

[sakaki]

Code: Select all

root@raspberrypi:/home/admin# losetup --show -P -f ./root/openwrt-RaspberryPi3-GO2020-03-01.img
...
root@raspberrypi:/home/admin# systemd-nspawn --directory=/mnt/rooterroot/
Spawning container rooterroot on /mnt/rooterroot.
Press ^] three times within 1s to kill container.
execv(/bin/bash, /bin/sh) failed: Exec format error
Container rooterroot failed with error code 1.
root@raspberrypi:/home/admin#

In this context, an exec format error normally indicates you are trying to chroot (or as here, non-boot nspawn, which is very roughly the same thing) into a 64-bit userspace guest image, from a host system running a 32-bit kernel.

That's because although the ARMv8 architecture allows a 64-bit kernel to natively run both 32-bit and 64-bit userspaces, it restricts 32-bit kernels to only (natively) run 32-bit userspaces.

So, when the nspawn command tries to start a shell inside your image, your 32-bit (Raspbian, host) kernel finds a 64-bit executable (e.g. /bin/bash inside the target image rootfs) and throws this error.

Incidentally, this issue doesn't arise for the original instructions, as they were targeted at modifying a Raspbian image (which has a 32-bit userspace) rather than an arbitrary one (such as openwrt).

The good news though, is that (assuming you are running on a 3B, 3B+ or 4B) you can easily switch your Raspbian host to use the 64-bit (official) kernel; after doing so ,your existing (32-bit) Raspbian userland will continue to work fine, but you should then also be able to chroot/nspawn into aarch64 userlands too (such as, presumably, your openwrt image, although I haven't tried it myself).

To do so, on your host system edit the file /boot/config.txt, and add to the end:

Code: Select all

[all]
arm_64bit=1

Save this file, and reboot. You will now be running a 64-bit kernel.

Now try setting up a loop mount and entering your image again. Any more luck this time?

hth, sakaki

[blee0518]

Code: Select all

root@raspberrypi:/home/admin# losetup --show -P -f ./root/openwrt-RaspberryPi3-GO2020-03-01.img
...
root@raspberrypi:/home/admin# systemd-nspawn --directory=/mnt/rooterroot/
Spawning container rooterroot on /mnt/rooterroot.
Press ^] three times within 1s to kill container.
execv(/bin/bash, /bin/sh) failed: Exec format error
Container rooterroot failed with error code 1.
root@raspberrypi:/home/admin#

The good news though, is that (assuming you are running on a 3B, 3B+ or 4B) you can easily switch your Raspbian host to use the 64-bit (official) kernel; after doing so ,your existing (32-bit) Raspbian userland will continue to work fine, but you should then also be able to chroot/nspawn into aarch64 userlands too (such as, presumably, your openwrt image, although I haven't tried it myself).

To do so, on your host system edit the file /boot/config.txt, and add to the end:

Code: Select all

[all]
arm_64bit=1

Save this file, and reboot. You will now be running a 64-bit kernel.

Now try setting up a loop mount and entering your image again. Any more luck this time?

hth, sakaki

https://i.pinimg.com/236x/09/3a/49/093a ... -scott.jpg

sorry for delay i wasnt subscribed
My next question would be: (yes i am using openwrt image, i am also trying to use the fork Goldenorb rooter as it gives me great cellular module tools, trying to create a cellular gateway in container)
Would this be the route to take when setting up this container as a service, machinectl (systemd.nspawn) ?

I really want this to boot up as a service when raspbian starts up.
I did get it setup as a machine `/var/lib/machines/openwrt` and ` /etc/systemd/nspawn/openwrt.nspawn` and setup a bridge to the container.
The issue I am having with this particular setup is from inside container, openwt, I can not access USB ports on the host RPi4.
Which is the underlying road block i am at now.

These are the 2 tutorials i am following:
https://openwrt.org/docs/guide-user/net ... /ltedongle

https://openwrt.org/docs/guide-user/sto ... quickstart

When i get to ls /dev/cdc-wdm0 the dir does not exist. but should of been added during installing packages.

I see you replied to some of my posts, thank you and i have now read them all. The reason for this madness is this.
The RPi4 is prices in a very affordable means to make it cellular router. the downfall is not being able to use most of the hardware abilities. That is why i want to run openwrt and a light linux server( file server / media center/etc...) I may even implement a VNC setup to play games(long shot to ever really doing this) . i do use the terminal for most everything. I am also in a position i need this to be usable by average person (yea, right at this point lol)
Could I run a file server / media server using openwrt ? could i use systemd to create a raspbian container on openwrt ?

Oh and GoldenOrb does not support natively RPi4 yet, as openwrt is only running a snapshot of it. I have been able to start container using RPi2 openwrt firmware. RPi3 doesnt play nice when starting container. but this maybe going back to the 32 - 64 bit thing.

[ganzgustav22]

Does anybody know how to script commands while inside the raspbian image with systemd-nspawn?

Basically, I hope for a script to setup an image like below. That is, mount raspbian image, nspawn into it, do stuff there, exit, unmount, e2fsck, done. But this doesn't work because all the commands after the "systemd-nspawn ..." command are not executed inside the raspbian image. The script just waits until I type "exit" to get out of the raspbian image and then executes the commands (while already being "back" on the host machine) which is of course not what I need.

Code: Select all

#!/bin/bash

losetup --show -P -f 2020-02-13-raspbian-buster-lite.img

mount -v /dev/loop0p2 /raspberry-image-maker/raspbian
mount -v /dev/loop0p1 /raspberry-image-maker/raspbian/boot

systemd-nspawn --directory=/raspberry-image-maker/raspbian

# --- do stuff inside the raspbian image here ---
apt-get update
apt-get install yadda yadda
cp something /etc/somewhere

exit
# --- done with stuff inside the raspbian image ---

umount -v /raspberry-image-maker/raspbian/{boot,}
dosfsck -a -v /dev/loop0p1
e2fsck -f /dev/loop0p2

zerofree -v /dev/loop0p2

losetup -d /dev/loop0
[code]

[sakaki]

Does anybody know how to script commands while inside the raspbian image with systemd-nspawn?

Well, per e.g. these notes, you could provide your desired command 'payload' as input to a secondary shell started inside the container.

So, for your example:

Code: Select all

#!/bin/bash
<... snip - set up loop mount etc ...>

systemd-nspawn --directory=/raspberry-image-maker/raspbian /bin/bash <<EOF
# --- do stuff inside the raspbian image here ---
apt-get update
apt-get -y install yadda yadda
cp something /etc/somewhere
# will exit automatically when done
EOF
# --- done with stuff inside the raspbian image ---

<... snip - close out loop mount etc...>

Note use of -y to apt-get, so large packages don't stall waiting for your confirmation.

There are other ways to achieve what you want, but the above is relatively straightforward for non-interactive command sequences.

With more modern versions of systemd-nspawn (that that in the current Raspbian) you should use the --pipe command line option, but that is moot at the moment.

hth,

sakaki

[ganzgustav22]

That works, thanks a lot.

[Rukbat]

Cool! Can I do this in Windows?

Goes to show you what you can do if you don't know anything. Ext2 Volume Manager allows you to mount the root image as a drive in Windows. EditPadLite (or Edit Plus, which I've been using since dinosaurs) lets you save a text file with *nix EOLs. And just opening Notepad, then saving it as a file (don't even put a space in it, or it takes an entire block) works to create the ssl file (basically like "touch ssl").

So I did the command.txt and the cmdline.txt changes, added wpa_supplicant.conf (which I shouldn't have for this RPi, because it's connected with eth0 and I don't want any WiFi entry hole into it) all in Windows. (Since I didn't want to do anything else, I didn't, but I could have. No x86 version of Raspbian needed, no VM running the x86 version, just raw Windows.

When people tell you that you can't, you don't. When you haven't been told, you just look for a way to do it. (The RPi 4 is running my WordPress site for a lot less electricity than this 800 Watt desktop now. I waited until it was running to install VNC because, even though I have a monitor plugged into the Pi, I only have one fully-working Unicomp keyboard [if you don't hunt and peck, you need a Unicomp or an expensive old real IBM keyboard] and RealVNC client on Windows just works with the Pi VNC server.)