| View previous topic :: View next topic |
| Author |
Message |
finalturismo
Guru
Joined: 06 Jan 2020
Posts: 410
|
 Posted: Fri Jul 01, 2022 8:05 pm Post subject: Building a custom initramfs /dev/disk missing? |
 |
|
So have an interesting situation.... and I am not sure what is causing it or what makes it unique to the system.
But i generated an initramfs on 1 computer and i extracted the files and slimmed it down. I than merged it inside the kernel and did a few other things. .......
Anyway when push comes to shove when ever i boot into the initramfs fs on any other computer besides the host it was created on the entire /dev/disk subset is missing from the device tree.
The disks show fine, the kernel modules are backed into the kernel.... but for some /dev/disk/by-id is missing /dev/disk/by-label. Anything that is under /disk/disk* is gone....
Anyone know what this is?
Trying to make a portable initramfs.... |
|
| Back to top |
|
 |
pingtoo
l33t
Joined: 10 Sep 2021
Posts: 916
Location: Richmond Hill, Canada
|
| Posted: Fri Jul 01, 2022 8:16 pm Post subject: |
|
|
| The "/dev/disk/" is create by udev. So do your initrd have udev binary? and your /init script call udev binary? |
|
| Back to top |
|
|
finalturismo
Guru
Joined: 06 Jan 2020
Posts: 410
|
| Posted: Fri Jul 01, 2022 8:37 pm Post subject: |
|
|
| pingtoo wrote: | | The "/dev/disk/" is create by udev. So do your initrd have udev binary? and your /init script call udev binary? |
ya i saw that and yes it does have udev and it works fine as long as it is being used on the host otherwise it seems udev doesnt do anything... |
|
| Back to top |
|
|
pingtoo
l33t
Joined: 10 Sep 2021
Posts: 916
Location: Richmond Hill, Canada
|
| Posted: Fri Jul 01, 2022 8:54 pm Post subject: |
|
|
Prior to invoke udev did you make udev update hwdb? i.e. | Code: | | udevadm hwdb --update |
How you generate initramfs? and what did you "slimmed it down"? |
|
| Back to top |
|
|
NeddySeagoon
Administrator
Joined: 05 Jul 2003
Posts: 54214
Location: 56N 3W
|
| Posted: Fri Jul 01, 2022 9:49 pm Post subject: |
|
|
finalturismo,
I would start from the bottom up, not by taking things out of an existing initrd.
What is the real problem you are trying to solve any why are the /dev/disk/... symlinks the solution?
_________________
Regards,
NeddySeagoon
Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail. |
|
| Back to top |
|
|
finalturismo
Guru
Joined: 06 Jan 2020
Posts: 410
|
| Posted: Fri Jul 01, 2022 11:11 pm Post subject: |
|
|
| NeddySeagoon wrote: | finalturismo,
I would start from the bottom up, not by taking things out of an existing initrd.
What is the real problem you are trying to solve any why are the /dev/disk/... symlinks the solution? |
Iam working on making a portable initramfs for a ZFS installation for quick dirty imports and exports of pools but in a smart way. So far i have a gutted kernel running with all the common hardware built in.
I have specific kernel parameters set that block the spa cache files and set a host id of zero for all hosts. But the risk iam running into is messing up the labels that are generated via UUID and create symlinks to the /dev/disk/by-id
Because of the race conditions of the standard block devices, not using a the UUIDs can cause a mix up of the device labels and cause the pool the be broken and at the point you lose some data.
Anyway i have a git project going for a custom kernel builder that builds an initramfs with ZFS built in and it works pretty good so far. I got the kernel , initram merged and set as an EFI stub partition with the EFI bootmgr creating a stub entry in the bios.
My 1 blockade seems to be the initramfs right now. I would attempt to get rid of the initramfs and go straight kernel but i need to some how call a pool import which i think mostly needs the initramfs. Iam sure its possible i just dont have enough knowledge for that part anyway.
So when i get stuck with stuff i come here , because you guys seem to be pretty smart. XD
I was doing some reading and it seems that alot of distros might have a host id of some sort attached to the initramfs that may create some issues. i dont know much where to start..... but at this point it might just be easier for me to downoad busy box and attempt to build my own?
| pingtoo wrote: | Prior to invoke udev did you make udev update hwdb? i.e. | Code: | | udevadm hwdb --update |
How you generate initramfs? and what did you "slimmed it down"? |
I just deleted the firmware folders, i don't need any of it.
And its a standard devuan initramfs , the same as debians for the most part. I just unpacked it with lsinitrd --unpack |
|
| Back to top |
|
|
NeddySeagoon
Administrator
Joined: 05 Jul 2003
Posts: 54214
Location: 56N 3W
|
| Posted: Sat Jul 02, 2022 11:18 am Post subject: |
|
|
finalturismo,
When your initrd fails, arrange for it to drop you to the busybox shell.
Check for the existence of /dev/... and friends.
They should be there by the time you type the command.
If that is true, you are trying to use them before udev has created them.
That's a well known and possibly forgotten thing by now but its why the handbook was changed to not use disk-by in fstab.
If they are still missing, udev is not creating them.
You may need the openrc sysinit scripts. I don't have udev, so I don't know their names.
udev-settle comes to mind and I think that there are a couple of others.
_________________
Regards,
NeddySeagoon
Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail. |
|
| Back to top |
|
|
pingtoo
l33t
Joined: 10 Sep 2021
Posts: 916
Location: Richmond Hill, Canada
|
| Posted: Sat Jul 02, 2022 7:51 pm Post subject: |
|
|
finalturismo,
I am afraid I will not be able to help much here.
I am not able to find where devuan initramfs-tools source code to check.
I look debian initramfs-tools source code, but I am not able to figure out how debian initramfs call udev, it seems only wait for udev event to settle. debian initramfs-tools seems depend on a lot of building host environment to create a initramfs. So without look at the building host I don't know where to begin.
BTW, Devuan seems to focus on create a system without systemd, so I wonder if they use a different logic for their initramfs creation.
If you write your own /init script for initramfs, maybe call blkid within your script after /sys is mounted will do the trick to get list of UUID. |
|
| Back to top |
|
|
dantrell
l33t
Joined: 01 Jun 2007
Posts: 915
Location: Earth
|
| Posted: Sun Jan 01, 2023 6:48 pm Post subject: |
|
|
| finalturismo wrote: | | Anything that is under /[dev]/disk* is gone.... |
| pingtoo wrote: | | The "/dev/disk/" is create by udev. |
The answer lies in Giles's answer on the Unix & Linux Stack Exchange.
Adapting his answer for the presence of sys-apps/systemd-utils in Gentoo, what you want to do is:
- Bundle udevadm (older initramfs' called this udevd and systemd-variants call it systemd-udevd so create symlinks accordingly if applicable).
- Bundle /lib/udev and the relevant rules (at least 1 persistent-storage rule) as well as whatever imported programs those rules mention (you don't need to bundle builtin programs, however, you probably want to bundle blkid anyway).
- Forget about calling udevadm as a daemon. Get a restrictive list of block devices (you can bundle lsblk for this) and use "udevadm test [options] devpath" on each block device (feel free to redirect the noise to /dev/null).
- Don't forget the step about moving /dev (if relevant).
Now, you can move on to the next issue. 
_________________
Dantrell B. |
|
| Back to top |
|
|
|
Portage & Programming
