add example how to mount the correct data partition for redundant layout

[FabianKnapp]

This is one possible solution how to select the correct data partition when using redundant data slots.

As talked with ejoerns at #rauc.

Feel free to modify.

[codecov]

Codecov Report

Merging #370 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #370   +/-   ##
=======================================
  Coverage   66.53%   66.53%           
=======================================
  Files          24       24           
  Lines        6455     6455           
=======================================
  Hits         4295     4295           
  Misses       2160     2160

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 5c35d0d...ba07781. Read the comment docs.

[ejoerns]

Thanks for the contribution!

The example implementation provided is still very use-case specific.

I would suggest boiling this down to 2 lines of udev as you can use the expr magic etc. also inside the shell script.

Then having %n only is possibl in this case, but will not work when having data volumes on separate volumes for example, because names are ubi0_x ubi1_y then.

Instead I would use %k, i.e. the udev 'kernel' name for this device. A bit of cut and sed will extract the name you need for comparsion.

[FabianKnapp]

Yes, you are right, either the shell script or the udev rules are highly use-case specific:

Individual logic is always needed to find the correct rootfs slot for a given data slot.

I decided to make the udev rule more specific since I dont like shell scripting that much ;)

Will provide an example with the individual logic inside the shell script.

[ejoerns]

I decided to make the udev rule more specific since I dont like shell scripting that much ;)

You'd be the first one I know who likes udev scripting more than shell scripting, but.. ok.. 😏

[FabianKnapp]

For my case the udev was indeed easier since I get the partition for free from udev.

Another thing:
Barebox uses ubi0:$ubiroot notation for booting from nand (which is also returned by cat /proc/cmdline):

bootargs="$bootargs root=ubi0:$ubiroot ubi.mtd=$rootfs_mtdblock"
line 89

But udev uses ubiX_Y notation from the kernel.

Any idea how to solve this without patching / scripting barebox?

[FabianKnapp]

Ok, logic inside the shell script.

/usr/bin/is-parent-active :

#!/bin/bash

cmdline=$( cat /proc/cmdline )

# Modify this to get the corresponding rootfs partition from a data partition.
rootfs=$( echo $1 | sed 's/_3/_2/' )
rootfs=$( echo $rootfs | sed 's/_7/_6/' )

if [[ $cmdline == *"root=$rootfs "* ]]; then
	echo 1
else
	echo 0
fi

udev rules:

KERNEL=="ubi[0-9]_[0-9]", PROGRAM="/usr/bin/is-parent-active %k", ENV{isroot}="%c"
KERNEL=="ubi[0-9]_[0-9]", ENV{isroot}=="1", SYMLINK+="data"

Should we use this for documentation?

[ejoerns]

The udev rules now looks fine for me (maybe we can add another example for (more common) sdX devices, such as

KERNEL=="sd?[0-9]", PROGRAM="/usr/bin/is-parent-active %k", ENV{isroot}="%c"

but basically that's it.

The script however is a bit too specifc for an example and I also realized that using %k is not required in any case, but there's nothing bad to about having it ;)

If I remember correctly, you also have access to udev variables (e.g. $ID_PART_TABLE_UUID) from within the script.

One check I've used on a similar setup is

DISK_ID=$(cat /proc/cmdline | sed -e 's/^.*root=PARTUUID=\([0-9a-z-]*\).*/\1/')

[ "$DISK_ID" == "$ID_PART_TABLE_UUID" ]  ...

Maybe we can sum it up to something like

ROOTFS_DEV=<determine rootfs by using proc cmdline or mount>
TEST_DEV=<obtain expected rootfs device for processed device (%k)>

if [[ $ROOTFS_DEV == $TEST_DEV]]; then
	echo 1
else
	echo 0
fi

The actual handling is probably too use-case specific?

[FabianKnapp]

Yep, in my opinion

ROOTFS_DEV=<determine rootfs by using proc cmdline or mount>
TEST_DEV=<obtain expected rootfs device for processed device (%k)>

if [[ $ROOTFS_DEV == $TEST_DEV]]; then
	echo 1
else
	echo 0
fi

is the best for documentation. Maybe, we can add one specific example.

[jluebbe]

I'm wondering if this can't be done by letting udev look at the kernel cmdline:

IMPORT{cmdline}="root"
SUBSYSTEM="ubi", ENV{root}=="/dev/ubi0:aroot", ATTRS{mtd_num}=="1", ATTRS{name}=="adata", SYMLINK+="data"
SUBSYSTEM="ubi", ENV{root}=="/dev/ubi0:broot", ATTRS{mtd_num}=="1", ATTRS{name}=="bdata", SYMLINK+="data"

That way you have an explicit mapping by name. I've not tried this very recently, so I'm not sure I got the details correct.

Also often useful is:

SUBSYSTEM=="ubi", KERNEL=="ubi*_*", SYMLINK+="ubi_mtd%s{mtd_num}_%s{name}"

[ejoerns]

@jluebbe is right for this case and maybe we should use this by default.

It will not work when dealing with PARTUUID's as I had it in my setup, there we need a solution similar to what is in @FabianKnapp's last comment.

So, we should maybe start with the simple ubifs example and shortly note that other as we already did.

@FabianKnapp would you like to?

[FabianKnapp]

As always: Feel free to modify.

[jluebbe]

As always: Feel free to modify.

I've added another commit on top, which simplifies the udev rules to get rid of the temporary variable. That makes it possible to use just a single rule.
I don't have a system ready to try this, but it should work like this. :)
If you have a chance to test this successfully, I'd squash these commits into one and merge it.
Were you able to try the approach based on IMPORT{cmdline}="root" and no external tool?

[FabianKnapp]

Were you able to try the approach based on IMPORT{cmdline}="root" and no external tool?

Not yet, but it is on my todo's

[thirtythreeforty]

We hit this problem, so we solved it with a small tool and a simple configuration file:

https://github.com/KopisMobile/bootslot-mounts

It does require systemd, which is a lot of complexity on a very small system. But if you're already using systemd, it's pretty straightforward to use.

[jluebbe]

I've updated the branch to current master and changed some minor wording.

[ejoerns]

Added a minor fix in the example script and updated CHANGES.

Looks fine for me now.