[RFC] 'Fixes' for initramfs creation

[cwedgwood]

---

INCOMPLETE*

This 'kinda' works for me.

I'm pushing this out now in the hope someone can help work out what the right approach is in a few places.

---

Motivation and Context

initframfs creation on Debian systems does not presently work because a critical file is missing

Partially addresses: #7904

Description

RFC right now, changes are for illustrative purposes. It will however build a functional initramfs.

How Has This Been Tested?

Local build testing, inspection of output and limited VM boot testing.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Performance enhancement (non-breaking change which improves efficiency)
• Code cleanup (non-breaking change which makes code smaller or more readable)
• Breaking change (fix or feature that would cause existing functionality to change)
• Documentation (a change to man pages or other documentation)

Checklist:

• My code follows the ZFS on Linux code style requirements.
• I have updated the documentation accordingly.
• I have read the contributing document.
• I have added tests to cover my changes.
• All new and existing tests passed.
• All commit messages are properly formatted and contain Signed-off-by.

[codecov]

Codecov Report

Merging #8796 into master will increase coverage by 0.11%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #8796      +/-   ##
==========================================
+ Coverage   78.73%   78.84%   +0.11%     
==========================================
  Files         382      382              
  Lines      117812   117809       -3     
==========================================
+ Hits        92758    92889     +131     
+ Misses      25054    24920     -134

Flag	Coverage Δ
#kernel	79.32% <ø> (-0.02%)	⬇️
#user	67.8% <ø> (+0.54%)	⬆️

---

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 841a7a9...6004973. Read the comment docs.

[rlaager]

Fixing the missing file seems clearly correct. Adding set -e seems reasonable. (Personally, I use set -eu in every shell script I write.) I'm less certain about the "error if missing" approach. Having zpool.cache be required is wrong.

[cwedgwood]

@rlaager thanks for the comments:

• agree set -eu seems better
• removed zpool.cache

re: "error if missing" - if we don't do that, we build initramfs which don't work; surely an error at such a time as someone can address/fix it it better than finding out after a reboot it's not working?

force-pushed the update, not sure if github noticed...

[cwedgwood]

what i pushed earlier wasn't supposed to work, it was for people to eyeball and comment on

however, in trying to make it work cleanly ... well, it's not clear or working

i'm wondering if instead we should just remove contrib/initramfs entirely?

[rlaager]

• agree set -eu seems better

If it works with set -u, that's fine. I wasn't suggesting set -u, just saying I use it personally for everything in addition to set -e. So your idea of using set -e is generally reasonable. (Though it's always possible that the initramfs script, or initramfs-tools generally, won't work with set -e without significant changes.)

• removed zpool.cache

My point was that zpool.cache should not be required. It should be in the optional list.

re: "error if missing" - if we don't do that, we build initramfs which don't work; surely an error at such a time as someone can address/fix it it better than finding out after a reboot it's not working?

Sure. If whatever problem you're catching is definitely fatal on boot, yes, it's better to catch it earlier.

i'm wondering if instead we should just remove contrib/initramfs entirely?

No. This is the initramfs script used by Debian and Ubuntu, at least.

[dreamcat4]

Can this one get some higher priority attention? Because otherise the missing file causes a kernel panic during boot. And that leads to broken (not bootable) system. Without any prior warnings / error messages to indicate anything was missing during previous building of initramfs.

[cwedgwood]

@dreamcat4 IIRC, I couldn't find a robust way to make sure zfs-functions was always present for packaging so this patch is incomplete (the change to add it work, but the file isn't always generated).

[c0d3z3r0]

@dreamcat4 @cwedgwood could you both test, if #9089 solves this for you, please?

[cwedgwood]

I'm going to close this in favor of #9089 which is more appropriate.