Add support for poweroff time, or at least not error out on "now"

[xqrs]

We do not usually accept feature requests, please see the README. You may open a feature request
if you are a distribution/package maintainer and have a clear need for the feature, or if you are
willing to develop the feature yourself and would like to discuss it first.
It shouldn't be hard, I can hack around with C if you want.

Your background
A developer of a script that uses kexec.

The problem
kexec depends on the command: shutdown -r now to reboot the system, NOW!
But dinit's shutdown command does not support now as an argument so it errors out instead of rebooting.
Actually dinit's shutdown command does not support any timing at all, it just does it NOW!

The solution
Add in the shutdown command source code: if you found a stray "now" argument, just ignore it.

Additional context
See: https://github.com/horms/kexec-tools/blob/daa29443819d3045338792b5ba950ed90e79d7a5/kexec/kexec.c#L937

[davmac314]

Note that you can use kexec -l <kernel-to-load> (which doesn't try to execute "shutdown -r now") followed by shutdown -k.

[davmac314]

Isn't this really a problem with kexec rather than Dinit? It's assuming a particular shutdown command will be valid. You kind of went ahead with an implementation (first in #495 and then #497) without us ever getting past the "do we really want this" part of the process (i.e. the "discuss it first" part of the issue template which is quoted at the top).

While I think I'm ok with allowing "now" as an argument for general compatibility reasons, I'm less certain that we want logic to change shutdown based on whether a kexec-kernel has been loaded. If the user asks for a reboot (which is what "shutdown -r" is), why shouldn't they get a reboot?

You could always put "kexec if a kernel is loaded" logic into a shutdown hook script (see the shutdown man page).

[xqrs]

Isn't this really a problem with kexec rather than Dinit? It's assuming a particular shutdown command will be valid. You kind of went ahead with an implementation (first in #495 and then #497) without us ever getting past the "do we really want this" part of the process (i.e. the "discuss it first" part of the issue template which is quoted at the top).

You're right, I should have discussed it first, but it was fun anyway trying to hack around with Dinit.

While I think I'm ok with allowing "now" as an argument for general compatibility reasons, I'm less certain that we want logic to change shutdown based on whether a kexec-kernel has been loaded. If the user asks for a reboot (which is what "shutdown -r" is), why shouldn't they get a reboot?

Your logic makes sense about -r should just stands for a normal reboot, but what about other systems?

Should we simply not care about SystemD's shutdown behavior even if it's expected by most users and programs? Most people assume that the shutdown command works SystemD-style because that's the most common implementation by far. The fact that the kexec-tools maintainers expect that behavior since kexec-tools 1.0^[1] does give a strong argument about "expectedness" of -r's meaning of trying to pickup a loaded kexec kernel.

If a user loaded a kexec kernel (not an easy job or "done-without-knowing" mistake) and then they did a "shutdown -r", they could (and most people DO, even advanced users like the kexec-tools maintainers) expect the system to automatically pickup the loaded kexec kernel.

Personally, I thought -r just picks up the loaded kexec kernel just like SystemD. And I think most people believe that too because that's what SystemD does.

You could always put "kexec if a kernel is loaded" logic into a shutdown hook script (see the shutdown man page).

I don't think it's good for a script to just edit the shutdown hooks just for when trying to kexec a kernel.

Should I submit a patch for the kexec-tools maintainers to replace -r with -k (and add the -k option to OpenRC's shutdown command because they use a capital -K instead^[2]) instead?

[1] That's more than 19 years ago: https://github.com/horms/kexec-tools/blame/eb5c1bab95722a56cc44beddd1eba8dd9470c82f/kexec/kexec.c#L711

[2] That's actually weird, because they initially used a lowercase -k but then 8 years ago this patch got commited for no good reason..?
I think that's a bug because their documentation and their reboot script expect a lowercase -k.

[davmac314]

Sorry for the slow reply.

You could always put "kexec if a kernel is loaded" logic into a shutdown hook script (see the shutdown man page).

I don't think it's good for a script to just edit the shutdown hooks just for when trying to kexec a kernel.

What I meant was: if people (individuals, or distributions) want dinit to behave as systemd does in regards to kexec-on-reboot, they can implement it via a shutdown hook. I don't think an arbitrary script should try to set up this situation either, but your script could depend on "shutdown -r now" being "kexec aware" and one way to achieve that, with Dinit, is via shutdown hooks. I hope that makes sense. You just document the requirement; it's up to the end user to ensure that their system complies.

Should we simply not care about SystemD's shutdown behavior even if it's expected by most users and programs? Most people assume that the shutdown command works SystemD-style

The problem I have with this is that you can extend this argument to any aspect of Systemd's behaviour and come to the conclusion that Dinit should behave that way, too. But then there's hardly any point having a separate piece of software.

I suspect that most people don't use kexec and so don't actually expect this particular behaviour, anyway.

The fact that the kexec-tools maintainers expect that behavior since kexec-tools 1.0

My suspicion is that even with Sys-V init you could implement the functionality as part of a shutdown script; it didn't have to be implemented by the init system itself. So what the Kexec-tools maintainers were relying on was quite possibility particular support via SysV init configuration, and then Systemd shutdown just implemented it directly - perhaps. Kexec has been around longer than Systemd, anyway.

Should I submit a patch for the kexec-tools maintainers to replace -r with -k (and add the -k option to OpenRC's shutdown command because they use a capital -K instead^[2]) instead?

I suppose that this would break compatibility with Systemd and so probably isn't going to be accepted either (in kexec-tools), but perhaps kexec should support being able to configure (either at run time or build time) how shutdown is called, for example. Especially if OpenRC has different requirements again.

The same might be true with your script. Maybe you execute a kexec-loaded-kernel one way or another depending on a run-time option and allow easily editing the default somehow.

[xqrs]

Sorry for the slow reply.

That's ok, No hurries.

You could always put "kexec if a kernel is loaded" logic into a shutdown hook script (see the shutdown man page).

I don't think it's good for a script to just edit the shutdown hooks just for when trying to kexec a kernel.

What I meant was: if people (individuals, or distributions) want dinit to behave as systemd does in regards to kexec-on-reboot, they can implement it via a shutdown hook. I don't think an arbitrary script should try to set up this situation either, but your script could depend on "shutdown -r now" being "kexec aware" and one way to achieve that, with Dinit, is via shutdown hooks. I hope that makes sense. You just document the requirement; it's up to the end user to ensure that their system complies.

Should we simply not care about SystemD's shutdown behavior even if it's expected by most users and programs? Most people assume that the shutdown command works SystemD-style

The problem I have with this is that you can extend this argument to any aspect of Systemd's behaviour and come to the conclusion that Dinit should behave that way, too. But then there's hardly any point having a separate piece of software.

I suspect that most people don't use kexec and so don't actually expect this particular behaviour, anyway.

The fact that the kexec-tools maintainers expect that behavior since kexec-tools 1.0

My suspicion is that even with Sys-V init you could implement the functionality as part of a shutdown script; it didn't have to be implemented by the init system itself. So what the Kexec-tools maintainers were relying on was quite possibility particular support via SysV init configuration, and then Systemd shutdown just implemented it directly - perhaps. Kexec has been around longer than Systemd, anyway.

I mostly agree with your reasoning, we ain't a cheap copy of SystemD.

Should I submit a patch for the kexec-tools maintainers to replace -r with -k (and add the -k option to OpenRC's shutdown command because they use a capital -K instead^[2]) instead?

I suppose that this would break compatibility with Systemd and so probably isn't going to be accepted either (in kexec-tools), but perhaps kexec should support being able to configure (either at run time or build time) how shutdown is called, for example. Especially if OpenRC has different requirements again.

The same might be true with your script. Maybe you execute a kexec-loaded-kernel one way or another depending on a run-time option and allow easily editing the default somehow.

K.

Note that the kexec-reboot perl script^[1] (not mine), used by many, does depend on check-for-kexec behavior.
It's up to you if you want to actually merge it. And for my script, I would just do init-system detection then choose the proper option.

[1] https://github.com/error10/kexec-reboot/blob/master/kexec-reboot#L260