Bug 869902 – app-containers/lxc-5.0.1 do not respond on SIGPWR or lxc-stop -n name-of-container

• Status: UNCONFIRMED
• Product: Gentoo Linux

[Description Yuriy Dmitriev 2022-09-12 19:23:05 UTC]

lxc-stop infinity wait when issue # lxc-stop -n cont_name
And do not respond on signal SIGPWR.

[Comment 1 Yuriy Dmitriev 2022-09-13 00:02:48 UTC]

Problem localized. Workaround is compile lxc without io-uring flag.
Kernel 5.15.67
sys0libs/liburing-2.2

Seems like kernel uring systems bug. May be badly backported patch.

[Comment 2 Yuriy Dmitriev 2022-09-13 15:34:55 UTC]

More detail:

vanilla-sources-5.15.64 work as expected. LXC compiled with io-uring.
vanilla-sources-5.15.67 DO NOT work as expected. LXC compiled with io-uring is temporary workaround solution.

5.15.67 work ONLY IF compiled without io-uring USE flag.

System fully updated, no add/remove packages along tests changed.

Move method to new kernel as usual:
make oldconfig && make -j17 && make -j17 modules_install

Resume:
This is kernel bug. Version between 5.15.[65-67] at this moment 67 is latest LTS.


With best wishes - YD.

[Comment 3 Joonas Niilola 2022-09-14 17:10:36 UTC]

If you edit your container's /etc/inittab adding:
  pf:12345:powerwait:/sbin/halt

to the bottom of it, does it help?

Anyway I guess these are related:
  https://github.com/lxc/lxc/issues/4118
  https://github.com/lxc/lxc/issues/4114
  https://github.com/lxc/lxc/issues/4112 has a patch you could maybe try if it helps.

[Comment 4 Yuriy Dmitriev 2022-09-14 18:42:00 UTC]

(In reply to Joonas Niilola from comment #3)
> If you edit your container's /etc/inittab adding:
>   pf:12345:powerwait:/sbin/halt
> 
> to the bottom of it, does it help?
> 
> Anyway I guess these are related:
>   https://github.com/lxc/lxc/issues/4118
>   https://github.com/lxc/lxc/issues/4114
>   https://github.com/lxc/lxc/issues/4112 has a patch you could maybe try if
> it helps.

Of course!)
I have this record)
 pf:12345:powerwait:/sbin/halt

This container I made from gentoo template script. It live few years, all works fine, as expected. Normally I can reboot, shutdown, poweroff container from containers shell. Or lxc-stop, lxc-attach, lxc-console from host OS.

Now, all listed above stop working. I only can connect via ssh. Container itself also work. May be partialy, I am not shure.

If I boot server from vanilla-sources-5.15.67, I must recompile with (disabled)-io-uring USE flag. 

> Anyway I guess these are related:
>   https://github.com/lxc/lxc/issues/4118
>   https://github.com/lxc/lxc/issues/4114
>   https://github.com/lxc/lxc/issues/4112 has a patch you could maybe try if
> it helps.

 Thanks, I investigate it....


With best wishes - YD.

[Comment 5 Yuriy Dmitriev 2022-09-14 18:52:10 UTC]

*Anyway I guess these are related:
*  https://github.com/lxc/lxc/issues/4118
*  https://github.com/lxc/lxc/issues/4114
*  https://github.com/lxc/lxc/issues/4112 has a patch you could maybe try if it *helps.

Yes yes yes, This is like same. 

strace show me readmsg.....infinity.

We are wait patch from upstream.

I reported few critical bugs around LXC (io-uring related). Upstream developers fix it around working day.

But for example, qemu also use io-uring. Absolutely correct code, completely no problem with io-uring.

With best wishes - YD.

[Comment 6 Yuriy Dmitriev 2022-09-18 20:48:33 UTC]

Last updates:
vanilla-sources-5.15.68 (LTS)
app-containers/lxc-5.0.1-r1

No changes.

# lxc-start -n cont_name -F
.
.
Console have output from container. Container normal boot. But no input into container(login/pass).

Roll back to vanilla-sources-5.15.64

With best wishes - YD.