Unbootable system with kmod-31

Unbootable system with kmod-31

[Kris Karas (Bug Reporting)]

[ Resend due to HTML not making it through ]

Hello Lucas, et al,

I upgraded from kmod-30 to kmod-31, and could not boot due to an
inability to load any module.  Any invocation of insmod bombs with,
"decompression failed with status 6".  I strace'd the issue to the
call to finit_module().  In kmod 30, the call to init_module()
works fine.

All modules are xz-compressed, for which support is present in the
kernel (using mainline kernel 6.5 here).

     CONFIG_HAVE_KERNEL_XZ=y
     CONFIG_KERNEL_XZ=y
     CONFIG_MODULE_COMPRESS_XZ=y
     CONFIG_FW_LOADER_COMPRESS_XZ=y
     CONFIG_XZ_DEC=y

However, on a whim, I tried "make nconfig", searched for any
kernel config options containing "XZ", and at the top of the list:

     # CONFIG_DECOMPRESS_XZ is not set

That appears to be an automatically-set option triggered by the
presence of initrd; but:

     # CONFIG_BLK_DEV_INITRD is not set

I don't know if (de)compression on an initrd is related to same
on a module, but it was the only thing I could find where XZ was
not =Y.

Kris

Re: Unbootable system with kmod-31

[Marco d'Itri]

On Oct 03, "Kris Karas (Bug Reporting)" <bugs-a21@moonlit-rail.com> wrote:

> I upgraded from kmod-30 to kmod-31, and could not boot due to an
> inability to load any module.  Any invocation of insmod bombs with,
> "decompression failed with status 6".  I strace'd the issue to the
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1050586

-- 
ciao,
Marco

Re: Unbootable system with kmod-31

[Lucas De Marchi]

On Tue, Oct 03, 2023 at 11:08:29PM +0200, Marco d'Itri wrote:
>On Oct 03, "Kris Karas (Bug Reporting)" <bugs-a21@moonlit-rail.com> wrote:
>
>> I upgraded from kmod-30 to kmod-31, and could not boot due to an
>> inability to load any module.  Any invocation of insmod bombs with,
>> "decompression failed with status 6".  I strace'd the issue to the
>https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1050586

duplicate of https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1050582 ?

Maybe that patch needs to propagate back to 6.5?

Lucas De Marchi

>
>-- 
>ciao,
>Marco
>

Re: Unbootable system with kmod-31

[Kris Karas (Bug Reporting)]

Lucas De Marchi wrote:
> On Tue, Oct 03, 2023 at 11:08:29PM +0200, Marco d'Itri wrote:
>> https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1050586
>
> duplicate of https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1050582 ?

Ah, that does seem the better bug report, as it triages it down to the 
dictionary size.

> Maybe that patch needs to propagate back to 6.5?

If you mean that modules_install doesn't pass the right parameters to 
/bin/xz, then definitely.  I've reported on bugzilla variously over the 
years about things "fixed in Linus's tree" that never got queued into 
Greg KH's tree.

Perhaps kmod should be updated to only call finit_module() when the 
kernel_version is >= known good, or when not using xz compression, and 
use init_module() elsewhere?

In any case, I didn't see the Debian bug report earlier.  Using 
Slackware here with custom-config mainline kernels.

Kris

Re: Unbootable system with kmod-31

[Lucas De Marchi]

On Tue, Oct 03, 2023 at 09:02:32PM -0400, Kris Karas (Bug Reporting) wrote:
>Lucas De Marchi wrote:
>>On Tue, Oct 03, 2023 at 11:08:29PM +0200, Marco d'Itri wrote:
>>>https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1050586
>>
>>duplicate of https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1050582 ?
>
>Ah, that does seem the better bug report, as it triages it down to the 
>dictionary size.
>
>>Maybe that patch needs to propagate back to 6.5?
>
>If you mean that modules_install doesn't pass the right parameters to 
>/bin/xz, then definitely.  I've reported on bugzilla variously over 
>the years about things "fixed in Linus's tree" that never got queued 
>into Greg KH's tree.

well.. it's bug reports like this that trigger an "oops, we didn't think
it would affect a stable released kernel, we really should".

>
>Perhaps kmod should be updated to only call finit_module() when the 
>kernel_version is >= known good, or when not using xz compression, and 
>use init_module() elsewhere?

checking by version is never ever correct as fixes are backported and
even features if you think about "corporate" distros. We do check by
"does the kernel support decompression and is it the same algo than what
we are expecting?" If the answer is no, then we fallback to
init_module().  But if the kernel tells us it supports something that it
doesn't really, then it's a kernel bug to be fixed, not to be
workarounded by kmod.

thanks
Lucas De Marchi

>
>In any case, I didn't see the Debian bug report earlier.  Using 
>Slackware here with custom-config mainline kernels.
>
>Kris
>