Re: glibc VETO for kernel version SUBLEVEL >= 255

[Salvatore Bonaccorso <carnil@debian.org>]

Hi,

On Sun, Sep 26, 2021 at 09:28:58AM +0200, Greg Kroah-Hartman wrote:
> On Sun, Sep 26, 2021 at 07:23:33AM +0000, Jari Ruusu wrote:
> > Earlier this year there was some discussion about kernel version numbers
> > after 4.9.255 and 4.4.255. Problem was 8-bit limitation for SUBLEVEL
> > number in stable kernel versions. The fix was to freeze LINUX_VERSION_CODE
> > number at x.x.255 and to continue incrementing SUBLEVEL number. Seems
> > there are more more fallout from that decision. At least some versions of
> > glibc do not play well with larger SUBLEVEL numbers.
> > 
> > 
> > # uname -s -r -m
> > Linux 4.9.283-QEMU armv6l
> > # apt upgrade
> > Reading package lists... Done
> > Building dependency tree
> > Reading state information... Done
> > Calculating upgrade... Done
> > The following packages will be upgraded:
> >  [SNIP]
> > Fetched 145 MB in 1min 57s (1244 kB/s)
> > Reading changelogs... Done
> > Preconfiguring packages ...
> > (Reading database ... 39028 files and directories currently installed.)
> > Preparing to unpack .../libc6-dbg_2.28-10+rpt2+rpi1_armhf.deb ...
> > Unpacking libc6-dbg:armhf (2.28-10+rpt2+rpi1) over (2.28-10+rpi1) ...
> > Preparing to unpack .../libc6-dev_2.28-10+rpt2+rpi1_armhf.deb ...
> > Unpacking libc6-dev:armhf (2.28-10+rpt2+rpi1) over (2.28-10+rpi1) ...
> > Preparing to unpack .../libc-dev-bin_2.28-10+rpt2+rpi1_armhf.deb ...
> > Unpacking libc-dev-bin (2.28-10+rpt2+rpi1) over (2.28-10+rpi1) ...
> > Preparing to unpack .../linux-libc-dev_1%3a1.20210831-3~buster_armhf.deb ...
> > Unpacking linux-libc-dev:armhf (1:1.20210831-3~buster) over (1:1.20210527-1) ...
> > Preparing to unpack .../libc6_2.28-10+rpt2+rpi1_armhf.deb ...
> > ERROR: Your kernel version indicates a revision number
> > of 255 or greater.  Glibc has a number of built in
> > assumptions that this revision number is less than 255.
> > If you\'ve built your own kernel, please make sure that any
> > custom version numbers are appended to the upstream
> > kernel number with a dash or some other delimiter.
> > 
> > dpkg: error processing archive /var/cache/apt/archives/libc6_2.28-10+rpt2+rpi1_armhf.deb (--unpack):
> >  new libc6:armhf package pre-installation script subprocess returned error exit status 1
> > Errors were encountered while processing:
> >  /var/cache/apt/archives/libc6_2.28-10+rpt2+rpi1_armhf.deb
> > E: Sub-process /usr/bin/dpkg returned an error code (1)
> > 
> > 
> > 
> > Above upgrade works normally if I edit top level Linux source Makefile to
> > say "SUBLEVEL = 0" and re-compile new kernel.
> > 
> > I am not pointing any fingers here, but it seems that either glibc code or
> > stable kernel versioning is messed up.
> 
> Are you sure this isn't just a warning coming from a script that apt is
> running when trying to install glibc?  Or is this from the glibc package
> itself?
> 
> And what exactly is it testing?  We fixed the build time detection of
> the kernel version here, so you should be able to build glibc properly.
> 
> This is the first time we've seen this reported, are people using the
> newer kernels on systems that are not using glibc?

They are probably not using a problematic combination or a
distribution kernel on those systems. Looking from the mentioned
versions above this looks like a version derived from Debian buster.

Recently prompted due to https://bugs.debian.org/987266 the check was
removed in the postinst script of libc in Debian:
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=987266 .

Regards,
Salvatore