June 1, 2026 at 5:37 AM, "Borja Tarraso" <[email protected] 
mailto:[email protected]?to=%22Borja%20Tarraso%22%20%3Cborja.tarraso%40member.fsf.org%3E
 > wrote:



> 
> Hi,
> 
> On canonical Debian GNU/Hurd 0.9, an apt-get install of around
> 14 MB of packages onto the underlying ext2fs /var reproducibly
> trips an ext2fs translator assertion within a minute or two and
> wedges the rootfs.

Perhaps try again with the "ext3/4fs" that just landed, and see if
the bug is still there...just a thought.

> Reproducer:
> 
>  1. Boot a canonical Debian GNU/Hurd 0.9 image (gnumach
>  2:1.8+git20260224-8, hurd 1:0.9.git20251029-7+b1). /var is
>  mounted via /hurd/tmpfs (size 256M) by default and masks the
>  underlying ext2fs /var on the rootfs.
> 
>  2. Drop the tmpfs to expose the underlying /var:
> 
>  settrans -fg /var
> 
>  3. apt-install a multi-package set, for example:
> 
>  apt-get install -y --no-install-recommends \
>  xvfb emacs-lucid elpa-exwm elpa-xelb pulseaudio
> 
>  dpkg unpacks a few packages and then the kernel log emits:
> 
>  ext2fs: ../../ext2fs/pager.c:455:
>  file_pager_write_pages: Assertion 'blk' failed.
> 
>  The translator does not recover; subsequent writes to /
>  hang and the VM has to be reset.
> 
> The same install runs to completion against the /hurd/tmpfs
> overlay (nothing persists in that case, of course). The rootfs
> is well under any size limit (5 GB qcow2 with around 3.6 GB
> free per df before the install). Reading pager.c around line
> 455 the failing condition looks like an indirect-block lookup
> returning zero for a block the pager is about to write out,
> which on the face of it reads like a metadata coherency issue
> between the inode block map and the pager's view of the dirty
> page; I am not an ext2fs maintainer, so that reading may be
> wrong.
> 
> I would appreciate:
> 
>  - Confirmation that this reproduces, or a pointer if it is
>  a known bug already in flight.
>  - Guidance on whether the root cause sits in ext2fs's pager,
>  in inode block-map maintenance under the apt-install
>  workload, or upstream of both.
>  - If a patch exists or is planned, which release it lands in.
> 
> Happy to test patches in a qemu loop, run the workload against
> a debug build of ext2fs with extra logging, or send the full
> serial trace if any of that helps.
> 
> Alternative filing channel if a tracker entry is preferred:
> https://savannah.gnu.org/bugs/?group=hurd
> 
> Thanks,
> Borja Tarraso
> [email protected] mailto:[email protected]
>

Reply via email to