[zfs-discuss] Unlistable files - REPRODUCED

Fabian Grünbichler f.gruenbichler at proxmox.com
Wed Apr 11 03:12:07 EDT 2018


On Wed, Apr 11, 2018 at 08:30:32AM +1000, Lindsay Mathieson via zfs-discuss wrote:
> On 7/04/2018 8:54 AM, Gionatan Danti via zfs-discuss wrote:
> > I *CONFIRM* the problem, which does *NOT* happen on ZoL 0.7.6
> > At this point, I think we should discuss the (worrying) problem here:
> > https://github.com/zfsonlinux/zfs/issues/7401
> 
> Proxmox is running the 0.7.7 packages and updated their debian stretch
> kernel, they stated the following:

we actually don't use a 'Debian Stretch' kernel, but our own based on
Ubuntu's 4.13/4.15 kernel series (for PVE 5).

> > affected (only those specific packages and versions!):
> > pve-kernel-4.15.15-1-pve in version 4.15.15-5
> > pve-kernel-4.13.16-1-pve in version 4.13.16-46
> > 
> > fixed (and everything afterwards):
> > pve-kernel-4.15.15-1-pve in version 4.15.15-6
> > pve-kernel-4.13.16-2-pve in version 4.13.16-47
> > *the bug is only in the kernel module code, so the ZFS user space
> > packages have not been updated (and don't need to be in PVE's case).*
> 
> (emphasis mine) Would this be the case?

$ git show --stat 9a2e90c9fc469d377c14eb863952261f9ec12d2c                                                                                        ✔
commit 9a2e90c9fc469d377c14eb863952261f9ec12d2c
Author:     Tony Hutter <hutter2 at llnl.gov>
AuthorDate: Mon Apr 9 14:24:46 2018 -0700
Commit:     Tony Hutter <hutter2 at llnl.gov>
CommitDate: Mon Apr 9 17:29:59 2018 -0400

    Revert "Handle zap_add() failures in mixed ... "

    This reverts commit cc63068e95ee725cce03b1b7ce50179825a6cda5.

    Under certain circumstances this change can result in an ENOSPC
    error when adding new files to a directory.  See #7401 for full
    details.

    Reviewed-by: Brian Behlendorf <behlendorf1 at llnl.gov>
    Signed-off-by: Tony Hutter <hutter2 at llnl.gov>
    Issue #7401
    Closes #7416

 tests/zfs-tests/tests/functional/casenorm/Makefile.am              |   1 -
 include/sys/zap_leaf.h                                             |  15 ++++-----------
 module/zfs/zap.c                                                   |  25 +------------------------
 module/zfs/zap_leaf.c                                              |   2 +-
 module/zfs/zap_micro.c                                             |  38 +-------------------------------------
 module/zfs/zfs_dir.c                                               |  29 ++++++-----------------------
 module/zfs/zfs_vnops.c                                             |  73 ++++++++++++++++++-------------------------------------------------------
 tests/runfiles/linux.run                                           |   2 +-
 tests/zfs-tests/tests/functional/casenorm/mixed_create_failure.ksh | 136 ----------------------------------------------------------------------------------------------------------------------------------------
 9 files changed, 32 insertions(+), 289 deletions(-)

the modified header file is only included in the zfs.ko module code
(zap.c, zap_leaf.c, zap_micro.c to be specific). we don't ship any ZFS
module source or DKMS package which would otherwise contain a copy of
the broken code. the only thing in our user space packages that is
affected by the revert is the zfs-test package, which is not installed
by default (and will simply fail one test case until the next regular
upgrade).

we did our revert Monday morning and pushed out fixed kernel packages to
the public Monday afternoon (both CEST, which is GMT+2 atm).

this was around 13 resp. 7 hours before 0.7.8 was available, so the
version numbering is a bit confusing (0.7.7-pve2 in PVE is essentially
upstream's 0.7.8, modulo the actual version bump).

our reaction also happened before this issue blew up media-wise, and
since we only ever had the affected 0.7.7 in testing packages, the
changelog is a bit non-descriptive in retrospective.

all of this would be more appropriate on the pve-users list IMHO, so if
you want to follow up with PVE specifics, please do so in one of the
threads there ;)



More information about the zfs-discuss mailing list