[OpenAFS] Possible cache corruption in Debian 10 with openafs 1.8.6

Jose M Calhariz jose.calhariz@tecnico.ulisboa.pt
Tue, 27 Oct 2020 18:34:42 +0000

Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable


I seeking for help on how to collect more information for a bug report
on openafs client 1.8.6.

I manage several openafs clients on 1.6.x and 1.8.x and servers on
1.6.x.  One of this clients, this morning, stop serving one very
important file by apache, the apache error was:

message=3DAH01630: client denied by server configuration: /afs/

istXXXYZ is a username of a user that I need to protect the privacy.

By login into my openafs client reference machine everything was OK.
But login into the apache server machine I found some strange things:

www-data@apache-server:~$ ls -alF /afs/ist.utl.pt/users/Y/X/istXXXYZ/web
ls: cannot access '/afs/ist.utl.pt/users/Y/Z/istXXXYZ/web': No such file or=
www-data@apache-server:~$ ls -alF /afs/ist.utl.pt/users/Y/Z/istXXXYZ/
ls: cannot access '/afs/ist.utl.pt/users/Y/Z/istXXXYZ/.mysql_history': Perm=
ission denied
ls: cannot access '/afs/ist.utl.pt/users/Y/Z/istXXXYZ/web': No such file or=
ls: cannot access '/afs/ist.utl.pt/users/Y/Z/istXXXYZ/.bash_history': Permi=
ssion denied

For some files, OK for others files.

The www-data user had a good openafs token, no reported problems in the last
"kinit -k -t keytab www-data ; aklog"

By looking into dmesg output I see no errors from openafs.

So as root I have done "systemctl restart openafs-client.service" and
as www-data I have done "kinit -k -t keytab www-data ; aklog" and
everything was OK again.

Some more notes:

The file in error is requested very frequently by an HAProxy to check
if the server is alive.

The openafs-client in error runs Debian 10.5, the openafs-client is
1.8.6 is a private backport of the latest openafs in future Debian 11
made by myself.

The Debian changelog of the package is:

openafs (1.8.6-2~dsi10+2) buster; urgency=3Dmedium

  * According to bugreport #969376 "Openafs cache erros on the logs",
    applied to patchs from gerrit.openafs.org:
    * Linux: use override_creds when available
    * LINUX: Properly revert creds in osi_UFSTruncate

 -- Jose M Calhariz <jose.calhariz@tecnico.ulisboa.pt>  Fri, 04 Sep 2020 13=
:42:52 +0100

openafs (1.8.6-2~dsi10+1) buster; urgency=3Dmedium

  * Rebuild for buster.

 -- Jose M Calhariz <jose.calhariz@tecnico.ulisboa.pt>  Fri, 04 Sep 2020 11=
:34:14 +0100

openafs (1.8.6-2) unstable; urgency=3Dmedium

  * Pull in upstream patches for -fcommon support (Closes: #966881)

 -- Benjamin Kaduk <kaduk@mit.edu>  Sun, 30 Aug 2020 19:39:18 -0700

Some info about openafs client:

OpenAFS client
@(#)OpenAFS 1.8.6-2~dsi10+2-debian 2020-09-04
@(#)OpenAFS 1.8.6-2~dsi10+2-debian 2020-09-04
OpenAFS modules
Trying (port 7001):
AFS version: OpenAFS 1.8.6-2~dsi10+2-debian 2020-09-04 root@apache-server
OpenAFS cache settings
Chunk files:   245760
Stat caches:   15000
Data caches:   10000
Volume caches: 200
Chunk size:    1048576
Cache size:    7864320 kB
Set time:      no
Cache type:    disk

When this bug happens again, what more information I should collect to
do a proper bug report.

Kind regards
Jose M Calhariz

	Os miseraveis n=E3o tem outro remedio a n=E3o ser a esperanca
		--  William Shakespeare

Content-Type: application/pgp-signature; name="signature.asc"