Clamd fails on reboot

[Arthur Dent]

Hello All,

When i do a reboot (about once per month) I carefully check all the
essential services are running (I especially needed to do this as
systemd was, until now, not starting sendmail - a bug that is now
fixed).

Clamd will not start after a reboot. A little investigation reveals that
it won't start because it tries to create a pid file
in /var/run/clamd.clamd/ - but /var/run/clamd.clamd/ does not exist.

mkdir /var/run/clamd.clamd && chown clamd:clamd /var/run/clamd.clamd/
will allow the service to be started with no problems.

The next reboot and... poof... /var/run/clamd.clamd/ is gone again!

I have grepped the whole of /etc/* to see if there are any references to
this directory with the only entries being the sensible ones
in /etc/init.d/clamd-wrapper.

In that init script the "stop" stanza does have the following command:
    "[ $RETVAL -eq 0 ] && rm -f $lockfile"
but "$lockfile" refers to
lockfile=/var/lock/subsys/clamd.${CLAMD_SERVICE}

There are no other "rm" commands.

So what deletes the /var/run/clamd.clamd/ directory on reboot?

Thanks in advance

Mark

[Genes MailLists]

On 07/20/2011 06:17 AM, Arthur Dent wrote:
> There are no other "rm" commands.
>
> So what deletes the /var/run/clamd.clamd/ directory on reboot?

I wonder if you should look for what should be creating it instead -
because of the fact that /run = /var/run is now tmpfs.

gene

[Arthur Dent]

On Wed, 2011-07-20 at 07:41 -0400, Genes MailLists wrote:
> On 07/20/2011 06:17 AM, Arthur Dent wrote:
> > There are no other "rm" commands.
> >
> > So what deletes the /var/run/clamd.clamd/ directory on reboot?
>
> I wonder if you should look for what should be creating it instead -
> because of the fact that /run = /var/run is now tmpfs.
>
> gene

Good point.

Well I attach the whole of the init script to this email in case anyone
can see any problems - but the key parts are:

CLAMD_PIDFILE=/var/run/clamd.${CLAMD_SERVICE}/clamd.pid

and

start () {
    echo -n $"Starting $prog: "
    daemon --pidfile=${CLAMD_PIDFILE} \
        exec -a $procname /usr/sbin/clamd \
        ${CLAMD_CONFIGFILE:+-c $CLAMD_CONFIGFILE} ${CLAMD_OPTIONS} --pid ${CLAMD_PIDFILE}
    RETVAL=$?
    echo
    [ $RETVAL -eq 0 ] && touch $lockfile
    return $RETVAL
}

Should I put a "touch $PIDFILE" in there? Or would I also need a mkdir
command?

This is a Fedora yum package. Shouldn't this just work?

Thanks for your help so far...

Mark

[Genes MailLists]

On 07/20/2011 07:59 AM, Arthur Dent wrote:

...
> Should I put a "touch $PIDFILE" in there? Or would I also need a mkdir
> command?
>
> This is a Fedora yum package. Shouldn't this just work?

1) I think clamd should be doing the mkdir .. not the start script ..
but ..

2) For now I'd put (you sure about the dir having .{${CLAMD_SERVICE}
} on the end?

piddir =  /var/run/clamd.${CLAMD_SERVICE} ## fix if needed

if [ !-d $piddir ] ; then
         mkdir piddir}

fi

in the file ...

3) i'd file a bug report against clamav on this - needing to be
updated for /var/run being tmpfs.

gene/

[Arthur Dent]

* On Wed, 2011-07-20 at 08:56 -0400, Genes MailLists wrote:
> On 07/20/2011 07:59 AM, Arthur Dent wrote:
>
> ...
> > Should I put a "touch $PIDFILE" in there? Or would I also need a mkdir
> > command?
> >
> > This is a Fedora yum package. Shouldn't this just work?
> I think clamd should be doing the mkdir .. not the start script ..
>
> but ..
> For now I'd put (you sure about the dir having .{${CLAMD_SERVICE}
>
> } on the end?
>  piddir =  /var/run/clamd.${CLAMD_SERVICE} ## fix if needed
>
>  if [ !-d $piddir ] ; then
>      mkdir piddir}
>
>  fi
>
> in the file ...
> i'd file a bug report against clamav on this - needing to be
>
> updated for /var/run being tmpfs.
>
> gene/

Thanks Gene.

https://bugzilla.redhat.com/show_bug.cgi?id=723544

Also, I found that in order for your suggested modification to work I
needed to put a space between the ! and the -d and also I needed to
chown the new directory.

For the benefit of the archives this is what now works for me:

#============8<===================================================
# In the definitions section:
CLAMD_PIDDIR=/var/run/clamd.${CLAMD_SERVICE}

# and in the "start" stanza:

if [ ! -d $CLAMD_PIDDIR ] ; then
         mkdir $CLAMD_PIDDIR && chown clamd:clamd $CLAMD_PIDDIR
      fi
#============8<===================================================

Thanks again for your help Gene. Much appreciated.

Mark

[Genes MailLists]

On 07/20/2011 11:11 AM, Arthur Dent wrote:
> Also, I found that in order for your suggested modification to work I
> needed to put a space between the ! and the -d and also I needed to
> chown the new directory.

...
> Thanks again for your help Gene. Much appreciated.

Hah well glad you got it - I should have said - "untested" :-)

[Arthur Dent]

On Wed, 2011-07-20 at 16:11 +0100, Arthur Dent wrote:
> For the benefit of the archives this is what now works for me:
>
> #============8<===================================================
> # In the definitions section:
> CLAMD_PIDDIR=/var/run/clamd.${CLAMD_SERVICE}
>
> # and in the "start" stanza:
>
> if [ ! -d $CLAMD_PIDDIR ] ; then
>          mkdir $CLAMD_PIDDIR && chown clamd:clamd $CLAMD_PIDDIR
>       fi
> #============8<===================================================

For those who aren't following the bug
https://bugzilla.redhat.com/show_bug.cgi?id=723544 Apparently the
correct solution to the problem of /var/run now being tempfs (and
therefore wiped clean with each reboot) is to create a /etc/tmpfiles.d
file with instructions on how to create the directory with the
appropriate permissions. This simply involves (for me) creating
a /etc/tmpfiles.d/clamd.conf file with the following line in it:

d /var/run/clamd.clamd 0700 clamd clamd -

At reboot the directory /var/run/clamd.clamd/ is then created with 0700
permissions and clamd:clamd ownership. The clamd package will then start
properly.

I hope that helps anyone else suffering the same problem.

Mark