Bug 248116

Perhaps we should start a survey on Fedora / Red Hat server system administrator
 to run the lsof part of rpm-check to see, how many library & application
updates are still not used because services aren't restarted.

BTW: a harder part would be implementing the same for Perl modules (think about
the last perl-Net-DNS update...it is not used, if e.g. postgrey or amavisd
aren't restarted). Unfortunately, lsof doesn't help here...is there a Perl
helper program available which can analyse running Perl program what they are
currently using?

BTW2: current lsof check does not use option "-n", this can lead to hanging
while DNS resolving, will be fixed next.


+++ This bug was initially created as a clone of Bug #247251 +++

Description of problem:
From security point of view it makes not much sense to update supplied libraries
all the time without restarting the programs which uses this libraries, because
still the old library code would be used and e.g. a network service can be still
exploitable. Reboot after each library update is overkill.

Version-Release number of selected component (if applicable):
all current version

How reproducible:
After each library package update

Steps to Reproduce:
Update e.g. on a systems krb5-libs
Take a look on open but deleted files

Actual results:
You will see that programs (e.g. daemons listening on network sockets) still
using the old but deleted library.

Expected results:
rpm at least suggest to restart related programs.

Additional info:
I've created now a cron job, which does this check on regular basis all days and
suggest me a list of services to restart or to reboot the system, if nothing
else would help.

Take this solution as start for discussion, how to solve this issue completly.

For middle term, perhaps every library package can get an additional program
call which (if exists) run the open-but-deleted-file check and automagically
suggests the services which should be restarted.
In case of using yum or up2date, this should only run once after the upgrade, so
the additional program has to detect, whether it was called by postinstall from
standalone rpm usage or triggered via yum|up2date->rpm->postinstall


I've created a rpm package for that and additional checks:
ftp://ftp.aerasec.de/pub/linux/repository/public/redhat/enterprise/4ES/SRPMS/rpm-check-0.3.1-1.RHEL4.AERAsec.1.src.rpm
ftp://ftp.aerasec.de/pub/linux/repository/public/redhat/enterprise/4ES/i386/rpm-check-0.3.1-1.RHEL4.AERAsec.1.noarch.rpm

cron.daily/rpm-check-updates.cron:
Checks for available packages, if automatic update is not active, search for
open but deleted files and check latest installed against running kernel version.

cron.weekly/rpm-list-extras.cron
Checks for packages which are not covered by yum or up2date channels - so
administrator have to look for updates manually here

-- Additional comment from pb on 2007-07-06 07:32 EST --
Created an attachment (id=158653)
Daily check of open files, et.al.


-- Additional comment from n3npq on 2007-07-07 19:38 EST --
Restarting running processes to insure that libraries with security updates
are used by persistent processes is outside the scope of package management.

Your goal can be accomplisehed without any assistance from rpm applications, as
you have shown.

-- Additional comment from pb on 2007-07-08 04:24 EST --
But the current problem is, that the "check" process needs to be started by rpm,
up2date or yum at the end of each run, so a hook or trigger would be needed - or
an optional wrapper.

Currently, my check is started by cron each day, this can be too late in
important security cases.