Bug 2092033

The option `-s` only reports affected systemd services, but doesn't report whether a reboot is required. What does `dnf needs-restarting -r` report?

I should have included that at the start.

~~~~~~~~~~~~~~~~
[root@r8 ~]# needs-restarting -r
Updating Subscription Management repositories.
No core libraries or services have been updated since boot-up.
Reboot should not be necessary.

[root@r8 ~]# needs-restarting -s
Updating Subscription Management repositories.
rsyslog.service
sshd.service
pmlogger.service
gssproxy.service
rpcbind.service
firewalld.service
rngd.service
auditd.service
sssd.service
systemd-logind.service
NetworkManager.service
pmcd.service
systemd-journald.service
pmie_farm.service
pmlogger_farm.service
systemd-udevd.service
pmie.service
~~~~~~~~~~~~~~~~

The reboot will only show when certain packages are updated.

REBOOTPKGS = ['kernel', 'kernel-rt', 'glibc', 'linux-firmware', 'systemd', 'udev', 'openssl-libs', 'gnutls', 'dbus']

In this case, zlib is triggering a need to restart dbus, but not saying it needs to reboot.

Reproduces (using expat instead ov zlib, though - unlike zlib, expat is direct dependency of dbus):

# dnf repoquery --requires dbus-daemon
Updating Subscription Management repositories.
Last metadata expiration check: 0:17:16 ago on Mon Jun 13 08:18:55 2022.
--------------------- 8< ----------------------
libexpat.so.1()(64bit)
--------------------- 8< ----------------------


# dnf downgrade expat
--------------------- 8< ----------------------
Downgraded:
  expat-2.2.5-4.el8_5.3.x86_64

# dnf needs-restarting -s
Updating Subscription Management repositories.
polkit.service
tuned.service
dbus.service

# dnf needs-restarting -r
Updating Subscription Management repositories.
No core libraries or services have been updated since boot-up.
Reboot should not be necessary.

# echo $?
0

The binary package is built from dnf-plugins-core. Moving a component there.

I think there is a flaw in a logic for the "-r" option in needs_restarting.py:

    def run(self):
        process_start = ProcessStart()
        owning_pkg_fn = functools.partial(owning_package, self.base.sack)
        owning_pkg_fn = memoize(owning_pkg_fn)

        opt = get_options_from_dir(os.path.join(
            self.base.conf.installroot,
            "etc/dnf/plugins/needs-restarting.d/"),
            self.base)
        NEED_REBOOT.extend(opt)
        if self.opts.reboothint:
            need_reboot = set()
            installed = self.base.sack.query().installed()
            for pkg in installed.filter(name=NEED_REBOOT):
→               if pkg.installtime > process_start.boot_time:
                    need_reboot.add(pkg.name)
            if need_reboot:
                print(_('Core libraries or services have been updated '
                        'since boot-up:'))
                for name in sorted(need_reboot):
                    print('  * %s' % name)
                print()
                print(_('Reboot is required to fully utilize these updates.'))
                print(_('More information:'),
                      'https://access.redhat.com/solutions/27943')
                raise dnf.exceptions.Error()  # Sets exit code 1
            else:
                print(_('No core libraries or services have been updated '
                        'since boot-up.'))

The current code ends up with "No core libraries or services have been updated since boot-up." message because all it does is checking an installation time of the very few packages listed in NEED_REBOOT array against uptime of the system.

In this case a library used by dbus program was updated. But that library is not part of dbus RPM package. It's comes from a different package (e.g. zlib) which is not on the list.

A proper fix needs to enumerate affected processes, locate their programs, map them to binary RPM package, consult the packages against the NEED_REBOOT array.

needs-restaring without options already locates the program ("needs-restarting | grep dbus"). All you need is query a sack for a package providing the program file and then match NEED_REBOOT without checking any installation times.

Now I realize that replacing the original approach would introduce regressions like "I upgraded linux-firmware and it does report to reboot because on process holds open firmware file". I think we need to combine both approaches. Checking for open files and checking for installation time, do union on both sets.

Hello, unfortunately, we are postponing this fix to 8.9. The risk of this bugfix having collateral side effects is high. Therefore, we would like to have more time to test it in Fedora.