When checking the validity of the pf.conf file also include the user supplied
pf_flags. These flags might overrule macros or specify anchors, which we will
apply when actually applying the pf.conf file, so we must also take them into
account when verifying the validity.
Submitted by: Andreas Longwitz <longwitz at incore.de>
MFC after: 3 weeks
pfctl only takes the last '-F' argument into account, so this never did what
was intended.
Moreover, there is no reason to flush rules before reloading, because pf keeps
track of the rule which created a given state. That means that existing
connections will keep being processed according to the rule which originally
created them. Simply reloading the (new) rules suffices. The new rules will
apply to new connections.
PR: 127814
Submitted by: Andreas Longwitz <longwitz at incore.de>
MFC after: 3 weeks
After some tests, here are the services that run into a vnet jail:
- defaultroute
- dhclient
- ip6addrctl
- natd
- pf
- pfsync
- pflog (deamon runs, pflog0 interface usable, but /var/log/pflog not filled)
- rarpd
- route6d (do nothing anyway because obsolete)
- routed (do nothing anyway because obsolete)
- rtsold
- static_arp
- static_ndp
PR: 220530
Submitted by: olivier@freebsd.org
Check if pf.ko is loaded (i.e. /dev/pf exists) before trying to use it. This
means that '/etc/rc.d/pf status' will no longer return 'pfctl: /dev/pf: No such
file or directory' but 'pf.ko is not loaded'.
PR: 205671
Submitted by: Johannes Jost Meixner <xmj@FreeBSD.org>
assignments to the literal values it would have returned.
The concept of set_rcvar() was nice in theory, but the forks
it creates are a drag on the startup process, which is especially
noticeable on slower systems, such as embedded ones.
During the discussion on freebsd-rc@ a preference was expressed for
using ${name}_enable instead of the literal values. However the
code portability concept doesn't really apply since there are so
many other places where the literal name has to be searched for
and replaced. Also, using the literal value is also a tiny bit
faster than dereferencing the variables, and every little bit helps.
employ a more generic solution, and use it in the individual rc.d scripts
that also have an $rc_quiet test:
1. Add check_startmsgs() to rc.subr.
2. In the rc.d scripts that use rc_quiet (and rc.subr) substitute
variations of [ -z "$rc_quiet" ] with check_startmsgs
3. In savecore add a trailing '.' to the end of the message to make it
more consistent with other scripts.
4. In newsyslog remove a : before the terminal '.' since we do not expect
there to be anything printed out in between to make it more consistent.
5. In the following scripts change "quotes" to 'quotes' where no variables
exist in the message: savecore pf newsyslog
6. In the following scripts substitute if/then/fi for the simpler (and
more consistent) check_startmsgs &&: faith stf
7. In the following scripts separate the "Starting foo:" from the terminal
'.' to make them more consistent: moused hostname pf
8. In nfsclient move the message to its own line to avoid a style bug
9. In pf rc_quiet does not apply to the _stop method, so remove the
test there.
10. In motd add 'quotes' around the terminal '.' for consistency
allow them to start after netif. There were too many problems reported
with this change in the short period of time that it lived in HEAD, and
we are too late in the release cycle to properly shake it out.
IMO the issue of having the firewalls up before the network is still a
valid concern, particularly for pf whose default state is wide open.
However properly solving this issue is going to take some investment
on the part of the people who actually use those tools.
This is not a strict reversion of all the changes for r193198 since it
also included some simplification of the BEFORE/REQUIRE logic which is
still valid for ipfilter and ip6fw.
and takes over mountcritlocal's role as the early / late divider. This
makes it far easier to add rc scripts which need to run early, such as a
startup script for zfs, which is right around the corner.
This change should be a no-op; I have verified that the only change in
rcorder's output is the insertion of FILESYSTEMS immediately after
mountcritlocal.
MFC after: 3 weeks
> > There is no need to explicitly add "status" to $extra_commands in
> > the /etc/rc.d/pf script as it is implicitly added by /etc/rc.subr's
> > run_rc_command() because of the existing $pf_program.
> >
> > Submitted by: Christoph Schug <chris@schug.net>
...because as yar@ points out: "[...] you were relying on evil
side-effects of the variable being named *_program. hose side-effect
have been eliminated since rc.subr rev. 1.42. [...] The point is that
the default "status" method is for rc.d scripts that handle startup and
shutdown of conventional daemons, and not for custom tasks like the pf
case."
The change is still valid in RELENG_6 (and still doesn't have to be
backed out) as long as rc.subr:r1.42 is not MFC'ed to RELENG_6, too.
the /etc/rc.d/pf script as it is implicitly added by /etc/rc.subr's
run_rc_command() because of the existing $pf_program.
Submitted by: Christoph Schug <chris@schug.net>
MFC after: 1 week
Start before routing for better system protection.
(pf used to start late during system boot, after
many a network daemon have started already, which
sucked from security POV.)
Remark: For maximum security, pf should start before
netif, but it would create a dependency loop because
pfsync has to start after netif, yet before pf.
Discussed with: mlaier on -pf
MFC after: 5 days
