freebsd-dev/share/man/man4/pfsync.4
Kristof Provost 4fc65bcbe3 pfsync: Performance improvement
pfsync code is called for every new state, state update and state
deletion in pf. While pf itself can operate on multiple states at the
same time (on different cores, assuming the states hash to a different
hashrow), pfsync only had a single lock.
This greatly reduced throughput on multicore systems.

Address this by splitting the pfsync queues into buckets, based on the
state id. This ensures that updates for a given connection always end up
in the same bucket, which allows pfsync to still collapse multiple
updates into one, while allowing multiple cores to proceed at the same
time.

The number of buckets is tunable, but defaults to 2 x number of cpus.
Benchmarking has shown improvement, depending on hardware and setup, from ~30%
to ~100%.

MFC after:	1 week
Sponsored by:	Orange Business Services
Differential Revision:	https://reviews.freebsd.org/D18373
2018-12-06 19:27:15 +00:00

236 lines
6.7 KiB
Groff

.\" $OpenBSD: pfsync.4,v 1.28 2009/02/17 10:05:18 dlg Exp $
.\"
.\" Copyright (c) 2002 Michael Shalayeff
.\" Copyright (c) 2003-2004 Ryan McBride
.\" All rights reserved.
.\"
.\" Redistribution and use in source and binary forms, with or without
.\" modification, are permitted provided that the following conditions
.\" are met:
.\" 1. Redistributions of source code must retain the above copyright
.\" notice, this list of conditions and the following disclaimer.
.\" 2. Redistributions in binary form must reproduce the above copyright
.\" notice, this list of conditions and the following disclaimer in the
.\" documentation and/or other materials provided with the distribution.
.\"
.\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
.\" IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
.\" OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
.\" IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
.\" INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
.\" NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF MIND,
.\" USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
.\" THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
.\" (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
.\" THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
.\"
.\" $FreeBSD$
.\"
.Dd December 6, 2018
.Dt PFSYNC 4
.Os
.Sh NAME
.Nm pfsync
.Nd packet filter state table sychronisation interface
.Sh SYNOPSIS
.Cd "device pfsync"
.Sh DESCRIPTION
The
.Nm
interface is a pseudo-device which exposes certain changes to the state
table used by
.Xr pf 4 .
State changes can be viewed by invoking
.Xr tcpdump 1
on the
.Nm
interface.
If configured with a physical synchronisation interface,
.Nm
will also send state changes out on that interface,
and insert state changes received on that interface from other systems
into the state table.
.Pp
By default, all local changes to the state table are exposed via
.Nm .
State changes from packets received by
.Nm
over the network are not rebroadcast.
Updates to states created by a rule marked with the
.Ar no-sync
keyword are ignored by the
.Nm
interface (see
.Xr pf.conf 5
for details).
.Pp
The
.Nm
interface will attempt to collapse multiple state updates into a single
packet where possible.
The maximum number of times a single state can be updated before a
.Nm
packet will be sent out is controlled by the
.Ar maxupd
parameter to ifconfig
(see
.Xr ifconfig 8
and the example below for more details).
The sending out of a
.Nm
packet will be delayed by a maximum of one second.
.Sh NETWORK SYNCHRONISATION
States can be synchronised between two or more firewalls using this
interface, by specifying a synchronisation interface using
.Xr ifconfig 8 .
For example, the following command sets fxp0 as the synchronisation
interface:
.Bd -literal -offset indent
# ifconfig pfsync0 syncdev fxp0
.Ed
.Pp
By default, state change messages are sent out on the synchronisation
interface using IP multicast packets to the 224.0.0.240 group address.
An alternative destination address for
.Nm
packets can be specified using the
.Ic syncpeer
keyword.
This can be used in combination with
.Xr ipsec 4
to protect the synchronisation traffic.
In such a configuration, the syncdev should be set to the
.Xr enc 4
interface, as this is where the traffic arrives when it is decapsulated,
e.g.:
.Bd -literal -offset indent
# ifconfig pfsync0 syncpeer 10.0.0.2 syncdev enc0
.Ed
.Pp
It is important that the pfsync traffic be well secured
as there is no authentication on the protocol and it would
be trivial to spoof packets which create states, bypassing the pf ruleset.
Either run the pfsync protocol on a trusted network \- ideally a network
dedicated to pfsync messages such as a crossover cable between two firewalls,
or specify a peer address and protect the traffic with
.Xr ipsec 4 .
.Pp
.Nm
has the following
.Xr sysctl 8
tunables:
.Bl -tag -width ".Va net.pfsync"
.It Va net.pfsync.carp_demotion_factor
Value added to
.Va net.inet.carp.demotion
while
.Nm
tries to perform its bulk update.
See
.Xr carp 4
for more information.
Default value is 240.
.It Va net.pfsync.pfsync_buckets
The number of
.Nm
buckets.
This affects the performance and memory tradeoff.
Defaults to twice the number of CPUs.
Change only if benchmarks show this helps on your workload.
.El
.Sh EXAMPLES
.Nm
and
.Xr carp 4
can be used together to provide automatic failover of a pair of firewalls
configured in parallel.
One firewall will handle all traffic until it dies, is shut down, or is
manually demoted, at which point the second firewall will take over
automatically.
.Pp
Both firewalls in this example have three
.Xr sis 4
interfaces.
sis0 is the external interface, on the 10.0.0.0/24 subnet; sis1 is the
internal interface, on the 192.168.0.0/24 subnet; and sis2 is the
.Nm
interface, using the 192.168.254.0/24 subnet.
A crossover cable connects the two firewalls via their sis2 interfaces.
On all three interfaces, firewall A uses the .254 address, while firewall B
uses .253.
The interfaces are configured as follows (firewall A unless otherwise
indicated):
.Pp
Interfaces configuration in
.Pa /etc/rc.conf :
.Bd -literal -offset indent
network_interfaces="lo0 sis0 sis1 sis2"
ifconfig_sis0="10.0.0.254/24"
ifconfig_sis0_alias0="inet 10.0.0.1/24 vhid 1 pass foo"
ifconfig_sis1="192.168.0.254/24"
ifconfig_sis1_alias0="inet 192.168.0.1/24 vhid 2 pass bar"
ifconfig_sis2="192.168.254.254/24"
pfsync_enable="YES"
pfsync_syncdev="sis2"
.Ed
.Pp
.Xr pf 4
must also be configured to allow
.Nm
and
.Xr carp 4
traffic through.
The following should be added to the top of
.Pa /etc/pf.conf :
.Bd -literal -offset indent
pass quick on { sis2 } proto pfsync keep state (no-sync)
pass on { sis0 sis1 } proto carp keep state (no-sync)
.Ed
.Pp
It is preferable that one firewall handle the forwarding of all the traffic,
therefore the
.Ar advskew
on the backup firewall's
.Xr carp 4
vhids should be set to something higher than
the primary's.
For example, if firewall B is the backup, its
carp1 configuration would look like this:
.Bd -literal -offset indent
ifconfig_sis1_alias0="inet 192.168.0.1/24 vhid 2 pass bar advskew 100"
.Ed
.Pp
The following must also be added to
.Pa /etc/sysctl.conf :
.Bd -literal -offset indent
net.inet.carp.preempt=1
.Ed
.Sh SEE ALSO
.Xr tcpdump 1 ,
.Xr bpf 4 ,
.Xr carp 4 ,
.Xr enc 4 ,
.Xr inet 4 ,
.Xr inet6 4 ,
.Xr ipsec 4 ,
.Xr netintro 4 ,
.Xr pf 4 ,
.Xr pf.conf 5 ,
.Xr protocols 5 ,
.Xr rc.conf 5 ,
.Xr ifconfig 8
.Sh HISTORY
The
.Nm
device first appeared in
.Ox 3.3 .
It was first imported to
.Fx 5.3 .
.Pp
The
.Nm
protocol and kernel implementation were significantly modified in
.Fx 9.0 .
The newer protocol is not compatible with older one and will not interoperate
with it.