[sliver-openvswitch.git] / utilities / ovs-openflowd.8.in

.TH ovs\-openflowd 8 "March 2009" "Open vSwitch" "Open vSwitch Manual"
.ds PN ovs\-openflowd

.SH NAME
ovs\-openflowd \- OpenFlow switch implementation

.SH SYNOPSIS
.B ovs\-openflowd
[\fIoptions\fR] \fIdatapath\fR [\fIcontroller\fR]

.SH DESCRIPTION
The \fBovs\-openflowd\fR program implements an OpenFlow switch using a
flow-based datapath.  \fBovs\-openflowd\fR connects to an OpenFlow controller
over TCP or SSL.

The mandatory \fIdatapath\fR argument argument specifies the local datapath
to relay.  It takes one of the following forms:

.so lib/dpif.man

.PP
The optional \fIcontroller\fR argument specifies how to connect to
the OpenFlow controller.  It takes one of the following forms:

.RS
.IP "\fBssl:\fIip\fR[\fB:\fIport\fR]"
The specified SSL \fIport\fR (default: 6633) on the host at the given
\fIip\fR, which must be expressed as an IP address (not a DNS name).
The \fB--private-key\fR, \fB--certificate\fR, and \fB--ca-cert\fR
options are mandatory when this form is used.

.IP "\fBtcp:\fIip\fR[\fB:\fIport\fR]"
The specified TCP \fIport\fR (default: 6633) on the host at the given
\fIip\fR, which must be expressed as an IP address (not a DNS name).

.TP
\fBunix:\fIfile\fR
The Unix domain server socket named \fIfile\fR.
.RE

.PP
If \fIcontroller\fR is omitted, \fBovs\-openflowd\fR attempts to discover the
location of the controller automatically (see below).

.SS "Contacting the Controller"
The OpenFlow switch must be able to contact the OpenFlow controller
over the network.  It can do so in one of two ways:

.IP out-of-band
In this configuration, OpenFlow traffic uses a network separate from
the data traffic that it controls, that is, the switch does not use
any of the network devices added to the datapath with \fBovs\-dpctl
add\-if\fR in its communication with the controller.

To use \fBovs\-openflowd\fR in a network with out-of-band control, specify
\fB--out-of-band\fR on the \fBovs\-openflowd\fR command line.  The control
network must be configured separately, before or after \fBovs\-openflowd\fR
is started.

.IP in-band
In this configuration, a single network is used for OpenFlow traffic
and other data traffic, that is, the switch contacts the controller
over one of the network devices added to the datapath with \fBovs\-dpctl
add\-if\fR.  This configuration is often more convenient than
out-of-band control, because it is not necessary to maintain two
independent networks.

In-band control is the default for \fBovs\-openflowd\fR, so no special
command-line option is required.

With in-band control, the location of the controller can be configured
manually or discovered automatically:

.RS
.IP "controller discovery"
To make \fBovs\-openflowd\fR discover the location of the controller
automatically, do not specify the location of the controller on the
\fBovs\-openflowd\fR command line.

In this mode, \fBovs\-openflowd\fR will broadcast a DHCP request with vendor
class identifier \fBOpenFlow\fR across the network devices added to
the datapath with \fBovs\-dpctl add\-if\fR.  It will accept any valid DHCP
reply that has the same vendor class identifier and includes a
vendor-specific option with code 1 whose contents are a string
specifying the location of the controller in the same format used on
the \fBovs\-openflowd\fR command line (e.g. \fBssl:192.168.0.1\fR).

The DHCP reply may also, optionally, include a vendor-specific option
with code 2 whose contents are a string specifying the URI to the base
of the OpenFlow PKI (e.g. \fBhttp://192.168.0.1/openflow/pki\fR).
This URI is used only for bootstrapping the OpenFlow PKI at initial
switch setup; \fBovs\-openflowd\fR does not use it at all.

The following ISC DHCP server configuration file assigns the IP
address range 192.168.0.20 through 192.168.0.30 to OpenFlow switches
that follow the switch protocol and addresses 192.168.0.1 through
192.168.0.10 to all other DHCP clients:

default-lease-time 600;
.br
max-lease-time 7200;
.br
option space openflow;
.br
option openflow.controller-vconn code 1 = text;
.br
option openflow.pki-uri code 2 = text;
.br
class "OpenFlow" {
.br
  match if option vendor-class-identifier = "OpenFlow";
.br
  vendor-option-space openflow;
.br
  option openflow.controller-vconn "tcp:192.168.0.10";
.br
  option openflow.pki-uri "http://192.168.0.10/openflow/pki";
.br
  option vendor-class-identifier "OpenFlow";
.br
}
.br
subnet 192.168.0.0 netmask 255.255.255.0 {
.br
    pool {
.br
        allow members of "OpenFlow";
.br
        range 192.168.0.20 192.168.0.30;
.br
    }
.br
    pool {
.br
        deny members of "OpenFlow";
.br
        range 192.168.0.1 192.168.0.10;
.br
    }
.br
}
.br

.IP "manual configuration"
To configure in-band control manually, specify the location of the
controller on the \fBovs\-openflowd\fR command line as the \fIcontroller\fR
argument.  You must also configure the network device for the OpenFlow
``local port'' to allow \fBovs\-openflowd\fR to connect to that controller.
The OpenFlow local port is a virtual network port that \fBovs\-openflowd\fR
bridges to the physical switch ports.  The name of the local port for
a given \fIdatapath\fR may be seen by running \fBovs\-dpctl show
\fIdatapath\fR; the local port is listed as port 0 in \fBshow\fR's
output.

.IP
Before \fBovs\-openflowd\fR starts, the local port network device is not
bridged to any physical network, so the next step depends on whether
connectivity is required to configure the device's IP address.  If the
switch has a static IP address, you may configure its IP address now
with a command such as 
.B ifconfig of0 192.168.1.1
and then invoke \fBovs\-openflowd\fR.

On the other hand, if the switch does not have a static IP address,
e.g. it obtains its IP address dynamically via DHCP, the DHCP client
will not be able to contact the DHCP server until the OpenFlow switch
has started up.  Thus, start \fBovs\-openflowd\fR without configuring
the local port network device, and start the DHCP client afterward.
.RE

.SH OPTIONS
.SS "Controller Discovery Options"
.TP
\fB--accept-vconn=\fIregex\fR
When \fBovs\-openflowd\fR performs controller discovery (see \fBContacting
the Controller\fR, above, for more information about controller
discovery), it validates the controller location obtained via DHCP
with a POSIX extended regular expression.  Only controllers whose
names match the regular expression will be accepted.

The default regular expression is \fBssl:.*\fR (meaning that only SSL
controller connections will be accepted) when any of the SSL
configuration options \fB--private-key\fR, \fB--certificate\fR, or
\fB--ca-cert\fR is specified.  The default is \fB^tcp:.*\fR otherwise
(meaning that only TCP controller connections will be accepted).

The \fIregex\fR is implicitly anchored at the beginning of the
controller location string, as if it begins with \fB^\fR.

When controller discovery is not performed, this option has no effect.

.TP
\fB--no-resolv-conf\fR
When \fBovs\-openflowd\fR performs controller discovery (see \fBContacting
the Controller\fR, above, for more information about controller
discovery), by default it overwrites the system's
\fB/etc/resolv.conf\fR with domain information and DNS servers
obtained via DHCP.  If the location of the controller is specified
using a hostname, rather than an IP address, and the network's DNS
servers ever change, this behavior is essential.  But because it also
interferes with any administrator or process that manages
\fB/etc/resolv.conf\fR, when this option is specified, \fBovs\-openflowd\fR
will not modify \fB/etc/resolv.conf\fR.

\fBovs\-openflowd\fR will only modify \fBresolv.conf\fR if the DHCP response
that it receives specifies one or more DNS servers.

When controller discovery is not performed, this option has no effect.

.SS "Networking Options"
.TP
\fB--datapath-id=\fIdpid\fR
Sets \fIdpid\fR, which must consist of exactly 12 hexadecimal digits,
as the datapath ID that the switch will use to identify itself to the
OpenFlow controller.

If this option is omitted, the default datapath ID is taken from the
Ethernet address of the datapath's local port (which is typically
randomly generated).

.TP
\fB--mgmt-id=\fImgmtid\fR
Sets \fImgmtid\fR, which must consist of exactly 12 hexadecimal
digits, as the switch's management ID.

If this option is omitted, the management ID defaults to 0, signaling
to the controller that management is supported but not configured.

.TP
\fB--fail=\fR[\fBopen\fR|\fBclosed\fR]
The controller is, ordinarily, responsible for setting up all flows on
the OpenFlow switch.  Thus, if the connection to the controller fails,
no new network connections can be set up.  If the connection to the
controller stays down long enough, no packets can pass through the
switch at all.

If this option is set to \fBopen\fR (the default), \fBovs\-openflowd\fR will
take over responsibility for setting up flows in the local datapath
when no message has been received from the controller for three times
the inactivity probe interval (see below), or 45 seconds by default.
In this ``fail open'' mode, \fBovs\-openflowd\fR causes the datapath to act
like an ordinary MAC-learning switch.  \fBovs\-openflowd\fR will continue to
retry connection to the controller in the background and, when the
connection succeeds, it discontinues its fail-open behavior.

If this option is set to \fBclosed\fR, then \fBovs\-openflowd\fR will not
set up flows on its own when the controller connection fails.

.TP
\fB--inactivity-probe=\fIsecs\fR
When the OpenFlow switch is connected to the controller, the
switch waits for a message to be received from the controller for
\fIsecs\fR seconds before it sends a inactivity probe to the
controller.  After sending the inactivity probe, if no response is
received for an additional \fIsecs\fR seconds, the switch
assumes that the connection has been broken and attempts to reconnect.
The default and the minimum value are both 5 seconds.

When fail-open mode is configured, changing the inactivity probe
interval also changes the interval before entering fail-open mode (see
above).

.TP
\fB--max-idle=\fIsecs\fR|\fBpermanent\fR
Sets \fIsecs\fR as the number of seconds that a flow set up by the
OpenFlow switch will remain in the switch's flow table without any
matching packets being seen.  If \fBpermanent\fR is specified, which
is not recommended, flows set up by the switch will never
expire.  The default is 15 seconds.

Most flows are set up by the OpenFlow controller, not by the
switch.  This option affects only the following flows, which the
OpenFlow switch sets up itself:

.RS
.IP \(bu
When \fB--fail=open\fR is specified, flows set up when the
switch has not been able to contact the controller for the configured
fail-open delay.

.IP \(bu
When in-band control is in use, flows set up to bootstrap contacting
the controller (see \fBContacting the Controller\fR, above, for
more information about in-band control).
.RE

.IP
As a result, when both \fB--fail=closed\fR and \fB--out-of-band\fR are
specified, this option has no effect.

.TP
\fB--max-backoff=\fIsecs\fR
Sets the maximum time between attempts to connect to the controller to
\fIsecs\fR, which must be at least 1.  The actual interval between
connection attempts starts at 1 second and doubles on each failing
attempt until it reaches the maximum.  The default maximum backoff
time is 8 seconds.

.TP
\fB-l\fR, \fB--listen=\fImethod\fR
By default, the switch listens for OpenFlow management connections on a
Unix domain socket named \fB@RUNDIR@/\fIdatapath\fB.mgmt\fR.  This socket 
can be used to perform local OpenFlow monitoring and administration with
tools such as \fBovs\-ofctl\fR.  

This option may be used to override the default listener.  The \fImethod\fR
must be given as one of the passive OpenFlow connection methods listed
below.  This option may be specified multiple times to listen to
multiple connection methods.  If a single \fImethod\fR of \fBnone\fR is
used, no listeners will be created.

.RS
.TP
\fBpssl:\fR[\fIport\fR][\fB:\fIip\fR]
Listens for SSL connections on \fIport\fR (default: 6633).  The
\fB--private-key\fR, \fB--certificate\fR, and \fB--ca-cert\fR options
are mandatory when this form is used.
By default, \fB\*(PN\fR listens for connections to any local IP
address, but \fIip\fR may be specified to listen only for connections
to the given \fIip\fR.

.TP
\fBptcp:\fR[\fIport\fR][\fB:\fIip\fR]
Listens for TCP connections on \fIport\fR (default: 6633).
By default, \fB\*(PN\fR listens for connections to any local IP
address, but \fIip\fR may be specified to listen only for connections
to the given \fIip\fR.

.TP
\fBpunix:\fIfile\fR
Listens for connections on Unix domain server socket named \fIfile\fR.
.RE

.TP
\fB--snoop=\fImethod\fR
Configures the switch to additionally listen for incoming OpenFlow
connections for controller connection snooping.  The \fImethod\fR must
be given as one of the passive OpenFlow connection methods listed
under the \fB--listen\fR option above.  This option may be specified
multiple times to listen to multiple connection methods.

If \fBovs\-ofctl monitor\fR is used to connect to \fImethod\fR specified on
\fB--snoop\fR, it will display all the OpenFlow messages traveling
between the switch and its controller on the primary OpenFlow
connection.  This can be useful for debugging switch and controller
problems.

.TP
\fB--in-band\fR, \fB--out-of-band\fR
Configures \fBovs\-openflowd\fR to operate in in-band or out-of-band control
mode (see \fBContacting the Controller\fR above).  When neither option
is given, the default is in-band control.

.TP
\fB--netflow=\fIip\fB:\fIport\fR
Configures the given UDP \fIport\fR on the specified IP \fIip\fR as
a recipient of NetFlow messages for expired flows.  The \fIip\fR must
be specified numerically, not as a DNS name.

This option may be specified multiple times to configure additional
NetFlow collectors.

.SS "Rate-Limiting Options"

These options configure how the switch applies a ``token bucket'' to
limit the rate at which packets in unknown flows are forwarded to an
OpenFlow controller for flow-setup processing.  This feature prevents
a single OpenFlow switch from overwhelming a controller.

.TP
\fB--rate-limit\fR[\fB=\fIrate\fR]
.
Limits the maximum rate at which packets will be forwarded to the
OpenFlow controller to \fIrate\fR packets per second.  If \fIrate\fR
is not specified then the default of 1,000 packets per second is used.

If \fB--rate-limit\fR is not used, then the switch does not limit the
rate at which packets are forwarded to the controller.

.TP
\fB--burst-limit=\fIburst\fR
.
Sets the maximum number of unused packet credits that the switch will
allow to accumulate during time in which no packets are being
forwarded to the OpenFlow controller to \fIburst\fR (measured in
packets).  The default \fIburst\fR is one-quarter of the \fIrate\fR
specified on \fB--rate-limit\fR.

This option takes effect only when \fB--rate-limit\fR is also specified.

.SS "Remote Command Execution Options"

.TP
\fB--command-acl=\fR[\fB!\fR]\fIglob\fR[\fB,\fR[\fB!\fR]\fIglob\fR...]
Configures the commands that remote OpenFlow connections are allowed
to invoke using (e.g.) \fBovs\-ofctl execute\fR.  The argument is a
comma-separated sequence of shell glob patterns.  A glob pattern
specified without a leading \fB!\fR is a ``whitelist'' that specifies
a set of commands that are that may be invoked, whereas a pattern that
does begin with \fB!\fR is a ``blacklist'' that specifies commands
that may not be invoked.  To be permitted, a command name must be
whitelisted and must not be blacklisted;
e.g. \fB--command-acl=up*,!upgrade\fR would allow any command whose name
begins with \fBup\fR except for the command named \fBupgrade\fR.
Command names that include characters other than upper- and lower-case
English letters, digits, and the underscore and hyphen characters are
unconditionally disallowed.

When the whitelist and blacklist permit a command name, \fBovs\-openflowd\fR
looks for a program with the same name as the command in the commands
directory (see below).  Other directories are not searched.

.TP
\fB--command-dir=\fIdirectory\fR
Sets the directory searched for remote command execution to
\fBdirectory\fR.  The default directory is
\fB@pkgdatadir@/commands\fR.

.SS "Datapath Options"
.
.IP "\fB\-\-ports=\fIport\fR[\fB,\fIport\fR...]"
Ordinarily, \fBovs\-openflowd\fR expects the administrator to create
the specified \fIdatapath\fR and add ports to it externally with a
utility such as \fBovs\-dpctl\fR.  However, the userspace switch
datapath is implemented inside \fBovs\-openflowd\fR itself and does
not (currently) have any external interface for \fBovs\-dpctl\fR to
access.  As a stopgap measure, this option specifies one or more ports
to add to the datapath at \fBovs\-openflowd\fR startup time.  Multiple
ports may be specified as a comma-separated list or by specifying
\fB\-\-ports\fR multiple times.
.IP
See \fBINSTALL.userspace\fR for more information about userspace
switching.

.SS "Daemon Options"
.so lib/daemon.man

.SS "Public Key Infrastructure Options"

.TP
\fB-p\fR, \fB--private-key=\fIprivkey.pem\fR
Specifies a PEM file containing the private key used as the switch's
identity for SSL connections to the controller.

.TP
\fB-c\fR, \fB--certificate=\fIcert.pem\fR
Specifies a PEM file containing a certificate, signed by the
controller's certificate authority (CA), that certifies the switch's
private key to identify a trustworthy switch.

.TP
\fB-C\fR, \fB--ca-cert=\fIcacert.pem\fR
Specifies a PEM file containing the CA certificate used to verify that
the switch is connected to a trustworthy controller.

.TP
\fB--bootstrap-ca-cert=\fIcacert.pem\fR
When \fIcacert.pem\fR exists, this option has the same effect as
\fB-C\fR or \fB--ca-cert\fR.  If it does not exist, then \fBovs\-openflowd\fR
will attempt to obtain the CA certificate from the controller on its
first SSL connection and save it to the named PEM file.  If it is
successful, it will immediately drop the connection and reconnect, and
from then on all SSL connections must be authenticated by a
certificate signed by the CA certificate thus obtained.

\fBThis option exposes the SSL connection to a man-in-the-middle
attack obtaining the initial CA certificate\fR, but it may be useful
for bootstrapping.

This option is only useful if the controller sends its CA certificate
as part of the SSL certificate chain.  The SSL protocol does not
require the controller to send the CA certificate, but
\fBcontroller\fR(8) can be configured to do so with the
\fB--peer-ca-cert\fR option.

.SS "Logging Options"
.so lib/vlog.man
.SS "Other Options"
.so lib/common.man
.so lib/leak-checker.man

.SH "SEE ALSO"

.BR ovs\-appctl (8),
.BR ovs\-controller (8),
.BR ovs\-discover (8),
.BR ovs\-dpctl (8),
.BR ovs\-ofctl (8),
.BR ovs\-pki (8),
.BR ovs\-vswitchd.conf (5)