ng_one2many
—
packet multiplexing netgraph node type
The one2many
provides a simple mechanism
for routing packets over several links in a one-to-many (and in the reverse
direction, many-to-one) fashion. There is a single hook named
one
, and multiple hooks named
many0
, many1
, etc. Packets
received on any of the many
hooks are forwarded out
the one
hook. Packets received on the
one
hook are forwarded out one or more of the
many
hooks; which hook(s) is determined by the
node's configured transmit algorithm. Packets are not altered in any
way.
Each of the connected many links may be considered to be up or
down. Packets are never delivered out a many hook that is down. How a link
is determined to be up or down depends on the node's configured link failure
detection algorithm.
Before an interface or link can be plumbed into a group, its
status must be marked as being “up”. This is normally setup
during the initial boot stages by
rc.conf(5).
It is also possible to change an interface's status to “up” by
using the
ifconfig(8)
utility.
NG_ONE2MANY_XMIT_ROUNDROBIN
- Packets are delivered out the many hooks in sequential order. Each packet
goes out on a different
many
hook.
NG_ONE2MANY_XMIT_ALL
- Packets are delivered out all the
many
hooks. Each
packet goes out each many
hook.
NG_ONE2MANY_XMIT_FAILOVER
- Packets are delivered out the first active
many
hook.
In the future other algorithms may be added as well.
The node distinguishes between active and failed links. Data is
sent only to active links. The following link failure detection algorithms
are available:
NG_ONE2MANY_FAIL_MANUAL
- The node is explicitly told which of the links are up via the
NGM_ONE2MANY_SET_CONFIG
control message (see
below). Newly connected links are down until configured otherwise.
NG_ONE2MANY_FAIL_NOTIFY
- The node listens to flow control message from many
hooks, and considers link failed if
NGM_LINK_IS_DOWN
is received. If the
NGM_LINK_IS_UP
message is received, node considers
link active.
In the future other algorithms may be added as well.
When all links are considered failed, node sends the
NGM_LINK_IS_DOWN
message towards the
one hook. When at least one link comes up, node sends
the NGM_LINK_IS_UP
message towards the
one hook.
This node type supports up to
NG_ONE2MANY_MAX_LINKS
hooks named
many0
, many1
, etc., plus a
single hook named one
.
This node type supports the generic control messages, plus the
following:
NGM_ONE2MANY_SET_CONFIG
(setconfig
)
- Sets the node configuration using a
struct
ng_one2many_link_config
as the control message argument:
/* Node configuration structure */
struct ng_one2many_config {
uint32_t xmitAlg; /* how to distribute packets */
uint32_t failAlg; /* how to detect link failure */
u_char enabledLinks[NG_ONE2MANY_MAX_LINKS];
};
Currently, the valid settings for the
xmitAlg
field are
NG_ONE2MANY_XMIT_ROUNDROBIN
(default) or
NG_ONE2MANY_XMIT_ALL
. The valid settings for
failAlg
are
NG_ONE2MANY_FAIL_MANUAL
(default) or
NG_ONE2MANY_FAIL_NOTIFY
.
NGM_ONE2MANY_GET_CONFIG
(getconfig
)
- Returns the current node configuration in a
struct
ng_one2many_link_config
.
NGM_ONE2MANY_GET_STATS
(getstats
)
- This command takes a 32 bit link number as an argument and returns a
struct ng_one2many_link_stats
containing
statistics for the corresponding many
link, which
may or may not be currently connected:
/* Statistics structure (one for each link) */
struct ng_one2many_link_stats {
uint64_t recvOctets; /* total octets rec'd on link */
uint64_t recvPackets; /* total pkts rec'd on link */
uint64_t xmitOctets; /* total octets xmit'd on link */
uint64_t xmitPackets; /* total pkts xmit'd on link */
uint64_t memoryFailures; /* times couldn't get mem or mbuf */
};
To access statistics for the one
link,
use the link number -1
.
NGM_ONE2MANY_CLR_STATS
(clrstats
)
- This command takes a 32 bit link number as an argument and clears the
statistics for that link.
NGM_ONE2MANY_GETCLR_STATS
(getclrstats
)
- Same as
NGM_ONE2MANY_GET_STATS
, but also
atomically clears the statistics for the link as well.
This node shuts down upon receipt of a
NGM_SHUTDOWN
control message, or when all hooks have
been disconnected.
The following commands will set up Ethernet interfaces
fxp0
to deliver packets alternating over the
physical interfaces corresponding to networking interfaces
fxp0
through fxp3
:
# Plumb nodes together
ngctl mkpeer fxp0: one2many upper one
ngctl connect fxp0: fxp0:upper lower many0
ngctl connect fxp1: fxp0:upper lower many1
ngctl connect fxp2: fxp0:upper lower many2
ngctl connect fxp3: fxp0:upper lower many3
# Allow fxp1 through fxp3 to xmit/recv fxp0 frames
ngctl msg fxp1: setpromisc 1
ngctl msg fxp2: setpromisc 1
ngctl msg fxp3: setpromisc 1
ngctl msg fxp1: setautosrc 0
ngctl msg fxp2: setautosrc 0
ngctl msg fxp3: setautosrc 0
# Configure all four links as up
ngctl msg fxp0:upper \
setconfig "{ xmitAlg=1 failAlg=1 enabledLinks=[ 1 1 1 1 ] }"
# Bring up interface
ifconfig fxp0 192.168.1.1 netmask 0xfffffffc
With a similar setup on a peer machine (using the address
192.168.1.2), a point-to-point Ethernet connection with four times normal
bandwidth is achieved.
The ng_one2many
node type was implemented
in FreeBSD 4.2.
More transmit and link failure algorithms should be supported. A
good candidate is Cisco's Etherchannel.