Fwd: LACP: active aggregator selection bug

[Boris Astardzhiev]

Hi,

I've been investigating the LACP implementation in FreeBSD and have
encountered a bug. Here's the set:

"                      ---------          ----------              "
"                      | lagg1 |          | bond0  |              "
"   ---------          |      xl0--------eth0      |   ---------  "
"   | hosts |----b1----1 FBSD rl0--------eth1 Linux|---| hosts |  "
"   ---------          |  9.1 rl1--------eth2      |   ---------  "
"                      |       |          |        |              "
"                      ---------          ----------              "

On a FreeBSD 9.1-RELEASE #0 r243826 system a lagg is created and three
interfaces are added to it:
- xl0
- rl0
- rl1

On a Linux system a bonding interface is added *ONLY ONE* interface:
- eth0
Note: I think the Linux may be substituted with any other LACP
implementation.

The lagg protocol on both of the systems is LACP.

LACPDUs transmission/reception takes place only between xl0 and eth0.
Here's the result:

root at freebsd91:/root # ifconfig lagg1
lagg1: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> metric 0 mtu 1500
    options=2008<VLAN_MTU,WOL_MAGIC>
    ether 00:10:b5:7f:97:fb
    inet6 fe80::210:b5ff:fe7f:97fb%lagg1 prefixlen 64 scopeid 0x9
    nd6 options=21<PERFORMNUD,AUTO_LINKLOCAL>
    media: Ethernet autoselect
    status: active
    laggproto lacp lagghash l2,l3,l4
    laggport: xl0 flags=18<COLLECTING,DISTRIBUTING>
    laggport: rl1 flags=1c<ACTIVE,COLLECTING,DISTRIBUTING>
    laggport: rl0 flags=1c<ACTIVE,COLLECTING,DISTRIBUTING>

I consider that xl0 is the only available link therefor the aggregation must
rely on it. However the lacp implementation has chosen the other two links
that haven't received a single LACPDU.

I think the problem is related to the selection of best active aggregator -
in lacp_select_active_aggregator(). I've attached the debug output of
sysctl net.lacp_debug.

... snippet ...
Jun 24 10:41:43 freebsd91 kernel: xl0: new pstate
3f<ACTIVITY,TIMEOUT,AGGREGATION,SYNC,COLLECTING,DISTRIBUTING>
Jun 24 10:41:43 freebsd91 kernel: rl0: lacp_sm_mux: state 4
Jun 24 10:41:43 freebsd91 kernel: rl1: lacp_sm_mux: state 4
Jun 24 10:41:43 freebsd91 kernel: xl0: lacp_sm_mux: state 3
Jun 24 10:41:43 freebsd91 kernel: xl0: enable distributing on aggregator
[(8000,00-10-B5-7F-97-FB,0126,0000,0000),(FFFF,E0-8F-EC-00-B5-2F,0009,0000,0000)],
nports 0 -> 1
Jun 24 10:41:43 freebsd91 kernel: lacp_select_active_aggregator
Jun 24 10:41:43 freebsd91 kernel:
[(8000,00-10-B5-7F-97-FB,0126,0000,0000),(FFFF,00-00-00-00-00-00,0000,0000,0000)],
speed=200000000, nports=2
Jun 24 10:41:43 freebsd91 kernel:
[(8000,00-10-B5-7F-97-FB,0126,0000,0000),(FFFF,E0-8F-EC-00-B5-2F,0009,0000,0000)],
speed=100000000, nports=1
Jun 24 10:41:43 freebsd91 kernel: active aggregator not changed
Jun 24 10:41:43 freebsd91 kernel: new
[(8000,00-10-B5-7F-97-FB,0126,0000,0000),(FFFF,00-00-00-00-00-00,0000,0000,0000)]
Jun 24 10:41:43 freebsd91 kernel: xl0: mux_state 3 -> 4
Jun 24 10:41:43 freebsd91 kernel: xl0: lacpdu transmit
... snippet ...

Though there is an aggregator with an active partner the implementation has
chosen
the other aggregator:
Jun 24 10:41:43 freebsd91 kernel: new
[(8000,00-10-B5-7F-97-FB,0126,0000,0000),(FFFF,00-00-00-00-00-00,0000,0000,0000)]

Do you think that such aggregators must be skipped in favour of aggregators
with active
partners? I've applied a patch that fixes this issue and xl0 remains the
only active link
but I'm not sure it is correct and it has the correct approach.

I've posted a PR - http://www.freebsd.org/cgi/query-pr.cgi?pr=179926

Any comments are appreciated.

Greetings,
Boris Astardzhiev,
Smartcom Bulgaria AD
-------------- next part --------------
A non-text attachment was scrubbed...
Name: lacp_debug-pr
Type: application/octet-stream
Size: 23808 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-net/attachments/20130624/13243a49/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: lacp-aggregator-select-skip-zero-partner.diff
Type: application/octet-stream
Size: 1283 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-net/attachments/20130624/13243a49/attachment-0001.obj>