When monitor(port-mirroring) traffic from other lacp port-channel, rx_machine_update may recieving other lacp negotiation packets. Thus bond mode 4 will negotiation failed. Fixes: 112891cd27e5 ("net/bonding: add dedicated HW queues for LACP control") Cc: stable@dpdk.org Signed-off-by: Liang Zhang <zhangliang@bigo.sg> --- drivers/net/bonding/rte_eth_bond_8023ad.c | 12 ++++++++++-- drivers/net/bonding/rte_eth_bond_8023ad_private.h | 1 + 2 files changed, 11 insertions(+), 2 deletions(-) diff --git a/drivers/net/bonding/rte_eth_bond_8023ad.c b/drivers/net/bonding/rte_eth_bond_8023ad.c index dd847c6..dac23ac 100644 --- a/drivers/net/bonding/rte_eth_bond_8023ad.c +++ b/drivers/net/bonding/rte_eth_bond_8023ad.c @@ -784,13 +784,20 @@ static void rx_machine_update(struct bond_dev_private *internals, uint8_t slave_id, struct rte_mbuf *lacp_pkt) { struct lacpdu_header *lacp; + struct lacpdu_actor_partner_params *partner; if (lacp_pkt != NULL) { lacp = rte_pktmbuf_mtod(lacp_pkt, struct lacpdu_header *); RTE_ASSERT(lacp->lacpdu.subtype == SLOW_SUBTYPE_LACP); - /* This is LACP frame so pass it to rx_machine */ - rx_machine(internals, slave_id, &lacp->lacpdu); + partner = &lacp->lacpdu.partner; + if (is_same_ether_addr(&partner->port_params.system, + &internals->mode4.mac_addr)) { + /* This LACP frame is sending to the bonding port + * so pass it to rx_machine. + */ + rx_machine(internals, slave_id, &lacp->lacpdu); + } rte_pktmbuf_free(lacp_pkt); } else rx_machine(internals, slave_id, NULL); @@ -1165,6 +1172,7 @@ bond_mode_8023ad_start(struct rte_eth_dev *bond_dev) struct mode8023ad_private *mode4 = &internals->mode4; static const uint64_t us = BOND_MODE_8023AX_UPDATE_TIMEOUT_MS * 1000; + rte_eth_macaddr_get(internals->port_id, &mode4->mac_addr); if (mode4->slowrx_cb) return rte_eal_alarm_set(us, &bond_mode_8023ad_ext_periodic_cb, bond_dev); diff --git a/drivers/net/bonding/rte_eth_bond_8023ad_private.h b/drivers/net/bonding/rte_eth_bond_8023ad_private.h index c51426b..f91902e 100644 --- a/drivers/net/bonding/rte_eth_bond_8023ad_private.h +++ b/drivers/net/bonding/rte_eth_bond_8023ad_private.h @@ -150,6 +150,7 @@ struct mode8023ad_private { uint64_t update_timeout_us; rte_eth_bond_8023ad_ext_slowrx_fn slowrx_cb; uint8_t external_sm; + struct ether_addr mac_addr; struct rte_eth_link slave_link; /***< slave link properties */ -- 2.7.4
From: Yicai Lu <luyicai@huawei.com> When two host is connected directly without any devices like switch, rx_machine_update would recieving partner lacp negotiation packets, which partner's port mac filled with zeros. So in this situation, it would never go rx_machine branch with correct mac! Thus bond mode 4 will negotiation failed. Fixes: 56cbc0817399 ("net/bonding: fix LACP negotiation") Cc: stable@dpdk.org Signed-off-by: luyicai <luyicai@huawei.com> --- v1 -> v2: Adjust commit info style --- drivers/net/bonding/rte_eth_bond_8023ad.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/net/bonding/rte_eth_bond_8023ad.c b/drivers/net/bonding/rte_eth_bond_8023ad.c index b77a37d..2002ec0 100644 --- a/drivers/net/bonding/rte_eth_bond_8023ad.c +++ b/drivers/net/bonding/rte_eth_bond_8023ad.c @@ -798,7 +798,8 @@ RTE_ASSERT(lacp->lacpdu.subtype == SLOW_SUBTYPE_LACP); partner = &lacp->lacpdu.partner; - if (rte_is_same_ether_addr(&partner->port_params.system, + if (rte_is_zero_ether_addr(&partner->port_params.system) || + rte_is_same_ether_addr(&partner->port_params.system, &internals->mode4.mac_addr)) { /* This LACP frame is sending to the bonding port * so pass it to rx_machine. -- 1.9.5.msysgit.1
From: Yicai Lu <luyicai@huawei.com> When two host is connected directly without any devices like switch, rx_machine_update would recieving partner lacp negotiation packets, and partner's port mac is filled with zeros in this packet, which is different with internal's mode4 mac. So in this situation, it would never go rx_machine branch and then execute mac swap for negotiation! Thus bond mode 4 will negotiation failed. Fixes: 56cbc0817399 ("net/bonding: fix LACP negotiation") Cc: stable@dpdk.org Signed-off-by: luyicai <luyicai@huawei.com> --- v1 -> v2: Adjust commit info style --- drivers/net/bonding/rte_eth_bond_8023ad.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/net/bonding/rte_eth_bond_8023ad.c b/drivers/net/bonding/rte_eth_bond_8023ad.c index b77a37d..2002ec0 100644 --- a/drivers/net/bonding/rte_eth_bond_8023ad.c +++ b/drivers/net/bonding/rte_eth_bond_8023ad.c @@ -798,7 +798,8 @@ RTE_ASSERT(lacp->lacpdu.subtype == SLOW_SUBTYPE_LACP); partner = &lacp->lacpdu.partner; - if (rte_is_same_ether_addr(&partner->port_params.system, + if (rte_is_zero_ether_addr(&partner->port_params.system) || + rte_is_same_ether_addr(&partner->port_params.system, &internals->mode4.mac_addr)) { /* This LACP frame is sending to the bonding port * so pass it to rx_machine. -- 1.9.5.msysgit.1
Reviewed-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
On 2020/7/10 11:29, luyicai wrote:
> From: Yicai Lu <luyicai@huawei.com>
>
> When two host is connected directly without any devices like switch,
> rx_machine_update would recieving partner lacp negotiation packets,
> and partner's port mac is filled with zeros in this packet,
> which is different with internal's mode4 mac. So in this situation,
> it would never go rx_machine branch and then execute mac swap for negotiation!
> Thus bond mode 4 will negotiation failed.
>
> Fixes: 56cbc0817399 ("net/bonding: fix LACP negotiation")
> Cc: stable@dpdk.org
>
> Signed-off-by: luyicai <luyicai@huawei.com>
> ---
> v1 -> v2: Adjust commit info style
> ---
> drivers/net/bonding/rte_eth_bond_8023ad.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/net/bonding/rte_eth_bond_8023ad.c b/drivers/net/bonding/rte_eth_bond_8023ad.c
> index b77a37d..2002ec0 100644
> --- a/drivers/net/bonding/rte_eth_bond_8023ad.c
> +++ b/drivers/net/bonding/rte_eth_bond_8023ad.c
> @@ -798,7 +798,8 @@
> RTE_ASSERT(lacp->lacpdu.subtype == SLOW_SUBTYPE_LACP);
>
> partner = &lacp->lacpdu.partner;
> - if (rte_is_same_ether_addr(&partner->port_params.system,
> + if (rte_is_zero_ether_addr(&partner->port_params.system) ||
> + rte_is_same_ether_addr(&partner->port_params.system,
> &internals->mode4.mac_addr)) {
> /* This LACP frame is sending to the bonding port
> * so pass it to rx_machine.
On 7/11/2020 2:20 AM, Wei Hu (Xavier) wrote:
>
> On 2020/7/10 11:29, luyicai wrote:
>> From: Yicai Lu <luyicai@huawei.com>
>>
>> When two host is connected directly without any devices like switch,
>> rx_machine_update would recieving partner lacp negotiation packets,
>> and partner's port mac is filled with zeros in this packet,
>> which is different with internal's mode4 mac. So in this situation,
>> it would never go rx_machine branch and then execute mac swap for negotiation!
>> Thus bond mode 4 will negotiation failed.
>>
>> Fixes: 56cbc0817399 ("net/bonding: fix LACP negotiation")
>> Cc: stable@dpdk.org
>>
>> Signed-off-by: luyicai <luyicai@huawei.com>
>
> Reviewed-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
>
Applied to dpdk-next-net/master, thanks.