[dpdk-dev] net/bonding: strengthen the judgment of lacp packets
Checks
Commit Message
When the nic does not support vlan rx offload may be wrong, resulting in
lacp packets will not be processed.
Signed-off-by: ZengGanghui <zengganghui@huawei.com>
---
drivers/net/bonding/rte_eth_bond_pmd.c | 7 ++++---
1 file changed, 4 insertions(+), 3 deletions(-)
Comments
On 8/30/2017 4:46 AM, ZengGanghui wrote:
> When the nic does not support vlan rx offload may be wrong, resulting in
> lacp packets will not be processed.
>
> Signed-off-by: ZengGanghui <zengganghui@huawei.com>
Reviewed-by: Radu Nicolau <radu.nicolau@intel.com>
On 30/08/2017 4:46 AM, ZengGanghui wrote:
> When the nic does not support vlan rx offload may be wrong, resulting in
> lacp packets will not be processed.
>
> Signed-off-by: ZengGanghui <zengganghui@huawei.com>
> ---
...
>
Acked-by: Declan Doherty <declan.doherty@intel.com>
For example, when packets received from an MLX network card, the value of mbuf->vlan_tci is a random value. So that this value cannot be used to determine whether VLAN packets . We need to judgment mbuf->ol_flags first.
BR.
Zeng Ganghui
Huawei Technologies Co., Ltd.
-----Original Message-----
From: Doherty, Declan [mailto:declan.doherty@intel.com]
Sent: Monday, September 18, 2017 5:14 PM
To: zengganghui; dev@dpdk.org
Subject: Re: [PATCH] net/bonding: strengthen the judgment of lacp packets
On 30/08/2017 4:46 AM, ZengGanghui wrote:
> When the nic does not support vlan rx offload may be wrong, resulting
> in lacp packets will not be processed.
>
> Signed-off-by: ZengGanghui <zengganghui@huawei.com>
> ---
...
>
Acked-by: Declan Doherty <declan.doherty@intel.com>
On 18/09/2017 12:12 PM, zengganghui wrote:
> For example, when packets received from an MLX network card, the value of mbuf->vlan_tci is a random value. So that this value cannot be used to determine whether VLAN packets . We need to judgment mbuf->ol_flags first.
>
> BR.
> Zeng Ganghui
> Huawei Technologies Co., Ltd.
>
> -----Original Message-----
> From: Doherty, Declan [mailto:declan.doherty@intel.com]
> Sent: Monday, September 18, 2017 5:14 PM
> To: zengganghui; dev@dpdk.org
> Subject: Re: [PATCH] net/bonding: strengthen the judgment of lacp packets
>
> On 30/08/2017 4:46 AM, ZengGanghui wrote:
>> When the nic does not support vlan rx offload may be wrong, resulting
>> in lacp packets will not be processed.
>>
>> Signed-off-by: ZengGanghui <zengganghui@huawei.com>
>> ---
> ...
>>
>
> Acked-by: Declan Doherty <declan.doherty@intel.com>
>
Ok, I see your point. A LACP PDU can't be encapsulated in a VLAN packet
anyway, as it is link local traffic. So a check for ol_flags &
PKT_RX_VLAN_PKT != 0 should be sufficient, otherwise if the
PKT_RX_VLAN_PKT flag is true the packet cannot be link local and
therefore a LACP PDU. I think that it's safe to assume all PMDs must set
this flag if VLAN stripping is enabled?
All mbuf packets have been init to zero when pktmbuf pool create. So judgment this flag is safe, whether or not support VLAN stripping.
BR.
Zeng Ganghui
Huawei Technologies Co., Ltd.
-----Original Message-----
From: Doherty, Declan [mailto:declan.doherty@intel.com]
Sent: Monday, September 18, 2017 8:34 PM
To: zengganghui; dev@dpdk.org
Subject: Re: [PATCH] net/bonding: strengthen the judgment of lacp packets
On 18/09/2017 12:12 PM, zengganghui wrote:
> For example, when packets received from an MLX network card, the value of mbuf->vlan_tci is a random value. So that this value cannot be used to determine whether VLAN packets . We need to judgment mbuf->ol_flags first.
>
> BR.
> Zeng Ganghui
> Huawei Technologies Co., Ltd.
>
> -----Original Message-----
> From: Doherty, Declan [mailto:declan.doherty@intel.com]
> Sent: Monday, September 18, 2017 5:14 PM
> To: zengganghui; dev@dpdk.org
> Subject: Re: [PATCH] net/bonding: strengthen the judgment of lacp
> packets
>
> On 30/08/2017 4:46 AM, ZengGanghui wrote:
>> When the nic does not support vlan rx offload may be wrong, resulting
>> in lacp packets will not be processed.
>>
>> Signed-off-by: ZengGanghui <zengganghui@huawei.com>
>> ---
> ...
>>
>
> Acked-by: Declan Doherty <declan.doherty@intel.com>
>
Ok, I see your point. A LACP PDU can't be encapsulated in a VLAN packet anyway, as it is link local traffic. So a check for ol_flags & PKT_RX_VLAN_PKT != 0 should be sufficient, otherwise if the PKT_RX_VLAN_PKT flag is true the packet cannot be link local and therefore a LACP PDU. I think that it's safe to assume all PMDs must set this flag if VLAN stripping is enabled?
On 18/09/2017 1:50 PM, zengganghui wrote:
> All mbuf packets have been init to zero when pktmbuf pool create. So judgment this flag is safe, whether or not support VLAN stripping.
Ok, but is there any need to check the ol_flags for PKT_RX_VLAN_PKT or
check the vlan_tci at all. I haven't come across anything in the
specification which allows LACP links to be formed on top of VLANs but I
may be missing something? So if the ethertype is not ETHER_TYPE_SLOW it
is irrelevant whether the packet has a VLAN tag or not.
Also on the basis that you could have LAG groups on top of VLANs, if the
NIC doesn't support VLAN stripping/insertion then we would miss all the
ingress LACP PDU at the moment now anyway, since the ethertype would be
VLAN and not ETHER_TYPE_SLOW, so is_lacp_packet() would always return 0,
and we would also fail to encapsulate the LACP PDU in the correct VLAN
on egress as that isn't supported in the bonding implementation.
>
> BR.
> Zeng Ganghui
> Huawei Technologies Co., Ltd.
>
> -----Original Message-----
> From: Doherty, Declan [mailto:declan.doherty@intel.com]
> Sent: Monday, September 18, 2017 8:34 PM
> To: zengganghui; dev@dpdk.org
> Subject: Re: [PATCH] net/bonding: strengthen the judgment of lacp packets
>
> On 18/09/2017 12:12 PM, zengganghui wrote:
>> For example, when packets received from an MLX network card, the value of mbuf->vlan_tci is a random value. So that this value cannot be used to determine whether VLAN packets . We need to judgment mbuf->ol_flags first.
>>
>> BR.
>> Zeng Ganghui
>> Huawei Technologies Co., Ltd.
>>
>> -----Original Message-----
>> From: Doherty, Declan [mailto:declan.doherty@intel.com]
>> Sent: Monday, September 18, 2017 5:14 PM
>> To: zengganghui; dev@dpdk.org
>> Subject: Re: [PATCH] net/bonding: strengthen the judgment of lacp
>> packets
>>
>> On 30/08/2017 4:46 AM, ZengGanghui wrote:
>>> When the nic does not support vlan rx offload may be wrong, resulting
>>> in lacp packets will not be processed.
>>>
>>> Signed-off-by: ZengGanghui <zengganghui@huawei.com>
>>> ---
>> ...
>>>
>>
>> Acked-by: Declan Doherty <declan.doherty@intel.com>
>>
>
> Ok, I see your point. A LACP PDU can't be encapsulated in a VLAN packet anyway, as it is link local traffic. So a check for ol_flags & PKT_RX_VLAN_PKT != 0 should be sufficient, otherwise if the PKT_RX_VLAN_PKT flag is true the packet cannot be link local and therefore a LACP PDU. I think that it's safe to assume all PMDs must set this flag if VLAN stripping is enabled?
>
Local LACP packets do not have VLANs, and ethertype must be ETHER_TYPE_SLOW. But when the PMD supports VLAN strip, you cannot directly determine the ethertype from the packet, but depends on whether the VLAN is stripped. If a VLAN is stripped, then this is not a local LACP packet, but it may be a need to pass through packet. The previous code was not rigorous in determining whether the VLAN was being stripped.
Did I answer your question?
BR.
Zeng Ganghui
Huawei Technologies Co., Ltd.
-----Original Message-----
From: Doherty, Declan [mailto:declan.doherty@intel.com]
Sent: Monday, September 18, 2017 9:11 PM
To: zengganghui; dev@dpdk.org
Subject: Re: [PATCH] net/bonding: strengthen the judgment of lacp packets
On 18/09/2017 1:50 PM, zengganghui wrote:
> All mbuf packets have been init to zero when pktmbuf pool create. So judgment this flag is safe, whether or not support VLAN stripping.
Ok, but is there any need to check the ol_flags for PKT_RX_VLAN_PKT or check the vlan_tci at all. I haven't come across anything in the specification which allows LACP links to be formed on top of VLANs but I may be missing something? So if the ethertype is not ETHER_TYPE_SLOW it is irrelevant whether the packet has a VLAN tag or not.
Also on the basis that you could have LAG groups on top of VLANs, if the NIC doesn't support VLAN stripping/insertion then we would miss all the ingress LACP PDU at the moment now anyway, since the ethertype would be VLAN and not ETHER_TYPE_SLOW, so is_lacp_packet() would always return 0, and we would also fail to encapsulate the LACP PDU in the correct VLAN on egress as that isn't supported in the bonding implementation.
>
> BR.
> Zeng Ganghui
> Huawei Technologies Co., Ltd.
>
> -----Original Message-----
> From: Doherty, Declan [mailto:declan.doherty@intel.com]
> Sent: Monday, September 18, 2017 8:34 PM
> To: zengganghui; dev@dpdk.org
> Subject: Re: [PATCH] net/bonding: strengthen the judgment of lacp
> packets
>
> On 18/09/2017 12:12 PM, zengganghui wrote:
>> For example, when packets received from an MLX network card, the value of mbuf->vlan_tci is a random value. So that this value cannot be used to determine whether VLAN packets . We need to judgment mbuf->ol_flags first.
>>
>> BR.
>> Zeng Ganghui
>> Huawei Technologies Co., Ltd.
>>
>> -----Original Message-----
>> From: Doherty, Declan [mailto:declan.doherty@intel.com]
>> Sent: Monday, September 18, 2017 5:14 PM
>> To: zengganghui; dev@dpdk.org
>> Subject: Re: [PATCH] net/bonding: strengthen the judgment of lacp
>> packets
>>
>> On 30/08/2017 4:46 AM, ZengGanghui wrote:
>>> When the nic does not support vlan rx offload may be wrong,
>>> resulting in lacp packets will not be processed.
>>>
>>> Signed-off-by: ZengGanghui <zengganghui@huawei.com>
>>> ---
>> ...
>>>
>>
>> Acked-by: Declan Doherty <declan.doherty@intel.com>
>>
>
> Ok, I see your point. A LACP PDU can't be encapsulated in a VLAN packet anyway, as it is link local traffic. So a check for ol_flags & PKT_RX_VLAN_PKT != 0 should be sufficient, otherwise if the PKT_RX_VLAN_PKT flag is true the packet cannot be link local and therefore a LACP PDU. I think that it's safe to assume all PMDs must set this flag if VLAN stripping is enabled?
>
On 9/19/2017 5:09 AM, zengganghui wrote:
> Local LACP packets do not have VLANs, and ethertype must be ETHER_TYPE_SLOW. But when the PMD supports VLAN strip, you cannot directly determine the ethertype from the packet, but depends on whether the VLAN is stripped. If a VLAN is stripped, then this is not a local LACP packet, but it may be a need to pass through packet. The previous code was not rigorous in determining whether the VLAN was being stripped.
> Did I answer your question?
Hi Declan,
Are you OK with the patch?
You already have your ack on patch but discussion was going on...
>
> BR.
> Zeng Ganghui
> Huawei Technologies Co., Ltd.
>
> -----Original Message-----
> From: Doherty, Declan [mailto:declan.doherty@intel.com]
> Sent: Monday, September 18, 2017 9:11 PM
> To: zengganghui; dev@dpdk.org
> Subject: Re: [PATCH] net/bonding: strengthen the judgment of lacp packets
>
> On 18/09/2017 1:50 PM, zengganghui wrote:
>> All mbuf packets have been init to zero when pktmbuf pool create. So judgment this flag is safe, whether or not support VLAN stripping.
>
> Ok, but is there any need to check the ol_flags for PKT_RX_VLAN_PKT or check the vlan_tci at all. I haven't come across anything in the specification which allows LACP links to be formed on top of VLANs but I may be missing something? So if the ethertype is not ETHER_TYPE_SLOW it is irrelevant whether the packet has a VLAN tag or not.
>
> Also on the basis that you could have LAG groups on top of VLANs, if the NIC doesn't support VLAN stripping/insertion then we would miss all the ingress LACP PDU at the moment now anyway, since the ethertype would be VLAN and not ETHER_TYPE_SLOW, so is_lacp_packet() would always return 0, and we would also fail to encapsulate the LACP PDU in the correct VLAN on egress as that isn't supported in the bonding implementation.
>
>>
>> BR.
>> Zeng Ganghui
>> Huawei Technologies Co., Ltd.
>>
>> -----Original Message-----
>> From: Doherty, Declan [mailto:declan.doherty@intel.com]
>> Sent: Monday, September 18, 2017 8:34 PM
>> To: zengganghui; dev@dpdk.org
>> Subject: Re: [PATCH] net/bonding: strengthen the judgment of lacp
>> packets
>>
>> On 18/09/2017 12:12 PM, zengganghui wrote:
>>> For example, when packets received from an MLX network card, the value of mbuf->vlan_tci is a random value. So that this value cannot be used to determine whether VLAN packets . We need to judgment mbuf->ol_flags first.
>>>
>>> BR.
>>> Zeng Ganghui
>>> Huawei Technologies Co., Ltd.
>>>
>>> -----Original Message-----
>>> From: Doherty, Declan [mailto:declan.doherty@intel.com]
>>> Sent: Monday, September 18, 2017 5:14 PM
>>> To: zengganghui; dev@dpdk.org
>>> Subject: Re: [PATCH] net/bonding: strengthen the judgment of lacp
>>> packets
>>>
>>> On 30/08/2017 4:46 AM, ZengGanghui wrote:
>>>> When the nic does not support vlan rx offload may be wrong,
>>>> resulting in lacp packets will not be processed.
>>>>
>>>> Signed-off-by: ZengGanghui <zengganghui@huawei.com>
>>>> ---
>>> ...
>>>>
>>>
>>> Acked-by: Declan Doherty <declan.doherty@intel.com>
>>>
>>
>> Ok, I see your point. A LACP PDU can't be encapsulated in a VLAN packet anyway, as it is link local traffic. So a check for ol_flags & PKT_RX_VLAN_PKT != 0 should be sufficient, otherwise if the PKT_RX_VLAN_PKT flag is true the packet cannot be link local and therefore a LACP PDU. I think that it's safe to assume all PMDs must set this flag if VLAN stripping is enabled?
>>
On 03/10/2017 4:49 PM, Ferruh Yigit wrote:
> On 9/19/2017 5:09 AM, zengganghui wrote:
>> Local LACP packets do not have VLANs, and ethertype must be ETHER_TYPE_SLOW. But when the PMD supports VLAN strip, you cannot directly determine the ethertype from the packet, but depends on whether the VLAN is stripped. If a VLAN is stripped, then this is not a local LACP packet, but it may be a need to pass through packet. The previous code was not rigorous in determining whether the VLAN was being stripped.
>> Did I answer your question?
>
> Hi Declan,
>
> Are you OK with the patch?
> You already have your ack on patch but discussion was going on...
>
>
...
>
Hey Ferruh, sorry I missed this mail, yes I'm happy for this to be applied.
On 9/4/2017 2:19 PM, Radu Nicolau wrote:
>
> On 8/30/2017 4:46 AM, ZengGanghui wrote:
>> When the nic does not support vlan rx offload may be wrong, resulting in
>> lacp packets will not be processed.
>>
>> Signed-off-by: ZengGanghui <zengganghui@huawei.com>
> Reviewed-by: Radu Nicolau <radu.nicolau@intel.com>
Acked-by: Declan Doherty <declan.doherty@intel.com>
Applied to dpdk-next-net/master, thanks.
@@ -125,11 +125,12 @@
}
static inline uint8_t
-is_lacp_packets(uint16_t ethertype, uint8_t subtype, uint16_t vlan_tci)
+is_lacp_packets(uint16_t ethertype, uint8_t subtype, struct rte_mbuf *mbuf)
{
const uint16_t ether_type_slow_be = rte_be_to_cpu_16(ETHER_TYPE_SLOW);
- return !vlan_tci && (ethertype == ether_type_slow_be &&
+ return !((mbuf->ol_flags & PKT_RX_VLAN_PKT) ? mbuf->vlan_tci : 0) &&
+ (ethertype == ether_type_slow_be &&
(subtype == SLOW_SUBTYPE_MARKER || subtype == SLOW_SUBTYPE_LACP));
}
@@ -444,7 +445,7 @@
/* Remove packet from array if it is slow packet or slave is not
* in collecting state or bonding interface is not in promiscuous
* mode and packet address does not match. */
- if (unlikely(is_lacp_packets(hdr->ether_type, subtype, bufs[j]->vlan_tci) ||
+ if (unlikely(is_lacp_packets(hdr->ether_type, subtype, bufs[j]) ||
!collecting || (!promisc &&
!is_multicast_ether_addr(&hdr->d_addr) &&
!is_same_ether_addr(&bond_mac, &hdr->d_addr)))) {