[PATCH] net/netvsc: force TX VLAN offload on 801.2Q packet

[PATCH] net/netvsc: force TX VLAN offload on 801.2Q packet

[longli]

From: Long Li <longli@microsoft.com>

The VSP assumes the packet doesn't have VLAN tags. When VLAN tag is
present in a TX packet, always strip it and use PPI to send VLAN info
through VSP packet.

Fixes: 4e9c73e96e83 ("net/netvsc: add Hyper-V network device")
Cc: Hui (Hui) Ling <hling@SonicWall.com>
Cc: Stephen Hemminger <stephen@networkplumber.org>
Cc: Wei Hu <weh@microsoft.com>
Cc: stable@dpdk.org
Signed-off-by: Long Li <longli@microsoft.com>
---
 drivers/net/netvsc/hn_rxtx.c | 20 +++++++++++++++++++-
 1 file changed, 19 insertions(+), 1 deletion(-)

diff --git a/drivers/net/netvsc/hn_rxtx.c b/drivers/net/netvsc/hn_rxtx.c
index 9bf1ec5509..e58efe793e 100644
--- a/drivers/net/netvsc/hn_rxtx.c
+++ b/drivers/net/netvsc/hn_rxtx.c
@@ -1518,14 +1518,32 @@ hn_xmit_pkts(void *ptxq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 
 	for (nb_tx = 0; nb_tx < nb_pkts; nb_tx++) {
 		struct rte_mbuf *m = tx_pkts[nb_tx];
-		uint32_t pkt_size = m->pkt_len + HN_RNDIS_PKT_LEN;
 		struct rndis_packet_msg *pkt;
 		struct hn_txdesc *txd;
+		uint32_t pkt_size;
 
 		txd = hn_txd_get(txq);
 		if (txd == NULL)
 			break;
 
+		if (!(m->ol_flags & RTE_MBUF_F_TX_VLAN)) {
+			struct rte_ether_hdr *eh =
+				rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
+			struct rte_vlan_hdr *vh;
+
+			/* Force TX vlan offloading for 801.2Q packet */
+			if (eh->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_VLAN)) {
+				vh = (struct rte_vlan_hdr *)(eh + 1);
+				m->ol_flags |= RTE_MBUF_F_TX_VLAN;
+				m->vlan_tci = rte_be_to_cpu_16(vh->vlan_tci);
+
+				/* Copy ether header over */
+				memmove(rte_pktmbuf_adj(m, sizeof(struct rte_vlan_hdr)),
+					eh, 2 * RTE_ETHER_ADDR_LEN);
+			}
+		}
+		pkt_size = m->pkt_len + HN_RNDIS_PKT_LEN;
+
 		/* For small packets aggregate them in chimney buffer */
 		if (m->pkt_len <= hv->tx_copybreak &&
 		    pkt_size <= txq->agg_szmax) {
-- 
2.25.1

Re: [PATCH] net/netvsc: force TX VLAN offload on 801.2Q packet

[Stephen Hemminger]

On Fri, 18 Oct 2024 11:13:50 -0700
longli@linuxonhyperv.com wrote:

> icrosoft.com>
> 
> The VSP assumes the packet doesn't have VLAN tags. When VLAN tag is
> present in a TX packet, always strip it and use PPI to send VLAN info
> through VSP packet.
> 
> Fixes: 4e9c73e96e83 ("net/netvsc: add Hyper-V network device")
> Cc: Hui (Hui) Ling <hling@SonicWall.com>
> Cc: Stephen Hemminger <stephen@networkplumber.org>
> Cc: Wei Hu <weh@microsoft.com>
> Cc: stable@dpdk.org
> Signed-off-by: Long Li <longli@microsoft.com>

This was discovered from an network application that was using VLAN's
but not setting the offload bits in the mbuf.


Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>

Re: [PATCH] net/netvsc: force TX VLAN offload on 801.2Q packet

[Ferruh Yigit]

On 10/19/2024 4:32 PM, Stephen Hemminger wrote:
> On Fri, 18 Oct 2024 11:13:50 -0700
> longli@linuxonhyperv.com wrote:
> 
>> icrosoft.com>
>>
>> The VSP assumes the packet doesn't have VLAN tags. When VLAN tag is
>> present in a TX packet, always strip it and use PPI to send VLAN info
>> through VSP packet.
>>
>> Fixes: 4e9c73e96e83 ("net/netvsc: add Hyper-V network device")
>> Cc: Hui (Hui) Ling <hling@SonicWall.com>
>> Cc: Stephen Hemminger <stephen@networkplumber.org>
>> Cc: Wei Hu <weh@microsoft.com>
>> Cc: stable@dpdk.org
>> Signed-off-by: Long Li <longli@microsoft.com>
> 
> This was discovered from an network application that was using VLAN's
> but not setting the offload bits in the mbuf.
> 
> 
> Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
>

Applied to dpdk-next-net/main, thanks.