Skip to contentSkip to conversation

Fix issues with Heper-V network offload code

PATCH In review K. Y. Srinivasan 2 participants 10 messages Apr 8, 2014 – Apr 9, 2014

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover Fix issues with Heper-V network offload code

WS2008 R2 does not support udp checksum offload. Furthermore, ws2012 and ws2012 r2 have issues offloading udp checksum from Linux guests. This patch-set addresses these issues as well as other bug fixes. Please apply.

K. Y. Srinivasan (3):
Drivers: net: hyperv: Allocate memory for all possible per-pecket
information
Drivers: net: hyperv: Negotiate suitable ndis version for offload
support
Drivers: net: hyperv: Address UDP checksum issues

 drivers/net/hyperv/hyperv_net.h   |    1 +
 drivers/net/hyperv/netvsc.c       |    2 +-
 drivers/net/hyperv/netvsc_drv.c   |   23 +++++++++++++++++++++--
 drivers/net/hyperv/rndis_filter.c |   12 +++++++++++-
 4 files changed, 34 insertions(+), 4 deletions(-)
Signature · 1 line
1.7.4.1
#1 lore raw

cover Fix issues with Heper-V network offload code

WS2008 R2 does not support udp checksum offload. Furthermore, ws2012 and ws2012 r2 have issues offloading udp checksum from Linux guests. This patch-set addresses these issues as well as other bug fixes. Please apply.

K. Y. Srinivasan (3):
Drivers: net: hyperv: Allocate memory for all possible per-pecket
information
Drivers: net: hyperv: Negotiate suitable ndis version for offload
support
Drivers: net: hyperv: Address UDP checksum issues

 drivers/net/hyperv/hyperv_net.h   |    1 +
 drivers/net/hyperv/netvsc.c       |    2 +-
 drivers/net/hyperv/netvsc_drv.c   |   23 +++++++++++++++++++++--
 drivers/net/hyperv/rndis_filter.c |   12 +++++++++++-
 4 files changed, 34 insertions(+), 4 deletions(-)
Signature · 1 line
1.7.4.1
#3 lore raw

PATCH 2/3 Drivers: net: hyperv: Negotiate suitable ndis version for offload support

Ws2008R2 supports ndis_version 6.1 and 6.1 is the minimal version required for various offloads. Negotiate ndis_version 6.1 when on ws2008r2.

---
 drivers/net/hyperv/netvsc.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/net/hyperv/netvsc.c +1 -1
index daddea2..f7629ec 100644
@@ -344,7 +344,7 @@ static int netvsc_connect_vsp(struct hv_device *device)
 	memset(init_packet, 0, sizeof(struct nvsp_message));
 
 	if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
-		ndis_version = 0x00050001;
+		ndis_version = 0x00060001;
 	else
 		ndis_version = 0x0006001e;
 
Signature · 1 line
1.7.4.1
#4 lore raw

PATCH 3/3 Drivers: net: hyperv: Address UDP checksum issues

In reply to K. Y. Srinivasan

ws2008r2 does not support UDP checksum offload. Thus, we cannnot turn on UDP offload in the host. Also, on ws2012 and ws2012 r2, there appear to be an issue with UDP checksum offload.
Fix this issue by computing the UDP checksum in the Hyper-V driver.

---
 drivers/net/hyperv/hyperv_net.h   |    1 +
 drivers/net/hyperv/netvsc_drv.c   |   19 ++++++++++++++++++-
 drivers/net/hyperv/rndis_filter.c |   12 +++++++++++-
 3 files changed, 30 insertions(+), 2 deletions(-)
drivers/net/hyperv/hyperv_net.h +1 -0
index 13010b4..d18f711d 100644
@@ -747,6 +747,7 @@ struct ndis_oject_header {
 #define NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4	0
 #define NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6	1
 
+#define VERSION_4_OFFLOAD_SIZE			22
 /*
  * New offload OIDs for NDIS 6
  */
drivers/net/hyperv/netvsc_drv.c +18 -1
index 6f39baa..8e3919b 100644
@@ -398,7 +398,24 @@ static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 		csum_info->transmit.tcp_checksum = 1;
 		csum_info->transmit.tcp_header_offset = hdr_offset;
 	} else if (net_trans_info & INFO_UDP) {
-		csum_info->transmit.udp_checksum = 1;
+		/* UDP checksum offload is not supported on ws2008r2.
+		 * Furthermore, on ws2012 and ws2012r2, there are some
+		 * issues with udp checksum offload from Linux guests.
+		 * (these are host issues).
+		 * For now compute the checksum here.
+		 */
+		struct udphdr *uh = udp_hdr(skb);
+		u16 udp_len = ntohs(uh->len);
+
+		uh->check = 0;
+		uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
+					      ip_hdr(skb)->daddr,
+					      udp_len, IPPROTO_UDP,
+					      csum_partial(uh, udp_len, 0));
+		if (uh->check == 0)
+			uh->check = CSUM_MANGLED_0;
+
+		csum_info->transmit.udp_checksum = 0;
 	}
 	goto do_send;
 
drivers/net/hyperv/rndis_filter.c +11 -1
index 4a37e3d..143a98c 100644
@@ -641,6 +641,16 @@ int rndis_filter_set_offload_params(struct hv_device *hdev,
 	struct rndis_set_complete *set_complete;
 	u32 extlen = sizeof(struct ndis_offload_params);
 	int ret, t;
+	u32 vsp_version = nvdev->nvsp_version;
+
+	if (vsp_version <= NVSP_PROTOCOL_VERSION_4) {
+		extlen = VERSION_4_OFFLOAD_SIZE;
+		/* On NVSP_PROTOCOL_VERSION_4 and below, we do not support
+		 * UDP checksum offload.
+		 */
+		req_offloads->udp_ip_v4_csum = 0;
+		req_offloads->udp_ip_v6_csum = 0;
+	}
 
 	request = get_rndis_request(rdev, RNDIS_MSG_SET,
 		RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
@@ -674,7 +684,7 @@ int rndis_filter_set_offload_params(struct hv_device *hdev,
 	} else {
 		set_complete = &request->response_msg.msg.set_complete;
 		if (set_complete->status != RNDIS_STATUS_SUCCESS) {
-			netdev_err(ndev, "Fail to set MAC on host side:0x%x\n",
+			netdev_err(ndev, "Fail to set offload on host side:0x%x\n",
 				   set_complete->status);
 			ret = -EINVAL;
 		}
Signature · 1 line
1.7.4.1
#2 lore raw

PATCH 1/3 Drivers: net: hyperv: Allocate memory for all possible per-pecket information

In reply to K. Y. Srinivasan

An outgoing packet can potentially need per-packet information for all the offloads and VLAN tagging. Fix this issue.

---
 drivers/net/hyperv/netvsc_drv.c |    4 +++-
 1 files changed, 3 insertions(+), 1 deletions(-)
drivers/net/hyperv/netvsc_drv.c +3 -1
index 4e4cf9e..6f39baa 100644
@@ -319,7 +319,9 @@ static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 	packet = kzalloc(sizeof(struct hv_netvsc_packet) +
 			 (num_data_pgs * sizeof(struct hv_page_buffer)) +
 			 sizeof(struct rndis_message) +
-			 NDIS_VLAN_PPI_SIZE, GFP_ATOMIC);
+			 NDIS_VLAN_PPI_SIZE +
+			 NDIS_CSUM_PPI_SIZE +
+			 NDIS_LSO_PPI_SIZE, GFP_ATOMIC);
 	if (!packet) {
 		/* out of memory, drop packet */
 		netdev_err(net, "unable to allocate hv_netvsc_packet\n");
Signature · 1 line
1.7.4.1
maintainer
#5 lore raw

Re: [PATCH net 3/3] Drivers: net: hyperv: Address UDP checksum issues

In reply to K. Y. Srinivasan

From: "K. Y. Srinivasan" <[email protected]>
Date: Tue, 8 Apr 2014 12:46:11 -0700

Quoted 10 lines
drivers/net/hyperv/netvsc_drv.c +8 -1
 	} else if (net_trans_info & INFO_UDP) {
-		csum_info->transmit.udp_checksum = 1;
+		/* UDP checksum offload is not supported on ws2008r2.
+		 * Furthermore, on ws2012 and ws2012r2, there are some
+		 * issues with udp checksum offload from Linux guests.
+		 * (these are host issues).
+		 * For now compute the checksum here.
+		 */
+		struct udphdr *uh = udp_hdr(skb);
+		u16 udp_len = ntohs(uh->len);

drivers/net/hyperv/netvsc_drv.c static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net) · line 418

Quoted 10 lines
drivers/net/hyperv/netvsc_drv.c +10 -0
+
+		uh->check = 0;
+		uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
+					      ip_hdr(skb)->daddr,
+					      udp_len, IPPROTO_UDP,
+					      csum_partial(uh, udp_len, 0));
+		if (uh->check == 0)
+			uh->check = CSUM_MANGLED_0;
+
+		csum_info->transmit.udp_checksum = 0;

You absolutely cannot mangle packet header contents without first COW'ing the SKB to make sure it's writable.

Otherwise network taps like tcpdump, and any other entity with a reference to the packet, can see inconsistent state.

#6 lore raw

RE: [PATCH net 3/3] Drivers: net: hyperv: Address UDP checksum issues

Quoted 39 lines

-----Original Message-----
From: David Miller [mailto:[email protected]]
Sent: Wednesday, April 9, 2014 10:01 AM
To: KY Srinivasan
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH net 3/3] Drivers: net: hyperv: Address UDP checksum issues

From: "K. Y. Srinivasan" <[email protected]>
Date: Tue, 8 Apr 2014 12:46:11 -0700

Quoted 10 lines
drivers/net/hyperv/netvsc_drv.c +8 -1
 	} else if (net_trans_info & INFO_UDP) {
-		csum_info->transmit.udp_checksum = 1;
+		/* UDP checksum offload is not supported on ws2008r2.
+		 * Furthermore, on ws2012 and ws2012r2, there are some
+		 * issues with udp checksum offload from Linux guests.
+		 * (these are host issues).
+		 * For now compute the checksum here.
+		 */
+		struct udphdr *uh = udp_hdr(skb);
+		u16 udp_len = ntohs(uh->len);

drivers/net/hyperv/netvsc_drv.c static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net) · line 418

Quoted 10 lines
drivers/net/hyperv/netvsc_drv.c +10 -0
+
+		uh->check = 0;
+		uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
+					      ip_hdr(skb)->daddr,
+					      udp_len, IPPROTO_UDP,
+					      csum_partial(uh, udp_len, 0));
+		if (uh->check == 0)
+			uh->check = CSUM_MANGLED_0;
+
+		csum_info->transmit.udp_checksum = 0;

You absolutely cannot mangle packet header contents without first COW'ing the SKB to make sure it's writable.

Quoted 2 lines

Otherwise network taps like tcpdump, and any other entity with a reference to the packet, can see inconsistent state.

Thanks Dave. I will COW the SKB before changing the checksum value. I do have a question though. I looked at a bunch of hardware drivers and they modify the header information, specially checksum field and I could not see where they had COWed the skb. I was grepping for skb_cow in these drivers.

Regards,

K. Y

maintainer
#7 lore raw

Re: [PATCH net 3/3] Drivers: net: hyperv: Address UDP checksum issues

From: KY Srinivasan <[email protected]>
Date: Wed, 9 Apr 2014 18:08:47 +0000

Quoted 5 lines

Thanks Dave. I will COW the SKB before changing the checksum value. I do have a question though. I looked at a bunch of hardware drivers and they modify the header information, specially checksum field and I could not see where they had COWed the skb. I was grepping for skb_cow in these drivers.

Look for skb_cow_head(), for example, the tg3.c driver uses this.

#8 lore raw

RE: [PATCH net 3/3] Drivers: net: hyperv: Address UDP checksum issues

Quoted 15 lines

-----Original Message-----
From: David Miller [mailto:[email protected]]
Sent: Wednesday, April 9, 2014 11:37 AM
To: KY Srinivasan
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH net 3/3] Drivers: net: hyperv: Address UDP checksum issues

From: KY Srinivasan <[email protected]>
Date: Wed, 9 Apr 2014 18:08:47 +0000

Quoted 8 lines
Quoted 5 lines

Thanks Dave. I will COW the SKB before changing the checksum value. I do have a question though. I looked at a bunch of hardware drivers and they modify the header information, specially checksum field and I could not see where they had COWed the skb. I was grepping for skb_cow in these drivers.

Look for skb_cow_head(), for example, the tg3.c driver uses this.

Thank you. I will update this patch. Do you want me to resend the whole set or just this patch.

K. Y

maintainer
#9 lore raw

Re: [PATCH net 3/3] Drivers: net: hyperv: Address UDP checksum issues

From: KY Srinivasan <[email protected]>
Date: Wed, 9 Apr 2014 18:39:57 +0000

Quoted 21 lines
Quoted 21 lines

-----Original Message-----
From: David Miller [mailto:[email protected]]
Sent: Wednesday, April 9, 2014 11:37 AM
To: KY Srinivasan
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCH net 3/3] Drivers: net: hyperv: Address UDP checksum issues

From: KY Srinivasan <[email protected]>
Date: Wed, 9 Apr 2014 18:08:47 +0000

Quoted 5 lines

Thanks Dave. I will COW the SKB before changing the checksum value. I do have a question though. I looked at a bunch of hardware drivers and they modify the header information, specially checksum field and I could not see where they had COWed the skb. I was grepping for skb_cow in these drivers.

Quoted 4 lines
Quoted 2 lines

Look for skb_cow_head(), for example, the tg3.c driver uses this.

Thank you. I will update this patch. Do you want me to resend the whole set or just this patch.

You never need to ask me this.

Always, when there is an update to any part of a series, re-submit the entire series.