Skip to contentSkip to conversation

gve: ignore nonrelevant GSO type bits when processing TSO headers

PATCH Superseded [email protected] 6 participants 11 messages Jun 6, 2024 – Jun 12, 2024

Subsystems (3)
lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

PATCH 1/1 gve: ignore nonrelevant GSO type bits when processing TSO headers

From: Joshua Washington <[email protected]>

TSO currently fails when the skb's gso_type field has more than one bit set.

TSO packets can be passed from userspace using PF_PACKET, TUNTAP and a few others, using virtio_net_hdr (e.g., PACKET_VNET_HDR). This includes virtualization, such as QEMU, a real use-case.

The gso_type and gso_size fields as passed from userspace in virtio_net_hdr are not trusted blindly by the kernel. It adds gso_type |= SKB_GSO_DODGY to force the packet to enter the software GSO stack for verification.

This issue might similarly come up when the CWR bit is set in the TCP header for congestion control, causing the SKB_GSO_TCP_ECN gso_type bit to be set.

Fixes a57e5de476be ("gve: DQO: Add TX path")
---
 drivers/net/ethernet/google/gve/gve_tx_dqo.c | 18 +++++++-----------
 1 file changed, 7 insertions(+), 11 deletions(-)
drivers/net/ethernet/google/gve/gve_tx_dqo.c +7 -11
index fe1b26a4d736..04cb43a97c96 100644
@@ -555,6 +555,10 @@ static int gve_prep_tso(struct sk_buff *skb)
 	if (unlikely(skb_shinfo(skb)->gso_size < GVE_TX_MIN_TSO_MSS_DQO))
 		return -1;
 
+	/* We only deal with TCP at this point. */
+	if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
+		return -EINVAL;
+
 	/* Needed because we will modify header. */
 	err = skb_cow_head(skb, 0);
 	if (err < 0)
@@ -565,18 +569,10 @@ static int gve_prep_tso(struct sk_buff *skb)
 	/* Remove payload length from checksum. */
 	paylen = skb->len - skb_transport_offset(skb);
 
-	switch (skb_shinfo(skb)->gso_type) {
-	case SKB_GSO_TCPV4:
-	case SKB_GSO_TCPV6:
-		csum_replace_by_diff(&tcp->check,
-				     (__force __wsum)htonl(paylen));
+	csum_replace_by_diff(&tcp->check, (__force __wsum)htonl(paylen));
 
-		/* Compute length of segmentation header. */
-		header_len = skb_tcp_all_headers(skb);
-		break;
-	default:
-		return -EINVAL;
-	}
+	/* Compute length of segmentation header. */
+	header_len = skb_tcp_all_headers(skb);
 
 	if (unlikely(header_len > GVE_TX_MAX_HDR_SIZE_DQO))
 		return -EINVAL;
Signature · 1 line
2.45.1.288.g0e0cd299f1-goog
Reviewed-by on v1 1/1
#1 lore raw
Quoted 14 lines · joshwash@ wrote:

From: Joshua Washington <[email protected]>

TSO currently fails when the skb's gso_type field has more than one bit set.

TSO packets can be passed from userspace using PF_PACKET, TUNTAP and a few others, using virtio_net_hdr (e.g., PACKET_VNET_HDR). This includes virtualization, such as QEMU, a real use-case.

The gso_type and gso_size fields as passed from userspace in virtio_net_hdr are not trusted blindly by the kernel. It adds gso_type |= SKB_GSO_DODGY to force the packet to enter the software GSO stack for verification.

Quoted 5 lines

This issue might similarly come up when the CWR bit is set in the TCP header for congestion control, causing the SKB_GSO_TCP_ECN gso_type bit to be set.

Fixes a57e5de476be ("gve: DQO: Add TX path")

nit: no empty line

Quoted 4 lines
Quoted 9 lines
---
 drivers/net/ethernet/google/gve/gve_tx_dqo.c | 18 +++++++-----------
 1 file changed, 7 insertions(+), 11 deletions(-)
drivers/net/ethernet/google/gve/gve_tx_dqo.c +0 -0
index fe1b26a4d736..04cb43a97c96 100644

drivers/net/ethernet/google/gve/gve_tx_dqo.c static int gve_prep_tso(struct sk_buff *skb) · line 561

Quoted 7 lines
drivers/net/ethernet/google/gve/gve_tx_dqo.c +4 -0
@@ -555,6 +555,10 @@ static int gve_prep_tso(struct sk_buff *skb)
 	if (unlikely(skb_shinfo(skb)->gso_size < GVE_TX_MIN_TSO_MSS_DQO))
 		return -1;
 
+	/* We only deal with TCP at this point. */
+	if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
+		return -EINVAL;
+

NETIF_F_TSO and NETIF_F_TSO6 are the only terminal/L4 segmentation offload types that gve advertises in hw_features. So I think that this will always be true.

If nothing else, it documents the assumption, so fine to keep.

Careful about comments that just repeat what the code does. More informative are comments that why non-obvious code exists (where applicable, which is not here).

Quoted 28 lines
drivers/net/ethernet/google/gve/gve_tx_dqo.c +3 -11
 	/* Needed because we will modify header. */
 	err = skb_cow_head(skb, 0);
 	if (err < 0)
@@ -565,18 +569,10 @@ static int gve_prep_tso(struct sk_buff *skb)
 	/* Remove payload length from checksum. */
 	paylen = skb->len - skb_transport_offset(skb);
 
-	switch (skb_shinfo(skb)->gso_type) {
-	case SKB_GSO_TCPV4:
-	case SKB_GSO_TCPV6:
-		csum_replace_by_diff(&tcp->check,
-				     (__force __wsum)htonl(paylen));
+	csum_replace_by_diff(&tcp->check, (__force __wsum)htonl(paylen));
 
-		/* Compute length of segmentation header. */
-		header_len = skb_tcp_all_headers(skb);
-		break;
-	default:
-		return -EINVAL;
-	}
+	/* Compute length of segmentation header. */
+	header_len = skb_tcp_all_headers(skb);
 
 	if (unlikely(header_len > GVE_TX_MAX_HDR_SIZE_DQO))
 		return -EINVAL;
Signature · 1 line
2.45.1.288.g0e0cd299f1-goog
Acked-by on v1 1/1
#2 lore raw
In reply to [email protected]
Quoted 9 lines · On Thu, Jun 6, 2024 at 12:22 PM <[email protected]> wrote:

From: Joshua Washington <[email protected]>

TSO currently fails when the skb's gso_type field has more than one bit set.

TSO packets can be passed from userspace using PF_PACKET, TUNTAP and a few others, using virtio_net_hdr (e.g., PACKET_VNET_HDR). This includes virtualization, such as QEMU, a real use-case.

Here is the bug report where this issue was triggered by gVisor:
https://github.com/google/gvisor/issues/10344

Quoted 10 lines

The gso_type and gso_size fields as passed from userspace in virtio_net_hdr are not trusted blindly by the kernel. It adds gso_type |= SKB_GSO_DODGY to force the packet to enter the software GSO stack for verification.

This issue might similarly come up when the CWR bit is set in the TCP header for congestion control, causing the SKB_GSO_TCP_ECN gso_type bit to be set.

Quoted 6 lines
Fixes a57e5de476be ("gve: DQO: Add TX path")

Thanks,
Andrei

author maintainer
#3 lore raw

PATCH v2 1/1 gve: ignore nonrelevant GSO type bits when processing TSO headers

In reply to [email protected]

From: Joshua Washington <[email protected]>

TSO currently fails when the skb's gso_type field has more than one bit set.

TSO packets can be passed from userspace using PF_PACKET, TUNTAP and a few others, using virtio_net_hdr (e.g., PACKET_VNET_HDR). This includes virtualization, such as QEMU, a real use-case.

The gso_type and gso_size fields as passed from userspace in virtio_net_hdr are not trusted blindly by the kernel. It adds gso_type |= SKB_GSO_DODGY to force the packet to enter the software GSO stack for verification.

This issue might similarly come up when the CWR bit is set in the TCP header for congestion control, causing the SKB_GSO_TCP_ECN gso_type bit to be set.

---
 drivers/net/ethernet/google/gve/gve_tx_dqo.c | 21 +++++---------------
 1 file changed, 5 insertions(+), 16 deletions(-)
drivers/net/ethernet/google/gve/gve_tx_dqo.c +5 -16
index fe1b26a4d736..a76b407a981b 100644
@@ -551,32 +551,21 @@ static int gve_prep_tso(struct sk_buff *skb)
 	 * - Hypervisor enforces a limit of 9K MTU
 	 * - Kernel will not produce a TSO larger than 64k
 	 */
-
 	if (unlikely(skb_shinfo(skb)->gso_size < GVE_TX_MIN_TSO_MSS_DQO))
 		return -1;
 
+	if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
+		return -EINVAL;
+
 	/* Needed because we will modify header. */
 	err = skb_cow_head(skb, 0);
 	if (err < 0)
 		return err;
 
 	tcp = tcp_hdr(skb);
-
-	/* Remove payload length from checksum. */
 	paylen = skb->len - skb_transport_offset(skb);
-
-	switch (skb_shinfo(skb)->gso_type) {
-	case SKB_GSO_TCPV4:
-	case SKB_GSO_TCPV6:
-		csum_replace_by_diff(&tcp->check,
-				     (__force __wsum)htonl(paylen));
-
-		/* Compute length of segmentation header. */
-		header_len = skb_tcp_all_headers(skb);
-		break;
-	default:
-		return -EINVAL;
-	}
+	csum_replace_by_diff(&tcp->check, (__force __wsum)htonl(paylen));
+	header_len = skb_tcp_all_headers(skb);
 
 	if (unlikely(header_len > GVE_TX_MAX_HDR_SIZE_DQO))
 		return -EINVAL;
Signature · 1 line
2.45.2.505.gda0bf45e8d-goog
Reviewed-by on v2 1/1
#4 lore raw
Quoted 19 lines · On Fri, Jun 7, 2024 at 8:10 AM <[email protected]> wrote:

From: Joshua Washington <[email protected]>

TSO currently fails when the skb's gso_type field has more than one bit set.

TSO packets can be passed from userspace using PF_PACKET, TUNTAP and a few others, using virtio_net_hdr (e.g., PACKET_VNET_HDR). This includes virtualization, such as QEMU, a real use-case.

The gso_type and gso_size fields as passed from userspace in virtio_net_hdr are not trusted blindly by the kernel. It adds gso_type |= SKB_GSO_DODGY to force the packet to enter the software GSO stack for verification.

This issue might similarly come up when the CWR bit is set in the TCP header for congestion control, causing the SKB_GSO_TCP_ECN gso_type bit to be set.

Quoted 7 lines
#5 lore raw
In reply to [email protected]
Quoted 18 lines · joshwash@ wrote:

From: Joshua Washington <[email protected]>

TSO currently fails when the skb's gso_type field has more than one bit set.

TSO packets can be passed from userspace using PF_PACKET, TUNTAP and a few others, using virtio_net_hdr (e.g., PACKET_VNET_HDR). This includes virtualization, such as QEMU, a real use-case.

The gso_type and gso_size fields as passed from userspace in virtio_net_hdr are not trusted blindly by the kernel. It adds gso_type |= SKB_GSO_DODGY to force the packet to enter the software GSO stack for verification.

This issue might similarly come up when the CWR bit is set in the TCP header for congestion control, causing the SKB_GSO_TCP_ECN gso_type bit to be set.

Quoted 7 lines

I did not mean to ask for a revision. When you send a v2, please do include a changelog

drivers/net/ethernet/google/gve/gve_tx_dqo.c static int gve_prep_tso(struct sk_buff *skb) · line 554

Quoted 13 lines
---
 drivers/net/ethernet/google/gve/gve_tx_dqo.c | 21 +++++---------------
 1 file changed, 5 insertions(+), 16 deletions(-)
drivers/net/ethernet/google/gve/gve_tx_dqo.c +0 -1
index fe1b26a4d736..a76b407a981b 100644
@@ -551,32 +551,21 @@ static int gve_prep_tso(struct sk_buff *skb)
 	 * - Hypervisor enforces a limit of 9K MTU
 	 * - Kernel will not produce a TSO larger than 64k
 	 */
-

Accidental removal?

Quoted 6 lines
drivers/net/ethernet/google/gve/gve_tx_dqo.c +3 -0
 	if (unlikely(skb_shinfo(skb)->gso_size < GVE_TX_MIN_TSO_MSS_DQO))
 		return -1;
 
+	if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
+		return -EINVAL;
+
author maintainer
#6 lore raw

PATCH v3 1/1 gve: ignore nonrelevant GSO type bits when processing TSO headers

In reply to [email protected]

From: Joshua Washington <[email protected]>

TSO currently fails when the skb's gso_type field has more than one bit set.

TSO packets can be passed from userspace using PF_PACKET, TUNTAP and a few others, using virtio_net_hdr (e.g., PACKET_VNET_HDR). This includes virtualization, such as QEMU, a real use-case.

The gso_type and gso_size fields as passed from userspace in virtio_net_hdr are not trusted blindly by the kernel. It adds gso_type |= SKB_GSO_DODGY to force the packet to enter the software GSO stack for verification.

This issue might similarly come up when the CWR bit is set in the TCP header for congestion control, causing the SKB_GSO_TCP_ECN gso_type bit to be set.

v2 - Remove unnecessary comments, remove line break between fixes tag and signoffs.

v3 - Add back unrelated empty line removal.

---
 drivers/net/ethernet/google/gve/gve_tx_dqo.c | 20 +++++---------------
 1 file changed, 5 insertions(+), 15 deletions(-)
drivers/net/ethernet/google/gve/gve_tx_dqo.c +5 -15
index fe1b26a4d736..0b3cca3fc792 100644
@@ -555,28 +555,18 @@ static int gve_prep_tso(struct sk_buff *skb)
 	if (unlikely(skb_shinfo(skb)->gso_size < GVE_TX_MIN_TSO_MSS_DQO))
 		return -1;
 
+	if (!(skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
+		return -EINVAL;
+
 	/* Needed because we will modify header. */
 	err = skb_cow_head(skb, 0);
 	if (err < 0)
 		return err;
 
 	tcp = tcp_hdr(skb);
-
-	/* Remove payload length from checksum. */
 	paylen = skb->len - skb_transport_offset(skb);
-
-	switch (skb_shinfo(skb)->gso_type) {
-	case SKB_GSO_TCPV4:
-	case SKB_GSO_TCPV6:
-		csum_replace_by_diff(&tcp->check,
-				     (__force __wsum)htonl(paylen));
-
-		/* Compute length of segmentation header. */
-		header_len = skb_tcp_all_headers(skb);
-		break;
-	default:
-		return -EINVAL;
-	}
+	csum_replace_by_diff(&tcp->check, (__force __wsum)htonl(paylen));
+	header_len = skb_tcp_all_headers(skb);
 
 	if (unlikely(header_len > GVE_TX_MAX_HDR_SIZE_DQO))
 		return -EINVAL;
Signature · 1 line
2.45.2.505.gda0bf45e8d-goog
author maintainer
#8 lore raw

My apologies. I'll send an updated patch tomorrow without --in-reply-to.

Quoted 10 lines · Jakub Kicinski
Quoted 4 lines · On Mon, 10 Jun 2024 15:57:18 -0700 [email protected] wrote:

v2 - Remove unnecessary comments, remove line break between fixes tag and signoffs.

v3 - Add back unrelated empty line removal.

Read the maintainer info again, please:
https://www.kernel.org/doc/html/next/process/maintainer-netdev.html
we prefer no in-reply to postings.

Signature · 1 line
Joshua Washington | Software Engineer | [email protected] | (414) 366-4423
maintainer
#9 lore raw
Quoted 1 line · On Mon, 10 Jun 2024 19:26:32 -0700 Joshua Washington wrote:

My apologies. I'll send an updated patch tomorrow without --in-reply-to.

No need, it's still in patchwork, it was just a note for the future. I should have made that more clear, I realized that after hitting send.

#10 lore raw
In reply to [email protected]
This patch was applied to netdev/net.git (main) by Jakub Kicinski <[email protected]>:

Hello:

This patch was applied to netdev/net.git (main)
by Jakub Kicinski <[email protected]>:

Quoted 10 lines · On Mon, 10 Jun 2024 15:57:18 -0700 you wrote:

From: Joshua Washington <[email protected]>

TSO currently fails when the skb's gso_type field has more than one bit set.

TSO packets can be passed from userspace using PF_PACKET, TUNTAP and a few others, using virtio_net_hdr (e.g., PACKET_VNET_HDR). This includes virtualization, such as QEMU, a real use-case.

[...]

Here is the summary with links:
- [net,v3] gve: ignore nonrelevant GSO type bits when processing TSO headers

    https://git.kernel.org/netdev/net/c/1b9f75634441

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html