netdev - [PATCH v2] net: Make skb_segment not to compute checksum if network controller supports checksumming

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <1583131910-29260-1-git-send-email-kyk.segfault@gmail.com>
Date:   Mon,  2 Mar 2020 12:21:50 +0530
From:   Yadu Kishore <kyk.segfault@...il.com>
To:     davem@...emloft.net
Cc:     willemdebruijn.kernel@...il.com, netdev@...r.kernel.org,
        Yadu Kishore <kyk.segfault@...il.com>
Subject: [PATCH v2] net: Make skb_segment not to compute checksum if network controller supports checksumming

> > Can you contrast this against a run with your changes? The thought is
> > that the majority of this cost is due to the memory loads and stores, not
> > the arithmetic ops to compute the checksum. When enabling checksum
> > offload, the same stalls will occur, but will simply be attributed to
> > memcpy instead of to do_csum.

> Agreed.

Below is the data from perf with and without the patch for the same
TCP Tx iperf run: (network driver has NETIF_F_HW_CSUM enabled)

Without patch :-
============
[Function = %cpu cycles]
skb_mac_gso_segment = 0.05
inet_gso_segment = 0.26
tcp4_gso_segment = 0.05
tcp_gso_segment = 0.17
skb_segment = 0.55
skb_copy_and_csum_bits = 0.60
do_csum = 7.43
memcpy = 3.81
__alloc_skb = 0.93
==================
SUM = 13.85

With patch :-
============
[Function = %cpu cycles]
skb_mac_gso_segment = 0.05
inet_gso_segment = 0.34
tcp4_gso_segment = 0.06
tcp_gso_segment = 0.26
skb_segment = 0.55
skb_copy_and_csum_bits = 0.00
do_csum = 0.04
memcpy = 4.29
__alloc_skb = 0.73
==================
SUM = 6.32

So, with the patch, from the above data we can see
that the percentage of cpu cycles spent in do_csum
has come down from 7.43% to 0.04%.

> > > Is this not already handled by __copy_skb_header above? If ip_summed
> > > has to be initialized, so have csum_start and csum_offset. That call
> > > should have initialized all three.

> > Thanks, I will look into why even though __copy_skb_header is being
> > called, I am still
> > seeing skb->ip_summed set to CHECKSUM_NONE in the network driver.

> Thanks.

My mistake. I had removed the 'skb->ip_summed = CHECKSUM_PARTIAL' line
from the patch but had forgotten to enable NETIF_F_HW_CSUM in the network
driver. Hence I was still seeing skb->ip_summed set to CHECKSUM_NONE.
After re-enabling NETIF_F_HW_CSUM in the driver, I now see that
skb->ip_summed is being set correctly to CHECKSUM_PARTIAL.
So as suggested, the __copy_skb_header is indeed taking care of doing
this and hence there is no need to explicitly set 'ip_summed' in the patch.
Below is V2 of the patch with the changes.

Problem:
TCP checksum in the output path is not being offloaded during GSO
in the following case:
The network driver does not support scatter-gather but supports
checksum offload with NETIF_F_HW_CSUM.

Cause:
skb_segment calls skb_copy_and_csum_bits if the network driver
does not announce NETIF_F_SG. It does not check if the driver
supports NETIF_F_HW_CSUM.
So for devices which might want to offload checksum but do not support SG
there is currently no way to do so if GSO is enabled.

Solution:
In skb_segment check if the network controller does checksum and if so
call skb_copy_bits instead of skb_copy_and_csum_bits.

Testing:
Without the patch, ran iperf TCP traffic with NETIF_F_HW_CSUM enabled
in the network driver. Observed the TCP checksum offload is not happening
since the skbs received by the driver in the output path have
skb->ip_summed set to CHECKSUM_NONE.

With the patch ran iperf TCP traffic and observed that TCP checksum
is being offloaded with skb->ip_summed set to CHECKSUM_PARTIAL.
Also tested with the patch by disabling NETIF_F_HW_CSUM in the driver
to cover the newly introduced if-else code path in skb_segment.

Link: https://lore.kernel.org/netdev/CA+FuTSeYGYr3Umij+Mezk9CUcaxYwqEe5sPSuXF8jPE2yMFJAw@mail.gmail.com
Signed-off-by: Yadu Kishore <kyk.segfault@...il.com>
---
 net/core/skbuff.c | 23 +++++++++++++++--------
 1 file changed, 15 insertions(+), 8 deletions(-)

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 1365a55..eca72bc 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -3926,14 +3926,21 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
 			goto perform_csum_check;

 		if (!sg) {
-			if (!nskb->remcsum_offload)
-				nskb->ip_summed = CHECKSUM_NONE;
-			SKB_GSO_CB(nskb)->csum =
-				skb_copy_and_csum_bits(head_skb, offset,
-						       skb_put(nskb, len),
-						       len, 0);
-			SKB_GSO_CB(nskb)->csum_start =
-				skb_headroom(nskb) + doffset;
+			if (!csum) {
+				if (!nskb->remcsum_offload)
+					nskb->ip_summed = CHECKSUM_NONE;
+				SKB_GSO_CB(nskb)->csum =
+					skb_copy_and_csum_bits(head_skb, offset,
+							       skb_put(nskb,
+								       len),
+							       len, 0);
+				SKB_GSO_CB(nskb)->csum_start =
+					skb_headroom(nskb) + doffset;
+			} else {
+				skb_copy_bits(head_skb, offset,
+					      skb_put(nskb, len),
+					      len);
+			}
 			continue;
 		}

-- 
2.7.4