[dpdk-dev] [PATCH] lib/librte_net: fix bug for checksum calculating

[dpdk-dev] [PATCH] lib/librte_net: fix bug for checksum calculating

[guohongzhi]

1.rte_ipv4_cksum will return invalid checksum value like 0xffff.(rfc1624)
2.__rte_raw_cksum should consider Big Endian.
3.checksum 0x0000 is invalid for udp, but valid for tcp.(rfc768)

Signed-off-by: guohongzhi <guohongzhi1@huawei.com>
---
 lib/librte_net/rte_ip.h | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)

diff --git a/lib/librte_net/rte_ip.h b/lib/librte_net/rte_ip.h
index 1ceb7b7..4177ea7 100644
--- a/lib/librte_net/rte_ip.h
+++ b/lib/librte_net/rte_ip.h
@@ -140,7 +140,11 @@ __rte_raw_cksum(const void *buf, size_t len, uint32_t sum)
 
 	/* if length is in odd bytes */
 	if (len == 1)
+#if (RTE_BYTE_ORDER == RTE_BIG_ENDIAN)
+		sum += *((const uint8_t *)u16_buf) << 8;
+#else
 		sum += *((const uint8_t *)u16_buf);
+#endif
 
 	return sum;
 }
@@ -267,7 +271,7 @@ rte_ipv4_cksum(const struct rte_ipv4_hdr *ipv4_hdr)
 {
 	uint16_t cksum;
 	cksum = rte_raw_cksum(ipv4_hdr, sizeof(struct rte_ipv4_hdr));
-	return (cksum == 0xffff) ? cksum : (uint16_t)~cksum;
+	return (uint16_t)~cksum;
 }
 
 /**
@@ -324,8 +328,7 @@ rte_ipv4_phdr_cksum(const struct rte_ipv4_hdr *ipv4_hdr, uint64_t ol_flags)
  * @param l4_hdr
  *   The pointer to the beginning of the L4 header.
  * @return
- *   The complemented checksum to set in the IP packet
- *   or 0 on error
+ *   The complemented checksum to set in the IP packet.
  */
 static inline uint16_t
 rte_ipv4_udptcp_cksum(const struct rte_ipv4_hdr *ipv4_hdr, const void *l4_hdr)
@@ -344,7 +347,8 @@ rte_ipv4_udptcp_cksum(const struct rte_ipv4_hdr *ipv4_hdr, const void *l4_hdr)
 
 	cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 	cksum = (~cksum) & 0xffff;
-	if (cksum == 0)
+	/* 0x0000 is invalid for udp, but valid for tcp. rfc768 */
+	if (cksum == 0 && ipv4_hdr->next_proto_id == IPPROTO_UDP)
 		cksum = 0xffff;
 
 	return (uint16_t)cksum;
@@ -436,7 +440,8 @@ rte_ipv6_udptcp_cksum(const struct rte_ipv6_hdr *ipv6_hdr, const void *l4_hdr)
 
 	cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
 	cksum = (~cksum) & 0xffff;
-	if (cksum == 0)
+	/* 0x0000 is invalid for udp, but valid for tcp. rfc768 */
+	if (cksum == 0 && ipv6_hdr->proto == IPPROTO_UDP)
 		cksum = 0xffff;
 
 	return (uint16_t)cksum;
-- 
2.21.0.windows.1

Re: [dpdk-dev] [PATCH] lib/librte_net: fix bug for checksum calculating

[Morten Brørup]

> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of guohongzhi
> Sent: Friday, May 15, 2020 3:09 PM
> 
> 1.rte_ipv4_cksum will return invalid checksum value like
> 0xffff.(rfc1624)
> 2.__rte_raw_cksum should consider Big Endian.
> 3.checksum 0x0000 is invalid for udp, but valid for tcp.(rfc768)
> 
> Signed-off-by: guohongzhi <guohongzhi1@huawei.com>
> ---
>  lib/librte_net/rte_ip.h | 15 ++++++++++-----
>  1 file changed, 10 insertions(+), 5 deletions(-)
> 
> diff --git a/lib/librte_net/rte_ip.h b/lib/librte_net/rte_ip.h
> index 1ceb7b7..4177ea7 100644
> --- a/lib/librte_net/rte_ip.h
> +++ b/lib/librte_net/rte_ip.h
> @@ -140,7 +140,11 @@ __rte_raw_cksum(const void *buf, size_t len,
> uint32_t sum)
> 
>  	/* if length is in odd bytes */
>  	if (len == 1)
> +#if (RTE_BYTE_ORDER == RTE_BIG_ENDIAN)
> +		sum += *((const uint8_t *)u16_buf) << 8;
> +#else
>  		sum += *((const uint8_t *)u16_buf);
> +#endif
> 
>  	return sum;
>  }
> @@ -267,7 +271,7 @@ rte_ipv4_cksum(const struct rte_ipv4_hdr *ipv4_hdr)
>  {
>  	uint16_t cksum;
>  	cksum = rte_raw_cksum(ipv4_hdr, sizeof(struct rte_ipv4_hdr));
> -	return (cksum == 0xffff) ? cksum : (uint16_t)~cksum;
> +	return (uint16_t)~cksum;
>  }
> 
>  /**
> @@ -324,8 +328,7 @@ rte_ipv4_phdr_cksum(const struct rte_ipv4_hdr
> *ipv4_hdr, uint64_t ol_flags)
>   * @param l4_hdr
>   *   The pointer to the beginning of the L4 header.
>   * @return
> - *   The complemented checksum to set in the IP packet
> - *   or 0 on error
> + *   The complemented checksum to set in the IP packet.
>   */
>  static inline uint16_t
>  rte_ipv4_udptcp_cksum(const struct rte_ipv4_hdr *ipv4_hdr, const void
> *l4_hdr)
> @@ -344,7 +347,8 @@ rte_ipv4_udptcp_cksum(const struct rte_ipv4_hdr
> *ipv4_hdr, const void *l4_hdr)
> 
>  	cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
>  	cksum = (~cksum) & 0xffff;
> -	if (cksum == 0)
> +	/* 0x0000 is invalid for udp, but valid for tcp. rfc768 */
> +	if (cksum == 0 && ipv4_hdr->next_proto_id == IPPROTO_UDP)
>  		cksum = 0xffff;
> 
>  	return (uint16_t)cksum;
> @@ -436,7 +440,8 @@ rte_ipv6_udptcp_cksum(const struct rte_ipv6_hdr
> *ipv6_hdr, const void *l4_hdr)
> 
>  	cksum = ((cksum & 0xffff0000) >> 16) + (cksum & 0xffff);
>  	cksum = (~cksum) & 0xffff;
> -	if (cksum == 0)
> +	/* 0x0000 is invalid for udp, but valid for tcp. rfc768 */
> +	if (cksum == 0 && ipv6_hdr->proto == IPPROTO_UDP)
>  		cksum = 0xffff;
> 
>  	return (uint16_t)cksum;
> --
> 2.21.0.windows.1
> 
> 

Reviewed-by: Morten Brørup <mb@smartsharesystems.com>

Re: [dpdk-dev] [PATCH] lib/librte_net: fix bug for checksum calculating

[Thomas Monjalon]

15/05/2020 15:08, guohongzhi:
> 1.rte_ipv4_cksum will return invalid checksum value like 0xffff.(rfc1624)
> 2.__rte_raw_cksum should consider Big Endian.
> 3.checksum 0x0000 is invalid for udp, but valid for tcp.(rfc768)

Please consider doing 3 patches, thanks.

It was sent late for 20.05, I prefer not taking any risk,
and wait for maintainers approval.
The patches must be reviewed and discussed in 20.08 timeframe.

Re: [dpdk-dev] [PATCH] lib/librte_net: fix bug for checksum calculating

[Morten Brørup]

> From: Thomas Monjalon [mailto:thomas@monjalon.net]
> Sent: Sunday, May 24, 2020 5:12 PM
> 
> 15/05/2020 15:08, guohongzhi:
> > 1.rte_ipv4_cksum will return invalid checksum value like 0xffff.(rfc1624)
> > 2.__rte_raw_cksum should consider Big Endian.
> > 3.checksum 0x0000 is invalid for udp, but valid for tcp.(rfc768)
> 
> Please consider doing 3 patches, thanks.
> 
> It was sent late for 20.05, I prefer not taking any risk,
> and wait for maintainers approval.
> The patches must be reviewed and discussed in 20.08 timeframe.

This patch fixes a bug, which is incompatible with some very popular Broadcom switch silicon, so the patch should be treated as such.

However, I agree that the review responsibility is not Thomas' alone; maintainers should step up if they consider this bugfix critical.

Re: [dpdk-dev] [PATCH] lib/librte_net: fix bug for checksum calculating

[Thomas Monjalon]

24/05/2020 17:31, Morten Brørup:
> > From: Thomas Monjalon [mailto:thomas@monjalon.net]
> > Sent: Sunday, May 24, 2020 5:12 PM
> > 
> > 15/05/2020 15:08, guohongzhi:
> > > 1.rte_ipv4_cksum will return invalid checksum value like 0xffff.(rfc1624)
> > > 2.__rte_raw_cksum should consider Big Endian.
> > > 3.checksum 0x0000 is invalid for udp, but valid for tcp.(rfc768)
> > 
> > Please consider doing 3 patches, thanks.
> > 
> > It was sent late for 20.05, I prefer not taking any risk,
> > and wait for maintainers approval.
> > The patches must be reviewed and discussed in 20.08 timeframe.
> 
> This patch fixes a bug, which is incompatible with some very popular Broadcom switch silicon, so the patch should be treated as such.
> 
> However, I agree that the review responsibility is not Thomas' alone; maintainers should step up if they consider this bugfix critical.

Unfortunately, the lack of "Signed-off-by" is really blocking the merge.
I am not allowed adding such thing.

Given the bug was already there for a long time,
it is probably not so much critical anyway.

Re: [dpdk-dev] [PATCH] lib/librte_net: fix bug for checksumcalculating

[Morten Brørup]

> From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Thomas Monjalon
> Sent: Sunday, May 24, 2020 5:39 PM
> 
> 24/05/2020 17:31, Morten Brørup:
> > > From: Thomas Monjalon [mailto:thomas@monjalon.net]
> > > Sent: Sunday, May 24, 2020 5:12 PM
> > >
> > > 15/05/2020 15:08, guohongzhi:
> > > > 1.rte_ipv4_cksum will return invalid checksum value like
> 0xffff.(rfc1624)
> > > > 2.__rte_raw_cksum should consider Big Endian.
> > > > 3.checksum 0x0000 is invalid for udp, but valid for tcp.(rfc768)
> > >
> > > Please consider doing 3 patches, thanks.
> > >
> > > It was sent late for 20.05, I prefer not taking any risk,
> > > and wait for maintainers approval.
> > > The patches must be reviewed and discussed in 20.08 timeframe.
> >
> > This patch fixes a bug, which is incompatible with some very popular
> Broadcom switch silicon, so the patch should be treated as such.
> >
> > However, I agree that the review responsibility is not Thomas' alone;
> maintainers should step up if they consider this bugfix critical.
> 
> Unfortunately, the lack of "Signed-off-by" is really blocking the merge.
> I am not allowed adding such thing.

Agreed. A valid "Signed-off-by" with the contributor's actual name a legal requirement - and cannot be ignored.

> 
> Given the bug was already there for a long time,
> it is probably not so much critical anyway.

Or someone has been debugging a production network for years and months, and finally managed to find the root cause of the problems experienced with the affected Broadcom switching silicon. Speaking from experience, the distance from production networks to network silicon software developers can be extremely long. ;-)