net-gre-gro: Fix a bug that breaks the forwarding path
Subsystems (3)
PATCH 1/1 net-gre-gro: Fix a bug that breaks the forwarding path
From: Jerry Chu <[email protected]>
I stumbled across a bug that was introduced by my own GRE-GRO patch (bf5a755f5e9186406bbf50f4087100af5bd68e40
net-gre-gro: Add GRE support to the GRO stack) submitted a while back that breaks the forwarding path because various GSO related fields were not set. This will cause on the egress path either the GSO code to fail, or a GRE-TSO capable (NETIF_F_GSO_GRE) NICs to choke. The following fix has been tested for both cases.
--- net/core/dev.c | 2 ++ net/ipv4/af_inet.c | 3 +++ net/ipv4/gre_offload.c | 3 +++ net/ipv4/tcp_offload.c | 2 +- net/ipv6/tcpv6_offload.c | 2 +- 5 files changed, 10 insertions(+), 2 deletions(-)
net/core/dev.c +2 -0
@@ -4045,6 +4045,8 @@ static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) skb->vlan_tci =0 ; skb->dev = napi->dev; skb->skb_iif =0 ; + skb->encapsulation =0 ; +skb_shinfo (skb)->gso_type =0 ; napi->skb = skb; }
net/ipv4/af_inet.c +3 -0
@@ -1431,6 +1431,9 @@ static int inet_gro_complete(struct sk_buff *skb, int nhoff)int proto = iph->protocol;int err = -ENOSYS ; +if (skb->encapsulation) +skb_set_inner_network_header (skb, nhoff); +csum_replace2 (&iph->check, iph->tot_len, newlen); iph->tot_len = newlen;
net/ipv4/gre_offload.c +3 -0
@@ -255,6 +255,9 @@ static int gre_gro_complete(struct sk_buff *skb, int nhoff)int err = -ENOENT ;__be16 type; + skb->encapsulation =1 ; +skb_shinfo (skb)->gso_type =SKB_GSO_GRE ; + type = greh->protocol;if (greh->flags &GRE_KEY ) grehlen +=GRE_HEADER_SECTION ;
net/ipv4/tcp_offload.c +1 -1
@@ -310,7 +310,7 @@ static int tcp4_gro_complete(struct sk_buff *skb, int thoff) th->check = ~tcp_v4_check (skb->len - thoff, iph->saddr, iph->daddr,0 ); -skb_shinfo (skb)->gso_type =SKB_GSO_TCPV4 ; +skb_shinfo (skb)->gso_type |=SKB_GSO_TCPV4 ;return tcp_gro_complete (skb); }
net/ipv6/tcpv6_offload.c +1 -1
@@ -73,7 +73,7 @@ static int tcp6_gro_complete(struct sk_buff *skb, int thoff) th->check = ~tcp_v6_check (skb->len - thoff, &iph->saddr, &iph->daddr,0 ); -skb_shinfo (skb)->gso_type =SKB_GSO_TCPV6 ; +skb_shinfo (skb)->gso_type |=SKB_GSO_TCPV6 ;return tcp_gro_complete (skb); }
Signature · 1 line
1.9.0.279.gdc9e3eb
Quoted 37 lines · H.K. Jerry Chu
From: Jerry Chu <[email protected]>
I stumbled across a bug that was introduced by my own GRE-GRO patch (
bf5a755f5e9186406bbf50f4087100af5bd68e40
net-gre-gro: Add GRE support to the GRO stack) submitted a while back that breaks the forwarding path because various GSO related fields were not set. This will cause on the egress path either the GSO code to fail, or a GRE-TSO capable (NETIF_F_GSO_GRE) NICs to choke. The following fix has been tested for both cases.Signed-off-by H.K. Jerry Chu <[email protected]>--- net/core/dev.c | 2 ++ net/ipv4/af_inet.c | 3 +++ net/ipv4/gre_offload.c | 3 +++ net/ipv4/tcp_offload.c | 2 +- net/ipv6/tcpv6_offload.c | 2 +- 5 files changed, 10 insertions(+), 2 deletions(-)net/core/dev.c +2
-0@@ -4045,6 +4045,8 @@ static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) skb->vlan_tci =0 ; skb->dev = napi->dev; skb->skb_iif =0 ; + skb->encapsulation =0 ; +skb_shinfo (skb)->gso_type =0 ; napi->skb = skb; }net/ipv4/af_inet.c +0
-0
net/ipv4/af_inet.c static int inet_gro_complete(struct sk_buff *skb, int nhoff) · line 1436
Quoted 6 lines
net/ipv4/af_inet.c +3
-0@@ -1431,6 +1431,9 @@ static int inet_gro_complete(struct sk_buff *skb, int nhoff)int proto = iph->protocol;int err = -ENOSYS ; +if (skb->encapsulation) +skb_set_inner_network_header (skb, nhoff); +
Note that packets (e.g those who are UDP encapsulated) will pass here twice, once for the external IP header and once for the internal, so if the driver marks the encapsulation sign the inner header setting will happen twice, is that what we want?
Quoted 2 lines
net/ipv4/af_inet.c +0
-0csum_replace2 (&iph->check, iph->tot_len, newlen); iph->tot_len = newlen;
Quoted 42 lines · Or Gerlitz
Quoted 40 lines · H.K. Jerry Chu
From: Jerry Chu <[email protected]>
I stumbled across a bug that was introduced by my own GRE-GRO patch (
bf5a755f5e9186406bbf50f4087100af5bd68e40
net-gre-gro: Add GRE support to the GRO stack) submitted a while back that breaks the forwarding path because various GSO related fields were not set. This will cause on the egress path either the GSO code to fail, or a GRE-TSO capable (NETIF_F_GSO_GRE) NICs to choke. The following fix has been tested for both cases.Signed-off-by H.K. Jerry Chu <[email protected]>--- net/core/dev.c | 2 ++ net/ipv4/af_inet.c | 3 +++ net/ipv4/gre_offload.c | 3 +++ net/ipv4/tcp_offload.c | 2 +- net/ipv6/tcpv6_offload.c | 2 +- 5 files changed, 10 insertions(+), 2 deletions(-)net/core/dev.c +2
-0@@ -4045,6 +4045,8 @@ static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) skb->vlan_tci =0 ; skb->dev = napi->dev; skb->skb_iif =0 ; + skb->encapsulation =0 ; +skb_shinfo (skb)->gso_type =0 ; napi->skb = skb; }net/ipv4/af_inet.c +0
-0@@ -1431,6 +1431,9 @@ static int inet_gro_complete(struct sk_buff *skb, int nhoff)int proto = iph->protocol;int err = -ENOSYS ;
Quoted 8 lines
net/ipv4/af_inet.c static int inet_gro_complete(struct sk_buff *skb, int nhoff) · line 1436
Quoted 3 lines
net/ipv4/af_inet.c +3
-0+if (skb->encapsulation) +skb_set_inner_network_header (skb, nhoff); +Note that packets (e.g those who are UDP encapsulated) will pass here twice, once for the external IP header and once for the internal, so if the driver marks the encapsulation sign the inner header setting will happen twice, is that what we want?
The inner most hdr will be the last to set. But I don't know
all the rules regarding skb->encapsulation. E.g., in this patch napi_reuse_skb() will reset skb->encapsulation. If that's problematic then I could call skb_set_inner_network_header() inside gre_gro_receive() (but that will require checking against "type" in order to compute the right offset).
Jerry
Quoted 4 lines
Quoted 4 lines
net/ipv4/af_inet.c +0
-0csum_replace2 (&iph->check, iph->tot_len, newlen); iph->tot_len = newlen;
David Miller
maintainerFrom: Jerry Chu <[email protected]>
Date: Thu, 27 Feb 2014 15:39:47 -0800
Quoted 52 lines · Or Gerlitz
Quoted 42 lines · H.K. Jerry Chu
Quoted 40 lines
From: Jerry Chu <[email protected]>
I stumbled across a bug that was introduced by my own GRE-GRO patch (
bf5a755f5e9186406bbf50f4087100af5bd68e40
net-gre-gro: Add GRE support to the GRO stack) submitted a while back that breaks the forwarding path because various GSO related fields were not set. This will cause on the egress path either the GSO code to fail, or a GRE-TSO capable (NETIF_F_GSO_GRE) NICs to choke. The following fix has been tested for both cases.Signed-off-by H.K. Jerry Chu <[email protected]>--- net/core/dev.c | 2 ++ net/ipv4/af_inet.c | 3 +++ net/ipv4/gre_offload.c | 3 +++ net/ipv4/tcp_offload.c | 2 +- net/ipv6/tcpv6_offload.c | 2 +- 5 files changed, 10 insertions(+), 2 deletions(-)net/core/dev.c +2
-0@@ -4045,6 +4045,8 @@ static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) skb->vlan_tci =0 ; skb->dev = napi->dev; skb->skb_iif =0 ; + skb->encapsulation =0 ; +skb_shinfo (skb)->gso_type =0 ; napi->skb = skb; }net/ipv4/af_inet.c +0
-0@@ -1431,6 +1431,9 @@ static int inet_gro_complete(struct sk_buff *skb, int nhoff)int proto = iph->protocol;int err = -ENOSYS ;Quoted 8 lines
net/ipv4/af_inet.c static int inet_gro_complete(struct sk_buff *skb, int nhoff) · line 1436
Quoted 3 lines
net/ipv4/af_inet.c +3
-0+if (skb->encapsulation) +skb_set_inner_network_header (skb, nhoff); +Note that packets (e.g those who are UDP encapsulated) will pass here twice, once for the external IP header and once for the internal, so if the driver marks the encapsulation sign the inner header setting will happen twice, is that what we want?
Quoted 6 lines
The inner most hdr will be the last to set. But I don't know
all the rules regardingskb->encapsulation. E.g., in this patchnapi_reuse_skb()will resetskb->encapsulation. If that's problematic then I could callskb_set_inner_network_header()insidegre_gro_receive()(but that will require checking against "type" in order to compute the right offset).
The topic of the skb->encapsulation semantics has come up several times in the past few weeks.
We cannot move forward on any changes in this area until the semantics are well defined, and documented.
Can someone work on a patch which documents skb->encapsulation properly, and then we can come back to fixing this bug?
Thanks.
Quoted 62 lines · David Miller
From: Jerry Chu <[email protected]>
Date: Thu, 27 Feb 2014 15:39:47 -0800Quoted 52 lines · Or Gerlitz
Quoted 42 lines · H.K. Jerry Chu
Quoted 40 lines
From: Jerry Chu <[email protected]>
I stumbled across a bug that was introduced by my own GRE-GRO patch (
bf5a755f5e9186406bbf50f4087100af5bd68e40
net-gre-gro: Add GRE support to the GRO stack) submitted a while back that breaks the forwarding path because various GSO related fields were not set. This will cause on the egress path either the GSO code to fail, or a GRE-TSO capable (NETIF_F_GSO_GRE) NICs to choke. The following fix has been tested for both cases.Signed-off-by H.K. Jerry Chu <[email protected]>--- net/core/dev.c | 2 ++ net/ipv4/af_inet.c | 3 +++ net/ipv4/gre_offload.c | 3 +++ net/ipv4/tcp_offload.c | 2 +- net/ipv6/tcpv6_offload.c | 2 +- 5 files changed, 10 insertions(+), 2 deletions(-)net/core/dev.c +2
-0@@ -4045,6 +4045,8 @@ static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb) skb->vlan_tci =0 ; skb->dev = napi->dev; skb->skb_iif =0 ; + skb->encapsulation =0 ; +skb_shinfo (skb)->gso_type =0 ; napi->skb = skb; }net/ipv4/af_inet.c +0
-0@@ -1431,6 +1431,9 @@ static int inet_gro_complete(struct sk_buff *skb, int nhoff)int proto = iph->protocol;int err = -ENOSYS ;Quoted 8 lines
net/ipv4/af_inet.c static int inet_gro_complete(struct sk_buff *skb, int nhoff) · line 1436
Quoted 3 lines
net/ipv4/af_inet.c +3
-0+if (skb->encapsulation) +skb_set_inner_network_header (skb, nhoff); +Note that packets (e.g those who are UDP encapsulated) will pass here twice, once for the external IP header and once for the internal, so if the driver marks the encapsulation sign the inner header setting will happen twice, is that what we want?
Quoted 6 lines
The inner most hdr will be the last to set. But I don't know
all the rules regardingskb->encapsulation. E.g., in this patchnapi_reuse_skb()will resetskb->encapsulation. If that's problematic then I could callskb_set_inner_network_header()insidegre_gro_receive()(but that will require checking against "type" in order to compute the right offset).
Quoted 8 lines
The topic of the
skb->encapsulationsemantics has come up several times in the past few weeks.We cannot move forward on any changes in this area until the semantics are well defined, and documented.
Can someone work on a patch which documents
skb->encapsulationproperly, and then we can come back to fixing this bug?
Yep, seems we need to be a bit more formal/precise here, we started to discuss that couple of weeks ago on this thread
http://marc.info/?l=linux-netdev&m=139233048521647&w=2
Quoted 1 line · David Miller
The topic of the
skb->encapsulationsemantics has come up several times in the past few weeks. We cannot move forward on any changes in this area until the semantics are well defined, and documented. Can someone work on a patch which documentsskb->encapsulationproperly, and then we can come back to fixing this bug? Thanks.
Lets try... the skb->encapsulation flag was introduced and used in 3.8 by the sequence of these three commits
0afb166 vxlan: Add capability of Rx checksum offload for inner packet d6727fe vxlan: capture inner headers during encapsulation 6a674e9 net: Add support for hardware-offloaded encapsulation
When discussed earlier on the list in the context of the skb->ip_summed field, Tom Herbert came with the following interpretation for the semantics which Joseph confirmed
"when skb->encapsulation is set the ip_summed is valid for both the inner and outer header
(e.g. CHECKSUM_COMPLETE is always assumed okay for both layers). If skb->encapsulation is not set then ip_summed is only valid for outer header"
As for the TX side of things, the change-log of commit 6a674e9 states
"For Tx encapsulation offload, the driver will need to set the right bits in netdev->hw_enc_features. The protocol driver will have to set the skb->encapsulation bit and populate the inner headers, so the NIC driver will use those inner headers to calculate the csum in hardware."
So in higher level, it seems that the role of the skb->encapsulation field is to mark the skb to carry encapsulated packet for the code path between the time the packet is encapsulated by the protocol driver (e.g vxlan/ipip) to the time driver xmit is called. Or from the time driver rx code runs till the the time the packet is decapsulated.
Further, my personal interpretation was that on the rx path, skb should carry the encapsulation flag **only** if the HW was able to offload the inner checksum.
Joseph, what's your thinking here?
Or.
Hi Or,
Thanks for writing this up.
Quoted 22 lines · Or Gerlitz
Quoted 6 lines · David Miller
The topic of the
skb->encapsulationsemantics has come up several times in the past few weeks. We cannot move forward on any changes in this area until the semantics are well defined, and documented. Can someone work on a patch which documentsskb->encapsulationproperly, and then we can come back to fixing this bug? Thanks.Lets try... the
skb->encapsulationflag was introduced and used in 3.8 by the
sequence of these three commits0afb166 vxlan: Add capability of Rx checksum offload for inner packet d6727fe vxlan: capture inner headers during encapsulation 6a674e9 net: Add support for hardware-offloaded encapsulation
When discussed earlier on the list in the context of the
skb->ip_summedfield,
Tom Herbert came with the following interpretation for the semantics which Joseph confirmed
Quoted 4 lines
"when
skb->encapsulationis set theip_summedis valid for both the inner and outer header
(e.g.CHECKSUM_COMPLETEis always assumed okay for both layers). Ifskb->encapsulationis not set thenip_summedis only valid for outer header"
For GRE encapped pkts is the following interpretation correct?
1) ip_summed == CHECKSUM_COMPLETE
=> csum covers IP payload csum of the outer IP datagram
2) ip_summed == CHECKSUM_UNNECESSARY
2.1. skb->encapsulation is on => both GRE csum (if one is present) and TCP/UDP csum have been validated (assuming inner is a TCP or UDP pkt)
2.2. skb->encapsulation is off => only GRE csum (if one is present) is validated.
Quoted 7 lines
As for the TX side of things, the change-log of commit 6a674e9 states
"For Tx encapsulation offload, the driver will need to set the right bits in
netdev->hw_enc_features. The protocol driver will have to set theskb->encapsulationbit and populate the inner headers, so the NIC driver will use those inner headers to calculate the csum in hardware."
So we only support/care about csum offload for the inner pkts, which makes sense.
Quoted 10 lines
So in higher level, it seems that the role of the
skb->encapsulationfield is to mark the skb to carry encapsulated packet for the code path between the time the packet is encapsulated by the protocol driver (e.g vxlan/ipip) to the time driver xmit is called. Or from the time driver rx code runs till the the time the packet is decapsulated.Further, my personal interpretation was that on the rx path, skb should carry the encapsulation flag **only** if the HW was able to offload the inner checksum.
SGTM.
Jerry
Quoted 6 lines
Joseph, what's your thinking here?
Or.
Quoted 32 lines · Jerry Chu
Hi Or,
Thanks for writing this up.
Quoted 22 lines · Or Gerlitz
Quoted 6 lines · David Miller
The topic of the
skb->encapsulationsemantics has come up several times in the past few weeks. We cannot move forward on any changes in this area until the semantics are well defined, and documented. Can someone work on a patch which documentsskb->encapsulationproperly, and then we can come back to fixing this bug? Thanks.Lets try... the
skb->encapsulationflag was introduced and used in 3.8 by the
sequence of these three commits0afb166 vxlan: Add capability of Rx checksum offload for inner packet d6727fe vxlan: capture inner headers during encapsulation 6a674e9 net: Add support for hardware-offloaded encapsulation
When discussed earlier on the list in the context of the
skb->ip_summedfield,
Tom Herbert came with the following interpretation for the semantics which Joseph confirmedQuoted 4 lines
"when
skb->encapsulationis set theip_summedis valid for both the inner and outer header
(e.g.CHECKSUM_COMPLETEis always assumed okay for both layers). Ifskb->encapsulationis not set thenip_summedis only valid for outer header"
Quoted 8 lines
For GRE encapped pkts is the following interpretation correct?
1)
ip_summed==CHECKSUM_COMPLETE
=> csum covers IP payload csum of the outer IP datagram2)
ip_summed==CHECKSUM_UNNECESSARY
2.1.skb->encapsulationis on => both GRE csum (if one is present) and TCP/UDP csum have been validated (assuming inner is a TCP or UDP pkt)
i40e also supports SCTP csumming for the inner packet, too.
Quoted 4 lines
2.2.
skb->encapsulationis off => only GRE csum (if one is present) is validated.
Apart for the SCTP request for inclusion, it looks reasonable to me.
Quoted 10 lines
Quoted 7 lines
As for the TX side of things, the change-log of commit 6a674e9 states
"For Tx encapsulation offload, the driver will need to set the right bits in
netdev->hw_enc_features. The protocol driver will have to set theskb->encapsulationbit and populate the inner headers, so the NIC driver will use those inner headers to calculate the csum in hardware."So we only support/care about csum offload for the inner pkts, which makes sense.
Well, we care about outer too. It is just that the inner headers are for the inner csum, the outer headers are for the outer csum. The outer headers will be always there anyway, so the patch introduced the inner ones. But I guess this is what you meant, right?
Quoted 23 lines
Quoted 11 lines
So in higher level, it seems that the role of the
skb->encapsulationfield is to mark the skb to carry encapsulated packet for the code path between the time the packet is encapsulated by the protocol driver (e.g vxlan/ipip) to the time driver xmit is called. Or from the time driver rx code runs till the the time the packet is decapsulated.Further, my personal interpretation was that on the rx path, skb should carry the encapsulation flag **only** if the HW was able to offload the inner checksum.
SGTM.
Jerry
Quoted 6 lines
Joseph, what's your thinking here?
Or.
Quoted 62 lines · Joseph Gasparakis
Quoted 32 lines · Jerry Chu
Hi Or,
Thanks for writing this up.
Quoted 22 lines · Or Gerlitz
Quoted 6 lines · David Miller
The topic of the
skb->encapsulationsemantics has come up several times in the past few weeks. We cannot move forward on any changes in this area until the semantics are well defined, and documented. Can someone work on a patch which documentsskb->encapsulationproperly, and then we can come back to fixing this bug? Thanks.Lets try... the
skb->encapsulationflag was introduced and used in 3.8 by the
sequence of these three commits0afb166 vxlan: Add capability of Rx checksum offload for inner packet d6727fe vxlan: capture inner headers during encapsulation 6a674e9 net: Add support for hardware-offloaded encapsulation
When discussed earlier on the list in the context of the
skb->ip_summedfield,
Tom Herbert came with the following interpretation for the semantics which Joseph confirmedQuoted 4 lines
"when
skb->encapsulationis set theip_summedis valid for both the inner and outer header
(e.g.CHECKSUM_COMPLETEis always assumed okay for both layers). Ifskb->encapsulationis not set thenip_summedis only valid for outer header"Quoted 8 lines
For GRE encapped pkts is the following interpretation correct?
1)
ip_summed==CHECKSUM_COMPLETE
=> csum covers IP payload csum of the outer IP datagram2)
ip_summed==CHECKSUM_UNNECESSARY
2.1.skb->encapsulationis on => both GRE csum (if one is present) and TCP/UDP csum have been validated (assuming inner is a TCP or UDP pkt)i40e also supports SCTP csumming for the inner packet, too.
Quoted 4 lines
2.2.
skb->encapsulationis off => only GRE csum (if one is present) is validated.Apart for the SCTP request for inclusion, it looks reasonable to me.
Quoted 8 lines
Quoted 7 lines
As for the TX side of things, the change-log of commit 6a674e9 states
"For Tx encapsulation offload, the driver will need to set the right bits in
netdev->hw_enc_features. The protocol driver will have to set theskb->encapsulationbit and populate the inner headers, so the NIC driver will use those inner headers to calculate the csum in hardware."
Quoted 7 lines
Quoted 2 lines
So we only support/care about csum offload for the inner pkts, which makes sense.
Well, we care about outer too. It is just that the inner headers are for the inner csum, the outer headers are for the outer csum. The outer headers will be always there anyway, so the patch introduced the inner ones. But I guess this is what you meant, right?
Actually i was wondering that since there is only one set csum_start/csum_offset fields per skb, which would you support CHECKSUM_PARTIAL for both inner and outer? You can only support one, right? (I haven't checked the UDP encap code to see how this works though.)
Jerry
Quoted 24 lines
Quoted 24 lines
Quoted 9 lines
So in higher level, it seems that the role of the
skb->encapsulationfield is to mark the skb to carry encapsulated packet for the code path between the time the packet is encapsulated by the protocol driver (e.g vxlan/ipip) to the time driver xmit is called. Or from the time driver rx code runs till the the time the packet is decapsulated.Quoted 3 lines
Further, my personal interpretation was that on the rx path, skb should carry the encapsulation flag **only** if the HW was able to offload the inner checksum.
SGTM.
Jerry
Quoted 6 lines
Joseph, what's your thinking here?
Or.
Quoted 72 lines · Jerry Chu
Quoted 62 lines · Joseph Gasparakis
Quoted 32 lines · Jerry Chu
Hi Or,
Thanks for writing this up.
Quoted 22 lines · Or Gerlitz
Quoted 6 lines · David Miller
The topic of the
skb->encapsulationsemantics has come up several times in the past few weeks. We cannot move forward on any changes in this area until the semantics are well defined, and documented. Can someone work on a patch which documentsskb->encapsulationproperly, and then we can come back to fixing this bug? Thanks.Lets try... the
skb->encapsulationflag was introduced and used in 3.8 by the
sequence of these three commits0afb166 vxlan: Add capability of Rx checksum offload for inner packet d6727fe vxlan: capture inner headers during encapsulation 6a674e9 net: Add support for hardware-offloaded encapsulation
When discussed earlier on the list in the context of the
skb->ip_summedfield,
Tom Herbert came with the following interpretation for the semantics which Joseph confirmedQuoted 4 lines
"when
skb->encapsulationis set theip_summedis valid for both the inner and outer header
(e.g.CHECKSUM_COMPLETEis always assumed okay for both layers). Ifskb->encapsulationis not set thenip_summedis only valid for outer header"Quoted 8 lines
For GRE encapped pkts is the following interpretation correct?
1)
ip_summed==CHECKSUM_COMPLETE
=> csum covers IP payload csum of the outer IP datagram2)
ip_summed==CHECKSUM_UNNECESSARY
2.1.skb->encapsulationis on => both GRE csum (if one is present) and TCP/UDP csum have been validated (assuming inner is a TCP or UDP pkt)i40e also supports SCTP csumming for the inner packet, too.
Quoted 4 lines
2.2.
skb->encapsulationis off => only GRE csum (if one is present) is validated.Apart for the SCTP request for inclusion, it looks reasonable to me.
Quoted 8 lines
Quoted 7 lines
As for the TX side of things, the change-log of commit 6a674e9 states
"For Tx encapsulation offload, the driver will need to set the right bits in
netdev->hw_enc_features. The protocol driver will have to set theskb->encapsulationbit and populate the inner headers, so the NIC driver will use those inner headers to calculate the csum in hardware."Quoted 7 lines
Quoted 2 lines
So we only support/care about csum offload for the inner pkts, which makes sense.
Well, we care about outer too. It is just that the inner headers are for the inner csum, the outer headers are for the outer csum. The outer headers will be always there anyway, so the patch introduced the inner ones. But I guess this is what you meant, right?
Quoted 6 lines
Actually i was wondering that since there is only one set csum_start/csum_offset fields per skb, which would you support
CHECKSUM_PARTIALfor both inner and outer? You can only support one, right? (I haven't checked the UDP encap code to see how this works though.)Jerry
I am not sure I understand the question, Jerry... Are you asking in Tx path if the ip_summed==CHECKSUM_PARTIAL, which headers will the csum_start/csum_offset refer to?
Quoted 25 lines
Quoted 24 lines
Quoted 24 lines
Quoted 9 lines
So in higher level, it seems that the role of the
skb->encapsulationfield is to mark the skb to carry encapsulated packet for the code path between the time the packet is encapsulated by the protocol driver (e.g vxlan/ipip) to the time driver xmit is called. Or from the time driver rx code runs till the the time the packet is decapsulated.Quoted 3 lines
Further, my personal interpretation was that on the rx path, skb should carry the encapsulation flag **only** if the HW was able to offload the inner checksum.
SGTM.
Jerry
Quoted 6 lines
Joseph, what's your thinking here?
Or.
Quoted 81 lines · Joseph Gasparakis
Quoted 77 lines · Jerry Chu
Quoted 62 lines · Joseph Gasparakis
Quoted 32 lines · Jerry Chu
Hi Or,
Thanks for writing this up.
Quoted 22 lines · Or Gerlitz
Quoted 6 lines · David Miller
The topic of the
skb->encapsulationsemantics has come up several times in the past few weeks. We cannot move forward on any changes in this area until the semantics are well defined, and documented. Can someone work on a patch which documentsskb->encapsulationproperly, and then we can come back to fixing this bug? Thanks.Lets try... the
skb->encapsulationflag was introduced and used in 3.8 by the
sequence of these three commits0afb166 vxlan: Add capability of Rx checksum offload for inner packet d6727fe vxlan: capture inner headers during encapsulation 6a674e9 net: Add support for hardware-offloaded encapsulation
When discussed earlier on the list in the context of the
skb->ip_summedfield,
Tom Herbert came with the following interpretation for the semantics which Joseph confirmedQuoted 4 lines
"when
skb->encapsulationis set theip_summedis valid for both the inner and outer header
(e.g.CHECKSUM_COMPLETEis always assumed okay for both layers). Ifskb->encapsulationis not set thenip_summedis only valid for outer header"Quoted 8 lines
For GRE encapped pkts is the following interpretation correct?
1)
ip_summed==CHECKSUM_COMPLETE
=> csum covers IP payload csum of the outer IP datagram2)
ip_summed==CHECKSUM_UNNECESSARY
2.1.skb->encapsulationis on => both GRE csum (if one is present) and TCP/UDP csum have been validated (assuming inner is a TCP or UDP pkt)i40e also supports SCTP csumming for the inner packet, too.
Quoted 4 lines
2.2.
skb->encapsulationis off => only GRE csum (if one is present) is validated.Apart for the SCTP request for inclusion, it looks reasonable to me.
Quoted 8 lines
Quoted 7 lines
As for the TX side of things, the change-log of commit 6a674e9 states
"For Tx encapsulation offload, the driver will need to set the right bits in
netdev->hw_enc_features. The protocol driver will have to set theskb->encapsulationbit and populate the inner headers, so the NIC driver will use those inner headers to calculate the csum in hardware."Quoted 7 lines
Quoted 2 lines
So we only support/care about csum offload for the inner pkts, which makes sense.
Well, we care about outer too. It is just that the inner headers are for the inner csum, the outer headers are for the outer csum. The outer headers will be always there anyway, so the patch introduced the inner ones. But I guess this is what you meant, right?
Actually i was wondering that since there is only one set csum_start/csum_offset fields per skb, which would you support
CHECKSUM_PARTIALfor both inner and outer? You can only support one, right? (I haven't checked the UDP encap code to see how this works though.)
Quoted 5 lines
Quoted 1 line
Jerry
I am not sure I understand the question, Jerry... Are you asking in Tx path if the ip_summed==CHECKSUM_PARTIAL, which headers will the csum_start/csum_offset refer to?
Yes (if the question makes sense since i haven't read any UDP encap code - will the outer (UDP) csum be enabled?).
Thanks,
Jerry
Quoted 26 lines
Quoted 26 lines
Quoted 25 lines
Quoted 25 lines
Quoted 10 lines
So in higher level, it seems that the role of the
skb->encapsulationfield is to mark the skb to carry encapsulated packet for the code path between the time the packet is encapsulated by the protocol driver (e.g vxlan/ipip) to the time driver xmit is called. Or from the time driver rx code runs till the the time the packet is decapsulated.Quoted 3 lines
Further, my personal interpretation was that on the rx path, skb should carry the encapsulation flag **only** if the HW was able to offload the inner checksum.
SGTM.
Jerry
Quoted 6 lines
Joseph, what's your thinking here?
Or.
Quoted 89 lines · Jerry Chu
Quoted 81 lines · Joseph Gasparakis
Quoted 77 lines · Jerry Chu
Quoted 62 lines · Joseph Gasparakis
Quoted 32 lines · Jerry Chu
Hi Or,
Thanks for writing this up.
Quoted 22 lines · Or Gerlitz
Quoted 6 lines · David Miller
The topic of the
skb->encapsulationsemantics has come up several times in the past few weeks. We cannot move forward on any changes in this area until the semantics are well defined, and documented. Can someone work on a patch which documentsskb->encapsulationproperly, and then we can come back to fixing this bug? Thanks.Lets try... the
skb->encapsulationflag was introduced and used in 3.8 by the
sequence of these three commits0afb166 vxlan: Add capability of Rx checksum offload for inner packet d6727fe vxlan: capture inner headers during encapsulation 6a674e9 net: Add support for hardware-offloaded encapsulation
When discussed earlier on the list in the context of the
skb->ip_summedfield,
Tom Herbert came with the following interpretation for the semantics which Joseph confirmedQuoted 4 lines
"when
skb->encapsulationis set theip_summedis valid for both the inner and outer header
(e.g.CHECKSUM_COMPLETEis always assumed okay for both layers). Ifskb->encapsulationis not set thenip_summedis only valid for outer header"Quoted 8 lines
For GRE encapped pkts is the following interpretation correct?
1)
ip_summed==CHECKSUM_COMPLETE
=> csum covers IP payload csum of the outer IP datagram2)
ip_summed==CHECKSUM_UNNECESSARY
2.1.skb->encapsulationis on => both GRE csum (if one is present) and TCP/UDP csum have been validated (assuming inner is a TCP or UDP pkt)i40e also supports SCTP csumming for the inner packet, too.
Quoted 4 lines
2.2.
skb->encapsulationis off => only GRE csum (if one is present) is validated.Apart for the SCTP request for inclusion, it looks reasonable to me.
Quoted 8 lines
Quoted 7 lines
As for the TX side of things, the change-log of commit 6a674e9 states
"For Tx encapsulation offload, the driver will need to set the right bits in
netdev->hw_enc_features. The protocol driver will have to set theskb->encapsulationbit and populate the inner headers, so the NIC driver will use those inner headers to calculate the csum in hardware."Quoted 7 lines
Quoted 2 lines
So we only support/care about csum offload for the inner pkts, which makes sense.
Well, we care about outer too. It is just that the inner headers are for the inner csum, the outer headers are for the outer csum. The outer headers will be always there anyway, so the patch introduced the inner ones. But I guess this is what you meant, right?
Actually i was wondering that since there is only one set csum_start/csum_offset fields per skb, which would you support
CHECKSUM_PARTIALfor both inner and outer? You can only support one, right? (I haven't checked the UDP encap code to see how this works though.)Quoted 5 lines
Quoted 1 line
Jerry
I am not sure I understand the question, Jerry... Are you asking in Tx path if the ip_summed==CHECKSUM_PARTIAL, which headers will the csum_start/csum_offset refer to?
Quoted 6 lines
Yes (if the question makes sense since i haven't read any UDP encap code - will the outer (UDP) csum be enabled?).
Thanks,
Jerry
I believe they will refer to the inner headers as csum_start and csum_offset do not get changed during encapsulation and they refer to the original (inner) packet, but please verify this.
Quoted 28 lines
Quoted 27 lines
Quoted 27 lines
Quoted 26 lines
Quoted 26 lines
Quoted 11 lines
So in higher level, it seems that the role of the
skb->encapsulationfield is to mark the skb to carry encapsulated packet for the code path between the time the packet is encapsulated by the protocol driver (e.g vxlan/ipip) to the time driver xmit is called. Or from the time driver rx code runs till the the time the packet is decapsulated.Quoted 3 lines
Further, my personal interpretation was that on the rx path, skb should carry the encapsulation flag **only** if the HW was able to offload the inner checksum.
SGTM.
Jerry
Quoted 6 lines
Joseph, what's your thinking here?
Or.
Quoted 32 lines · Jerry Chu
Hi Or,
Thanks for writing this up.
Quoted 22 lines · Or Gerlitz
Quoted 6 lines · David Miller
The topic of the
skb->encapsulationsemantics has come up several times in the past few weeks. We cannot move forward on any changes in this area until the semantics are well defined, and documented. Can someone work on a patch which documentsskb->encapsulationproperly, and then we can come back to fixing this bug? Thanks.Lets try... the
skb->encapsulationflag was introduced and used in 3.8 by the
sequence of these three commits0afb166 vxlan: Add capability of Rx checksum offload for inner packet d6727fe vxlan: capture inner headers during encapsulation 6a674e9 net: Add support for hardware-offloaded encapsulation
When discussed earlier on the list in the context of the
skb->ip_summedfield,
Tom Herbert came with the following interpretation for the semantics which Joseph confirmedQuoted 4 lines
"when
skb->encapsulationis set theip_summedis valid for both the inner and outer header
(e.g.CHECKSUM_COMPLETEis always assumed okay for both layers). Ifskb->encapsulationis not set thenip_summedis only valid for outer header"
Quoted 4 lines
For GRE encapped pkts is the following interpretation correct?
1)
ip_summed==CHECKSUM_COMPLETE
=> csum covers IP payload csum of the outer IP datagram
my understanding here is that if ip_summed == CHECKSUM_COMPLETE the hw computed skb->csum covers the whole packet except mac header. vlan_untag() will adjust csum
then ip_local_deliver_finish() will pop iphdr without touching skb->csum, since for the valid iphdr, it's a nop operation.
then gre_rcv()-> gre_cisco_rcv()->parse_gre_header()->iptunnel_pull_header() will adjust skb->csum after pulling gre header.
and so on.
xoring csum at every pull is not cheap, so when HW sets CHECKSUM_UNNECESSARY and verifies inner tcp/udp checksum it's the best. Ideally HW would have a fallback to checksum_complete if it cannot parse encap.
Quoted 43 lines
2)
ip_summed==CHECKSUM_UNNECESSARY
2.1.skb->encapsulationis on => both GRE csum (if one is present) and TCP/UDP csum have been validated (assuming inner is a TCP or UDP pkt)2.2.
skb->encapsulationis off => only GRE csum (if one is present) is validated.Quoted 7 lines
As for the TX side of things, the change-log of commit 6a674e9 states
"For Tx encapsulation offload, the driver will need to set the right bits in
netdev->hw_enc_features. The protocol driver will have to set theskb->encapsulationbit and populate the inner headers, so the NIC driver will use those inner headers to calculate the csum in hardware."So we only support/care about csum offload for the inner pkts, which makes sense.
Quoted 10 lines
So in higher level, it seems that the role of the
skb->encapsulationfield is to mark the skb to carry encapsulated packet for the code path between the time the packet is encapsulated by the protocol driver (e.g vxlan/ipip) to the time driver xmit is called. Or from the time driver rx code runs till the the time the packet is decapsulated.Further, my personal interpretation was that on the rx path, skb should carry the encapsulation flag **only** if the HW was able to offload the inner checksum.
SGTM.
Jerry
Quoted 6 lines
Joseph, what's your thinking here?
Or.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoted 14 lines · Or Gerlitz
Quoted 5 lines · David Miller
The topic of the
skb->encapsulationsemantics has come up several times in the past few weeks. We cannot move forward on any changes in this area until the semantics are well defined, and documented. Can someone work on a patch which documentsskb->encapsulationproperly, and then we can come back to fixing this bug? Thanks.Lets try... the
skb->encapsulationflag was introduced and used in 3.8 by the sequence of these three commits0afb166 vxlan: Add capability of Rx checksum offload for inner packet d6727fe vxlan: capture inner headers during encapsulation 6a674e9 net: Add support for hardware-offloaded encapsulation
Quoted 9 lines
When discussed earlier on the list in the context of the
skb->ip_summedfield, Tom Herbert came with the following interpretation for the semantics which Joseph confirmed"when
skb->encapsulationis set theip_summedis valid for both the inner and outer header
(e.g.CHECKSUM_COMPLETEis always assumed okay for both layers). Ifskb->encapsulationis not set thenip_summedis only valid for outer header"
Agree. This should be valid for both Rx and Tx.
Quoted 7 lines
As for the TX side of things, the change-log of commit 6a674e9 states
"For Tx encapsulation offload, the driver will need to set the right bits in
netdev->hw_enc_features. The protocol driver will have to set theskb->encapsulationbit and populate the inner headers, so the NIC driver will use those inner headers to calculate the csum in hardware."
Quoted 5 lines
So in higher level, it seems that the role of the
skb->encapsulationfield is to mark the skb to carry encapsulated packet for the code path between the time the packet is encapsulated by the protocol driver (e.g vxlan/ipip) to the time driver xmit is called. Or from the time driver rx code runs till the the time the packet is decapsulated.
Correct. Here is a little bit more explanation about the though behind these statements:
When the packet gets decapsulated skb->encapsulation should be reset to 0 as all that is left is the (previously to decap) inner packet. For the same reason the inner headers also will not be valid any more: there are no inner headers as such. Personaly in Rx I assume that when the skb leaves the driver, and the hardware has detected encapsulation and hence the csums have been verified (or not), the skb->encapsulation is on and skb->ip_summed is set accordingly for both layers, but the inner headers are not set and even if they are they are not valid.
Also for Tx, skb->encapsulation should be the indication to the driver that it can use the inner headers (i.e. they are valid) in the skb in order to offload the inner csum.
Quoted 6 lines
Further, my personal interpretation was that on the rx path, skb should carry the encapsulation flag **only** if the HW was able to offload the inner checksum.
Joseph, what's your thinking here?
Yes, I agree. If the hardware cannot offload the inner checksum most probably it couldn't even detect the encapsulation.
Quoted 5 lines
Or.
Quoted 45 lines · Joseph Gasparakis
Quoted 14 lines · Or Gerlitz
Quoted 5 lines · David Miller
The topic of the
skb->encapsulationsemantics has come up several times in the past few weeks. We cannot move forward on any changes in this area until the semantics are well defined, and documented. Can someone work on a patch which documentsskb->encapsulationproperly, and then we can come back to fixing this bug? Thanks.Lets try... the
skb->encapsulationflag was introduced and used in 3.8 by the sequence of these three commits0afb166 vxlan: Add capability of Rx checksum offload for inner packet d6727fe vxlan: capture inner headers during encapsulation 6a674e9 net: Add support for hardware-offloaded encapsulation
Quoted 9 lines
When discussed earlier on the list in the context of the
skb->ip_summedfield, Tom Herbert came with the following interpretation for the semantics which Joseph confirmed"when
skb->encapsulationis set theip_summedis valid for both the inner and outer header
(e.g.CHECKSUM_COMPLETEis always assumed okay for both layers). Ifskb->encapsulationis not set thenip_summedis only valid for outer header"Agree. This should be valid for both Rx and Tx.
Quoted 7 lines
As for the TX side of things, the change-log of commit 6a674e9 states
"For Tx encapsulation offload, the driver will need to set the right bits in
netdev->hw_enc_features. The protocol driver will have to set theskb->encapsulationbit and populate the inner headers, so the NIC driver will use those inner headers to calculate the csum in hardware."Quoted 5 lines
So in higher level, it seems that the role of the
skb->encapsulationfield is to mark the skb to carry encapsulated packet for the code path between the time the packet is encapsulated by the protocol driver (e.g vxlan/ipip) to the time driver xmit is called. Or from the time driver rx code runs till the the time the packet is decapsulated.Correct. Here is a little bit more explanation about the though behind these statements:
Quoted 8 lines
When the packet gets decapsulated
skb->encapsulationshould be reset to 0 as all that is left is the (previously to decap) inner packet. For the same reason the inner headers also will not be valid any more: there are no inner headers as such. Personaly in Rx I assume that when the skb leaves the driver, and the hardware has detected encapsulation and hence the csums have been verified (or not), theskb->encapsulationis on andskb->ip_summedis set accordingly for both layers, but the inner headers are not set and even if they are they are not valid.
I am concerned that we are overloading the skb->encpasulation, for instance if this is set in TX path meaning inner headers are valid, this should also be true of RX. It might be better if this field indicated characteristics of the packet independent of being in RX or TX path. For checksum, maybe we should have a separate encap_checksum field. Also, to be future proof may this should be two bits for the devices that can verify checksums in multiple levels of encapsulations (yes, I know this sounds absurd, but it's no more absurd than devices vendors taking it upon themselves to parse to some restricted set of encapsulations instead of just giving us the full packet checksum! :-) ).
I hope to have patches soon on fixing CHECKSUM_COMPLETE, I think it entails some more state in skb indicating offset and extent of checksum value in the packet. An like a said, we need to get out of the habit of stashing pseudo csums there (unless CHECKSUM_PARTIAL is set).
Quoted 19 lines
Also for Tx,
skb->encapsulationshould be the indication to the driver that it can use the inner headers (i.e. they are valid) in the skb in order to offload the inner csum.Quoted 6 lines
Further, my personal interpretation was that on the rx path, skb should carry the encapsulation flag **only** if the HW was able to offload the inner checksum.
Joseph, what's your thinking here?
Yes, I agree. If the hardware cannot offload the inner checksum most probably it couldn't even detect the encapsulation.
Quoted 5 lines
Or.
Quoted 55 lines · Tom Herbert
Quoted 45 lines · Joseph Gasparakis
Quoted 14 lines · Or Gerlitz
Quoted 5 lines · David Miller
The topic of the
skb->encapsulationsemantics has come up several times in the past few weeks. We cannot move forward on any changes in this area until the semantics are well defined, and documented. Can someone work on a patch which documentsskb->encapsulationproperly, and then we can come back to fixing this bug? Thanks.Lets try... the
skb->encapsulationflag was introduced and used in 3.8 by the sequence of these three commits0afb166 vxlan: Add capability of Rx checksum offload for inner packet d6727fe vxlan: capture inner headers during encapsulation 6a674e9 net: Add support for hardware-offloaded encapsulation
Quoted 9 lines
When discussed earlier on the list in the context of the
skb->ip_summedfield, Tom Herbert came with the following interpretation for the semantics which Joseph confirmed"when
skb->encapsulationis set theip_summedis valid for both the inner and outer header
(e.g.CHECKSUM_COMPLETEis always assumed okay for both layers). Ifskb->encapsulationis not set thenip_summedis only valid for outer header"Agree. This should be valid for both Rx and Tx.
Quoted 7 lines
As for the TX side of things, the change-log of commit 6a674e9 states
"For Tx encapsulation offload, the driver will need to set the right bits in
netdev->hw_enc_features. The protocol driver will have to set theskb->encapsulationbit and populate the inner headers, so the NIC driver will use those inner headers to calculate the csum in hardware."Quoted 5 lines
So in higher level, it seems that the role of the
skb->encapsulationfield is to mark the skb to carry encapsulated packet for the code path between the time the packet is encapsulated by the protocol driver (e.g vxlan/ipip) to the time driver xmit is called. Or from the time driver rx code runs till the the time the packet is decapsulated.Correct. Here is a little bit more explanation about the though behind these statements:
Quoted 8 lines
When the packet gets decapsulated
skb->encapsulationshould be reset to 0 as all that is left is the (previously to decap) inner packet. For the same reason the inner headers also will not be valid any more: there are no inner headers as such. Personaly in Rx I assume that when the skb leaves the driver, and the hardware has detected encapsulation and hence the csums have been verified (or not), theskb->encapsulationis on andskb->ip_summedis set accordingly for both layers, but the inner headers are not set and even if they are they are not valid.
Quoted 5 lines
I am concerned that we are overloading the
skb->encpasulation, for instance if this is set in TX path meaning inner headers are valid, this should also be true of RX. It might be better if this field indicated characteristics of the packet independent of being in RX or TX path.
Well, the fact that we cannot have valid inner headers in Rx is simply because hardware does not provide them to the driver, so it doesn't make sense for the driver to parse the pkt and populate them. Do you have in mind a place in Rx path before the protocol (decapsulation) driver that we would do something with inner headers and we need them valid?
Quoted 34 lines
For checksum, maybe we should have a separate
encap_checksumfield. Also, to be future proof may this should be two bits for the devices that can verify checksums in multiple levels of encapsulations (yes, I know this sounds absurd, but it's no more absurd than devices vendors taking it upon themselves to parse to some restricted set of encapsulations instead of just giving us the full packet checksum! :-) ).I hope to have patches soon on fixing
CHECKSUM_COMPLETE, I think it entails some more state in skb indicating offset and extent of checksum value in the packet. An like a said, we need to get out of the habit of stashing pseudo csums there (unlessCHECKSUM_PARTIALis set).Quoted 19 lines
Also for Tx,
skb->encapsulationshould be the indication to the driver that it can use the inner headers (i.e. they are valid) in the skb in order to offload the inner csum.Quoted 6 lines
Further, my personal interpretation was that on the rx path, skb should carry the encapsulation flag **only** if the HW was able to offload the inner checksum.
Joseph, what's your thinking here?
Yes, I agree. If the hardware cannot offload the inner checksum most probably it couldn't even detect the encapsulation.
Quoted 5 lines
Or.
Quoted 1 line · Tom Herbert
I am concerned that we are overloading the
skb->encpasulation
Guys, so to sum up yesterday's messages on this thread -- do folks think we can come up with better/proper definition for the skb->encapsulation bit, or we have to make this multiple bits, or we have (hopefully not too non linear) dependancy on proper definitions/usage of the skb->ip_summed field?
Or.
Quoted 19 lines
for instance if this is set in TX path meaning inner headers are valid, this should also be true of RX. It might be better if this field indicated characteristics of the packet independent of being in RX or TX path. For checksum, maybe we should have a separate
encap_checksumfield. Also, to be future proof may this should be two bits for the devices that can verify checksums in multiple levels of encapsulations (yes, I know this sounds absurd, but it's no more absurd than devices vendors taking it upon themselves to parse to some restricted set of encapsulations instead of just giving us the full packet checksum! :-)).I hope to have patches soon on fixing
CHECKSUM_COMPLETE, I think it entails some more state in skb indicating offset and extent of checksum value in the packet. An like a said, we need to get out of the habit of stashing pseudo csums there (unlessCHECKSUM_PARTIALis set).Quoted 3 lines
Also for Tx,
skb->encapsulationshould be the indication to the driver that it can use the inner headers (i.e. they are valid) in the skb in order to offload the inner csum.
Quoted 10 lines · Or Gerlitz
Quoted 1 line · Tom Herbert
I am concerned that we are overloading the
skb->encpasulationGuys, so to sum up yesterday's messages on this thread -- do folks think we can come up with better/proper definition for the
skb->encapsulationbit, or we have to make this multiple bits, or we have (hopefully not too non linear) dependancy on proper definitions/usage of theskb->ip_summedfield?
Quoted 4 lines
Or.
From my perspective I do not see a gap with the mentioned fields, but I am
open if anyone has something to suggest.
Joseph
Quoted 20 lines
Quoted 19 lines
for instance if this is set in TX path meaning inner headers are valid, this should also be true of RX. It might be better if this field indicated characteristics of the packet independent of being in RX or TX path. For checksum, maybe we should have a separate
encap_checksumfield. Also, to be future proof may this should be two bits for the devices that can verify checksums in multiple levels of encapsulations (yes, I know this sounds absurd, but it's no more absurd than devices vendors taking it upon themselves to parse to some restricted set of encapsulations instead of just giving us the full packet checksum! :-)).I hope to have patches soon on fixing
CHECKSUM_COMPLETE, I think it entails some more state in skb indicating offset and extent of checksum value in the packet. An like a said, we need to get out of the habit of stashing pseudo csums there (unlessCHECKSUM_PARTIALis set).Quoted 3 lines
Also for Tx,
skb->encapsulationshould be the indication to the driver that it can use the inner headers (i.e. they are valid) in the skb in order to offload the inner csum.
Quoted 14 lines · Joseph Gasparakis
Quoted 10 lines · Or Gerlitz
Quoted 1 line · Tom Herbert
I am concerned that we are overloading the
skb->encpasulationGuys, so to sum up yesterday's messages on this thread -- do folks think we can come up with better/proper definition for the
skb->encapsulationbit, or we have to make this multiple bits, or we have (hopefully not too non linear) dependancy on proper definitions/usage of theskb->ip_summedfield?
Quoted 5 lines
Quoted 2 lines
Or.
From my perspective I do not see a gap with the mentioned fields, but I am open if anyone has something to suggest.
I am still unsure if we want to deal with segmentation in nested tunnels. In that case I fear we would need additional markers when to actual segment the UDP packet and update the fragments.
This came up in commit 91a48a2e85a3b7 ("ipv4: ipv6: better estimate tunnel header cut for correct ufo handling").
Greetings,
Hannes