Skip to contentSkip to conversation

net-gro: Prepare GRO stack for the upcoming tunneling support

PATCH Applied H.K. Jerry Chu 3 participants 9 messages Dec 12, 2013 – Dec 16, 2013

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 net-gro: Prepare GRO stack for the upcoming tunneling support

From: Jerry Chu <[email protected]>

This patch modifies the GRO stack to avoid the use of "network_header" and associated macros like ip_hdr() and ipv6_hdr() in order to allow an arbitary number of IP hdrs (v4 or v6) to be used in the encapsulation chain. This lays the foundation for various IP tunneling support (IP-in-IP, GRE, VXLAN, SIT,...) to be added later.

With this patch, the GRO stack traversing now is mostly based on skb_gro_offset rather than special hdr offsets saved in skb (e.g., skb->network_header). As a result all but the top layer (i.e., the the transport layer) must have hdrs of the same length in order for a pkt to be considered for aggregation. Therefore when adding a new encap layer (e.g., for tunneling), one must check and skip flows (e.g., by setting NAPI_GRO_CB(p)->same_flow to 0) that have a different hdr length.

Note that unlike the network header, the transport header can and will continue to be set by the GRO code since there will be at most one "transport layer" in the encap chain.

---
 include/linux/netdevice.h |  2 +-
 net/core/dev.c            | 76 ++++++++++++++++-------------------------------
 net/ipv4/af_inet.c        | 25 ++++++++++++----
 net/ipv4/tcp_offload.c    |  9 +++---
 net/ipv6/ip6_offload.c    | 54 ++++++++++++++++++++++++++-------
 net/ipv6/tcpv6_offload.c  |  6 ++--
 6 files changed, 97 insertions(+), 75 deletions(-)
include/linux/netdevice.h +1 -1
index 9d55e51..ba321f1b 100644
@@ -1673,7 +1673,7 @@ struct offload_callbacks {
 	int			(*gso_send_check)(struct sk_buff *skb);
 	struct sk_buff		**(*gro_receive)(struct sk_buff **head,
 					       struct sk_buff *skb);
-	int			(*gro_complete)(struct sk_buff *skb);
+	int			(*gro_complete)(struct sk_buff *skb, int nhoff);
 };
 
 struct packet_offload {
net/core/dev.c +25 -51

125 lines not on this page · raw message

net/ipv4/af_inet.c +19 -6
index 70011e0..ef4f9df 100644
@@ -1377,8 +1377,12 @@ static struct sk_buff **inet_gro_receive(struct sk_buff **head,
 		if (!NAPI_GRO_CB(p)->same_flow)
 			continue;
 
-		iph2 = ip_hdr(p);
-
+		iph2 = (struct iphdr *)(p->data + off);
+		/* The above works because, with the exception of the top
+		 * (inner most) layer, we only aggregate pkts with the same
+		 * hdr length so all the hdrs we'll need to verify will start
+		 * at the same offset.
+		 */
 		if ((iph->protocol ^ iph2->protocol) |
 		    ((__force u32)iph->saddr ^ (__force u32)iph2->saddr) |
 		    ((__force u32)iph->daddr ^ (__force u32)iph2->daddr)) {
@@ -1397,6 +1401,11 @@ static struct sk_buff **inet_gro_receive(struct sk_buff **head,
 	}
 
 	NAPI_GRO_CB(skb)->flush |= flush;
+	skb_set_network_header(skb, off);
+	/* The above will be needed by the transport layer if there is one
+	 * immediately following this IP hdr.
+	 */
+
 	skb_gro_pull(skb, sizeof(*iph));
 	skb_set_transport_header(skb, skb_gro_offset(skb));
 
@@ -1411,10 +1420,10 @@ out:
 	return pp;
 }
 
-static int inet_gro_complete(struct sk_buff *skb)
+static int inet_gro_complete(struct sk_buff *skb, int nhoff)
 {
-	__be16 newlen = htons(skb->len - skb_network_offset(skb));
-	struct iphdr *iph = ip_hdr(skb);
+	__be16 newlen = htons(skb->len - nhoff);
+	struct iphdr *iph = (struct iphdr *)(skb->data + nhoff);
 	const struct net_offload *ops;
 	int proto = iph->protocol;
 	int err = -ENOSYS;
@@ -1427,7 +1436,11 @@ static int inet_gro_complete(struct sk_buff *skb)
 	if (WARN_ON(!ops || !ops->callbacks.gro_complete))
 		goto out_unlock;
 
-	err = ops->callbacks.gro_complete(skb);
+	/* Only need to add sizeof(*iph) to get to the next hdr below
+	 * because any hdr with option will have been flushed in
+	 * inet_gro_receive().
+	 */
+	err = ops->callbacks.gro_complete(skb, nhoff + sizeof(*iph));
 
 out_unlock:
 	rcu_read_unlock();
net/ipv4/tcp_offload.c +5 -4
index 0560635..2658a27 100644
@@ -240,7 +240,7 @@ int tcp_gro_complete(struct sk_buff *skb)
 {
 	struct tcphdr *th = tcp_hdr(skb);
 
-	skb->csum_start = skb_transport_header(skb) - skb->head;
+	skb->csum_start = (unsigned char *)th - skb->head;
 	skb->csum_offset = offsetof(struct tcphdr, check);
 	skb->ip_summed = CHECKSUM_PARTIAL;
 
@@ -272,6 +272,7 @@ static int tcp_v4_gso_send_check(struct sk_buff *skb)
 
 static struct sk_buff **tcp4_gro_receive(struct sk_buff **head, struct sk_buff *skb)
 {
+	/* Use the IP hdr immediately proceeding for this transport */
 	const struct iphdr *iph = skb_gro_network_header(skb);
 	__wsum wsum;
 
@@ -303,13 +304,13 @@ skip_csum:
 	return tcp_gro_receive(head, skb);
 }
 
-static int tcp4_gro_complete(struct sk_buff *skb)
+static int tcp4_gro_complete(struct sk_buff *skb, int thoff)
 {
 	const struct iphdr *iph = ip_hdr(skb);
 	struct tcphdr *th = tcp_hdr(skb);
 
-	th->check = ~tcp_v4_check(skb->len - skb_transport_offset(skb),
-				  iph->saddr, iph->daddr, 0);
+	th->check = ~tcp_v4_check(skb->len - thoff, iph->saddr,
+				  iph->daddr, 0);
 	skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 
 	return tcp_gro_complete(skb);
net/ipv6/ip6_offload.c +44 -10

89 lines not on this page · raw message

net/ipv6/tcpv6_offload.c +3 -3
index 6d18157..0d78132 100644
@@ -66,13 +66,13 @@ skip_csum:
 	return tcp_gro_receive(head, skb);
 }
 
-static int tcp6_gro_complete(struct sk_buff *skb)
+static int tcp6_gro_complete(struct sk_buff *skb, int thoff)
 {
 	const struct ipv6hdr *iph = ipv6_hdr(skb);
 	struct tcphdr *th = tcp_hdr(skb);
 
-	th->check = ~tcp_v6_check(skb->len - skb_transport_offset(skb),
-				  &iph->saddr, &iph->daddr, 0);
+	th->check = ~tcp_v6_check(skb->len - thoff, &iph->saddr,
+				  &iph->daddr, 0);
 	skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
 
 	return tcp_gro_complete(skb);
Signature · 1 line
1.8.5.1
maintainer Applied
#1 lore raw

From: "H.K. Jerry Chu" <[email protected]>
Date: Wed, 11 Dec 2013 20:53:45 -0800

Quoted 17 lines

From: Jerry Chu <[email protected]>

This patch modifies the GRO stack to avoid the use of "network_header" and associated macros like ip_hdr() and ipv6_hdr() in order to allow an arbitary number of IP hdrs (v4 or v6) to be used in the encapsulation chain. This lays the foundation for various IP tunneling support (IP-in-IP, GRE, VXLAN, SIT,...) to be added later.

With this patch, the GRO stack traversing now is mostly based on skb_gro_offset rather than special hdr offsets saved in skb (e.g., skb->network_header). As a result all but the top layer (i.e., the the transport layer) must have hdrs of the same length in order for a pkt to be considered for aggregation. Therefore when adding a new encap layer (e.g., for tunneling), one must check and skip flows (e.g., by setting NAPI_GRO_CB(p)->same_flow to 0) that have a different hdr length.

Quoted 7 lines

Note that unlike the network header, the transport header can and will continue to be set by the GRO code since there will be at most one "transport layer" in the encap chain.

Applied, thanks.

author
#2 lore raw
Quoted 24 lines · David Miller

From: "H.K. Jerry Chu" <[email protected]>
Date: Wed, 11 Dec 2013 20:53:45 -0800

Quoted 21 lines

From: Jerry Chu <[email protected]>

This patch modifies the GRO stack to avoid the use of "network_header" and associated macros like ip_hdr() and ipv6_hdr() in order to allow an arbitary number of IP hdrs (v4 or v6) to be used in the encapsulation chain. This lays the foundation for various IP tunneling support (IP-in-IP, GRE, VXLAN, SIT,...) to be added later.

With this patch, the GRO stack traversing now is mostly based on skb_gro_offset rather than special hdr offsets saved in skb (e.g., skb->network_header). As a result all but the top layer (i.e., the the transport layer) must have hdrs of the same length in order for a pkt to be considered for aggregation. Therefore when adding a new encap layer (e.g., for tunneling), one must check and skip flows (e.g., by setting NAPI_GRO_CB(p)->same_flow to 0) that have a different hdr length.

Note that unlike the network header, the transport header can and will continue to be set by the GRO code since there will be at most one "transport layer" in the encap chain.

Quoted 5 lines
Quoted 3 lines

Applied, thanks.

Thanks, David. I will submit a patch for GRE soon to demonstrate just how easy it is to add tunneling support to GRO after the assumption of only one IP hdr in the encap chain is removed. (I already have the code but it doesn't support the CSUM flag. I'm thinking adding it plus csum offload support before submitting the patch.)

Best,

Jerry

#3 lore raw

PATCH 1/1 ipv6: fix compiler warning in ipv6_exthdrs_len

In reply to H.K. Jerry Chu

Commit 299603e8370a93dd5d8e8d800f0dff1ce2c53d36 ("net-gro: Prepare GRO stack for the upcoming tunneling support") used an uninitialized variable which leads to the following compiler warning:

net/ipv6/ip6_offload.c: In function ‘ipv6_gro_complete’:
net/ipv6/ip6_offload.c:178:24: warning: ‘optlen’ may be used uninitialized in this function [-Wmaybe-uninitialized]

    opth = (void *)opth + optlen;
                        ^
net/ipv6/ip6_offload.c:164:22: note: ‘optlen’ was declared here
  int len = 0, proto, optlen;
                      ^

Fix it up.

---

I just noticed this and tried to fix it. Jerry, could you please review? I hope I got it right. ;)

 net/ipv6/ip6_offload.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)
net/ipv6/ip6_offload.c +5 -4
index 7540a0e..08861f1 100644
@@ -161,7 +161,7 @@ static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 			    const struct net_offload **opps)
 {
 	struct ipv6_opt_hdr *opth = NULL;
-	int len = 0, proto, optlen;
+	int len = 0, optlen = 0, proto;
 
 	proto = iph->nexthdr;
 	for (;;) {
@@ -172,11 +172,12 @@ static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 			if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 				break;
 		}
-		if (opth == NULL)
+		if (opth == NULL) {
 			opth = (void *)(iph+1);
-		else
+		} else {
+			optlen = ipv6_optlen(opth);
 			opth = (void *)opth + optlen;
-		optlen = ipv6_optlen(opth);
+		}
 		len += optlen;
 		proto = opth->nexthdr;
 	}
Signature · 1 line
1.8.3.1
maintainer Applied
#4 lore raw

Re: [PATCH net-next] ipv6: fix compiler warning in ipv6_exthdrs_len

From: Hannes Frederic Sowa <[email protected]>
Date: Sat, 14 Dec 2013 07:29:29 +0100

Quoted 11 lines

Commit 299603e8370a93dd5d8e8d800f0dff1ce2c53d36 ("net-gro: Prepare GRO stack for the upcoming tunneling support") used an uninitialized variable which leads to the following compiler warning:

net/ipv6/ip6_offload.c: In function ‘ipv6_gro_complete’:
net/ipv6/ip6_offload.c:178:24: warning: ‘optlen’ may be used uninitialized in this function [-Wmaybe-uninitialized]

    opth = (void *)opth + optlen;
                        ^
net/ipv6/ip6_offload.c:164:22: note: ‘optlen’ was declared here
  int len = 0, proto, optlen;
                      ^
Quoted 4 lines

Fix it up.

Applied.

author
#5 lore raw

Re: [PATCH net-next] ipv6: fix compiler warning in ipv6_exthdrs_len

Quoted 16 lines · David Miller

From: Hannes Frederic Sowa <[email protected]>
Date: Sat, 14 Dec 2013 07:29:29 +0100

Quoted 13 lines

Commit 299603e8370a93dd5d8e8d800f0dff1ce2c53d36 ("net-gro: Prepare GRO stack for the upcoming tunneling support") used an uninitialized variable which leads to the following compiler warning:

net/ipv6/ip6_offload.c: In function ‘ipv6_gro_complete’:
net/ipv6/ip6_offload.c:178:24: warning: ‘optlen’ may be used uninitialized in this function [-Wmaybe-uninitialized]

    opth = (void *)opth + optlen;
                        ^
net/ipv6/ip6_offload.c:164:22: note: ‘optlen’ was declared here
  int len = 0, proto, optlen;
                      ^

Fix it up.

Quoted 4 lines
Quoted 2 lines

Applied.

Please back out the above "fix" - it actually breaks the original code because the fix will cause exthdrs_len to be one optlen short.

There was no bug in the original code, just the compiler warning (and I do not get any warning, although I admit I did not run with the -W flag).

In any case I will submit a revision shortly that will silent any warning if exists.

Jerry

#6 lore raw

PATCH 1/1 ipv6: revert misguided compiler warning fix on ipv6_exthdrs_len

This essentlally reverts commit f52d81dc27c3456c702e83183035142c222acdc7 ("ipv6: fix compiler warning in ipv6_exthdrs_len") and silences the warning with uninitialized_var.

---

Quoted 20 lines · Jerry Chu
Quoted 19 lines · David Miller

From: Hannes Frederic Sowa <[email protected]>
Date: Sat, 14 Dec 2013 07:29:29 +0100

Quoted 15 lines

Commit 299603e8370a93dd5d8e8d800f0dff1ce2c53d36 ("net-gro: Prepare GRO stack for the upcoming tunneling support") used an uninitialized variable which leads to the following compiler warning:

net/ipv6/ip6_offload.c: In function ‘ipv6_gro_complete’:
net/ipv6/ip6_offload.c:178:24: warning: ‘optlen’ may be used uninitialized in this function [-Wmaybe-uninitialized]

    opth = (void *)opth + optlen;
                        ^
net/ipv6/ip6_offload.c:164:22: note: ‘optlen’ was declared here
  int len = 0, proto, optlen;
                      ^

Fix it up.

Quoted 7 lines
Quoted 1 line

Applied.

Please back out the above "fix" - it actually breaks the original code because the fix will cause exthdrs_len to be one optlen short.

There was no bug in the original code, just the compiler warning (and I do not get any warning, although I admit I did not run with the -W flag).

Sorry, I see now, of course. Btw. I didn't use any special flags, just a normal kernel build with gcc 4.8.2.

Quoted 2 lines

In any case I will submit a revision shortly that will silent any warning if exists.

I hope you are ok with this.

Sorry,

Hannes

 net/ipv6/ip6_offload.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)
net/ipv6/ip6_offload.c +4 -5
index 08861f1..0a578fc 100644
@@ -161,7 +161,7 @@ static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 			    const struct net_offload **opps)
 {
 	struct ipv6_opt_hdr *opth = NULL;
-	int len = 0, optlen = 0, proto;
+	int len = 0, proto, uninitialized_var(optlen);
 
 	proto = iph->nexthdr;
 	for (;;) {
@@ -172,12 +172,11 @@ static int ipv6_exthdrs_len(struct ipv6hdr *iph,
 			if (!((*opps)->flags & INET6_PROTO_GSO_EXTHDR))
 				break;
 		}
-		if (opth == NULL) {
+		if (opth == NULL)
 			opth = (void *)(iph+1);
-		} else {
-			optlen = ipv6_optlen(opth);
+		else
 			opth = (void *)opth + optlen;
-		}
+		optlen = ipv6_optlen(opth);
 		len += optlen;
 		proto = opth->nexthdr;
 	}
Signature · 1 line
1.8.3.1
#7 lore raw

Re: [PATCH net-next] ipv6: revert misguided compiler warning fix on ipv6_exthdrs_len

Quoted 3 lines · Hannes Frederic Sowa

This essentlally reverts commit f52d81dc27c3456c702e83183035142c222acdc7 ("ipv6: fix compiler warning in ipv6_exthdrs_len") and silences the warning with uninitialized_var.

David, please discard this. Jerry already posted a nicer cleanup.

Sorry,

Hannes

maintainer
#8 lore raw

Re: [PATCH net-next] ipv6: revert misguided compiler warning fix on ipv6_exthdrs_len

From: Hannes Frederic Sowa <[email protected]>
Date: Sun, 15 Dec 2013 18:23:22 +0100

Quoted 6 lines · Hannes Frederic Sowa
Quoted 3 lines

This essentlally reverts commit f52d81dc27c3456c702e83183035142c222acdc7 ("ipv6: fix compiler warning in ipv6_exthdrs_len") and silences the warning with uninitialized_var.

David, please discard this. Jerry already posted a nicer cleanup.

I can't resolve these two patches in any reasonable way.

Specifically, whether I revert your patch or not, I cannot get Jerry's patch to apply cleanly.

I can't figure out what Jerry's patch is against at all.