[PATCH RFC v2 ipsec-next 0/6] IPsec GRO layer decapsulation

[PATCH RFC v2 ipsec-next 0/6] IPsec GRO layer decapsulation

[Steffen Klassert]

This patchset adds a software GRO codepath for IPsec ESP.
The ESP gro_receive callback functions decapsulate the
ESP packets at the GRO layer and reinject them back with
gro_cells_receive(). This saves a complete round through
the stack for IPsec ESP packets.

We also need this for ESP HW offload, because HW decrypt but
does not decapsulate the packet. We need to decapsulate before
the inbound policy check, otherwise this check will fail.

Patches 2 an 3 prepare the generic code for packet consuming
gro callbacks.

Changes from v1:

- Add the 'xfrm: Extend the sec_path for IPsec offloading'
  patch and use the secpath to encode GRO calls.

- Coding style fixes.

[PATCH RFC v2 ipsec-next 1/6] xfrm: Add a secpath_set helper.

[Steffen Klassert]

Add a new helper to set the secpath to the skb.
This avoids code duplication, as this is used
in multiple places.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
---
 include/net/xfrm.h     |  1 +
 net/ipv6/xfrm6_input.c | 15 +++------------
 net/xfrm/xfrm_input.c  | 34 ++++++++++++++++++++++------------
 3 files changed, 26 insertions(+), 24 deletions(-)

diff --git a/include/net/xfrm.h b/include/net/xfrm.h
index 6e06130..287635d 100644
--- a/include/net/xfrm.h
+++ b/include/net/xfrm.h
@@ -1006,6 +1006,7 @@ static inline int secpath_exists(struct sk_buff *skb)
 }
 
 struct sec_path *secpath_dup(struct sec_path *src);
+int secpath_set(struct sk_buff *skb);
 
 static inline void
 secpath_reset(struct sk_buff *skb)
diff --git a/net/ipv6/xfrm6_input.c b/net/ipv6/xfrm6_input.c
index b578956..662fb2c 100644
--- a/net/ipv6/xfrm6_input.c
+++ b/net/ipv6/xfrm6_input.c
@@ -69,18 +69,9 @@ int xfrm6_input_addr(struct sk_buff *skb, xfrm_address_t *daddr,
 	struct xfrm_state *x = NULL;
 	int i = 0;
 
-	/* Allocate new secpath or COW existing one. */
-	if (!skb->sp || atomic_read(&skb->sp->refcnt) != 1) {
-		struct sec_path *sp;
-
-		sp = secpath_dup(skb->sp);
-		if (!sp) {
-			XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
-			goto drop;
-		}
-		if (skb->sp)
-			secpath_put(skb->sp);
-		skb->sp = sp;
+	if (secpath_set(skb)) {
+		XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
+		goto drop;
 	}
 
 	if (1 + skb->sp->len == XFRM_MAX_DEPTH) {
diff --git a/net/xfrm/xfrm_input.c b/net/xfrm/xfrm_input.c
index 8722294..d8f913b 100644
--- a/net/xfrm/xfrm_input.c
+++ b/net/xfrm/xfrm_input.c
@@ -117,6 +117,24 @@ struct sec_path *secpath_dup(struct sec_path *src)
 }
 EXPORT_SYMBOL(secpath_dup);
 
+int secpath_set(struct sk_buff *skb)
+{
+	struct sec_path *sp;
+
+	/* Allocate new secpath or COW existing one. */
+	if (!skb->sp || atomic_read(&skb->sp->refcnt) != 1) {
+		sp = secpath_dup(skb->sp);
+		if (!sp)
+			return -ENOMEM;
+
+		if (skb->sp)
+			secpath_put(skb->sp);
+		skb->sp = sp;
+	}
+	return 0;
+}
+EXPORT_SYMBOL(secpath_set);
+
 /* Fetch spi and seq from ipsec header */
 
 int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
@@ -212,18 +230,10 @@ int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
 		break;
 	}
 
-	/* Allocate new secpath or COW existing one. */
-	if (!skb->sp || atomic_read(&skb->sp->refcnt) != 1) {
-		struct sec_path *sp;
-
-		sp = secpath_dup(skb->sp);
-		if (!sp) {
-			XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
-			goto drop;
-		}
-		if (skb->sp)
-			secpath_put(skb->sp);
-		skb->sp = sp;
+	err = secpath_set(skb);
+	if (err) {
+		XFRM_INC_STATS(net, LINUX_MIB_XFRMINERROR);
+		goto drop;
 	}
 
 	seq = 0;
-- 
1.9.1

[PATCH RFC v2 ipsec-next 2/6] net: Add a skb_gro_flush_final helper.

[Steffen Klassert]

Add a skb_gro_flush_final helper to prepare for  consuming
skbs in call_gro_receive. We will extend this helper to not
touch the skb if the skb is consumed by a gro callback with
a followup patch. We need this to handle the upcomming IPsec
ESP callbacks as they reinject the skb to the napi_gro_receive
asynchronous. The handler is used in all gro_receive functions
that can call the ESP gro handlers.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
---
 include/linux/netdevice.h | 5 +++++
 net/ethernet/eth.c        | 2 +-
 net/ipv4/af_inet.c        | 2 +-
 net/ipv6/ip6_offload.c    | 2 +-
 4 files changed, 8 insertions(+), 3 deletions(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 58afbd1..f9da3ac 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -2661,6 +2661,11 @@ static inline void skb_gro_remcsum_cleanup(struct sk_buff *skb,
 	remcsum_unadjust((__sum16 *)ptr, grc->delta);
 }
 
+static inline void skb_gro_flush_final(struct sk_buff *skb, struct sk_buff **pp, int flush)
+{
+	NAPI_GRO_CB(skb)->flush |= flush;
+}
+
 static inline int dev_hard_header(struct sk_buff *skb, struct net_device *dev,
 				  unsigned short type,
 				  const void *daddr, const void *saddr,
diff --git a/net/ethernet/eth.c b/net/ethernet/eth.c
index efdaaab..c666ff0 100644
--- a/net/ethernet/eth.c
+++ b/net/ethernet/eth.c
@@ -474,7 +474,7 @@ struct sk_buff **eth_gro_receive(struct sk_buff **head,
 out_unlock:
 	rcu_read_unlock();
 out:
-	NAPI_GRO_CB(skb)->flush |= flush;
+	skb_gro_flush_final(skb, pp, flush);
 
 	return pp;
 }
diff --git a/net/ipv4/af_inet.c b/net/ipv4/af_inet.c
index 685ba53..602d40f 100644
--- a/net/ipv4/af_inet.c
+++ b/net/ipv4/af_inet.c
@@ -1423,7 +1423,7 @@ struct sk_buff **inet_gro_receive(struct sk_buff **head, struct sk_buff *skb)
 	rcu_read_unlock();
 
 out:
-	NAPI_GRO_CB(skb)->flush |= flush;
+	skb_gro_flush_final(skb, pp, flush);
 
 	return pp;
 }
diff --git a/net/ipv6/ip6_offload.c b/net/ipv6/ip6_offload.c
index fc7b401..0838e6d 100644
--- a/net/ipv6/ip6_offload.c
+++ b/net/ipv6/ip6_offload.c
@@ -253,7 +253,7 @@ static struct sk_buff **ipv6_gro_receive(struct sk_buff **head,
 	rcu_read_unlock();
 
 out:
-	NAPI_GRO_CB(skb)->flush |= flush;
+	skb_gro_flush_final(skb, pp, flush);
 
 	return pp;
 }
-- 
1.9.1

[PATCH RFC v2 ipsec-next 3/6] net: Prepare gro for packet consuming gro callbacks

[Steffen Klassert]

The upcomming IPsec ESP gro callbacks will consume the skb,
so prepare for that.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
---
 include/linux/netdevice.h | 9 +++++++++
 net/core/dev.c            | 7 +++++++
 net/xfrm/Kconfig          | 4 ++++
 3 files changed, 20 insertions(+)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index f9da3ac..9e5d1cd 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -352,6 +352,7 @@ enum gro_result {
 	GRO_HELD,
 	GRO_NORMAL,
 	GRO_DROP,
+	GRO_CONSUMED,
 };
 typedef enum gro_result gro_result_t;
 
@@ -2661,10 +2662,18 @@ static inline void skb_gro_remcsum_cleanup(struct sk_buff *skb,
 	remcsum_unadjust((__sum16 *)ptr, grc->delta);
 }
 
+#ifdef CONFIG_XFRM_OFFLOAD
+static inline void skb_gro_flush_final(struct sk_buff *skb, struct sk_buff **pp, int flush)
+{
+	if (PTR_ERR(pp) != -EINPROGRESS)
+		NAPI_GRO_CB(skb)->flush |= flush;
+}
+#else
 static inline void skb_gro_flush_final(struct sk_buff *skb, struct sk_buff **pp, int flush)
 {
 	NAPI_GRO_CB(skb)->flush |= flush;
 }
+#endif
 
 static inline int dev_hard_header(struct sk_buff *skb, struct net_device *dev,
 				  unsigned short type,
diff --git a/net/core/dev.c b/net/core/dev.c
index 3e1a601..64efbb9 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4505,6 +4505,11 @@ static enum gro_result dev_gro_receive(struct napi_struct *napi, struct sk_buff
 	if (&ptype->list == head)
 		goto normal;
 
+	if (IS_ERR(pp) && PTR_ERR(pp) == -EINPROGRESS) {
+		ret = GRO_CONSUMED;
+		goto ok;
+	}
+
 	same_flow = NAPI_GRO_CB(skb)->same_flow;
 	ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
 
@@ -4609,6 +4614,7 @@ static gro_result_t napi_skb_finish(gro_result_t ret, struct sk_buff *skb)
 
 	case GRO_HELD:
 	case GRO_MERGED:
+	case GRO_CONSUMED:
 		break;
 	}
 
@@ -4680,6 +4686,7 @@ static gro_result_t napi_frags_finish(struct napi_struct *napi,
 		break;
 
 	case GRO_MERGED:
+	case GRO_CONSUMED:
 		break;
 	}
 
diff --git a/net/xfrm/Kconfig b/net/xfrm/Kconfig
index bda1a13..a484451 100644
--- a/net/xfrm/Kconfig
+++ b/net/xfrm/Kconfig
@@ -5,6 +5,10 @@ config XFRM
        bool
        depends on NET
 
+config XFRM_OFFLOAD
+       bool
+       depends on XFRM
+
 config XFRM_ALGO
 	tristate
 	select XFRM
-- 
1.9.1

[PATCH RFC v2 ipsec-next 4/6] xfrm: Export xfrm_parse_spi.

[Steffen Klassert]

We need it in the ESP offload handlers, so export it.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
---
 include/net/xfrm.h    | 1 +
 net/xfrm/xfrm_input.c | 1 +
 2 files changed, 2 insertions(+)

diff --git a/include/net/xfrm.h b/include/net/xfrm.h
index 287635d..fe8db3d 100644
--- a/include/net/xfrm.h
+++ b/include/net/xfrm.h
@@ -1520,6 +1520,7 @@ int xfrm4_rcv_encap(struct sk_buff *skb, int nexthdr, __be32 spi,
 		    int encap_type);
 int xfrm4_transport_finish(struct sk_buff *skb, int async);
 int xfrm4_rcv(struct sk_buff *skb);
+int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq);
 
 static inline int xfrm4_rcv_spi(struct sk_buff *skb, int nexthdr, __be32 spi)
 {
diff --git a/net/xfrm/xfrm_input.c b/net/xfrm/xfrm_input.c
index d8f913b..86f8a8d 100644
--- a/net/xfrm/xfrm_input.c
+++ b/net/xfrm/xfrm_input.c
@@ -170,6 +170,7 @@ int xfrm_parse_spi(struct sk_buff *skb, u8 nexthdr, __be32 *spi, __be32 *seq)
 	*seq = *(__be32 *)(skb_transport_header(skb) + offset_seq);
 	return 0;
 }
+EXPORT_SYMBOL(xfrm_parse_spi);
 
 int xfrm_prepare_input(struct xfrm_state *x, struct sk_buff *skb)
 {
-- 
1.9.1

[PATCH RFC v2 ipsec-next 5/6] xfrm: Extend the sec_path for IPsec offloading

[Steffen Klassert]

We need to keep per packet offloading informations across
the layers. So we extend the sec_path to carry these for
the input and output offload codepath.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
---
 include/net/xfrm.h    | 41 +++++++++++++++++++++++++++++++++++++++++
 net/xfrm/xfrm_input.c |  2 ++
 2 files changed, 43 insertions(+)

diff --git a/include/net/xfrm.h b/include/net/xfrm.h
index fe8db3d..11f63f2 100644
--- a/include/net/xfrm.h
+++ b/include/net/xfrm.h
@@ -498,6 +498,7 @@ struct xfrm_tmpl {
 };
 
 #define XFRM_MAX_DEPTH		6
+#define XFRM_MAX_OFFLOAD_DEPTH	1
 
 struct xfrm_policy_walk_entry {
 	struct list_head	all;
@@ -973,10 +974,41 @@ static inline void xfrm_dst_destroy(struct xfrm_dst *xdst)
 
 void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev);
 
+struct xfrm_offload {
+        /* Output sequence number for replay protection on offloading. */
+	struct {
+		__u32 low;
+		__u32 hi;
+	} seq;
+
+	__u32			flags;
+#define	SA_DELETE_REQ		1
+#define	CRYPTO_DONE		2
+#define	CRYPTO_NEXT_DONE	4
+#define	CRYPTO_FALLBACK		8
+#define	XFRM_GSO_SEGMENT	16
+#define	XFRM_GRO		32
+
+	__u32			status;
+#define CRYPTO_SUCCESS				1
+#define CRYPTO_GENERIC_ERROR			2
+#define CRYPTO_TRANSPORT_AH_AUTH_FAILED		4
+#define CRYPTO_TRANSPORT_ESP_AUTH_FAILED	8
+#define CRYPTO_TUNNEL_AH_AUTH_FAILED		16
+#define CRYPTO_TUNNEL_ESP_AUTH_FAILED		32
+#define CRYPTO_INVALID_PACKET_SYNTAX		64
+#define CRYPTO_INVALID_PROTOCOL			128
+
+	__u8			proto;
+};
+
 struct sec_path {
 	atomic_t		refcnt;
 	int			len;
+	int			olen;
+
 	struct xfrm_state	*xvec[XFRM_MAX_DEPTH];
+	struct xfrm_offload	ovec[XFRM_MAX_OFFLOAD_DEPTH];
 };
 
 static inline int secpath_exists(struct sk_buff *skb)
@@ -1776,6 +1808,15 @@ static inline struct xfrm_state *xfrm_input_state(struct sk_buff *skb)
 {
 	return skb->sp->xvec[skb->sp->len - 1];
 }
+static inline struct xfrm_offload *xfrm_offload(struct sk_buff *skb)
+{
+	struct sec_path *sp = skb->sp;
+
+	if (!sp || !sp->olen || sp->len != sp->olen)
+		return NULL;
+
+	return &sp->ovec[sp->olen - 1];
+}
 #endif
 
 static inline int xfrm_mark_get(struct nlattr **attrs, struct xfrm_mark *m)
diff --git a/net/xfrm/xfrm_input.c b/net/xfrm/xfrm_input.c
index 86f8a8d..d2ff712 100644
--- a/net/xfrm/xfrm_input.c
+++ b/net/xfrm/xfrm_input.c
@@ -105,6 +105,8 @@ struct sec_path *secpath_dup(struct sec_path *src)
 		return NULL;
 
 	sp->len = 0;
+	sp->olen = 0;
+
 	if (src) {
 		int i;
 
-- 
1.9.1

[PATCH RFC v2 ipsec-next 6/6] esp: Add a software GRO codepath

[Steffen Klassert]

This patch adds GRO ifrastructure and callbacks for ESP on
ipv4 and ipv6.

In case the GRO layer detects an ESP packet, the
esp{4,6}_gro_receive() function does a xfrm state lookup
and calls the xfrm input layer if it finds a matching state.
The packet will be decapsulated and reinjected it into layer 2.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
---
 include/net/xfrm.h              |   1 +
 net/ipv4/Kconfig                |  13 +++++
 net/ipv4/Makefile               |   1 +
 net/ipv4/esp4_offload.c         | 106 +++++++++++++++++++++++++++++++++++++++
 net/ipv4/xfrm4_input.c          |   6 +++
 net/ipv4/xfrm4_mode_transport.c |   4 +-
 net/ipv6/Kconfig                |  13 +++++
 net/ipv6/Makefile               |   1 +
 net/ipv6/esp6_offload.c         | 108 ++++++++++++++++++++++++++++++++++++++++
 net/ipv6/xfrm6_input.c          |   7 +++
 net/ipv6/xfrm6_mode_transport.c |   4 +-
 net/xfrm/xfrm_input.c           |  31 ++++++++++--
 12 files changed, 288 insertions(+), 7 deletions(-)
 create mode 100644 net/ipv4/esp4_offload.c
 create mode 100644 net/ipv6/esp6_offload.c

diff --git a/include/net/xfrm.h b/include/net/xfrm.h
index 11f63f2..e45837e 100644
--- a/include/net/xfrm.h
+++ b/include/net/xfrm.h
@@ -682,6 +682,7 @@ struct xfrm_spi_skb_cb {
 
 	unsigned int daddroff;
 	unsigned int family;
+	__be32 seq;
 };
 
 #define XFRM_SPI_SKB_CB(__skb) ((struct xfrm_spi_skb_cb *)&((__skb)->cb[0]))
diff --git a/net/ipv4/Kconfig b/net/ipv4/Kconfig
index 6e7baaf..e0878c3 100644
--- a/net/ipv4/Kconfig
+++ b/net/ipv4/Kconfig
@@ -360,6 +360,19 @@ config INET_ESP
 
 	  If unsure, say Y.
 
+config INET_ESP_OFFLOAD
+	tristate "IP: ESP transformation offload"
+	depends on INET_ESP
+	select XFRM_OFFLOAD
+	default n
+	---help---
+	  Support for ESP transformation offload. This makes sense
+	  only if this system really does IPsec and want to do it
+	  with high throughput. A typical desktop system does not
+	  need it, even if it does IPsec.
+
+	  If unsure, say N.
+
 config INET_IPCOMP
 	tristate "IP: IPComp transformation"
 	select INET_XFRM_TUNNEL
diff --git a/net/ipv4/Makefile b/net/ipv4/Makefile
index 48af58a..c6d4238 100644
--- a/net/ipv4/Makefile
+++ b/net/ipv4/Makefile
@@ -29,6 +29,7 @@ obj-$(CONFIG_NET_IPVTI) += ip_vti.o
 obj-$(CONFIG_SYN_COOKIES) += syncookies.o
 obj-$(CONFIG_INET_AH) += ah4.o
 obj-$(CONFIG_INET_ESP) += esp4.o
+obj-$(CONFIG_INET_ESP_OFFLOAD) += esp4_offload.o
 obj-$(CONFIG_INET_IPCOMP) += ipcomp.o
 obj-$(CONFIG_INET_XFRM_TUNNEL) += xfrm4_tunnel.o
 obj-$(CONFIG_INET_XFRM_MODE_BEET) += xfrm4_mode_beet.o
diff --git a/net/ipv4/esp4_offload.c b/net/ipv4/esp4_offload.c
new file mode 100644
index 0000000..1de4426
--- /dev/null
+++ b/net/ipv4/esp4_offload.c
@@ -0,0 +1,106 @@
+/*
+ * IPV4 GSO/GRO offload support
+ * Linux INET implementation
+ *
+ * Copyright (C) 2016 secunet Security Networks AG
+ * Author: Steffen Klassert <steffen.klassert@secunet.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * ESP GRO support
+ */
+
+#include <linux/skbuff.h>
+#include <linux/init.h>
+#include <net/protocol.h>
+#include <crypto/aead.h>
+#include <crypto/authenc.h>
+#include <linux/err.h>
+#include <linux/module.h>
+#include <net/ip.h>
+#include <net/xfrm.h>
+#include <net/esp.h>
+#include <linux/scatterlist.h>
+#include <linux/kernel.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <net/udp.h>
+
+static struct sk_buff **esp4_gro_receive(struct sk_buff **head,
+					 struct sk_buff *skb)
+{
+	int offset = skb_gro_offset(skb);
+	struct xfrm_offload *xo;
+	struct xfrm_state *x;
+	__be32 seq;
+	__be32 spi;
+	int err;
+
+	skb_pull(skb, offset);
+
+	if ((err = xfrm_parse_spi(skb, IPPROTO_ESP, &spi, &seq)) != 0)
+		goto out;
+
+	err = secpath_set(skb);
+	if (err)
+		goto out;
+
+	if (skb->sp->len == XFRM_MAX_DEPTH)
+		goto out;
+
+	x = xfrm_state_lookup(dev_net(skb->dev), skb->mark,
+			      (xfrm_address_t *)&ip_hdr(skb)->daddr,
+			      spi, IPPROTO_ESP, AF_INET);
+	if (!x)
+		goto out;
+
+	skb->sp->xvec[skb->sp->len++] = x;
+	skb->sp->olen++;
+
+	xo = xfrm_offload(skb);
+	if (!xo) {
+		xfrm_state_put(x);
+		goto out;
+	}
+	xo->flags |= XFRM_GRO;
+
+	XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip4 = NULL;
+	XFRM_SPI_SKB_CB(skb)->family = AF_INET;
+	XFRM_SPI_SKB_CB(skb)->daddroff = offsetof(struct iphdr, daddr);
+	XFRM_SPI_SKB_CB(skb)->seq = seq;
+
+	/* We don't need to handle errors from xfrm_input, it does all
+	 * the error handling and frees the resources on error. */
+	xfrm_input(skb, IPPROTO_ESP, spi, -2);
+
+	return ERR_PTR(-EINPROGRESS);
+out:
+	skb_push(skb, offset);
+	NAPI_GRO_CB(skb)->same_flow = 0;
+	NAPI_GRO_CB(skb)->flush = 1;
+
+	return NULL;
+}
+
+static const struct net_offload esp4_offload = {
+	.callbacks = {
+		.gro_receive = esp4_gro_receive,
+	},
+};
+
+static int __init esp4_offload_init(void)
+{
+	return inet_add_offload(&esp4_offload, IPPROTO_ESP);
+}
+
+static void __exit esp4_offload_exit(void)
+{
+	inet_del_offload(&esp4_offload, IPPROTO_ESP);
+}
+
+module_init(esp4_offload_init);
+module_exit(esp4_offload_exit);
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Steffen Klassert <steffen.klassert@secunet.com>");
diff --git a/net/ipv4/xfrm4_input.c b/net/ipv4/xfrm4_input.c
index 62e1e72..1fc6841 100644
--- a/net/ipv4/xfrm4_input.c
+++ b/net/ipv4/xfrm4_input.c
@@ -40,6 +40,7 @@ static inline int xfrm4_rcv_encap_finish(struct net *net, struct sock *sk,
 
 int xfrm4_transport_finish(struct sk_buff *skb, int async)
 {
+	struct xfrm_offload *xo = xfrm_offload(skb);
 	struct iphdr *iph = ip_hdr(skb);
 
 	iph->protocol = XFRM_MODE_SKB_CB(skb)->protocol;
@@ -53,6 +54,11 @@ int xfrm4_transport_finish(struct sk_buff *skb, int async)
 	iph->tot_len = htons(skb->len);
 	ip_send_check(iph);
 
+	if (xo && (xo->flags & XFRM_GRO)) {
+		skb_mac_header_rebuild(skb);
+		return 0;
+	}
+
 	NF_HOOK(NFPROTO_IPV4, NF_INET_PRE_ROUTING,
 		dev_net(skb->dev), NULL, skb, skb->dev, NULL,
 		xfrm4_rcv_encap_finish);
diff --git a/net/ipv4/xfrm4_mode_transport.c b/net/ipv4/xfrm4_mode_transport.c
index fd840c7..4acc050 100644
--- a/net/ipv4/xfrm4_mode_transport.c
+++ b/net/ipv4/xfrm4_mode_transport.c
@@ -43,6 +43,7 @@ static int xfrm4_transport_output(struct xfrm_state *x, struct sk_buff *skb)
 static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
 {
 	int ihl = skb->data - skb_transport_header(skb);
+	struct xfrm_offload *xo = xfrm_offload(skb);
 
 	if (skb->transport_header != skb->network_header) {
 		memmove(skb_transport_header(skb),
@@ -50,7 +51,8 @@ static int xfrm4_transport_input(struct xfrm_state *x, struct sk_buff *skb)
 		skb->network_header = skb->transport_header;
 	}
 	ip_hdr(skb)->tot_len = htons(skb->len + ihl);
-	skb_reset_transport_header(skb);
+	if (!xo || !(xo->flags & XFRM_GRO))
+		skb_reset_transport_header(skb);
 	return 0;
 }
 
diff --git a/net/ipv6/Kconfig b/net/ipv6/Kconfig
index ec1267e..b2a85cc 100644
--- a/net/ipv6/Kconfig
+++ b/net/ipv6/Kconfig
@@ -75,6 +75,19 @@ config INET6_ESP
 
 	  If unsure, say Y.
 
+config INET6_ESP_OFFLOAD
+	tristate "IPv6: ESP transformation offload"
+	depends on INET6_ESP
+	select XFRM_OFFLOAD
+	default n
+	---help---
+	  Support for ESP transformation offload. This makes sense
+	  only if this system really does IPsec and want to do it
+	  with high throughput. A typical desktop system does not
+	  need it, even if it does IPsec.
+
+	  If unsure, say N.
+
 config INET6_IPCOMP
 	tristate "IPv6: IPComp transformation"
 	select INET6_XFRM_TUNNEL
diff --git a/net/ipv6/Makefile b/net/ipv6/Makefile
index a9e9fec..217e9ff 100644
--- a/net/ipv6/Makefile
+++ b/net/ipv6/Makefile
@@ -30,6 +30,7 @@ ipv6-objs += $(ipv6-y)
 
 obj-$(CONFIG_INET6_AH) += ah6.o
 obj-$(CONFIG_INET6_ESP) += esp6.o
+obj-$(CONFIG_INET6_ESP_OFFLOAD) += esp6_offload.o
 obj-$(CONFIG_INET6_IPCOMP) += ipcomp6.o
 obj-$(CONFIG_INET6_XFRM_TUNNEL) += xfrm6_tunnel.o
 obj-$(CONFIG_INET6_TUNNEL) += tunnel6.o
diff --git a/net/ipv6/esp6_offload.c b/net/ipv6/esp6_offload.c
new file mode 100644
index 0000000..d914eb9
--- /dev/null
+++ b/net/ipv6/esp6_offload.c
@@ -0,0 +1,108 @@
+/*
+ * IPV6 GSO/GRO offload support
+ * Linux INET implementation
+ *
+ * Copyright (C) 2016 secunet Security Networks AG
+ * Author: Steffen Klassert <steffen.klassert@secunet.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * ESP GRO support
+ */
+
+#include <linux/skbuff.h>
+#include <linux/init.h>
+#include <net/protocol.h>
+#include <crypto/aead.h>
+#include <crypto/authenc.h>
+#include <linux/err.h>
+#include <linux/module.h>
+#include <net/ip.h>
+#include <net/xfrm.h>
+#include <net/esp.h>
+#include <linux/scatterlist.h>
+#include <linux/kernel.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <net/ip6_route.h>
+#include <net/ipv6.h>
+#include <linux/icmpv6.h>
+
+static struct sk_buff **esp6_gro_receive(struct sk_buff **head,
+					 struct sk_buff *skb)
+{
+	int offset = skb_gro_offset(skb);
+	struct xfrm_offload *xo;
+	struct xfrm_state *x;
+	__be32 seq;
+	__be32 spi;
+	int err;
+
+	skb_pull(skb, offset);
+
+	if ((err = xfrm_parse_spi(skb, IPPROTO_ESP, &spi, &seq)) != 0)
+		goto out;
+
+	err = secpath_set(skb);
+	if (err)
+		goto out;
+
+	if (skb->sp->len == XFRM_MAX_DEPTH)
+		goto out;
+
+	x = xfrm_state_lookup(dev_net(skb->dev), skb->mark,
+			      (xfrm_address_t *)&ipv6_hdr(skb)->daddr,
+			      spi, IPPROTO_ESP, AF_INET6);
+	if (!x)
+		goto out;
+
+	skb->sp->xvec[skb->sp->len++] = x;
+	skb->sp->olen++;
+
+	xo = xfrm_offload(skb);
+	if (!xo) {
+		xfrm_state_put(x);
+		goto out;
+	}
+	xo->flags |= XFRM_GRO;
+
+	XFRM_TUNNEL_SKB_CB(skb)->tunnel.ip6 = NULL;
+	XFRM_SPI_SKB_CB(skb)->family = AF_INET6;
+	XFRM_SPI_SKB_CB(skb)->daddroff = offsetof(struct ipv6hdr, daddr);
+	XFRM_SPI_SKB_CB(skb)->seq = seq;
+
+	/* We don't need to handle errors from xfrm_input, it does all
+	 * the error handling and frees the resources on error. */
+	xfrm_input(skb, IPPROTO_ESP, spi, -2);
+
+	return ERR_PTR(-EINPROGRESS);
+out:
+	skb_push(skb, offset);
+	NAPI_GRO_CB(skb)->same_flow = 0;
+	NAPI_GRO_CB(skb)->flush = 1;
+
+	return NULL;
+}
+
+static const struct net_offload esp6_offload = {
+	.callbacks = {
+		.gro_receive = esp6_gro_receive,
+	},
+};
+
+static int __init esp6_offload_init(void)
+{
+	return inet6_add_offload(&esp6_offload, IPPROTO_ESP);
+}
+
+static void __exit esp6_offload_exit(void)
+{
+	inet6_del_offload(&esp6_offload, IPPROTO_ESP);
+}
+
+module_init(esp6_offload_init);
+module_exit(esp6_offload_exit);
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Steffen Klassert <steffen.klassert@secunet.com>");
diff --git a/net/ipv6/xfrm6_input.c b/net/ipv6/xfrm6_input.c
index 662fb2c..08a807b 100644
--- a/net/ipv6/xfrm6_input.c
+++ b/net/ipv6/xfrm6_input.c
@@ -33,6 +33,8 @@ int xfrm6_rcv_spi(struct sk_buff *skb, int nexthdr, __be32 spi,
 
 int xfrm6_transport_finish(struct sk_buff *skb, int async)
 {
+	struct xfrm_offload *xo = xfrm_offload(skb);
+
 	skb_network_header(skb)[IP6CB(skb)->nhoff] =
 		XFRM_MODE_SKB_CB(skb)->protocol;
 
@@ -44,6 +46,11 @@ int xfrm6_transport_finish(struct sk_buff *skb, int async)
 	ipv6_hdr(skb)->payload_len = htons(skb->len);
 	__skb_push(skb, skb->data - skb_network_header(skb));
 
+	if (xo && (xo->flags & XFRM_GRO)) {
+		skb_mac_header_rebuild(skb);
+		return -1;
+	}
+
 	NF_HOOK(NFPROTO_IPV6, NF_INET_PRE_ROUTING,
 		dev_net(skb->dev), NULL, skb, skb->dev, NULL,
 		ip6_rcv_finish);
diff --git a/net/ipv6/xfrm6_mode_transport.c b/net/ipv6/xfrm6_mode_transport.c
index 4e34410..4439ee4 100644
--- a/net/ipv6/xfrm6_mode_transport.c
+++ b/net/ipv6/xfrm6_mode_transport.c
@@ -47,6 +47,7 @@ static int xfrm6_transport_output(struct xfrm_state *x, struct sk_buff *skb)
 static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
 {
 	int ihl = skb->data - skb_transport_header(skb);
+	struct xfrm_offload *xo = xfrm_offload(skb);
 
 	if (skb->transport_header != skb->network_header) {
 		memmove(skb_transport_header(skb),
@@ -55,7 +56,8 @@ static int xfrm6_transport_input(struct xfrm_state *x, struct sk_buff *skb)
 	}
 	ipv6_hdr(skb)->payload_len = htons(skb->len + ihl -
 					   sizeof(struct ipv6hdr));
-	skb_reset_transport_header(skb);
+	if (!xo || !(xo->flags & XFRM_GRO))
+		skb_reset_transport_header(skb);
 	return 0;
 }
 
diff --git a/net/xfrm/xfrm_input.c b/net/xfrm/xfrm_input.c
index d2ff712..46bdb4f 100644
--- a/net/xfrm/xfrm_input.c
+++ b/net/xfrm/xfrm_input.c
@@ -207,14 +207,23 @@ int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
 	unsigned int family;
 	int decaps = 0;
 	int async = 0;
+	struct xfrm_offload *xo;
+	bool xfrm_gro = false;
 
-	/* A negative encap_type indicates async resumption. */
 	if (encap_type < 0) {
-		async = 1;
 		x = xfrm_input_state(skb);
-		seq = XFRM_SKB_CB(skb)->seq.input.low;
 		family = x->outer_mode->afinfo->family;
-		goto resume;
+
+		/* An encap_type of -1 indicates async resumption. */
+		if (encap_type == -1) {
+			async = 1;
+			seq = XFRM_SKB_CB(skb)->seq.input.low;
+			goto resume;
+		}
+		/* encap_type < -1 indicates a GRO call. */
+		encap_type = 0;
+		seq = XFRM_SPI_SKB_CB(skb)->seq;
+		goto lock;
 	}
 
 	daddr = (xfrm_address_t *)(skb_network_header(skb) +
@@ -260,6 +269,7 @@ int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
 
 		skb->sp->xvec[skb->sp->len++] = x;
 
+lock:
 		spin_lock(&x->lock);
 
 		if (unlikely(x->km.state != XFRM_STATE_VALID)) {
@@ -381,7 +391,18 @@ int xfrm_input(struct sk_buff *skb, int nexthdr, __be32 spi, int encap_type)
 		gro_cells_receive(&gro_cells, skb);
 		return 0;
 	} else {
-		return x->inner_mode->afinfo->transport_finish(skb, async);
+		xo = xfrm_offload(skb);
+		if (xo)
+			xfrm_gro = xo->flags & XFRM_GRO;
+
+		err = x->inner_mode->afinfo->transport_finish(skb, async);
+		if (xfrm_gro) {
+			skb_dst_drop(skb);
+			gro_cells_receive(&gro_cells, skb);
+			return err;
+		}
+
+		return err;
 	}
 
 drop_unlock:
-- 
1.9.1

Re: [PATCH RFC v2 ipsec-next 0/6] IPsec GRO layer decapsulation

[Steffen Klassert]

On Mon, Feb 13, 2017 at 10:57:55AM +0100, Steffen Klassert wrote:
> This patchset adds a software GRO codepath for IPsec ESP.
> The ESP gro_receive callback functions decapsulate the
> ESP packets at the GRO layer and reinject them back with
> gro_cells_receive(). This saves a complete round through
> the stack for IPsec ESP packets.
> 
> We also need this for ESP HW offload, because HW decrypt but
> does not decapsulate the packet. We need to decapsulate before
> the inbound policy check, otherwise this check will fail.
> 
> Patches 2 an 3 prepare the generic code for packet consuming
> gro callbacks.
> 
> Changes from v1:
> 
> - Add the 'xfrm: Extend the sec_path for IPsec offloading'
>   patch and use the secpath to encode GRO calls.
> 
> - Coding style fixes.

I have not got any further feedback, so I applied this
series to ipsec-next.