[linux-2.6-block.git] / net / ipv4 / ip_tunnel_core.c

// SPDX-License-Identifier: GPL-2.0-only
/*
 * Copyright (c) 2013 Nicira, Inc.
 */

#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

#include <linux/types.h>
#include <linux/kernel.h>
#include <linux/skbuff.h>
#include <linux/netdevice.h>
#include <linux/in.h>
#include <linux/if_arp.h>
#include <linux/init.h>
#include <linux/in6.h>
#include <linux/inetdevice.h>
#include <linux/netfilter_ipv4.h>
#include <linux/etherdevice.h>
#include <linux/if_ether.h>
#include <linux/if_vlan.h>
#include <linux/static_key.h>

#include <net/ip.h>
#include <net/icmp.h>
#include <net/protocol.h>
#include <net/ip_tunnels.h>
#include <net/ip6_tunnel.h>
#include <net/arp.h>
#include <net/checksum.h>
#include <net/dsfield.h>
#include <net/inet_ecn.h>
#include <net/xfrm.h>
#include <net/net_namespace.h>
#include <net/netns/generic.h>
#include <net/rtnetlink.h>
#include <net/dst_metadata.h>

const struct ip_tunnel_encap_ops __rcu *
		iptun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
EXPORT_SYMBOL(iptun_encaps);

const struct ip6_tnl_encap_ops __rcu *
		ip6tun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
EXPORT_SYMBOL(ip6tun_encaps);

void iptunnel_xmit(struct sock *sk, struct rtable *rt, struct sk_buff *skb,
		   __be32 src, __be32 dst, __u8 proto,
		   __u8 tos, __u8 ttl, __be16 df, bool xnet)
{
	int pkt_len = skb->len - skb_inner_network_offset(skb);
	struct net *net = dev_net(rt->dst.dev);
	struct net_device *dev = skb->dev;
	struct iphdr *iph;
	int err;

	skb_scrub_packet(skb, xnet);

	skb_clear_hash_if_not_l4(skb);
	skb_dst_set(skb, &rt->dst);
	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));

	/* Push down and install the IP header. */
	skb_push(skb, sizeof(struct iphdr));
	skb_reset_network_header(skb);

	iph = ip_hdr(skb);

	iph->version	=	4;
	iph->ihl	=	sizeof(struct iphdr) >> 2;
	iph->frag_off	=	ip_mtu_locked(&rt->dst) ? 0 : df;
	iph->protocol	=	proto;
	iph->tos	=	tos;
	iph->daddr	=	dst;
	iph->saddr	=	src;
	iph->ttl	=	ttl;
	__ip_select_ident(net, iph, skb_shinfo(skb)->gso_segs ?: 1);

	err = ip_local_out(net, sk, skb);

	if (dev) {
		if (unlikely(net_xmit_eval(err)))
			pkt_len = 0;
		iptunnel_xmit_stats(dev, pkt_len);
	}
}
EXPORT_SYMBOL_GPL(iptunnel_xmit);

int __iptunnel_pull_header(struct sk_buff *skb, int hdr_len,
			   __be16 inner_proto, bool raw_proto, bool xnet)
{
	if (unlikely(!pskb_may_pull(skb, hdr_len)))
		return -ENOMEM;

	skb_pull_rcsum(skb, hdr_len);

	if (!raw_proto && inner_proto == htons(ETH_P_TEB)) {
		struct ethhdr *eh;

		if (unlikely(!pskb_may_pull(skb, ETH_HLEN)))
			return -ENOMEM;

		eh = (struct ethhdr *)skb->data;
		if (likely(eth_proto_is_802_3(eh->h_proto)))
			skb->protocol = eh->h_proto;
		else
			skb->protocol = htons(ETH_P_802_2);

	} else {
		skb->protocol = inner_proto;
	}

	skb_clear_hash_if_not_l4(skb);
	__vlan_hwaccel_clear_tag(skb);
	skb_set_queue_mapping(skb, 0);
	skb_scrub_packet(skb, xnet);

	return iptunnel_pull_offloads(skb);
}
EXPORT_SYMBOL_GPL(__iptunnel_pull_header);

struct metadata_dst *iptunnel_metadata_reply(struct metadata_dst *md,
					     gfp_t flags)
{
	struct metadata_dst *res;
	struct ip_tunnel_info *dst, *src;

	if (!md || md->type != METADATA_IP_TUNNEL ||
	    md->u.tun_info.mode & IP_TUNNEL_INFO_TX)

		return NULL;

	res = metadata_dst_alloc(0, METADATA_IP_TUNNEL, flags);
	if (!res)
		return NULL;

	dst = &res->u.tun_info;
	src = &md->u.tun_info;
	dst->key.tun_id = src->key.tun_id;
	if (src->mode & IP_TUNNEL_INFO_IPV6)
		memcpy(&dst->key.u.ipv6.dst, &src->key.u.ipv6.src,
		       sizeof(struct in6_addr));
	else
		dst->key.u.ipv4.dst = src->key.u.ipv4.src;
	dst->key.tun_flags = src->key.tun_flags;
	dst->mode = src->mode | IP_TUNNEL_INFO_TX;

	return res;
}
EXPORT_SYMBOL_GPL(iptunnel_metadata_reply);

int iptunnel_handle_offloads(struct sk_buff *skb,
			     int gso_type_mask)
{
	int err;

	if (likely(!skb->encapsulation)) {
		skb_reset_inner_headers(skb);
		skb->encapsulation = 1;
	}

	if (skb_is_gso(skb)) {
		err = skb_header_unclone(skb, GFP_ATOMIC);
		if (unlikely(err))
			return err;
		skb_shinfo(skb)->gso_type |= gso_type_mask;
		return 0;
	}

	if (skb->ip_summed != CHECKSUM_PARTIAL) {
		skb->ip_summed = CHECKSUM_NONE;
		/* We clear encapsulation here to prevent badly-written
		 * drivers potentially deciding to offload an inner checksum
		 * if we set CHECKSUM_PARTIAL on the outer header.
		 * This should go away when the drivers are all fixed.
		 */
		skb->encapsulation = 0;
	}

	return 0;
}
EXPORT_SYMBOL_GPL(iptunnel_handle_offloads);

/* Often modified stats are per cpu, other are shared (netdev->stats) */
void ip_tunnel_get_stats64(struct net_device *dev,
			   struct rtnl_link_stats64 *tot)
{
	int i;

	netdev_stats_to_stats64(tot, &dev->stats);

	for_each_possible_cpu(i) {
		const struct pcpu_sw_netstats *tstats =
						   per_cpu_ptr(dev->tstats, i);
		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
		unsigned int start;

		do {
			start = u64_stats_fetch_begin_irq(&tstats->syncp);
			rx_packets = tstats->rx_packets;
			tx_packets = tstats->tx_packets;
			rx_bytes = tstats->rx_bytes;
			tx_bytes = tstats->tx_bytes;
		} while (u64_stats_fetch_retry_irq(&tstats->syncp, start));

		tot->rx_packets += rx_packets;
		tot->tx_packets += tx_packets;
		tot->rx_bytes   += rx_bytes;
		tot->tx_bytes   += tx_bytes;
	}
}
EXPORT_SYMBOL_GPL(ip_tunnel_get_stats64);

static const struct nla_policy ip_tun_policy[LWTUNNEL_IP_MAX + 1] = {
	[LWTUNNEL_IP_ID]	= { .type = NLA_U64 },
	[LWTUNNEL_IP_DST]	= { .type = NLA_U32 },
	[LWTUNNEL_IP_SRC]	= { .type = NLA_U32 },
	[LWTUNNEL_IP_TTL]	= { .type = NLA_U8 },
	[LWTUNNEL_IP_TOS]	= { .type = NLA_U8 },
	[LWTUNNEL_IP_FLAGS]	= { .type = NLA_U16 },
};

static int ip_tun_build_state(struct nlattr *attr,
			      unsigned int family, const void *cfg,
			      struct lwtunnel_state **ts,
			      struct netlink_ext_ack *extack)
{
	struct ip_tunnel_info *tun_info;
	struct lwtunnel_state *new_state;
	struct nlattr *tb[LWTUNNEL_IP_MAX + 1];
	int err;

	err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP_MAX, attr,
					  ip_tun_policy, extack);
	if (err < 0)
		return err;

	new_state = lwtunnel_state_alloc(sizeof(*tun_info));
	if (!new_state)
		return -ENOMEM;

	new_state->type = LWTUNNEL_ENCAP_IP;

	tun_info = lwt_tun_info(new_state);

#ifdef CONFIG_DST_CACHE
	err = dst_cache_init(&tun_info->dst_cache, GFP_KERNEL);
	if (err) {
		lwtstate_free(new_state);
		return err;
	}
#endif

	if (tb[LWTUNNEL_IP_ID])
		tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP_ID]);

	if (tb[LWTUNNEL_IP_DST])
		tun_info->key.u.ipv4.dst = nla_get_in_addr(tb[LWTUNNEL_IP_DST]);

	if (tb[LWTUNNEL_IP_SRC])
		tun_info->key.u.ipv4.src = nla_get_in_addr(tb[LWTUNNEL_IP_SRC]);

	if (tb[LWTUNNEL_IP_TTL])
		tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP_TTL]);

	if (tb[LWTUNNEL_IP_TOS])
		tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP_TOS]);

	if (tb[LWTUNNEL_IP_FLAGS])
		tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP_FLAGS]);

	tun_info->mode = IP_TUNNEL_INFO_TX;
	tun_info->options_len = 0;

	*ts = new_state;

	return 0;
}

static void ip_tun_destroy_state(struct lwtunnel_state *lwtstate)
{
#ifdef CONFIG_DST_CACHE
	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);

	dst_cache_destroy(&tun_info->dst_cache);
#endif
}

static int ip_tun_fill_encap_info(struct sk_buff *skb,
				  struct lwtunnel_state *lwtstate)
{
	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);

	if (nla_put_be64(skb, LWTUNNEL_IP_ID, tun_info->key.tun_id,
			 LWTUNNEL_IP_PAD) ||
	    nla_put_in_addr(skb, LWTUNNEL_IP_DST, tun_info->key.u.ipv4.dst) ||
	    nla_put_in_addr(skb, LWTUNNEL_IP_SRC, tun_info->key.u.ipv4.src) ||
	    nla_put_u8(skb, LWTUNNEL_IP_TOS, tun_info->key.tos) ||
	    nla_put_u8(skb, LWTUNNEL_IP_TTL, tun_info->key.ttl) ||
	    nla_put_be16(skb, LWTUNNEL_IP_FLAGS, tun_info->key.tun_flags))
		return -ENOMEM;

	return 0;
}

static int ip_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
{
	return nla_total_size_64bit(8)	/* LWTUNNEL_IP_ID */
		+ nla_total_size(4)	/* LWTUNNEL_IP_DST */
		+ nla_total_size(4)	/* LWTUNNEL_IP_SRC */
		+ nla_total_size(1)	/* LWTUNNEL_IP_TOS */
		+ nla_total_size(1)	/* LWTUNNEL_IP_TTL */
		+ nla_total_size(2);	/* LWTUNNEL_IP_FLAGS */
}

static int ip_tun_cmp_encap(struct lwtunnel_state *a, struct lwtunnel_state *b)
{
	return memcmp(lwt_tun_info(a), lwt_tun_info(b),
		      sizeof(struct ip_tunnel_info));
}

static const struct lwtunnel_encap_ops ip_tun_lwt_ops = {
	.build_state = ip_tun_build_state,
	.destroy_state = ip_tun_destroy_state,
	.fill_encap = ip_tun_fill_encap_info,
	.get_encap_size = ip_tun_encap_nlsize,
	.cmp_encap = ip_tun_cmp_encap,
	.owner = THIS_MODULE,
};

static const struct nla_policy ip6_tun_policy[LWTUNNEL_IP6_MAX + 1] = {
	[LWTUNNEL_IP6_ID]		= { .type = NLA_U64 },
	[LWTUNNEL_IP6_DST]		= { .len = sizeof(struct in6_addr) },
	[LWTUNNEL_IP6_SRC]		= { .len = sizeof(struct in6_addr) },
	[LWTUNNEL_IP6_HOPLIMIT]		= { .type = NLA_U8 },
	[LWTUNNEL_IP6_TC]		= { .type = NLA_U8 },
	[LWTUNNEL_IP6_FLAGS]		= { .type = NLA_U16 },
};

static int ip6_tun_build_state(struct nlattr *attr,
			       unsigned int family, const void *cfg,
			       struct lwtunnel_state **ts,
			       struct netlink_ext_ack *extack)
{
	struct ip_tunnel_info *tun_info;
	struct lwtunnel_state *new_state;
	struct nlattr *tb[LWTUNNEL_IP6_MAX + 1];
	int err;

	err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP6_MAX, attr,
					  ip6_tun_policy, extack);
	if (err < 0)
		return err;

	new_state = lwtunnel_state_alloc(sizeof(*tun_info));
	if (!new_state)
		return -ENOMEM;

	new_state->type = LWTUNNEL_ENCAP_IP6;

	tun_info = lwt_tun_info(new_state);

	if (tb[LWTUNNEL_IP6_ID])
		tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP6_ID]);

	if (tb[LWTUNNEL_IP6_DST])
		tun_info->key.u.ipv6.dst = nla_get_in6_addr(tb[LWTUNNEL_IP6_DST]);

	if (tb[LWTUNNEL_IP6_SRC])
		tun_info->key.u.ipv6.src = nla_get_in6_addr(tb[LWTUNNEL_IP6_SRC]);

	if (tb[LWTUNNEL_IP6_HOPLIMIT])
		tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP6_HOPLIMIT]);

	if (tb[LWTUNNEL_IP6_TC])
		tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP6_TC]);

	if (tb[LWTUNNEL_IP6_FLAGS])
		tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP6_FLAGS]);

	tun_info->mode = IP_TUNNEL_INFO_TX | IP_TUNNEL_INFO_IPV6;
	tun_info->options_len = 0;

	*ts = new_state;

	return 0;
}

static int ip6_tun_fill_encap_info(struct sk_buff *skb,
				   struct lwtunnel_state *lwtstate)
{
	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);

	if (nla_put_be64(skb, LWTUNNEL_IP6_ID, tun_info->key.tun_id,
			 LWTUNNEL_IP6_PAD) ||
	    nla_put_in6_addr(skb, LWTUNNEL_IP6_DST, &tun_info->key.u.ipv6.dst) ||
	    nla_put_in6_addr(skb, LWTUNNEL_IP6_SRC, &tun_info->key.u.ipv6.src) ||
	    nla_put_u8(skb, LWTUNNEL_IP6_TC, tun_info->key.tos) ||
	    nla_put_u8(skb, LWTUNNEL_IP6_HOPLIMIT, tun_info->key.ttl) ||
	    nla_put_be16(skb, LWTUNNEL_IP6_FLAGS, tun_info->key.tun_flags))
		return -ENOMEM;

	return 0;
}

static int ip6_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
{
	return nla_total_size_64bit(8)	/* LWTUNNEL_IP6_ID */
		+ nla_total_size(16)	/* LWTUNNEL_IP6_DST */
		+ nla_total_size(16)	/* LWTUNNEL_IP6_SRC */
		+ nla_total_size(1)	/* LWTUNNEL_IP6_HOPLIMIT */
		+ nla_total_size(1)	/* LWTUNNEL_IP6_TC */
		+ nla_total_size(2);	/* LWTUNNEL_IP6_FLAGS */
}

static const struct lwtunnel_encap_ops ip6_tun_lwt_ops = {
	.build_state = ip6_tun_build_state,
	.fill_encap = ip6_tun_fill_encap_info,
	.get_encap_size = ip6_tun_encap_nlsize,
	.cmp_encap = ip_tun_cmp_encap,
	.owner = THIS_MODULE,
};

void __init ip_tunnel_core_init(void)
{
	/* If you land here, make sure whether increasing ip_tunnel_info's
	 * options_len is a reasonable choice with its usage in front ends
	 * (f.e., it's part of flow keys, etc).
	 */
	BUILD_BUG_ON(IP_TUNNEL_OPTS_MAX != 255);

	lwtunnel_encap_add_ops(&ip_tun_lwt_ops, LWTUNNEL_ENCAP_IP);
	lwtunnel_encap_add_ops(&ip6_tun_lwt_ops, LWTUNNEL_ENCAP_IP6);
}

DEFINE_STATIC_KEY_FALSE(ip_tunnel_metadata_cnt);
EXPORT_SYMBOL(ip_tunnel_metadata_cnt);

void ip_tunnel_need_metadata(void)
{
	static_branch_inc(&ip_tunnel_metadata_cnt);
}
EXPORT_SYMBOL_GPL(ip_tunnel_need_metadata);

void ip_tunnel_unneed_metadata(void)
{
	static_branch_dec(&ip_tunnel_metadata_cnt);
}
EXPORT_SYMBOL_GPL(ip_tunnel_unneed_metadata);
Commit	Line	Data
c9422999	1	// SPDX-License-Identifier: GPL-2.0-only
0e6fbc5b PS	2	/*
0e6fbc5b PS	3	* Copyright (c) 2013 Nicira, Inc.
0e6fbc5b PS	4	*/
	5
	6	#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
	7
	8	#include <linux/types.h>
	9	#include <linux/kernel.h>
	10	#include <linux/skbuff.h>
	11	#include <linux/netdevice.h>
	12	#include <linux/in.h>
	13	#include <linux/if_arp.h>
0e6fbc5b PS	14	#include <linux/init.h>
	15	#include <linux/in6.h>
	16	#include <linux/inetdevice.h>
	17	#include <linux/netfilter_ipv4.h>
	18	#include <linux/etherdevice.h>
	19	#include <linux/if_ether.h>
	20	#include <linux/if_vlan.h>
e7030878	21	#include <linux/static_key.h>
0e6fbc5b PS	22
	23	#include <net/ip.h>
	24	#include <net/icmp.h>
	25	#include <net/protocol.h>
	26	#include <net/ip_tunnels.h>
058214a4	27	#include <net/ip6_tunnel.h>
0e6fbc5b PS	28	#include <net/arp.h>
	29	#include <net/checksum.h>
	30	#include <net/dsfield.h>
	31	#include <net/inet_ecn.h>
	32	#include <net/xfrm.h>
	33	#include <net/net_namespace.h>
	34	#include <net/netns/generic.h>
	35	#include <net/rtnetlink.h>
63d008a4	36	#include <net/dst_metadata.h>
0e6fbc5b	37
55c2bc14 TH	38	const struct ip_tunnel_encap_ops __rcu *
	39	iptun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
	40	EXPORT_SYMBOL(iptun_encaps);
	41
058214a4 TH	42	const struct ip6_tnl_encap_ops __rcu *
	43	ip6tun_encaps[MAX_IPTUN_ENCAP_OPS] __read_mostly;
	44	EXPORT_SYMBOL(ip6tun_encaps);
	45
039f5062 PS	46	void iptunnel_xmit(struct sock sk, struct rtable rt, struct sk_buff *skb,
	47	__be32 src, __be32 dst, __u8 proto,
	48	__u8 tos, __u8 ttl, __be16 df, bool xnet)
0e6fbc5b	49	{
bc22a0e2	50	int pkt_len = skb->len - skb_inner_network_offset(skb);
f859b0f6	51	struct net *net = dev_net(rt->dst.dev);
039f5062	52	struct net_device *dev = skb->dev;
0e6fbc5b PS	53	struct iphdr *iph;
	54	int err;
	55
963a88b3 ND	56	skb_scrub_packet(skb, xnet);
963a88b3 ND	57
bf8d85d4	58	skb_clear_hash_if_not_l4(skb);
0e6fbc5b PS	59	skb_dst_set(skb, &rt->dst);
	60	memset(IPCB(skb), 0, sizeof(*IPCB(skb)));
	61
	62	/* Push down and install the IP header. */
78a3694d	63	skb_push(skb, sizeof(struct iphdr));
0e6fbc5b PS	64	skb_reset_network_header(skb);
	65
	66	iph = ip_hdr(skb);
	67
	68	iph->version = 4;
	69	iph->ihl = sizeof(struct iphdr) >> 2;
16f7eb2b	70	iph->frag_off = ip_mtu_locked(&rt->dst) ? 0 : df;
0e6fbc5b PS	71	iph->protocol = proto;
	72	iph->tos = tos;
	73	iph->daddr = dst;
	74	iph->saddr = src;
	75	iph->ttl = ttl;
f859b0f6	76	__ip_select_ident(net, iph, skb_shinfo(skb)->gso_segs ?: 1);
0e6fbc5b	77
33224b16	78	err = ip_local_out(net, sk, skb);
5684abf7 XL	79
	80	if (dev) {
	81	if (unlikely(net_xmit_eval(err)))
	82	pkt_len = 0;
	83	iptunnel_xmit_stats(dev, pkt_len);
	84	}
0e6fbc5b PS	85	}
0e6fbc5b PS	86	EXPORT_SYMBOL_GPL(iptunnel_xmit);
3d7b46cd	87
a6d5bbf3 JB	88	int __iptunnel_pull_header(struct sk_buff *skb, int hdr_len,
a6d5bbf3 JB	89	__be16 inner_proto, bool raw_proto, bool xnet)
3d7b46cd PS	90	{
	91	if (unlikely(!pskb_may_pull(skb, hdr_len)))
	92	return -ENOMEM;
	93
	94	skb_pull_rcsum(skb, hdr_len);
	95
a6d5bbf3	96	if (!raw_proto && inner_proto == htons(ETH_P_TEB)) {
1245dfc8	97	struct ethhdr *eh;
3d7b46cd PS	98
	99	if (unlikely(!pskb_may_pull(skb, ETH_HLEN)))
	100	return -ENOMEM;
	101
1245dfc8	102	eh = (struct ethhdr *)skb->data;
d181ddca	103	if (likely(eth_proto_is_802_3(eh->h_proto)))
3d7b46cd PS	104	skb->protocol = eh->h_proto;
	105	else
	106	skb->protocol = htons(ETH_P_802_2);
	107
	108	} else {
	109	skb->protocol = inner_proto;
	110	}
	111
7539fadc	112	skb_clear_hash_if_not_l4(skb);
3e2ed0c2	113	__vlan_hwaccel_clear_tag(skb);
3d7b46cd	114	skb_set_queue_mapping(skb, 0);
7f290c94	115	skb_scrub_packet(skb, xnet);
a09a4c8d JG	116
a09a4c8d JG	117	return iptunnel_pull_offloads(skb);
3d7b46cd	118	}
a6d5bbf3	119	EXPORT_SYMBOL_GPL(__iptunnel_pull_header);
2d26f0a3	120
63d008a4 JB	121	struct metadata_dst iptunnel_metadata_reply(struct metadata_dst md,
	122	gfp_t flags)
	123	{
	124	struct metadata_dst *res;
	125	struct ip_tunnel_info dst, src;
	126
3fcece12 JK	127	if (!md \|\| md->type != METADATA_IP_TUNNEL \|\|
	128	md->u.tun_info.mode & IP_TUNNEL_INFO_TX)
	129
63d008a4 JB	130	return NULL;
63d008a4 JB	131
3fcece12	132	res = metadata_dst_alloc(0, METADATA_IP_TUNNEL, flags);
63d008a4 JB	133	if (!res)
	134	return NULL;
	135
	136	dst = &res->u.tun_info;
	137	src = &md->u.tun_info;
	138	dst->key.tun_id = src->key.tun_id;
	139	if (src->mode & IP_TUNNEL_INFO_IPV6)
	140	memcpy(&dst->key.u.ipv6.dst, &src->key.u.ipv6.src,
	141	sizeof(struct in6_addr));
	142	else
	143	dst->key.u.ipv4.dst = src->key.u.ipv4.src;
7bdca378	144	dst->key.tun_flags = src->key.tun_flags;
63d008a4 JB	145	dst->mode = src->mode \| IP_TUNNEL_INFO_TX;
	146
	147	return res;
	148	}
	149	EXPORT_SYMBOL_GPL(iptunnel_metadata_reply);
	150
aed069df AD	151	int iptunnel_handle_offloads(struct sk_buff *skb,
aed069df AD	152	int gso_type_mask)
2d26f0a3 ED	153	{
	154	int err;
	155
	156	if (likely(!skb->encapsulation)) {
	157	skb_reset_inner_headers(skb);
	158	skb->encapsulation = 1;
	159	}
	160
	161	if (skb_is_gso(skb)) {
9580bf2e	162	err = skb_header_unclone(skb, GFP_ATOMIC);
2d26f0a3	163	if (unlikely(err))
aed069df	164	return err;
2d26f0a3	165	skb_shinfo(skb)->gso_type \|= gso_type_mask;
aed069df	166	return 0;
2d26f0a3 ED	167	}
2d26f0a3 ED	168
6fa79666	169	if (skb->ip_summed != CHECKSUM_PARTIAL) {
2d26f0a3	170	skb->ip_summed = CHECKSUM_NONE;
6fa79666 EC	171	/* We clear encapsulation here to prevent badly-written
	172	* drivers potentially deciding to offload an inner checksum
	173	* if we set CHECKSUM_PARTIAL on the outer header.
	174	* This should go away when the drivers are all fixed.
	175	*/
179bc67f EC	176	skb->encapsulation = 0;
179bc67f EC	177	}
2d26f0a3	178
aed069df	179	return 0;
2d26f0a3 ED	180	}
2d26f0a3 ED	181	EXPORT_SYMBOL_GPL(iptunnel_handle_offloads);
ebe44f35 DM	182
ebe44f35 DM	183	/* Often modified stats are per cpu, other are shared (netdev->stats) */
bc1f4470	184	void ip_tunnel_get_stats64(struct net_device *dev,
bc1f4470	185	struct rtnl_link_stats64 *tot)
ebe44f35 DM	186	{
	187	int i;
	188
c24a5964 AD	189	netdev_stats_to_stats64(tot, &dev->stats);
c24a5964 AD	190
ebe44f35 DM	191	for_each_possible_cpu(i) {
	192	const struct pcpu_sw_netstats *tstats =
	193	per_cpu_ptr(dev->tstats, i);
	194	u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
	195	unsigned int start;
	196
	197	do {
57a7744e	198	start = u64_stats_fetch_begin_irq(&tstats->syncp);
ebe44f35 DM	199	rx_packets = tstats->rx_packets;
	200	tx_packets = tstats->tx_packets;
	201	rx_bytes = tstats->rx_bytes;
	202	tx_bytes = tstats->tx_bytes;
57a7744e	203	} while (u64_stats_fetch_retry_irq(&tstats->syncp, start));
ebe44f35 DM	204
	205	tot->rx_packets += rx_packets;
	206	tot->tx_packets += tx_packets;
	207	tot->rx_bytes += rx_bytes;
	208	tot->tx_bytes += tx_bytes;
	209	}
ebe44f35 DM	210	}
ebe44f35 DM	211	EXPORT_SYMBOL_GPL(ip_tunnel_get_stats64);
3093fbe7	212
a1c234f9 JB	213	static const struct nla_policy ip_tun_policy[LWTUNNEL_IP_MAX + 1] = {
	214	[LWTUNNEL_IP_ID] = { .type = NLA_U64 },
	215	[LWTUNNEL_IP_DST] = { .type = NLA_U32 },
	216	[LWTUNNEL_IP_SRC] = { .type = NLA_U32 },
	217	[LWTUNNEL_IP_TTL] = { .type = NLA_U8 },
	218	[LWTUNNEL_IP_TOS] = { .type = NLA_U8 },
a1c234f9	219	[LWTUNNEL_IP_FLAGS] = { .type = NLA_U16 },
3093fbe7 TG	220	};
3093fbe7 TG	221
30357d7d	222	static int ip_tun_build_state(struct nlattr *attr,
127eb7cd	223	unsigned int family, const void *cfg,
9ae28727 DA	224	struct lwtunnel_state **ts,
9ae28727 DA	225	struct netlink_ext_ack *extack)
3093fbe7 TG	226	{
	227	struct ip_tunnel_info *tun_info;
	228	struct lwtunnel_state *new_state;
a1c234f9	229	struct nlattr *tb[LWTUNNEL_IP_MAX + 1];
3093fbe7 TG	230	int err;
3093fbe7 TG	231
8cb08174 JB	232	err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP_MAX, attr,
8cb08174 JB	233	ip_tun_policy, extack);
3093fbe7 TG	234	if (err < 0)
	235	return err;
	236
	237	new_state = lwtunnel_state_alloc(sizeof(*tun_info));
	238	if (!new_state)
	239	return -ENOMEM;
	240
	241	new_state->type = LWTUNNEL_ENCAP_IP;
	242
	243	tun_info = lwt_tun_info(new_state);
	244
3d25eabb	245	#ifdef CONFIG_DST_CACHE
	246	err = dst_cache_init(&tun_info->dst_cache, GFP_KERNEL);
	247	if (err) {
	248	lwtstate_free(new_state);
	249	return err;
	250	}
	251	#endif
	252
a1c234f9	253	if (tb[LWTUNNEL_IP_ID])
30d3d83a	254	tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP_ID]);
3093fbe7	255
a1c234f9	256	if (tb[LWTUNNEL_IP_DST])
7822ce73	257	tun_info->key.u.ipv4.dst = nla_get_in_addr(tb[LWTUNNEL_IP_DST]);
3093fbe7	258
a1c234f9	259	if (tb[LWTUNNEL_IP_SRC])
7822ce73	260	tun_info->key.u.ipv4.src = nla_get_in_addr(tb[LWTUNNEL_IP_SRC]);
3093fbe7	261
a1c234f9	262	if (tb[LWTUNNEL_IP_TTL])
7c383fb2	263	tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP_TTL]);
3093fbe7	264
a1c234f9	265	if (tb[LWTUNNEL_IP_TOS])
7c383fb2	266	tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP_TOS]);
3093fbe7	267
a1c234f9	268	if (tb[LWTUNNEL_IP_FLAGS])
30d3d83a	269	tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP_FLAGS]);
3093fbe7 TG	270
3093fbe7 TG	271	tun_info->mode = IP_TUNNEL_INFO_TX;
3093fbe7 TG	272	tun_info->options_len = 0;
	273
	274	*ts = new_state;
	275
	276	return 0;
	277	}
	278
3d25eabb	279	static void ip_tun_destroy_state(struct lwtunnel_state *lwtstate)
	280	{
	281	#ifdef CONFIG_DST_CACHE
	282	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
	283
	284	dst_cache_destroy(&tun_info->dst_cache);
	285	#endif
	286	}
	287
3093fbe7 TG	288	static int ip_tun_fill_encap_info(struct sk_buff *skb,
	289	struct lwtunnel_state *lwtstate)
	290	{
	291	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
	292
b46f6ded ND	293	if (nla_put_be64(skb, LWTUNNEL_IP_ID, tun_info->key.tun_id,
b46f6ded ND	294	LWTUNNEL_IP_PAD) \|\|
7822ce73 HY	295	nla_put_in_addr(skb, LWTUNNEL_IP_DST, tun_info->key.u.ipv4.dst) \|\|
7822ce73 HY	296	nla_put_in_addr(skb, LWTUNNEL_IP_SRC, tun_info->key.u.ipv4.src) \|\|
7c383fb2 JB	297	nla_put_u8(skb, LWTUNNEL_IP_TOS, tun_info->key.tos) \|\|
7c383fb2 JB	298	nla_put_u8(skb, LWTUNNEL_IP_TTL, tun_info->key.ttl) \|\|
30d3d83a	299	nla_put_be16(skb, LWTUNNEL_IP_FLAGS, tun_info->key.tun_flags))
3093fbe7 TG	300	return -ENOMEM;
	301
	302	return 0;
	303	}
	304
	305	static int ip_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
	306	{
b46f6ded	307	return nla_total_size_64bit(8) /* LWTUNNEL_IP_ID */
a1c234f9 JB	308	+ nla_total_size(4) /* LWTUNNEL_IP_DST */
	309	+ nla_total_size(4) /* LWTUNNEL_IP_SRC */
	310	+ nla_total_size(1) /* LWTUNNEL_IP_TOS */
	311	+ nla_total_size(1) /* LWTUNNEL_IP_TTL */
a1c234f9	312	+ nla_total_size(2); /* LWTUNNEL_IP_FLAGS */
3093fbe7 TG	313	}
3093fbe7 TG	314
2d798499 JB	315	static int ip_tun_cmp_encap(struct lwtunnel_state a, struct lwtunnel_state b)
	316	{
	317	return memcmp(lwt_tun_info(a), lwt_tun_info(b),
	318	sizeof(struct ip_tunnel_info));
	319	}
	320
3093fbe7 TG	321	static const struct lwtunnel_encap_ops ip_tun_lwt_ops = {
3093fbe7 TG	322	.build_state = ip_tun_build_state,
3d25eabb	323	.destroy_state = ip_tun_destroy_state,
3093fbe7 TG	324	.fill_encap = ip_tun_fill_encap_info,
3093fbe7 TG	325	.get_encap_size = ip_tun_encap_nlsize,
2d798499	326	.cmp_encap = ip_tun_cmp_encap,
88ff7334	327	.owner = THIS_MODULE,
3093fbe7 TG	328	};
3093fbe7 TG	329
32a2b002 JB	330	static const struct nla_policy ip6_tun_policy[LWTUNNEL_IP6_MAX + 1] = {
	331	[LWTUNNEL_IP6_ID] = { .type = NLA_U64 },
	332	[LWTUNNEL_IP6_DST] = { .len = sizeof(struct in6_addr) },
	333	[LWTUNNEL_IP6_SRC] = { .len = sizeof(struct in6_addr) },
	334	[LWTUNNEL_IP6_HOPLIMIT] = { .type = NLA_U8 },
	335	[LWTUNNEL_IP6_TC] = { .type = NLA_U8 },
32a2b002 JB	336	[LWTUNNEL_IP6_FLAGS] = { .type = NLA_U16 },
	337	};
	338
30357d7d	339	static int ip6_tun_build_state(struct nlattr *attr,
127eb7cd	340	unsigned int family, const void *cfg,
9ae28727 DA	341	struct lwtunnel_state **ts,
9ae28727 DA	342	struct netlink_ext_ack *extack)
32a2b002 JB	343	{
	344	struct ip_tunnel_info *tun_info;
	345	struct lwtunnel_state *new_state;
	346	struct nlattr *tb[LWTUNNEL_IP6_MAX + 1];
	347	int err;
	348
8cb08174 JB	349	err = nla_parse_nested_deprecated(tb, LWTUNNEL_IP6_MAX, attr,
8cb08174 JB	350	ip6_tun_policy, extack);
32a2b002 JB	351	if (err < 0)
	352	return err;
	353
	354	new_state = lwtunnel_state_alloc(sizeof(*tun_info));
	355	if (!new_state)
	356	return -ENOMEM;
	357
	358	new_state->type = LWTUNNEL_ENCAP_IP6;
	359
	360	tun_info = lwt_tun_info(new_state);
	361
	362	if (tb[LWTUNNEL_IP6_ID])
30d3d83a	363	tun_info->key.tun_id = nla_get_be64(tb[LWTUNNEL_IP6_ID]);
32a2b002 JB	364
	365	if (tb[LWTUNNEL_IP6_DST])
	366	tun_info->key.u.ipv6.dst = nla_get_in6_addr(tb[LWTUNNEL_IP6_DST]);
	367
	368	if (tb[LWTUNNEL_IP6_SRC])
	369	tun_info->key.u.ipv6.src = nla_get_in6_addr(tb[LWTUNNEL_IP6_SRC]);
	370
	371	if (tb[LWTUNNEL_IP6_HOPLIMIT])
	372	tun_info->key.ttl = nla_get_u8(tb[LWTUNNEL_IP6_HOPLIMIT]);
	373
	374	if (tb[LWTUNNEL_IP6_TC])
	375	tun_info->key.tos = nla_get_u8(tb[LWTUNNEL_IP6_TC]);
	376
32a2b002	377	if (tb[LWTUNNEL_IP6_FLAGS])
30d3d83a	378	tun_info->key.tun_flags = nla_get_be16(tb[LWTUNNEL_IP6_FLAGS]);
32a2b002	379
7f9562a1	380	tun_info->mode = IP_TUNNEL_INFO_TX \| IP_TUNNEL_INFO_IPV6;
32a2b002 JB	381	tun_info->options_len = 0;
	382
	383	*ts = new_state;
	384
	385	return 0;
	386	}
	387
	388	static int ip6_tun_fill_encap_info(struct sk_buff *skb,
	389	struct lwtunnel_state *lwtstate)
	390	{
	391	struct ip_tunnel_info *tun_info = lwt_tun_info(lwtstate);
	392
b46f6ded ND	393	if (nla_put_be64(skb, LWTUNNEL_IP6_ID, tun_info->key.tun_id,
b46f6ded ND	394	LWTUNNEL_IP6_PAD) \|\|
32a2b002 JB	395	nla_put_in6_addr(skb, LWTUNNEL_IP6_DST, &tun_info->key.u.ipv6.dst) \|\|
32a2b002 JB	396	nla_put_in6_addr(skb, LWTUNNEL_IP6_SRC, &tun_info->key.u.ipv6.src) \|\|
995096a0 QA	397	nla_put_u8(skb, LWTUNNEL_IP6_TC, tun_info->key.tos) \|\|
995096a0 QA	398	nla_put_u8(skb, LWTUNNEL_IP6_HOPLIMIT, tun_info->key.ttl) \|\|
30d3d83a	399	nla_put_be16(skb, LWTUNNEL_IP6_FLAGS, tun_info->key.tun_flags))
32a2b002 JB	400	return -ENOMEM;
	401
	402	return 0;
	403	}
	404
	405	static int ip6_tun_encap_nlsize(struct lwtunnel_state *lwtstate)
	406	{
b46f6ded	407	return nla_total_size_64bit(8) /* LWTUNNEL_IP6_ID */
32a2b002 JB	408	+ nla_total_size(16) /* LWTUNNEL_IP6_DST */
	409	+ nla_total_size(16) /* LWTUNNEL_IP6_SRC */
	410	+ nla_total_size(1) /* LWTUNNEL_IP6_HOPLIMIT */
	411	+ nla_total_size(1) /* LWTUNNEL_IP6_TC */
32a2b002 JB	412	+ nla_total_size(2); /* LWTUNNEL_IP6_FLAGS */
	413	}
	414
	415	static const struct lwtunnel_encap_ops ip6_tun_lwt_ops = {
	416	.build_state = ip6_tun_build_state,
	417	.fill_encap = ip6_tun_fill_encap_info,
	418	.get_encap_size = ip6_tun_encap_nlsize,
	419	.cmp_encap = ip_tun_cmp_encap,
88ff7334	420	.owner = THIS_MODULE,
32a2b002 JB	421	};
32a2b002 JB	422
045a0fa0	423	void __init ip_tunnel_core_init(void)
3093fbe7	424	{
fca5fdf6 DB	425	/* If you land here, make sure whether increasing ip_tunnel_info's
	426	* options_len is a reasonable choice with its usage in front ends
	427	* (f.e., it's part of flow keys, etc).
	428	*/
	429	BUILD_BUG_ON(IP_TUNNEL_OPTS_MAX != 255);
	430
3093fbe7	431	lwtunnel_encap_add_ops(&ip_tun_lwt_ops, LWTUNNEL_ENCAP_IP);
32a2b002	432	lwtunnel_encap_add_ops(&ip6_tun_lwt_ops, LWTUNNEL_ENCAP_IP6);
3093fbe7	433	}
e7030878	434
5263a98f	435	DEFINE_STATIC_KEY_FALSE(ip_tunnel_metadata_cnt);
e7030878 TG	436	EXPORT_SYMBOL(ip_tunnel_metadata_cnt);
	437
	438	void ip_tunnel_need_metadata(void)
	439	{
5263a98f	440	static_branch_inc(&ip_tunnel_metadata_cnt);
e7030878 TG	441	}
	442	EXPORT_SYMBOL_GPL(ip_tunnel_need_metadata);
	443
	444	void ip_tunnel_unneed_metadata(void)
	445	{
5263a98f	446	static_branch_dec(&ip_tunnel_metadata_cnt);
e7030878 TG	447	}
e7030878 TG	448	EXPORT_SYMBOL_GPL(ip_tunnel_unneed_metadata);