1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * ip_vs_xmit.c: various packet transmitters for IPVS
5 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
6 * Julian Anastasov <ja@ssi.bg>
10 * Description of forwarding methods:
11 * - all transmitters are called from LOCAL_IN (remote clients) and
12 * LOCAL_OUT (local clients) but for ICMP can be called from FORWARD
13 * - not all connections have destination server, for example,
14 * connections in backup server when fwmark is used
15 * - bypass connections use daddr from packet
16 * - we can use dst without ref while sending in RCU section, we use
17 * ref when returning NF_ACCEPT for NAT-ed packet via loopback
19 * - skb->dev is NULL, skb->protocol is not set (both are set in POST_ROUTING)
20 * - skb->pkt_type is not set yet
21 * - the only place where we can see skb->sk != NULL
24 #define KMSG_COMPONENT "IPVS"
25 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
27 #include <linux/kernel.h>
28 #include <linux/slab.h>
29 #include <linux/tcp.h> /* for tcphdr */
33 #include <net/tcp.h> /* for csum_tcpudp_magic */
35 #include <net/icmp.h> /* for icmp_send */
36 #include <net/route.h> /* for ip_route_output */
38 #include <net/ip6_route.h>
39 #include <net/ip_tunnels.h>
40 #include <net/ip6_checksum.h>
41 #include <net/addrconf.h>
42 #include <linux/icmpv6.h>
43 #include <linux/netfilter.h>
44 #include <linux/netfilter_ipv4.h>
46 #include <net/ip_vs.h>
49 IP_VS_RT_MODE_LOCAL = 1, /* Allow local dest */
50 IP_VS_RT_MODE_NON_LOCAL = 2, /* Allow non-local dest */
51 IP_VS_RT_MODE_RDR = 4, /* Allow redirect from remote daddr to
54 IP_VS_RT_MODE_CONNECT = 8, /* Always bind route to saddr */
55 IP_VS_RT_MODE_KNOWN_NH = 16,/* Route via remote addr */
56 IP_VS_RT_MODE_TUNNEL = 32,/* Tunnel mode */
59 static inline struct ip_vs_dest_dst *ip_vs_dest_dst_alloc(void)
61 return kmalloc(sizeof(struct ip_vs_dest_dst), GFP_ATOMIC);
64 static inline void ip_vs_dest_dst_free(struct ip_vs_dest_dst *dest_dst)
70 * Destination cache to speed up outgoing route lookup
73 __ip_vs_dst_set(struct ip_vs_dest *dest, struct ip_vs_dest_dst *dest_dst,
74 struct dst_entry *dst, u32 dst_cookie)
76 struct ip_vs_dest_dst *old;
78 old = rcu_dereference_protected(dest->dest_dst,
79 lockdep_is_held(&dest->dst_lock));
82 dest_dst->dst_cache = dst;
83 dest_dst->dst_cookie = dst_cookie;
85 rcu_assign_pointer(dest->dest_dst, dest_dst);
88 call_rcu(&old->rcu_head, ip_vs_dest_dst_rcu_free);
91 static inline struct ip_vs_dest_dst *
92 __ip_vs_dst_check(struct ip_vs_dest *dest)
94 struct ip_vs_dest_dst *dest_dst = rcu_dereference(dest->dest_dst);
95 struct dst_entry *dst;
99 dst = dest_dst->dst_cache;
101 dst->ops->check(dst, dest_dst->dst_cookie) == NULL)
107 __mtu_check_toobig_v6(const struct sk_buff *skb, u32 mtu)
109 if (IP6CB(skb)->frag_max_size) {
110 /* frag_max_size tell us that, this packet have been
111 * defragmented by netfilter IPv6 conntrack module.
113 if (IP6CB(skb)->frag_max_size > mtu)
114 return true; /* largest fragment violate MTU */
116 else if (skb->len > mtu && !skb_is_gso(skb)) {
117 return true; /* Packet size violate MTU size */
122 /* Get route to daddr, update *saddr, optionally bind route to saddr */
123 static struct rtable *do_output_route4(struct net *net, __be32 daddr,
124 int rt_mode, __be32 *saddr)
130 memset(&fl4, 0, sizeof(fl4));
132 fl4.flowi4_flags = (rt_mode & IP_VS_RT_MODE_KNOWN_NH) ?
133 FLOWI_FLAG_KNOWN_NH : 0;
136 rt = ip_route_output_key(net, &fl4);
138 /* Invalid saddr ? */
139 if (PTR_ERR(rt) == -EINVAL && *saddr &&
140 rt_mode & IP_VS_RT_MODE_CONNECT && !loop) {
142 flowi4_update_output(&fl4, 0, 0, daddr, 0);
145 IP_VS_DBG_RL("ip_route_output error, dest: %pI4\n", &daddr);
147 } else if (!*saddr && rt_mode & IP_VS_RT_MODE_CONNECT && fl4.saddr) {
150 flowi4_update_output(&fl4, 0, 0, daddr, fl4.saddr);
158 #ifdef CONFIG_IP_VS_IPV6
159 static inline int __ip_vs_is_local_route6(struct rt6_info *rt)
161 return rt->dst.dev && rt->dst.dev->flags & IFF_LOOPBACK;
165 static inline bool crosses_local_route_boundary(int skb_af, struct sk_buff *skb,
167 bool new_rt_is_local)
169 bool rt_mode_allow_local = !!(rt_mode & IP_VS_RT_MODE_LOCAL);
170 bool rt_mode_allow_non_local = !!(rt_mode & IP_VS_RT_MODE_NON_LOCAL);
171 bool rt_mode_allow_redirect = !!(rt_mode & IP_VS_RT_MODE_RDR);
172 bool source_is_loopback;
173 bool old_rt_is_local;
175 #ifdef CONFIG_IP_VS_IPV6
176 if (skb_af == AF_INET6) {
177 int addr_type = ipv6_addr_type(&ipv6_hdr(skb)->saddr);
180 (!skb->dev || skb->dev->flags & IFF_LOOPBACK) &&
181 (addr_type & IPV6_ADDR_LOOPBACK);
182 old_rt_is_local = __ip_vs_is_local_route6(
183 (struct rt6_info *)skb_dst(skb));
187 source_is_loopback = ipv4_is_loopback(ip_hdr(skb)->saddr);
188 old_rt_is_local = skb_rtable(skb)->rt_flags & RTCF_LOCAL;
191 if (unlikely(new_rt_is_local)) {
192 if (!rt_mode_allow_local)
194 if (!rt_mode_allow_redirect && !old_rt_is_local)
197 if (!rt_mode_allow_non_local)
199 if (source_is_loopback)
205 static inline void maybe_update_pmtu(int skb_af, struct sk_buff *skb, int mtu)
207 struct sock *sk = skb->sk;
208 struct rtable *ort = skb_rtable(skb);
210 if (!skb->dev && sk && sk_fullsock(sk))
211 ort->dst.ops->update_pmtu(&ort->dst, sk, NULL, mtu, true);
214 static inline bool ensure_mtu_is_adequate(struct netns_ipvs *ipvs, int skb_af,
216 struct ip_vs_iphdr *ipvsh,
217 struct sk_buff *skb, int mtu)
219 #ifdef CONFIG_IP_VS_IPV6
220 if (skb_af == AF_INET6) {
221 struct net *net = ipvs->net;
223 if (unlikely(__mtu_check_toobig_v6(skb, mtu))) {
225 skb->dev = net->loopback_dev;
226 /* only send ICMP too big on first fragment */
227 if (!ipvsh->fragoffs && !ip_vs_iph_icmp(ipvsh))
228 icmpv6_send(skb, ICMPV6_PKT_TOOBIG, 0, mtu);
229 IP_VS_DBG(1, "frag needed for %pI6c\n",
230 &ipv6_hdr(skb)->saddr);
236 /* If we're going to tunnel the packet and pmtu discovery
237 * is disabled, we'll just fragment it anyway
239 if ((rt_mode & IP_VS_RT_MODE_TUNNEL) && !sysctl_pmtu_disc(ipvs))
242 if (unlikely(ip_hdr(skb)->frag_off & htons(IP_DF) &&
243 skb->len > mtu && !skb_is_gso(skb) &&
244 !ip_vs_iph_icmp(ipvsh))) {
245 icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED,
247 IP_VS_DBG(1, "frag needed for %pI4\n",
248 &ip_hdr(skb)->saddr);
256 static inline bool decrement_ttl(struct netns_ipvs *ipvs,
260 struct net *net = ipvs->net;
262 #ifdef CONFIG_IP_VS_IPV6
263 if (skb_af == AF_INET6) {
264 struct dst_entry *dst = skb_dst(skb);
266 /* check and decrement ttl */
267 if (ipv6_hdr(skb)->hop_limit <= 1) {
268 struct inet6_dev *idev = __in6_dev_get_safely(skb->dev);
270 /* Force OUTPUT device used as source address */
272 icmpv6_send(skb, ICMPV6_TIME_EXCEED,
273 ICMPV6_EXC_HOPLIMIT, 0);
274 __IP6_INC_STATS(net, idev, IPSTATS_MIB_INHDRERRORS);
279 /* don't propagate ttl change to cloned packets */
280 if (skb_ensure_writable(skb, sizeof(struct ipv6hdr)))
283 ipv6_hdr(skb)->hop_limit--;
287 if (ip_hdr(skb)->ttl <= 1) {
288 /* Tell the sender its packet died... */
289 __IP_INC_STATS(net, IPSTATS_MIB_INHDRERRORS);
290 icmp_send(skb, ICMP_TIME_EXCEEDED, ICMP_EXC_TTL, 0);
294 /* don't propagate ttl change to cloned packets */
295 if (skb_ensure_writable(skb, sizeof(struct iphdr)))
299 ip_decrease_ttl(ip_hdr(skb));
305 /* Get route to destination or remote server */
307 __ip_vs_get_out_rt(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
308 struct ip_vs_dest *dest,
309 __be32 daddr, int rt_mode, __be32 *ret_saddr,
310 struct ip_vs_iphdr *ipvsh)
312 struct net *net = ipvs->net;
313 struct ip_vs_dest_dst *dest_dst;
314 struct rtable *rt; /* Route to the other host */
316 int local, noref = 1;
319 dest_dst = __ip_vs_dst_check(dest);
320 if (likely(dest_dst))
321 rt = (struct rtable *) dest_dst->dst_cache;
323 dest_dst = ip_vs_dest_dst_alloc();
324 spin_lock_bh(&dest->dst_lock);
326 __ip_vs_dst_set(dest, NULL, NULL, 0);
327 spin_unlock_bh(&dest->dst_lock);
330 rt = do_output_route4(net, dest->addr.ip, rt_mode,
331 &dest_dst->dst_saddr.ip);
333 __ip_vs_dst_set(dest, NULL, NULL, 0);
334 spin_unlock_bh(&dest->dst_lock);
335 ip_vs_dest_dst_free(dest_dst);
338 __ip_vs_dst_set(dest, dest_dst, &rt->dst, 0);
339 spin_unlock_bh(&dest->dst_lock);
340 IP_VS_DBG(10, "new dst %pI4, src %pI4, refcnt=%d\n",
341 &dest->addr.ip, &dest_dst->dst_saddr.ip,
342 atomic_read(&rt->dst.__refcnt));
345 *ret_saddr = dest_dst->dst_saddr.ip;
347 __be32 saddr = htonl(INADDR_ANY);
351 /* For such unconfigured boxes avoid many route lookups
352 * for performance reasons because we do not remember saddr
354 rt_mode &= ~IP_VS_RT_MODE_CONNECT;
355 rt = do_output_route4(net, daddr, rt_mode, &saddr);
362 local = (rt->rt_flags & RTCF_LOCAL) ? 1 : 0;
363 if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
365 IP_VS_DBG_RL("We are crossing local and non-local addresses"
366 " daddr=%pI4\n", &daddr);
370 if (unlikely(local)) {
371 /* skb to local stack, preserve old route */
377 if (!decrement_ttl(ipvs, skb_af, skb))
380 if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL))) {
381 mtu = dst_mtu(&rt->dst);
383 mtu = dst_mtu(&rt->dst) - sizeof(struct iphdr);
386 if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
387 mtu -= sizeof(struct udphdr) + sizeof(struct guehdr);
388 if ((dest->tun_flags &
389 IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
390 skb->ip_summed == CHECKSUM_PARTIAL)
391 mtu -= GUE_PLEN_REMCSUM + GUE_LEN_PRIV;
392 } else if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
395 if (dest->tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
396 tflags |= TUNNEL_CSUM;
397 mtu -= gre_calc_hlen(tflags);
400 IP_VS_DBG_RL("%s(): mtu less than 68\n", __func__);
403 maybe_update_pmtu(skb_af, skb, mtu);
406 if (!ensure_mtu_is_adequate(ipvs, skb_af, rt_mode, ipvsh, skb, mtu))
411 skb_dst_set_noref(skb, &rt->dst);
413 skb_dst_set(skb, &rt->dst);
423 dst_link_failure(skb);
427 #ifdef CONFIG_IP_VS_IPV6
428 static struct dst_entry *
429 __ip_vs_route_output_v6(struct net *net, struct in6_addr *daddr,
430 struct in6_addr *ret_saddr, int do_xfrm, int rt_mode)
432 struct dst_entry *dst;
433 struct flowi6 fl6 = {
437 if (rt_mode & IP_VS_RT_MODE_KNOWN_NH)
438 fl6.flowi6_flags = FLOWI_FLAG_KNOWN_NH;
440 dst = ip6_route_output(net, NULL, &fl6);
445 if (ipv6_addr_any(&fl6.saddr) &&
446 ipv6_dev_get_saddr(net, ip6_dst_idev(dst)->dev,
447 &fl6.daddr, 0, &fl6.saddr) < 0)
450 dst = xfrm_lookup(net, dst, flowi6_to_flowi(&fl6), NULL, 0);
456 *ret_saddr = fl6.saddr;
461 IP_VS_DBG_RL("ip6_route_output error, dest: %pI6\n", daddr);
466 * Get route to destination or remote server
469 __ip_vs_get_out_rt_v6(struct netns_ipvs *ipvs, int skb_af, struct sk_buff *skb,
470 struct ip_vs_dest *dest,
471 struct in6_addr *daddr, struct in6_addr *ret_saddr,
472 struct ip_vs_iphdr *ipvsh, int do_xfrm, int rt_mode)
474 struct net *net = ipvs->net;
475 struct ip_vs_dest_dst *dest_dst;
476 struct rt6_info *rt; /* Route to the other host */
477 struct dst_entry *dst;
479 int local, noref = 1;
482 dest_dst = __ip_vs_dst_check(dest);
483 if (likely(dest_dst))
484 rt = (struct rt6_info *) dest_dst->dst_cache;
488 dest_dst = ip_vs_dest_dst_alloc();
489 spin_lock_bh(&dest->dst_lock);
491 __ip_vs_dst_set(dest, NULL, NULL, 0);
492 spin_unlock_bh(&dest->dst_lock);
495 dst = __ip_vs_route_output_v6(net, &dest->addr.in6,
496 &dest_dst->dst_saddr.in6,
499 __ip_vs_dst_set(dest, NULL, NULL, 0);
500 spin_unlock_bh(&dest->dst_lock);
501 ip_vs_dest_dst_free(dest_dst);
504 rt = (struct rt6_info *) dst;
505 cookie = rt6_get_cookie(rt);
506 __ip_vs_dst_set(dest, dest_dst, &rt->dst, cookie);
507 spin_unlock_bh(&dest->dst_lock);
508 IP_VS_DBG(10, "new dst %pI6, src %pI6, refcnt=%d\n",
509 &dest->addr.in6, &dest_dst->dst_saddr.in6,
510 atomic_read(&rt->dst.__refcnt));
513 *ret_saddr = dest_dst->dst_saddr.in6;
516 dst = __ip_vs_route_output_v6(net, daddr, ret_saddr, do_xfrm,
520 rt = (struct rt6_info *) dst;
523 local = __ip_vs_is_local_route6(rt);
525 if (unlikely(crosses_local_route_boundary(skb_af, skb, rt_mode,
527 IP_VS_DBG_RL("We are crossing local and non-local addresses"
528 " daddr=%pI6\n", daddr);
532 if (unlikely(local)) {
533 /* skb to local stack, preserve old route */
535 dst_release(&rt->dst);
539 if (!decrement_ttl(ipvs, skb_af, skb))
543 if (likely(!(rt_mode & IP_VS_RT_MODE_TUNNEL)))
544 mtu = dst_mtu(&rt->dst);
546 mtu = dst_mtu(&rt->dst) - sizeof(struct ipv6hdr);
549 if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
550 mtu -= sizeof(struct udphdr) + sizeof(struct guehdr);
551 if ((dest->tun_flags &
552 IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
553 skb->ip_summed == CHECKSUM_PARTIAL)
554 mtu -= GUE_PLEN_REMCSUM + GUE_LEN_PRIV;
555 } else if (dest->tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
558 if (dest->tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
559 tflags |= TUNNEL_CSUM;
560 mtu -= gre_calc_hlen(tflags);
562 if (mtu < IPV6_MIN_MTU) {
563 IP_VS_DBG_RL("%s(): mtu less than %d\n", __func__,
567 maybe_update_pmtu(skb_af, skb, mtu);
570 if (!ensure_mtu_is_adequate(ipvs, skb_af, rt_mode, ipvsh, skb, mtu))
575 skb_dst_set_noref(skb, &rt->dst);
577 skb_dst_set(skb, &rt->dst);
583 dst_release(&rt->dst);
587 /* The ip6_link_failure function requires the dev field to be set
588 * in order to get the net (further for the sake of fwmark
592 skb->dev = skb_dst(skb)->dev;
594 dst_link_failure(skb);
600 /* return NF_ACCEPT to allow forwarding or other NF_xxx on error */
601 static inline int ip_vs_tunnel_xmit_prepare(struct sk_buff *skb,
602 struct ip_vs_conn *cp)
606 skb->ipvs_property = 1;
607 if (unlikely(cp->flags & IP_VS_CONN_F_NFCT))
608 ret = ip_vs_confirm_conntrack(skb);
609 if (ret == NF_ACCEPT) {
611 skb_forward_csum(skb);
616 /* In the event of a remote destination, it's possible that we would have
617 * matches against an old socket (particularly a TIME-WAIT socket). This
618 * causes havoc down the line (ip_local_out et. al. expect regular sockets
619 * and invalid memory accesses will happen) so simply drop the association
622 static inline void ip_vs_drop_early_demux_sk(struct sk_buff *skb)
624 /* If dev is set, the packet came from the LOCAL_IN callback and
625 * not from a local TCP socket.
631 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
632 static inline int ip_vs_nat_send_or_cont(int pf, struct sk_buff *skb,
633 struct ip_vs_conn *cp, int local)
637 skb->ipvs_property = 1;
638 if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
641 ip_vs_update_conntrack(skb, cp, 1);
643 /* Remove the early_demux association unless it's bound for the
644 * exact same port and address on this host after translation.
646 if (!local || cp->vport != cp->dport ||
647 !ip_vs_addr_equal(cp->af, &cp->vaddr, &cp->daddr))
648 ip_vs_drop_early_demux_sk(skb);
651 skb_forward_csum(skb);
652 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
653 NULL, skb_dst(skb)->dev, dst_output);
660 /* return NF_STOLEN (sent) or NF_ACCEPT if local=1 (not sent) */
661 static inline int ip_vs_send_or_cont(int pf, struct sk_buff *skb,
662 struct ip_vs_conn *cp, int local)
666 skb->ipvs_property = 1;
667 if (likely(!(cp->flags & IP_VS_CONN_F_NFCT)))
670 ip_vs_drop_early_demux_sk(skb);
671 skb_forward_csum(skb);
672 NF_HOOK(pf, NF_INET_LOCAL_OUT, cp->ipvs->net, NULL, skb,
673 NULL, skb_dst(skb)->dev, dst_output);
681 * NULL transmitter (do nothing except return NF_ACCEPT)
684 ip_vs_null_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
685 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
687 /* we do not touch skb and do not need pskb ptr */
688 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
694 * Let packets bypass the destination when the destination is not
695 * available, it may be only used in transparent cache cluster.
698 ip_vs_bypass_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
699 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
701 struct iphdr *iph = ip_hdr(skb);
705 if (__ip_vs_get_out_rt(cp->ipvs, cp->af, skb, NULL, iph->daddr,
706 IP_VS_RT_MODE_NON_LOCAL, NULL, ipvsh) < 0)
711 /* Another hack: avoid icmp_send in ip_fragment */
714 ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
725 #ifdef CONFIG_IP_VS_IPV6
727 ip_vs_bypass_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
728 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
730 struct ipv6hdr *iph = ipv6_hdr(skb);
734 if (__ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, NULL,
736 ipvsh, 0, IP_VS_RT_MODE_NON_LOCAL) < 0)
739 /* Another hack: avoid icmp_send in ip_fragment */
742 ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
755 * NAT transmitter (only for outside-to-inside nat forwarding)
756 * Not used for related ICMP
759 ip_vs_nat_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
760 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
762 struct rtable *rt; /* Route to the other host */
763 int local, rc, was_input;
767 /* check if it is a connection of no-client-port */
768 if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT)) {
771 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
774 ip_vs_conn_fill_cport(cp, *p);
775 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
778 was_input = rt_is_input_route(skb_rtable(skb));
779 local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
780 IP_VS_RT_MODE_LOCAL |
781 IP_VS_RT_MODE_NON_LOCAL |
782 IP_VS_RT_MODE_RDR, NULL, ipvsh);
785 rt = skb_rtable(skb);
787 * Avoid duplicate tuple in reply direction for NAT traffic
788 * to local address when connection is sync-ed
790 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
791 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
792 enum ip_conntrack_info ctinfo;
793 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
796 IP_VS_DBG_RL_PKT(10, AF_INET, pp, skb, ipvsh->off,
798 "stopping DNAT to local address");
804 /* From world but DNAT to loopback address? */
805 if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
806 IP_VS_DBG_RL_PKT(1, AF_INET, pp, skb, ipvsh->off,
807 "ip_vs_nat_xmit(): stopping DNAT to loopback "
812 /* copy-on-write the packet before mangling it */
813 if (skb_ensure_writable(skb, sizeof(struct iphdr)))
816 if (skb_cow(skb, rt->dst.dev->hard_header_len))
819 /* mangle the packet */
820 if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
822 ip_hdr(skb)->daddr = cp->daddr.ip;
823 ip_send_check(ip_hdr(skb));
825 IP_VS_DBG_PKT(10, AF_INET, pp, skb, ipvsh->off, "After DNAT");
827 /* FIXME: when application helper enlarges the packet and the length
828 is larger than the MTU of outgoing device, there will be still
831 /* Another hack: avoid icmp_send in ip_fragment */
834 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
845 #ifdef CONFIG_IP_VS_IPV6
847 ip_vs_nat_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
848 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
850 struct rt6_info *rt; /* Route to the other host */
855 /* check if it is a connection of no-client-port */
856 if (unlikely(cp->flags & IP_VS_CONN_F_NO_CPORT && !ipvsh->fragoffs)) {
858 p = skb_header_pointer(skb, ipvsh->len, sizeof(_pt), &_pt);
861 ip_vs_conn_fill_cport(cp, *p);
862 IP_VS_DBG(10, "filled cport=%d\n", ntohs(*p));
865 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
868 IP_VS_RT_MODE_LOCAL |
869 IP_VS_RT_MODE_NON_LOCAL |
873 rt = (struct rt6_info *) skb_dst(skb);
875 * Avoid duplicate tuple in reply direction for NAT traffic
876 * to local address when connection is sync-ed
878 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
879 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
880 enum ip_conntrack_info ctinfo;
881 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
884 IP_VS_DBG_RL_PKT(10, AF_INET6, pp, skb, ipvsh->off,
885 "ip_vs_nat_xmit_v6(): "
886 "stopping DNAT to local address");
892 /* From world but DNAT to loopback address? */
893 if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
894 ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
895 IP_VS_DBG_RL_PKT(1, AF_INET6, pp, skb, ipvsh->off,
896 "ip_vs_nat_xmit_v6(): "
897 "stopping DNAT to loopback address");
901 /* copy-on-write the packet before mangling it */
902 if (skb_ensure_writable(skb, sizeof(struct ipv6hdr)))
905 if (skb_cow(skb, rt->dst.dev->hard_header_len))
908 /* mangle the packet */
909 if (pp->dnat_handler && !pp->dnat_handler(skb, pp, cp, ipvsh))
911 ipv6_hdr(skb)->daddr = cp->daddr.in6;
913 IP_VS_DBG_PKT(10, AF_INET6, pp, skb, ipvsh->off, "After DNAT");
915 /* FIXME: when application helper enlarges the packet and the length
916 is larger than the MTU of outgoing device, there will be still
919 /* Another hack: avoid icmp_send in ip_fragment */
922 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);
934 /* When forwarding a packet, we must ensure that we've got enough headroom
935 * for the encapsulation packet in the skb. This also gives us an
936 * opportunity to figure out what the payload_len, dsfield, ttl, and df
937 * values should be, so that we won't need to look at the old ip header
940 static struct sk_buff *
941 ip_vs_prepare_tunneled_skb(struct sk_buff *skb, int skb_af,
942 unsigned int max_headroom, __u8 *next_protocol,
943 __u32 *payload_len, __u8 *dsfield, __u8 *ttl,
946 struct sk_buff *new_skb = NULL;
947 struct iphdr *old_iph = NULL;
949 #ifdef CONFIG_IP_VS_IPV6
950 struct ipv6hdr *old_ipv6h = NULL;
953 ip_vs_drop_early_demux_sk(skb);
955 if (skb_headroom(skb) < max_headroom || skb_cloned(skb)) {
956 new_skb = skb_realloc_headroom(skb, max_headroom);
960 skb_set_owner_w(new_skb, skb->sk);
965 #ifdef CONFIG_IP_VS_IPV6
966 if (skb_af == AF_INET6) {
967 old_ipv6h = ipv6_hdr(skb);
968 *next_protocol = IPPROTO_IPV6;
971 ntohs(old_ipv6h->payload_len) +
973 old_dsfield = ipv6_get_dsfield(old_ipv6h);
974 *ttl = old_ipv6h->hop_limit;
980 old_iph = ip_hdr(skb);
981 /* Copy DF, reset fragment offset and MF */
983 *df = (old_iph->frag_off & htons(IP_DF));
984 *next_protocol = IPPROTO_IPIP;
986 /* fix old IP header checksum */
987 ip_send_check(old_iph);
988 old_dsfield = ipv4_get_dsfield(old_iph);
991 *payload_len = ntohs(old_iph->tot_len);
994 /* Implement full-functionality option for ECN encapsulation */
995 *dsfield = INET_ECN_encapsulate(old_dsfield, old_dsfield);
1000 return ERR_PTR(-ENOMEM);
1003 static inline int __tun_gso_type_mask(int encaps_af, int orig_af)
1005 switch (encaps_af) {
1007 return SKB_GSO_IPXIP4;
1009 return SKB_GSO_IPXIP6;
1016 ipvs_gue_encap(struct net *net, struct sk_buff *skb,
1017 struct ip_vs_conn *cp, __u8 *next_protocol)
1020 __be16 sport = udp_flow_src_port(net, skb, 0, 0, false);
1021 struct udphdr *udph; /* Our new UDP header */
1022 struct guehdr *gueh; /* Our new GUE header */
1023 size_t hdrlen, optlen = 0;
1025 bool need_priv = false;
1027 if ((cp->dest->tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
1028 skb->ip_summed == CHECKSUM_PARTIAL) {
1029 optlen += GUE_PLEN_REMCSUM + GUE_LEN_PRIV;
1033 hdrlen = sizeof(struct guehdr) + optlen;
1035 skb_push(skb, hdrlen);
1037 gueh = (struct guehdr *)skb->data;
1041 gueh->hlen = optlen >> 2;
1043 gueh->proto_ctype = *next_protocol;
1048 __be32 *flags = data;
1049 u16 csum_start = skb_checksum_start_offset(skb);
1052 gueh->flags |= GUE_FLAG_PRIV;
1054 data += GUE_LEN_PRIV;
1056 if (csum_start < hdrlen)
1059 csum_start -= hdrlen;
1061 pd[0] = htons(csum_start);
1062 pd[1] = htons(csum_start + skb->csum_offset);
1064 if (!skb_is_gso(skb)) {
1065 skb->ip_summed = CHECKSUM_NONE;
1066 skb->encapsulation = 0;
1069 *flags |= GUE_PFLAG_REMCSUM;
1070 data += GUE_PLEN_REMCSUM;
1073 skb_push(skb, sizeof(struct udphdr));
1074 skb_reset_transport_header(skb);
1076 udph = udp_hdr(skb);
1078 dport = cp->dest->tun_port;
1080 udph->source = sport;
1081 udph->len = htons(skb->len);
1084 *next_protocol = IPPROTO_UDP;
1090 ipvs_gre_encap(struct net *net, struct sk_buff *skb,
1091 struct ip_vs_conn *cp, __u8 *next_protocol)
1093 __be16 proto = *next_protocol == IPPROTO_IPIP ?
1094 htons(ETH_P_IP) : htons(ETH_P_IPV6);
1098 if (cp->dest->tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
1099 tflags |= TUNNEL_CSUM;
1101 hdrlen = gre_calc_hlen(tflags);
1102 gre_build_header(skb, hdrlen, tflags, proto, 0, 0);
1104 *next_protocol = IPPROTO_GRE;
1108 * IP Tunneling transmitter
1110 * This function encapsulates the packet in a new IP packet, its
1111 * destination will be set to cp->daddr. Most code of this function
1112 * is taken from ipip.c.
1114 * It is used in VS/TUN cluster. The load balancer selects a real
1115 * server from a cluster based on a scheduling algorithm,
1116 * encapsulates the request packet and forwards it to the selected
1117 * server. For example, all real servers are configured with
1118 * "ifconfig tunl0 <Virtual IP Address> up". When the server receives
1119 * the encapsulated packet, it will decapsulate the packet, processe
1120 * the request and return the response packets directly to the client
1121 * without passing the load balancer. This can greatly increase the
1122 * scalability of virtual server.
1124 * Used for ANY protocol
1127 ip_vs_tunnel_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1128 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1130 struct netns_ipvs *ipvs = cp->ipvs;
1131 struct net *net = ipvs->net;
1132 struct rtable *rt; /* Route to the other host */
1133 __be32 saddr; /* Source for tunnel */
1134 struct net_device *tdev; /* Device to other host */
1135 __u8 next_protocol = 0;
1140 struct iphdr *iph; /* Our new IP header */
1141 unsigned int max_headroom; /* The extra header space needed */
1143 int tun_type, gso_type;
1148 local = __ip_vs_get_out_rt(ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
1149 IP_VS_RT_MODE_LOCAL |
1150 IP_VS_RT_MODE_NON_LOCAL |
1151 IP_VS_RT_MODE_CONNECT |
1152 IP_VS_RT_MODE_TUNNEL, &saddr, ipvsh);
1156 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1158 rt = skb_rtable(skb);
1162 * Okay, now see if we can stuff it in the buffer as-is.
1164 max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct iphdr);
1166 tun_type = cp->dest->tun_type;
1167 tun_flags = cp->dest->tun_flags;
1169 if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1170 size_t gue_hdrlen, gue_optlen = 0;
1172 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
1173 skb->ip_summed == CHECKSUM_PARTIAL) {
1174 gue_optlen += GUE_PLEN_REMCSUM + GUE_LEN_PRIV;
1176 gue_hdrlen = sizeof(struct guehdr) + gue_optlen;
1178 max_headroom += sizeof(struct udphdr) + gue_hdrlen;
1179 } else if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
1183 if (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
1184 tflags |= TUNNEL_CSUM;
1185 gre_hdrlen = gre_calc_hlen(tflags);
1187 max_headroom += gre_hdrlen;
1190 /* We only care about the df field if sysctl_pmtu_disc(ipvs) is set */
1191 dfp = sysctl_pmtu_disc(ipvs) ? &df : NULL;
1192 skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1193 &next_protocol, NULL, &dsfield,
1198 gso_type = __tun_gso_type_mask(AF_INET, cp->af);
1199 if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1200 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM) ||
1201 (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM))
1202 gso_type |= SKB_GSO_UDP_TUNNEL_CSUM;
1204 gso_type |= SKB_GSO_UDP_TUNNEL;
1205 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
1206 skb->ip_summed == CHECKSUM_PARTIAL) {
1207 gso_type |= SKB_GSO_TUNNEL_REMCSUM;
1209 } else if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
1210 if (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
1211 gso_type |= SKB_GSO_GRE_CSUM;
1213 gso_type |= SKB_GSO_GRE;
1216 if (iptunnel_handle_offloads(skb, gso_type))
1219 skb->transport_header = skb->network_header;
1221 skb_set_inner_ipproto(skb, next_protocol);
1223 if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1226 if (ipvs_gue_encap(net, skb, cp, &next_protocol))
1229 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM) ||
1230 (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM))
1233 udp_set_csum(!check, skb, saddr, cp->daddr.ip, skb->len);
1234 } else if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE)
1235 ipvs_gre_encap(net, skb, cp, &next_protocol);
1237 skb_push(skb, sizeof(struct iphdr));
1238 skb_reset_network_header(skb);
1239 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1242 * Push down and install the IPIP header.
1246 iph->ihl = sizeof(struct iphdr)>>2;
1248 iph->protocol = next_protocol;
1250 iph->daddr = cp->daddr.ip;
1253 ip_select_ident(net, skb, NULL);
1255 /* Another hack: avoid icmp_send in ip_fragment */
1258 ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1259 if (ret == NF_ACCEPT)
1260 ip_local_out(net, skb->sk, skb);
1261 else if (ret == NF_DROP)
1275 #ifdef CONFIG_IP_VS_IPV6
1277 ip_vs_tunnel_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1278 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1280 struct netns_ipvs *ipvs = cp->ipvs;
1281 struct net *net = ipvs->net;
1282 struct rt6_info *rt; /* Route to the other host */
1283 struct in6_addr saddr; /* Source for tunnel */
1284 struct net_device *tdev; /* Device to other host */
1285 __u8 next_protocol = 0;
1286 __u32 payload_len = 0;
1289 struct ipv6hdr *iph; /* Our new IP header */
1290 unsigned int max_headroom; /* The extra header space needed */
1292 int tun_type, gso_type;
1297 local = __ip_vs_get_out_rt_v6(ipvs, cp->af, skb, cp->dest,
1300 IP_VS_RT_MODE_LOCAL |
1301 IP_VS_RT_MODE_NON_LOCAL |
1302 IP_VS_RT_MODE_TUNNEL);
1306 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1308 rt = (struct rt6_info *) skb_dst(skb);
1312 * Okay, now see if we can stuff it in the buffer as-is.
1314 max_headroom = LL_RESERVED_SPACE(tdev) + sizeof(struct ipv6hdr);
1316 tun_type = cp->dest->tun_type;
1317 tun_flags = cp->dest->tun_flags;
1319 if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1320 size_t gue_hdrlen, gue_optlen = 0;
1322 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
1323 skb->ip_summed == CHECKSUM_PARTIAL) {
1324 gue_optlen += GUE_PLEN_REMCSUM + GUE_LEN_PRIV;
1326 gue_hdrlen = sizeof(struct guehdr) + gue_optlen;
1328 max_headroom += sizeof(struct udphdr) + gue_hdrlen;
1329 } else if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
1333 if (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
1334 tflags |= TUNNEL_CSUM;
1335 gre_hdrlen = gre_calc_hlen(tflags);
1337 max_headroom += gre_hdrlen;
1340 skb = ip_vs_prepare_tunneled_skb(skb, cp->af, max_headroom,
1341 &next_protocol, &payload_len,
1342 &dsfield, &ttl, NULL);
1346 gso_type = __tun_gso_type_mask(AF_INET6, cp->af);
1347 if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1348 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM) ||
1349 (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM))
1350 gso_type |= SKB_GSO_UDP_TUNNEL_CSUM;
1352 gso_type |= SKB_GSO_UDP_TUNNEL;
1353 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM) &&
1354 skb->ip_summed == CHECKSUM_PARTIAL) {
1355 gso_type |= SKB_GSO_TUNNEL_REMCSUM;
1357 } else if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE) {
1358 if (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM)
1359 gso_type |= SKB_GSO_GRE_CSUM;
1361 gso_type |= SKB_GSO_GRE;
1364 if (iptunnel_handle_offloads(skb, gso_type))
1367 skb->transport_header = skb->network_header;
1369 skb_set_inner_ipproto(skb, next_protocol);
1371 if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GUE) {
1374 if (ipvs_gue_encap(net, skb, cp, &next_protocol))
1377 if ((tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_CSUM) ||
1378 (tun_flags & IP_VS_TUNNEL_ENCAP_FLAG_REMCSUM))
1381 udp6_set_csum(!check, skb, &saddr, &cp->daddr.in6, skb->len);
1382 } else if (tun_type == IP_VS_CONN_F_TUNNEL_TYPE_GRE)
1383 ipvs_gre_encap(net, skb, cp, &next_protocol);
1385 skb_push(skb, sizeof(struct ipv6hdr));
1386 skb_reset_network_header(skb);
1387 memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1390 * Push down and install the IPIP header.
1392 iph = ipv6_hdr(skb);
1394 iph->nexthdr = next_protocol;
1395 iph->payload_len = htons(payload_len);
1396 memset(&iph->flow_lbl, 0, sizeof(iph->flow_lbl));
1397 ipv6_change_dsfield(iph, 0, dsfield);
1398 iph->daddr = cp->daddr.in6;
1400 iph->hop_limit = ttl;
1402 /* Another hack: avoid icmp_send in ip_fragment */
1405 ret = ip_vs_tunnel_xmit_prepare(skb, cp);
1406 if (ret == NF_ACCEPT)
1407 ip6_local_out(net, skb->sk, skb);
1408 else if (ret == NF_DROP)
1425 * Direct Routing transmitter
1426 * Used for ANY protocol
1429 ip_vs_dr_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1430 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1436 local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip,
1437 IP_VS_RT_MODE_LOCAL |
1438 IP_VS_RT_MODE_NON_LOCAL |
1439 IP_VS_RT_MODE_KNOWN_NH, NULL, ipvsh);
1443 return ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 1);
1445 ip_send_check(ip_hdr(skb));
1447 /* Another hack: avoid icmp_send in ip_fragment */
1450 ip_vs_send_or_cont(NFPROTO_IPV4, skb, cp, 0);
1461 #ifdef CONFIG_IP_VS_IPV6
1463 ip_vs_dr_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1464 struct ip_vs_protocol *pp, struct ip_vs_iphdr *ipvsh)
1470 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1473 IP_VS_RT_MODE_LOCAL |
1474 IP_VS_RT_MODE_NON_LOCAL |
1475 IP_VS_RT_MODE_KNOWN_NH);
1479 return ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 1);
1481 /* Another hack: avoid icmp_send in ip_fragment */
1484 ip_vs_send_or_cont(NFPROTO_IPV6, skb, cp, 0);
1498 * ICMP packet transmitter
1499 * called by the ip_vs_in_icmp
1502 ip_vs_icmp_xmit(struct sk_buff *skb, struct ip_vs_conn *cp,
1503 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1504 struct ip_vs_iphdr *iph)
1506 struct rtable *rt; /* Route to the other host */
1509 int rt_mode, was_input;
1513 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1514 forwarded directly here, because there is no need to
1515 translate address/port back */
1516 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1517 if (cp->packet_xmit)
1518 rc = cp->packet_xmit(skb, cp, pp, iph);
1521 /* do not touch skb anymore */
1522 atomic_inc(&cp->in_pkts);
1527 * mangle and send the packet here (only for VS/NAT)
1529 was_input = rt_is_input_route(skb_rtable(skb));
1531 /* LOCALNODE from FORWARD hook is not supported */
1532 rt_mode = (hooknum != NF_INET_FORWARD) ?
1533 IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1534 IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1535 local = __ip_vs_get_out_rt(cp->ipvs, cp->af, skb, cp->dest, cp->daddr.ip, rt_mode,
1539 rt = skb_rtable(skb);
1542 * Avoid duplicate tuple in reply direction for NAT traffic
1543 * to local address when connection is sync-ed
1545 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1546 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1547 enum ip_conntrack_info ctinfo;
1548 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1551 IP_VS_DBG(10, "%s(): "
1552 "stopping DNAT to local address %pI4\n",
1553 __func__, &cp->daddr.ip);
1559 /* From world but DNAT to loopback address? */
1560 if (local && ipv4_is_loopback(cp->daddr.ip) && was_input) {
1561 IP_VS_DBG(1, "%s(): "
1562 "stopping DNAT to loopback %pI4\n",
1563 __func__, &cp->daddr.ip);
1567 /* copy-on-write the packet before mangling it */
1568 if (skb_ensure_writable(skb, offset))
1571 if (skb_cow(skb, rt->dst.dev->hard_header_len))
1574 ip_vs_nat_icmp(skb, pp, cp, 0);
1576 /* Another hack: avoid icmp_send in ip_fragment */
1579 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV4, skb, cp, local);
1590 #ifdef CONFIG_IP_VS_IPV6
1592 ip_vs_icmp_xmit_v6(struct sk_buff *skb, struct ip_vs_conn *cp,
1593 struct ip_vs_protocol *pp, int offset, unsigned int hooknum,
1594 struct ip_vs_iphdr *ipvsh)
1596 struct rt6_info *rt; /* Route to the other host */
1603 /* The ICMP packet for VS/TUN, VS/DR and LOCALNODE will be
1604 forwarded directly here, because there is no need to
1605 translate address/port back */
1606 if (IP_VS_FWD_METHOD(cp) != IP_VS_CONN_F_MASQ) {
1607 if (cp->packet_xmit)
1608 rc = cp->packet_xmit(skb, cp, pp, ipvsh);
1611 /* do not touch skb anymore */
1612 atomic_inc(&cp->in_pkts);
1617 * mangle and send the packet here (only for VS/NAT)
1620 /* LOCALNODE from FORWARD hook is not supported */
1621 rt_mode = (hooknum != NF_INET_FORWARD) ?
1622 IP_VS_RT_MODE_LOCAL | IP_VS_RT_MODE_NON_LOCAL |
1623 IP_VS_RT_MODE_RDR : IP_VS_RT_MODE_NON_LOCAL;
1624 local = __ip_vs_get_out_rt_v6(cp->ipvs, cp->af, skb, cp->dest,
1625 &cp->daddr.in6, NULL, ipvsh, 0, rt_mode);
1628 rt = (struct rt6_info *) skb_dst(skb);
1630 * Avoid duplicate tuple in reply direction for NAT traffic
1631 * to local address when connection is sync-ed
1633 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1634 if (cp->flags & IP_VS_CONN_F_SYNC && local) {
1635 enum ip_conntrack_info ctinfo;
1636 struct nf_conn *ct = nf_ct_get(skb, &ctinfo);
1639 IP_VS_DBG(10, "%s(): "
1640 "stopping DNAT to local address %pI6\n",
1641 __func__, &cp->daddr.in6);
1647 /* From world but DNAT to loopback address? */
1648 if (local && skb->dev && !(skb->dev->flags & IFF_LOOPBACK) &&
1649 ipv6_addr_type(&cp->daddr.in6) & IPV6_ADDR_LOOPBACK) {
1650 IP_VS_DBG(1, "%s(): "
1651 "stopping DNAT to loopback %pI6\n",
1652 __func__, &cp->daddr.in6);
1656 /* copy-on-write the packet before mangling it */
1657 if (skb_ensure_writable(skb, offset))
1660 if (skb_cow(skb, rt->dst.dev->hard_header_len))
1663 ip_vs_nat_icmp_v6(skb, pp, cp, 0);
1665 /* Another hack: avoid icmp_send in ip_fragment */
1668 rc = ip_vs_nat_send_or_cont(NFPROTO_IPV6, skb, cp, local);